xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 83d3c4f2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
62 
63 #include <trace/events/kvm.h>
64 
65 #include <asm/debugreg.h>
66 #include <asm/msr.h>
67 #include <asm/desc.h>
68 #include <asm/mce.h>
69 #include <asm/pkru.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/internal.h> /* Ugh! */
72 #include <asm/pvclock.h>
73 #include <asm/div64.h>
74 #include <asm/irq_remapping.h>
75 #include <asm/mshyperv.h>
76 #include <asm/hypervisor.h>
77 #include <asm/tlbflush.h>
78 #include <asm/intel_pt.h>
79 #include <asm/emulate_prefix.h>
80 #include <asm/sgx.h>
81 #include <clocksource/hyperv_timer.h>
82 
83 #define CREATE_TRACE_POINTS
84 #include "trace.h"
85 
86 #define MAX_IO_MSRS 256
87 #define KVM_MAX_MCE_BANKS 32
88 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
89 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
90 
91 #define emul_to_vcpu(ctxt) \
92 	((struct kvm_vcpu *)(ctxt)->vcpu)
93 
94 /* EFER defaults:
95  * - enable syscall per default because its emulated by KVM
96  * - enable LME and LMA per default on 64 bit KVM
97  */
98 #ifdef CONFIG_X86_64
99 static
100 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
101 #else
102 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
103 #endif
104 
105 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
106 
107 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
108 
109 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
110                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
111 
112 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
113 static void process_nmi(struct kvm_vcpu *vcpu);
114 static void process_smi(struct kvm_vcpu *vcpu);
115 static void enter_smm(struct kvm_vcpu *vcpu);
116 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
117 static void store_regs(struct kvm_vcpu *vcpu);
118 static int sync_regs(struct kvm_vcpu *vcpu);
119 
120 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
121 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
122 
123 struct kvm_x86_ops kvm_x86_ops __read_mostly;
124 EXPORT_SYMBOL_GPL(kvm_x86_ops);
125 
126 #define KVM_X86_OP(func)					     \
127 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
128 				*(((struct kvm_x86_ops *)0)->func));
129 #define KVM_X86_OP_NULL KVM_X86_OP
130 #include <asm/kvm-x86-ops.h>
131 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
132 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
133 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
134 
135 static bool __read_mostly ignore_msrs = 0;
136 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
137 
138 bool __read_mostly report_ignored_msrs = true;
139 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
140 EXPORT_SYMBOL_GPL(report_ignored_msrs);
141 
142 unsigned int min_timer_period_us = 200;
143 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
144 
145 static bool __read_mostly kvmclock_periodic_sync = true;
146 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
147 
148 bool __read_mostly kvm_has_tsc_control;
149 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
150 u32  __read_mostly kvm_max_guest_tsc_khz;
151 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
152 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
153 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
154 u64  __read_mostly kvm_max_tsc_scaling_ratio;
155 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
156 u64 __read_mostly kvm_default_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
158 bool __read_mostly kvm_has_bus_lock_exit;
159 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
160 
161 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
162 static u32 __read_mostly tsc_tolerance_ppm = 250;
163 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
164 
165 /*
166  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
167  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
168  * advancement entirely.  Any other value is used as-is and disables adaptive
169  * tuning, i.e. allows privileged userspace to set an exact advancement time.
170  */
171 static int __read_mostly lapic_timer_advance_ns = -1;
172 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
173 
174 static bool __read_mostly vector_hashing = true;
175 module_param(vector_hashing, bool, S_IRUGO);
176 
177 bool __read_mostly enable_vmware_backdoor = false;
178 module_param(enable_vmware_backdoor, bool, S_IRUGO);
179 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
180 
181 static bool __read_mostly force_emulation_prefix = false;
182 module_param(force_emulation_prefix, bool, S_IRUGO);
183 
184 int __read_mostly pi_inject_timer = -1;
185 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
186 
187 /*
188  * Restoring the host value for MSRs that are only consumed when running in
189  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
190  * returns to userspace, i.e. the kernel can run with the guest's value.
191  */
192 #define KVM_MAX_NR_USER_RETURN_MSRS 16
193 
194 struct kvm_user_return_msrs {
195 	struct user_return_notifier urn;
196 	bool registered;
197 	struct kvm_user_return_msr_values {
198 		u64 host;
199 		u64 curr;
200 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
201 };
202 
203 u32 __read_mostly kvm_nr_uret_msrs;
204 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
205 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
206 static struct kvm_user_return_msrs __percpu *user_return_msrs;
207 
208 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
209 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
210 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
211 				| XFEATURE_MASK_PKRU)
212 
213 u64 __read_mostly host_efer;
214 EXPORT_SYMBOL_GPL(host_efer);
215 
216 bool __read_mostly allow_smaller_maxphyaddr = 0;
217 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
218 
219 bool __read_mostly enable_apicv = true;
220 EXPORT_SYMBOL_GPL(enable_apicv);
221 
222 u64 __read_mostly host_xss;
223 EXPORT_SYMBOL_GPL(host_xss);
224 u64 __read_mostly supported_xss;
225 EXPORT_SYMBOL_GPL(supported_xss);
226 
227 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
228 	KVM_GENERIC_VM_STATS(),
229 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
230 	STATS_DESC_COUNTER(VM, mmu_pte_write),
231 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
232 	STATS_DESC_COUNTER(VM, mmu_flooded),
233 	STATS_DESC_COUNTER(VM, mmu_recycled),
234 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
235 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
236 	STATS_DESC_ICOUNTER(VM, pages_4k),
237 	STATS_DESC_ICOUNTER(VM, pages_2m),
238 	STATS_DESC_ICOUNTER(VM, pages_1g),
239 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
240 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
241 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
242 };
243 
244 const struct kvm_stats_header kvm_vm_stats_header = {
245 	.name_size = KVM_STATS_NAME_SIZE,
246 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
247 	.id_offset = sizeof(struct kvm_stats_header),
248 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
249 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
250 		       sizeof(kvm_vm_stats_desc),
251 };
252 
253 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
254 	KVM_GENERIC_VCPU_STATS(),
255 	STATS_DESC_COUNTER(VCPU, pf_fixed),
256 	STATS_DESC_COUNTER(VCPU, pf_guest),
257 	STATS_DESC_COUNTER(VCPU, tlb_flush),
258 	STATS_DESC_COUNTER(VCPU, invlpg),
259 	STATS_DESC_COUNTER(VCPU, exits),
260 	STATS_DESC_COUNTER(VCPU, io_exits),
261 	STATS_DESC_COUNTER(VCPU, mmio_exits),
262 	STATS_DESC_COUNTER(VCPU, signal_exits),
263 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
264 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
265 	STATS_DESC_COUNTER(VCPU, l1d_flush),
266 	STATS_DESC_COUNTER(VCPU, halt_exits),
267 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
268 	STATS_DESC_COUNTER(VCPU, irq_exits),
269 	STATS_DESC_COUNTER(VCPU, host_state_reload),
270 	STATS_DESC_COUNTER(VCPU, fpu_reload),
271 	STATS_DESC_COUNTER(VCPU, insn_emulation),
272 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
273 	STATS_DESC_COUNTER(VCPU, hypercalls),
274 	STATS_DESC_COUNTER(VCPU, irq_injections),
275 	STATS_DESC_COUNTER(VCPU, nmi_injections),
276 	STATS_DESC_COUNTER(VCPU, req_event),
277 	STATS_DESC_COUNTER(VCPU, nested_run),
278 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
279 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
280 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
281 };
282 
283 const struct kvm_stats_header kvm_vcpu_stats_header = {
284 	.name_size = KVM_STATS_NAME_SIZE,
285 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
286 	.id_offset = sizeof(struct kvm_stats_header),
287 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
288 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
289 		       sizeof(kvm_vcpu_stats_desc),
290 };
291 
292 u64 __read_mostly host_xcr0;
293 u64 __read_mostly supported_xcr0;
294 EXPORT_SYMBOL_GPL(supported_xcr0);
295 
296 static struct kmem_cache *x86_fpu_cache;
297 
298 static struct kmem_cache *x86_emulator_cache;
299 
300 /*
301  * When called, it means the previous get/set msr reached an invalid msr.
302  * Return true if we want to ignore/silent this failed msr access.
303  */
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
305 {
306 	const char *op = write ? "wrmsr" : "rdmsr";
307 
308 	if (ignore_msrs) {
309 		if (report_ignored_msrs)
310 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
311 				      op, msr, data);
312 		/* Mask the error */
313 		return true;
314 	} else {
315 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
316 				      op, msr, data);
317 		return false;
318 	}
319 }
320 
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
322 {
323 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 	unsigned int size = sizeof(struct x86_emulate_ctxt);
325 
326 	return kmem_cache_create_usercopy("x86_emulator", size,
327 					  __alignof__(struct x86_emulate_ctxt),
328 					  SLAB_ACCOUNT, useroffset,
329 					  size - useroffset, NULL);
330 }
331 
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
333 
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
335 {
336 	int i;
337 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 		vcpu->arch.apf.gfns[i] = ~0;
339 }
340 
341 static void kvm_on_user_return(struct user_return_notifier *urn)
342 {
343 	unsigned slot;
344 	struct kvm_user_return_msrs *msrs
345 		= container_of(urn, struct kvm_user_return_msrs, urn);
346 	struct kvm_user_return_msr_values *values;
347 	unsigned long flags;
348 
349 	/*
350 	 * Disabling irqs at this point since the following code could be
351 	 * interrupted and executed through kvm_arch_hardware_disable()
352 	 */
353 	local_irq_save(flags);
354 	if (msrs->registered) {
355 		msrs->registered = false;
356 		user_return_notifier_unregister(urn);
357 	}
358 	local_irq_restore(flags);
359 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 		values = &msrs->values[slot];
361 		if (values->host != values->curr) {
362 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 			values->curr = values->host;
364 		}
365 	}
366 }
367 
368 static int kvm_probe_user_return_msr(u32 msr)
369 {
370 	u64 val;
371 	int ret;
372 
373 	preempt_disable();
374 	ret = rdmsrl_safe(msr, &val);
375 	if (ret)
376 		goto out;
377 	ret = wrmsrl_safe(msr, val);
378 out:
379 	preempt_enable();
380 	return ret;
381 }
382 
383 int kvm_add_user_return_msr(u32 msr)
384 {
385 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
386 
387 	if (kvm_probe_user_return_msr(msr))
388 		return -1;
389 
390 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 	return kvm_nr_uret_msrs++;
392 }
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
394 
395 int kvm_find_user_return_msr(u32 msr)
396 {
397 	int i;
398 
399 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 		if (kvm_uret_msrs_list[i] == msr)
401 			return i;
402 	}
403 	return -1;
404 }
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
406 
407 static void kvm_user_return_msr_cpu_online(void)
408 {
409 	unsigned int cpu = smp_processor_id();
410 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
411 	u64 value;
412 	int i;
413 
414 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 		msrs->values[i].host = value;
417 		msrs->values[i].curr = value;
418 	}
419 }
420 
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
422 {
423 	unsigned int cpu = smp_processor_id();
424 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 	int err;
426 
427 	value = (value & mask) | (msrs->values[slot].host & ~mask);
428 	if (value == msrs->values[slot].curr)
429 		return 0;
430 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
431 	if (err)
432 		return 1;
433 
434 	msrs->values[slot].curr = value;
435 	if (!msrs->registered) {
436 		msrs->urn.on_user_return = kvm_on_user_return;
437 		user_return_notifier_register(&msrs->urn);
438 		msrs->registered = true;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
443 
444 static void drop_user_return_notifiers(void)
445 {
446 	unsigned int cpu = smp_processor_id();
447 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
448 
449 	if (msrs->registered)
450 		kvm_on_user_return(&msrs->urn);
451 }
452 
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
454 {
455 	return vcpu->arch.apic_base;
456 }
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
458 
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
460 {
461 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
462 }
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
464 
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
466 {
467 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
471 
472 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
473 		return 1;
474 	if (!msr_info->host_initiated) {
475 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
476 			return 1;
477 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
478 			return 1;
479 	}
480 
481 	kvm_lapic_set_base(vcpu, msr_info->data);
482 	kvm_recalculate_apic_map(vcpu->kvm);
483 	return 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
486 
487 /*
488  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
489  *
490  * Hardware virtualization extension instructions may fault if a reboot turns
491  * off virtualization while processes are running.  Usually after catching the
492  * fault we just panic; during reboot instead the instruction is ignored.
493  */
494 noinstr void kvm_spurious_fault(void)
495 {
496 	/* Fault while not rebooting.  We want the trace. */
497 	BUG_ON(!kvm_rebooting);
498 }
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
500 
501 #define EXCPT_BENIGN		0
502 #define EXCPT_CONTRIBUTORY	1
503 #define EXCPT_PF		2
504 
505 static int exception_class(int vector)
506 {
507 	switch (vector) {
508 	case PF_VECTOR:
509 		return EXCPT_PF;
510 	case DE_VECTOR:
511 	case TS_VECTOR:
512 	case NP_VECTOR:
513 	case SS_VECTOR:
514 	case GP_VECTOR:
515 		return EXCPT_CONTRIBUTORY;
516 	default:
517 		break;
518 	}
519 	return EXCPT_BENIGN;
520 }
521 
522 #define EXCPT_FAULT		0
523 #define EXCPT_TRAP		1
524 #define EXCPT_ABORT		2
525 #define EXCPT_INTERRUPT		3
526 
527 static int exception_type(int vector)
528 {
529 	unsigned int mask;
530 
531 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 		return EXCPT_INTERRUPT;
533 
534 	mask = 1 << vector;
535 
536 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
537 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
538 		return EXCPT_TRAP;
539 
540 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
541 		return EXCPT_ABORT;
542 
543 	/* Reserved exceptions will result in fault */
544 	return EXCPT_FAULT;
545 }
546 
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
548 {
549 	unsigned nr = vcpu->arch.exception.nr;
550 	bool has_payload = vcpu->arch.exception.has_payload;
551 	unsigned long payload = vcpu->arch.exception.payload;
552 
553 	if (!has_payload)
554 		return;
555 
556 	switch (nr) {
557 	case DB_VECTOR:
558 		/*
559 		 * "Certain debug exceptions may clear bit 0-3.  The
560 		 * remaining contents of the DR6 register are never
561 		 * cleared by the processor".
562 		 */
563 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
564 		/*
565 		 * In order to reflect the #DB exception payload in guest
566 		 * dr6, three components need to be considered: active low
567 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
568 		 * DR6_BS and DR6_BT)
569 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 		 * In the target guest dr6:
571 		 * FIXED_1 bits should always be set.
572 		 * Active low bits should be cleared if 1-setting in payload.
573 		 * Active high bits should be set if 1-setting in payload.
574 		 *
575 		 * Note, the payload is compatible with the pending debug
576 		 * exceptions/exit qualification under VMX, that active_low bits
577 		 * are active high in payload.
578 		 * So they need to be flipped for DR6.
579 		 */
580 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 		vcpu->arch.dr6 |= payload;
582 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
583 
584 		/*
585 		 * The #DB payload is defined as compatible with the 'pending
586 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 		 * defined in the 'pending debug exceptions' field (enabled
588 		 * breakpoint), it is reserved and must be zero in DR6.
589 		 */
590 		vcpu->arch.dr6 &= ~BIT(12);
591 		break;
592 	case PF_VECTOR:
593 		vcpu->arch.cr2 = payload;
594 		break;
595 	}
596 
597 	vcpu->arch.exception.has_payload = false;
598 	vcpu->arch.exception.payload = 0;
599 }
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
601 
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 		unsigned nr, bool has_error, u32 error_code,
604 	        bool has_payload, unsigned long payload, bool reinject)
605 {
606 	u32 prev_nr;
607 	int class1, class2;
608 
609 	kvm_make_request(KVM_REQ_EVENT, vcpu);
610 
611 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
612 	queue:
613 		if (reinject) {
614 			/*
615 			 * On vmentry, vcpu->arch.exception.pending is only
616 			 * true if an event injection was blocked by
617 			 * nested_run_pending.  In that case, however,
618 			 * vcpu_enter_guest requests an immediate exit,
619 			 * and the guest shouldn't proceed far enough to
620 			 * need reinjection.
621 			 */
622 			WARN_ON_ONCE(vcpu->arch.exception.pending);
623 			vcpu->arch.exception.injected = true;
624 			if (WARN_ON_ONCE(has_payload)) {
625 				/*
626 				 * A reinjected event has already
627 				 * delivered its payload.
628 				 */
629 				has_payload = false;
630 				payload = 0;
631 			}
632 		} else {
633 			vcpu->arch.exception.pending = true;
634 			vcpu->arch.exception.injected = false;
635 		}
636 		vcpu->arch.exception.has_error_code = has_error;
637 		vcpu->arch.exception.nr = nr;
638 		vcpu->arch.exception.error_code = error_code;
639 		vcpu->arch.exception.has_payload = has_payload;
640 		vcpu->arch.exception.payload = payload;
641 		if (!is_guest_mode(vcpu))
642 			kvm_deliver_exception_payload(vcpu);
643 		return;
644 	}
645 
646 	/* to check exception */
647 	prev_nr = vcpu->arch.exception.nr;
648 	if (prev_nr == DF_VECTOR) {
649 		/* triple fault -> shutdown */
650 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
651 		return;
652 	}
653 	class1 = exception_class(prev_nr);
654 	class2 = exception_class(nr);
655 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
657 		/*
658 		 * Generate double fault per SDM Table 5-5.  Set
659 		 * exception.pending = true so that the double fault
660 		 * can trigger a nested vmexit.
661 		 */
662 		vcpu->arch.exception.pending = true;
663 		vcpu->arch.exception.injected = false;
664 		vcpu->arch.exception.has_error_code = true;
665 		vcpu->arch.exception.nr = DF_VECTOR;
666 		vcpu->arch.exception.error_code = 0;
667 		vcpu->arch.exception.has_payload = false;
668 		vcpu->arch.exception.payload = 0;
669 	} else
670 		/* replace previous exception with a new one in a hope
671 		   that instruction re-execution will regenerate lost
672 		   exception */
673 		goto queue;
674 }
675 
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
677 {
678 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
679 }
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
681 
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
683 {
684 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
685 }
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
687 
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 			   unsigned long payload)
690 {
691 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
692 }
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
694 
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 				    u32 error_code, unsigned long payload)
697 {
698 	kvm_multiple_exception(vcpu, nr, true, error_code,
699 			       true, payload, false);
700 }
701 
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
703 {
704 	if (err)
705 		kvm_inject_gp(vcpu, 0);
706 	else
707 		return kvm_skip_emulated_instruction(vcpu);
708 
709 	return 1;
710 }
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
712 
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
714 {
715 	++vcpu->stat.pf_guest;
716 	vcpu->arch.exception.nested_apf =
717 		is_guest_mode(vcpu) && fault->async_page_fault;
718 	if (vcpu->arch.exception.nested_apf) {
719 		vcpu->arch.apf.nested_apf_token = fault->address;
720 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
721 	} else {
722 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
723 					fault->address);
724 	}
725 }
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
727 
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 				    struct x86_exception *fault)
730 {
731 	struct kvm_mmu *fault_mmu;
732 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
733 
734 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
735 					       vcpu->arch.walk_mmu;
736 
737 	/*
738 	 * Invalidate the TLB entry for the faulting address, if it exists,
739 	 * else the access will fault indefinitely (and to emulate hardware).
740 	 */
741 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 	    !(fault->error_code & PFERR_RSVD_MASK))
743 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 				       fault_mmu->root_hpa);
745 
746 	fault_mmu->inject_page_fault(vcpu, fault);
747 	return fault->nested_page_fault;
748 }
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
750 
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
752 {
753 	atomic_inc(&vcpu->arch.nmi_queued);
754 	kvm_make_request(KVM_REQ_NMI, vcpu);
755 }
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
757 
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
759 {
760 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
761 }
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
763 
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
765 {
766 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
767 }
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
769 
770 /*
771  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
772  * a #GP and return false.
773  */
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
775 {
776 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
777 		return true;
778 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
779 	return false;
780 }
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
782 
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
784 {
785 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 		return true;
787 
788 	kvm_queue_exception(vcpu, UD_VECTOR);
789 	return false;
790 }
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
792 
793 /*
794  * This function will be used to read from the physical memory of the currently
795  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
796  * can read from guest physical or from the guest's guest physical memory.
797  */
798 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
799 			    gfn_t ngfn, void *data, int offset, int len,
800 			    u32 access)
801 {
802 	struct x86_exception exception;
803 	gfn_t real_gfn;
804 	gpa_t ngpa;
805 
806 	ngpa     = gfn_to_gpa(ngfn);
807 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
808 	if (real_gfn == UNMAPPED_GVA)
809 		return -EFAULT;
810 
811 	real_gfn = gpa_to_gfn(real_gfn);
812 
813 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
814 }
815 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
816 
817 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
818 {
819 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
820 }
821 
822 /*
823  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
824  */
825 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
826 {
827 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
828 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
829 	int i;
830 	int ret;
831 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
832 
833 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
834 				      offset * sizeof(u64), sizeof(pdpte),
835 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
836 	if (ret < 0) {
837 		ret = 0;
838 		goto out;
839 	}
840 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
841 		if ((pdpte[i] & PT_PRESENT_MASK) &&
842 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
843 			ret = 0;
844 			goto out;
845 		}
846 	}
847 	ret = 1;
848 
849 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
850 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
851 	vcpu->arch.pdptrs_from_userspace = false;
852 
853 out:
854 
855 	return ret;
856 }
857 EXPORT_SYMBOL_GPL(load_pdptrs);
858 
859 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
860 {
861 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
862 		kvm_clear_async_pf_completion_queue(vcpu);
863 		kvm_async_pf_hash_reset(vcpu);
864 	}
865 
866 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
867 		kvm_mmu_reset_context(vcpu);
868 
869 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
870 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
871 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
872 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
873 }
874 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
875 
876 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
877 {
878 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
879 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
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) & pdptr_bits) &&
910 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
911 		return 1;
912 
913 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
914 		return 1;
915 
916 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
917 
918 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
919 
920 	return 0;
921 }
922 EXPORT_SYMBOL_GPL(kvm_set_cr0);
923 
924 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
925 {
926 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
927 }
928 EXPORT_SYMBOL_GPL(kvm_lmsw);
929 
930 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
931 {
932 	if (vcpu->arch.guest_state_protected)
933 		return;
934 
935 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
936 
937 		if (vcpu->arch.xcr0 != host_xcr0)
938 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
939 
940 		if (vcpu->arch.xsaves_enabled &&
941 		    vcpu->arch.ia32_xss != host_xss)
942 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
943 	}
944 
945 	if (static_cpu_has(X86_FEATURE_PKU) &&
946 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
947 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
948 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
949 		write_pkru(vcpu->arch.pkru);
950 }
951 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
952 
953 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
954 {
955 	if (vcpu->arch.guest_state_protected)
956 		return;
957 
958 	if (static_cpu_has(X86_FEATURE_PKU) &&
959 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
960 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
961 		vcpu->arch.pkru = rdpkru();
962 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
963 			write_pkru(vcpu->arch.host_pkru);
964 	}
965 
966 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
967 
968 		if (vcpu->arch.xcr0 != host_xcr0)
969 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
970 
971 		if (vcpu->arch.xsaves_enabled &&
972 		    vcpu->arch.ia32_xss != host_xss)
973 			wrmsrl(MSR_IA32_XSS, host_xss);
974 	}
975 
976 }
977 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
978 
979 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
980 {
981 	u64 xcr0 = xcr;
982 	u64 old_xcr0 = vcpu->arch.xcr0;
983 	u64 valid_bits;
984 
985 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
986 	if (index != XCR_XFEATURE_ENABLED_MASK)
987 		return 1;
988 	if (!(xcr0 & XFEATURE_MASK_FP))
989 		return 1;
990 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
991 		return 1;
992 
993 	/*
994 	 * Do not allow the guest to set bits that we do not support
995 	 * saving.  However, xcr0 bit 0 is always set, even if the
996 	 * emulated CPU does not support XSAVE (see fx_init).
997 	 */
998 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
999 	if (xcr0 & ~valid_bits)
1000 		return 1;
1001 
1002 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1003 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1004 		return 1;
1005 
1006 	if (xcr0 & XFEATURE_MASK_AVX512) {
1007 		if (!(xcr0 & XFEATURE_MASK_YMM))
1008 			return 1;
1009 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1010 			return 1;
1011 	}
1012 	vcpu->arch.xcr0 = xcr0;
1013 
1014 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1015 		kvm_update_cpuid_runtime(vcpu);
1016 	return 0;
1017 }
1018 
1019 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1020 {
1021 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1022 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1023 		kvm_inject_gp(vcpu, 0);
1024 		return 1;
1025 	}
1026 
1027 	return kvm_skip_emulated_instruction(vcpu);
1028 }
1029 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1030 
1031 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1032 {
1033 	if (cr4 & cr4_reserved_bits)
1034 		return false;
1035 
1036 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1037 		return false;
1038 
1039 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1040 }
1041 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1042 
1043 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1044 {
1045 	if (((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS) ||
1046 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1047 		kvm_mmu_reset_context(vcpu);
1048 }
1049 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1050 
1051 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1052 {
1053 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1054 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1055 				   X86_CR4_SMEP;
1056 
1057 	if (!kvm_is_valid_cr4(vcpu, cr4))
1058 		return 1;
1059 
1060 	if (is_long_mode(vcpu)) {
1061 		if (!(cr4 & X86_CR4_PAE))
1062 			return 1;
1063 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1064 			return 1;
1065 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1066 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1067 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1068 				   kvm_read_cr3(vcpu)))
1069 		return 1;
1070 
1071 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1072 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1073 			return 1;
1074 
1075 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1076 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1077 			return 1;
1078 	}
1079 
1080 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1081 
1082 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1083 
1084 	return 0;
1085 }
1086 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1087 
1088 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1089 {
1090 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1091 	unsigned long roots_to_free = 0;
1092 	int i;
1093 
1094 	/*
1095 	 * If neither the current CR3 nor any of the prev_roots use the given
1096 	 * PCID, then nothing needs to be done here because a resync will
1097 	 * happen anyway before switching to any other CR3.
1098 	 */
1099 	if (kvm_get_active_pcid(vcpu) == pcid) {
1100 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1101 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1102 	}
1103 
1104 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1105 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1106 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1107 
1108 	kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1109 }
1110 
1111 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1112 {
1113 	bool skip_tlb_flush = false;
1114 	unsigned long pcid = 0;
1115 #ifdef CONFIG_X86_64
1116 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1117 
1118 	if (pcid_enabled) {
1119 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1120 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1121 		pcid = cr3 & X86_CR3_PCID_MASK;
1122 	}
1123 #endif
1124 
1125 	/* PDPTRs are always reloaded for PAE paging. */
1126 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1127 		goto handle_tlb_flush;
1128 
1129 	/*
1130 	 * Do not condition the GPA check on long mode, this helper is used to
1131 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1132 	 * the current vCPU mode is accurate.
1133 	 */
1134 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1135 		return 1;
1136 
1137 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1138 		return 1;
1139 
1140 	if (cr3 != kvm_read_cr3(vcpu))
1141 		kvm_mmu_new_pgd(vcpu, cr3);
1142 
1143 	vcpu->arch.cr3 = cr3;
1144 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1145 
1146 handle_tlb_flush:
1147 	/*
1148 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1149 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1150 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1151 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1152 	 * i.e. only PCID=0 can be relevant.
1153 	 */
1154 	if (!skip_tlb_flush)
1155 		kvm_invalidate_pcid(vcpu, pcid);
1156 
1157 	return 0;
1158 }
1159 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1160 
1161 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1162 {
1163 	if (cr8 & CR8_RESERVED_BITS)
1164 		return 1;
1165 	if (lapic_in_kernel(vcpu))
1166 		kvm_lapic_set_tpr(vcpu, cr8);
1167 	else
1168 		vcpu->arch.cr8 = cr8;
1169 	return 0;
1170 }
1171 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1172 
1173 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1174 {
1175 	if (lapic_in_kernel(vcpu))
1176 		return kvm_lapic_get_cr8(vcpu);
1177 	else
1178 		return vcpu->arch.cr8;
1179 }
1180 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1181 
1182 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1183 {
1184 	int i;
1185 
1186 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1187 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1188 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1189 	}
1190 }
1191 
1192 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1193 {
1194 	unsigned long dr7;
1195 
1196 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1197 		dr7 = vcpu->arch.guest_debug_dr7;
1198 	else
1199 		dr7 = vcpu->arch.dr7;
1200 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1201 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1202 	if (dr7 & DR7_BP_EN_MASK)
1203 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1204 }
1205 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1206 
1207 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1208 {
1209 	u64 fixed = DR6_FIXED_1;
1210 
1211 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1212 		fixed |= DR6_RTM;
1213 
1214 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1215 		fixed |= DR6_BUS_LOCK;
1216 	return fixed;
1217 }
1218 
1219 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1220 {
1221 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1222 
1223 	switch (dr) {
1224 	case 0 ... 3:
1225 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1226 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1227 			vcpu->arch.eff_db[dr] = val;
1228 		break;
1229 	case 4:
1230 	case 6:
1231 		if (!kvm_dr6_valid(val))
1232 			return 1; /* #GP */
1233 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1234 		break;
1235 	case 5:
1236 	default: /* 7 */
1237 		if (!kvm_dr7_valid(val))
1238 			return 1; /* #GP */
1239 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1240 		kvm_update_dr7(vcpu);
1241 		break;
1242 	}
1243 
1244 	return 0;
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_set_dr);
1247 
1248 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1249 {
1250 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1251 
1252 	switch (dr) {
1253 	case 0 ... 3:
1254 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1255 		break;
1256 	case 4:
1257 	case 6:
1258 		*val = vcpu->arch.dr6;
1259 		break;
1260 	case 5:
1261 	default: /* 7 */
1262 		*val = vcpu->arch.dr7;
1263 		break;
1264 	}
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_get_dr);
1267 
1268 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1269 {
1270 	u32 ecx = kvm_rcx_read(vcpu);
1271 	u64 data;
1272 
1273 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1274 		kvm_inject_gp(vcpu, 0);
1275 		return 1;
1276 	}
1277 
1278 	kvm_rax_write(vcpu, (u32)data);
1279 	kvm_rdx_write(vcpu, data >> 32);
1280 	return kvm_skip_emulated_instruction(vcpu);
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1283 
1284 /*
1285  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1286  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1287  *
1288  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1289  * extract the supported MSRs from the related const lists.
1290  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1291  * capabilities of the host cpu. This capabilities test skips MSRs that are
1292  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1293  * may depend on host virtualization features rather than host cpu features.
1294  */
1295 
1296 static const u32 msrs_to_save_all[] = {
1297 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1298 	MSR_STAR,
1299 #ifdef CONFIG_X86_64
1300 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1301 #endif
1302 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1303 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1304 	MSR_IA32_SPEC_CTRL,
1305 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1306 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1307 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1308 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1309 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1310 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1311 	MSR_IA32_UMWAIT_CONTROL,
1312 
1313 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1314 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1315 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1316 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1317 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1318 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1319 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1320 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1321 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1322 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1323 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1324 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1325 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1326 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1327 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1328 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1329 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1330 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1331 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1332 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1333 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1334 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1335 
1336 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1337 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1338 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1339 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1340 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1341 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1342 };
1343 
1344 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1345 static unsigned num_msrs_to_save;
1346 
1347 static const u32 emulated_msrs_all[] = {
1348 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1349 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1350 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1351 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1352 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1353 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1354 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1355 	HV_X64_MSR_RESET,
1356 	HV_X64_MSR_VP_INDEX,
1357 	HV_X64_MSR_VP_RUNTIME,
1358 	HV_X64_MSR_SCONTROL,
1359 	HV_X64_MSR_STIMER0_CONFIG,
1360 	HV_X64_MSR_VP_ASSIST_PAGE,
1361 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1362 	HV_X64_MSR_TSC_EMULATION_STATUS,
1363 	HV_X64_MSR_SYNDBG_OPTIONS,
1364 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1365 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1366 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1367 
1368 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1369 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1370 
1371 	MSR_IA32_TSC_ADJUST,
1372 	MSR_IA32_TSC_DEADLINE,
1373 	MSR_IA32_ARCH_CAPABILITIES,
1374 	MSR_IA32_PERF_CAPABILITIES,
1375 	MSR_IA32_MISC_ENABLE,
1376 	MSR_IA32_MCG_STATUS,
1377 	MSR_IA32_MCG_CTL,
1378 	MSR_IA32_MCG_EXT_CTL,
1379 	MSR_IA32_SMBASE,
1380 	MSR_SMI_COUNT,
1381 	MSR_PLATFORM_INFO,
1382 	MSR_MISC_FEATURES_ENABLES,
1383 	MSR_AMD64_VIRT_SPEC_CTRL,
1384 	MSR_IA32_POWER_CTL,
1385 	MSR_IA32_UCODE_REV,
1386 
1387 	/*
1388 	 * The following list leaves out MSRs whose values are determined
1389 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1390 	 * We always support the "true" VMX control MSRs, even if the host
1391 	 * processor does not, so I am putting these registers here rather
1392 	 * than in msrs_to_save_all.
1393 	 */
1394 	MSR_IA32_VMX_BASIC,
1395 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1396 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1397 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1398 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1399 	MSR_IA32_VMX_MISC,
1400 	MSR_IA32_VMX_CR0_FIXED0,
1401 	MSR_IA32_VMX_CR4_FIXED0,
1402 	MSR_IA32_VMX_VMCS_ENUM,
1403 	MSR_IA32_VMX_PROCBASED_CTLS2,
1404 	MSR_IA32_VMX_EPT_VPID_CAP,
1405 	MSR_IA32_VMX_VMFUNC,
1406 
1407 	MSR_K7_HWCR,
1408 	MSR_KVM_POLL_CONTROL,
1409 };
1410 
1411 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1412 static unsigned num_emulated_msrs;
1413 
1414 /*
1415  * List of msr numbers which are used to expose MSR-based features that
1416  * can be used by a hypervisor to validate requested CPU features.
1417  */
1418 static const u32 msr_based_features_all[] = {
1419 	MSR_IA32_VMX_BASIC,
1420 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1421 	MSR_IA32_VMX_PINBASED_CTLS,
1422 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1423 	MSR_IA32_VMX_PROCBASED_CTLS,
1424 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1425 	MSR_IA32_VMX_EXIT_CTLS,
1426 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1427 	MSR_IA32_VMX_ENTRY_CTLS,
1428 	MSR_IA32_VMX_MISC,
1429 	MSR_IA32_VMX_CR0_FIXED0,
1430 	MSR_IA32_VMX_CR0_FIXED1,
1431 	MSR_IA32_VMX_CR4_FIXED0,
1432 	MSR_IA32_VMX_CR4_FIXED1,
1433 	MSR_IA32_VMX_VMCS_ENUM,
1434 	MSR_IA32_VMX_PROCBASED_CTLS2,
1435 	MSR_IA32_VMX_EPT_VPID_CAP,
1436 	MSR_IA32_VMX_VMFUNC,
1437 
1438 	MSR_F10H_DECFG,
1439 	MSR_IA32_UCODE_REV,
1440 	MSR_IA32_ARCH_CAPABILITIES,
1441 	MSR_IA32_PERF_CAPABILITIES,
1442 };
1443 
1444 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1445 static unsigned int num_msr_based_features;
1446 
1447 static u64 kvm_get_arch_capabilities(void)
1448 {
1449 	u64 data = 0;
1450 
1451 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1452 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1453 
1454 	/*
1455 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1456 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1457 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1458 	 * L1 guests, so it need not worry about its own (L2) guests.
1459 	 */
1460 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1461 
1462 	/*
1463 	 * If we're doing cache flushes (either "always" or "cond")
1464 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1465 	 * If an outer hypervisor is doing the cache flush for us
1466 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1467 	 * capability to the guest too, and if EPT is disabled we're not
1468 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1469 	 * require a nested hypervisor to do a flush of its own.
1470 	 */
1471 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1472 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1473 
1474 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1475 		data |= ARCH_CAP_RDCL_NO;
1476 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1477 		data |= ARCH_CAP_SSB_NO;
1478 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1479 		data |= ARCH_CAP_MDS_NO;
1480 
1481 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1482 		/*
1483 		 * If RTM=0 because the kernel has disabled TSX, the host might
1484 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1485 		 * and therefore knows that there cannot be TAA) but keep
1486 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1487 		 * and we want to allow migrating those guests to tsx=off hosts.
1488 		 */
1489 		data &= ~ARCH_CAP_TAA_NO;
1490 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1491 		data |= ARCH_CAP_TAA_NO;
1492 	} else {
1493 		/*
1494 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1495 		 * host so the guest can choose between disabling TSX or
1496 		 * using VERW to clear CPU buffers.
1497 		 */
1498 	}
1499 
1500 	return data;
1501 }
1502 
1503 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1504 {
1505 	switch (msr->index) {
1506 	case MSR_IA32_ARCH_CAPABILITIES:
1507 		msr->data = kvm_get_arch_capabilities();
1508 		break;
1509 	case MSR_IA32_UCODE_REV:
1510 		rdmsrl_safe(msr->index, &msr->data);
1511 		break;
1512 	default:
1513 		return static_call(kvm_x86_get_msr_feature)(msr);
1514 	}
1515 	return 0;
1516 }
1517 
1518 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1519 {
1520 	struct kvm_msr_entry msr;
1521 	int r;
1522 
1523 	msr.index = index;
1524 	r = kvm_get_msr_feature(&msr);
1525 
1526 	if (r == KVM_MSR_RET_INVALID) {
1527 		/* Unconditionally clear the output for simplicity */
1528 		*data = 0;
1529 		if (kvm_msr_ignored_check(index, 0, false))
1530 			r = 0;
1531 	}
1532 
1533 	if (r)
1534 		return r;
1535 
1536 	*data = msr.data;
1537 
1538 	return 0;
1539 }
1540 
1541 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1542 {
1543 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1544 		return false;
1545 
1546 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1547 		return false;
1548 
1549 	if (efer & (EFER_LME | EFER_LMA) &&
1550 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1551 		return false;
1552 
1553 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1554 		return false;
1555 
1556 	return true;
1557 
1558 }
1559 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1560 {
1561 	if (efer & efer_reserved_bits)
1562 		return false;
1563 
1564 	return __kvm_valid_efer(vcpu, efer);
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1567 
1568 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1569 {
1570 	u64 old_efer = vcpu->arch.efer;
1571 	u64 efer = msr_info->data;
1572 	int r;
1573 
1574 	if (efer & efer_reserved_bits)
1575 		return 1;
1576 
1577 	if (!msr_info->host_initiated) {
1578 		if (!__kvm_valid_efer(vcpu, efer))
1579 			return 1;
1580 
1581 		if (is_paging(vcpu) &&
1582 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1583 			return 1;
1584 	}
1585 
1586 	efer &= ~EFER_LMA;
1587 	efer |= vcpu->arch.efer & EFER_LMA;
1588 
1589 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1590 	if (r) {
1591 		WARN_ON(r > 0);
1592 		return r;
1593 	}
1594 
1595 	/* Update reserved bits */
1596 	if ((efer ^ old_efer) & EFER_NX)
1597 		kvm_mmu_reset_context(vcpu);
1598 
1599 	return 0;
1600 }
1601 
1602 void kvm_enable_efer_bits(u64 mask)
1603 {
1604        efer_reserved_bits &= ~mask;
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1607 
1608 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1609 {
1610 	struct kvm_x86_msr_filter *msr_filter;
1611 	struct msr_bitmap_range *ranges;
1612 	struct kvm *kvm = vcpu->kvm;
1613 	bool allowed;
1614 	int idx;
1615 	u32 i;
1616 
1617 	/* x2APIC MSRs do not support filtering. */
1618 	if (index >= 0x800 && index <= 0x8ff)
1619 		return true;
1620 
1621 	idx = srcu_read_lock(&kvm->srcu);
1622 
1623 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1624 	if (!msr_filter) {
1625 		allowed = true;
1626 		goto out;
1627 	}
1628 
1629 	allowed = msr_filter->default_allow;
1630 	ranges = msr_filter->ranges;
1631 
1632 	for (i = 0; i < msr_filter->count; i++) {
1633 		u32 start = ranges[i].base;
1634 		u32 end = start + ranges[i].nmsrs;
1635 		u32 flags = ranges[i].flags;
1636 		unsigned long *bitmap = ranges[i].bitmap;
1637 
1638 		if ((index >= start) && (index < end) && (flags & type)) {
1639 			allowed = !!test_bit(index - start, bitmap);
1640 			break;
1641 		}
1642 	}
1643 
1644 out:
1645 	srcu_read_unlock(&kvm->srcu, idx);
1646 
1647 	return allowed;
1648 }
1649 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1650 
1651 /*
1652  * Write @data into the MSR specified by @index.  Select MSR specific fault
1653  * checks are bypassed if @host_initiated is %true.
1654  * Returns 0 on success, non-0 otherwise.
1655  * Assumes vcpu_load() was already called.
1656  */
1657 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1658 			 bool host_initiated)
1659 {
1660 	struct msr_data msr;
1661 
1662 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1663 		return KVM_MSR_RET_FILTERED;
1664 
1665 	switch (index) {
1666 	case MSR_FS_BASE:
1667 	case MSR_GS_BASE:
1668 	case MSR_KERNEL_GS_BASE:
1669 	case MSR_CSTAR:
1670 	case MSR_LSTAR:
1671 		if (is_noncanonical_address(data, vcpu))
1672 			return 1;
1673 		break;
1674 	case MSR_IA32_SYSENTER_EIP:
1675 	case MSR_IA32_SYSENTER_ESP:
1676 		/*
1677 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1678 		 * non-canonical address is written on Intel but not on
1679 		 * AMD (which ignores the top 32-bits, because it does
1680 		 * not implement 64-bit SYSENTER).
1681 		 *
1682 		 * 64-bit code should hence be able to write a non-canonical
1683 		 * value on AMD.  Making the address canonical ensures that
1684 		 * vmentry does not fail on Intel after writing a non-canonical
1685 		 * value, and that something deterministic happens if the guest
1686 		 * invokes 64-bit SYSENTER.
1687 		 */
1688 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1689 		break;
1690 	case MSR_TSC_AUX:
1691 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1692 			return 1;
1693 
1694 		if (!host_initiated &&
1695 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1696 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1697 			return 1;
1698 
1699 		/*
1700 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1701 		 * incomplete and conflicting architectural behavior.  Current
1702 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1703 		 * reserved and always read as zeros.  Enforce Intel's reserved
1704 		 * bits check if and only if the guest CPU is Intel, and clear
1705 		 * the bits in all other cases.  This ensures cross-vendor
1706 		 * migration will provide consistent behavior for the guest.
1707 		 */
1708 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1709 			return 1;
1710 
1711 		data = (u32)data;
1712 		break;
1713 	}
1714 
1715 	msr.data = data;
1716 	msr.index = index;
1717 	msr.host_initiated = host_initiated;
1718 
1719 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1720 }
1721 
1722 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1723 				     u32 index, u64 data, bool host_initiated)
1724 {
1725 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1726 
1727 	if (ret == KVM_MSR_RET_INVALID)
1728 		if (kvm_msr_ignored_check(index, data, true))
1729 			ret = 0;
1730 
1731 	return ret;
1732 }
1733 
1734 /*
1735  * Read the MSR specified by @index into @data.  Select MSR specific fault
1736  * checks are bypassed if @host_initiated is %true.
1737  * Returns 0 on success, non-0 otherwise.
1738  * Assumes vcpu_load() was already called.
1739  */
1740 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1741 		  bool host_initiated)
1742 {
1743 	struct msr_data msr;
1744 	int ret;
1745 
1746 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1747 		return KVM_MSR_RET_FILTERED;
1748 
1749 	switch (index) {
1750 	case MSR_TSC_AUX:
1751 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1752 			return 1;
1753 
1754 		if (!host_initiated &&
1755 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1756 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1757 			return 1;
1758 		break;
1759 	}
1760 
1761 	msr.index = index;
1762 	msr.host_initiated = host_initiated;
1763 
1764 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1765 	if (!ret)
1766 		*data = msr.data;
1767 	return ret;
1768 }
1769 
1770 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1771 				     u32 index, u64 *data, bool host_initiated)
1772 {
1773 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1774 
1775 	if (ret == KVM_MSR_RET_INVALID) {
1776 		/* Unconditionally clear *data for simplicity */
1777 		*data = 0;
1778 		if (kvm_msr_ignored_check(index, 0, false))
1779 			ret = 0;
1780 	}
1781 
1782 	return ret;
1783 }
1784 
1785 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1786 {
1787 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1788 }
1789 EXPORT_SYMBOL_GPL(kvm_get_msr);
1790 
1791 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1792 {
1793 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1794 }
1795 EXPORT_SYMBOL_GPL(kvm_set_msr);
1796 
1797 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1798 {
1799 	int err = vcpu->run->msr.error;
1800 	if (!err) {
1801 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1802 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1803 	}
1804 
1805 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1806 }
1807 
1808 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1809 {
1810 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1811 }
1812 
1813 static u64 kvm_msr_reason(int r)
1814 {
1815 	switch (r) {
1816 	case KVM_MSR_RET_INVALID:
1817 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1818 	case KVM_MSR_RET_FILTERED:
1819 		return KVM_MSR_EXIT_REASON_FILTER;
1820 	default:
1821 		return KVM_MSR_EXIT_REASON_INVAL;
1822 	}
1823 }
1824 
1825 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1826 			      u32 exit_reason, u64 data,
1827 			      int (*completion)(struct kvm_vcpu *vcpu),
1828 			      int r)
1829 {
1830 	u64 msr_reason = kvm_msr_reason(r);
1831 
1832 	/* Check if the user wanted to know about this MSR fault */
1833 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1834 		return 0;
1835 
1836 	vcpu->run->exit_reason = exit_reason;
1837 	vcpu->run->msr.error = 0;
1838 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1839 	vcpu->run->msr.reason = msr_reason;
1840 	vcpu->run->msr.index = index;
1841 	vcpu->run->msr.data = data;
1842 	vcpu->arch.complete_userspace_io = completion;
1843 
1844 	return 1;
1845 }
1846 
1847 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1848 {
1849 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1850 				   complete_emulated_rdmsr, r);
1851 }
1852 
1853 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1854 {
1855 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1856 				   complete_emulated_wrmsr, r);
1857 }
1858 
1859 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1860 {
1861 	u32 ecx = kvm_rcx_read(vcpu);
1862 	u64 data;
1863 	int r;
1864 
1865 	r = kvm_get_msr(vcpu, ecx, &data);
1866 
1867 	/* MSR read failed? See if we should ask user space */
1868 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1869 		/* Bounce to user space */
1870 		return 0;
1871 	}
1872 
1873 	if (!r) {
1874 		trace_kvm_msr_read(ecx, data);
1875 
1876 		kvm_rax_write(vcpu, data & -1u);
1877 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1878 	} else {
1879 		trace_kvm_msr_read_ex(ecx);
1880 	}
1881 
1882 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1883 }
1884 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1885 
1886 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1887 {
1888 	u32 ecx = kvm_rcx_read(vcpu);
1889 	u64 data = kvm_read_edx_eax(vcpu);
1890 	int r;
1891 
1892 	r = kvm_set_msr(vcpu, ecx, data);
1893 
1894 	/* MSR write failed? See if we should ask user space */
1895 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1896 		/* Bounce to user space */
1897 		return 0;
1898 
1899 	/* Signal all other negative errors to userspace */
1900 	if (r < 0)
1901 		return r;
1902 
1903 	if (!r)
1904 		trace_kvm_msr_write(ecx, data);
1905 	else
1906 		trace_kvm_msr_write_ex(ecx, data);
1907 
1908 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1909 }
1910 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1911 
1912 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1913 {
1914 	return kvm_skip_emulated_instruction(vcpu);
1915 }
1916 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1917 
1918 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1919 {
1920 	/* Treat an INVD instruction as a NOP and just skip it. */
1921 	return kvm_emulate_as_nop(vcpu);
1922 }
1923 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1924 
1925 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1926 {
1927 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1928 	return kvm_emulate_as_nop(vcpu);
1929 }
1930 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1931 
1932 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1933 {
1934 	kvm_queue_exception(vcpu, UD_VECTOR);
1935 	return 1;
1936 }
1937 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1938 
1939 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1940 {
1941 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1942 	return kvm_emulate_as_nop(vcpu);
1943 }
1944 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1945 
1946 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1947 {
1948 	xfer_to_guest_mode_prepare();
1949 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1950 		xfer_to_guest_mode_work_pending();
1951 }
1952 
1953 /*
1954  * The fast path for frequent and performance sensitive wrmsr emulation,
1955  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1956  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1957  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1958  * other cases which must be called after interrupts are enabled on the host.
1959  */
1960 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1961 {
1962 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1963 		return 1;
1964 
1965 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1966 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1967 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1968 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1969 
1970 		data &= ~(1 << 12);
1971 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1972 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1973 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1974 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1975 		return 0;
1976 	}
1977 
1978 	return 1;
1979 }
1980 
1981 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1982 {
1983 	if (!kvm_can_use_hv_timer(vcpu))
1984 		return 1;
1985 
1986 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1987 	return 0;
1988 }
1989 
1990 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1991 {
1992 	u32 msr = kvm_rcx_read(vcpu);
1993 	u64 data;
1994 	fastpath_t ret = EXIT_FASTPATH_NONE;
1995 
1996 	switch (msr) {
1997 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1998 		data = kvm_read_edx_eax(vcpu);
1999 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2000 			kvm_skip_emulated_instruction(vcpu);
2001 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2002 		}
2003 		break;
2004 	case MSR_IA32_TSC_DEADLINE:
2005 		data = kvm_read_edx_eax(vcpu);
2006 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2007 			kvm_skip_emulated_instruction(vcpu);
2008 			ret = EXIT_FASTPATH_REENTER_GUEST;
2009 		}
2010 		break;
2011 	default:
2012 		break;
2013 	}
2014 
2015 	if (ret != EXIT_FASTPATH_NONE)
2016 		trace_kvm_msr_write(msr, data);
2017 
2018 	return ret;
2019 }
2020 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2021 
2022 /*
2023  * Adapt set_msr() to msr_io()'s calling convention
2024  */
2025 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2026 {
2027 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2028 }
2029 
2030 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2031 {
2032 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2033 }
2034 
2035 #ifdef CONFIG_X86_64
2036 struct pvclock_clock {
2037 	int vclock_mode;
2038 	u64 cycle_last;
2039 	u64 mask;
2040 	u32 mult;
2041 	u32 shift;
2042 	u64 base_cycles;
2043 	u64 offset;
2044 };
2045 
2046 struct pvclock_gtod_data {
2047 	seqcount_t	seq;
2048 
2049 	struct pvclock_clock clock; /* extract of a clocksource struct */
2050 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2051 
2052 	ktime_t		offs_boot;
2053 	u64		wall_time_sec;
2054 };
2055 
2056 static struct pvclock_gtod_data pvclock_gtod_data;
2057 
2058 static void update_pvclock_gtod(struct timekeeper *tk)
2059 {
2060 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2061 
2062 	write_seqcount_begin(&vdata->seq);
2063 
2064 	/* copy pvclock gtod data */
2065 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2066 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2067 	vdata->clock.mask		= tk->tkr_mono.mask;
2068 	vdata->clock.mult		= tk->tkr_mono.mult;
2069 	vdata->clock.shift		= tk->tkr_mono.shift;
2070 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2071 	vdata->clock.offset		= tk->tkr_mono.base;
2072 
2073 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2074 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2075 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2076 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2077 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2078 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2079 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2080 
2081 	vdata->wall_time_sec            = tk->xtime_sec;
2082 
2083 	vdata->offs_boot		= tk->offs_boot;
2084 
2085 	write_seqcount_end(&vdata->seq);
2086 }
2087 
2088 static s64 get_kvmclock_base_ns(void)
2089 {
2090 	/* Count up from boot time, but with the frequency of the raw clock.  */
2091 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2092 }
2093 #else
2094 static s64 get_kvmclock_base_ns(void)
2095 {
2096 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2097 	return ktime_get_boottime_ns();
2098 }
2099 #endif
2100 
2101 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2102 {
2103 	int version;
2104 	int r;
2105 	struct pvclock_wall_clock wc;
2106 	u32 wc_sec_hi;
2107 	u64 wall_nsec;
2108 
2109 	if (!wall_clock)
2110 		return;
2111 
2112 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2113 	if (r)
2114 		return;
2115 
2116 	if (version & 1)
2117 		++version;  /* first time write, random junk */
2118 
2119 	++version;
2120 
2121 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2122 		return;
2123 
2124 	/*
2125 	 * The guest calculates current wall clock time by adding
2126 	 * system time (updated by kvm_guest_time_update below) to the
2127 	 * wall clock specified here.  We do the reverse here.
2128 	 */
2129 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2130 
2131 	wc.nsec = do_div(wall_nsec, 1000000000);
2132 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2133 	wc.version = version;
2134 
2135 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2136 
2137 	if (sec_hi_ofs) {
2138 		wc_sec_hi = wall_nsec >> 32;
2139 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2140 				&wc_sec_hi, sizeof(wc_sec_hi));
2141 	}
2142 
2143 	version++;
2144 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2145 }
2146 
2147 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2148 				  bool old_msr, bool host_initiated)
2149 {
2150 	struct kvm_arch *ka = &vcpu->kvm->arch;
2151 
2152 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2153 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2154 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2155 
2156 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2157 	}
2158 
2159 	vcpu->arch.time = system_time;
2160 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2161 
2162 	/* we verify if the enable bit is set... */
2163 	vcpu->arch.pv_time_enabled = false;
2164 	if (!(system_time & 1))
2165 		return;
2166 
2167 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2168 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2169 				       sizeof(struct pvclock_vcpu_time_info)))
2170 		vcpu->arch.pv_time_enabled = true;
2171 
2172 	return;
2173 }
2174 
2175 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2176 {
2177 	do_shl32_div32(dividend, divisor);
2178 	return dividend;
2179 }
2180 
2181 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2182 			       s8 *pshift, u32 *pmultiplier)
2183 {
2184 	uint64_t scaled64;
2185 	int32_t  shift = 0;
2186 	uint64_t tps64;
2187 	uint32_t tps32;
2188 
2189 	tps64 = base_hz;
2190 	scaled64 = scaled_hz;
2191 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2192 		tps64 >>= 1;
2193 		shift--;
2194 	}
2195 
2196 	tps32 = (uint32_t)tps64;
2197 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2198 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2199 			scaled64 >>= 1;
2200 		else
2201 			tps32 <<= 1;
2202 		shift++;
2203 	}
2204 
2205 	*pshift = shift;
2206 	*pmultiplier = div_frac(scaled64, tps32);
2207 }
2208 
2209 #ifdef CONFIG_X86_64
2210 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2211 #endif
2212 
2213 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2214 static unsigned long max_tsc_khz;
2215 
2216 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2217 {
2218 	u64 v = (u64)khz * (1000000 + ppm);
2219 	do_div(v, 1000000);
2220 	return v;
2221 }
2222 
2223 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2224 
2225 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2226 {
2227 	u64 ratio;
2228 
2229 	/* Guest TSC same frequency as host TSC? */
2230 	if (!scale) {
2231 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2232 		return 0;
2233 	}
2234 
2235 	/* TSC scaling supported? */
2236 	if (!kvm_has_tsc_control) {
2237 		if (user_tsc_khz > tsc_khz) {
2238 			vcpu->arch.tsc_catchup = 1;
2239 			vcpu->arch.tsc_always_catchup = 1;
2240 			return 0;
2241 		} else {
2242 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2243 			return -1;
2244 		}
2245 	}
2246 
2247 	/* TSC scaling required  - calculate ratio */
2248 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2249 				user_tsc_khz, tsc_khz);
2250 
2251 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2252 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2253 			            user_tsc_khz);
2254 		return -1;
2255 	}
2256 
2257 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2258 	return 0;
2259 }
2260 
2261 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2262 {
2263 	u32 thresh_lo, thresh_hi;
2264 	int use_scaling = 0;
2265 
2266 	/* tsc_khz can be zero if TSC calibration fails */
2267 	if (user_tsc_khz == 0) {
2268 		/* set tsc_scaling_ratio to a safe value */
2269 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2270 		return -1;
2271 	}
2272 
2273 	/* Compute a scale to convert nanoseconds in TSC cycles */
2274 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2275 			   &vcpu->arch.virtual_tsc_shift,
2276 			   &vcpu->arch.virtual_tsc_mult);
2277 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2278 
2279 	/*
2280 	 * Compute the variation in TSC rate which is acceptable
2281 	 * within the range of tolerance and decide if the
2282 	 * rate being applied is within that bounds of the hardware
2283 	 * rate.  If so, no scaling or compensation need be done.
2284 	 */
2285 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2286 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2287 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2288 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2289 		use_scaling = 1;
2290 	}
2291 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2292 }
2293 
2294 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2295 {
2296 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2297 				      vcpu->arch.virtual_tsc_mult,
2298 				      vcpu->arch.virtual_tsc_shift);
2299 	tsc += vcpu->arch.this_tsc_write;
2300 	return tsc;
2301 }
2302 
2303 static inline int gtod_is_based_on_tsc(int mode)
2304 {
2305 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2306 }
2307 
2308 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2309 {
2310 #ifdef CONFIG_X86_64
2311 	bool vcpus_matched;
2312 	struct kvm_arch *ka = &vcpu->kvm->arch;
2313 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2314 
2315 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2316 			 atomic_read(&vcpu->kvm->online_vcpus));
2317 
2318 	/*
2319 	 * Once the masterclock is enabled, always perform request in
2320 	 * order to update it.
2321 	 *
2322 	 * In order to enable masterclock, the host clocksource must be TSC
2323 	 * and the vcpus need to have matched TSCs.  When that happens,
2324 	 * perform request to enable masterclock.
2325 	 */
2326 	if (ka->use_master_clock ||
2327 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2328 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2329 
2330 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2331 			    atomic_read(&vcpu->kvm->online_vcpus),
2332 		            ka->use_master_clock, gtod->clock.vclock_mode);
2333 #endif
2334 }
2335 
2336 /*
2337  * Multiply tsc by a fixed point number represented by ratio.
2338  *
2339  * The most significant 64-N bits (mult) of ratio represent the
2340  * integral part of the fixed point number; the remaining N bits
2341  * (frac) represent the fractional part, ie. ratio represents a fixed
2342  * point number (mult + frac * 2^(-N)).
2343  *
2344  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2345  */
2346 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2347 {
2348 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2349 }
2350 
2351 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2352 {
2353 	u64 _tsc = tsc;
2354 
2355 	if (ratio != kvm_default_tsc_scaling_ratio)
2356 		_tsc = __scale_tsc(ratio, tsc);
2357 
2358 	return _tsc;
2359 }
2360 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2361 
2362 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2363 {
2364 	u64 tsc;
2365 
2366 	tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2367 
2368 	return target_tsc - tsc;
2369 }
2370 
2371 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2372 {
2373 	return vcpu->arch.l1_tsc_offset +
2374 		kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2375 }
2376 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2377 
2378 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2379 {
2380 	u64 nested_offset;
2381 
2382 	if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2383 		nested_offset = l1_offset;
2384 	else
2385 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2386 						kvm_tsc_scaling_ratio_frac_bits);
2387 
2388 	nested_offset += l2_offset;
2389 	return nested_offset;
2390 }
2391 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2392 
2393 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2394 {
2395 	if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2396 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2397 				       kvm_tsc_scaling_ratio_frac_bits);
2398 
2399 	return l1_multiplier;
2400 }
2401 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2402 
2403 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2404 {
2405 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2406 				   vcpu->arch.l1_tsc_offset,
2407 				   l1_offset);
2408 
2409 	vcpu->arch.l1_tsc_offset = l1_offset;
2410 
2411 	/*
2412 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2413 	 * according to the spec this should set L1's TSC (as opposed to
2414 	 * setting L1's offset for L2).
2415 	 */
2416 	if (is_guest_mode(vcpu))
2417 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2418 			l1_offset,
2419 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2420 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2421 	else
2422 		vcpu->arch.tsc_offset = l1_offset;
2423 
2424 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2425 }
2426 
2427 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2428 {
2429 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2430 
2431 	/* Userspace is changing the multiplier while L2 is active */
2432 	if (is_guest_mode(vcpu))
2433 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2434 			l1_multiplier,
2435 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2436 	else
2437 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2438 
2439 	if (kvm_has_tsc_control)
2440 		static_call(kvm_x86_write_tsc_multiplier)(
2441 			vcpu, vcpu->arch.tsc_scaling_ratio);
2442 }
2443 
2444 static inline bool kvm_check_tsc_unstable(void)
2445 {
2446 #ifdef CONFIG_X86_64
2447 	/*
2448 	 * TSC is marked unstable when we're running on Hyper-V,
2449 	 * 'TSC page' clocksource is good.
2450 	 */
2451 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2452 		return false;
2453 #endif
2454 	return check_tsc_unstable();
2455 }
2456 
2457 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2458 {
2459 	struct kvm *kvm = vcpu->kvm;
2460 	u64 offset, ns, elapsed;
2461 	unsigned long flags;
2462 	bool matched;
2463 	bool already_matched;
2464 	bool synchronizing = false;
2465 
2466 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2467 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2468 	ns = get_kvmclock_base_ns();
2469 	elapsed = ns - kvm->arch.last_tsc_nsec;
2470 
2471 	if (vcpu->arch.virtual_tsc_khz) {
2472 		if (data == 0) {
2473 			/*
2474 			 * detection of vcpu initialization -- need to sync
2475 			 * with other vCPUs. This particularly helps to keep
2476 			 * kvm_clock stable after CPU hotplug
2477 			 */
2478 			synchronizing = true;
2479 		} else {
2480 			u64 tsc_exp = kvm->arch.last_tsc_write +
2481 						nsec_to_cycles(vcpu, elapsed);
2482 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2483 			/*
2484 			 * Special case: TSC write with a small delta (1 second)
2485 			 * of virtual cycle time against real time is
2486 			 * interpreted as an attempt to synchronize the CPU.
2487 			 */
2488 			synchronizing = data < tsc_exp + tsc_hz &&
2489 					data + tsc_hz > tsc_exp;
2490 		}
2491 	}
2492 
2493 	/*
2494 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2495 	 * TSC, we add elapsed time in this computation.  We could let the
2496 	 * compensation code attempt to catch up if we fall behind, but
2497 	 * it's better to try to match offsets from the beginning.
2498          */
2499 	if (synchronizing &&
2500 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2501 		if (!kvm_check_tsc_unstable()) {
2502 			offset = kvm->arch.cur_tsc_offset;
2503 		} else {
2504 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2505 			data += delta;
2506 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2507 		}
2508 		matched = true;
2509 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2510 	} else {
2511 		/*
2512 		 * We split periods of matched TSC writes into generations.
2513 		 * For each generation, we track the original measured
2514 		 * nanosecond time, offset, and write, so if TSCs are in
2515 		 * sync, we can match exact offset, and if not, we can match
2516 		 * exact software computation in compute_guest_tsc()
2517 		 *
2518 		 * These values are tracked in kvm->arch.cur_xxx variables.
2519 		 */
2520 		kvm->arch.cur_tsc_generation++;
2521 		kvm->arch.cur_tsc_nsec = ns;
2522 		kvm->arch.cur_tsc_write = data;
2523 		kvm->arch.cur_tsc_offset = offset;
2524 		matched = false;
2525 	}
2526 
2527 	/*
2528 	 * We also track th most recent recorded KHZ, write and time to
2529 	 * allow the matching interval to be extended at each write.
2530 	 */
2531 	kvm->arch.last_tsc_nsec = ns;
2532 	kvm->arch.last_tsc_write = data;
2533 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2534 
2535 	vcpu->arch.last_guest_tsc = data;
2536 
2537 	/* Keep track of which generation this VCPU has synchronized to */
2538 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2539 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2540 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2541 
2542 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2543 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2544 
2545 	spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2546 	if (!matched) {
2547 		kvm->arch.nr_vcpus_matched_tsc = 0;
2548 	} else if (!already_matched) {
2549 		kvm->arch.nr_vcpus_matched_tsc++;
2550 	}
2551 
2552 	kvm_track_tsc_matching(vcpu);
2553 	spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2554 }
2555 
2556 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2557 					   s64 adjustment)
2558 {
2559 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2560 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2561 }
2562 
2563 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2564 {
2565 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2566 		WARN_ON(adjustment < 0);
2567 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2568 				   vcpu->arch.l1_tsc_scaling_ratio);
2569 	adjust_tsc_offset_guest(vcpu, adjustment);
2570 }
2571 
2572 #ifdef CONFIG_X86_64
2573 
2574 static u64 read_tsc(void)
2575 {
2576 	u64 ret = (u64)rdtsc_ordered();
2577 	u64 last = pvclock_gtod_data.clock.cycle_last;
2578 
2579 	if (likely(ret >= last))
2580 		return ret;
2581 
2582 	/*
2583 	 * GCC likes to generate cmov here, but this branch is extremely
2584 	 * predictable (it's just a function of time and the likely is
2585 	 * very likely) and there's a data dependence, so force GCC
2586 	 * to generate a branch instead.  I don't barrier() because
2587 	 * we don't actually need a barrier, and if this function
2588 	 * ever gets inlined it will generate worse code.
2589 	 */
2590 	asm volatile ("");
2591 	return last;
2592 }
2593 
2594 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2595 			  int *mode)
2596 {
2597 	long v;
2598 	u64 tsc_pg_val;
2599 
2600 	switch (clock->vclock_mode) {
2601 	case VDSO_CLOCKMODE_HVCLOCK:
2602 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2603 						  tsc_timestamp);
2604 		if (tsc_pg_val != U64_MAX) {
2605 			/* TSC page valid */
2606 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2607 			v = (tsc_pg_val - clock->cycle_last) &
2608 				clock->mask;
2609 		} else {
2610 			/* TSC page invalid */
2611 			*mode = VDSO_CLOCKMODE_NONE;
2612 		}
2613 		break;
2614 	case VDSO_CLOCKMODE_TSC:
2615 		*mode = VDSO_CLOCKMODE_TSC;
2616 		*tsc_timestamp = read_tsc();
2617 		v = (*tsc_timestamp - clock->cycle_last) &
2618 			clock->mask;
2619 		break;
2620 	default:
2621 		*mode = VDSO_CLOCKMODE_NONE;
2622 	}
2623 
2624 	if (*mode == VDSO_CLOCKMODE_NONE)
2625 		*tsc_timestamp = v = 0;
2626 
2627 	return v * clock->mult;
2628 }
2629 
2630 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2631 {
2632 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2633 	unsigned long seq;
2634 	int mode;
2635 	u64 ns;
2636 
2637 	do {
2638 		seq = read_seqcount_begin(&gtod->seq);
2639 		ns = gtod->raw_clock.base_cycles;
2640 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2641 		ns >>= gtod->raw_clock.shift;
2642 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2643 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2644 	*t = ns;
2645 
2646 	return mode;
2647 }
2648 
2649 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2650 {
2651 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2652 	unsigned long seq;
2653 	int mode;
2654 	u64 ns;
2655 
2656 	do {
2657 		seq = read_seqcount_begin(&gtod->seq);
2658 		ts->tv_sec = gtod->wall_time_sec;
2659 		ns = gtod->clock.base_cycles;
2660 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2661 		ns >>= gtod->clock.shift;
2662 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2663 
2664 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2665 	ts->tv_nsec = ns;
2666 
2667 	return mode;
2668 }
2669 
2670 /* returns true if host is using TSC based clocksource */
2671 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2672 {
2673 	/* checked again under seqlock below */
2674 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2675 		return false;
2676 
2677 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2678 						      tsc_timestamp));
2679 }
2680 
2681 /* returns true if host is using TSC based clocksource */
2682 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2683 					   u64 *tsc_timestamp)
2684 {
2685 	/* checked again under seqlock below */
2686 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2687 		return false;
2688 
2689 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2690 }
2691 #endif
2692 
2693 /*
2694  *
2695  * Assuming a stable TSC across physical CPUS, and a stable TSC
2696  * across virtual CPUs, the following condition is possible.
2697  * Each numbered line represents an event visible to both
2698  * CPUs at the next numbered event.
2699  *
2700  * "timespecX" represents host monotonic time. "tscX" represents
2701  * RDTSC value.
2702  *
2703  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2704  *
2705  * 1.  read timespec0,tsc0
2706  * 2.					| timespec1 = timespec0 + N
2707  * 					| tsc1 = tsc0 + M
2708  * 3. transition to guest		| transition to guest
2709  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2710  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2711  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2712  *
2713  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2714  *
2715  * 	- ret0 < ret1
2716  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2717  *		...
2718  *	- 0 < N - M => M < N
2719  *
2720  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2721  * always the case (the difference between two distinct xtime instances
2722  * might be smaller then the difference between corresponding TSC reads,
2723  * when updating guest vcpus pvclock areas).
2724  *
2725  * To avoid that problem, do not allow visibility of distinct
2726  * system_timestamp/tsc_timestamp values simultaneously: use a master
2727  * copy of host monotonic time values. Update that master copy
2728  * in lockstep.
2729  *
2730  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2731  *
2732  */
2733 
2734 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2735 {
2736 #ifdef CONFIG_X86_64
2737 	struct kvm_arch *ka = &kvm->arch;
2738 	int vclock_mode;
2739 	bool host_tsc_clocksource, vcpus_matched;
2740 
2741 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2742 			atomic_read(&kvm->online_vcpus));
2743 
2744 	/*
2745 	 * If the host uses TSC clock, then passthrough TSC as stable
2746 	 * to the guest.
2747 	 */
2748 	host_tsc_clocksource = kvm_get_time_and_clockread(
2749 					&ka->master_kernel_ns,
2750 					&ka->master_cycle_now);
2751 
2752 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2753 				&& !ka->backwards_tsc_observed
2754 				&& !ka->boot_vcpu_runs_old_kvmclock;
2755 
2756 	if (ka->use_master_clock)
2757 		atomic_set(&kvm_guest_has_master_clock, 1);
2758 
2759 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2760 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2761 					vcpus_matched);
2762 #endif
2763 }
2764 
2765 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2766 {
2767 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2768 }
2769 
2770 static void kvm_gen_update_masterclock(struct kvm *kvm)
2771 {
2772 #ifdef CONFIG_X86_64
2773 	int i;
2774 	struct kvm_vcpu *vcpu;
2775 	struct kvm_arch *ka = &kvm->arch;
2776 	unsigned long flags;
2777 
2778 	kvm_hv_invalidate_tsc_page(kvm);
2779 
2780 	kvm_make_mclock_inprogress_request(kvm);
2781 
2782 	/* no guest entries from this point */
2783 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2784 	pvclock_update_vm_gtod_copy(kvm);
2785 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2786 
2787 	kvm_for_each_vcpu(i, vcpu, kvm)
2788 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2789 
2790 	/* guest entries allowed */
2791 	kvm_for_each_vcpu(i, vcpu, kvm)
2792 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2793 #endif
2794 }
2795 
2796 u64 get_kvmclock_ns(struct kvm *kvm)
2797 {
2798 	struct kvm_arch *ka = &kvm->arch;
2799 	struct pvclock_vcpu_time_info hv_clock;
2800 	unsigned long flags;
2801 	u64 ret;
2802 
2803 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2804 	if (!ka->use_master_clock) {
2805 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2806 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2807 	}
2808 
2809 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2810 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2811 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2812 
2813 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2814 	get_cpu();
2815 
2816 	if (__this_cpu_read(cpu_tsc_khz)) {
2817 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2818 				   &hv_clock.tsc_shift,
2819 				   &hv_clock.tsc_to_system_mul);
2820 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2821 	} else
2822 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2823 
2824 	put_cpu();
2825 
2826 	return ret;
2827 }
2828 
2829 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2830 				   struct gfn_to_hva_cache *cache,
2831 				   unsigned int offset)
2832 {
2833 	struct kvm_vcpu_arch *vcpu = &v->arch;
2834 	struct pvclock_vcpu_time_info guest_hv_clock;
2835 
2836 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2837 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2838 		return;
2839 
2840 	/* This VCPU is paused, but it's legal for a guest to read another
2841 	 * VCPU's kvmclock, so we really have to follow the specification where
2842 	 * it says that version is odd if data is being modified, and even after
2843 	 * it is consistent.
2844 	 *
2845 	 * Version field updates must be kept separate.  This is because
2846 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2847 	 * writes within a string instruction are weakly ordered.  So there
2848 	 * are three writes overall.
2849 	 *
2850 	 * As a small optimization, only write the version field in the first
2851 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2852 	 * version field is the first in the struct.
2853 	 */
2854 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2855 
2856 	if (guest_hv_clock.version & 1)
2857 		++guest_hv_clock.version;  /* first time write, random junk */
2858 
2859 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2860 	kvm_write_guest_offset_cached(v->kvm, cache,
2861 				      &vcpu->hv_clock, offset,
2862 				      sizeof(vcpu->hv_clock.version));
2863 
2864 	smp_wmb();
2865 
2866 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2867 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2868 
2869 	if (vcpu->pvclock_set_guest_stopped_request) {
2870 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2871 		vcpu->pvclock_set_guest_stopped_request = false;
2872 	}
2873 
2874 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2875 
2876 	kvm_write_guest_offset_cached(v->kvm, cache,
2877 				      &vcpu->hv_clock, offset,
2878 				      sizeof(vcpu->hv_clock));
2879 
2880 	smp_wmb();
2881 
2882 	vcpu->hv_clock.version++;
2883 	kvm_write_guest_offset_cached(v->kvm, cache,
2884 				     &vcpu->hv_clock, offset,
2885 				     sizeof(vcpu->hv_clock.version));
2886 }
2887 
2888 static int kvm_guest_time_update(struct kvm_vcpu *v)
2889 {
2890 	unsigned long flags, tgt_tsc_khz;
2891 	struct kvm_vcpu_arch *vcpu = &v->arch;
2892 	struct kvm_arch *ka = &v->kvm->arch;
2893 	s64 kernel_ns;
2894 	u64 tsc_timestamp, host_tsc;
2895 	u8 pvclock_flags;
2896 	bool use_master_clock;
2897 
2898 	kernel_ns = 0;
2899 	host_tsc = 0;
2900 
2901 	/*
2902 	 * If the host uses TSC clock, then passthrough TSC as stable
2903 	 * to the guest.
2904 	 */
2905 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2906 	use_master_clock = ka->use_master_clock;
2907 	if (use_master_clock) {
2908 		host_tsc = ka->master_cycle_now;
2909 		kernel_ns = ka->master_kernel_ns;
2910 	}
2911 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2912 
2913 	/* Keep irq disabled to prevent changes to the clock */
2914 	local_irq_save(flags);
2915 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2916 	if (unlikely(tgt_tsc_khz == 0)) {
2917 		local_irq_restore(flags);
2918 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2919 		return 1;
2920 	}
2921 	if (!use_master_clock) {
2922 		host_tsc = rdtsc();
2923 		kernel_ns = get_kvmclock_base_ns();
2924 	}
2925 
2926 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2927 
2928 	/*
2929 	 * We may have to catch up the TSC to match elapsed wall clock
2930 	 * time for two reasons, even if kvmclock is used.
2931 	 *   1) CPU could have been running below the maximum TSC rate
2932 	 *   2) Broken TSC compensation resets the base at each VCPU
2933 	 *      entry to avoid unknown leaps of TSC even when running
2934 	 *      again on the same CPU.  This may cause apparent elapsed
2935 	 *      time to disappear, and the guest to stand still or run
2936 	 *	very slowly.
2937 	 */
2938 	if (vcpu->tsc_catchup) {
2939 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2940 		if (tsc > tsc_timestamp) {
2941 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2942 			tsc_timestamp = tsc;
2943 		}
2944 	}
2945 
2946 	local_irq_restore(flags);
2947 
2948 	/* With all the info we got, fill in the values */
2949 
2950 	if (kvm_has_tsc_control)
2951 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2952 					    v->arch.l1_tsc_scaling_ratio);
2953 
2954 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2955 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2956 				   &vcpu->hv_clock.tsc_shift,
2957 				   &vcpu->hv_clock.tsc_to_system_mul);
2958 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2959 	}
2960 
2961 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2962 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2963 	vcpu->last_guest_tsc = tsc_timestamp;
2964 
2965 	/* If the host uses TSC clocksource, then it is stable */
2966 	pvclock_flags = 0;
2967 	if (use_master_clock)
2968 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2969 
2970 	vcpu->hv_clock.flags = pvclock_flags;
2971 
2972 	if (vcpu->pv_time_enabled)
2973 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2974 	if (vcpu->xen.vcpu_info_set)
2975 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2976 				       offsetof(struct compat_vcpu_info, time));
2977 	if (vcpu->xen.vcpu_time_info_set)
2978 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2979 	if (!v->vcpu_idx)
2980 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2981 	return 0;
2982 }
2983 
2984 /*
2985  * kvmclock updates which are isolated to a given vcpu, such as
2986  * vcpu->cpu migration, should not allow system_timestamp from
2987  * the rest of the vcpus to remain static. Otherwise ntp frequency
2988  * correction applies to one vcpu's system_timestamp but not
2989  * the others.
2990  *
2991  * So in those cases, request a kvmclock update for all vcpus.
2992  * We need to rate-limit these requests though, as they can
2993  * considerably slow guests that have a large number of vcpus.
2994  * The time for a remote vcpu to update its kvmclock is bound
2995  * by the delay we use to rate-limit the updates.
2996  */
2997 
2998 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2999 
3000 static void kvmclock_update_fn(struct work_struct *work)
3001 {
3002 	int i;
3003 	struct delayed_work *dwork = to_delayed_work(work);
3004 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3005 					   kvmclock_update_work);
3006 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3007 	struct kvm_vcpu *vcpu;
3008 
3009 	kvm_for_each_vcpu(i, vcpu, kvm) {
3010 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3011 		kvm_vcpu_kick(vcpu);
3012 	}
3013 }
3014 
3015 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3016 {
3017 	struct kvm *kvm = v->kvm;
3018 
3019 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3020 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3021 					KVMCLOCK_UPDATE_DELAY);
3022 }
3023 
3024 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3025 
3026 static void kvmclock_sync_fn(struct work_struct *work)
3027 {
3028 	struct delayed_work *dwork = to_delayed_work(work);
3029 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3030 					   kvmclock_sync_work);
3031 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3032 
3033 	if (!kvmclock_periodic_sync)
3034 		return;
3035 
3036 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3037 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3038 					KVMCLOCK_SYNC_PERIOD);
3039 }
3040 
3041 /*
3042  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3043  */
3044 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3045 {
3046 	/* McStatusWrEn enabled? */
3047 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3048 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3049 
3050 	return false;
3051 }
3052 
3053 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3054 {
3055 	u64 mcg_cap = vcpu->arch.mcg_cap;
3056 	unsigned bank_num = mcg_cap & 0xff;
3057 	u32 msr = msr_info->index;
3058 	u64 data = msr_info->data;
3059 
3060 	switch (msr) {
3061 	case MSR_IA32_MCG_STATUS:
3062 		vcpu->arch.mcg_status = data;
3063 		break;
3064 	case MSR_IA32_MCG_CTL:
3065 		if (!(mcg_cap & MCG_CTL_P) &&
3066 		    (data || !msr_info->host_initiated))
3067 			return 1;
3068 		if (data != 0 && data != ~(u64)0)
3069 			return 1;
3070 		vcpu->arch.mcg_ctl = data;
3071 		break;
3072 	default:
3073 		if (msr >= MSR_IA32_MC0_CTL &&
3074 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3075 			u32 offset = array_index_nospec(
3076 				msr - MSR_IA32_MC0_CTL,
3077 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3078 
3079 			/* only 0 or all 1s can be written to IA32_MCi_CTL
3080 			 * some Linux kernels though clear bit 10 in bank 4 to
3081 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3082 			 * this to avoid an uncatched #GP in the guest
3083 			 */
3084 			if ((offset & 0x3) == 0 &&
3085 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
3086 				return -1;
3087 
3088 			/* MCi_STATUS */
3089 			if (!msr_info->host_initiated &&
3090 			    (offset & 0x3) == 1 && data != 0) {
3091 				if (!can_set_mci_status(vcpu))
3092 					return -1;
3093 			}
3094 
3095 			vcpu->arch.mce_banks[offset] = data;
3096 			break;
3097 		}
3098 		return 1;
3099 	}
3100 	return 0;
3101 }
3102 
3103 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3104 {
3105 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3106 
3107 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3108 }
3109 
3110 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3111 {
3112 	gpa_t gpa = data & ~0x3f;
3113 
3114 	/* Bits 4:5 are reserved, Should be zero */
3115 	if (data & 0x30)
3116 		return 1;
3117 
3118 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3119 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3120 		return 1;
3121 
3122 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3123 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3124 		return 1;
3125 
3126 	if (!lapic_in_kernel(vcpu))
3127 		return data ? 1 : 0;
3128 
3129 	vcpu->arch.apf.msr_en_val = data;
3130 
3131 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3132 		kvm_clear_async_pf_completion_queue(vcpu);
3133 		kvm_async_pf_hash_reset(vcpu);
3134 		return 0;
3135 	}
3136 
3137 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3138 					sizeof(u64)))
3139 		return 1;
3140 
3141 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3142 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3143 
3144 	kvm_async_pf_wakeup_all(vcpu);
3145 
3146 	return 0;
3147 }
3148 
3149 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3150 {
3151 	/* Bits 8-63 are reserved */
3152 	if (data >> 8)
3153 		return 1;
3154 
3155 	if (!lapic_in_kernel(vcpu))
3156 		return 1;
3157 
3158 	vcpu->arch.apf.msr_int_val = data;
3159 
3160 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3161 
3162 	return 0;
3163 }
3164 
3165 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3166 {
3167 	vcpu->arch.pv_time_enabled = false;
3168 	vcpu->arch.time = 0;
3169 }
3170 
3171 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3172 {
3173 	++vcpu->stat.tlb_flush;
3174 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3175 }
3176 
3177 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3178 {
3179 	++vcpu->stat.tlb_flush;
3180 
3181 	if (!tdp_enabled) {
3182                /*
3183 		 * A TLB flush on behalf of the guest is equivalent to
3184 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3185 		 * a forced sync of the shadow page tables.  Unload the
3186 		 * entire MMU here and the subsequent load will sync the
3187 		 * shadow page tables, and also flush the TLB.
3188 		 */
3189 		kvm_mmu_unload(vcpu);
3190 		return;
3191 	}
3192 
3193 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3194 }
3195 
3196 static void record_steal_time(struct kvm_vcpu *vcpu)
3197 {
3198 	struct kvm_host_map map;
3199 	struct kvm_steal_time *st;
3200 
3201 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3202 		kvm_xen_runstate_set_running(vcpu);
3203 		return;
3204 	}
3205 
3206 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3207 		return;
3208 
3209 	/* -EAGAIN is returned in atomic context so we can just return. */
3210 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3211 			&map, &vcpu->arch.st.cache, false))
3212 		return;
3213 
3214 	st = map.hva +
3215 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3216 
3217 	/*
3218 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3219 	 * expensive IPIs.
3220 	 */
3221 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3222 		u8 st_preempted = xchg(&st->preempted, 0);
3223 
3224 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3225 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3226 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3227 			kvm_vcpu_flush_tlb_guest(vcpu);
3228 	} else {
3229 		st->preempted = 0;
3230 	}
3231 
3232 	vcpu->arch.st.preempted = 0;
3233 
3234 	if (st->version & 1)
3235 		st->version += 1;  /* first time write, random junk */
3236 
3237 	st->version += 1;
3238 
3239 	smp_wmb();
3240 
3241 	st->steal += current->sched_info.run_delay -
3242 		vcpu->arch.st.last_steal;
3243 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3244 
3245 	smp_wmb();
3246 
3247 	st->version += 1;
3248 
3249 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3250 }
3251 
3252 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3253 {
3254 	bool pr = false;
3255 	u32 msr = msr_info->index;
3256 	u64 data = msr_info->data;
3257 
3258 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3259 		return kvm_xen_write_hypercall_page(vcpu, data);
3260 
3261 	switch (msr) {
3262 	case MSR_AMD64_NB_CFG:
3263 	case MSR_IA32_UCODE_WRITE:
3264 	case MSR_VM_HSAVE_PA:
3265 	case MSR_AMD64_PATCH_LOADER:
3266 	case MSR_AMD64_BU_CFG2:
3267 	case MSR_AMD64_DC_CFG:
3268 	case MSR_F15H_EX_CFG:
3269 		break;
3270 
3271 	case MSR_IA32_UCODE_REV:
3272 		if (msr_info->host_initiated)
3273 			vcpu->arch.microcode_version = data;
3274 		break;
3275 	case MSR_IA32_ARCH_CAPABILITIES:
3276 		if (!msr_info->host_initiated)
3277 			return 1;
3278 		vcpu->arch.arch_capabilities = data;
3279 		break;
3280 	case MSR_IA32_PERF_CAPABILITIES: {
3281 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3282 
3283 		if (!msr_info->host_initiated)
3284 			return 1;
3285 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3286 			return 1;
3287 		if (data & ~msr_ent.data)
3288 			return 1;
3289 
3290 		vcpu->arch.perf_capabilities = data;
3291 
3292 		return 0;
3293 		}
3294 	case MSR_EFER:
3295 		return set_efer(vcpu, msr_info);
3296 	case MSR_K7_HWCR:
3297 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3298 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3299 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3300 
3301 		/* Handle McStatusWrEn */
3302 		if (data == BIT_ULL(18)) {
3303 			vcpu->arch.msr_hwcr = data;
3304 		} else if (data != 0) {
3305 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3306 				    data);
3307 			return 1;
3308 		}
3309 		break;
3310 	case MSR_FAM10H_MMIO_CONF_BASE:
3311 		if (data != 0) {
3312 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3313 				    "0x%llx\n", data);
3314 			return 1;
3315 		}
3316 		break;
3317 	case 0x200 ... 0x2ff:
3318 		return kvm_mtrr_set_msr(vcpu, msr, data);
3319 	case MSR_IA32_APICBASE:
3320 		return kvm_set_apic_base(vcpu, msr_info);
3321 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3322 		return kvm_x2apic_msr_write(vcpu, msr, data);
3323 	case MSR_IA32_TSC_DEADLINE:
3324 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3325 		break;
3326 	case MSR_IA32_TSC_ADJUST:
3327 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3328 			if (!msr_info->host_initiated) {
3329 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3330 				adjust_tsc_offset_guest(vcpu, adj);
3331 				/* Before back to guest, tsc_timestamp must be adjusted
3332 				 * as well, otherwise guest's percpu pvclock time could jump.
3333 				 */
3334 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3335 			}
3336 			vcpu->arch.ia32_tsc_adjust_msr = data;
3337 		}
3338 		break;
3339 	case MSR_IA32_MISC_ENABLE:
3340 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3341 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3342 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3343 				return 1;
3344 			vcpu->arch.ia32_misc_enable_msr = data;
3345 			kvm_update_cpuid_runtime(vcpu);
3346 		} else {
3347 			vcpu->arch.ia32_misc_enable_msr = data;
3348 		}
3349 		break;
3350 	case MSR_IA32_SMBASE:
3351 		if (!msr_info->host_initiated)
3352 			return 1;
3353 		vcpu->arch.smbase = data;
3354 		break;
3355 	case MSR_IA32_POWER_CTL:
3356 		vcpu->arch.msr_ia32_power_ctl = data;
3357 		break;
3358 	case MSR_IA32_TSC:
3359 		if (msr_info->host_initiated) {
3360 			kvm_synchronize_tsc(vcpu, data);
3361 		} else {
3362 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3363 			adjust_tsc_offset_guest(vcpu, adj);
3364 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3365 		}
3366 		break;
3367 	case MSR_IA32_XSS:
3368 		if (!msr_info->host_initiated &&
3369 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3370 			return 1;
3371 		/*
3372 		 * KVM supports exposing PT to the guest, but does not support
3373 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3374 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3375 		 */
3376 		if (data & ~supported_xss)
3377 			return 1;
3378 		vcpu->arch.ia32_xss = data;
3379 		break;
3380 	case MSR_SMI_COUNT:
3381 		if (!msr_info->host_initiated)
3382 			return 1;
3383 		vcpu->arch.smi_count = data;
3384 		break;
3385 	case MSR_KVM_WALL_CLOCK_NEW:
3386 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3387 			return 1;
3388 
3389 		vcpu->kvm->arch.wall_clock = data;
3390 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3391 		break;
3392 	case MSR_KVM_WALL_CLOCK:
3393 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3394 			return 1;
3395 
3396 		vcpu->kvm->arch.wall_clock = data;
3397 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3398 		break;
3399 	case MSR_KVM_SYSTEM_TIME_NEW:
3400 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3401 			return 1;
3402 
3403 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3404 		break;
3405 	case MSR_KVM_SYSTEM_TIME:
3406 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3407 			return 1;
3408 
3409 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3410 		break;
3411 	case MSR_KVM_ASYNC_PF_EN:
3412 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3413 			return 1;
3414 
3415 		if (kvm_pv_enable_async_pf(vcpu, data))
3416 			return 1;
3417 		break;
3418 	case MSR_KVM_ASYNC_PF_INT:
3419 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3420 			return 1;
3421 
3422 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3423 			return 1;
3424 		break;
3425 	case MSR_KVM_ASYNC_PF_ACK:
3426 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3427 			return 1;
3428 		if (data & 0x1) {
3429 			vcpu->arch.apf.pageready_pending = false;
3430 			kvm_check_async_pf_completion(vcpu);
3431 		}
3432 		break;
3433 	case MSR_KVM_STEAL_TIME:
3434 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3435 			return 1;
3436 
3437 		if (unlikely(!sched_info_on()))
3438 			return 1;
3439 
3440 		if (data & KVM_STEAL_RESERVED_MASK)
3441 			return 1;
3442 
3443 		vcpu->arch.st.msr_val = data;
3444 
3445 		if (!(data & KVM_MSR_ENABLED))
3446 			break;
3447 
3448 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3449 
3450 		break;
3451 	case MSR_KVM_PV_EOI_EN:
3452 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3453 			return 1;
3454 
3455 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3456 			return 1;
3457 		break;
3458 
3459 	case MSR_KVM_POLL_CONTROL:
3460 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3461 			return 1;
3462 
3463 		/* only enable bit supported */
3464 		if (data & (-1ULL << 1))
3465 			return 1;
3466 
3467 		vcpu->arch.msr_kvm_poll_control = data;
3468 		break;
3469 
3470 	case MSR_IA32_MCG_CTL:
3471 	case MSR_IA32_MCG_STATUS:
3472 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3473 		return set_msr_mce(vcpu, msr_info);
3474 
3475 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3476 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3477 		pr = true;
3478 		fallthrough;
3479 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3480 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3481 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3482 			return kvm_pmu_set_msr(vcpu, msr_info);
3483 
3484 		if (pr || data != 0)
3485 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3486 				    "0x%x data 0x%llx\n", msr, data);
3487 		break;
3488 	case MSR_K7_CLK_CTL:
3489 		/*
3490 		 * Ignore all writes to this no longer documented MSR.
3491 		 * Writes are only relevant for old K7 processors,
3492 		 * all pre-dating SVM, but a recommended workaround from
3493 		 * AMD for these chips. It is possible to specify the
3494 		 * affected processor models on the command line, hence
3495 		 * the need to ignore the workaround.
3496 		 */
3497 		break;
3498 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3499 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3500 	case HV_X64_MSR_SYNDBG_OPTIONS:
3501 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3502 	case HV_X64_MSR_CRASH_CTL:
3503 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3504 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3505 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3506 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3507 		return kvm_hv_set_msr_common(vcpu, msr, data,
3508 					     msr_info->host_initiated);
3509 	case MSR_IA32_BBL_CR_CTL3:
3510 		/* Drop writes to this legacy MSR -- see rdmsr
3511 		 * counterpart for further detail.
3512 		 */
3513 		if (report_ignored_msrs)
3514 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3515 				msr, data);
3516 		break;
3517 	case MSR_AMD64_OSVW_ID_LENGTH:
3518 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3519 			return 1;
3520 		vcpu->arch.osvw.length = data;
3521 		break;
3522 	case MSR_AMD64_OSVW_STATUS:
3523 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3524 			return 1;
3525 		vcpu->arch.osvw.status = data;
3526 		break;
3527 	case MSR_PLATFORM_INFO:
3528 		if (!msr_info->host_initiated ||
3529 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3530 		     cpuid_fault_enabled(vcpu)))
3531 			return 1;
3532 		vcpu->arch.msr_platform_info = data;
3533 		break;
3534 	case MSR_MISC_FEATURES_ENABLES:
3535 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3536 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3537 		     !supports_cpuid_fault(vcpu)))
3538 			return 1;
3539 		vcpu->arch.msr_misc_features_enables = data;
3540 		break;
3541 	default:
3542 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3543 			return kvm_pmu_set_msr(vcpu, msr_info);
3544 		return KVM_MSR_RET_INVALID;
3545 	}
3546 	return 0;
3547 }
3548 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3549 
3550 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3551 {
3552 	u64 data;
3553 	u64 mcg_cap = vcpu->arch.mcg_cap;
3554 	unsigned bank_num = mcg_cap & 0xff;
3555 
3556 	switch (msr) {
3557 	case MSR_IA32_P5_MC_ADDR:
3558 	case MSR_IA32_P5_MC_TYPE:
3559 		data = 0;
3560 		break;
3561 	case MSR_IA32_MCG_CAP:
3562 		data = vcpu->arch.mcg_cap;
3563 		break;
3564 	case MSR_IA32_MCG_CTL:
3565 		if (!(mcg_cap & MCG_CTL_P) && !host)
3566 			return 1;
3567 		data = vcpu->arch.mcg_ctl;
3568 		break;
3569 	case MSR_IA32_MCG_STATUS:
3570 		data = vcpu->arch.mcg_status;
3571 		break;
3572 	default:
3573 		if (msr >= MSR_IA32_MC0_CTL &&
3574 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3575 			u32 offset = array_index_nospec(
3576 				msr - MSR_IA32_MC0_CTL,
3577 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3578 
3579 			data = vcpu->arch.mce_banks[offset];
3580 			break;
3581 		}
3582 		return 1;
3583 	}
3584 	*pdata = data;
3585 	return 0;
3586 }
3587 
3588 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3589 {
3590 	switch (msr_info->index) {
3591 	case MSR_IA32_PLATFORM_ID:
3592 	case MSR_IA32_EBL_CR_POWERON:
3593 	case MSR_IA32_LASTBRANCHFROMIP:
3594 	case MSR_IA32_LASTBRANCHTOIP:
3595 	case MSR_IA32_LASTINTFROMIP:
3596 	case MSR_IA32_LASTINTTOIP:
3597 	case MSR_AMD64_SYSCFG:
3598 	case MSR_K8_TSEG_ADDR:
3599 	case MSR_K8_TSEG_MASK:
3600 	case MSR_VM_HSAVE_PA:
3601 	case MSR_K8_INT_PENDING_MSG:
3602 	case MSR_AMD64_NB_CFG:
3603 	case MSR_FAM10H_MMIO_CONF_BASE:
3604 	case MSR_AMD64_BU_CFG2:
3605 	case MSR_IA32_PERF_CTL:
3606 	case MSR_AMD64_DC_CFG:
3607 	case MSR_F15H_EX_CFG:
3608 	/*
3609 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3610 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3611 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3612 	 * so for existing CPU-specific MSRs.
3613 	 */
3614 	case MSR_RAPL_POWER_UNIT:
3615 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3616 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3617 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3618 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3619 		msr_info->data = 0;
3620 		break;
3621 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3622 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3623 			return kvm_pmu_get_msr(vcpu, msr_info);
3624 		if (!msr_info->host_initiated)
3625 			return 1;
3626 		msr_info->data = 0;
3627 		break;
3628 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3629 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3630 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3631 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3632 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3633 			return kvm_pmu_get_msr(vcpu, msr_info);
3634 		msr_info->data = 0;
3635 		break;
3636 	case MSR_IA32_UCODE_REV:
3637 		msr_info->data = vcpu->arch.microcode_version;
3638 		break;
3639 	case MSR_IA32_ARCH_CAPABILITIES:
3640 		if (!msr_info->host_initiated &&
3641 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3642 			return 1;
3643 		msr_info->data = vcpu->arch.arch_capabilities;
3644 		break;
3645 	case MSR_IA32_PERF_CAPABILITIES:
3646 		if (!msr_info->host_initiated &&
3647 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3648 			return 1;
3649 		msr_info->data = vcpu->arch.perf_capabilities;
3650 		break;
3651 	case MSR_IA32_POWER_CTL:
3652 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3653 		break;
3654 	case MSR_IA32_TSC: {
3655 		/*
3656 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3657 		 * even when not intercepted. AMD manual doesn't explicitly
3658 		 * state this but appears to behave the same.
3659 		 *
3660 		 * On userspace reads and writes, however, we unconditionally
3661 		 * return L1's TSC value to ensure backwards-compatible
3662 		 * behavior for migration.
3663 		 */
3664 		u64 offset, ratio;
3665 
3666 		if (msr_info->host_initiated) {
3667 			offset = vcpu->arch.l1_tsc_offset;
3668 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3669 		} else {
3670 			offset = vcpu->arch.tsc_offset;
3671 			ratio = vcpu->arch.tsc_scaling_ratio;
3672 		}
3673 
3674 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3675 		break;
3676 	}
3677 	case MSR_MTRRcap:
3678 	case 0x200 ... 0x2ff:
3679 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3680 	case 0xcd: /* fsb frequency */
3681 		msr_info->data = 3;
3682 		break;
3683 		/*
3684 		 * MSR_EBC_FREQUENCY_ID
3685 		 * Conservative value valid for even the basic CPU models.
3686 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3687 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3688 		 * and 266MHz for model 3, or 4. Set Core Clock
3689 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3690 		 * 31:24) even though these are only valid for CPU
3691 		 * models > 2, however guests may end up dividing or
3692 		 * multiplying by zero otherwise.
3693 		 */
3694 	case MSR_EBC_FREQUENCY_ID:
3695 		msr_info->data = 1 << 24;
3696 		break;
3697 	case MSR_IA32_APICBASE:
3698 		msr_info->data = kvm_get_apic_base(vcpu);
3699 		break;
3700 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3701 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3702 	case MSR_IA32_TSC_DEADLINE:
3703 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3704 		break;
3705 	case MSR_IA32_TSC_ADJUST:
3706 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3707 		break;
3708 	case MSR_IA32_MISC_ENABLE:
3709 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3710 		break;
3711 	case MSR_IA32_SMBASE:
3712 		if (!msr_info->host_initiated)
3713 			return 1;
3714 		msr_info->data = vcpu->arch.smbase;
3715 		break;
3716 	case MSR_SMI_COUNT:
3717 		msr_info->data = vcpu->arch.smi_count;
3718 		break;
3719 	case MSR_IA32_PERF_STATUS:
3720 		/* TSC increment by tick */
3721 		msr_info->data = 1000ULL;
3722 		/* CPU multiplier */
3723 		msr_info->data |= (((uint64_t)4ULL) << 40);
3724 		break;
3725 	case MSR_EFER:
3726 		msr_info->data = vcpu->arch.efer;
3727 		break;
3728 	case MSR_KVM_WALL_CLOCK:
3729 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3730 			return 1;
3731 
3732 		msr_info->data = vcpu->kvm->arch.wall_clock;
3733 		break;
3734 	case MSR_KVM_WALL_CLOCK_NEW:
3735 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3736 			return 1;
3737 
3738 		msr_info->data = vcpu->kvm->arch.wall_clock;
3739 		break;
3740 	case MSR_KVM_SYSTEM_TIME:
3741 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3742 			return 1;
3743 
3744 		msr_info->data = vcpu->arch.time;
3745 		break;
3746 	case MSR_KVM_SYSTEM_TIME_NEW:
3747 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3748 			return 1;
3749 
3750 		msr_info->data = vcpu->arch.time;
3751 		break;
3752 	case MSR_KVM_ASYNC_PF_EN:
3753 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3754 			return 1;
3755 
3756 		msr_info->data = vcpu->arch.apf.msr_en_val;
3757 		break;
3758 	case MSR_KVM_ASYNC_PF_INT:
3759 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3760 			return 1;
3761 
3762 		msr_info->data = vcpu->arch.apf.msr_int_val;
3763 		break;
3764 	case MSR_KVM_ASYNC_PF_ACK:
3765 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3766 			return 1;
3767 
3768 		msr_info->data = 0;
3769 		break;
3770 	case MSR_KVM_STEAL_TIME:
3771 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3772 			return 1;
3773 
3774 		msr_info->data = vcpu->arch.st.msr_val;
3775 		break;
3776 	case MSR_KVM_PV_EOI_EN:
3777 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3778 			return 1;
3779 
3780 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3781 		break;
3782 	case MSR_KVM_POLL_CONTROL:
3783 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3784 			return 1;
3785 
3786 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3787 		break;
3788 	case MSR_IA32_P5_MC_ADDR:
3789 	case MSR_IA32_P5_MC_TYPE:
3790 	case MSR_IA32_MCG_CAP:
3791 	case MSR_IA32_MCG_CTL:
3792 	case MSR_IA32_MCG_STATUS:
3793 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3794 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3795 				   msr_info->host_initiated);
3796 	case MSR_IA32_XSS:
3797 		if (!msr_info->host_initiated &&
3798 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3799 			return 1;
3800 		msr_info->data = vcpu->arch.ia32_xss;
3801 		break;
3802 	case MSR_K7_CLK_CTL:
3803 		/*
3804 		 * Provide expected ramp-up count for K7. All other
3805 		 * are set to zero, indicating minimum divisors for
3806 		 * every field.
3807 		 *
3808 		 * This prevents guest kernels on AMD host with CPU
3809 		 * type 6, model 8 and higher from exploding due to
3810 		 * the rdmsr failing.
3811 		 */
3812 		msr_info->data = 0x20000000;
3813 		break;
3814 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3815 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3816 	case HV_X64_MSR_SYNDBG_OPTIONS:
3817 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3818 	case HV_X64_MSR_CRASH_CTL:
3819 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3820 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3821 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3822 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3823 		return kvm_hv_get_msr_common(vcpu,
3824 					     msr_info->index, &msr_info->data,
3825 					     msr_info->host_initiated);
3826 	case MSR_IA32_BBL_CR_CTL3:
3827 		/* This legacy MSR exists but isn't fully documented in current
3828 		 * silicon.  It is however accessed by winxp in very narrow
3829 		 * scenarios where it sets bit #19, itself documented as
3830 		 * a "reserved" bit.  Best effort attempt to source coherent
3831 		 * read data here should the balance of the register be
3832 		 * interpreted by the guest:
3833 		 *
3834 		 * L2 cache control register 3: 64GB range, 256KB size,
3835 		 * enabled, latency 0x1, configured
3836 		 */
3837 		msr_info->data = 0xbe702111;
3838 		break;
3839 	case MSR_AMD64_OSVW_ID_LENGTH:
3840 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3841 			return 1;
3842 		msr_info->data = vcpu->arch.osvw.length;
3843 		break;
3844 	case MSR_AMD64_OSVW_STATUS:
3845 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3846 			return 1;
3847 		msr_info->data = vcpu->arch.osvw.status;
3848 		break;
3849 	case MSR_PLATFORM_INFO:
3850 		if (!msr_info->host_initiated &&
3851 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3852 			return 1;
3853 		msr_info->data = vcpu->arch.msr_platform_info;
3854 		break;
3855 	case MSR_MISC_FEATURES_ENABLES:
3856 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3857 		break;
3858 	case MSR_K7_HWCR:
3859 		msr_info->data = vcpu->arch.msr_hwcr;
3860 		break;
3861 	default:
3862 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3863 			return kvm_pmu_get_msr(vcpu, msr_info);
3864 		return KVM_MSR_RET_INVALID;
3865 	}
3866 	return 0;
3867 }
3868 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3869 
3870 /*
3871  * Read or write a bunch of msrs. All parameters are kernel addresses.
3872  *
3873  * @return number of msrs set successfully.
3874  */
3875 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3876 		    struct kvm_msr_entry *entries,
3877 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3878 				  unsigned index, u64 *data))
3879 {
3880 	int i;
3881 
3882 	for (i = 0; i < msrs->nmsrs; ++i)
3883 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3884 			break;
3885 
3886 	return i;
3887 }
3888 
3889 /*
3890  * Read or write a bunch of msrs. Parameters are user addresses.
3891  *
3892  * @return number of msrs set successfully.
3893  */
3894 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3895 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3896 				unsigned index, u64 *data),
3897 		  int writeback)
3898 {
3899 	struct kvm_msrs msrs;
3900 	struct kvm_msr_entry *entries;
3901 	int r, n;
3902 	unsigned size;
3903 
3904 	r = -EFAULT;
3905 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3906 		goto out;
3907 
3908 	r = -E2BIG;
3909 	if (msrs.nmsrs >= MAX_IO_MSRS)
3910 		goto out;
3911 
3912 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3913 	entries = memdup_user(user_msrs->entries, size);
3914 	if (IS_ERR(entries)) {
3915 		r = PTR_ERR(entries);
3916 		goto out;
3917 	}
3918 
3919 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3920 	if (r < 0)
3921 		goto out_free;
3922 
3923 	r = -EFAULT;
3924 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3925 		goto out_free;
3926 
3927 	r = n;
3928 
3929 out_free:
3930 	kfree(entries);
3931 out:
3932 	return r;
3933 }
3934 
3935 static inline bool kvm_can_mwait_in_guest(void)
3936 {
3937 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3938 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3939 		boot_cpu_has(X86_FEATURE_ARAT);
3940 }
3941 
3942 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3943 					    struct kvm_cpuid2 __user *cpuid_arg)
3944 {
3945 	struct kvm_cpuid2 cpuid;
3946 	int r;
3947 
3948 	r = -EFAULT;
3949 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3950 		return r;
3951 
3952 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3953 	if (r)
3954 		return r;
3955 
3956 	r = -EFAULT;
3957 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3958 		return r;
3959 
3960 	return 0;
3961 }
3962 
3963 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3964 {
3965 	int r = 0;
3966 
3967 	switch (ext) {
3968 	case KVM_CAP_IRQCHIP:
3969 	case KVM_CAP_HLT:
3970 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3971 	case KVM_CAP_SET_TSS_ADDR:
3972 	case KVM_CAP_EXT_CPUID:
3973 	case KVM_CAP_EXT_EMUL_CPUID:
3974 	case KVM_CAP_CLOCKSOURCE:
3975 	case KVM_CAP_PIT:
3976 	case KVM_CAP_NOP_IO_DELAY:
3977 	case KVM_CAP_MP_STATE:
3978 	case KVM_CAP_SYNC_MMU:
3979 	case KVM_CAP_USER_NMI:
3980 	case KVM_CAP_REINJECT_CONTROL:
3981 	case KVM_CAP_IRQ_INJECT_STATUS:
3982 	case KVM_CAP_IOEVENTFD:
3983 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3984 	case KVM_CAP_PIT2:
3985 	case KVM_CAP_PIT_STATE2:
3986 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3987 	case KVM_CAP_VCPU_EVENTS:
3988 	case KVM_CAP_HYPERV:
3989 	case KVM_CAP_HYPERV_VAPIC:
3990 	case KVM_CAP_HYPERV_SPIN:
3991 	case KVM_CAP_HYPERV_SYNIC:
3992 	case KVM_CAP_HYPERV_SYNIC2:
3993 	case KVM_CAP_HYPERV_VP_INDEX:
3994 	case KVM_CAP_HYPERV_EVENTFD:
3995 	case KVM_CAP_HYPERV_TLBFLUSH:
3996 	case KVM_CAP_HYPERV_SEND_IPI:
3997 	case KVM_CAP_HYPERV_CPUID:
3998 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
3999 	case KVM_CAP_SYS_HYPERV_CPUID:
4000 	case KVM_CAP_PCI_SEGMENT:
4001 	case KVM_CAP_DEBUGREGS:
4002 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4003 	case KVM_CAP_XSAVE:
4004 	case KVM_CAP_ASYNC_PF:
4005 	case KVM_CAP_ASYNC_PF_INT:
4006 	case KVM_CAP_GET_TSC_KHZ:
4007 	case KVM_CAP_KVMCLOCK_CTRL:
4008 	case KVM_CAP_READONLY_MEM:
4009 	case KVM_CAP_HYPERV_TIME:
4010 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4011 	case KVM_CAP_TSC_DEADLINE_TIMER:
4012 	case KVM_CAP_DISABLE_QUIRKS:
4013 	case KVM_CAP_SET_BOOT_CPU_ID:
4014  	case KVM_CAP_SPLIT_IRQCHIP:
4015 	case KVM_CAP_IMMEDIATE_EXIT:
4016 	case KVM_CAP_PMU_EVENT_FILTER:
4017 	case KVM_CAP_GET_MSR_FEATURES:
4018 	case KVM_CAP_MSR_PLATFORM_INFO:
4019 	case KVM_CAP_EXCEPTION_PAYLOAD:
4020 	case KVM_CAP_SET_GUEST_DEBUG:
4021 	case KVM_CAP_LAST_CPU:
4022 	case KVM_CAP_X86_USER_SPACE_MSR:
4023 	case KVM_CAP_X86_MSR_FILTER:
4024 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4025 #ifdef CONFIG_X86_SGX_KVM
4026 	case KVM_CAP_SGX_ATTRIBUTE:
4027 #endif
4028 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4029 	case KVM_CAP_SREGS2:
4030 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4031 		r = 1;
4032 		break;
4033 	case KVM_CAP_EXIT_HYPERCALL:
4034 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4035 		break;
4036 	case KVM_CAP_SET_GUEST_DEBUG2:
4037 		return KVM_GUESTDBG_VALID_MASK;
4038 #ifdef CONFIG_KVM_XEN
4039 	case KVM_CAP_XEN_HVM:
4040 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4041 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4042 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
4043 		if (sched_info_on())
4044 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4045 		break;
4046 #endif
4047 	case KVM_CAP_SYNC_REGS:
4048 		r = KVM_SYNC_X86_VALID_FIELDS;
4049 		break;
4050 	case KVM_CAP_ADJUST_CLOCK:
4051 		r = KVM_CLOCK_TSC_STABLE;
4052 		break;
4053 	case KVM_CAP_X86_DISABLE_EXITS:
4054 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4055 		      KVM_X86_DISABLE_EXITS_CSTATE;
4056 		if(kvm_can_mwait_in_guest())
4057 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4058 		break;
4059 	case KVM_CAP_X86_SMM:
4060 		/* SMBASE is usually relocated above 1M on modern chipsets,
4061 		 * and SMM handlers might indeed rely on 4G segment limits,
4062 		 * so do not report SMM to be available if real mode is
4063 		 * emulated via vm86 mode.  Still, do not go to great lengths
4064 		 * to avoid userspace's usage of the feature, because it is a
4065 		 * fringe case that is not enabled except via specific settings
4066 		 * of the module parameters.
4067 		 */
4068 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4069 		break;
4070 	case KVM_CAP_VAPIC:
4071 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4072 		break;
4073 	case KVM_CAP_NR_VCPUS:
4074 		r = KVM_SOFT_MAX_VCPUS;
4075 		break;
4076 	case KVM_CAP_MAX_VCPUS:
4077 		r = KVM_MAX_VCPUS;
4078 		break;
4079 	case KVM_CAP_MAX_VCPU_ID:
4080 		r = KVM_MAX_VCPU_ID;
4081 		break;
4082 	case KVM_CAP_PV_MMU:	/* obsolete */
4083 		r = 0;
4084 		break;
4085 	case KVM_CAP_MCE:
4086 		r = KVM_MAX_MCE_BANKS;
4087 		break;
4088 	case KVM_CAP_XCRS:
4089 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4090 		break;
4091 	case KVM_CAP_TSC_CONTROL:
4092 		r = kvm_has_tsc_control;
4093 		break;
4094 	case KVM_CAP_X2APIC_API:
4095 		r = KVM_X2APIC_API_VALID_FLAGS;
4096 		break;
4097 	case KVM_CAP_NESTED_STATE:
4098 		r = kvm_x86_ops.nested_ops->get_state ?
4099 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4100 		break;
4101 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4102 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4103 		break;
4104 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4105 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4106 		break;
4107 	case KVM_CAP_SMALLER_MAXPHYADDR:
4108 		r = (int) allow_smaller_maxphyaddr;
4109 		break;
4110 	case KVM_CAP_STEAL_TIME:
4111 		r = sched_info_on();
4112 		break;
4113 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4114 		if (kvm_has_bus_lock_exit)
4115 			r = KVM_BUS_LOCK_DETECTION_OFF |
4116 			    KVM_BUS_LOCK_DETECTION_EXIT;
4117 		else
4118 			r = 0;
4119 		break;
4120 	default:
4121 		break;
4122 	}
4123 	return r;
4124 
4125 }
4126 
4127 long kvm_arch_dev_ioctl(struct file *filp,
4128 			unsigned int ioctl, unsigned long arg)
4129 {
4130 	void __user *argp = (void __user *)arg;
4131 	long r;
4132 
4133 	switch (ioctl) {
4134 	case KVM_GET_MSR_INDEX_LIST: {
4135 		struct kvm_msr_list __user *user_msr_list = argp;
4136 		struct kvm_msr_list msr_list;
4137 		unsigned n;
4138 
4139 		r = -EFAULT;
4140 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4141 			goto out;
4142 		n = msr_list.nmsrs;
4143 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4144 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4145 			goto out;
4146 		r = -E2BIG;
4147 		if (n < msr_list.nmsrs)
4148 			goto out;
4149 		r = -EFAULT;
4150 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4151 				 num_msrs_to_save * sizeof(u32)))
4152 			goto out;
4153 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4154 				 &emulated_msrs,
4155 				 num_emulated_msrs * sizeof(u32)))
4156 			goto out;
4157 		r = 0;
4158 		break;
4159 	}
4160 	case KVM_GET_SUPPORTED_CPUID:
4161 	case KVM_GET_EMULATED_CPUID: {
4162 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4163 		struct kvm_cpuid2 cpuid;
4164 
4165 		r = -EFAULT;
4166 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4167 			goto out;
4168 
4169 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4170 					    ioctl);
4171 		if (r)
4172 			goto out;
4173 
4174 		r = -EFAULT;
4175 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4176 			goto out;
4177 		r = 0;
4178 		break;
4179 	}
4180 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4181 		r = -EFAULT;
4182 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4183 				 sizeof(kvm_mce_cap_supported)))
4184 			goto out;
4185 		r = 0;
4186 		break;
4187 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4188 		struct kvm_msr_list __user *user_msr_list = argp;
4189 		struct kvm_msr_list msr_list;
4190 		unsigned int n;
4191 
4192 		r = -EFAULT;
4193 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4194 			goto out;
4195 		n = msr_list.nmsrs;
4196 		msr_list.nmsrs = num_msr_based_features;
4197 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4198 			goto out;
4199 		r = -E2BIG;
4200 		if (n < msr_list.nmsrs)
4201 			goto out;
4202 		r = -EFAULT;
4203 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4204 				 num_msr_based_features * sizeof(u32)))
4205 			goto out;
4206 		r = 0;
4207 		break;
4208 	}
4209 	case KVM_GET_MSRS:
4210 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4211 		break;
4212 	case KVM_GET_SUPPORTED_HV_CPUID:
4213 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4214 		break;
4215 	default:
4216 		r = -EINVAL;
4217 		break;
4218 	}
4219 out:
4220 	return r;
4221 }
4222 
4223 static void wbinvd_ipi(void *garbage)
4224 {
4225 	wbinvd();
4226 }
4227 
4228 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4229 {
4230 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4231 }
4232 
4233 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4234 {
4235 	/* Address WBINVD may be executed by guest */
4236 	if (need_emulate_wbinvd(vcpu)) {
4237 		if (static_call(kvm_x86_has_wbinvd_exit)())
4238 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4239 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4240 			smp_call_function_single(vcpu->cpu,
4241 					wbinvd_ipi, NULL, 1);
4242 	}
4243 
4244 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4245 
4246 	/* Save host pkru register if supported */
4247 	vcpu->arch.host_pkru = read_pkru();
4248 
4249 	/* Apply any externally detected TSC adjustments (due to suspend) */
4250 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4251 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4252 		vcpu->arch.tsc_offset_adjustment = 0;
4253 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4254 	}
4255 
4256 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4257 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4258 				rdtsc() - vcpu->arch.last_host_tsc;
4259 		if (tsc_delta < 0)
4260 			mark_tsc_unstable("KVM discovered backwards TSC");
4261 
4262 		if (kvm_check_tsc_unstable()) {
4263 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4264 						vcpu->arch.last_guest_tsc);
4265 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4266 			vcpu->arch.tsc_catchup = 1;
4267 		}
4268 
4269 		if (kvm_lapic_hv_timer_in_use(vcpu))
4270 			kvm_lapic_restart_hv_timer(vcpu);
4271 
4272 		/*
4273 		 * On a host with synchronized TSC, there is no need to update
4274 		 * kvmclock on vcpu->cpu migration
4275 		 */
4276 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4277 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4278 		if (vcpu->cpu != cpu)
4279 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4280 		vcpu->cpu = cpu;
4281 	}
4282 
4283 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4284 }
4285 
4286 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4287 {
4288 	struct kvm_host_map map;
4289 	struct kvm_steal_time *st;
4290 
4291 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4292 		return;
4293 
4294 	if (vcpu->arch.st.preempted)
4295 		return;
4296 
4297 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4298 			&vcpu->arch.st.cache, true))
4299 		return;
4300 
4301 	st = map.hva +
4302 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4303 
4304 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4305 
4306 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4307 }
4308 
4309 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4310 {
4311 	int idx;
4312 
4313 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4314 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4315 
4316 	/*
4317 	 * Take the srcu lock as memslots will be accessed to check the gfn
4318 	 * cache generation against the memslots generation.
4319 	 */
4320 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4321 	if (kvm_xen_msr_enabled(vcpu->kvm))
4322 		kvm_xen_runstate_set_preempted(vcpu);
4323 	else
4324 		kvm_steal_time_set_preempted(vcpu);
4325 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4326 
4327 	static_call(kvm_x86_vcpu_put)(vcpu);
4328 	vcpu->arch.last_host_tsc = rdtsc();
4329 }
4330 
4331 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4332 				    struct kvm_lapic_state *s)
4333 {
4334 	if (vcpu->arch.apicv_active)
4335 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4336 
4337 	return kvm_apic_get_state(vcpu, s);
4338 }
4339 
4340 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4341 				    struct kvm_lapic_state *s)
4342 {
4343 	int r;
4344 
4345 	r = kvm_apic_set_state(vcpu, s);
4346 	if (r)
4347 		return r;
4348 	update_cr8_intercept(vcpu);
4349 
4350 	return 0;
4351 }
4352 
4353 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4354 {
4355 	/*
4356 	 * We can accept userspace's request for interrupt injection
4357 	 * as long as we have a place to store the interrupt number.
4358 	 * The actual injection will happen when the CPU is able to
4359 	 * deliver the interrupt.
4360 	 */
4361 	if (kvm_cpu_has_extint(vcpu))
4362 		return false;
4363 
4364 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4365 	return (!lapic_in_kernel(vcpu) ||
4366 		kvm_apic_accept_pic_intr(vcpu));
4367 }
4368 
4369 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4370 {
4371 	/*
4372 	 * Do not cause an interrupt window exit if an exception
4373 	 * is pending or an event needs reinjection; userspace
4374 	 * might want to inject the interrupt manually using KVM_SET_REGS
4375 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4376 	 * instruction boundary and with no events half-injected.
4377 	 */
4378 	return (kvm_arch_interrupt_allowed(vcpu) &&
4379 		kvm_cpu_accept_dm_intr(vcpu) &&
4380 		!kvm_event_needs_reinjection(vcpu) &&
4381 		!vcpu->arch.exception.pending);
4382 }
4383 
4384 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4385 				    struct kvm_interrupt *irq)
4386 {
4387 	if (irq->irq >= KVM_NR_INTERRUPTS)
4388 		return -EINVAL;
4389 
4390 	if (!irqchip_in_kernel(vcpu->kvm)) {
4391 		kvm_queue_interrupt(vcpu, irq->irq, false);
4392 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4393 		return 0;
4394 	}
4395 
4396 	/*
4397 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4398 	 * fail for in-kernel 8259.
4399 	 */
4400 	if (pic_in_kernel(vcpu->kvm))
4401 		return -ENXIO;
4402 
4403 	if (vcpu->arch.pending_external_vector != -1)
4404 		return -EEXIST;
4405 
4406 	vcpu->arch.pending_external_vector = irq->irq;
4407 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4408 	return 0;
4409 }
4410 
4411 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4412 {
4413 	kvm_inject_nmi(vcpu);
4414 
4415 	return 0;
4416 }
4417 
4418 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4419 {
4420 	kvm_make_request(KVM_REQ_SMI, vcpu);
4421 
4422 	return 0;
4423 }
4424 
4425 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4426 					   struct kvm_tpr_access_ctl *tac)
4427 {
4428 	if (tac->flags)
4429 		return -EINVAL;
4430 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4431 	return 0;
4432 }
4433 
4434 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4435 					u64 mcg_cap)
4436 {
4437 	int r;
4438 	unsigned bank_num = mcg_cap & 0xff, bank;
4439 
4440 	r = -EINVAL;
4441 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4442 		goto out;
4443 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4444 		goto out;
4445 	r = 0;
4446 	vcpu->arch.mcg_cap = mcg_cap;
4447 	/* Init IA32_MCG_CTL to all 1s */
4448 	if (mcg_cap & MCG_CTL_P)
4449 		vcpu->arch.mcg_ctl = ~(u64)0;
4450 	/* Init IA32_MCi_CTL to all 1s */
4451 	for (bank = 0; bank < bank_num; bank++)
4452 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4453 
4454 	static_call(kvm_x86_setup_mce)(vcpu);
4455 out:
4456 	return r;
4457 }
4458 
4459 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4460 				      struct kvm_x86_mce *mce)
4461 {
4462 	u64 mcg_cap = vcpu->arch.mcg_cap;
4463 	unsigned bank_num = mcg_cap & 0xff;
4464 	u64 *banks = vcpu->arch.mce_banks;
4465 
4466 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4467 		return -EINVAL;
4468 	/*
4469 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4470 	 * reporting is disabled
4471 	 */
4472 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4473 	    vcpu->arch.mcg_ctl != ~(u64)0)
4474 		return 0;
4475 	banks += 4 * mce->bank;
4476 	/*
4477 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4478 	 * reporting is disabled for the bank
4479 	 */
4480 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4481 		return 0;
4482 	if (mce->status & MCI_STATUS_UC) {
4483 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4484 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4485 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4486 			return 0;
4487 		}
4488 		if (banks[1] & MCI_STATUS_VAL)
4489 			mce->status |= MCI_STATUS_OVER;
4490 		banks[2] = mce->addr;
4491 		banks[3] = mce->misc;
4492 		vcpu->arch.mcg_status = mce->mcg_status;
4493 		banks[1] = mce->status;
4494 		kvm_queue_exception(vcpu, MC_VECTOR);
4495 	} else if (!(banks[1] & MCI_STATUS_VAL)
4496 		   || !(banks[1] & MCI_STATUS_UC)) {
4497 		if (banks[1] & MCI_STATUS_VAL)
4498 			mce->status |= MCI_STATUS_OVER;
4499 		banks[2] = mce->addr;
4500 		banks[3] = mce->misc;
4501 		banks[1] = mce->status;
4502 	} else
4503 		banks[1] |= MCI_STATUS_OVER;
4504 	return 0;
4505 }
4506 
4507 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4508 					       struct kvm_vcpu_events *events)
4509 {
4510 	process_nmi(vcpu);
4511 
4512 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4513 		process_smi(vcpu);
4514 
4515 	/*
4516 	 * In guest mode, payload delivery should be deferred,
4517 	 * so that the L1 hypervisor can intercept #PF before
4518 	 * CR2 is modified (or intercept #DB before DR6 is
4519 	 * modified under nVMX). Unless the per-VM capability,
4520 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4521 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4522 	 * opportunistically defer the exception payload, deliver it if the
4523 	 * capability hasn't been requested before processing a
4524 	 * KVM_GET_VCPU_EVENTS.
4525 	 */
4526 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4527 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4528 		kvm_deliver_exception_payload(vcpu);
4529 
4530 	/*
4531 	 * The API doesn't provide the instruction length for software
4532 	 * exceptions, so don't report them. As long as the guest RIP
4533 	 * isn't advanced, we should expect to encounter the exception
4534 	 * again.
4535 	 */
4536 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4537 		events->exception.injected = 0;
4538 		events->exception.pending = 0;
4539 	} else {
4540 		events->exception.injected = vcpu->arch.exception.injected;
4541 		events->exception.pending = vcpu->arch.exception.pending;
4542 		/*
4543 		 * For ABI compatibility, deliberately conflate
4544 		 * pending and injected exceptions when
4545 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4546 		 */
4547 		if (!vcpu->kvm->arch.exception_payload_enabled)
4548 			events->exception.injected |=
4549 				vcpu->arch.exception.pending;
4550 	}
4551 	events->exception.nr = vcpu->arch.exception.nr;
4552 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4553 	events->exception.error_code = vcpu->arch.exception.error_code;
4554 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4555 	events->exception_payload = vcpu->arch.exception.payload;
4556 
4557 	events->interrupt.injected =
4558 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4559 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4560 	events->interrupt.soft = 0;
4561 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4562 
4563 	events->nmi.injected = vcpu->arch.nmi_injected;
4564 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4565 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4566 	events->nmi.pad = 0;
4567 
4568 	events->sipi_vector = 0; /* never valid when reporting to user space */
4569 
4570 	events->smi.smm = is_smm(vcpu);
4571 	events->smi.pending = vcpu->arch.smi_pending;
4572 	events->smi.smm_inside_nmi =
4573 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4574 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4575 
4576 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4577 			 | KVM_VCPUEVENT_VALID_SHADOW
4578 			 | KVM_VCPUEVENT_VALID_SMM);
4579 	if (vcpu->kvm->arch.exception_payload_enabled)
4580 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4581 
4582 	memset(&events->reserved, 0, sizeof(events->reserved));
4583 }
4584 
4585 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4586 
4587 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4588 					      struct kvm_vcpu_events *events)
4589 {
4590 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4591 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4592 			      | KVM_VCPUEVENT_VALID_SHADOW
4593 			      | KVM_VCPUEVENT_VALID_SMM
4594 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4595 		return -EINVAL;
4596 
4597 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4598 		if (!vcpu->kvm->arch.exception_payload_enabled)
4599 			return -EINVAL;
4600 		if (events->exception.pending)
4601 			events->exception.injected = 0;
4602 		else
4603 			events->exception_has_payload = 0;
4604 	} else {
4605 		events->exception.pending = 0;
4606 		events->exception_has_payload = 0;
4607 	}
4608 
4609 	if ((events->exception.injected || events->exception.pending) &&
4610 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4611 		return -EINVAL;
4612 
4613 	/* INITs are latched while in SMM */
4614 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4615 	    (events->smi.smm || events->smi.pending) &&
4616 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4617 		return -EINVAL;
4618 
4619 	process_nmi(vcpu);
4620 	vcpu->arch.exception.injected = events->exception.injected;
4621 	vcpu->arch.exception.pending = events->exception.pending;
4622 	vcpu->arch.exception.nr = events->exception.nr;
4623 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4624 	vcpu->arch.exception.error_code = events->exception.error_code;
4625 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4626 	vcpu->arch.exception.payload = events->exception_payload;
4627 
4628 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4629 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4630 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4631 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4632 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4633 						events->interrupt.shadow);
4634 
4635 	vcpu->arch.nmi_injected = events->nmi.injected;
4636 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4637 		vcpu->arch.nmi_pending = events->nmi.pending;
4638 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4639 
4640 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4641 	    lapic_in_kernel(vcpu))
4642 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4643 
4644 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4645 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4646 			kvm_smm_changed(vcpu, events->smi.smm);
4647 
4648 		vcpu->arch.smi_pending = events->smi.pending;
4649 
4650 		if (events->smi.smm) {
4651 			if (events->smi.smm_inside_nmi)
4652 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4653 			else
4654 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4655 		}
4656 
4657 		if (lapic_in_kernel(vcpu)) {
4658 			if (events->smi.latched_init)
4659 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4660 			else
4661 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4662 		}
4663 	}
4664 
4665 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4666 
4667 	return 0;
4668 }
4669 
4670 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4671 					     struct kvm_debugregs *dbgregs)
4672 {
4673 	unsigned long val;
4674 
4675 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4676 	kvm_get_dr(vcpu, 6, &val);
4677 	dbgregs->dr6 = val;
4678 	dbgregs->dr7 = vcpu->arch.dr7;
4679 	dbgregs->flags = 0;
4680 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4681 }
4682 
4683 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4684 					    struct kvm_debugregs *dbgregs)
4685 {
4686 	if (dbgregs->flags)
4687 		return -EINVAL;
4688 
4689 	if (!kvm_dr6_valid(dbgregs->dr6))
4690 		return -EINVAL;
4691 	if (!kvm_dr7_valid(dbgregs->dr7))
4692 		return -EINVAL;
4693 
4694 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4695 	kvm_update_dr0123(vcpu);
4696 	vcpu->arch.dr6 = dbgregs->dr6;
4697 	vcpu->arch.dr7 = dbgregs->dr7;
4698 	kvm_update_dr7(vcpu);
4699 
4700 	return 0;
4701 }
4702 
4703 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4704 
4705 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4706 {
4707 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4708 	u64 xstate_bv = xsave->header.xfeatures;
4709 	u64 valid;
4710 
4711 	/*
4712 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4713 	 * leaves 0 and 1 in the loop below.
4714 	 */
4715 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4716 
4717 	/* Set XSTATE_BV */
4718 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4719 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4720 
4721 	/*
4722 	 * Copy each region from the possibly compacted offset to the
4723 	 * non-compacted offset.
4724 	 */
4725 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4726 	while (valid) {
4727 		u32 size, offset, ecx, edx;
4728 		u64 xfeature_mask = valid & -valid;
4729 		int xfeature_nr = fls64(xfeature_mask) - 1;
4730 		void *src;
4731 
4732 		cpuid_count(XSTATE_CPUID, xfeature_nr,
4733 			    &size, &offset, &ecx, &edx);
4734 
4735 		if (xfeature_nr == XFEATURE_PKRU) {
4736 			memcpy(dest + offset, &vcpu->arch.pkru,
4737 			       sizeof(vcpu->arch.pkru));
4738 		} else {
4739 			src = get_xsave_addr(xsave, xfeature_nr);
4740 			if (src)
4741 				memcpy(dest + offset, src, size);
4742 		}
4743 
4744 		valid -= xfeature_mask;
4745 	}
4746 }
4747 
4748 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4749 {
4750 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4751 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4752 	u64 valid;
4753 
4754 	/*
4755 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4756 	 * leaves 0 and 1 in the loop below.
4757 	 */
4758 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4759 
4760 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4761 	xsave->header.xfeatures = xstate_bv;
4762 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4763 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4764 
4765 	/*
4766 	 * Copy each region from the non-compacted offset to the
4767 	 * possibly compacted offset.
4768 	 */
4769 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4770 	while (valid) {
4771 		u32 size, offset, ecx, edx;
4772 		u64 xfeature_mask = valid & -valid;
4773 		int xfeature_nr = fls64(xfeature_mask) - 1;
4774 
4775 		cpuid_count(XSTATE_CPUID, xfeature_nr,
4776 			    &size, &offset, &ecx, &edx);
4777 
4778 		if (xfeature_nr == XFEATURE_PKRU) {
4779 			memcpy(&vcpu->arch.pkru, src + offset,
4780 			       sizeof(vcpu->arch.pkru));
4781 		} else {
4782 			void *dest = get_xsave_addr(xsave, xfeature_nr);
4783 
4784 			if (dest)
4785 				memcpy(dest, src + offset, size);
4786 		}
4787 
4788 		valid -= xfeature_mask;
4789 	}
4790 }
4791 
4792 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4793 					 struct kvm_xsave *guest_xsave)
4794 {
4795 	if (!vcpu->arch.guest_fpu)
4796 		return;
4797 
4798 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4799 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4800 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4801 	} else {
4802 		memcpy(guest_xsave->region,
4803 			&vcpu->arch.guest_fpu->state.fxsave,
4804 			sizeof(struct fxregs_state));
4805 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4806 			XFEATURE_MASK_FPSSE;
4807 	}
4808 }
4809 
4810 #define XSAVE_MXCSR_OFFSET 24
4811 
4812 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4813 					struct kvm_xsave *guest_xsave)
4814 {
4815 	u64 xstate_bv;
4816 	u32 mxcsr;
4817 
4818 	if (!vcpu->arch.guest_fpu)
4819 		return 0;
4820 
4821 	xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4822 	mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4823 
4824 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4825 		/*
4826 		 * Here we allow setting states that are not present in
4827 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4828 		 * with old userspace.
4829 		 */
4830 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4831 			return -EINVAL;
4832 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4833 	} else {
4834 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4835 			mxcsr & ~mxcsr_feature_mask)
4836 			return -EINVAL;
4837 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4838 			guest_xsave->region, sizeof(struct fxregs_state));
4839 	}
4840 	return 0;
4841 }
4842 
4843 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4844 					struct kvm_xcrs *guest_xcrs)
4845 {
4846 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4847 		guest_xcrs->nr_xcrs = 0;
4848 		return;
4849 	}
4850 
4851 	guest_xcrs->nr_xcrs = 1;
4852 	guest_xcrs->flags = 0;
4853 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4854 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4855 }
4856 
4857 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4858 				       struct kvm_xcrs *guest_xcrs)
4859 {
4860 	int i, r = 0;
4861 
4862 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4863 		return -EINVAL;
4864 
4865 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4866 		return -EINVAL;
4867 
4868 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4869 		/* Only support XCR0 currently */
4870 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4871 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4872 				guest_xcrs->xcrs[i].value);
4873 			break;
4874 		}
4875 	if (r)
4876 		r = -EINVAL;
4877 	return r;
4878 }
4879 
4880 /*
4881  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4882  * stopped by the hypervisor.  This function will be called from the host only.
4883  * EINVAL is returned when the host attempts to set the flag for a guest that
4884  * does not support pv clocks.
4885  */
4886 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4887 {
4888 	if (!vcpu->arch.pv_time_enabled)
4889 		return -EINVAL;
4890 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4891 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4892 	return 0;
4893 }
4894 
4895 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4896 				     struct kvm_enable_cap *cap)
4897 {
4898 	int r;
4899 	uint16_t vmcs_version;
4900 	void __user *user_ptr;
4901 
4902 	if (cap->flags)
4903 		return -EINVAL;
4904 
4905 	switch (cap->cap) {
4906 	case KVM_CAP_HYPERV_SYNIC2:
4907 		if (cap->args[0])
4908 			return -EINVAL;
4909 		fallthrough;
4910 
4911 	case KVM_CAP_HYPERV_SYNIC:
4912 		if (!irqchip_in_kernel(vcpu->kvm))
4913 			return -EINVAL;
4914 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4915 					     KVM_CAP_HYPERV_SYNIC2);
4916 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4917 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4918 			return -ENOTTY;
4919 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4920 		if (!r) {
4921 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4922 			if (copy_to_user(user_ptr, &vmcs_version,
4923 					 sizeof(vmcs_version)))
4924 				r = -EFAULT;
4925 		}
4926 		return r;
4927 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4928 		if (!kvm_x86_ops.enable_direct_tlbflush)
4929 			return -ENOTTY;
4930 
4931 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4932 
4933 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4934 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4935 
4936 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4937 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4938 		if (vcpu->arch.pv_cpuid.enforce)
4939 			kvm_update_pv_runtime(vcpu);
4940 
4941 		return 0;
4942 	default:
4943 		return -EINVAL;
4944 	}
4945 }
4946 
4947 long kvm_arch_vcpu_ioctl(struct file *filp,
4948 			 unsigned int ioctl, unsigned long arg)
4949 {
4950 	struct kvm_vcpu *vcpu = filp->private_data;
4951 	void __user *argp = (void __user *)arg;
4952 	int r;
4953 	union {
4954 		struct kvm_sregs2 *sregs2;
4955 		struct kvm_lapic_state *lapic;
4956 		struct kvm_xsave *xsave;
4957 		struct kvm_xcrs *xcrs;
4958 		void *buffer;
4959 	} u;
4960 
4961 	vcpu_load(vcpu);
4962 
4963 	u.buffer = NULL;
4964 	switch (ioctl) {
4965 	case KVM_GET_LAPIC: {
4966 		r = -EINVAL;
4967 		if (!lapic_in_kernel(vcpu))
4968 			goto out;
4969 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4970 				GFP_KERNEL_ACCOUNT);
4971 
4972 		r = -ENOMEM;
4973 		if (!u.lapic)
4974 			goto out;
4975 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4976 		if (r)
4977 			goto out;
4978 		r = -EFAULT;
4979 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4980 			goto out;
4981 		r = 0;
4982 		break;
4983 	}
4984 	case KVM_SET_LAPIC: {
4985 		r = -EINVAL;
4986 		if (!lapic_in_kernel(vcpu))
4987 			goto out;
4988 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4989 		if (IS_ERR(u.lapic)) {
4990 			r = PTR_ERR(u.lapic);
4991 			goto out_nofree;
4992 		}
4993 
4994 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4995 		break;
4996 	}
4997 	case KVM_INTERRUPT: {
4998 		struct kvm_interrupt irq;
4999 
5000 		r = -EFAULT;
5001 		if (copy_from_user(&irq, argp, sizeof(irq)))
5002 			goto out;
5003 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5004 		break;
5005 	}
5006 	case KVM_NMI: {
5007 		r = kvm_vcpu_ioctl_nmi(vcpu);
5008 		break;
5009 	}
5010 	case KVM_SMI: {
5011 		r = kvm_vcpu_ioctl_smi(vcpu);
5012 		break;
5013 	}
5014 	case KVM_SET_CPUID: {
5015 		struct kvm_cpuid __user *cpuid_arg = argp;
5016 		struct kvm_cpuid cpuid;
5017 
5018 		r = -EFAULT;
5019 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5020 			goto out;
5021 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5022 		break;
5023 	}
5024 	case KVM_SET_CPUID2: {
5025 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5026 		struct kvm_cpuid2 cpuid;
5027 
5028 		r = -EFAULT;
5029 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5030 			goto out;
5031 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5032 					      cpuid_arg->entries);
5033 		break;
5034 	}
5035 	case KVM_GET_CPUID2: {
5036 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5037 		struct kvm_cpuid2 cpuid;
5038 
5039 		r = -EFAULT;
5040 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5041 			goto out;
5042 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5043 					      cpuid_arg->entries);
5044 		if (r)
5045 			goto out;
5046 		r = -EFAULT;
5047 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5048 			goto out;
5049 		r = 0;
5050 		break;
5051 	}
5052 	case KVM_GET_MSRS: {
5053 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5054 		r = msr_io(vcpu, argp, do_get_msr, 1);
5055 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5056 		break;
5057 	}
5058 	case KVM_SET_MSRS: {
5059 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5060 		r = msr_io(vcpu, argp, do_set_msr, 0);
5061 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5062 		break;
5063 	}
5064 	case KVM_TPR_ACCESS_REPORTING: {
5065 		struct kvm_tpr_access_ctl tac;
5066 
5067 		r = -EFAULT;
5068 		if (copy_from_user(&tac, argp, sizeof(tac)))
5069 			goto out;
5070 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5071 		if (r)
5072 			goto out;
5073 		r = -EFAULT;
5074 		if (copy_to_user(argp, &tac, sizeof(tac)))
5075 			goto out;
5076 		r = 0;
5077 		break;
5078 	};
5079 	case KVM_SET_VAPIC_ADDR: {
5080 		struct kvm_vapic_addr va;
5081 		int idx;
5082 
5083 		r = -EINVAL;
5084 		if (!lapic_in_kernel(vcpu))
5085 			goto out;
5086 		r = -EFAULT;
5087 		if (copy_from_user(&va, argp, sizeof(va)))
5088 			goto out;
5089 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5090 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5091 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5092 		break;
5093 	}
5094 	case KVM_X86_SETUP_MCE: {
5095 		u64 mcg_cap;
5096 
5097 		r = -EFAULT;
5098 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5099 			goto out;
5100 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5101 		break;
5102 	}
5103 	case KVM_X86_SET_MCE: {
5104 		struct kvm_x86_mce mce;
5105 
5106 		r = -EFAULT;
5107 		if (copy_from_user(&mce, argp, sizeof(mce)))
5108 			goto out;
5109 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5110 		break;
5111 	}
5112 	case KVM_GET_VCPU_EVENTS: {
5113 		struct kvm_vcpu_events events;
5114 
5115 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5116 
5117 		r = -EFAULT;
5118 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5119 			break;
5120 		r = 0;
5121 		break;
5122 	}
5123 	case KVM_SET_VCPU_EVENTS: {
5124 		struct kvm_vcpu_events events;
5125 
5126 		r = -EFAULT;
5127 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5128 			break;
5129 
5130 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5131 		break;
5132 	}
5133 	case KVM_GET_DEBUGREGS: {
5134 		struct kvm_debugregs dbgregs;
5135 
5136 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5137 
5138 		r = -EFAULT;
5139 		if (copy_to_user(argp, &dbgregs,
5140 				 sizeof(struct kvm_debugregs)))
5141 			break;
5142 		r = 0;
5143 		break;
5144 	}
5145 	case KVM_SET_DEBUGREGS: {
5146 		struct kvm_debugregs dbgregs;
5147 
5148 		r = -EFAULT;
5149 		if (copy_from_user(&dbgregs, argp,
5150 				   sizeof(struct kvm_debugregs)))
5151 			break;
5152 
5153 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5154 		break;
5155 	}
5156 	case KVM_GET_XSAVE: {
5157 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5158 		r = -ENOMEM;
5159 		if (!u.xsave)
5160 			break;
5161 
5162 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5163 
5164 		r = -EFAULT;
5165 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5166 			break;
5167 		r = 0;
5168 		break;
5169 	}
5170 	case KVM_SET_XSAVE: {
5171 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5172 		if (IS_ERR(u.xsave)) {
5173 			r = PTR_ERR(u.xsave);
5174 			goto out_nofree;
5175 		}
5176 
5177 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5178 		break;
5179 	}
5180 	case KVM_GET_XCRS: {
5181 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5182 		r = -ENOMEM;
5183 		if (!u.xcrs)
5184 			break;
5185 
5186 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5187 
5188 		r = -EFAULT;
5189 		if (copy_to_user(argp, u.xcrs,
5190 				 sizeof(struct kvm_xcrs)))
5191 			break;
5192 		r = 0;
5193 		break;
5194 	}
5195 	case KVM_SET_XCRS: {
5196 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5197 		if (IS_ERR(u.xcrs)) {
5198 			r = PTR_ERR(u.xcrs);
5199 			goto out_nofree;
5200 		}
5201 
5202 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5203 		break;
5204 	}
5205 	case KVM_SET_TSC_KHZ: {
5206 		u32 user_tsc_khz;
5207 
5208 		r = -EINVAL;
5209 		user_tsc_khz = (u32)arg;
5210 
5211 		if (kvm_has_tsc_control &&
5212 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5213 			goto out;
5214 
5215 		if (user_tsc_khz == 0)
5216 			user_tsc_khz = tsc_khz;
5217 
5218 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5219 			r = 0;
5220 
5221 		goto out;
5222 	}
5223 	case KVM_GET_TSC_KHZ: {
5224 		r = vcpu->arch.virtual_tsc_khz;
5225 		goto out;
5226 	}
5227 	case KVM_KVMCLOCK_CTRL: {
5228 		r = kvm_set_guest_paused(vcpu);
5229 		goto out;
5230 	}
5231 	case KVM_ENABLE_CAP: {
5232 		struct kvm_enable_cap cap;
5233 
5234 		r = -EFAULT;
5235 		if (copy_from_user(&cap, argp, sizeof(cap)))
5236 			goto out;
5237 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5238 		break;
5239 	}
5240 	case KVM_GET_NESTED_STATE: {
5241 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5242 		u32 user_data_size;
5243 
5244 		r = -EINVAL;
5245 		if (!kvm_x86_ops.nested_ops->get_state)
5246 			break;
5247 
5248 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5249 		r = -EFAULT;
5250 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5251 			break;
5252 
5253 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5254 						     user_data_size);
5255 		if (r < 0)
5256 			break;
5257 
5258 		if (r > user_data_size) {
5259 			if (put_user(r, &user_kvm_nested_state->size))
5260 				r = -EFAULT;
5261 			else
5262 				r = -E2BIG;
5263 			break;
5264 		}
5265 
5266 		r = 0;
5267 		break;
5268 	}
5269 	case KVM_SET_NESTED_STATE: {
5270 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5271 		struct kvm_nested_state kvm_state;
5272 		int idx;
5273 
5274 		r = -EINVAL;
5275 		if (!kvm_x86_ops.nested_ops->set_state)
5276 			break;
5277 
5278 		r = -EFAULT;
5279 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5280 			break;
5281 
5282 		r = -EINVAL;
5283 		if (kvm_state.size < sizeof(kvm_state))
5284 			break;
5285 
5286 		if (kvm_state.flags &
5287 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5288 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5289 		      | KVM_STATE_NESTED_GIF_SET))
5290 			break;
5291 
5292 		/* nested_run_pending implies guest_mode.  */
5293 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5294 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5295 			break;
5296 
5297 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5298 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5299 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5300 		break;
5301 	}
5302 	case KVM_GET_SUPPORTED_HV_CPUID:
5303 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5304 		break;
5305 #ifdef CONFIG_KVM_XEN
5306 	case KVM_XEN_VCPU_GET_ATTR: {
5307 		struct kvm_xen_vcpu_attr xva;
5308 
5309 		r = -EFAULT;
5310 		if (copy_from_user(&xva, argp, sizeof(xva)))
5311 			goto out;
5312 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5313 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5314 			r = -EFAULT;
5315 		break;
5316 	}
5317 	case KVM_XEN_VCPU_SET_ATTR: {
5318 		struct kvm_xen_vcpu_attr xva;
5319 
5320 		r = -EFAULT;
5321 		if (copy_from_user(&xva, argp, sizeof(xva)))
5322 			goto out;
5323 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5324 		break;
5325 	}
5326 #endif
5327 	case KVM_GET_SREGS2: {
5328 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5329 		r = -ENOMEM;
5330 		if (!u.sregs2)
5331 			goto out;
5332 		__get_sregs2(vcpu, u.sregs2);
5333 		r = -EFAULT;
5334 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5335 			goto out;
5336 		r = 0;
5337 		break;
5338 	}
5339 	case KVM_SET_SREGS2: {
5340 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5341 		if (IS_ERR(u.sregs2)) {
5342 			r = PTR_ERR(u.sregs2);
5343 			u.sregs2 = NULL;
5344 			goto out;
5345 		}
5346 		r = __set_sregs2(vcpu, u.sregs2);
5347 		break;
5348 	}
5349 	default:
5350 		r = -EINVAL;
5351 	}
5352 out:
5353 	kfree(u.buffer);
5354 out_nofree:
5355 	vcpu_put(vcpu);
5356 	return r;
5357 }
5358 
5359 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5360 {
5361 	return VM_FAULT_SIGBUS;
5362 }
5363 
5364 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5365 {
5366 	int ret;
5367 
5368 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5369 		return -EINVAL;
5370 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5371 	return ret;
5372 }
5373 
5374 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5375 					      u64 ident_addr)
5376 {
5377 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5378 }
5379 
5380 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5381 					 unsigned long kvm_nr_mmu_pages)
5382 {
5383 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5384 		return -EINVAL;
5385 
5386 	mutex_lock(&kvm->slots_lock);
5387 
5388 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5389 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5390 
5391 	mutex_unlock(&kvm->slots_lock);
5392 	return 0;
5393 }
5394 
5395 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5396 {
5397 	return kvm->arch.n_max_mmu_pages;
5398 }
5399 
5400 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5401 {
5402 	struct kvm_pic *pic = kvm->arch.vpic;
5403 	int r;
5404 
5405 	r = 0;
5406 	switch (chip->chip_id) {
5407 	case KVM_IRQCHIP_PIC_MASTER:
5408 		memcpy(&chip->chip.pic, &pic->pics[0],
5409 			sizeof(struct kvm_pic_state));
5410 		break;
5411 	case KVM_IRQCHIP_PIC_SLAVE:
5412 		memcpy(&chip->chip.pic, &pic->pics[1],
5413 			sizeof(struct kvm_pic_state));
5414 		break;
5415 	case KVM_IRQCHIP_IOAPIC:
5416 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5417 		break;
5418 	default:
5419 		r = -EINVAL;
5420 		break;
5421 	}
5422 	return r;
5423 }
5424 
5425 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5426 {
5427 	struct kvm_pic *pic = kvm->arch.vpic;
5428 	int r;
5429 
5430 	r = 0;
5431 	switch (chip->chip_id) {
5432 	case KVM_IRQCHIP_PIC_MASTER:
5433 		spin_lock(&pic->lock);
5434 		memcpy(&pic->pics[0], &chip->chip.pic,
5435 			sizeof(struct kvm_pic_state));
5436 		spin_unlock(&pic->lock);
5437 		break;
5438 	case KVM_IRQCHIP_PIC_SLAVE:
5439 		spin_lock(&pic->lock);
5440 		memcpy(&pic->pics[1], &chip->chip.pic,
5441 			sizeof(struct kvm_pic_state));
5442 		spin_unlock(&pic->lock);
5443 		break;
5444 	case KVM_IRQCHIP_IOAPIC:
5445 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5446 		break;
5447 	default:
5448 		r = -EINVAL;
5449 		break;
5450 	}
5451 	kvm_pic_update_irq(pic);
5452 	return r;
5453 }
5454 
5455 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5456 {
5457 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5458 
5459 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5460 
5461 	mutex_lock(&kps->lock);
5462 	memcpy(ps, &kps->channels, sizeof(*ps));
5463 	mutex_unlock(&kps->lock);
5464 	return 0;
5465 }
5466 
5467 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5468 {
5469 	int i;
5470 	struct kvm_pit *pit = kvm->arch.vpit;
5471 
5472 	mutex_lock(&pit->pit_state.lock);
5473 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5474 	for (i = 0; i < 3; i++)
5475 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5476 	mutex_unlock(&pit->pit_state.lock);
5477 	return 0;
5478 }
5479 
5480 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5481 {
5482 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5483 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5484 		sizeof(ps->channels));
5485 	ps->flags = kvm->arch.vpit->pit_state.flags;
5486 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5487 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5488 	return 0;
5489 }
5490 
5491 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5492 {
5493 	int start = 0;
5494 	int i;
5495 	u32 prev_legacy, cur_legacy;
5496 	struct kvm_pit *pit = kvm->arch.vpit;
5497 
5498 	mutex_lock(&pit->pit_state.lock);
5499 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5500 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5501 	if (!prev_legacy && cur_legacy)
5502 		start = 1;
5503 	memcpy(&pit->pit_state.channels, &ps->channels,
5504 	       sizeof(pit->pit_state.channels));
5505 	pit->pit_state.flags = ps->flags;
5506 	for (i = 0; i < 3; i++)
5507 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5508 				   start && i == 0);
5509 	mutex_unlock(&pit->pit_state.lock);
5510 	return 0;
5511 }
5512 
5513 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5514 				 struct kvm_reinject_control *control)
5515 {
5516 	struct kvm_pit *pit = kvm->arch.vpit;
5517 
5518 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5519 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5520 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5521 	 */
5522 	mutex_lock(&pit->pit_state.lock);
5523 	kvm_pit_set_reinject(pit, control->pit_reinject);
5524 	mutex_unlock(&pit->pit_state.lock);
5525 
5526 	return 0;
5527 }
5528 
5529 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5530 {
5531 
5532 	/*
5533 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5534 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5535 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5536 	 * VM-Exit.
5537 	 */
5538 	struct kvm_vcpu *vcpu;
5539 	int i;
5540 
5541 	kvm_for_each_vcpu(i, vcpu, kvm)
5542 		kvm_vcpu_kick(vcpu);
5543 }
5544 
5545 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5546 			bool line_status)
5547 {
5548 	if (!irqchip_in_kernel(kvm))
5549 		return -ENXIO;
5550 
5551 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5552 					irq_event->irq, irq_event->level,
5553 					line_status);
5554 	return 0;
5555 }
5556 
5557 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5558 			    struct kvm_enable_cap *cap)
5559 {
5560 	int r;
5561 
5562 	if (cap->flags)
5563 		return -EINVAL;
5564 
5565 	switch (cap->cap) {
5566 	case KVM_CAP_DISABLE_QUIRKS:
5567 		kvm->arch.disabled_quirks = cap->args[0];
5568 		r = 0;
5569 		break;
5570 	case KVM_CAP_SPLIT_IRQCHIP: {
5571 		mutex_lock(&kvm->lock);
5572 		r = -EINVAL;
5573 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5574 			goto split_irqchip_unlock;
5575 		r = -EEXIST;
5576 		if (irqchip_in_kernel(kvm))
5577 			goto split_irqchip_unlock;
5578 		if (kvm->created_vcpus)
5579 			goto split_irqchip_unlock;
5580 		r = kvm_setup_empty_irq_routing(kvm);
5581 		if (r)
5582 			goto split_irqchip_unlock;
5583 		/* Pairs with irqchip_in_kernel. */
5584 		smp_wmb();
5585 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5586 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5587 		r = 0;
5588 split_irqchip_unlock:
5589 		mutex_unlock(&kvm->lock);
5590 		break;
5591 	}
5592 	case KVM_CAP_X2APIC_API:
5593 		r = -EINVAL;
5594 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5595 			break;
5596 
5597 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5598 			kvm->arch.x2apic_format = true;
5599 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5600 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5601 
5602 		r = 0;
5603 		break;
5604 	case KVM_CAP_X86_DISABLE_EXITS:
5605 		r = -EINVAL;
5606 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5607 			break;
5608 
5609 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5610 			kvm_can_mwait_in_guest())
5611 			kvm->arch.mwait_in_guest = true;
5612 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5613 			kvm->arch.hlt_in_guest = true;
5614 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5615 			kvm->arch.pause_in_guest = true;
5616 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5617 			kvm->arch.cstate_in_guest = true;
5618 		r = 0;
5619 		break;
5620 	case KVM_CAP_MSR_PLATFORM_INFO:
5621 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5622 		r = 0;
5623 		break;
5624 	case KVM_CAP_EXCEPTION_PAYLOAD:
5625 		kvm->arch.exception_payload_enabled = cap->args[0];
5626 		r = 0;
5627 		break;
5628 	case KVM_CAP_X86_USER_SPACE_MSR:
5629 		kvm->arch.user_space_msr_mask = cap->args[0];
5630 		r = 0;
5631 		break;
5632 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5633 		r = -EINVAL;
5634 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5635 			break;
5636 
5637 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5638 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5639 			break;
5640 
5641 		if (kvm_has_bus_lock_exit &&
5642 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5643 			kvm->arch.bus_lock_detection_enabled = true;
5644 		r = 0;
5645 		break;
5646 #ifdef CONFIG_X86_SGX_KVM
5647 	case KVM_CAP_SGX_ATTRIBUTE: {
5648 		unsigned long allowed_attributes = 0;
5649 
5650 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5651 		if (r)
5652 			break;
5653 
5654 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5655 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5656 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5657 			kvm->arch.sgx_provisioning_allowed = true;
5658 		else
5659 			r = -EINVAL;
5660 		break;
5661 	}
5662 #endif
5663 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5664 		r = -EINVAL;
5665 		if (kvm_x86_ops.vm_copy_enc_context_from)
5666 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5667 		return r;
5668 	case KVM_CAP_EXIT_HYPERCALL:
5669 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5670 			r = -EINVAL;
5671 			break;
5672 		}
5673 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5674 		r = 0;
5675 		break;
5676 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5677 		r = -EINVAL;
5678 		if (cap->args[0] & ~1)
5679 			break;
5680 		kvm->arch.exit_on_emulation_error = cap->args[0];
5681 		r = 0;
5682 		break;
5683 	default:
5684 		r = -EINVAL;
5685 		break;
5686 	}
5687 	return r;
5688 }
5689 
5690 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5691 {
5692 	struct kvm_x86_msr_filter *msr_filter;
5693 
5694 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5695 	if (!msr_filter)
5696 		return NULL;
5697 
5698 	msr_filter->default_allow = default_allow;
5699 	return msr_filter;
5700 }
5701 
5702 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5703 {
5704 	u32 i;
5705 
5706 	if (!msr_filter)
5707 		return;
5708 
5709 	for (i = 0; i < msr_filter->count; i++)
5710 		kfree(msr_filter->ranges[i].bitmap);
5711 
5712 	kfree(msr_filter);
5713 }
5714 
5715 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5716 			      struct kvm_msr_filter_range *user_range)
5717 {
5718 	unsigned long *bitmap = NULL;
5719 	size_t bitmap_size;
5720 
5721 	if (!user_range->nmsrs)
5722 		return 0;
5723 
5724 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5725 		return -EINVAL;
5726 
5727 	if (!user_range->flags)
5728 		return -EINVAL;
5729 
5730 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5731 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5732 		return -EINVAL;
5733 
5734 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5735 	if (IS_ERR(bitmap))
5736 		return PTR_ERR(bitmap);
5737 
5738 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5739 		.flags = user_range->flags,
5740 		.base = user_range->base,
5741 		.nmsrs = user_range->nmsrs,
5742 		.bitmap = bitmap,
5743 	};
5744 
5745 	msr_filter->count++;
5746 	return 0;
5747 }
5748 
5749 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5750 {
5751 	struct kvm_msr_filter __user *user_msr_filter = argp;
5752 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5753 	struct kvm_msr_filter filter;
5754 	bool default_allow;
5755 	bool empty = true;
5756 	int r = 0;
5757 	u32 i;
5758 
5759 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5760 		return -EFAULT;
5761 
5762 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5763 		empty &= !filter.ranges[i].nmsrs;
5764 
5765 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5766 	if (empty && !default_allow)
5767 		return -EINVAL;
5768 
5769 	new_filter = kvm_alloc_msr_filter(default_allow);
5770 	if (!new_filter)
5771 		return -ENOMEM;
5772 
5773 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5774 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5775 		if (r) {
5776 			kvm_free_msr_filter(new_filter);
5777 			return r;
5778 		}
5779 	}
5780 
5781 	mutex_lock(&kvm->lock);
5782 
5783 	/* The per-VM filter is protected by kvm->lock... */
5784 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5785 
5786 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5787 	synchronize_srcu(&kvm->srcu);
5788 
5789 	kvm_free_msr_filter(old_filter);
5790 
5791 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5792 	mutex_unlock(&kvm->lock);
5793 
5794 	return 0;
5795 }
5796 
5797 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5798 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5799 {
5800 	struct kvm_vcpu *vcpu;
5801 	int i, ret = 0;
5802 
5803 	mutex_lock(&kvm->lock);
5804 	kvm_for_each_vcpu(i, vcpu, kvm) {
5805 		if (!vcpu->arch.pv_time_enabled)
5806 			continue;
5807 
5808 		ret = kvm_set_guest_paused(vcpu);
5809 		if (ret) {
5810 			kvm_err("Failed to pause guest VCPU%d: %d\n",
5811 				vcpu->vcpu_id, ret);
5812 			break;
5813 		}
5814 	}
5815 	mutex_unlock(&kvm->lock);
5816 
5817 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
5818 }
5819 
5820 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5821 {
5822 	switch (state) {
5823 	case PM_HIBERNATION_PREPARE:
5824 	case PM_SUSPEND_PREPARE:
5825 		return kvm_arch_suspend_notifier(kvm);
5826 	}
5827 
5828 	return NOTIFY_DONE;
5829 }
5830 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5831 
5832 long kvm_arch_vm_ioctl(struct file *filp,
5833 		       unsigned int ioctl, unsigned long arg)
5834 {
5835 	struct kvm *kvm = filp->private_data;
5836 	void __user *argp = (void __user *)arg;
5837 	int r = -ENOTTY;
5838 	/*
5839 	 * This union makes it completely explicit to gcc-3.x
5840 	 * that these two variables' stack usage should be
5841 	 * combined, not added together.
5842 	 */
5843 	union {
5844 		struct kvm_pit_state ps;
5845 		struct kvm_pit_state2 ps2;
5846 		struct kvm_pit_config pit_config;
5847 	} u;
5848 
5849 	switch (ioctl) {
5850 	case KVM_SET_TSS_ADDR:
5851 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5852 		break;
5853 	case KVM_SET_IDENTITY_MAP_ADDR: {
5854 		u64 ident_addr;
5855 
5856 		mutex_lock(&kvm->lock);
5857 		r = -EINVAL;
5858 		if (kvm->created_vcpus)
5859 			goto set_identity_unlock;
5860 		r = -EFAULT;
5861 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5862 			goto set_identity_unlock;
5863 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5864 set_identity_unlock:
5865 		mutex_unlock(&kvm->lock);
5866 		break;
5867 	}
5868 	case KVM_SET_NR_MMU_PAGES:
5869 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5870 		break;
5871 	case KVM_GET_NR_MMU_PAGES:
5872 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5873 		break;
5874 	case KVM_CREATE_IRQCHIP: {
5875 		mutex_lock(&kvm->lock);
5876 
5877 		r = -EEXIST;
5878 		if (irqchip_in_kernel(kvm))
5879 			goto create_irqchip_unlock;
5880 
5881 		r = -EINVAL;
5882 		if (kvm->created_vcpus)
5883 			goto create_irqchip_unlock;
5884 
5885 		r = kvm_pic_init(kvm);
5886 		if (r)
5887 			goto create_irqchip_unlock;
5888 
5889 		r = kvm_ioapic_init(kvm);
5890 		if (r) {
5891 			kvm_pic_destroy(kvm);
5892 			goto create_irqchip_unlock;
5893 		}
5894 
5895 		r = kvm_setup_default_irq_routing(kvm);
5896 		if (r) {
5897 			kvm_ioapic_destroy(kvm);
5898 			kvm_pic_destroy(kvm);
5899 			goto create_irqchip_unlock;
5900 		}
5901 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5902 		smp_wmb();
5903 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5904 	create_irqchip_unlock:
5905 		mutex_unlock(&kvm->lock);
5906 		break;
5907 	}
5908 	case KVM_CREATE_PIT:
5909 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5910 		goto create_pit;
5911 	case KVM_CREATE_PIT2:
5912 		r = -EFAULT;
5913 		if (copy_from_user(&u.pit_config, argp,
5914 				   sizeof(struct kvm_pit_config)))
5915 			goto out;
5916 	create_pit:
5917 		mutex_lock(&kvm->lock);
5918 		r = -EEXIST;
5919 		if (kvm->arch.vpit)
5920 			goto create_pit_unlock;
5921 		r = -ENOMEM;
5922 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5923 		if (kvm->arch.vpit)
5924 			r = 0;
5925 	create_pit_unlock:
5926 		mutex_unlock(&kvm->lock);
5927 		break;
5928 	case KVM_GET_IRQCHIP: {
5929 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5930 		struct kvm_irqchip *chip;
5931 
5932 		chip = memdup_user(argp, sizeof(*chip));
5933 		if (IS_ERR(chip)) {
5934 			r = PTR_ERR(chip);
5935 			goto out;
5936 		}
5937 
5938 		r = -ENXIO;
5939 		if (!irqchip_kernel(kvm))
5940 			goto get_irqchip_out;
5941 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5942 		if (r)
5943 			goto get_irqchip_out;
5944 		r = -EFAULT;
5945 		if (copy_to_user(argp, chip, sizeof(*chip)))
5946 			goto get_irqchip_out;
5947 		r = 0;
5948 	get_irqchip_out:
5949 		kfree(chip);
5950 		break;
5951 	}
5952 	case KVM_SET_IRQCHIP: {
5953 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5954 		struct kvm_irqchip *chip;
5955 
5956 		chip = memdup_user(argp, sizeof(*chip));
5957 		if (IS_ERR(chip)) {
5958 			r = PTR_ERR(chip);
5959 			goto out;
5960 		}
5961 
5962 		r = -ENXIO;
5963 		if (!irqchip_kernel(kvm))
5964 			goto set_irqchip_out;
5965 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5966 	set_irqchip_out:
5967 		kfree(chip);
5968 		break;
5969 	}
5970 	case KVM_GET_PIT: {
5971 		r = -EFAULT;
5972 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5973 			goto out;
5974 		r = -ENXIO;
5975 		if (!kvm->arch.vpit)
5976 			goto out;
5977 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5978 		if (r)
5979 			goto out;
5980 		r = -EFAULT;
5981 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5982 			goto out;
5983 		r = 0;
5984 		break;
5985 	}
5986 	case KVM_SET_PIT: {
5987 		r = -EFAULT;
5988 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5989 			goto out;
5990 		mutex_lock(&kvm->lock);
5991 		r = -ENXIO;
5992 		if (!kvm->arch.vpit)
5993 			goto set_pit_out;
5994 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5995 set_pit_out:
5996 		mutex_unlock(&kvm->lock);
5997 		break;
5998 	}
5999 	case KVM_GET_PIT2: {
6000 		r = -ENXIO;
6001 		if (!kvm->arch.vpit)
6002 			goto out;
6003 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6004 		if (r)
6005 			goto out;
6006 		r = -EFAULT;
6007 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6008 			goto out;
6009 		r = 0;
6010 		break;
6011 	}
6012 	case KVM_SET_PIT2: {
6013 		r = -EFAULT;
6014 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6015 			goto out;
6016 		mutex_lock(&kvm->lock);
6017 		r = -ENXIO;
6018 		if (!kvm->arch.vpit)
6019 			goto set_pit2_out;
6020 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6021 set_pit2_out:
6022 		mutex_unlock(&kvm->lock);
6023 		break;
6024 	}
6025 	case KVM_REINJECT_CONTROL: {
6026 		struct kvm_reinject_control control;
6027 		r =  -EFAULT;
6028 		if (copy_from_user(&control, argp, sizeof(control)))
6029 			goto out;
6030 		r = -ENXIO;
6031 		if (!kvm->arch.vpit)
6032 			goto out;
6033 		r = kvm_vm_ioctl_reinject(kvm, &control);
6034 		break;
6035 	}
6036 	case KVM_SET_BOOT_CPU_ID:
6037 		r = 0;
6038 		mutex_lock(&kvm->lock);
6039 		if (kvm->created_vcpus)
6040 			r = -EBUSY;
6041 		else
6042 			kvm->arch.bsp_vcpu_id = arg;
6043 		mutex_unlock(&kvm->lock);
6044 		break;
6045 #ifdef CONFIG_KVM_XEN
6046 	case KVM_XEN_HVM_CONFIG: {
6047 		struct kvm_xen_hvm_config xhc;
6048 		r = -EFAULT;
6049 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6050 			goto out;
6051 		r = kvm_xen_hvm_config(kvm, &xhc);
6052 		break;
6053 	}
6054 	case KVM_XEN_HVM_GET_ATTR: {
6055 		struct kvm_xen_hvm_attr xha;
6056 
6057 		r = -EFAULT;
6058 		if (copy_from_user(&xha, argp, sizeof(xha)))
6059 			goto out;
6060 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6061 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6062 			r = -EFAULT;
6063 		break;
6064 	}
6065 	case KVM_XEN_HVM_SET_ATTR: {
6066 		struct kvm_xen_hvm_attr xha;
6067 
6068 		r = -EFAULT;
6069 		if (copy_from_user(&xha, argp, sizeof(xha)))
6070 			goto out;
6071 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6072 		break;
6073 	}
6074 #endif
6075 	case KVM_SET_CLOCK: {
6076 		struct kvm_arch *ka = &kvm->arch;
6077 		struct kvm_clock_data user_ns;
6078 		u64 now_ns;
6079 
6080 		r = -EFAULT;
6081 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6082 			goto out;
6083 
6084 		r = -EINVAL;
6085 		if (user_ns.flags)
6086 			goto out;
6087 
6088 		r = 0;
6089 		/*
6090 		 * TODO: userspace has to take care of races with VCPU_RUN, so
6091 		 * kvm_gen_update_masterclock() can be cut down to locked
6092 		 * pvclock_update_vm_gtod_copy().
6093 		 */
6094 		kvm_gen_update_masterclock(kvm);
6095 
6096 		/*
6097 		 * This pairs with kvm_guest_time_update(): when masterclock is
6098 		 * in use, we use master_kernel_ns + kvmclock_offset to set
6099 		 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6100 		 * is slightly ahead) here we risk going negative on unsigned
6101 		 * 'system_time' when 'user_ns.clock' is very small.
6102 		 */
6103 		spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6104 		if (kvm->arch.use_master_clock)
6105 			now_ns = ka->master_kernel_ns;
6106 		else
6107 			now_ns = get_kvmclock_base_ns();
6108 		ka->kvmclock_offset = user_ns.clock - now_ns;
6109 		spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6110 
6111 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6112 		break;
6113 	}
6114 	case KVM_GET_CLOCK: {
6115 		struct kvm_clock_data user_ns;
6116 		u64 now_ns;
6117 
6118 		now_ns = get_kvmclock_ns(kvm);
6119 		user_ns.clock = now_ns;
6120 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6121 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6122 
6123 		r = -EFAULT;
6124 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6125 			goto out;
6126 		r = 0;
6127 		break;
6128 	}
6129 	case KVM_MEMORY_ENCRYPT_OP: {
6130 		r = -ENOTTY;
6131 		if (kvm_x86_ops.mem_enc_op)
6132 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6133 		break;
6134 	}
6135 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6136 		struct kvm_enc_region region;
6137 
6138 		r = -EFAULT;
6139 		if (copy_from_user(&region, argp, sizeof(region)))
6140 			goto out;
6141 
6142 		r = -ENOTTY;
6143 		if (kvm_x86_ops.mem_enc_reg_region)
6144 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6145 		break;
6146 	}
6147 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6148 		struct kvm_enc_region region;
6149 
6150 		r = -EFAULT;
6151 		if (copy_from_user(&region, argp, sizeof(region)))
6152 			goto out;
6153 
6154 		r = -ENOTTY;
6155 		if (kvm_x86_ops.mem_enc_unreg_region)
6156 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6157 		break;
6158 	}
6159 	case KVM_HYPERV_EVENTFD: {
6160 		struct kvm_hyperv_eventfd hvevfd;
6161 
6162 		r = -EFAULT;
6163 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6164 			goto out;
6165 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6166 		break;
6167 	}
6168 	case KVM_SET_PMU_EVENT_FILTER:
6169 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6170 		break;
6171 	case KVM_X86_SET_MSR_FILTER:
6172 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6173 		break;
6174 	default:
6175 		r = -ENOTTY;
6176 	}
6177 out:
6178 	return r;
6179 }
6180 
6181 static void kvm_init_msr_list(void)
6182 {
6183 	struct x86_pmu_capability x86_pmu;
6184 	u32 dummy[2];
6185 	unsigned i;
6186 
6187 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6188 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6189 
6190 	perf_get_x86_pmu_capability(&x86_pmu);
6191 
6192 	num_msrs_to_save = 0;
6193 	num_emulated_msrs = 0;
6194 	num_msr_based_features = 0;
6195 
6196 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6197 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6198 			continue;
6199 
6200 		/*
6201 		 * Even MSRs that are valid in the host may not be exposed
6202 		 * to the guests in some cases.
6203 		 */
6204 		switch (msrs_to_save_all[i]) {
6205 		case MSR_IA32_BNDCFGS:
6206 			if (!kvm_mpx_supported())
6207 				continue;
6208 			break;
6209 		case MSR_TSC_AUX:
6210 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6211 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6212 				continue;
6213 			break;
6214 		case MSR_IA32_UMWAIT_CONTROL:
6215 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6216 				continue;
6217 			break;
6218 		case MSR_IA32_RTIT_CTL:
6219 		case MSR_IA32_RTIT_STATUS:
6220 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6221 				continue;
6222 			break;
6223 		case MSR_IA32_RTIT_CR3_MATCH:
6224 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6225 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6226 				continue;
6227 			break;
6228 		case MSR_IA32_RTIT_OUTPUT_BASE:
6229 		case MSR_IA32_RTIT_OUTPUT_MASK:
6230 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6231 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6232 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6233 				continue;
6234 			break;
6235 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6236 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6237 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6238 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6239 				continue;
6240 			break;
6241 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6242 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6243 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6244 				continue;
6245 			break;
6246 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6247 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6248 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6249 				continue;
6250 			break;
6251 		default:
6252 			break;
6253 		}
6254 
6255 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6256 	}
6257 
6258 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6259 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6260 			continue;
6261 
6262 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6263 	}
6264 
6265 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6266 		struct kvm_msr_entry msr;
6267 
6268 		msr.index = msr_based_features_all[i];
6269 		if (kvm_get_msr_feature(&msr))
6270 			continue;
6271 
6272 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6273 	}
6274 }
6275 
6276 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6277 			   const void *v)
6278 {
6279 	int handled = 0;
6280 	int n;
6281 
6282 	do {
6283 		n = min(len, 8);
6284 		if (!(lapic_in_kernel(vcpu) &&
6285 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6286 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6287 			break;
6288 		handled += n;
6289 		addr += n;
6290 		len -= n;
6291 		v += n;
6292 	} while (len);
6293 
6294 	return handled;
6295 }
6296 
6297 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6298 {
6299 	int handled = 0;
6300 	int n;
6301 
6302 	do {
6303 		n = min(len, 8);
6304 		if (!(lapic_in_kernel(vcpu) &&
6305 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6306 					 addr, n, v))
6307 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6308 			break;
6309 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6310 		handled += n;
6311 		addr += n;
6312 		len -= n;
6313 		v += n;
6314 	} while (len);
6315 
6316 	return handled;
6317 }
6318 
6319 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6320 			struct kvm_segment *var, int seg)
6321 {
6322 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6323 }
6324 
6325 void kvm_get_segment(struct kvm_vcpu *vcpu,
6326 		     struct kvm_segment *var, int seg)
6327 {
6328 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6329 }
6330 
6331 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6332 			   struct x86_exception *exception)
6333 {
6334 	gpa_t t_gpa;
6335 
6336 	BUG_ON(!mmu_is_nested(vcpu));
6337 
6338 	/* NPT walks are always user-walks */
6339 	access |= PFERR_USER_MASK;
6340 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6341 
6342 	return t_gpa;
6343 }
6344 
6345 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6346 			      struct x86_exception *exception)
6347 {
6348 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6349 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6350 }
6351 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6352 
6353  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6354 				struct x86_exception *exception)
6355 {
6356 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6357 	access |= PFERR_FETCH_MASK;
6358 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6359 }
6360 
6361 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6362 			       struct x86_exception *exception)
6363 {
6364 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6365 	access |= PFERR_WRITE_MASK;
6366 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6367 }
6368 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6369 
6370 /* uses this to access any guest's mapped memory without checking CPL */
6371 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6372 				struct x86_exception *exception)
6373 {
6374 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6375 }
6376 
6377 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6378 				      struct kvm_vcpu *vcpu, u32 access,
6379 				      struct x86_exception *exception)
6380 {
6381 	void *data = val;
6382 	int r = X86EMUL_CONTINUE;
6383 
6384 	while (bytes) {
6385 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6386 							    exception);
6387 		unsigned offset = addr & (PAGE_SIZE-1);
6388 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6389 		int ret;
6390 
6391 		if (gpa == UNMAPPED_GVA)
6392 			return X86EMUL_PROPAGATE_FAULT;
6393 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6394 					       offset, toread);
6395 		if (ret < 0) {
6396 			r = X86EMUL_IO_NEEDED;
6397 			goto out;
6398 		}
6399 
6400 		bytes -= toread;
6401 		data += toread;
6402 		addr += toread;
6403 	}
6404 out:
6405 	return r;
6406 }
6407 
6408 /* used for instruction fetching */
6409 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6410 				gva_t addr, void *val, unsigned int bytes,
6411 				struct x86_exception *exception)
6412 {
6413 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6414 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6415 	unsigned offset;
6416 	int ret;
6417 
6418 	/* Inline kvm_read_guest_virt_helper for speed.  */
6419 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6420 						    exception);
6421 	if (unlikely(gpa == UNMAPPED_GVA))
6422 		return X86EMUL_PROPAGATE_FAULT;
6423 
6424 	offset = addr & (PAGE_SIZE-1);
6425 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6426 		bytes = (unsigned)PAGE_SIZE - offset;
6427 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6428 				       offset, bytes);
6429 	if (unlikely(ret < 0))
6430 		return X86EMUL_IO_NEEDED;
6431 
6432 	return X86EMUL_CONTINUE;
6433 }
6434 
6435 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6436 			       gva_t addr, void *val, unsigned int bytes,
6437 			       struct x86_exception *exception)
6438 {
6439 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6440 
6441 	/*
6442 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6443 	 * is returned, but our callers are not ready for that and they blindly
6444 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6445 	 * uninitialized kernel stack memory into cr2 and error code.
6446 	 */
6447 	memset(exception, 0, sizeof(*exception));
6448 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6449 					  exception);
6450 }
6451 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6452 
6453 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6454 			     gva_t addr, void *val, unsigned int bytes,
6455 			     struct x86_exception *exception, bool system)
6456 {
6457 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6458 	u32 access = 0;
6459 
6460 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6461 		access |= PFERR_USER_MASK;
6462 
6463 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6464 }
6465 
6466 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6467 		unsigned long addr, void *val, unsigned int bytes)
6468 {
6469 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6470 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6471 
6472 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6473 }
6474 
6475 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6476 				      struct kvm_vcpu *vcpu, u32 access,
6477 				      struct x86_exception *exception)
6478 {
6479 	void *data = val;
6480 	int r = X86EMUL_CONTINUE;
6481 
6482 	while (bytes) {
6483 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6484 							     access,
6485 							     exception);
6486 		unsigned offset = addr & (PAGE_SIZE-1);
6487 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6488 		int ret;
6489 
6490 		if (gpa == UNMAPPED_GVA)
6491 			return X86EMUL_PROPAGATE_FAULT;
6492 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6493 		if (ret < 0) {
6494 			r = X86EMUL_IO_NEEDED;
6495 			goto out;
6496 		}
6497 
6498 		bytes -= towrite;
6499 		data += towrite;
6500 		addr += towrite;
6501 	}
6502 out:
6503 	return r;
6504 }
6505 
6506 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6507 			      unsigned int bytes, struct x86_exception *exception,
6508 			      bool system)
6509 {
6510 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6511 	u32 access = PFERR_WRITE_MASK;
6512 
6513 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6514 		access |= PFERR_USER_MASK;
6515 
6516 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6517 					   access, exception);
6518 }
6519 
6520 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6521 				unsigned int bytes, struct x86_exception *exception)
6522 {
6523 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6524 	vcpu->arch.l1tf_flush_l1d = true;
6525 
6526 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6527 					   PFERR_WRITE_MASK, exception);
6528 }
6529 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6530 
6531 int handle_ud(struct kvm_vcpu *vcpu)
6532 {
6533 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6534 	int emul_type = EMULTYPE_TRAP_UD;
6535 	char sig[5]; /* ud2; .ascii "kvm" */
6536 	struct x86_exception e;
6537 
6538 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6539 		return 1;
6540 
6541 	if (force_emulation_prefix &&
6542 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6543 				sig, sizeof(sig), &e) == 0 &&
6544 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6545 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6546 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6547 	}
6548 
6549 	return kvm_emulate_instruction(vcpu, emul_type);
6550 }
6551 EXPORT_SYMBOL_GPL(handle_ud);
6552 
6553 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6554 			    gpa_t gpa, bool write)
6555 {
6556 	/* For APIC access vmexit */
6557 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6558 		return 1;
6559 
6560 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6561 		trace_vcpu_match_mmio(gva, gpa, write, true);
6562 		return 1;
6563 	}
6564 
6565 	return 0;
6566 }
6567 
6568 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6569 				gpa_t *gpa, struct x86_exception *exception,
6570 				bool write)
6571 {
6572 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6573 		| (write ? PFERR_WRITE_MASK : 0);
6574 
6575 	/*
6576 	 * currently PKRU is only applied to ept enabled guest so
6577 	 * there is no pkey in EPT page table for L1 guest or EPT
6578 	 * shadow page table for L2 guest.
6579 	 */
6580 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6581 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
6582 			      vcpu->arch.mmio_access, 0, access))) {
6583 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6584 					(gva & (PAGE_SIZE - 1));
6585 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6586 		return 1;
6587 	}
6588 
6589 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6590 
6591 	if (*gpa == UNMAPPED_GVA)
6592 		return -1;
6593 
6594 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6595 }
6596 
6597 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6598 			const void *val, int bytes)
6599 {
6600 	int ret;
6601 
6602 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6603 	if (ret < 0)
6604 		return 0;
6605 	kvm_page_track_write(vcpu, gpa, val, bytes);
6606 	return 1;
6607 }
6608 
6609 struct read_write_emulator_ops {
6610 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6611 				  int bytes);
6612 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6613 				  void *val, int bytes);
6614 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6615 			       int bytes, void *val);
6616 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6617 				    void *val, int bytes);
6618 	bool write;
6619 };
6620 
6621 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6622 {
6623 	if (vcpu->mmio_read_completed) {
6624 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6625 			       vcpu->mmio_fragments[0].gpa, val);
6626 		vcpu->mmio_read_completed = 0;
6627 		return 1;
6628 	}
6629 
6630 	return 0;
6631 }
6632 
6633 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6634 			void *val, int bytes)
6635 {
6636 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6637 }
6638 
6639 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6640 			 void *val, int bytes)
6641 {
6642 	return emulator_write_phys(vcpu, gpa, val, bytes);
6643 }
6644 
6645 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6646 {
6647 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6648 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6649 }
6650 
6651 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6652 			  void *val, int bytes)
6653 {
6654 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6655 	return X86EMUL_IO_NEEDED;
6656 }
6657 
6658 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6659 			   void *val, int bytes)
6660 {
6661 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6662 
6663 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6664 	return X86EMUL_CONTINUE;
6665 }
6666 
6667 static const struct read_write_emulator_ops read_emultor = {
6668 	.read_write_prepare = read_prepare,
6669 	.read_write_emulate = read_emulate,
6670 	.read_write_mmio = vcpu_mmio_read,
6671 	.read_write_exit_mmio = read_exit_mmio,
6672 };
6673 
6674 static const struct read_write_emulator_ops write_emultor = {
6675 	.read_write_emulate = write_emulate,
6676 	.read_write_mmio = write_mmio,
6677 	.read_write_exit_mmio = write_exit_mmio,
6678 	.write = true,
6679 };
6680 
6681 static int emulator_read_write_onepage(unsigned long addr, void *val,
6682 				       unsigned int bytes,
6683 				       struct x86_exception *exception,
6684 				       struct kvm_vcpu *vcpu,
6685 				       const struct read_write_emulator_ops *ops)
6686 {
6687 	gpa_t gpa;
6688 	int handled, ret;
6689 	bool write = ops->write;
6690 	struct kvm_mmio_fragment *frag;
6691 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6692 
6693 	/*
6694 	 * If the exit was due to a NPF we may already have a GPA.
6695 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6696 	 * Note, this cannot be used on string operations since string
6697 	 * operation using rep will only have the initial GPA from the NPF
6698 	 * occurred.
6699 	 */
6700 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6701 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6702 		gpa = ctxt->gpa_val;
6703 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6704 	} else {
6705 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6706 		if (ret < 0)
6707 			return X86EMUL_PROPAGATE_FAULT;
6708 	}
6709 
6710 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6711 		return X86EMUL_CONTINUE;
6712 
6713 	/*
6714 	 * Is this MMIO handled locally?
6715 	 */
6716 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6717 	if (handled == bytes)
6718 		return X86EMUL_CONTINUE;
6719 
6720 	gpa += handled;
6721 	bytes -= handled;
6722 	val += handled;
6723 
6724 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6725 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6726 	frag->gpa = gpa;
6727 	frag->data = val;
6728 	frag->len = bytes;
6729 	return X86EMUL_CONTINUE;
6730 }
6731 
6732 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6733 			unsigned long addr,
6734 			void *val, unsigned int bytes,
6735 			struct x86_exception *exception,
6736 			const struct read_write_emulator_ops *ops)
6737 {
6738 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6739 	gpa_t gpa;
6740 	int rc;
6741 
6742 	if (ops->read_write_prepare &&
6743 		  ops->read_write_prepare(vcpu, val, bytes))
6744 		return X86EMUL_CONTINUE;
6745 
6746 	vcpu->mmio_nr_fragments = 0;
6747 
6748 	/* Crossing a page boundary? */
6749 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6750 		int now;
6751 
6752 		now = -addr & ~PAGE_MASK;
6753 		rc = emulator_read_write_onepage(addr, val, now, exception,
6754 						 vcpu, ops);
6755 
6756 		if (rc != X86EMUL_CONTINUE)
6757 			return rc;
6758 		addr += now;
6759 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6760 			addr = (u32)addr;
6761 		val += now;
6762 		bytes -= now;
6763 	}
6764 
6765 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6766 					 vcpu, ops);
6767 	if (rc != X86EMUL_CONTINUE)
6768 		return rc;
6769 
6770 	if (!vcpu->mmio_nr_fragments)
6771 		return rc;
6772 
6773 	gpa = vcpu->mmio_fragments[0].gpa;
6774 
6775 	vcpu->mmio_needed = 1;
6776 	vcpu->mmio_cur_fragment = 0;
6777 
6778 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6779 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6780 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6781 	vcpu->run->mmio.phys_addr = gpa;
6782 
6783 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6784 }
6785 
6786 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6787 				  unsigned long addr,
6788 				  void *val,
6789 				  unsigned int bytes,
6790 				  struct x86_exception *exception)
6791 {
6792 	return emulator_read_write(ctxt, addr, val, bytes,
6793 				   exception, &read_emultor);
6794 }
6795 
6796 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6797 			    unsigned long addr,
6798 			    const void *val,
6799 			    unsigned int bytes,
6800 			    struct x86_exception *exception)
6801 {
6802 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6803 				   exception, &write_emultor);
6804 }
6805 
6806 #define CMPXCHG_TYPE(t, ptr, old, new) \
6807 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6808 
6809 #ifdef CONFIG_X86_64
6810 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6811 #else
6812 #  define CMPXCHG64(ptr, old, new) \
6813 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6814 #endif
6815 
6816 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6817 				     unsigned long addr,
6818 				     const void *old,
6819 				     const void *new,
6820 				     unsigned int bytes,
6821 				     struct x86_exception *exception)
6822 {
6823 	struct kvm_host_map map;
6824 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6825 	u64 page_line_mask;
6826 	gpa_t gpa;
6827 	char *kaddr;
6828 	bool exchanged;
6829 
6830 	/* guests cmpxchg8b have to be emulated atomically */
6831 	if (bytes > 8 || (bytes & (bytes - 1)))
6832 		goto emul_write;
6833 
6834 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6835 
6836 	if (gpa == UNMAPPED_GVA ||
6837 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6838 		goto emul_write;
6839 
6840 	/*
6841 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6842 	 * enabled in the host and the access splits a cache line.
6843 	 */
6844 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6845 		page_line_mask = ~(cache_line_size() - 1);
6846 	else
6847 		page_line_mask = PAGE_MASK;
6848 
6849 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6850 		goto emul_write;
6851 
6852 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6853 		goto emul_write;
6854 
6855 	kaddr = map.hva + offset_in_page(gpa);
6856 
6857 	switch (bytes) {
6858 	case 1:
6859 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6860 		break;
6861 	case 2:
6862 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6863 		break;
6864 	case 4:
6865 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6866 		break;
6867 	case 8:
6868 		exchanged = CMPXCHG64(kaddr, old, new);
6869 		break;
6870 	default:
6871 		BUG();
6872 	}
6873 
6874 	kvm_vcpu_unmap(vcpu, &map, true);
6875 
6876 	if (!exchanged)
6877 		return X86EMUL_CMPXCHG_FAILED;
6878 
6879 	kvm_page_track_write(vcpu, gpa, new, bytes);
6880 
6881 	return X86EMUL_CONTINUE;
6882 
6883 emul_write:
6884 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6885 
6886 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6887 }
6888 
6889 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6890 {
6891 	int r = 0, i;
6892 
6893 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6894 		if (vcpu->arch.pio.in)
6895 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6896 					    vcpu->arch.pio.size, pd);
6897 		else
6898 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6899 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6900 					     pd);
6901 		if (r)
6902 			break;
6903 		pd += vcpu->arch.pio.size;
6904 	}
6905 	return r;
6906 }
6907 
6908 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6909 			       unsigned short port, void *val,
6910 			       unsigned int count, bool in)
6911 {
6912 	vcpu->arch.pio.port = port;
6913 	vcpu->arch.pio.in = in;
6914 	vcpu->arch.pio.count  = count;
6915 	vcpu->arch.pio.size = size;
6916 
6917 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6918 		vcpu->arch.pio.count = 0;
6919 		return 1;
6920 	}
6921 
6922 	vcpu->run->exit_reason = KVM_EXIT_IO;
6923 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6924 	vcpu->run->io.size = size;
6925 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6926 	vcpu->run->io.count = count;
6927 	vcpu->run->io.port = port;
6928 
6929 	return 0;
6930 }
6931 
6932 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6933 			   unsigned short port, void *val, unsigned int count)
6934 {
6935 	int ret;
6936 
6937 	if (vcpu->arch.pio.count)
6938 		goto data_avail;
6939 
6940 	memset(vcpu->arch.pio_data, 0, size * count);
6941 
6942 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6943 	if (ret) {
6944 data_avail:
6945 		memcpy(val, vcpu->arch.pio_data, size * count);
6946 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6947 		vcpu->arch.pio.count = 0;
6948 		return 1;
6949 	}
6950 
6951 	return 0;
6952 }
6953 
6954 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6955 				    int size, unsigned short port, void *val,
6956 				    unsigned int count)
6957 {
6958 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6959 
6960 }
6961 
6962 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6963 			    unsigned short port, const void *val,
6964 			    unsigned int count)
6965 {
6966 	memcpy(vcpu->arch.pio_data, val, size * count);
6967 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6968 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6969 }
6970 
6971 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6972 				     int size, unsigned short port,
6973 				     const void *val, unsigned int count)
6974 {
6975 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6976 }
6977 
6978 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6979 {
6980 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6981 }
6982 
6983 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6984 {
6985 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6986 }
6987 
6988 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6989 {
6990 	if (!need_emulate_wbinvd(vcpu))
6991 		return X86EMUL_CONTINUE;
6992 
6993 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
6994 		int cpu = get_cpu();
6995 
6996 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6997 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6998 				wbinvd_ipi, NULL, 1);
6999 		put_cpu();
7000 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7001 	} else
7002 		wbinvd();
7003 	return X86EMUL_CONTINUE;
7004 }
7005 
7006 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7007 {
7008 	kvm_emulate_wbinvd_noskip(vcpu);
7009 	return kvm_skip_emulated_instruction(vcpu);
7010 }
7011 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7012 
7013 
7014 
7015 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7016 {
7017 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7018 }
7019 
7020 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7021 			    unsigned long *dest)
7022 {
7023 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7024 }
7025 
7026 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7027 			   unsigned long value)
7028 {
7029 
7030 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7031 }
7032 
7033 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7034 {
7035 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7036 }
7037 
7038 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7039 {
7040 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7041 	unsigned long value;
7042 
7043 	switch (cr) {
7044 	case 0:
7045 		value = kvm_read_cr0(vcpu);
7046 		break;
7047 	case 2:
7048 		value = vcpu->arch.cr2;
7049 		break;
7050 	case 3:
7051 		value = kvm_read_cr3(vcpu);
7052 		break;
7053 	case 4:
7054 		value = kvm_read_cr4(vcpu);
7055 		break;
7056 	case 8:
7057 		value = kvm_get_cr8(vcpu);
7058 		break;
7059 	default:
7060 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7061 		return 0;
7062 	}
7063 
7064 	return value;
7065 }
7066 
7067 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7068 {
7069 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7070 	int res = 0;
7071 
7072 	switch (cr) {
7073 	case 0:
7074 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7075 		break;
7076 	case 2:
7077 		vcpu->arch.cr2 = val;
7078 		break;
7079 	case 3:
7080 		res = kvm_set_cr3(vcpu, val);
7081 		break;
7082 	case 4:
7083 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7084 		break;
7085 	case 8:
7086 		res = kvm_set_cr8(vcpu, val);
7087 		break;
7088 	default:
7089 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7090 		res = -1;
7091 	}
7092 
7093 	return res;
7094 }
7095 
7096 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7097 {
7098 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7099 }
7100 
7101 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7102 {
7103 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7104 }
7105 
7106 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7107 {
7108 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7109 }
7110 
7111 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7112 {
7113 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7114 }
7115 
7116 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7117 {
7118 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7119 }
7120 
7121 static unsigned long emulator_get_cached_segment_base(
7122 	struct x86_emulate_ctxt *ctxt, int seg)
7123 {
7124 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7125 }
7126 
7127 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7128 				 struct desc_struct *desc, u32 *base3,
7129 				 int seg)
7130 {
7131 	struct kvm_segment var;
7132 
7133 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7134 	*selector = var.selector;
7135 
7136 	if (var.unusable) {
7137 		memset(desc, 0, sizeof(*desc));
7138 		if (base3)
7139 			*base3 = 0;
7140 		return false;
7141 	}
7142 
7143 	if (var.g)
7144 		var.limit >>= 12;
7145 	set_desc_limit(desc, var.limit);
7146 	set_desc_base(desc, (unsigned long)var.base);
7147 #ifdef CONFIG_X86_64
7148 	if (base3)
7149 		*base3 = var.base >> 32;
7150 #endif
7151 	desc->type = var.type;
7152 	desc->s = var.s;
7153 	desc->dpl = var.dpl;
7154 	desc->p = var.present;
7155 	desc->avl = var.avl;
7156 	desc->l = var.l;
7157 	desc->d = var.db;
7158 	desc->g = var.g;
7159 
7160 	return true;
7161 }
7162 
7163 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7164 				 struct desc_struct *desc, u32 base3,
7165 				 int seg)
7166 {
7167 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7168 	struct kvm_segment var;
7169 
7170 	var.selector = selector;
7171 	var.base = get_desc_base(desc);
7172 #ifdef CONFIG_X86_64
7173 	var.base |= ((u64)base3) << 32;
7174 #endif
7175 	var.limit = get_desc_limit(desc);
7176 	if (desc->g)
7177 		var.limit = (var.limit << 12) | 0xfff;
7178 	var.type = desc->type;
7179 	var.dpl = desc->dpl;
7180 	var.db = desc->d;
7181 	var.s = desc->s;
7182 	var.l = desc->l;
7183 	var.g = desc->g;
7184 	var.avl = desc->avl;
7185 	var.present = desc->p;
7186 	var.unusable = !var.present;
7187 	var.padding = 0;
7188 
7189 	kvm_set_segment(vcpu, &var, seg);
7190 	return;
7191 }
7192 
7193 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7194 			    u32 msr_index, u64 *pdata)
7195 {
7196 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7197 	int r;
7198 
7199 	r = kvm_get_msr(vcpu, msr_index, pdata);
7200 
7201 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7202 		/* Bounce to user space */
7203 		return X86EMUL_IO_NEEDED;
7204 	}
7205 
7206 	return r;
7207 }
7208 
7209 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7210 			    u32 msr_index, u64 data)
7211 {
7212 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7213 	int r;
7214 
7215 	r = kvm_set_msr(vcpu, msr_index, data);
7216 
7217 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7218 		/* Bounce to user space */
7219 		return X86EMUL_IO_NEEDED;
7220 	}
7221 
7222 	return r;
7223 }
7224 
7225 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7226 {
7227 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7228 
7229 	return vcpu->arch.smbase;
7230 }
7231 
7232 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7233 {
7234 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7235 
7236 	vcpu->arch.smbase = smbase;
7237 }
7238 
7239 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7240 			      u32 pmc)
7241 {
7242 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7243 }
7244 
7245 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7246 			     u32 pmc, u64 *pdata)
7247 {
7248 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7249 }
7250 
7251 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7252 {
7253 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7254 }
7255 
7256 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7257 			      struct x86_instruction_info *info,
7258 			      enum x86_intercept_stage stage)
7259 {
7260 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7261 					    &ctxt->exception);
7262 }
7263 
7264 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7265 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7266 			      bool exact_only)
7267 {
7268 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7269 }
7270 
7271 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7272 {
7273 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7274 }
7275 
7276 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7277 {
7278 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7279 }
7280 
7281 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7282 {
7283 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7284 }
7285 
7286 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7287 {
7288 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7289 }
7290 
7291 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7292 {
7293 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7294 }
7295 
7296 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7297 {
7298 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7299 }
7300 
7301 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7302 {
7303 	return emul_to_vcpu(ctxt)->arch.hflags;
7304 }
7305 
7306 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7307 {
7308 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7309 
7310 	kvm_smm_changed(vcpu, false);
7311 }
7312 
7313 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7314 				  const char *smstate)
7315 {
7316 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7317 }
7318 
7319 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7320 {
7321 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7322 }
7323 
7324 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7325 {
7326 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7327 }
7328 
7329 static const struct x86_emulate_ops emulate_ops = {
7330 	.read_gpr            = emulator_read_gpr,
7331 	.write_gpr           = emulator_write_gpr,
7332 	.read_std            = emulator_read_std,
7333 	.write_std           = emulator_write_std,
7334 	.read_phys           = kvm_read_guest_phys_system,
7335 	.fetch               = kvm_fetch_guest_virt,
7336 	.read_emulated       = emulator_read_emulated,
7337 	.write_emulated      = emulator_write_emulated,
7338 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7339 	.invlpg              = emulator_invlpg,
7340 	.pio_in_emulated     = emulator_pio_in_emulated,
7341 	.pio_out_emulated    = emulator_pio_out_emulated,
7342 	.get_segment         = emulator_get_segment,
7343 	.set_segment         = emulator_set_segment,
7344 	.get_cached_segment_base = emulator_get_cached_segment_base,
7345 	.get_gdt             = emulator_get_gdt,
7346 	.get_idt	     = emulator_get_idt,
7347 	.set_gdt             = emulator_set_gdt,
7348 	.set_idt	     = emulator_set_idt,
7349 	.get_cr              = emulator_get_cr,
7350 	.set_cr              = emulator_set_cr,
7351 	.cpl                 = emulator_get_cpl,
7352 	.get_dr              = emulator_get_dr,
7353 	.set_dr              = emulator_set_dr,
7354 	.get_smbase          = emulator_get_smbase,
7355 	.set_smbase          = emulator_set_smbase,
7356 	.set_msr             = emulator_set_msr,
7357 	.get_msr             = emulator_get_msr,
7358 	.check_pmc	     = emulator_check_pmc,
7359 	.read_pmc            = emulator_read_pmc,
7360 	.halt                = emulator_halt,
7361 	.wbinvd              = emulator_wbinvd,
7362 	.fix_hypercall       = emulator_fix_hypercall,
7363 	.intercept           = emulator_intercept,
7364 	.get_cpuid           = emulator_get_cpuid,
7365 	.guest_has_long_mode = emulator_guest_has_long_mode,
7366 	.guest_has_movbe     = emulator_guest_has_movbe,
7367 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7368 	.set_nmi_mask        = emulator_set_nmi_mask,
7369 	.get_hflags          = emulator_get_hflags,
7370 	.exiting_smm         = emulator_exiting_smm,
7371 	.leave_smm           = emulator_leave_smm,
7372 	.triple_fault        = emulator_triple_fault,
7373 	.set_xcr             = emulator_set_xcr,
7374 };
7375 
7376 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7377 {
7378 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7379 	/*
7380 	 * an sti; sti; sequence only disable interrupts for the first
7381 	 * instruction. So, if the last instruction, be it emulated or
7382 	 * not, left the system with the INT_STI flag enabled, it
7383 	 * means that the last instruction is an sti. We should not
7384 	 * leave the flag on in this case. The same goes for mov ss
7385 	 */
7386 	if (int_shadow & mask)
7387 		mask = 0;
7388 	if (unlikely(int_shadow || mask)) {
7389 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7390 		if (!mask)
7391 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7392 	}
7393 }
7394 
7395 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7396 {
7397 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7398 	if (ctxt->exception.vector == PF_VECTOR)
7399 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7400 
7401 	if (ctxt->exception.error_code_valid)
7402 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7403 				      ctxt->exception.error_code);
7404 	else
7405 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7406 	return false;
7407 }
7408 
7409 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7410 {
7411 	struct x86_emulate_ctxt *ctxt;
7412 
7413 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7414 	if (!ctxt) {
7415 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7416 		return NULL;
7417 	}
7418 
7419 	ctxt->vcpu = vcpu;
7420 	ctxt->ops = &emulate_ops;
7421 	vcpu->arch.emulate_ctxt = ctxt;
7422 
7423 	return ctxt;
7424 }
7425 
7426 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7427 {
7428 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7429 	int cs_db, cs_l;
7430 
7431 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7432 
7433 	ctxt->gpa_available = false;
7434 	ctxt->eflags = kvm_get_rflags(vcpu);
7435 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7436 
7437 	ctxt->eip = kvm_rip_read(vcpu);
7438 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7439 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7440 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7441 		     cs_db				? X86EMUL_MODE_PROT32 :
7442 							  X86EMUL_MODE_PROT16;
7443 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7444 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7445 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7446 
7447 	ctxt->interruptibility = 0;
7448 	ctxt->have_exception = false;
7449 	ctxt->exception.vector = -1;
7450 	ctxt->perm_ok = false;
7451 
7452 	init_decode_cache(ctxt);
7453 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7454 }
7455 
7456 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7457 {
7458 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7459 	int ret;
7460 
7461 	init_emulate_ctxt(vcpu);
7462 
7463 	ctxt->op_bytes = 2;
7464 	ctxt->ad_bytes = 2;
7465 	ctxt->_eip = ctxt->eip + inc_eip;
7466 	ret = emulate_int_real(ctxt, irq);
7467 
7468 	if (ret != X86EMUL_CONTINUE) {
7469 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7470 	} else {
7471 		ctxt->eip = ctxt->_eip;
7472 		kvm_rip_write(vcpu, ctxt->eip);
7473 		kvm_set_rflags(vcpu, ctxt->eflags);
7474 	}
7475 }
7476 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7477 
7478 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7479 {
7480 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7481 	u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7482 	struct kvm_run *run = vcpu->run;
7483 
7484 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7485 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7486 	run->emulation_failure.ndata = 0;
7487 	run->emulation_failure.flags = 0;
7488 
7489 	if (insn_size) {
7490 		run->emulation_failure.ndata = 3;
7491 		run->emulation_failure.flags |=
7492 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7493 		run->emulation_failure.insn_size = insn_size;
7494 		memset(run->emulation_failure.insn_bytes, 0x90,
7495 		       sizeof(run->emulation_failure.insn_bytes));
7496 		memcpy(run->emulation_failure.insn_bytes,
7497 		       ctxt->fetch.data, insn_size);
7498 	}
7499 }
7500 
7501 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7502 {
7503 	struct kvm *kvm = vcpu->kvm;
7504 
7505 	++vcpu->stat.insn_emulation_fail;
7506 	trace_kvm_emulate_insn_failed(vcpu);
7507 
7508 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7509 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7510 		return 1;
7511 	}
7512 
7513 	if (kvm->arch.exit_on_emulation_error ||
7514 	    (emulation_type & EMULTYPE_SKIP)) {
7515 		prepare_emulation_failure_exit(vcpu);
7516 		return 0;
7517 	}
7518 
7519 	kvm_queue_exception(vcpu, UD_VECTOR);
7520 
7521 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7522 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7523 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7524 		vcpu->run->internal.ndata = 0;
7525 		return 0;
7526 	}
7527 
7528 	return 1;
7529 }
7530 
7531 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7532 				  bool write_fault_to_shadow_pgtable,
7533 				  int emulation_type)
7534 {
7535 	gpa_t gpa = cr2_or_gpa;
7536 	kvm_pfn_t pfn;
7537 
7538 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7539 		return false;
7540 
7541 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7542 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7543 		return false;
7544 
7545 	if (!vcpu->arch.mmu->direct_map) {
7546 		/*
7547 		 * Write permission should be allowed since only
7548 		 * write access need to be emulated.
7549 		 */
7550 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7551 
7552 		/*
7553 		 * If the mapping is invalid in guest, let cpu retry
7554 		 * it to generate fault.
7555 		 */
7556 		if (gpa == UNMAPPED_GVA)
7557 			return true;
7558 	}
7559 
7560 	/*
7561 	 * Do not retry the unhandleable instruction if it faults on the
7562 	 * readonly host memory, otherwise it will goto a infinite loop:
7563 	 * retry instruction -> write #PF -> emulation fail -> retry
7564 	 * instruction -> ...
7565 	 */
7566 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7567 
7568 	/*
7569 	 * If the instruction failed on the error pfn, it can not be fixed,
7570 	 * report the error to userspace.
7571 	 */
7572 	if (is_error_noslot_pfn(pfn))
7573 		return false;
7574 
7575 	kvm_release_pfn_clean(pfn);
7576 
7577 	/* The instructions are well-emulated on direct mmu. */
7578 	if (vcpu->arch.mmu->direct_map) {
7579 		unsigned int indirect_shadow_pages;
7580 
7581 		write_lock(&vcpu->kvm->mmu_lock);
7582 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7583 		write_unlock(&vcpu->kvm->mmu_lock);
7584 
7585 		if (indirect_shadow_pages)
7586 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7587 
7588 		return true;
7589 	}
7590 
7591 	/*
7592 	 * if emulation was due to access to shadowed page table
7593 	 * and it failed try to unshadow page and re-enter the
7594 	 * guest to let CPU execute the instruction.
7595 	 */
7596 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7597 
7598 	/*
7599 	 * If the access faults on its page table, it can not
7600 	 * be fixed by unprotecting shadow page and it should
7601 	 * be reported to userspace.
7602 	 */
7603 	return !write_fault_to_shadow_pgtable;
7604 }
7605 
7606 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7607 			      gpa_t cr2_or_gpa,  int emulation_type)
7608 {
7609 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7610 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7611 
7612 	last_retry_eip = vcpu->arch.last_retry_eip;
7613 	last_retry_addr = vcpu->arch.last_retry_addr;
7614 
7615 	/*
7616 	 * If the emulation is caused by #PF and it is non-page_table
7617 	 * writing instruction, it means the VM-EXIT is caused by shadow
7618 	 * page protected, we can zap the shadow page and retry this
7619 	 * instruction directly.
7620 	 *
7621 	 * Note: if the guest uses a non-page-table modifying instruction
7622 	 * on the PDE that points to the instruction, then we will unmap
7623 	 * the instruction and go to an infinite loop. So, we cache the
7624 	 * last retried eip and the last fault address, if we meet the eip
7625 	 * and the address again, we can break out of the potential infinite
7626 	 * loop.
7627 	 */
7628 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7629 
7630 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7631 		return false;
7632 
7633 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7634 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7635 		return false;
7636 
7637 	if (x86_page_table_writing_insn(ctxt))
7638 		return false;
7639 
7640 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7641 		return false;
7642 
7643 	vcpu->arch.last_retry_eip = ctxt->eip;
7644 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7645 
7646 	if (!vcpu->arch.mmu->direct_map)
7647 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7648 
7649 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7650 
7651 	return true;
7652 }
7653 
7654 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7655 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7656 
7657 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7658 {
7659 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7660 
7661 	if (entering_smm) {
7662 		vcpu->arch.hflags |= HF_SMM_MASK;
7663 	} else {
7664 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7665 
7666 		/* Process a latched INIT or SMI, if any.  */
7667 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7668 
7669 		/*
7670 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7671 		 * on SMM exit we still need to reload them from
7672 		 * guest memory
7673 		 */
7674 		vcpu->arch.pdptrs_from_userspace = false;
7675 	}
7676 
7677 	kvm_mmu_reset_context(vcpu);
7678 }
7679 
7680 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7681 				unsigned long *db)
7682 {
7683 	u32 dr6 = 0;
7684 	int i;
7685 	u32 enable, rwlen;
7686 
7687 	enable = dr7;
7688 	rwlen = dr7 >> 16;
7689 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7690 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7691 			dr6 |= (1 << i);
7692 	return dr6;
7693 }
7694 
7695 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7696 {
7697 	struct kvm_run *kvm_run = vcpu->run;
7698 
7699 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7700 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7701 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7702 		kvm_run->debug.arch.exception = DB_VECTOR;
7703 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7704 		return 0;
7705 	}
7706 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7707 	return 1;
7708 }
7709 
7710 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7711 {
7712 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7713 	int r;
7714 
7715 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7716 	if (unlikely(!r))
7717 		return 0;
7718 
7719 	/*
7720 	 * rflags is the old, "raw" value of the flags.  The new value has
7721 	 * not been saved yet.
7722 	 *
7723 	 * This is correct even for TF set by the guest, because "the
7724 	 * processor will not generate this exception after the instruction
7725 	 * that sets the TF flag".
7726 	 */
7727 	if (unlikely(rflags & X86_EFLAGS_TF))
7728 		r = kvm_vcpu_do_singlestep(vcpu);
7729 	return r;
7730 }
7731 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7732 
7733 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7734 {
7735 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7736 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7737 		struct kvm_run *kvm_run = vcpu->run;
7738 		unsigned long eip = kvm_get_linear_rip(vcpu);
7739 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7740 					   vcpu->arch.guest_debug_dr7,
7741 					   vcpu->arch.eff_db);
7742 
7743 		if (dr6 != 0) {
7744 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7745 			kvm_run->debug.arch.pc = eip;
7746 			kvm_run->debug.arch.exception = DB_VECTOR;
7747 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7748 			*r = 0;
7749 			return true;
7750 		}
7751 	}
7752 
7753 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7754 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7755 		unsigned long eip = kvm_get_linear_rip(vcpu);
7756 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7757 					   vcpu->arch.dr7,
7758 					   vcpu->arch.db);
7759 
7760 		if (dr6 != 0) {
7761 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7762 			*r = 1;
7763 			return true;
7764 		}
7765 	}
7766 
7767 	return false;
7768 }
7769 
7770 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7771 {
7772 	switch (ctxt->opcode_len) {
7773 	case 1:
7774 		switch (ctxt->b) {
7775 		case 0xe4:	/* IN */
7776 		case 0xe5:
7777 		case 0xec:
7778 		case 0xed:
7779 		case 0xe6:	/* OUT */
7780 		case 0xe7:
7781 		case 0xee:
7782 		case 0xef:
7783 		case 0x6c:	/* INS */
7784 		case 0x6d:
7785 		case 0x6e:	/* OUTS */
7786 		case 0x6f:
7787 			return true;
7788 		}
7789 		break;
7790 	case 2:
7791 		switch (ctxt->b) {
7792 		case 0x33:	/* RDPMC */
7793 			return true;
7794 		}
7795 		break;
7796 	}
7797 
7798 	return false;
7799 }
7800 
7801 /*
7802  * Decode to be emulated instruction. Return EMULATION_OK if success.
7803  */
7804 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7805 				    void *insn, int insn_len)
7806 {
7807 	int r = EMULATION_OK;
7808 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7809 
7810 	init_emulate_ctxt(vcpu);
7811 
7812 	/*
7813 	 * We will reenter on the same instruction since we do not set
7814 	 * complete_userspace_io. This does not handle watchpoints yet,
7815 	 * those would be handled in the emulate_ops.
7816 	 */
7817 	if (!(emulation_type & EMULTYPE_SKIP) &&
7818 	    kvm_vcpu_check_breakpoint(vcpu, &r))
7819 		return r;
7820 
7821 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7822 
7823 	trace_kvm_emulate_insn_start(vcpu);
7824 	++vcpu->stat.insn_emulation;
7825 
7826 	return r;
7827 }
7828 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7829 
7830 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7831 			    int emulation_type, void *insn, int insn_len)
7832 {
7833 	int r;
7834 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7835 	bool writeback = true;
7836 	bool write_fault_to_spt;
7837 
7838 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7839 		return 1;
7840 
7841 	vcpu->arch.l1tf_flush_l1d = true;
7842 
7843 	/*
7844 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7845 	 * never reused.
7846 	 */
7847 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7848 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7849 
7850 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7851 		kvm_clear_exception_queue(vcpu);
7852 
7853 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
7854 						    insn, insn_len);
7855 		if (r != EMULATION_OK)  {
7856 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7857 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7858 				kvm_queue_exception(vcpu, UD_VECTOR);
7859 				return 1;
7860 			}
7861 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7862 						  write_fault_to_spt,
7863 						  emulation_type))
7864 				return 1;
7865 			if (ctxt->have_exception) {
7866 				/*
7867 				 * #UD should result in just EMULATION_FAILED, and trap-like
7868 				 * exception should not be encountered during decode.
7869 				 */
7870 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7871 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7872 				inject_emulated_exception(vcpu);
7873 				return 1;
7874 			}
7875 			return handle_emulation_failure(vcpu, emulation_type);
7876 		}
7877 	}
7878 
7879 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7880 	    !is_vmware_backdoor_opcode(ctxt)) {
7881 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7882 		return 1;
7883 	}
7884 
7885 	/*
7886 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7887 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7888 	 * updating interruptibility state and injecting single-step #DBs.
7889 	 */
7890 	if (emulation_type & EMULTYPE_SKIP) {
7891 		kvm_rip_write(vcpu, ctxt->_eip);
7892 		if (ctxt->eflags & X86_EFLAGS_RF)
7893 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7894 		return 1;
7895 	}
7896 
7897 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7898 		return 1;
7899 
7900 	/* this is needed for vmware backdoor interface to work since it
7901 	   changes registers values  during IO operation */
7902 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7903 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7904 		emulator_invalidate_register_cache(ctxt);
7905 	}
7906 
7907 restart:
7908 	if (emulation_type & EMULTYPE_PF) {
7909 		/* Save the faulting GPA (cr2) in the address field */
7910 		ctxt->exception.address = cr2_or_gpa;
7911 
7912 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7913 		if (vcpu->arch.mmu->direct_map) {
7914 			ctxt->gpa_available = true;
7915 			ctxt->gpa_val = cr2_or_gpa;
7916 		}
7917 	} else {
7918 		/* Sanitize the address out of an abundance of paranoia. */
7919 		ctxt->exception.address = 0;
7920 	}
7921 
7922 	r = x86_emulate_insn(ctxt);
7923 
7924 	if (r == EMULATION_INTERCEPTED)
7925 		return 1;
7926 
7927 	if (r == EMULATION_FAILED) {
7928 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7929 					emulation_type))
7930 			return 1;
7931 
7932 		return handle_emulation_failure(vcpu, emulation_type);
7933 	}
7934 
7935 	if (ctxt->have_exception) {
7936 		r = 1;
7937 		if (inject_emulated_exception(vcpu))
7938 			return r;
7939 	} else if (vcpu->arch.pio.count) {
7940 		if (!vcpu->arch.pio.in) {
7941 			/* FIXME: return into emulator if single-stepping.  */
7942 			vcpu->arch.pio.count = 0;
7943 		} else {
7944 			writeback = false;
7945 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7946 		}
7947 		r = 0;
7948 	} else if (vcpu->mmio_needed) {
7949 		++vcpu->stat.mmio_exits;
7950 
7951 		if (!vcpu->mmio_is_write)
7952 			writeback = false;
7953 		r = 0;
7954 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7955 	} else if (r == EMULATION_RESTART)
7956 		goto restart;
7957 	else
7958 		r = 1;
7959 
7960 	if (writeback) {
7961 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7962 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7963 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7964 		if (!ctxt->have_exception ||
7965 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7966 			kvm_rip_write(vcpu, ctxt->eip);
7967 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7968 				r = kvm_vcpu_do_singlestep(vcpu);
7969 			if (kvm_x86_ops.update_emulated_instruction)
7970 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
7971 			__kvm_set_rflags(vcpu, ctxt->eflags);
7972 		}
7973 
7974 		/*
7975 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7976 		 * do nothing, and it will be requested again as soon as
7977 		 * the shadow expires.  But we still need to check here,
7978 		 * because POPF has no interrupt shadow.
7979 		 */
7980 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7981 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7982 	} else
7983 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7984 
7985 	return r;
7986 }
7987 
7988 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7989 {
7990 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7991 }
7992 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7993 
7994 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7995 					void *insn, int insn_len)
7996 {
7997 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7998 }
7999 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8000 
8001 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8002 {
8003 	vcpu->arch.pio.count = 0;
8004 	return 1;
8005 }
8006 
8007 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8008 {
8009 	vcpu->arch.pio.count = 0;
8010 
8011 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8012 		return 1;
8013 
8014 	return kvm_skip_emulated_instruction(vcpu);
8015 }
8016 
8017 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8018 			    unsigned short port)
8019 {
8020 	unsigned long val = kvm_rax_read(vcpu);
8021 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8022 
8023 	if (ret)
8024 		return ret;
8025 
8026 	/*
8027 	 * Workaround userspace that relies on old KVM behavior of %rip being
8028 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8029 	 */
8030 	if (port == 0x7e &&
8031 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8032 		vcpu->arch.complete_userspace_io =
8033 			complete_fast_pio_out_port_0x7e;
8034 		kvm_skip_emulated_instruction(vcpu);
8035 	} else {
8036 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8037 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8038 	}
8039 	return 0;
8040 }
8041 
8042 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8043 {
8044 	unsigned long val;
8045 
8046 	/* We should only ever be called with arch.pio.count equal to 1 */
8047 	BUG_ON(vcpu->arch.pio.count != 1);
8048 
8049 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8050 		vcpu->arch.pio.count = 0;
8051 		return 1;
8052 	}
8053 
8054 	/* For size less than 4 we merge, else we zero extend */
8055 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8056 
8057 	/*
8058 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8059 	 * the copy and tracing
8060 	 */
8061 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8062 	kvm_rax_write(vcpu, val);
8063 
8064 	return kvm_skip_emulated_instruction(vcpu);
8065 }
8066 
8067 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8068 			   unsigned short port)
8069 {
8070 	unsigned long val;
8071 	int ret;
8072 
8073 	/* For size less than 4 we merge, else we zero extend */
8074 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8075 
8076 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8077 	if (ret) {
8078 		kvm_rax_write(vcpu, val);
8079 		return ret;
8080 	}
8081 
8082 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8083 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8084 
8085 	return 0;
8086 }
8087 
8088 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8089 {
8090 	int ret;
8091 
8092 	if (in)
8093 		ret = kvm_fast_pio_in(vcpu, size, port);
8094 	else
8095 		ret = kvm_fast_pio_out(vcpu, size, port);
8096 	return ret && kvm_skip_emulated_instruction(vcpu);
8097 }
8098 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8099 
8100 static int kvmclock_cpu_down_prep(unsigned int cpu)
8101 {
8102 	__this_cpu_write(cpu_tsc_khz, 0);
8103 	return 0;
8104 }
8105 
8106 static void tsc_khz_changed(void *data)
8107 {
8108 	struct cpufreq_freqs *freq = data;
8109 	unsigned long khz = 0;
8110 
8111 	if (data)
8112 		khz = freq->new;
8113 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8114 		khz = cpufreq_quick_get(raw_smp_processor_id());
8115 	if (!khz)
8116 		khz = tsc_khz;
8117 	__this_cpu_write(cpu_tsc_khz, khz);
8118 }
8119 
8120 #ifdef CONFIG_X86_64
8121 static void kvm_hyperv_tsc_notifier(void)
8122 {
8123 	struct kvm *kvm;
8124 	struct kvm_vcpu *vcpu;
8125 	int cpu;
8126 	unsigned long flags;
8127 
8128 	mutex_lock(&kvm_lock);
8129 	list_for_each_entry(kvm, &vm_list, vm_list)
8130 		kvm_make_mclock_inprogress_request(kvm);
8131 
8132 	hyperv_stop_tsc_emulation();
8133 
8134 	/* TSC frequency always matches when on Hyper-V */
8135 	for_each_present_cpu(cpu)
8136 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8137 	kvm_max_guest_tsc_khz = tsc_khz;
8138 
8139 	list_for_each_entry(kvm, &vm_list, vm_list) {
8140 		struct kvm_arch *ka = &kvm->arch;
8141 
8142 		spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8143 		pvclock_update_vm_gtod_copy(kvm);
8144 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8145 
8146 		kvm_for_each_vcpu(cpu, vcpu, kvm)
8147 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8148 
8149 		kvm_for_each_vcpu(cpu, vcpu, kvm)
8150 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8151 	}
8152 	mutex_unlock(&kvm_lock);
8153 }
8154 #endif
8155 
8156 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8157 {
8158 	struct kvm *kvm;
8159 	struct kvm_vcpu *vcpu;
8160 	int i, send_ipi = 0;
8161 
8162 	/*
8163 	 * We allow guests to temporarily run on slowing clocks,
8164 	 * provided we notify them after, or to run on accelerating
8165 	 * clocks, provided we notify them before.  Thus time never
8166 	 * goes backwards.
8167 	 *
8168 	 * However, we have a problem.  We can't atomically update
8169 	 * the frequency of a given CPU from this function; it is
8170 	 * merely a notifier, which can be called from any CPU.
8171 	 * Changing the TSC frequency at arbitrary points in time
8172 	 * requires a recomputation of local variables related to
8173 	 * the TSC for each VCPU.  We must flag these local variables
8174 	 * to be updated and be sure the update takes place with the
8175 	 * new frequency before any guests proceed.
8176 	 *
8177 	 * Unfortunately, the combination of hotplug CPU and frequency
8178 	 * change creates an intractable locking scenario; the order
8179 	 * of when these callouts happen is undefined with respect to
8180 	 * CPU hotplug, and they can race with each other.  As such,
8181 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8182 	 * undefined; you can actually have a CPU frequency change take
8183 	 * place in between the computation of X and the setting of the
8184 	 * variable.  To protect against this problem, all updates of
8185 	 * the per_cpu tsc_khz variable are done in an interrupt
8186 	 * protected IPI, and all callers wishing to update the value
8187 	 * must wait for a synchronous IPI to complete (which is trivial
8188 	 * if the caller is on the CPU already).  This establishes the
8189 	 * necessary total order on variable updates.
8190 	 *
8191 	 * Note that because a guest time update may take place
8192 	 * anytime after the setting of the VCPU's request bit, the
8193 	 * correct TSC value must be set before the request.  However,
8194 	 * to ensure the update actually makes it to any guest which
8195 	 * starts running in hardware virtualization between the set
8196 	 * and the acquisition of the spinlock, we must also ping the
8197 	 * CPU after setting the request bit.
8198 	 *
8199 	 */
8200 
8201 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8202 
8203 	mutex_lock(&kvm_lock);
8204 	list_for_each_entry(kvm, &vm_list, vm_list) {
8205 		kvm_for_each_vcpu(i, vcpu, kvm) {
8206 			if (vcpu->cpu != cpu)
8207 				continue;
8208 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8209 			if (vcpu->cpu != raw_smp_processor_id())
8210 				send_ipi = 1;
8211 		}
8212 	}
8213 	mutex_unlock(&kvm_lock);
8214 
8215 	if (freq->old < freq->new && send_ipi) {
8216 		/*
8217 		 * We upscale the frequency.  Must make the guest
8218 		 * doesn't see old kvmclock values while running with
8219 		 * the new frequency, otherwise we risk the guest sees
8220 		 * time go backwards.
8221 		 *
8222 		 * In case we update the frequency for another cpu
8223 		 * (which might be in guest context) send an interrupt
8224 		 * to kick the cpu out of guest context.  Next time
8225 		 * guest context is entered kvmclock will be updated,
8226 		 * so the guest will not see stale values.
8227 		 */
8228 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8229 	}
8230 }
8231 
8232 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8233 				     void *data)
8234 {
8235 	struct cpufreq_freqs *freq = data;
8236 	int cpu;
8237 
8238 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8239 		return 0;
8240 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8241 		return 0;
8242 
8243 	for_each_cpu(cpu, freq->policy->cpus)
8244 		__kvmclock_cpufreq_notifier(freq, cpu);
8245 
8246 	return 0;
8247 }
8248 
8249 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8250 	.notifier_call  = kvmclock_cpufreq_notifier
8251 };
8252 
8253 static int kvmclock_cpu_online(unsigned int cpu)
8254 {
8255 	tsc_khz_changed(NULL);
8256 	return 0;
8257 }
8258 
8259 static void kvm_timer_init(void)
8260 {
8261 	max_tsc_khz = tsc_khz;
8262 
8263 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8264 #ifdef CONFIG_CPU_FREQ
8265 		struct cpufreq_policy *policy;
8266 		int cpu;
8267 
8268 		cpu = get_cpu();
8269 		policy = cpufreq_cpu_get(cpu);
8270 		if (policy) {
8271 			if (policy->cpuinfo.max_freq)
8272 				max_tsc_khz = policy->cpuinfo.max_freq;
8273 			cpufreq_cpu_put(policy);
8274 		}
8275 		put_cpu();
8276 #endif
8277 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8278 					  CPUFREQ_TRANSITION_NOTIFIER);
8279 	}
8280 
8281 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8282 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8283 }
8284 
8285 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8286 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8287 
8288 int kvm_is_in_guest(void)
8289 {
8290 	return __this_cpu_read(current_vcpu) != NULL;
8291 }
8292 
8293 static int kvm_is_user_mode(void)
8294 {
8295 	int user_mode = 3;
8296 
8297 	if (__this_cpu_read(current_vcpu))
8298 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8299 
8300 	return user_mode != 0;
8301 }
8302 
8303 static unsigned long kvm_get_guest_ip(void)
8304 {
8305 	unsigned long ip = 0;
8306 
8307 	if (__this_cpu_read(current_vcpu))
8308 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8309 
8310 	return ip;
8311 }
8312 
8313 static void kvm_handle_intel_pt_intr(void)
8314 {
8315 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8316 
8317 	kvm_make_request(KVM_REQ_PMI, vcpu);
8318 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8319 			(unsigned long *)&vcpu->arch.pmu.global_status);
8320 }
8321 
8322 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8323 	.is_in_guest		= kvm_is_in_guest,
8324 	.is_user_mode		= kvm_is_user_mode,
8325 	.get_guest_ip		= kvm_get_guest_ip,
8326 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8327 };
8328 
8329 #ifdef CONFIG_X86_64
8330 static void pvclock_gtod_update_fn(struct work_struct *work)
8331 {
8332 	struct kvm *kvm;
8333 
8334 	struct kvm_vcpu *vcpu;
8335 	int i;
8336 
8337 	mutex_lock(&kvm_lock);
8338 	list_for_each_entry(kvm, &vm_list, vm_list)
8339 		kvm_for_each_vcpu(i, vcpu, kvm)
8340 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8341 	atomic_set(&kvm_guest_has_master_clock, 0);
8342 	mutex_unlock(&kvm_lock);
8343 }
8344 
8345 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8346 
8347 /*
8348  * Indirection to move queue_work() out of the tk_core.seq write held
8349  * region to prevent possible deadlocks against time accessors which
8350  * are invoked with work related locks held.
8351  */
8352 static void pvclock_irq_work_fn(struct irq_work *w)
8353 {
8354 	queue_work(system_long_wq, &pvclock_gtod_work);
8355 }
8356 
8357 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8358 
8359 /*
8360  * Notification about pvclock gtod data update.
8361  */
8362 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8363 			       void *priv)
8364 {
8365 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8366 	struct timekeeper *tk = priv;
8367 
8368 	update_pvclock_gtod(tk);
8369 
8370 	/*
8371 	 * Disable master clock if host does not trust, or does not use,
8372 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8373 	 * this is invoked with tk_core.seq write held.
8374 	 */
8375 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8376 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8377 		irq_work_queue(&pvclock_irq_work);
8378 	return 0;
8379 }
8380 
8381 static struct notifier_block pvclock_gtod_notifier = {
8382 	.notifier_call = pvclock_gtod_notify,
8383 };
8384 #endif
8385 
8386 int kvm_arch_init(void *opaque)
8387 {
8388 	struct kvm_x86_init_ops *ops = opaque;
8389 	int r;
8390 
8391 	if (kvm_x86_ops.hardware_enable) {
8392 		printk(KERN_ERR "kvm: already loaded the other module\n");
8393 		r = -EEXIST;
8394 		goto out;
8395 	}
8396 
8397 	if (!ops->cpu_has_kvm_support()) {
8398 		pr_err_ratelimited("kvm: no hardware support\n");
8399 		r = -EOPNOTSUPP;
8400 		goto out;
8401 	}
8402 	if (ops->disabled_by_bios()) {
8403 		pr_err_ratelimited("kvm: disabled by bios\n");
8404 		r = -EOPNOTSUPP;
8405 		goto out;
8406 	}
8407 
8408 	/*
8409 	 * KVM explicitly assumes that the guest has an FPU and
8410 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8411 	 * vCPU's FPU state as a fxregs_state struct.
8412 	 */
8413 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8414 		printk(KERN_ERR "kvm: inadequate fpu\n");
8415 		r = -EOPNOTSUPP;
8416 		goto out;
8417 	}
8418 
8419 	r = -ENOMEM;
8420 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8421 					  __alignof__(struct fpu), SLAB_ACCOUNT,
8422 					  NULL);
8423 	if (!x86_fpu_cache) {
8424 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8425 		goto out;
8426 	}
8427 
8428 	x86_emulator_cache = kvm_alloc_emulator_cache();
8429 	if (!x86_emulator_cache) {
8430 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8431 		goto out_free_x86_fpu_cache;
8432 	}
8433 
8434 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8435 	if (!user_return_msrs) {
8436 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8437 		goto out_free_x86_emulator_cache;
8438 	}
8439 	kvm_nr_uret_msrs = 0;
8440 
8441 	r = kvm_mmu_module_init();
8442 	if (r)
8443 		goto out_free_percpu;
8444 
8445 	kvm_timer_init();
8446 
8447 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8448 
8449 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8450 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8451 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8452 	}
8453 
8454 	if (pi_inject_timer == -1)
8455 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8456 #ifdef CONFIG_X86_64
8457 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8458 
8459 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8460 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8461 #endif
8462 
8463 	return 0;
8464 
8465 out_free_percpu:
8466 	free_percpu(user_return_msrs);
8467 out_free_x86_emulator_cache:
8468 	kmem_cache_destroy(x86_emulator_cache);
8469 out_free_x86_fpu_cache:
8470 	kmem_cache_destroy(x86_fpu_cache);
8471 out:
8472 	return r;
8473 }
8474 
8475 void kvm_arch_exit(void)
8476 {
8477 #ifdef CONFIG_X86_64
8478 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8479 		clear_hv_tscchange_cb();
8480 #endif
8481 	kvm_lapic_exit();
8482 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8483 
8484 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8485 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8486 					    CPUFREQ_TRANSITION_NOTIFIER);
8487 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8488 #ifdef CONFIG_X86_64
8489 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8490 	irq_work_sync(&pvclock_irq_work);
8491 	cancel_work_sync(&pvclock_gtod_work);
8492 #endif
8493 	kvm_x86_ops.hardware_enable = NULL;
8494 	kvm_mmu_module_exit();
8495 	free_percpu(user_return_msrs);
8496 	kmem_cache_destroy(x86_emulator_cache);
8497 	kmem_cache_destroy(x86_fpu_cache);
8498 #ifdef CONFIG_KVM_XEN
8499 	static_key_deferred_flush(&kvm_xen_enabled);
8500 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8501 #endif
8502 }
8503 
8504 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8505 {
8506 	++vcpu->stat.halt_exits;
8507 	if (lapic_in_kernel(vcpu)) {
8508 		vcpu->arch.mp_state = state;
8509 		return 1;
8510 	} else {
8511 		vcpu->run->exit_reason = reason;
8512 		return 0;
8513 	}
8514 }
8515 
8516 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8517 {
8518 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8519 }
8520 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8521 
8522 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8523 {
8524 	int ret = kvm_skip_emulated_instruction(vcpu);
8525 	/*
8526 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8527 	 * KVM_EXIT_DEBUG here.
8528 	 */
8529 	return kvm_vcpu_halt(vcpu) && ret;
8530 }
8531 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8532 
8533 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8534 {
8535 	int ret = kvm_skip_emulated_instruction(vcpu);
8536 
8537 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8538 }
8539 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8540 
8541 #ifdef CONFIG_X86_64
8542 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8543 			        unsigned long clock_type)
8544 {
8545 	struct kvm_clock_pairing clock_pairing;
8546 	struct timespec64 ts;
8547 	u64 cycle;
8548 	int ret;
8549 
8550 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8551 		return -KVM_EOPNOTSUPP;
8552 
8553 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8554 		return -KVM_EOPNOTSUPP;
8555 
8556 	clock_pairing.sec = ts.tv_sec;
8557 	clock_pairing.nsec = ts.tv_nsec;
8558 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8559 	clock_pairing.flags = 0;
8560 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8561 
8562 	ret = 0;
8563 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8564 			    sizeof(struct kvm_clock_pairing)))
8565 		ret = -KVM_EFAULT;
8566 
8567 	return ret;
8568 }
8569 #endif
8570 
8571 /*
8572  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8573  *
8574  * @apicid - apicid of vcpu to be kicked.
8575  */
8576 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8577 {
8578 	struct kvm_lapic_irq lapic_irq;
8579 
8580 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8581 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8582 	lapic_irq.level = 0;
8583 	lapic_irq.dest_id = apicid;
8584 	lapic_irq.msi_redir_hint = false;
8585 
8586 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8587 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8588 }
8589 
8590 bool kvm_apicv_activated(struct kvm *kvm)
8591 {
8592 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8593 }
8594 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8595 
8596 static void kvm_apicv_init(struct kvm *kvm)
8597 {
8598 	mutex_init(&kvm->arch.apicv_update_lock);
8599 
8600 	if (enable_apicv)
8601 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8602 			  &kvm->arch.apicv_inhibit_reasons);
8603 	else
8604 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8605 			&kvm->arch.apicv_inhibit_reasons);
8606 }
8607 
8608 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8609 {
8610 	struct kvm_vcpu *target = NULL;
8611 	struct kvm_apic_map *map;
8612 
8613 	vcpu->stat.directed_yield_attempted++;
8614 
8615 	if (single_task_running())
8616 		goto no_yield;
8617 
8618 	rcu_read_lock();
8619 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8620 
8621 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8622 		target = map->phys_map[dest_id]->vcpu;
8623 
8624 	rcu_read_unlock();
8625 
8626 	if (!target || !READ_ONCE(target->ready))
8627 		goto no_yield;
8628 
8629 	/* Ignore requests to yield to self */
8630 	if (vcpu == target)
8631 		goto no_yield;
8632 
8633 	if (kvm_vcpu_yield_to(target) <= 0)
8634 		goto no_yield;
8635 
8636 	vcpu->stat.directed_yield_successful++;
8637 
8638 no_yield:
8639 	return;
8640 }
8641 
8642 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8643 {
8644 	u64 ret = vcpu->run->hypercall.ret;
8645 
8646 	if (!is_64_bit_mode(vcpu))
8647 		ret = (u32)ret;
8648 	kvm_rax_write(vcpu, ret);
8649 	++vcpu->stat.hypercalls;
8650 	return kvm_skip_emulated_instruction(vcpu);
8651 }
8652 
8653 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8654 {
8655 	unsigned long nr, a0, a1, a2, a3, ret;
8656 	int op_64_bit;
8657 
8658 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8659 		return kvm_xen_hypercall(vcpu);
8660 
8661 	if (kvm_hv_hypercall_enabled(vcpu))
8662 		return kvm_hv_hypercall(vcpu);
8663 
8664 	nr = kvm_rax_read(vcpu);
8665 	a0 = kvm_rbx_read(vcpu);
8666 	a1 = kvm_rcx_read(vcpu);
8667 	a2 = kvm_rdx_read(vcpu);
8668 	a3 = kvm_rsi_read(vcpu);
8669 
8670 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8671 
8672 	op_64_bit = is_64_bit_mode(vcpu);
8673 	if (!op_64_bit) {
8674 		nr &= 0xFFFFFFFF;
8675 		a0 &= 0xFFFFFFFF;
8676 		a1 &= 0xFFFFFFFF;
8677 		a2 &= 0xFFFFFFFF;
8678 		a3 &= 0xFFFFFFFF;
8679 	}
8680 
8681 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8682 		ret = -KVM_EPERM;
8683 		goto out;
8684 	}
8685 
8686 	ret = -KVM_ENOSYS;
8687 
8688 	switch (nr) {
8689 	case KVM_HC_VAPIC_POLL_IRQ:
8690 		ret = 0;
8691 		break;
8692 	case KVM_HC_KICK_CPU:
8693 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8694 			break;
8695 
8696 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8697 		kvm_sched_yield(vcpu, a1);
8698 		ret = 0;
8699 		break;
8700 #ifdef CONFIG_X86_64
8701 	case KVM_HC_CLOCK_PAIRING:
8702 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8703 		break;
8704 #endif
8705 	case KVM_HC_SEND_IPI:
8706 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8707 			break;
8708 
8709 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8710 		break;
8711 	case KVM_HC_SCHED_YIELD:
8712 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8713 			break;
8714 
8715 		kvm_sched_yield(vcpu, a0);
8716 		ret = 0;
8717 		break;
8718 	case KVM_HC_MAP_GPA_RANGE: {
8719 		u64 gpa = a0, npages = a1, attrs = a2;
8720 
8721 		ret = -KVM_ENOSYS;
8722 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8723 			break;
8724 
8725 		if (!PAGE_ALIGNED(gpa) || !npages ||
8726 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8727 			ret = -KVM_EINVAL;
8728 			break;
8729 		}
8730 
8731 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
8732 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
8733 		vcpu->run->hypercall.args[0]  = gpa;
8734 		vcpu->run->hypercall.args[1]  = npages;
8735 		vcpu->run->hypercall.args[2]  = attrs;
8736 		vcpu->run->hypercall.longmode = op_64_bit;
8737 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8738 		return 0;
8739 	}
8740 	default:
8741 		ret = -KVM_ENOSYS;
8742 		break;
8743 	}
8744 out:
8745 	if (!op_64_bit)
8746 		ret = (u32)ret;
8747 	kvm_rax_write(vcpu, ret);
8748 
8749 	++vcpu->stat.hypercalls;
8750 	return kvm_skip_emulated_instruction(vcpu);
8751 }
8752 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8753 
8754 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8755 {
8756 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8757 	char instruction[3];
8758 	unsigned long rip = kvm_rip_read(vcpu);
8759 
8760 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8761 
8762 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8763 		&ctxt->exception);
8764 }
8765 
8766 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8767 {
8768 	return vcpu->run->request_interrupt_window &&
8769 		likely(!pic_in_kernel(vcpu->kvm));
8770 }
8771 
8772 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8773 {
8774 	struct kvm_run *kvm_run = vcpu->run;
8775 
8776 	/*
8777 	 * if_flag is obsolete and useless, so do not bother
8778 	 * setting it for SEV-ES guests.  Userspace can just
8779 	 * use kvm_run->ready_for_interrupt_injection.
8780 	 */
8781 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8782 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8783 
8784 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8785 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8786 	kvm_run->ready_for_interrupt_injection =
8787 		pic_in_kernel(vcpu->kvm) ||
8788 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8789 
8790 	if (is_smm(vcpu))
8791 		kvm_run->flags |= KVM_RUN_X86_SMM;
8792 }
8793 
8794 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8795 {
8796 	int max_irr, tpr;
8797 
8798 	if (!kvm_x86_ops.update_cr8_intercept)
8799 		return;
8800 
8801 	if (!lapic_in_kernel(vcpu))
8802 		return;
8803 
8804 	if (vcpu->arch.apicv_active)
8805 		return;
8806 
8807 	if (!vcpu->arch.apic->vapic_addr)
8808 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8809 	else
8810 		max_irr = -1;
8811 
8812 	if (max_irr != -1)
8813 		max_irr >>= 4;
8814 
8815 	tpr = kvm_lapic_get_cr8(vcpu);
8816 
8817 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8818 }
8819 
8820 
8821 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8822 {
8823 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8824 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
8825 		return 1;
8826 	}
8827 
8828 	return kvm_x86_ops.nested_ops->check_events(vcpu);
8829 }
8830 
8831 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8832 {
8833 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8834 		vcpu->arch.exception.error_code = false;
8835 	static_call(kvm_x86_queue_exception)(vcpu);
8836 }
8837 
8838 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8839 {
8840 	int r;
8841 	bool can_inject = true;
8842 
8843 	/* try to reinject previous events if any */
8844 
8845 	if (vcpu->arch.exception.injected) {
8846 		kvm_inject_exception(vcpu);
8847 		can_inject = false;
8848 	}
8849 	/*
8850 	 * Do not inject an NMI or interrupt if there is a pending
8851 	 * exception.  Exceptions and interrupts are recognized at
8852 	 * instruction boundaries, i.e. the start of an instruction.
8853 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8854 	 * NMIs and interrupts, i.e. traps are recognized before an
8855 	 * NMI/interrupt that's pending on the same instruction.
8856 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8857 	 * priority, but are only generated (pended) during instruction
8858 	 * execution, i.e. a pending fault-like exception means the
8859 	 * fault occurred on the *previous* instruction and must be
8860 	 * serviced prior to recognizing any new events in order to
8861 	 * fully complete the previous instruction.
8862 	 */
8863 	else if (!vcpu->arch.exception.pending) {
8864 		if (vcpu->arch.nmi_injected) {
8865 			static_call(kvm_x86_set_nmi)(vcpu);
8866 			can_inject = false;
8867 		} else if (vcpu->arch.interrupt.injected) {
8868 			static_call(kvm_x86_set_irq)(vcpu);
8869 			can_inject = false;
8870 		}
8871 	}
8872 
8873 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8874 		     vcpu->arch.exception.pending);
8875 
8876 	/*
8877 	 * Call check_nested_events() even if we reinjected a previous event
8878 	 * in order for caller to determine if it should require immediate-exit
8879 	 * from L2 to L1 due to pending L1 events which require exit
8880 	 * from L2 to L1.
8881 	 */
8882 	if (is_guest_mode(vcpu)) {
8883 		r = kvm_check_nested_events(vcpu);
8884 		if (r < 0)
8885 			goto out;
8886 	}
8887 
8888 	/* try to inject new event if pending */
8889 	if (vcpu->arch.exception.pending) {
8890 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8891 					vcpu->arch.exception.has_error_code,
8892 					vcpu->arch.exception.error_code);
8893 
8894 		vcpu->arch.exception.pending = false;
8895 		vcpu->arch.exception.injected = true;
8896 
8897 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8898 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8899 					     X86_EFLAGS_RF);
8900 
8901 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8902 			kvm_deliver_exception_payload(vcpu);
8903 			if (vcpu->arch.dr7 & DR7_GD) {
8904 				vcpu->arch.dr7 &= ~DR7_GD;
8905 				kvm_update_dr7(vcpu);
8906 			}
8907 		}
8908 
8909 		kvm_inject_exception(vcpu);
8910 		can_inject = false;
8911 	}
8912 
8913 	/* Don't inject interrupts if the user asked to avoid doing so */
8914 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
8915 		return 0;
8916 
8917 	/*
8918 	 * Finally, inject interrupt events.  If an event cannot be injected
8919 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8920 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8921 	 * and can architecturally be injected, but we cannot do it right now:
8922 	 * an interrupt could have arrived just now and we have to inject it
8923 	 * as a vmexit, or there could already an event in the queue, which is
8924 	 * indicated by can_inject.  In that case we request an immediate exit
8925 	 * in order to make progress and get back here for another iteration.
8926 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8927 	 */
8928 	if (vcpu->arch.smi_pending) {
8929 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8930 		if (r < 0)
8931 			goto out;
8932 		if (r) {
8933 			vcpu->arch.smi_pending = false;
8934 			++vcpu->arch.smi_count;
8935 			enter_smm(vcpu);
8936 			can_inject = false;
8937 		} else
8938 			static_call(kvm_x86_enable_smi_window)(vcpu);
8939 	}
8940 
8941 	if (vcpu->arch.nmi_pending) {
8942 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8943 		if (r < 0)
8944 			goto out;
8945 		if (r) {
8946 			--vcpu->arch.nmi_pending;
8947 			vcpu->arch.nmi_injected = true;
8948 			static_call(kvm_x86_set_nmi)(vcpu);
8949 			can_inject = false;
8950 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8951 		}
8952 		if (vcpu->arch.nmi_pending)
8953 			static_call(kvm_x86_enable_nmi_window)(vcpu);
8954 	}
8955 
8956 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8957 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8958 		if (r < 0)
8959 			goto out;
8960 		if (r) {
8961 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8962 			static_call(kvm_x86_set_irq)(vcpu);
8963 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8964 		}
8965 		if (kvm_cpu_has_injectable_intr(vcpu))
8966 			static_call(kvm_x86_enable_irq_window)(vcpu);
8967 	}
8968 
8969 	if (is_guest_mode(vcpu) &&
8970 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8971 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8972 		*req_immediate_exit = true;
8973 
8974 	WARN_ON(vcpu->arch.exception.pending);
8975 	return 0;
8976 
8977 out:
8978 	if (r == -EBUSY) {
8979 		*req_immediate_exit = true;
8980 		r = 0;
8981 	}
8982 	return r;
8983 }
8984 
8985 static void process_nmi(struct kvm_vcpu *vcpu)
8986 {
8987 	unsigned limit = 2;
8988 
8989 	/*
8990 	 * x86 is limited to one NMI running, and one NMI pending after it.
8991 	 * If an NMI is already in progress, limit further NMIs to just one.
8992 	 * Otherwise, allow two (and we'll inject the first one immediately).
8993 	 */
8994 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8995 		limit = 1;
8996 
8997 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8998 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8999 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9000 }
9001 
9002 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9003 {
9004 	u32 flags = 0;
9005 	flags |= seg->g       << 23;
9006 	flags |= seg->db      << 22;
9007 	flags |= seg->l       << 21;
9008 	flags |= seg->avl     << 20;
9009 	flags |= seg->present << 15;
9010 	flags |= seg->dpl     << 13;
9011 	flags |= seg->s       << 12;
9012 	flags |= seg->type    << 8;
9013 	return flags;
9014 }
9015 
9016 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9017 {
9018 	struct kvm_segment seg;
9019 	int offset;
9020 
9021 	kvm_get_segment(vcpu, &seg, n);
9022 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9023 
9024 	if (n < 3)
9025 		offset = 0x7f84 + n * 12;
9026 	else
9027 		offset = 0x7f2c + (n - 3) * 12;
9028 
9029 	put_smstate(u32, buf, offset + 8, seg.base);
9030 	put_smstate(u32, buf, offset + 4, seg.limit);
9031 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9032 }
9033 
9034 #ifdef CONFIG_X86_64
9035 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9036 {
9037 	struct kvm_segment seg;
9038 	int offset;
9039 	u16 flags;
9040 
9041 	kvm_get_segment(vcpu, &seg, n);
9042 	offset = 0x7e00 + n * 16;
9043 
9044 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9045 	put_smstate(u16, buf, offset, seg.selector);
9046 	put_smstate(u16, buf, offset + 2, flags);
9047 	put_smstate(u32, buf, offset + 4, seg.limit);
9048 	put_smstate(u64, buf, offset + 8, seg.base);
9049 }
9050 #endif
9051 
9052 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9053 {
9054 	struct desc_ptr dt;
9055 	struct kvm_segment seg;
9056 	unsigned long val;
9057 	int i;
9058 
9059 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9060 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9061 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9062 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9063 
9064 	for (i = 0; i < 8; i++)
9065 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9066 
9067 	kvm_get_dr(vcpu, 6, &val);
9068 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9069 	kvm_get_dr(vcpu, 7, &val);
9070 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9071 
9072 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9073 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9074 	put_smstate(u32, buf, 0x7f64, seg.base);
9075 	put_smstate(u32, buf, 0x7f60, seg.limit);
9076 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9077 
9078 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9079 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9080 	put_smstate(u32, buf, 0x7f80, seg.base);
9081 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9082 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9083 
9084 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9085 	put_smstate(u32, buf, 0x7f74, dt.address);
9086 	put_smstate(u32, buf, 0x7f70, dt.size);
9087 
9088 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9089 	put_smstate(u32, buf, 0x7f58, dt.address);
9090 	put_smstate(u32, buf, 0x7f54, dt.size);
9091 
9092 	for (i = 0; i < 6; i++)
9093 		enter_smm_save_seg_32(vcpu, buf, i);
9094 
9095 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9096 
9097 	/* revision id */
9098 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9099 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9100 }
9101 
9102 #ifdef CONFIG_X86_64
9103 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9104 {
9105 	struct desc_ptr dt;
9106 	struct kvm_segment seg;
9107 	unsigned long val;
9108 	int i;
9109 
9110 	for (i = 0; i < 16; i++)
9111 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9112 
9113 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9114 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9115 
9116 	kvm_get_dr(vcpu, 6, &val);
9117 	put_smstate(u64, buf, 0x7f68, val);
9118 	kvm_get_dr(vcpu, 7, &val);
9119 	put_smstate(u64, buf, 0x7f60, val);
9120 
9121 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9122 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9123 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9124 
9125 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9126 
9127 	/* revision id */
9128 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9129 
9130 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9131 
9132 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9133 	put_smstate(u16, buf, 0x7e90, seg.selector);
9134 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9135 	put_smstate(u32, buf, 0x7e94, seg.limit);
9136 	put_smstate(u64, buf, 0x7e98, seg.base);
9137 
9138 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9139 	put_smstate(u32, buf, 0x7e84, dt.size);
9140 	put_smstate(u64, buf, 0x7e88, dt.address);
9141 
9142 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9143 	put_smstate(u16, buf, 0x7e70, seg.selector);
9144 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9145 	put_smstate(u32, buf, 0x7e74, seg.limit);
9146 	put_smstate(u64, buf, 0x7e78, seg.base);
9147 
9148 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9149 	put_smstate(u32, buf, 0x7e64, dt.size);
9150 	put_smstate(u64, buf, 0x7e68, dt.address);
9151 
9152 	for (i = 0; i < 6; i++)
9153 		enter_smm_save_seg_64(vcpu, buf, i);
9154 }
9155 #endif
9156 
9157 static void enter_smm(struct kvm_vcpu *vcpu)
9158 {
9159 	struct kvm_segment cs, ds;
9160 	struct desc_ptr dt;
9161 	unsigned long cr0;
9162 	char buf[512];
9163 
9164 	memset(buf, 0, 512);
9165 #ifdef CONFIG_X86_64
9166 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9167 		enter_smm_save_state_64(vcpu, buf);
9168 	else
9169 #endif
9170 		enter_smm_save_state_32(vcpu, buf);
9171 
9172 	/*
9173 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9174 	 * state (e.g. leave guest mode) after we've saved the state into the
9175 	 * SMM state-save area.
9176 	 */
9177 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9178 
9179 	kvm_smm_changed(vcpu, true);
9180 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9181 
9182 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9183 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9184 	else
9185 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9186 
9187 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9188 	kvm_rip_write(vcpu, 0x8000);
9189 
9190 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9191 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9192 	vcpu->arch.cr0 = cr0;
9193 
9194 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9195 
9196 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9197 	dt.address = dt.size = 0;
9198 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9199 
9200 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9201 
9202 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9203 	cs.base = vcpu->arch.smbase;
9204 
9205 	ds.selector = 0;
9206 	ds.base = 0;
9207 
9208 	cs.limit    = ds.limit = 0xffffffff;
9209 	cs.type     = ds.type = 0x3;
9210 	cs.dpl      = ds.dpl = 0;
9211 	cs.db       = ds.db = 0;
9212 	cs.s        = ds.s = 1;
9213 	cs.l        = ds.l = 0;
9214 	cs.g        = ds.g = 1;
9215 	cs.avl      = ds.avl = 0;
9216 	cs.present  = ds.present = 1;
9217 	cs.unusable = ds.unusable = 0;
9218 	cs.padding  = ds.padding = 0;
9219 
9220 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9221 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9222 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9223 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9224 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9225 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9226 
9227 #ifdef CONFIG_X86_64
9228 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9229 		static_call(kvm_x86_set_efer)(vcpu, 0);
9230 #endif
9231 
9232 	kvm_update_cpuid_runtime(vcpu);
9233 	kvm_mmu_reset_context(vcpu);
9234 }
9235 
9236 static void process_smi(struct kvm_vcpu *vcpu)
9237 {
9238 	vcpu->arch.smi_pending = true;
9239 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9240 }
9241 
9242 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9243 				       unsigned long *vcpu_bitmap)
9244 {
9245 	cpumask_var_t cpus;
9246 
9247 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9248 
9249 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9250 				    NULL, vcpu_bitmap, cpus);
9251 
9252 	free_cpumask_var(cpus);
9253 }
9254 
9255 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9256 {
9257 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9258 }
9259 
9260 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9261 {
9262 	bool activate;
9263 
9264 	if (!lapic_in_kernel(vcpu))
9265 		return;
9266 
9267 	mutex_lock(&vcpu->kvm->arch.apicv_update_lock);
9268 
9269 	activate = kvm_apicv_activated(vcpu->kvm);
9270 	if (vcpu->arch.apicv_active == activate)
9271 		goto out;
9272 
9273 	vcpu->arch.apicv_active = activate;
9274 	kvm_apic_update_apicv(vcpu);
9275 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9276 
9277 	/*
9278 	 * When APICv gets disabled, we may still have injected interrupts
9279 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9280 	 * still active when the interrupt got accepted. Make sure
9281 	 * inject_pending_event() is called to check for that.
9282 	 */
9283 	if (!vcpu->arch.apicv_active)
9284 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9285 
9286 out:
9287 	mutex_unlock(&vcpu->kvm->arch.apicv_update_lock);
9288 }
9289 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9290 
9291 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9292 {
9293 	unsigned long old, new;
9294 
9295 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9296 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9297 		return;
9298 
9299 	old = new = kvm->arch.apicv_inhibit_reasons;
9300 
9301 	if (activate)
9302 		__clear_bit(bit, &new);
9303 	else
9304 		__set_bit(bit, &new);
9305 
9306 	if (!!old != !!new) {
9307 		trace_kvm_apicv_update_request(activate, bit);
9308 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9309 		kvm->arch.apicv_inhibit_reasons = new;
9310 		if (new) {
9311 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9312 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
9313 		}
9314 	} else
9315 		kvm->arch.apicv_inhibit_reasons = new;
9316 }
9317 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9318 
9319 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9320 {
9321 	mutex_lock(&kvm->arch.apicv_update_lock);
9322 	__kvm_request_apicv_update(kvm, activate, bit);
9323 	mutex_unlock(&kvm->arch.apicv_update_lock);
9324 }
9325 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9326 
9327 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9328 {
9329 	if (!kvm_apic_present(vcpu))
9330 		return;
9331 
9332 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9333 
9334 	if (irqchip_split(vcpu->kvm))
9335 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9336 	else {
9337 		if (vcpu->arch.apicv_active)
9338 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9339 		if (ioapic_in_kernel(vcpu->kvm))
9340 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9341 	}
9342 
9343 	if (is_guest_mode(vcpu))
9344 		vcpu->arch.load_eoi_exitmap_pending = true;
9345 	else
9346 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9347 }
9348 
9349 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9350 {
9351 	u64 eoi_exit_bitmap[4];
9352 
9353 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9354 		return;
9355 
9356 	if (to_hv_vcpu(vcpu))
9357 		bitmap_or((ulong *)eoi_exit_bitmap,
9358 			  vcpu->arch.ioapic_handled_vectors,
9359 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9360 
9361 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9362 }
9363 
9364 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9365 					    unsigned long start, unsigned long end)
9366 {
9367 	unsigned long apic_address;
9368 
9369 	/*
9370 	 * The physical address of apic access page is stored in the VMCS.
9371 	 * Update it when it becomes invalid.
9372 	 */
9373 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9374 	if (start <= apic_address && apic_address < end)
9375 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9376 }
9377 
9378 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9379 {
9380 	if (!lapic_in_kernel(vcpu))
9381 		return;
9382 
9383 	if (!kvm_x86_ops.set_apic_access_page_addr)
9384 		return;
9385 
9386 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9387 }
9388 
9389 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9390 {
9391 	smp_send_reschedule(vcpu->cpu);
9392 }
9393 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9394 
9395 /*
9396  * Returns 1 to let vcpu_run() continue the guest execution loop without
9397  * exiting to the userspace.  Otherwise, the value will be returned to the
9398  * userspace.
9399  */
9400 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9401 {
9402 	int r;
9403 	bool req_int_win =
9404 		dm_request_for_irq_injection(vcpu) &&
9405 		kvm_cpu_accept_dm_intr(vcpu);
9406 	fastpath_t exit_fastpath;
9407 
9408 	bool req_immediate_exit = false;
9409 
9410 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9411 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9412 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9413 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9414 		trace_kvm_dirty_ring_exit(vcpu);
9415 		r = 0;
9416 		goto out;
9417 	}
9418 
9419 	if (kvm_request_pending(vcpu)) {
9420 		if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9421 			r = -EIO;
9422 			goto out;
9423 		}
9424 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9425 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9426 				r = 0;
9427 				goto out;
9428 			}
9429 		}
9430 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9431 			kvm_mmu_unload(vcpu);
9432 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9433 			__kvm_migrate_timers(vcpu);
9434 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9435 			kvm_gen_update_masterclock(vcpu->kvm);
9436 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9437 			kvm_gen_kvmclock_update(vcpu);
9438 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9439 			r = kvm_guest_time_update(vcpu);
9440 			if (unlikely(r))
9441 				goto out;
9442 		}
9443 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9444 			kvm_mmu_sync_roots(vcpu);
9445 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9446 			kvm_mmu_load_pgd(vcpu);
9447 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9448 			kvm_vcpu_flush_tlb_all(vcpu);
9449 
9450 			/* Flushing all ASIDs flushes the current ASID... */
9451 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9452 		}
9453 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9454 			kvm_vcpu_flush_tlb_current(vcpu);
9455 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9456 			kvm_vcpu_flush_tlb_guest(vcpu);
9457 
9458 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9459 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9460 			r = 0;
9461 			goto out;
9462 		}
9463 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9464 			if (is_guest_mode(vcpu)) {
9465 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9466 			} else {
9467 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9468 				vcpu->mmio_needed = 0;
9469 				r = 0;
9470 				goto out;
9471 			}
9472 		}
9473 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9474 			/* Page is swapped out. Do synthetic halt */
9475 			vcpu->arch.apf.halted = true;
9476 			r = 1;
9477 			goto out;
9478 		}
9479 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9480 			record_steal_time(vcpu);
9481 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9482 			process_smi(vcpu);
9483 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9484 			process_nmi(vcpu);
9485 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9486 			kvm_pmu_handle_event(vcpu);
9487 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9488 			kvm_pmu_deliver_pmi(vcpu);
9489 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9490 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9491 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9492 				     vcpu->arch.ioapic_handled_vectors)) {
9493 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9494 				vcpu->run->eoi.vector =
9495 						vcpu->arch.pending_ioapic_eoi;
9496 				r = 0;
9497 				goto out;
9498 			}
9499 		}
9500 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9501 			vcpu_scan_ioapic(vcpu);
9502 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9503 			vcpu_load_eoi_exitmap(vcpu);
9504 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9505 			kvm_vcpu_reload_apic_access_page(vcpu);
9506 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9507 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9508 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9509 			r = 0;
9510 			goto out;
9511 		}
9512 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9513 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9514 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9515 			r = 0;
9516 			goto out;
9517 		}
9518 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9519 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9520 
9521 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9522 			vcpu->run->hyperv = hv_vcpu->exit;
9523 			r = 0;
9524 			goto out;
9525 		}
9526 
9527 		/*
9528 		 * KVM_REQ_HV_STIMER has to be processed after
9529 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9530 		 * depend on the guest clock being up-to-date
9531 		 */
9532 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9533 			kvm_hv_process_stimers(vcpu);
9534 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9535 			kvm_vcpu_update_apicv(vcpu);
9536 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9537 			kvm_check_async_pf_completion(vcpu);
9538 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9539 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9540 
9541 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9542 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9543 	}
9544 
9545 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9546 	    kvm_xen_has_interrupt(vcpu)) {
9547 		++vcpu->stat.req_event;
9548 		r = kvm_apic_accept_events(vcpu);
9549 		if (r < 0) {
9550 			r = 0;
9551 			goto out;
9552 		}
9553 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9554 			r = 1;
9555 			goto out;
9556 		}
9557 
9558 		r = inject_pending_event(vcpu, &req_immediate_exit);
9559 		if (r < 0) {
9560 			r = 0;
9561 			goto out;
9562 		}
9563 		if (req_int_win)
9564 			static_call(kvm_x86_enable_irq_window)(vcpu);
9565 
9566 		if (kvm_lapic_enabled(vcpu)) {
9567 			update_cr8_intercept(vcpu);
9568 			kvm_lapic_sync_to_vapic(vcpu);
9569 		}
9570 	}
9571 
9572 	r = kvm_mmu_reload(vcpu);
9573 	if (unlikely(r)) {
9574 		goto cancel_injection;
9575 	}
9576 
9577 	preempt_disable();
9578 
9579 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9580 
9581 	/*
9582 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9583 	 * IPI are then delayed after guest entry, which ensures that they
9584 	 * result in virtual interrupt delivery.
9585 	 */
9586 	local_irq_disable();
9587 	vcpu->mode = IN_GUEST_MODE;
9588 
9589 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9590 
9591 	/*
9592 	 * 1) We should set ->mode before checking ->requests.  Please see
9593 	 * the comment in kvm_vcpu_exiting_guest_mode().
9594 	 *
9595 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9596 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9597 	 * (see vmx_deliver_posted_interrupt).
9598 	 *
9599 	 * 3) This also orders the write to mode from any reads to the page
9600 	 * tables done while the VCPU is running.  Please see the comment
9601 	 * in kvm_flush_remote_tlbs.
9602 	 */
9603 	smp_mb__after_srcu_read_unlock();
9604 
9605 	/*
9606 	 * This handles the case where a posted interrupt was
9607 	 * notified with kvm_vcpu_kick.
9608 	 */
9609 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9610 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9611 
9612 	if (kvm_vcpu_exit_request(vcpu)) {
9613 		vcpu->mode = OUTSIDE_GUEST_MODE;
9614 		smp_wmb();
9615 		local_irq_enable();
9616 		preempt_enable();
9617 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9618 		r = 1;
9619 		goto cancel_injection;
9620 	}
9621 
9622 	if (req_immediate_exit) {
9623 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9624 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9625 	}
9626 
9627 	fpregs_assert_state_consistent();
9628 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9629 		switch_fpu_return();
9630 
9631 	if (unlikely(vcpu->arch.switch_db_regs)) {
9632 		set_debugreg(0, 7);
9633 		set_debugreg(vcpu->arch.eff_db[0], 0);
9634 		set_debugreg(vcpu->arch.eff_db[1], 1);
9635 		set_debugreg(vcpu->arch.eff_db[2], 2);
9636 		set_debugreg(vcpu->arch.eff_db[3], 3);
9637 	} else if (unlikely(hw_breakpoint_active())) {
9638 		set_debugreg(0, 7);
9639 	}
9640 
9641 	for (;;) {
9642 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9643 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9644 			break;
9645 
9646                 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9647 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9648 			break;
9649 		}
9650 
9651 		if (vcpu->arch.apicv_active)
9652 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9653         }
9654 
9655 	/*
9656 	 * Do this here before restoring debug registers on the host.  And
9657 	 * since we do this before handling the vmexit, a DR access vmexit
9658 	 * can (a) read the correct value of the debug registers, (b) set
9659 	 * KVM_DEBUGREG_WONT_EXIT again.
9660 	 */
9661 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9662 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9663 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9664 		kvm_update_dr0123(vcpu);
9665 		kvm_update_dr7(vcpu);
9666 	}
9667 
9668 	/*
9669 	 * If the guest has used debug registers, at least dr7
9670 	 * will be disabled while returning to the host.
9671 	 * If we don't have active breakpoints in the host, we don't
9672 	 * care about the messed up debug address registers. But if
9673 	 * we have some of them active, restore the old state.
9674 	 */
9675 	if (hw_breakpoint_active())
9676 		hw_breakpoint_restore();
9677 
9678 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9679 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9680 
9681 	vcpu->mode = OUTSIDE_GUEST_MODE;
9682 	smp_wmb();
9683 
9684 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9685 
9686 	/*
9687 	 * Consume any pending interrupts, including the possible source of
9688 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9689 	 * An instruction is required after local_irq_enable() to fully unblock
9690 	 * interrupts on processors that implement an interrupt shadow, the
9691 	 * stat.exits increment will do nicely.
9692 	 */
9693 	kvm_before_interrupt(vcpu);
9694 	local_irq_enable();
9695 	++vcpu->stat.exits;
9696 	local_irq_disable();
9697 	kvm_after_interrupt(vcpu);
9698 
9699 	/*
9700 	 * Wait until after servicing IRQs to account guest time so that any
9701 	 * ticks that occurred while running the guest are properly accounted
9702 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9703 	 * of accounting via context tracking, but the loss of accuracy is
9704 	 * acceptable for all known use cases.
9705 	 */
9706 	vtime_account_guest_exit();
9707 
9708 	if (lapic_in_kernel(vcpu)) {
9709 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9710 		if (delta != S64_MIN) {
9711 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9712 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9713 		}
9714 	}
9715 
9716 	local_irq_enable();
9717 	preempt_enable();
9718 
9719 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9720 
9721 	/*
9722 	 * Profile KVM exit RIPs:
9723 	 */
9724 	if (unlikely(prof_on == KVM_PROFILING)) {
9725 		unsigned long rip = kvm_rip_read(vcpu);
9726 		profile_hit(KVM_PROFILING, (void *)rip);
9727 	}
9728 
9729 	if (unlikely(vcpu->arch.tsc_always_catchup))
9730 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9731 
9732 	if (vcpu->arch.apic_attention)
9733 		kvm_lapic_sync_from_vapic(vcpu);
9734 
9735 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9736 	return r;
9737 
9738 cancel_injection:
9739 	if (req_immediate_exit)
9740 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9741 	static_call(kvm_x86_cancel_injection)(vcpu);
9742 	if (unlikely(vcpu->arch.apic_attention))
9743 		kvm_lapic_sync_from_vapic(vcpu);
9744 out:
9745 	return r;
9746 }
9747 
9748 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9749 {
9750 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9751 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9752 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9753 		kvm_vcpu_block(vcpu);
9754 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9755 
9756 		if (kvm_x86_ops.post_block)
9757 			static_call(kvm_x86_post_block)(vcpu);
9758 
9759 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9760 			return 1;
9761 	}
9762 
9763 	if (kvm_apic_accept_events(vcpu) < 0)
9764 		return 0;
9765 	switch(vcpu->arch.mp_state) {
9766 	case KVM_MP_STATE_HALTED:
9767 	case KVM_MP_STATE_AP_RESET_HOLD:
9768 		vcpu->arch.pv.pv_unhalted = false;
9769 		vcpu->arch.mp_state =
9770 			KVM_MP_STATE_RUNNABLE;
9771 		fallthrough;
9772 	case KVM_MP_STATE_RUNNABLE:
9773 		vcpu->arch.apf.halted = false;
9774 		break;
9775 	case KVM_MP_STATE_INIT_RECEIVED:
9776 		break;
9777 	default:
9778 		return -EINTR;
9779 	}
9780 	return 1;
9781 }
9782 
9783 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9784 {
9785 	if (is_guest_mode(vcpu))
9786 		kvm_check_nested_events(vcpu);
9787 
9788 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9789 		!vcpu->arch.apf.halted);
9790 }
9791 
9792 static int vcpu_run(struct kvm_vcpu *vcpu)
9793 {
9794 	int r;
9795 	struct kvm *kvm = vcpu->kvm;
9796 
9797 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9798 	vcpu->arch.l1tf_flush_l1d = true;
9799 
9800 	for (;;) {
9801 		if (kvm_vcpu_running(vcpu)) {
9802 			r = vcpu_enter_guest(vcpu);
9803 		} else {
9804 			r = vcpu_block(kvm, vcpu);
9805 		}
9806 
9807 		if (r <= 0)
9808 			break;
9809 
9810 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9811 		if (kvm_cpu_has_pending_timer(vcpu))
9812 			kvm_inject_pending_timer_irqs(vcpu);
9813 
9814 		if (dm_request_for_irq_injection(vcpu) &&
9815 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9816 			r = 0;
9817 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9818 			++vcpu->stat.request_irq_exits;
9819 			break;
9820 		}
9821 
9822 		if (__xfer_to_guest_mode_work_pending()) {
9823 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9824 			r = xfer_to_guest_mode_handle_work(vcpu);
9825 			if (r)
9826 				return r;
9827 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9828 		}
9829 	}
9830 
9831 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9832 
9833 	return r;
9834 }
9835 
9836 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9837 {
9838 	int r;
9839 
9840 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9841 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9842 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9843 	return r;
9844 }
9845 
9846 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9847 {
9848 	BUG_ON(!vcpu->arch.pio.count);
9849 
9850 	return complete_emulated_io(vcpu);
9851 }
9852 
9853 /*
9854  * Implements the following, as a state machine:
9855  *
9856  * read:
9857  *   for each fragment
9858  *     for each mmio piece in the fragment
9859  *       write gpa, len
9860  *       exit
9861  *       copy data
9862  *   execute insn
9863  *
9864  * write:
9865  *   for each fragment
9866  *     for each mmio piece in the fragment
9867  *       write gpa, len
9868  *       copy data
9869  *       exit
9870  */
9871 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9872 {
9873 	struct kvm_run *run = vcpu->run;
9874 	struct kvm_mmio_fragment *frag;
9875 	unsigned len;
9876 
9877 	BUG_ON(!vcpu->mmio_needed);
9878 
9879 	/* Complete previous fragment */
9880 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9881 	len = min(8u, frag->len);
9882 	if (!vcpu->mmio_is_write)
9883 		memcpy(frag->data, run->mmio.data, len);
9884 
9885 	if (frag->len <= 8) {
9886 		/* Switch to the next fragment. */
9887 		frag++;
9888 		vcpu->mmio_cur_fragment++;
9889 	} else {
9890 		/* Go forward to the next mmio piece. */
9891 		frag->data += len;
9892 		frag->gpa += len;
9893 		frag->len -= len;
9894 	}
9895 
9896 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9897 		vcpu->mmio_needed = 0;
9898 
9899 		/* FIXME: return into emulator if single-stepping.  */
9900 		if (vcpu->mmio_is_write)
9901 			return 1;
9902 		vcpu->mmio_read_completed = 1;
9903 		return complete_emulated_io(vcpu);
9904 	}
9905 
9906 	run->exit_reason = KVM_EXIT_MMIO;
9907 	run->mmio.phys_addr = frag->gpa;
9908 	if (vcpu->mmio_is_write)
9909 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9910 	run->mmio.len = min(8u, frag->len);
9911 	run->mmio.is_write = vcpu->mmio_is_write;
9912 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9913 	return 0;
9914 }
9915 
9916 static void kvm_save_current_fpu(struct fpu *fpu)
9917 {
9918 	/*
9919 	 * If the target FPU state is not resident in the CPU registers, just
9920 	 * memcpy() from current, else save CPU state directly to the target.
9921 	 */
9922 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9923 		memcpy(&fpu->state, &current->thread.fpu.state,
9924 		       fpu_kernel_xstate_size);
9925 	else
9926 		save_fpregs_to_fpstate(fpu);
9927 }
9928 
9929 /* Swap (qemu) user FPU context for the guest FPU context. */
9930 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9931 {
9932 	fpregs_lock();
9933 
9934 	kvm_save_current_fpu(vcpu->arch.user_fpu);
9935 
9936 	/*
9937 	 * Guests with protected state can't have it set by the hypervisor,
9938 	 * so skip trying to set it.
9939 	 */
9940 	if (vcpu->arch.guest_fpu)
9941 		/* PKRU is separately restored in kvm_x86_ops.run. */
9942 		__restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
9943 					~XFEATURE_MASK_PKRU);
9944 
9945 	fpregs_mark_activate();
9946 	fpregs_unlock();
9947 
9948 	trace_kvm_fpu(1);
9949 }
9950 
9951 /* When vcpu_run ends, restore user space FPU context. */
9952 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9953 {
9954 	fpregs_lock();
9955 
9956 	/*
9957 	 * Guests with protected state can't have it read by the hypervisor,
9958 	 * so skip trying to save it.
9959 	 */
9960 	if (vcpu->arch.guest_fpu)
9961 		kvm_save_current_fpu(vcpu->arch.guest_fpu);
9962 
9963 	restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
9964 
9965 	fpregs_mark_activate();
9966 	fpregs_unlock();
9967 
9968 	++vcpu->stat.fpu_reload;
9969 	trace_kvm_fpu(0);
9970 }
9971 
9972 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9973 {
9974 	struct kvm_run *kvm_run = vcpu->run;
9975 	int r;
9976 
9977 	vcpu_load(vcpu);
9978 	kvm_sigset_activate(vcpu);
9979 	kvm_run->flags = 0;
9980 	kvm_load_guest_fpu(vcpu);
9981 
9982 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9983 		if (kvm_run->immediate_exit) {
9984 			r = -EINTR;
9985 			goto out;
9986 		}
9987 		kvm_vcpu_block(vcpu);
9988 		if (kvm_apic_accept_events(vcpu) < 0) {
9989 			r = 0;
9990 			goto out;
9991 		}
9992 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9993 		r = -EAGAIN;
9994 		if (signal_pending(current)) {
9995 			r = -EINTR;
9996 			kvm_run->exit_reason = KVM_EXIT_INTR;
9997 			++vcpu->stat.signal_exits;
9998 		}
9999 		goto out;
10000 	}
10001 
10002 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10003 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10004 		r = -EINVAL;
10005 		goto out;
10006 	}
10007 
10008 	if (kvm_run->kvm_dirty_regs) {
10009 		r = sync_regs(vcpu);
10010 		if (r != 0)
10011 			goto out;
10012 	}
10013 
10014 	/* re-sync apic's tpr */
10015 	if (!lapic_in_kernel(vcpu)) {
10016 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10017 			r = -EINVAL;
10018 			goto out;
10019 		}
10020 	}
10021 
10022 	if (unlikely(vcpu->arch.complete_userspace_io)) {
10023 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10024 		vcpu->arch.complete_userspace_io = NULL;
10025 		r = cui(vcpu);
10026 		if (r <= 0)
10027 			goto out;
10028 	} else
10029 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10030 
10031 	if (kvm_run->immediate_exit)
10032 		r = -EINTR;
10033 	else
10034 		r = vcpu_run(vcpu);
10035 
10036 out:
10037 	kvm_put_guest_fpu(vcpu);
10038 	if (kvm_run->kvm_valid_regs)
10039 		store_regs(vcpu);
10040 	post_kvm_run_save(vcpu);
10041 	kvm_sigset_deactivate(vcpu);
10042 
10043 	vcpu_put(vcpu);
10044 	return r;
10045 }
10046 
10047 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10048 {
10049 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10050 		/*
10051 		 * We are here if userspace calls get_regs() in the middle of
10052 		 * instruction emulation. Registers state needs to be copied
10053 		 * back from emulation context to vcpu. Userspace shouldn't do
10054 		 * that usually, but some bad designed PV devices (vmware
10055 		 * backdoor interface) need this to work
10056 		 */
10057 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10058 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10059 	}
10060 	regs->rax = kvm_rax_read(vcpu);
10061 	regs->rbx = kvm_rbx_read(vcpu);
10062 	regs->rcx = kvm_rcx_read(vcpu);
10063 	regs->rdx = kvm_rdx_read(vcpu);
10064 	regs->rsi = kvm_rsi_read(vcpu);
10065 	regs->rdi = kvm_rdi_read(vcpu);
10066 	regs->rsp = kvm_rsp_read(vcpu);
10067 	regs->rbp = kvm_rbp_read(vcpu);
10068 #ifdef CONFIG_X86_64
10069 	regs->r8 = kvm_r8_read(vcpu);
10070 	regs->r9 = kvm_r9_read(vcpu);
10071 	regs->r10 = kvm_r10_read(vcpu);
10072 	regs->r11 = kvm_r11_read(vcpu);
10073 	regs->r12 = kvm_r12_read(vcpu);
10074 	regs->r13 = kvm_r13_read(vcpu);
10075 	regs->r14 = kvm_r14_read(vcpu);
10076 	regs->r15 = kvm_r15_read(vcpu);
10077 #endif
10078 
10079 	regs->rip = kvm_rip_read(vcpu);
10080 	regs->rflags = kvm_get_rflags(vcpu);
10081 }
10082 
10083 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10084 {
10085 	vcpu_load(vcpu);
10086 	__get_regs(vcpu, regs);
10087 	vcpu_put(vcpu);
10088 	return 0;
10089 }
10090 
10091 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10092 {
10093 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10094 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10095 
10096 	kvm_rax_write(vcpu, regs->rax);
10097 	kvm_rbx_write(vcpu, regs->rbx);
10098 	kvm_rcx_write(vcpu, regs->rcx);
10099 	kvm_rdx_write(vcpu, regs->rdx);
10100 	kvm_rsi_write(vcpu, regs->rsi);
10101 	kvm_rdi_write(vcpu, regs->rdi);
10102 	kvm_rsp_write(vcpu, regs->rsp);
10103 	kvm_rbp_write(vcpu, regs->rbp);
10104 #ifdef CONFIG_X86_64
10105 	kvm_r8_write(vcpu, regs->r8);
10106 	kvm_r9_write(vcpu, regs->r9);
10107 	kvm_r10_write(vcpu, regs->r10);
10108 	kvm_r11_write(vcpu, regs->r11);
10109 	kvm_r12_write(vcpu, regs->r12);
10110 	kvm_r13_write(vcpu, regs->r13);
10111 	kvm_r14_write(vcpu, regs->r14);
10112 	kvm_r15_write(vcpu, regs->r15);
10113 #endif
10114 
10115 	kvm_rip_write(vcpu, regs->rip);
10116 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10117 
10118 	vcpu->arch.exception.pending = false;
10119 
10120 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10121 }
10122 
10123 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10124 {
10125 	vcpu_load(vcpu);
10126 	__set_regs(vcpu, regs);
10127 	vcpu_put(vcpu);
10128 	return 0;
10129 }
10130 
10131 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10132 {
10133 	struct kvm_segment cs;
10134 
10135 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10136 	*db = cs.db;
10137 	*l = cs.l;
10138 }
10139 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10140 
10141 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10142 {
10143 	struct desc_ptr dt;
10144 
10145 	if (vcpu->arch.guest_state_protected)
10146 		goto skip_protected_regs;
10147 
10148 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10149 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10150 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10151 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10152 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10153 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10154 
10155 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10156 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10157 
10158 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10159 	sregs->idt.limit = dt.size;
10160 	sregs->idt.base = dt.address;
10161 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10162 	sregs->gdt.limit = dt.size;
10163 	sregs->gdt.base = dt.address;
10164 
10165 	sregs->cr2 = vcpu->arch.cr2;
10166 	sregs->cr3 = kvm_read_cr3(vcpu);
10167 
10168 skip_protected_regs:
10169 	sregs->cr0 = kvm_read_cr0(vcpu);
10170 	sregs->cr4 = kvm_read_cr4(vcpu);
10171 	sregs->cr8 = kvm_get_cr8(vcpu);
10172 	sregs->efer = vcpu->arch.efer;
10173 	sregs->apic_base = kvm_get_apic_base(vcpu);
10174 }
10175 
10176 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10177 {
10178 	__get_sregs_common(vcpu, sregs);
10179 
10180 	if (vcpu->arch.guest_state_protected)
10181 		return;
10182 
10183 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10184 		set_bit(vcpu->arch.interrupt.nr,
10185 			(unsigned long *)sregs->interrupt_bitmap);
10186 }
10187 
10188 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10189 {
10190 	int i;
10191 
10192 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10193 
10194 	if (vcpu->arch.guest_state_protected)
10195 		return;
10196 
10197 	if (is_pae_paging(vcpu)) {
10198 		for (i = 0 ; i < 4 ; i++)
10199 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10200 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10201 	}
10202 }
10203 
10204 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10205 				  struct kvm_sregs *sregs)
10206 {
10207 	vcpu_load(vcpu);
10208 	__get_sregs(vcpu, sregs);
10209 	vcpu_put(vcpu);
10210 	return 0;
10211 }
10212 
10213 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10214 				    struct kvm_mp_state *mp_state)
10215 {
10216 	int r;
10217 
10218 	vcpu_load(vcpu);
10219 	if (kvm_mpx_supported())
10220 		kvm_load_guest_fpu(vcpu);
10221 
10222 	r = kvm_apic_accept_events(vcpu);
10223 	if (r < 0)
10224 		goto out;
10225 	r = 0;
10226 
10227 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10228 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10229 	    vcpu->arch.pv.pv_unhalted)
10230 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10231 	else
10232 		mp_state->mp_state = vcpu->arch.mp_state;
10233 
10234 out:
10235 	if (kvm_mpx_supported())
10236 		kvm_put_guest_fpu(vcpu);
10237 	vcpu_put(vcpu);
10238 	return r;
10239 }
10240 
10241 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10242 				    struct kvm_mp_state *mp_state)
10243 {
10244 	int ret = -EINVAL;
10245 
10246 	vcpu_load(vcpu);
10247 
10248 	if (!lapic_in_kernel(vcpu) &&
10249 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10250 		goto out;
10251 
10252 	/*
10253 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10254 	 * INIT state; latched init should be reported using
10255 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10256 	 */
10257 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10258 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10259 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10260 		goto out;
10261 
10262 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10263 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10264 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10265 	} else
10266 		vcpu->arch.mp_state = mp_state->mp_state;
10267 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10268 
10269 	ret = 0;
10270 out:
10271 	vcpu_put(vcpu);
10272 	return ret;
10273 }
10274 
10275 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10276 		    int reason, bool has_error_code, u32 error_code)
10277 {
10278 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10279 	int ret;
10280 
10281 	init_emulate_ctxt(vcpu);
10282 
10283 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10284 				   has_error_code, error_code);
10285 	if (ret) {
10286 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10287 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10288 		vcpu->run->internal.ndata = 0;
10289 		return 0;
10290 	}
10291 
10292 	kvm_rip_write(vcpu, ctxt->eip);
10293 	kvm_set_rflags(vcpu, ctxt->eflags);
10294 	return 1;
10295 }
10296 EXPORT_SYMBOL_GPL(kvm_task_switch);
10297 
10298 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10299 {
10300 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10301 		/*
10302 		 * When EFER.LME and CR0.PG are set, the processor is in
10303 		 * 64-bit mode (though maybe in a 32-bit code segment).
10304 		 * CR4.PAE and EFER.LMA must be set.
10305 		 */
10306 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10307 			return false;
10308 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10309 			return false;
10310 	} else {
10311 		/*
10312 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10313 		 * segment cannot be 64-bit.
10314 		 */
10315 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10316 			return false;
10317 	}
10318 
10319 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10320 }
10321 
10322 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10323 		int *mmu_reset_needed, bool update_pdptrs)
10324 {
10325 	struct msr_data apic_base_msr;
10326 	int idx;
10327 	struct desc_ptr dt;
10328 
10329 	if (!kvm_is_valid_sregs(vcpu, sregs))
10330 		return -EINVAL;
10331 
10332 	apic_base_msr.data = sregs->apic_base;
10333 	apic_base_msr.host_initiated = true;
10334 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10335 		return -EINVAL;
10336 
10337 	if (vcpu->arch.guest_state_protected)
10338 		return 0;
10339 
10340 	dt.size = sregs->idt.limit;
10341 	dt.address = sregs->idt.base;
10342 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10343 	dt.size = sregs->gdt.limit;
10344 	dt.address = sregs->gdt.base;
10345 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10346 
10347 	vcpu->arch.cr2 = sregs->cr2;
10348 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10349 	vcpu->arch.cr3 = sregs->cr3;
10350 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10351 
10352 	kvm_set_cr8(vcpu, sregs->cr8);
10353 
10354 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10355 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10356 
10357 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10358 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10359 	vcpu->arch.cr0 = sregs->cr0;
10360 
10361 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10362 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10363 
10364 	if (update_pdptrs) {
10365 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10366 		if (is_pae_paging(vcpu)) {
10367 			load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10368 			*mmu_reset_needed = 1;
10369 		}
10370 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10371 	}
10372 
10373 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10374 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10375 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10376 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10377 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10378 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10379 
10380 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10381 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10382 
10383 	update_cr8_intercept(vcpu);
10384 
10385 	/* Older userspace won't unhalt the vcpu on reset. */
10386 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10387 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10388 	    !is_protmode(vcpu))
10389 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10390 
10391 	return 0;
10392 }
10393 
10394 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10395 {
10396 	int pending_vec, max_bits;
10397 	int mmu_reset_needed = 0;
10398 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10399 
10400 	if (ret)
10401 		return ret;
10402 
10403 	if (mmu_reset_needed)
10404 		kvm_mmu_reset_context(vcpu);
10405 
10406 	max_bits = KVM_NR_INTERRUPTS;
10407 	pending_vec = find_first_bit(
10408 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10409 
10410 	if (pending_vec < max_bits) {
10411 		kvm_queue_interrupt(vcpu, pending_vec, false);
10412 		pr_debug("Set back pending irq %d\n", pending_vec);
10413 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10414 	}
10415 	return 0;
10416 }
10417 
10418 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10419 {
10420 	int mmu_reset_needed = 0;
10421 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10422 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10423 		!(sregs2->efer & EFER_LMA);
10424 	int i, ret;
10425 
10426 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10427 		return -EINVAL;
10428 
10429 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10430 		return -EINVAL;
10431 
10432 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10433 				 &mmu_reset_needed, !valid_pdptrs);
10434 	if (ret)
10435 		return ret;
10436 
10437 	if (valid_pdptrs) {
10438 		for (i = 0; i < 4 ; i++)
10439 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10440 
10441 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10442 		mmu_reset_needed = 1;
10443 		vcpu->arch.pdptrs_from_userspace = true;
10444 	}
10445 	if (mmu_reset_needed)
10446 		kvm_mmu_reset_context(vcpu);
10447 	return 0;
10448 }
10449 
10450 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10451 				  struct kvm_sregs *sregs)
10452 {
10453 	int ret;
10454 
10455 	vcpu_load(vcpu);
10456 	ret = __set_sregs(vcpu, sregs);
10457 	vcpu_put(vcpu);
10458 	return ret;
10459 }
10460 
10461 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10462 					struct kvm_guest_debug *dbg)
10463 {
10464 	unsigned long rflags;
10465 	int i, r;
10466 
10467 	if (vcpu->arch.guest_state_protected)
10468 		return -EINVAL;
10469 
10470 	vcpu_load(vcpu);
10471 
10472 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10473 		r = -EBUSY;
10474 		if (vcpu->arch.exception.pending)
10475 			goto out;
10476 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10477 			kvm_queue_exception(vcpu, DB_VECTOR);
10478 		else
10479 			kvm_queue_exception(vcpu, BP_VECTOR);
10480 	}
10481 
10482 	/*
10483 	 * Read rflags as long as potentially injected trace flags are still
10484 	 * filtered out.
10485 	 */
10486 	rflags = kvm_get_rflags(vcpu);
10487 
10488 	vcpu->guest_debug = dbg->control;
10489 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10490 		vcpu->guest_debug = 0;
10491 
10492 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10493 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10494 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10495 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10496 	} else {
10497 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10498 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10499 	}
10500 	kvm_update_dr7(vcpu);
10501 
10502 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10503 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10504 
10505 	/*
10506 	 * Trigger an rflags update that will inject or remove the trace
10507 	 * flags.
10508 	 */
10509 	kvm_set_rflags(vcpu, rflags);
10510 
10511 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10512 
10513 	r = 0;
10514 
10515 out:
10516 	vcpu_put(vcpu);
10517 	return r;
10518 }
10519 
10520 /*
10521  * Translate a guest virtual address to a guest physical address.
10522  */
10523 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10524 				    struct kvm_translation *tr)
10525 {
10526 	unsigned long vaddr = tr->linear_address;
10527 	gpa_t gpa;
10528 	int idx;
10529 
10530 	vcpu_load(vcpu);
10531 
10532 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10533 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10534 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10535 	tr->physical_address = gpa;
10536 	tr->valid = gpa != UNMAPPED_GVA;
10537 	tr->writeable = 1;
10538 	tr->usermode = 0;
10539 
10540 	vcpu_put(vcpu);
10541 	return 0;
10542 }
10543 
10544 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10545 {
10546 	struct fxregs_state *fxsave;
10547 
10548 	if (!vcpu->arch.guest_fpu)
10549 		return 0;
10550 
10551 	vcpu_load(vcpu);
10552 
10553 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10554 	memcpy(fpu->fpr, fxsave->st_space, 128);
10555 	fpu->fcw = fxsave->cwd;
10556 	fpu->fsw = fxsave->swd;
10557 	fpu->ftwx = fxsave->twd;
10558 	fpu->last_opcode = fxsave->fop;
10559 	fpu->last_ip = fxsave->rip;
10560 	fpu->last_dp = fxsave->rdp;
10561 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10562 
10563 	vcpu_put(vcpu);
10564 	return 0;
10565 }
10566 
10567 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10568 {
10569 	struct fxregs_state *fxsave;
10570 
10571 	if (!vcpu->arch.guest_fpu)
10572 		return 0;
10573 
10574 	vcpu_load(vcpu);
10575 
10576 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10577 
10578 	memcpy(fxsave->st_space, fpu->fpr, 128);
10579 	fxsave->cwd = fpu->fcw;
10580 	fxsave->swd = fpu->fsw;
10581 	fxsave->twd = fpu->ftwx;
10582 	fxsave->fop = fpu->last_opcode;
10583 	fxsave->rip = fpu->last_ip;
10584 	fxsave->rdp = fpu->last_dp;
10585 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10586 
10587 	vcpu_put(vcpu);
10588 	return 0;
10589 }
10590 
10591 static void store_regs(struct kvm_vcpu *vcpu)
10592 {
10593 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10594 
10595 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10596 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10597 
10598 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10599 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10600 
10601 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10602 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10603 				vcpu, &vcpu->run->s.regs.events);
10604 }
10605 
10606 static int sync_regs(struct kvm_vcpu *vcpu)
10607 {
10608 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10609 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10610 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10611 	}
10612 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10613 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10614 			return -EINVAL;
10615 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10616 	}
10617 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10618 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10619 				vcpu, &vcpu->run->s.regs.events))
10620 			return -EINVAL;
10621 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10622 	}
10623 
10624 	return 0;
10625 }
10626 
10627 static void fx_init(struct kvm_vcpu *vcpu)
10628 {
10629 	if (!vcpu->arch.guest_fpu)
10630 		return;
10631 
10632 	fpstate_init(&vcpu->arch.guest_fpu->state);
10633 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10634 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10635 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
10636 
10637 	/*
10638 	 * Ensure guest xcr0 is valid for loading
10639 	 */
10640 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10641 
10642 	vcpu->arch.cr0 |= X86_CR0_ET;
10643 }
10644 
10645 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10646 {
10647 	if (vcpu->arch.guest_fpu) {
10648 		kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10649 		vcpu->arch.guest_fpu = NULL;
10650 	}
10651 }
10652 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10653 
10654 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10655 {
10656 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10657 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10658 			     "guest TSC will not be reliable\n");
10659 
10660 	return 0;
10661 }
10662 
10663 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10664 {
10665 	struct page *page;
10666 	int r;
10667 
10668 	vcpu->arch.last_vmentry_cpu = -1;
10669 	vcpu->arch.regs_avail = ~0;
10670 	vcpu->arch.regs_dirty = ~0;
10671 
10672 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10673 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10674 	else
10675 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10676 
10677 	r = kvm_mmu_create(vcpu);
10678 	if (r < 0)
10679 		return r;
10680 
10681 	if (irqchip_in_kernel(vcpu->kvm)) {
10682 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10683 		if (r < 0)
10684 			goto fail_mmu_destroy;
10685 		if (kvm_apicv_activated(vcpu->kvm))
10686 			vcpu->arch.apicv_active = true;
10687 	} else
10688 		static_branch_inc(&kvm_has_noapic_vcpu);
10689 
10690 	r = -ENOMEM;
10691 
10692 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10693 	if (!page)
10694 		goto fail_free_lapic;
10695 	vcpu->arch.pio_data = page_address(page);
10696 
10697 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10698 				       GFP_KERNEL_ACCOUNT);
10699 	if (!vcpu->arch.mce_banks)
10700 		goto fail_free_pio_data;
10701 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10702 
10703 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10704 				GFP_KERNEL_ACCOUNT))
10705 		goto fail_free_mce_banks;
10706 
10707 	if (!alloc_emulate_ctxt(vcpu))
10708 		goto free_wbinvd_dirty_mask;
10709 
10710 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10711 						GFP_KERNEL_ACCOUNT);
10712 	if (!vcpu->arch.user_fpu) {
10713 		pr_err("kvm: failed to allocate userspace's fpu\n");
10714 		goto free_emulate_ctxt;
10715 	}
10716 
10717 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10718 						 GFP_KERNEL_ACCOUNT);
10719 	if (!vcpu->arch.guest_fpu) {
10720 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10721 		goto free_user_fpu;
10722 	}
10723 	fx_init(vcpu);
10724 
10725 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10726 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10727 
10728 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10729 
10730 	kvm_async_pf_hash_reset(vcpu);
10731 	kvm_pmu_init(vcpu);
10732 
10733 	vcpu->arch.pending_external_vector = -1;
10734 	vcpu->arch.preempted_in_kernel = false;
10735 
10736 #if IS_ENABLED(CONFIG_HYPERV)
10737 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
10738 #endif
10739 
10740 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10741 	if (r)
10742 		goto free_guest_fpu;
10743 
10744 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10745 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10746 	kvm_vcpu_mtrr_init(vcpu);
10747 	vcpu_load(vcpu);
10748 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10749 	kvm_vcpu_reset(vcpu, false);
10750 	kvm_init_mmu(vcpu);
10751 	vcpu_put(vcpu);
10752 	return 0;
10753 
10754 free_guest_fpu:
10755 	kvm_free_guest_fpu(vcpu);
10756 free_user_fpu:
10757 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10758 free_emulate_ctxt:
10759 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10760 free_wbinvd_dirty_mask:
10761 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10762 fail_free_mce_banks:
10763 	kfree(vcpu->arch.mce_banks);
10764 fail_free_pio_data:
10765 	free_page((unsigned long)vcpu->arch.pio_data);
10766 fail_free_lapic:
10767 	kvm_free_lapic(vcpu);
10768 fail_mmu_destroy:
10769 	kvm_mmu_destroy(vcpu);
10770 	return r;
10771 }
10772 
10773 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10774 {
10775 	struct kvm *kvm = vcpu->kvm;
10776 
10777 	if (mutex_lock_killable(&vcpu->mutex))
10778 		return;
10779 	vcpu_load(vcpu);
10780 	kvm_synchronize_tsc(vcpu, 0);
10781 	vcpu_put(vcpu);
10782 
10783 	/* poll control enabled by default */
10784 	vcpu->arch.msr_kvm_poll_control = 1;
10785 
10786 	mutex_unlock(&vcpu->mutex);
10787 
10788 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10789 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10790 						KVMCLOCK_SYNC_PERIOD);
10791 }
10792 
10793 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10794 {
10795 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10796 	int idx;
10797 
10798 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10799 
10800 	kvmclock_reset(vcpu);
10801 
10802 	static_call(kvm_x86_vcpu_free)(vcpu);
10803 
10804 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10805 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10806 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10807 	kvm_free_guest_fpu(vcpu);
10808 
10809 	kvm_hv_vcpu_uninit(vcpu);
10810 	kvm_pmu_destroy(vcpu);
10811 	kfree(vcpu->arch.mce_banks);
10812 	kvm_free_lapic(vcpu);
10813 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10814 	kvm_mmu_destroy(vcpu);
10815 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10816 	free_page((unsigned long)vcpu->arch.pio_data);
10817 	kvfree(vcpu->arch.cpuid_entries);
10818 	if (!lapic_in_kernel(vcpu))
10819 		static_branch_dec(&kvm_has_noapic_vcpu);
10820 }
10821 
10822 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10823 {
10824 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
10825 	unsigned long new_cr0;
10826 	u32 eax, dummy;
10827 
10828 	kvm_lapic_reset(vcpu, init_event);
10829 
10830 	vcpu->arch.hflags = 0;
10831 
10832 	vcpu->arch.smi_pending = 0;
10833 	vcpu->arch.smi_count = 0;
10834 	atomic_set(&vcpu->arch.nmi_queued, 0);
10835 	vcpu->arch.nmi_pending = 0;
10836 	vcpu->arch.nmi_injected = false;
10837 	kvm_clear_interrupt_queue(vcpu);
10838 	kvm_clear_exception_queue(vcpu);
10839 
10840 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10841 	kvm_update_dr0123(vcpu);
10842 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10843 	vcpu->arch.dr7 = DR7_FIXED_1;
10844 	kvm_update_dr7(vcpu);
10845 
10846 	vcpu->arch.cr2 = 0;
10847 
10848 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10849 	vcpu->arch.apf.msr_en_val = 0;
10850 	vcpu->arch.apf.msr_int_val = 0;
10851 	vcpu->arch.st.msr_val = 0;
10852 
10853 	kvmclock_reset(vcpu);
10854 
10855 	kvm_clear_async_pf_completion_queue(vcpu);
10856 	kvm_async_pf_hash_reset(vcpu);
10857 	vcpu->arch.apf.halted = false;
10858 
10859 	if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10860 		void *mpx_state_buffer;
10861 
10862 		/*
10863 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10864 		 * called with loaded FPU and does not let userspace fix the state.
10865 		 */
10866 		if (init_event)
10867 			kvm_put_guest_fpu(vcpu);
10868 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10869 					XFEATURE_BNDREGS);
10870 		if (mpx_state_buffer)
10871 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10872 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10873 					XFEATURE_BNDCSR);
10874 		if (mpx_state_buffer)
10875 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10876 		if (init_event)
10877 			kvm_load_guest_fpu(vcpu);
10878 	}
10879 
10880 	if (!init_event) {
10881 		kvm_pmu_reset(vcpu);
10882 		vcpu->arch.smbase = 0x30000;
10883 
10884 		vcpu->arch.msr_misc_features_enables = 0;
10885 
10886 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10887 	}
10888 
10889 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10890 	vcpu->arch.regs_avail = ~0;
10891 	vcpu->arch.regs_dirty = ~0;
10892 
10893 	/*
10894 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
10895 	 * if no CPUID match is found.  Note, it's impossible to get a match at
10896 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
10897 	 * i.e. it'simpossible for kvm_cpuid() to find a valid entry on RESET.
10898 	 * But, go through the motions in case that's ever remedied.
10899 	 */
10900 	eax = 1;
10901 	if (!kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true))
10902 		eax = 0x600;
10903 	kvm_rdx_write(vcpu, eax);
10904 
10905 	vcpu->arch.ia32_xss = 0;
10906 
10907 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10908 
10909 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
10910 	kvm_rip_write(vcpu, 0xfff0);
10911 
10912 	vcpu->arch.cr3 = 0;
10913 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10914 
10915 	/*
10916 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
10917 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
10918 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
10919 	 */
10920 	new_cr0 = X86_CR0_ET;
10921 	if (init_event)
10922 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
10923 	else
10924 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
10925 
10926 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
10927 	static_call(kvm_x86_set_cr4)(vcpu, 0);
10928 	static_call(kvm_x86_set_efer)(vcpu, 0);
10929 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10930 
10931 	/*
10932 	 * Reset the MMU context if paging was enabled prior to INIT (which is
10933 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
10934 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
10935 	 * checked because it is unconditionally cleared on INIT and all other
10936 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
10937 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
10938 	 */
10939 	if (old_cr0 & X86_CR0_PG)
10940 		kvm_mmu_reset_context(vcpu);
10941 
10942 	/*
10943 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
10944 	 * APM states the TLBs are untouched by INIT, but it also states that
10945 	 * the TLBs are flushed on "External initialization of the processor."
10946 	 * Flush the guest TLB regardless of vendor, there is no meaningful
10947 	 * benefit in relying on the guest to flush the TLB immediately after
10948 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
10949 	 * performance perspective.
10950 	 */
10951 	if (init_event)
10952 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
10953 }
10954 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
10955 
10956 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10957 {
10958 	struct kvm_segment cs;
10959 
10960 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10961 	cs.selector = vector << 8;
10962 	cs.base = vector << 12;
10963 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10964 	kvm_rip_write(vcpu, 0);
10965 }
10966 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10967 
10968 int kvm_arch_hardware_enable(void)
10969 {
10970 	struct kvm *kvm;
10971 	struct kvm_vcpu *vcpu;
10972 	int i;
10973 	int ret;
10974 	u64 local_tsc;
10975 	u64 max_tsc = 0;
10976 	bool stable, backwards_tsc = false;
10977 
10978 	kvm_user_return_msr_cpu_online();
10979 	ret = static_call(kvm_x86_hardware_enable)();
10980 	if (ret != 0)
10981 		return ret;
10982 
10983 	local_tsc = rdtsc();
10984 	stable = !kvm_check_tsc_unstable();
10985 	list_for_each_entry(kvm, &vm_list, vm_list) {
10986 		kvm_for_each_vcpu(i, vcpu, kvm) {
10987 			if (!stable && vcpu->cpu == smp_processor_id())
10988 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10989 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10990 				backwards_tsc = true;
10991 				if (vcpu->arch.last_host_tsc > max_tsc)
10992 					max_tsc = vcpu->arch.last_host_tsc;
10993 			}
10994 		}
10995 	}
10996 
10997 	/*
10998 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10999 	 * platforms that reset TSC during suspend or hibernate actions, but
11000 	 * maintain synchronization.  We must compensate.  Fortunately, we can
11001 	 * detect that condition here, which happens early in CPU bringup,
11002 	 * before any KVM threads can be running.  Unfortunately, we can't
11003 	 * bring the TSCs fully up to date with real time, as we aren't yet far
11004 	 * enough into CPU bringup that we know how much real time has actually
11005 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11006 	 * variables that haven't been updated yet.
11007 	 *
11008 	 * So we simply find the maximum observed TSC above, then record the
11009 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
11010 	 * the adjustment will be applied.  Note that we accumulate
11011 	 * adjustments, in case multiple suspend cycles happen before some VCPU
11012 	 * gets a chance to run again.  In the event that no KVM threads get a
11013 	 * chance to run, we will miss the entire elapsed period, as we'll have
11014 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11015 	 * loose cycle time.  This isn't too big a deal, since the loss will be
11016 	 * uniform across all VCPUs (not to mention the scenario is extremely
11017 	 * unlikely). It is possible that a second hibernate recovery happens
11018 	 * much faster than a first, causing the observed TSC here to be
11019 	 * smaller; this would require additional padding adjustment, which is
11020 	 * why we set last_host_tsc to the local tsc observed here.
11021 	 *
11022 	 * N.B. - this code below runs only on platforms with reliable TSC,
11023 	 * as that is the only way backwards_tsc is set above.  Also note
11024 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11025 	 * have the same delta_cyc adjustment applied if backwards_tsc
11026 	 * is detected.  Note further, this adjustment is only done once,
11027 	 * as we reset last_host_tsc on all VCPUs to stop this from being
11028 	 * called multiple times (one for each physical CPU bringup).
11029 	 *
11030 	 * Platforms with unreliable TSCs don't have to deal with this, they
11031 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
11032 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
11033 	 * guarantee that they stay in perfect synchronization.
11034 	 */
11035 	if (backwards_tsc) {
11036 		u64 delta_cyc = max_tsc - local_tsc;
11037 		list_for_each_entry(kvm, &vm_list, vm_list) {
11038 			kvm->arch.backwards_tsc_observed = true;
11039 			kvm_for_each_vcpu(i, vcpu, kvm) {
11040 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
11041 				vcpu->arch.last_host_tsc = local_tsc;
11042 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11043 			}
11044 
11045 			/*
11046 			 * We have to disable TSC offset matching.. if you were
11047 			 * booting a VM while issuing an S4 host suspend....
11048 			 * you may have some problem.  Solving this issue is
11049 			 * left as an exercise to the reader.
11050 			 */
11051 			kvm->arch.last_tsc_nsec = 0;
11052 			kvm->arch.last_tsc_write = 0;
11053 		}
11054 
11055 	}
11056 	return 0;
11057 }
11058 
11059 void kvm_arch_hardware_disable(void)
11060 {
11061 	static_call(kvm_x86_hardware_disable)();
11062 	drop_user_return_notifiers();
11063 }
11064 
11065 int kvm_arch_hardware_setup(void *opaque)
11066 {
11067 	struct kvm_x86_init_ops *ops = opaque;
11068 	int r;
11069 
11070 	rdmsrl_safe(MSR_EFER, &host_efer);
11071 
11072 	if (boot_cpu_has(X86_FEATURE_XSAVES))
11073 		rdmsrl(MSR_IA32_XSS, host_xss);
11074 
11075 	r = ops->hardware_setup();
11076 	if (r != 0)
11077 		return r;
11078 
11079 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11080 	kvm_ops_static_call_update();
11081 
11082 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11083 		supported_xss = 0;
11084 
11085 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11086 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11087 #undef __kvm_cpu_cap_has
11088 
11089 	if (kvm_has_tsc_control) {
11090 		/*
11091 		 * Make sure the user can only configure tsc_khz values that
11092 		 * fit into a signed integer.
11093 		 * A min value is not calculated because it will always
11094 		 * be 1 on all machines.
11095 		 */
11096 		u64 max = min(0x7fffffffULL,
11097 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11098 		kvm_max_guest_tsc_khz = max;
11099 
11100 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11101 	}
11102 
11103 	kvm_init_msr_list();
11104 	return 0;
11105 }
11106 
11107 void kvm_arch_hardware_unsetup(void)
11108 {
11109 	static_call(kvm_x86_hardware_unsetup)();
11110 }
11111 
11112 int kvm_arch_check_processor_compat(void *opaque)
11113 {
11114 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11115 	struct kvm_x86_init_ops *ops = opaque;
11116 
11117 	WARN_ON(!irqs_disabled());
11118 
11119 	if (__cr4_reserved_bits(cpu_has, c) !=
11120 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11121 		return -EIO;
11122 
11123 	return ops->check_processor_compatibility();
11124 }
11125 
11126 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11127 {
11128 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11129 }
11130 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11131 
11132 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11133 {
11134 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11135 }
11136 
11137 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11138 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11139 
11140 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11141 {
11142 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11143 
11144 	vcpu->arch.l1tf_flush_l1d = true;
11145 	if (pmu->version && unlikely(pmu->event_count)) {
11146 		pmu->need_cleanup = true;
11147 		kvm_make_request(KVM_REQ_PMU, vcpu);
11148 	}
11149 	static_call(kvm_x86_sched_in)(vcpu, cpu);
11150 }
11151 
11152 void kvm_arch_free_vm(struct kvm *kvm)
11153 {
11154 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
11155 	vfree(kvm);
11156 }
11157 
11158 
11159 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11160 {
11161 	int ret;
11162 
11163 	if (type)
11164 		return -EINVAL;
11165 
11166 	ret = kvm_page_track_init(kvm);
11167 	if (ret)
11168 		return ret;
11169 
11170 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11171 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11172 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11173 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11174 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11175 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11176 
11177 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11178 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11179 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11180 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11181 		&kvm->arch.irq_sources_bitmap);
11182 
11183 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11184 	mutex_init(&kvm->arch.apic_map_lock);
11185 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11186 
11187 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11188 	pvclock_update_vm_gtod_copy(kvm);
11189 
11190 	kvm->arch.guest_can_read_msr_platform_info = true;
11191 
11192 #if IS_ENABLED(CONFIG_HYPERV)
11193 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11194 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11195 #endif
11196 
11197 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11198 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11199 
11200 	kvm_apicv_init(kvm);
11201 	kvm_hv_init_vm(kvm);
11202 	kvm_mmu_init_vm(kvm);
11203 	kvm_xen_init_vm(kvm);
11204 
11205 	return static_call(kvm_x86_vm_init)(kvm);
11206 }
11207 
11208 int kvm_arch_post_init_vm(struct kvm *kvm)
11209 {
11210 	return kvm_mmu_post_init_vm(kvm);
11211 }
11212 
11213 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11214 {
11215 	vcpu_load(vcpu);
11216 	kvm_mmu_unload(vcpu);
11217 	vcpu_put(vcpu);
11218 }
11219 
11220 static void kvm_free_vcpus(struct kvm *kvm)
11221 {
11222 	unsigned int i;
11223 	struct kvm_vcpu *vcpu;
11224 
11225 	/*
11226 	 * Unpin any mmu pages first.
11227 	 */
11228 	kvm_for_each_vcpu(i, vcpu, kvm) {
11229 		kvm_clear_async_pf_completion_queue(vcpu);
11230 		kvm_unload_vcpu_mmu(vcpu);
11231 	}
11232 	kvm_for_each_vcpu(i, vcpu, kvm)
11233 		kvm_vcpu_destroy(vcpu);
11234 
11235 	mutex_lock(&kvm->lock);
11236 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11237 		kvm->vcpus[i] = NULL;
11238 
11239 	atomic_set(&kvm->online_vcpus, 0);
11240 	mutex_unlock(&kvm->lock);
11241 }
11242 
11243 void kvm_arch_sync_events(struct kvm *kvm)
11244 {
11245 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11246 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11247 	kvm_free_pit(kvm);
11248 }
11249 
11250 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11251 
11252 /**
11253  * __x86_set_memory_region: Setup KVM internal memory slot
11254  *
11255  * @kvm: the kvm pointer to the VM.
11256  * @id: the slot ID to setup.
11257  * @gpa: the GPA to install the slot (unused when @size == 0).
11258  * @size: the size of the slot. Set to zero to uninstall a slot.
11259  *
11260  * This function helps to setup a KVM internal memory slot.  Specify
11261  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11262  * slot.  The return code can be one of the following:
11263  *
11264  *   HVA:           on success (uninstall will return a bogus HVA)
11265  *   -errno:        on error
11266  *
11267  * The caller should always use IS_ERR() to check the return value
11268  * before use.  Note, the KVM internal memory slots are guaranteed to
11269  * remain valid and unchanged until the VM is destroyed, i.e., the
11270  * GPA->HVA translation will not change.  However, the HVA is a user
11271  * address, i.e. its accessibility is not guaranteed, and must be
11272  * accessed via __copy_{to,from}_user().
11273  */
11274 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11275 				      u32 size)
11276 {
11277 	int i, r;
11278 	unsigned long hva, old_npages;
11279 	struct kvm_memslots *slots = kvm_memslots(kvm);
11280 	struct kvm_memory_slot *slot;
11281 
11282 	/* Called with kvm->slots_lock held.  */
11283 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11284 		return ERR_PTR_USR(-EINVAL);
11285 
11286 	slot = id_to_memslot(slots, id);
11287 	if (size) {
11288 		if (slot && slot->npages)
11289 			return ERR_PTR_USR(-EEXIST);
11290 
11291 		/*
11292 		 * MAP_SHARED to prevent internal slot pages from being moved
11293 		 * by fork()/COW.
11294 		 */
11295 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11296 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11297 		if (IS_ERR((void *)hva))
11298 			return (void __user *)hva;
11299 	} else {
11300 		if (!slot || !slot->npages)
11301 			return NULL;
11302 
11303 		old_npages = slot->npages;
11304 		hva = slot->userspace_addr;
11305 	}
11306 
11307 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11308 		struct kvm_userspace_memory_region m;
11309 
11310 		m.slot = id | (i << 16);
11311 		m.flags = 0;
11312 		m.guest_phys_addr = gpa;
11313 		m.userspace_addr = hva;
11314 		m.memory_size = size;
11315 		r = __kvm_set_memory_region(kvm, &m);
11316 		if (r < 0)
11317 			return ERR_PTR_USR(r);
11318 	}
11319 
11320 	if (!size)
11321 		vm_munmap(hva, old_npages * PAGE_SIZE);
11322 
11323 	return (void __user *)hva;
11324 }
11325 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11326 
11327 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11328 {
11329 	kvm_mmu_pre_destroy_vm(kvm);
11330 }
11331 
11332 void kvm_arch_destroy_vm(struct kvm *kvm)
11333 {
11334 	if (current->mm == kvm->mm) {
11335 		/*
11336 		 * Free memory regions allocated on behalf of userspace,
11337 		 * unless the the memory map has changed due to process exit
11338 		 * or fd copying.
11339 		 */
11340 		mutex_lock(&kvm->slots_lock);
11341 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11342 					0, 0);
11343 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11344 					0, 0);
11345 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11346 		mutex_unlock(&kvm->slots_lock);
11347 	}
11348 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11349 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11350 	kvm_pic_destroy(kvm);
11351 	kvm_ioapic_destroy(kvm);
11352 	kvm_free_vcpus(kvm);
11353 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11354 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11355 	kvm_mmu_uninit_vm(kvm);
11356 	kvm_page_track_cleanup(kvm);
11357 	kvm_xen_destroy_vm(kvm);
11358 	kvm_hv_destroy_vm(kvm);
11359 }
11360 
11361 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11362 {
11363 	int i;
11364 
11365 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11366 		kvfree(slot->arch.rmap[i]);
11367 		slot->arch.rmap[i] = NULL;
11368 	}
11369 }
11370 
11371 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11372 {
11373 	int i;
11374 
11375 	memslot_rmap_free(slot);
11376 
11377 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11378 		kvfree(slot->arch.lpage_info[i - 1]);
11379 		slot->arch.lpage_info[i - 1] = NULL;
11380 	}
11381 
11382 	kvm_page_track_free_memslot(slot);
11383 }
11384 
11385 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11386 			      unsigned long npages)
11387 {
11388 	const int sz = sizeof(*slot->arch.rmap[0]);
11389 	int i;
11390 
11391 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11392 		int level = i + 1;
11393 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11394 
11395 		WARN_ON(slot->arch.rmap[i]);
11396 
11397 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11398 		if (!slot->arch.rmap[i]) {
11399 			memslot_rmap_free(slot);
11400 			return -ENOMEM;
11401 		}
11402 	}
11403 
11404 	return 0;
11405 }
11406 
11407 int alloc_all_memslots_rmaps(struct kvm *kvm)
11408 {
11409 	struct kvm_memslots *slots;
11410 	struct kvm_memory_slot *slot;
11411 	int r, i;
11412 
11413 	/*
11414 	 * Check if memslots alreday have rmaps early before acquiring
11415 	 * the slots_arch_lock below.
11416 	 */
11417 	if (kvm_memslots_have_rmaps(kvm))
11418 		return 0;
11419 
11420 	mutex_lock(&kvm->slots_arch_lock);
11421 
11422 	/*
11423 	 * Read memslots_have_rmaps again, under the slots arch lock,
11424 	 * before allocating the rmaps
11425 	 */
11426 	if (kvm_memslots_have_rmaps(kvm)) {
11427 		mutex_unlock(&kvm->slots_arch_lock);
11428 		return 0;
11429 	}
11430 
11431 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11432 		slots = __kvm_memslots(kvm, i);
11433 		kvm_for_each_memslot(slot, slots) {
11434 			r = memslot_rmap_alloc(slot, slot->npages);
11435 			if (r) {
11436 				mutex_unlock(&kvm->slots_arch_lock);
11437 				return r;
11438 			}
11439 		}
11440 	}
11441 
11442 	/*
11443 	 * Ensure that memslots_have_rmaps becomes true strictly after
11444 	 * all the rmap pointers are set.
11445 	 */
11446 	smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11447 	mutex_unlock(&kvm->slots_arch_lock);
11448 	return 0;
11449 }
11450 
11451 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11452 				      struct kvm_memory_slot *slot,
11453 				      unsigned long npages)
11454 {
11455 	int i, r;
11456 
11457 	/*
11458 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11459 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11460 	 * the new memslot is successful.
11461 	 */
11462 	memset(&slot->arch, 0, sizeof(slot->arch));
11463 
11464 	if (kvm_memslots_have_rmaps(kvm)) {
11465 		r = memslot_rmap_alloc(slot, npages);
11466 		if (r)
11467 			return r;
11468 	}
11469 
11470 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11471 		struct kvm_lpage_info *linfo;
11472 		unsigned long ugfn;
11473 		int lpages;
11474 		int level = i + 1;
11475 
11476 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11477 
11478 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11479 		if (!linfo)
11480 			goto out_free;
11481 
11482 		slot->arch.lpage_info[i - 1] = linfo;
11483 
11484 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11485 			linfo[0].disallow_lpage = 1;
11486 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11487 			linfo[lpages - 1].disallow_lpage = 1;
11488 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11489 		/*
11490 		 * If the gfn and userspace address are not aligned wrt each
11491 		 * other, disable large page support for this slot.
11492 		 */
11493 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11494 			unsigned long j;
11495 
11496 			for (j = 0; j < lpages; ++j)
11497 				linfo[j].disallow_lpage = 1;
11498 		}
11499 	}
11500 
11501 	if (kvm_page_track_create_memslot(slot, npages))
11502 		goto out_free;
11503 
11504 	return 0;
11505 
11506 out_free:
11507 	memslot_rmap_free(slot);
11508 
11509 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11510 		kvfree(slot->arch.lpage_info[i - 1]);
11511 		slot->arch.lpage_info[i - 1] = NULL;
11512 	}
11513 	return -ENOMEM;
11514 }
11515 
11516 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11517 {
11518 	struct kvm_vcpu *vcpu;
11519 	int i;
11520 
11521 	/*
11522 	 * memslots->generation has been incremented.
11523 	 * mmio generation may have reached its maximum value.
11524 	 */
11525 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11526 
11527 	/* Force re-initialization of steal_time cache */
11528 	kvm_for_each_vcpu(i, vcpu, kvm)
11529 		kvm_vcpu_kick(vcpu);
11530 }
11531 
11532 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11533 				struct kvm_memory_slot *memslot,
11534 				const struct kvm_userspace_memory_region *mem,
11535 				enum kvm_mr_change change)
11536 {
11537 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11538 		return kvm_alloc_memslot_metadata(kvm, memslot,
11539 						  mem->memory_size >> PAGE_SHIFT);
11540 	return 0;
11541 }
11542 
11543 
11544 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11545 {
11546 	struct kvm_arch *ka = &kvm->arch;
11547 
11548 	if (!kvm_x86_ops.cpu_dirty_log_size)
11549 		return;
11550 
11551 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11552 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11553 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11554 
11555 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11556 }
11557 
11558 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11559 				     struct kvm_memory_slot *old,
11560 				     const struct kvm_memory_slot *new,
11561 				     enum kvm_mr_change change)
11562 {
11563 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11564 
11565 	/*
11566 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11567 	 * applies to all operations.
11568 	 */
11569 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11570 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11571 
11572 	/*
11573 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11574 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11575 	 *
11576 	 * For a memslot with dirty logging disabled:
11577 	 * CREATE:      No dirty mappings will already exist.
11578 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11579 	 *		kvm_arch_flush_shadow_memslot()
11580 	 *
11581 	 * For a memslot with dirty logging enabled:
11582 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11583 	 *		and no dirty bits to clear.
11584 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11585 	 *		kvm_arch_flush_shadow_memslot().
11586 	 */
11587 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11588 		return;
11589 
11590 	/*
11591 	 * READONLY and non-flags changes were filtered out above, and the only
11592 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11593 	 * logging isn't being toggled on or off.
11594 	 */
11595 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11596 		return;
11597 
11598 	if (!log_dirty_pages) {
11599 		/*
11600 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11601 		 * large sptes have to be split.  If live migration succeeds,
11602 		 * the guest in the source machine will be destroyed and large
11603 		 * sptes will be created in the destination.  However, if the
11604 		 * guest continues to run in the source machine (for example if
11605 		 * live migration fails), small sptes will remain around and
11606 		 * cause bad performance.
11607 		 *
11608 		 * Scan sptes if dirty logging has been stopped, dropping those
11609 		 * which can be collapsed into a single large-page spte.  Later
11610 		 * page faults will create the large-page sptes.
11611 		 */
11612 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11613 	} else {
11614 		/*
11615 		 * Initially-all-set does not require write protecting any page,
11616 		 * because they're all assumed to be dirty.
11617 		 */
11618 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11619 			return;
11620 
11621 		if (kvm_x86_ops.cpu_dirty_log_size) {
11622 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11623 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11624 		} else {
11625 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11626 		}
11627 	}
11628 }
11629 
11630 void kvm_arch_commit_memory_region(struct kvm *kvm,
11631 				const struct kvm_userspace_memory_region *mem,
11632 				struct kvm_memory_slot *old,
11633 				const struct kvm_memory_slot *new,
11634 				enum kvm_mr_change change)
11635 {
11636 	if (!kvm->arch.n_requested_mmu_pages)
11637 		kvm_mmu_change_mmu_pages(kvm,
11638 				kvm_mmu_calculate_default_mmu_pages(kvm));
11639 
11640 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
11641 
11642 	/* Free the arrays associated with the old memslot. */
11643 	if (change == KVM_MR_MOVE)
11644 		kvm_arch_free_memslot(kvm, old);
11645 }
11646 
11647 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11648 {
11649 	kvm_mmu_zap_all(kvm);
11650 }
11651 
11652 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11653 				   struct kvm_memory_slot *slot)
11654 {
11655 	kvm_page_track_flush_slot(kvm, slot);
11656 }
11657 
11658 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11659 {
11660 	return (is_guest_mode(vcpu) &&
11661 			kvm_x86_ops.guest_apic_has_interrupt &&
11662 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11663 }
11664 
11665 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11666 {
11667 	if (!list_empty_careful(&vcpu->async_pf.done))
11668 		return true;
11669 
11670 	if (kvm_apic_has_events(vcpu))
11671 		return true;
11672 
11673 	if (vcpu->arch.pv.pv_unhalted)
11674 		return true;
11675 
11676 	if (vcpu->arch.exception.pending)
11677 		return true;
11678 
11679 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11680 	    (vcpu->arch.nmi_pending &&
11681 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11682 		return true;
11683 
11684 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11685 	    (vcpu->arch.smi_pending &&
11686 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11687 		return true;
11688 
11689 	if (kvm_arch_interrupt_allowed(vcpu) &&
11690 	    (kvm_cpu_has_interrupt(vcpu) ||
11691 	    kvm_guest_apic_has_interrupt(vcpu)))
11692 		return true;
11693 
11694 	if (kvm_hv_has_stimer_pending(vcpu))
11695 		return true;
11696 
11697 	if (is_guest_mode(vcpu) &&
11698 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11699 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11700 		return true;
11701 
11702 	return false;
11703 }
11704 
11705 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11706 {
11707 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11708 }
11709 
11710 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11711 {
11712 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11713 		return true;
11714 
11715 	return false;
11716 }
11717 
11718 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11719 {
11720 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11721 		return true;
11722 
11723 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11724 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11725 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11726 		return true;
11727 
11728 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11729 }
11730 
11731 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11732 {
11733 	if (vcpu->arch.guest_state_protected)
11734 		return true;
11735 
11736 	return vcpu->arch.preempted_in_kernel;
11737 }
11738 
11739 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11740 {
11741 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11742 }
11743 
11744 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11745 {
11746 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11747 }
11748 
11749 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11750 {
11751 	/* Can't read the RIP when guest state is protected, just return 0 */
11752 	if (vcpu->arch.guest_state_protected)
11753 		return 0;
11754 
11755 	if (is_64_bit_mode(vcpu))
11756 		return kvm_rip_read(vcpu);
11757 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11758 		     kvm_rip_read(vcpu));
11759 }
11760 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11761 
11762 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11763 {
11764 	return kvm_get_linear_rip(vcpu) == linear_rip;
11765 }
11766 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11767 
11768 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11769 {
11770 	unsigned long rflags;
11771 
11772 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11773 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11774 		rflags &= ~X86_EFLAGS_TF;
11775 	return rflags;
11776 }
11777 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11778 
11779 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11780 {
11781 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11782 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11783 		rflags |= X86_EFLAGS_TF;
11784 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11785 }
11786 
11787 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11788 {
11789 	__kvm_set_rflags(vcpu, rflags);
11790 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11791 }
11792 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11793 
11794 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11795 {
11796 	int r;
11797 
11798 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11799 	      work->wakeup_all)
11800 		return;
11801 
11802 	r = kvm_mmu_reload(vcpu);
11803 	if (unlikely(r))
11804 		return;
11805 
11806 	if (!vcpu->arch.mmu->direct_map &&
11807 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11808 		return;
11809 
11810 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11811 }
11812 
11813 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11814 {
11815 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11816 
11817 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11818 }
11819 
11820 static inline u32 kvm_async_pf_next_probe(u32 key)
11821 {
11822 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11823 }
11824 
11825 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11826 {
11827 	u32 key = kvm_async_pf_hash_fn(gfn);
11828 
11829 	while (vcpu->arch.apf.gfns[key] != ~0)
11830 		key = kvm_async_pf_next_probe(key);
11831 
11832 	vcpu->arch.apf.gfns[key] = gfn;
11833 }
11834 
11835 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11836 {
11837 	int i;
11838 	u32 key = kvm_async_pf_hash_fn(gfn);
11839 
11840 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11841 		     (vcpu->arch.apf.gfns[key] != gfn &&
11842 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11843 		key = kvm_async_pf_next_probe(key);
11844 
11845 	return key;
11846 }
11847 
11848 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11849 {
11850 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11851 }
11852 
11853 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11854 {
11855 	u32 i, j, k;
11856 
11857 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11858 
11859 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11860 		return;
11861 
11862 	while (true) {
11863 		vcpu->arch.apf.gfns[i] = ~0;
11864 		do {
11865 			j = kvm_async_pf_next_probe(j);
11866 			if (vcpu->arch.apf.gfns[j] == ~0)
11867 				return;
11868 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11869 			/*
11870 			 * k lies cyclically in ]i,j]
11871 			 * |    i.k.j |
11872 			 * |....j i.k.| or  |.k..j i...|
11873 			 */
11874 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11875 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11876 		i = j;
11877 	}
11878 }
11879 
11880 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11881 {
11882 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11883 
11884 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11885 				      sizeof(reason));
11886 }
11887 
11888 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11889 {
11890 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11891 
11892 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11893 					     &token, offset, sizeof(token));
11894 }
11895 
11896 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11897 {
11898 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11899 	u32 val;
11900 
11901 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11902 					 &val, offset, sizeof(val)))
11903 		return false;
11904 
11905 	return !val;
11906 }
11907 
11908 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11909 {
11910 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11911 		return false;
11912 
11913 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11914 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11915 		return false;
11916 
11917 	return true;
11918 }
11919 
11920 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11921 {
11922 	if (unlikely(!lapic_in_kernel(vcpu) ||
11923 		     kvm_event_needs_reinjection(vcpu) ||
11924 		     vcpu->arch.exception.pending))
11925 		return false;
11926 
11927 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11928 		return false;
11929 
11930 	/*
11931 	 * If interrupts are off we cannot even use an artificial
11932 	 * halt state.
11933 	 */
11934 	return kvm_arch_interrupt_allowed(vcpu);
11935 }
11936 
11937 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11938 				     struct kvm_async_pf *work)
11939 {
11940 	struct x86_exception fault;
11941 
11942 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11943 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11944 
11945 	if (kvm_can_deliver_async_pf(vcpu) &&
11946 	    !apf_put_user_notpresent(vcpu)) {
11947 		fault.vector = PF_VECTOR;
11948 		fault.error_code_valid = true;
11949 		fault.error_code = 0;
11950 		fault.nested_page_fault = false;
11951 		fault.address = work->arch.token;
11952 		fault.async_page_fault = true;
11953 		kvm_inject_page_fault(vcpu, &fault);
11954 		return true;
11955 	} else {
11956 		/*
11957 		 * It is not possible to deliver a paravirtualized asynchronous
11958 		 * page fault, but putting the guest in an artificial halt state
11959 		 * can be beneficial nevertheless: if an interrupt arrives, we
11960 		 * can deliver it timely and perhaps the guest will schedule
11961 		 * another process.  When the instruction that triggered a page
11962 		 * fault is retried, hopefully the page will be ready in the host.
11963 		 */
11964 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11965 		return false;
11966 	}
11967 }
11968 
11969 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11970 				 struct kvm_async_pf *work)
11971 {
11972 	struct kvm_lapic_irq irq = {
11973 		.delivery_mode = APIC_DM_FIXED,
11974 		.vector = vcpu->arch.apf.vec
11975 	};
11976 
11977 	if (work->wakeup_all)
11978 		work->arch.token = ~0; /* broadcast wakeup */
11979 	else
11980 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11981 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11982 
11983 	if ((work->wakeup_all || work->notpresent_injected) &&
11984 	    kvm_pv_async_pf_enabled(vcpu) &&
11985 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11986 		vcpu->arch.apf.pageready_pending = true;
11987 		kvm_apic_set_irq(vcpu, &irq, NULL);
11988 	}
11989 
11990 	vcpu->arch.apf.halted = false;
11991 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11992 }
11993 
11994 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11995 {
11996 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11997 	if (!vcpu->arch.apf.pageready_pending)
11998 		kvm_vcpu_kick(vcpu);
11999 }
12000 
12001 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12002 {
12003 	if (!kvm_pv_async_pf_enabled(vcpu))
12004 		return true;
12005 	else
12006 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12007 }
12008 
12009 void kvm_arch_start_assignment(struct kvm *kvm)
12010 {
12011 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12012 		static_call_cond(kvm_x86_start_assignment)(kvm);
12013 }
12014 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12015 
12016 void kvm_arch_end_assignment(struct kvm *kvm)
12017 {
12018 	atomic_dec(&kvm->arch.assigned_device_count);
12019 }
12020 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12021 
12022 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12023 {
12024 	return atomic_read(&kvm->arch.assigned_device_count);
12025 }
12026 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12027 
12028 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12029 {
12030 	atomic_inc(&kvm->arch.noncoherent_dma_count);
12031 }
12032 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12033 
12034 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12035 {
12036 	atomic_dec(&kvm->arch.noncoherent_dma_count);
12037 }
12038 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12039 
12040 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12041 {
12042 	return atomic_read(&kvm->arch.noncoherent_dma_count);
12043 }
12044 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12045 
12046 bool kvm_arch_has_irq_bypass(void)
12047 {
12048 	return true;
12049 }
12050 
12051 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12052 				      struct irq_bypass_producer *prod)
12053 {
12054 	struct kvm_kernel_irqfd *irqfd =
12055 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12056 	int ret;
12057 
12058 	irqfd->producer = prod;
12059 	kvm_arch_start_assignment(irqfd->kvm);
12060 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12061 					 prod->irq, irqfd->gsi, 1);
12062 
12063 	if (ret)
12064 		kvm_arch_end_assignment(irqfd->kvm);
12065 
12066 	return ret;
12067 }
12068 
12069 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12070 				      struct irq_bypass_producer *prod)
12071 {
12072 	int ret;
12073 	struct kvm_kernel_irqfd *irqfd =
12074 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12075 
12076 	WARN_ON(irqfd->producer != prod);
12077 	irqfd->producer = NULL;
12078 
12079 	/*
12080 	 * When producer of consumer is unregistered, we change back to
12081 	 * remapped mode, so we can re-use the current implementation
12082 	 * when the irq is masked/disabled or the consumer side (KVM
12083 	 * int this case doesn't want to receive the interrupts.
12084 	*/
12085 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12086 	if (ret)
12087 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12088 		       " fails: %d\n", irqfd->consumer.token, ret);
12089 
12090 	kvm_arch_end_assignment(irqfd->kvm);
12091 }
12092 
12093 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12094 				   uint32_t guest_irq, bool set)
12095 {
12096 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12097 }
12098 
12099 bool kvm_vector_hashing_enabled(void)
12100 {
12101 	return vector_hashing;
12102 }
12103 
12104 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12105 {
12106 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12107 }
12108 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12109 
12110 
12111 int kvm_spec_ctrl_test_value(u64 value)
12112 {
12113 	/*
12114 	 * test that setting IA32_SPEC_CTRL to given value
12115 	 * is allowed by the host processor
12116 	 */
12117 
12118 	u64 saved_value;
12119 	unsigned long flags;
12120 	int ret = 0;
12121 
12122 	local_irq_save(flags);
12123 
12124 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12125 		ret = 1;
12126 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12127 		ret = 1;
12128 	else
12129 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12130 
12131 	local_irq_restore(flags);
12132 
12133 	return ret;
12134 }
12135 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12136 
12137 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12138 {
12139 	struct x86_exception fault;
12140 	u32 access = error_code &
12141 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12142 
12143 	if (!(error_code & PFERR_PRESENT_MASK) ||
12144 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12145 		/*
12146 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12147 		 * tables probably do not match the TLB.  Just proceed
12148 		 * with the error code that the processor gave.
12149 		 */
12150 		fault.vector = PF_VECTOR;
12151 		fault.error_code_valid = true;
12152 		fault.error_code = error_code;
12153 		fault.nested_page_fault = false;
12154 		fault.address = gva;
12155 	}
12156 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12157 }
12158 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12159 
12160 /*
12161  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12162  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12163  * indicates whether exit to userspace is needed.
12164  */
12165 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12166 			      struct x86_exception *e)
12167 {
12168 	if (r == X86EMUL_PROPAGATE_FAULT) {
12169 		kvm_inject_emulated_page_fault(vcpu, e);
12170 		return 1;
12171 	}
12172 
12173 	/*
12174 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12175 	 * while handling a VMX instruction KVM could've handled the request
12176 	 * correctly by exiting to userspace and performing I/O but there
12177 	 * doesn't seem to be a real use-case behind such requests, just return
12178 	 * KVM_EXIT_INTERNAL_ERROR for now.
12179 	 */
12180 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12181 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12182 	vcpu->run->internal.ndata = 0;
12183 
12184 	return 0;
12185 }
12186 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12187 
12188 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12189 {
12190 	bool pcid_enabled;
12191 	struct x86_exception e;
12192 	struct {
12193 		u64 pcid;
12194 		u64 gla;
12195 	} operand;
12196 	int r;
12197 
12198 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12199 	if (r != X86EMUL_CONTINUE)
12200 		return kvm_handle_memory_failure(vcpu, r, &e);
12201 
12202 	if (operand.pcid >> 12 != 0) {
12203 		kvm_inject_gp(vcpu, 0);
12204 		return 1;
12205 	}
12206 
12207 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12208 
12209 	switch (type) {
12210 	case INVPCID_TYPE_INDIV_ADDR:
12211 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12212 		    is_noncanonical_address(operand.gla, vcpu)) {
12213 			kvm_inject_gp(vcpu, 0);
12214 			return 1;
12215 		}
12216 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12217 		return kvm_skip_emulated_instruction(vcpu);
12218 
12219 	case INVPCID_TYPE_SINGLE_CTXT:
12220 		if (!pcid_enabled && (operand.pcid != 0)) {
12221 			kvm_inject_gp(vcpu, 0);
12222 			return 1;
12223 		}
12224 
12225 		kvm_invalidate_pcid(vcpu, operand.pcid);
12226 		return kvm_skip_emulated_instruction(vcpu);
12227 
12228 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12229 		/*
12230 		 * Currently, KVM doesn't mark global entries in the shadow
12231 		 * page tables, so a non-global flush just degenerates to a
12232 		 * global flush. If needed, we could optimize this later by
12233 		 * keeping track of global entries in shadow page tables.
12234 		 */
12235 
12236 		fallthrough;
12237 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12238 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12239 		return kvm_skip_emulated_instruction(vcpu);
12240 
12241 	default:
12242 		BUG(); /* We have already checked above that type <= 3 */
12243 	}
12244 }
12245 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12246 
12247 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12248 {
12249 	struct kvm_run *run = vcpu->run;
12250 	struct kvm_mmio_fragment *frag;
12251 	unsigned int len;
12252 
12253 	BUG_ON(!vcpu->mmio_needed);
12254 
12255 	/* Complete previous fragment */
12256 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12257 	len = min(8u, frag->len);
12258 	if (!vcpu->mmio_is_write)
12259 		memcpy(frag->data, run->mmio.data, len);
12260 
12261 	if (frag->len <= 8) {
12262 		/* Switch to the next fragment. */
12263 		frag++;
12264 		vcpu->mmio_cur_fragment++;
12265 	} else {
12266 		/* Go forward to the next mmio piece. */
12267 		frag->data += len;
12268 		frag->gpa += len;
12269 		frag->len -= len;
12270 	}
12271 
12272 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12273 		vcpu->mmio_needed = 0;
12274 
12275 		// VMG change, at this point, we're always done
12276 		// RIP has already been advanced
12277 		return 1;
12278 	}
12279 
12280 	// More MMIO is needed
12281 	run->mmio.phys_addr = frag->gpa;
12282 	run->mmio.len = min(8u, frag->len);
12283 	run->mmio.is_write = vcpu->mmio_is_write;
12284 	if (run->mmio.is_write)
12285 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12286 	run->exit_reason = KVM_EXIT_MMIO;
12287 
12288 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12289 
12290 	return 0;
12291 }
12292 
12293 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12294 			  void *data)
12295 {
12296 	int handled;
12297 	struct kvm_mmio_fragment *frag;
12298 
12299 	if (!data)
12300 		return -EINVAL;
12301 
12302 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12303 	if (handled == bytes)
12304 		return 1;
12305 
12306 	bytes -= handled;
12307 	gpa += handled;
12308 	data += handled;
12309 
12310 	/*TODO: Check if need to increment number of frags */
12311 	frag = vcpu->mmio_fragments;
12312 	vcpu->mmio_nr_fragments = 1;
12313 	frag->len = bytes;
12314 	frag->gpa = gpa;
12315 	frag->data = data;
12316 
12317 	vcpu->mmio_needed = 1;
12318 	vcpu->mmio_cur_fragment = 0;
12319 
12320 	vcpu->run->mmio.phys_addr = gpa;
12321 	vcpu->run->mmio.len = min(8u, frag->len);
12322 	vcpu->run->mmio.is_write = 1;
12323 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12324 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12325 
12326 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12327 
12328 	return 0;
12329 }
12330 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12331 
12332 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12333 			 void *data)
12334 {
12335 	int handled;
12336 	struct kvm_mmio_fragment *frag;
12337 
12338 	if (!data)
12339 		return -EINVAL;
12340 
12341 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12342 	if (handled == bytes)
12343 		return 1;
12344 
12345 	bytes -= handled;
12346 	gpa += handled;
12347 	data += handled;
12348 
12349 	/*TODO: Check if need to increment number of frags */
12350 	frag = vcpu->mmio_fragments;
12351 	vcpu->mmio_nr_fragments = 1;
12352 	frag->len = bytes;
12353 	frag->gpa = gpa;
12354 	frag->data = data;
12355 
12356 	vcpu->mmio_needed = 1;
12357 	vcpu->mmio_cur_fragment = 0;
12358 
12359 	vcpu->run->mmio.phys_addr = gpa;
12360 	vcpu->run->mmio.len = min(8u, frag->len);
12361 	vcpu->run->mmio.is_write = 0;
12362 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12363 
12364 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12365 
12366 	return 0;
12367 }
12368 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12369 
12370 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12371 {
12372 	memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12373 	       vcpu->arch.pio.count * vcpu->arch.pio.size);
12374 	vcpu->arch.pio.count = 0;
12375 
12376 	return 1;
12377 }
12378 
12379 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12380 			   unsigned int port, void *data,  unsigned int count)
12381 {
12382 	int ret;
12383 
12384 	ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12385 					data, count);
12386 	if (ret)
12387 		return ret;
12388 
12389 	vcpu->arch.pio.count = 0;
12390 
12391 	return 0;
12392 }
12393 
12394 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12395 			  unsigned int port, void *data, unsigned int count)
12396 {
12397 	int ret;
12398 
12399 	ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12400 				       data, count);
12401 	if (ret) {
12402 		vcpu->arch.pio.count = 0;
12403 	} else {
12404 		vcpu->arch.guest_ins_data = data;
12405 		vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12406 	}
12407 
12408 	return 0;
12409 }
12410 
12411 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12412 			 unsigned int port, void *data,  unsigned int count,
12413 			 int in)
12414 {
12415 	return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12416 		  : kvm_sev_es_outs(vcpu, size, port, data, count);
12417 }
12418 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12419 
12420 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12421 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12422 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12423 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12424 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12425 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12426 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12427 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12428 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12429 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12430 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12432 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12433 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12434 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12435 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12447