xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 827beb77)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
62 
63 #include <trace/events/kvm.h>
64 
65 #include <asm/debugreg.h>
66 #include <asm/msr.h>
67 #include <asm/desc.h>
68 #include <asm/mce.h>
69 #include <asm/pkru.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/api.h>
72 #include <asm/fpu/xcr.h>
73 #include <asm/fpu/xstate.h>
74 #include <asm/pvclock.h>
75 #include <asm/div64.h>
76 #include <asm/irq_remapping.h>
77 #include <asm/mshyperv.h>
78 #include <asm/hypervisor.h>
79 #include <asm/tlbflush.h>
80 #include <asm/intel_pt.h>
81 #include <asm/emulate_prefix.h>
82 #include <asm/sgx.h>
83 #include <clocksource/hyperv_timer.h>
84 
85 #define CREATE_TRACE_POINTS
86 #include "trace.h"
87 
88 #define MAX_IO_MSRS 256
89 #define KVM_MAX_MCE_BANKS 32
90 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
91 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
92 
93 #define emul_to_vcpu(ctxt) \
94 	((struct kvm_vcpu *)(ctxt)->vcpu)
95 
96 /* EFER defaults:
97  * - enable syscall per default because its emulated by KVM
98  * - enable LME and LMA per default on 64 bit KVM
99  */
100 #ifdef CONFIG_X86_64
101 static
102 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
103 #else
104 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 #endif
106 
107 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
108 
109 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
110 
111 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
112                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
113 
114 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
115 static void process_nmi(struct kvm_vcpu *vcpu);
116 static void process_smi(struct kvm_vcpu *vcpu);
117 static void enter_smm(struct kvm_vcpu *vcpu);
118 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
119 static void store_regs(struct kvm_vcpu *vcpu);
120 static int sync_regs(struct kvm_vcpu *vcpu);
121 
122 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
123 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
124 
125 struct kvm_x86_ops kvm_x86_ops __read_mostly;
126 EXPORT_SYMBOL_GPL(kvm_x86_ops);
127 
128 #define KVM_X86_OP(func)					     \
129 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
130 				*(((struct kvm_x86_ops *)0)->func));
131 #define KVM_X86_OP_NULL KVM_X86_OP
132 #include <asm/kvm-x86-ops.h>
133 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
134 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
135 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
136 
137 static bool __read_mostly ignore_msrs = 0;
138 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
139 
140 bool __read_mostly report_ignored_msrs = true;
141 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(report_ignored_msrs);
143 
144 unsigned int min_timer_period_us = 200;
145 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
146 
147 static bool __read_mostly kvmclock_periodic_sync = true;
148 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
149 
150 bool __read_mostly kvm_has_tsc_control;
151 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
152 u32  __read_mostly kvm_max_guest_tsc_khz;
153 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
154 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
155 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
156 u64  __read_mostly kvm_max_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
158 u64 __read_mostly kvm_default_tsc_scaling_ratio;
159 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
160 bool __read_mostly kvm_has_bus_lock_exit;
161 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
162 
163 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
164 static u32 __read_mostly tsc_tolerance_ppm = 250;
165 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
166 
167 /*
168  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
169  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
170  * advancement entirely.  Any other value is used as-is and disables adaptive
171  * tuning, i.e. allows privileged userspace to set an exact advancement time.
172  */
173 static int __read_mostly lapic_timer_advance_ns = -1;
174 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
175 
176 static bool __read_mostly vector_hashing = true;
177 module_param(vector_hashing, bool, S_IRUGO);
178 
179 bool __read_mostly enable_vmware_backdoor = false;
180 module_param(enable_vmware_backdoor, bool, S_IRUGO);
181 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
182 
183 static bool __read_mostly force_emulation_prefix = false;
184 module_param(force_emulation_prefix, bool, S_IRUGO);
185 
186 int __read_mostly pi_inject_timer = -1;
187 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
188 
189 /*
190  * Restoring the host value for MSRs that are only consumed when running in
191  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
192  * returns to userspace, i.e. the kernel can run with the guest's value.
193  */
194 #define KVM_MAX_NR_USER_RETURN_MSRS 16
195 
196 struct kvm_user_return_msrs {
197 	struct user_return_notifier urn;
198 	bool registered;
199 	struct kvm_user_return_msr_values {
200 		u64 host;
201 		u64 curr;
202 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
203 };
204 
205 u32 __read_mostly kvm_nr_uret_msrs;
206 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
207 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
208 static struct kvm_user_return_msrs __percpu *user_return_msrs;
209 
210 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
211 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
212 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
213 				| XFEATURE_MASK_PKRU)
214 
215 u64 __read_mostly host_efer;
216 EXPORT_SYMBOL_GPL(host_efer);
217 
218 bool __read_mostly allow_smaller_maxphyaddr = 0;
219 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
220 
221 bool __read_mostly enable_apicv = true;
222 EXPORT_SYMBOL_GPL(enable_apicv);
223 
224 u64 __read_mostly host_xss;
225 EXPORT_SYMBOL_GPL(host_xss);
226 u64 __read_mostly supported_xss;
227 EXPORT_SYMBOL_GPL(supported_xss);
228 
229 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
230 	KVM_GENERIC_VM_STATS(),
231 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
232 	STATS_DESC_COUNTER(VM, mmu_pte_write),
233 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
234 	STATS_DESC_COUNTER(VM, mmu_flooded),
235 	STATS_DESC_COUNTER(VM, mmu_recycled),
236 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
237 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
238 	STATS_DESC_ICOUNTER(VM, pages_4k),
239 	STATS_DESC_ICOUNTER(VM, pages_2m),
240 	STATS_DESC_ICOUNTER(VM, pages_1g),
241 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
242 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
243 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
244 };
245 
246 const struct kvm_stats_header kvm_vm_stats_header = {
247 	.name_size = KVM_STATS_NAME_SIZE,
248 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
249 	.id_offset = sizeof(struct kvm_stats_header),
250 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
251 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
252 		       sizeof(kvm_vm_stats_desc),
253 };
254 
255 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
256 	KVM_GENERIC_VCPU_STATS(),
257 	STATS_DESC_COUNTER(VCPU, pf_fixed),
258 	STATS_DESC_COUNTER(VCPU, pf_guest),
259 	STATS_DESC_COUNTER(VCPU, tlb_flush),
260 	STATS_DESC_COUNTER(VCPU, invlpg),
261 	STATS_DESC_COUNTER(VCPU, exits),
262 	STATS_DESC_COUNTER(VCPU, io_exits),
263 	STATS_DESC_COUNTER(VCPU, mmio_exits),
264 	STATS_DESC_COUNTER(VCPU, signal_exits),
265 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
266 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
267 	STATS_DESC_COUNTER(VCPU, l1d_flush),
268 	STATS_DESC_COUNTER(VCPU, halt_exits),
269 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
270 	STATS_DESC_COUNTER(VCPU, irq_exits),
271 	STATS_DESC_COUNTER(VCPU, host_state_reload),
272 	STATS_DESC_COUNTER(VCPU, fpu_reload),
273 	STATS_DESC_COUNTER(VCPU, insn_emulation),
274 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
275 	STATS_DESC_COUNTER(VCPU, hypercalls),
276 	STATS_DESC_COUNTER(VCPU, irq_injections),
277 	STATS_DESC_COUNTER(VCPU, nmi_injections),
278 	STATS_DESC_COUNTER(VCPU, req_event),
279 	STATS_DESC_COUNTER(VCPU, nested_run),
280 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
281 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
282 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
283 };
284 
285 const struct kvm_stats_header kvm_vcpu_stats_header = {
286 	.name_size = KVM_STATS_NAME_SIZE,
287 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
288 	.id_offset = sizeof(struct kvm_stats_header),
289 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
290 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
291 		       sizeof(kvm_vcpu_stats_desc),
292 };
293 
294 u64 __read_mostly host_xcr0;
295 u64 __read_mostly supported_xcr0;
296 EXPORT_SYMBOL_GPL(supported_xcr0);
297 
298 static struct kmem_cache *x86_emulator_cache;
299 
300 /*
301  * When called, it means the previous get/set msr reached an invalid msr.
302  * Return true if we want to ignore/silent this failed msr access.
303  */
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
305 {
306 	const char *op = write ? "wrmsr" : "rdmsr";
307 
308 	if (ignore_msrs) {
309 		if (report_ignored_msrs)
310 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
311 				      op, msr, data);
312 		/* Mask the error */
313 		return true;
314 	} else {
315 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
316 				      op, msr, data);
317 		return false;
318 	}
319 }
320 
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
322 {
323 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 	unsigned int size = sizeof(struct x86_emulate_ctxt);
325 
326 	return kmem_cache_create_usercopy("x86_emulator", size,
327 					  __alignof__(struct x86_emulate_ctxt),
328 					  SLAB_ACCOUNT, useroffset,
329 					  size - useroffset, NULL);
330 }
331 
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
333 
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
335 {
336 	int i;
337 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 		vcpu->arch.apf.gfns[i] = ~0;
339 }
340 
341 static void kvm_on_user_return(struct user_return_notifier *urn)
342 {
343 	unsigned slot;
344 	struct kvm_user_return_msrs *msrs
345 		= container_of(urn, struct kvm_user_return_msrs, urn);
346 	struct kvm_user_return_msr_values *values;
347 	unsigned long flags;
348 
349 	/*
350 	 * Disabling irqs at this point since the following code could be
351 	 * interrupted and executed through kvm_arch_hardware_disable()
352 	 */
353 	local_irq_save(flags);
354 	if (msrs->registered) {
355 		msrs->registered = false;
356 		user_return_notifier_unregister(urn);
357 	}
358 	local_irq_restore(flags);
359 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 		values = &msrs->values[slot];
361 		if (values->host != values->curr) {
362 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 			values->curr = values->host;
364 		}
365 	}
366 }
367 
368 static int kvm_probe_user_return_msr(u32 msr)
369 {
370 	u64 val;
371 	int ret;
372 
373 	preempt_disable();
374 	ret = rdmsrl_safe(msr, &val);
375 	if (ret)
376 		goto out;
377 	ret = wrmsrl_safe(msr, val);
378 out:
379 	preempt_enable();
380 	return ret;
381 }
382 
383 int kvm_add_user_return_msr(u32 msr)
384 {
385 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
386 
387 	if (kvm_probe_user_return_msr(msr))
388 		return -1;
389 
390 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 	return kvm_nr_uret_msrs++;
392 }
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
394 
395 int kvm_find_user_return_msr(u32 msr)
396 {
397 	int i;
398 
399 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 		if (kvm_uret_msrs_list[i] == msr)
401 			return i;
402 	}
403 	return -1;
404 }
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
406 
407 static void kvm_user_return_msr_cpu_online(void)
408 {
409 	unsigned int cpu = smp_processor_id();
410 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
411 	u64 value;
412 	int i;
413 
414 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 		msrs->values[i].host = value;
417 		msrs->values[i].curr = value;
418 	}
419 }
420 
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
422 {
423 	unsigned int cpu = smp_processor_id();
424 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 	int err;
426 
427 	value = (value & mask) | (msrs->values[slot].host & ~mask);
428 	if (value == msrs->values[slot].curr)
429 		return 0;
430 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
431 	if (err)
432 		return 1;
433 
434 	msrs->values[slot].curr = value;
435 	if (!msrs->registered) {
436 		msrs->urn.on_user_return = kvm_on_user_return;
437 		user_return_notifier_register(&msrs->urn);
438 		msrs->registered = true;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
443 
444 static void drop_user_return_notifiers(void)
445 {
446 	unsigned int cpu = smp_processor_id();
447 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
448 
449 	if (msrs->registered)
450 		kvm_on_user_return(&msrs->urn);
451 }
452 
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
454 {
455 	return vcpu->arch.apic_base;
456 }
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
458 
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
460 {
461 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
462 }
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
464 
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
466 {
467 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
471 
472 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
473 		return 1;
474 	if (!msr_info->host_initiated) {
475 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
476 			return 1;
477 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
478 			return 1;
479 	}
480 
481 	kvm_lapic_set_base(vcpu, msr_info->data);
482 	kvm_recalculate_apic_map(vcpu->kvm);
483 	return 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
486 
487 /*
488  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
489  *
490  * Hardware virtualization extension instructions may fault if a reboot turns
491  * off virtualization while processes are running.  Usually after catching the
492  * fault we just panic; during reboot instead the instruction is ignored.
493  */
494 noinstr void kvm_spurious_fault(void)
495 {
496 	/* Fault while not rebooting.  We want the trace. */
497 	BUG_ON(!kvm_rebooting);
498 }
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
500 
501 #define EXCPT_BENIGN		0
502 #define EXCPT_CONTRIBUTORY	1
503 #define EXCPT_PF		2
504 
505 static int exception_class(int vector)
506 {
507 	switch (vector) {
508 	case PF_VECTOR:
509 		return EXCPT_PF;
510 	case DE_VECTOR:
511 	case TS_VECTOR:
512 	case NP_VECTOR:
513 	case SS_VECTOR:
514 	case GP_VECTOR:
515 		return EXCPT_CONTRIBUTORY;
516 	default:
517 		break;
518 	}
519 	return EXCPT_BENIGN;
520 }
521 
522 #define EXCPT_FAULT		0
523 #define EXCPT_TRAP		1
524 #define EXCPT_ABORT		2
525 #define EXCPT_INTERRUPT		3
526 
527 static int exception_type(int vector)
528 {
529 	unsigned int mask;
530 
531 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 		return EXCPT_INTERRUPT;
533 
534 	mask = 1 << vector;
535 
536 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
537 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
538 		return EXCPT_TRAP;
539 
540 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
541 		return EXCPT_ABORT;
542 
543 	/* Reserved exceptions will result in fault */
544 	return EXCPT_FAULT;
545 }
546 
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
548 {
549 	unsigned nr = vcpu->arch.exception.nr;
550 	bool has_payload = vcpu->arch.exception.has_payload;
551 	unsigned long payload = vcpu->arch.exception.payload;
552 
553 	if (!has_payload)
554 		return;
555 
556 	switch (nr) {
557 	case DB_VECTOR:
558 		/*
559 		 * "Certain debug exceptions may clear bit 0-3.  The
560 		 * remaining contents of the DR6 register are never
561 		 * cleared by the processor".
562 		 */
563 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
564 		/*
565 		 * In order to reflect the #DB exception payload in guest
566 		 * dr6, three components need to be considered: active low
567 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
568 		 * DR6_BS and DR6_BT)
569 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 		 * In the target guest dr6:
571 		 * FIXED_1 bits should always be set.
572 		 * Active low bits should be cleared if 1-setting in payload.
573 		 * Active high bits should be set if 1-setting in payload.
574 		 *
575 		 * Note, the payload is compatible with the pending debug
576 		 * exceptions/exit qualification under VMX, that active_low bits
577 		 * are active high in payload.
578 		 * So they need to be flipped for DR6.
579 		 */
580 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 		vcpu->arch.dr6 |= payload;
582 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
583 
584 		/*
585 		 * The #DB payload is defined as compatible with the 'pending
586 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 		 * defined in the 'pending debug exceptions' field (enabled
588 		 * breakpoint), it is reserved and must be zero in DR6.
589 		 */
590 		vcpu->arch.dr6 &= ~BIT(12);
591 		break;
592 	case PF_VECTOR:
593 		vcpu->arch.cr2 = payload;
594 		break;
595 	}
596 
597 	vcpu->arch.exception.has_payload = false;
598 	vcpu->arch.exception.payload = 0;
599 }
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
601 
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 		unsigned nr, bool has_error, u32 error_code,
604 	        bool has_payload, unsigned long payload, bool reinject)
605 {
606 	u32 prev_nr;
607 	int class1, class2;
608 
609 	kvm_make_request(KVM_REQ_EVENT, vcpu);
610 
611 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
612 	queue:
613 		if (reinject) {
614 			/*
615 			 * On vmentry, vcpu->arch.exception.pending is only
616 			 * true if an event injection was blocked by
617 			 * nested_run_pending.  In that case, however,
618 			 * vcpu_enter_guest requests an immediate exit,
619 			 * and the guest shouldn't proceed far enough to
620 			 * need reinjection.
621 			 */
622 			WARN_ON_ONCE(vcpu->arch.exception.pending);
623 			vcpu->arch.exception.injected = true;
624 			if (WARN_ON_ONCE(has_payload)) {
625 				/*
626 				 * A reinjected event has already
627 				 * delivered its payload.
628 				 */
629 				has_payload = false;
630 				payload = 0;
631 			}
632 		} else {
633 			vcpu->arch.exception.pending = true;
634 			vcpu->arch.exception.injected = false;
635 		}
636 		vcpu->arch.exception.has_error_code = has_error;
637 		vcpu->arch.exception.nr = nr;
638 		vcpu->arch.exception.error_code = error_code;
639 		vcpu->arch.exception.has_payload = has_payload;
640 		vcpu->arch.exception.payload = payload;
641 		if (!is_guest_mode(vcpu))
642 			kvm_deliver_exception_payload(vcpu);
643 		return;
644 	}
645 
646 	/* to check exception */
647 	prev_nr = vcpu->arch.exception.nr;
648 	if (prev_nr == DF_VECTOR) {
649 		/* triple fault -> shutdown */
650 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
651 		return;
652 	}
653 	class1 = exception_class(prev_nr);
654 	class2 = exception_class(nr);
655 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
657 		/*
658 		 * Generate double fault per SDM Table 5-5.  Set
659 		 * exception.pending = true so that the double fault
660 		 * can trigger a nested vmexit.
661 		 */
662 		vcpu->arch.exception.pending = true;
663 		vcpu->arch.exception.injected = false;
664 		vcpu->arch.exception.has_error_code = true;
665 		vcpu->arch.exception.nr = DF_VECTOR;
666 		vcpu->arch.exception.error_code = 0;
667 		vcpu->arch.exception.has_payload = false;
668 		vcpu->arch.exception.payload = 0;
669 	} else
670 		/* replace previous exception with a new one in a hope
671 		   that instruction re-execution will regenerate lost
672 		   exception */
673 		goto queue;
674 }
675 
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
677 {
678 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
679 }
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
681 
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
683 {
684 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
685 }
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
687 
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 			   unsigned long payload)
690 {
691 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
692 }
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
694 
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 				    u32 error_code, unsigned long payload)
697 {
698 	kvm_multiple_exception(vcpu, nr, true, error_code,
699 			       true, payload, false);
700 }
701 
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
703 {
704 	if (err)
705 		kvm_inject_gp(vcpu, 0);
706 	else
707 		return kvm_skip_emulated_instruction(vcpu);
708 
709 	return 1;
710 }
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
712 
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
714 {
715 	++vcpu->stat.pf_guest;
716 	vcpu->arch.exception.nested_apf =
717 		is_guest_mode(vcpu) && fault->async_page_fault;
718 	if (vcpu->arch.exception.nested_apf) {
719 		vcpu->arch.apf.nested_apf_token = fault->address;
720 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
721 	} else {
722 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
723 					fault->address);
724 	}
725 }
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
727 
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 				    struct x86_exception *fault)
730 {
731 	struct kvm_mmu *fault_mmu;
732 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
733 
734 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
735 					       vcpu->arch.walk_mmu;
736 
737 	/*
738 	 * Invalidate the TLB entry for the faulting address, if it exists,
739 	 * else the access will fault indefinitely (and to emulate hardware).
740 	 */
741 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 	    !(fault->error_code & PFERR_RSVD_MASK))
743 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 				       fault_mmu->root_hpa);
745 
746 	fault_mmu->inject_page_fault(vcpu, fault);
747 	return fault->nested_page_fault;
748 }
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
750 
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
752 {
753 	atomic_inc(&vcpu->arch.nmi_queued);
754 	kvm_make_request(KVM_REQ_NMI, vcpu);
755 }
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
757 
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
759 {
760 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
761 }
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
763 
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
765 {
766 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
767 }
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
769 
770 /*
771  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
772  * a #GP and return false.
773  */
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
775 {
776 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
777 		return true;
778 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
779 	return false;
780 }
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
782 
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
784 {
785 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 		return true;
787 
788 	kvm_queue_exception(vcpu, UD_VECTOR);
789 	return false;
790 }
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
792 
793 /*
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 	raw_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 	raw_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 	raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2784 	pvclock_update_vm_gtod_copy(kvm);
2785 	raw_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 	raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2804 	if (!ka->use_master_clock) {
2805 		raw_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 	raw_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 	raw_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 	raw_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 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4704 					 struct kvm_xsave *guest_xsave)
4705 {
4706 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4707 		return;
4708 
4709 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4710 				       guest_xsave->region,
4711 				       sizeof(guest_xsave->region),
4712 				       vcpu->arch.pkru);
4713 }
4714 
4715 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4716 					struct kvm_xsave *guest_xsave)
4717 {
4718 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4719 		return 0;
4720 
4721 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
4722 					      guest_xsave->region,
4723 					      supported_xcr0, &vcpu->arch.pkru);
4724 }
4725 
4726 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4727 					struct kvm_xcrs *guest_xcrs)
4728 {
4729 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4730 		guest_xcrs->nr_xcrs = 0;
4731 		return;
4732 	}
4733 
4734 	guest_xcrs->nr_xcrs = 1;
4735 	guest_xcrs->flags = 0;
4736 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4737 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4738 }
4739 
4740 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4741 				       struct kvm_xcrs *guest_xcrs)
4742 {
4743 	int i, r = 0;
4744 
4745 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4746 		return -EINVAL;
4747 
4748 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4749 		return -EINVAL;
4750 
4751 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4752 		/* Only support XCR0 currently */
4753 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4754 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4755 				guest_xcrs->xcrs[i].value);
4756 			break;
4757 		}
4758 	if (r)
4759 		r = -EINVAL;
4760 	return r;
4761 }
4762 
4763 /*
4764  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4765  * stopped by the hypervisor.  This function will be called from the host only.
4766  * EINVAL is returned when the host attempts to set the flag for a guest that
4767  * does not support pv clocks.
4768  */
4769 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4770 {
4771 	if (!vcpu->arch.pv_time_enabled)
4772 		return -EINVAL;
4773 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4774 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4775 	return 0;
4776 }
4777 
4778 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4779 				     struct kvm_enable_cap *cap)
4780 {
4781 	int r;
4782 	uint16_t vmcs_version;
4783 	void __user *user_ptr;
4784 
4785 	if (cap->flags)
4786 		return -EINVAL;
4787 
4788 	switch (cap->cap) {
4789 	case KVM_CAP_HYPERV_SYNIC2:
4790 		if (cap->args[0])
4791 			return -EINVAL;
4792 		fallthrough;
4793 
4794 	case KVM_CAP_HYPERV_SYNIC:
4795 		if (!irqchip_in_kernel(vcpu->kvm))
4796 			return -EINVAL;
4797 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4798 					     KVM_CAP_HYPERV_SYNIC2);
4799 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4800 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4801 			return -ENOTTY;
4802 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4803 		if (!r) {
4804 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4805 			if (copy_to_user(user_ptr, &vmcs_version,
4806 					 sizeof(vmcs_version)))
4807 				r = -EFAULT;
4808 		}
4809 		return r;
4810 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4811 		if (!kvm_x86_ops.enable_direct_tlbflush)
4812 			return -ENOTTY;
4813 
4814 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4815 
4816 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4817 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4818 
4819 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4820 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4821 		if (vcpu->arch.pv_cpuid.enforce)
4822 			kvm_update_pv_runtime(vcpu);
4823 
4824 		return 0;
4825 	default:
4826 		return -EINVAL;
4827 	}
4828 }
4829 
4830 long kvm_arch_vcpu_ioctl(struct file *filp,
4831 			 unsigned int ioctl, unsigned long arg)
4832 {
4833 	struct kvm_vcpu *vcpu = filp->private_data;
4834 	void __user *argp = (void __user *)arg;
4835 	int r;
4836 	union {
4837 		struct kvm_sregs2 *sregs2;
4838 		struct kvm_lapic_state *lapic;
4839 		struct kvm_xsave *xsave;
4840 		struct kvm_xcrs *xcrs;
4841 		void *buffer;
4842 	} u;
4843 
4844 	vcpu_load(vcpu);
4845 
4846 	u.buffer = NULL;
4847 	switch (ioctl) {
4848 	case KVM_GET_LAPIC: {
4849 		r = -EINVAL;
4850 		if (!lapic_in_kernel(vcpu))
4851 			goto out;
4852 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4853 				GFP_KERNEL_ACCOUNT);
4854 
4855 		r = -ENOMEM;
4856 		if (!u.lapic)
4857 			goto out;
4858 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4859 		if (r)
4860 			goto out;
4861 		r = -EFAULT;
4862 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4863 			goto out;
4864 		r = 0;
4865 		break;
4866 	}
4867 	case KVM_SET_LAPIC: {
4868 		r = -EINVAL;
4869 		if (!lapic_in_kernel(vcpu))
4870 			goto out;
4871 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4872 		if (IS_ERR(u.lapic)) {
4873 			r = PTR_ERR(u.lapic);
4874 			goto out_nofree;
4875 		}
4876 
4877 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4878 		break;
4879 	}
4880 	case KVM_INTERRUPT: {
4881 		struct kvm_interrupt irq;
4882 
4883 		r = -EFAULT;
4884 		if (copy_from_user(&irq, argp, sizeof(irq)))
4885 			goto out;
4886 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4887 		break;
4888 	}
4889 	case KVM_NMI: {
4890 		r = kvm_vcpu_ioctl_nmi(vcpu);
4891 		break;
4892 	}
4893 	case KVM_SMI: {
4894 		r = kvm_vcpu_ioctl_smi(vcpu);
4895 		break;
4896 	}
4897 	case KVM_SET_CPUID: {
4898 		struct kvm_cpuid __user *cpuid_arg = argp;
4899 		struct kvm_cpuid cpuid;
4900 
4901 		r = -EFAULT;
4902 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4903 			goto out;
4904 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4905 		break;
4906 	}
4907 	case KVM_SET_CPUID2: {
4908 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4909 		struct kvm_cpuid2 cpuid;
4910 
4911 		r = -EFAULT;
4912 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4913 			goto out;
4914 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4915 					      cpuid_arg->entries);
4916 		break;
4917 	}
4918 	case KVM_GET_CPUID2: {
4919 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4920 		struct kvm_cpuid2 cpuid;
4921 
4922 		r = -EFAULT;
4923 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4924 			goto out;
4925 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4926 					      cpuid_arg->entries);
4927 		if (r)
4928 			goto out;
4929 		r = -EFAULT;
4930 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4931 			goto out;
4932 		r = 0;
4933 		break;
4934 	}
4935 	case KVM_GET_MSRS: {
4936 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4937 		r = msr_io(vcpu, argp, do_get_msr, 1);
4938 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4939 		break;
4940 	}
4941 	case KVM_SET_MSRS: {
4942 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4943 		r = msr_io(vcpu, argp, do_set_msr, 0);
4944 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4945 		break;
4946 	}
4947 	case KVM_TPR_ACCESS_REPORTING: {
4948 		struct kvm_tpr_access_ctl tac;
4949 
4950 		r = -EFAULT;
4951 		if (copy_from_user(&tac, argp, sizeof(tac)))
4952 			goto out;
4953 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4954 		if (r)
4955 			goto out;
4956 		r = -EFAULT;
4957 		if (copy_to_user(argp, &tac, sizeof(tac)))
4958 			goto out;
4959 		r = 0;
4960 		break;
4961 	};
4962 	case KVM_SET_VAPIC_ADDR: {
4963 		struct kvm_vapic_addr va;
4964 		int idx;
4965 
4966 		r = -EINVAL;
4967 		if (!lapic_in_kernel(vcpu))
4968 			goto out;
4969 		r = -EFAULT;
4970 		if (copy_from_user(&va, argp, sizeof(va)))
4971 			goto out;
4972 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4973 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4974 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4975 		break;
4976 	}
4977 	case KVM_X86_SETUP_MCE: {
4978 		u64 mcg_cap;
4979 
4980 		r = -EFAULT;
4981 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4982 			goto out;
4983 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4984 		break;
4985 	}
4986 	case KVM_X86_SET_MCE: {
4987 		struct kvm_x86_mce mce;
4988 
4989 		r = -EFAULT;
4990 		if (copy_from_user(&mce, argp, sizeof(mce)))
4991 			goto out;
4992 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4993 		break;
4994 	}
4995 	case KVM_GET_VCPU_EVENTS: {
4996 		struct kvm_vcpu_events events;
4997 
4998 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4999 
5000 		r = -EFAULT;
5001 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5002 			break;
5003 		r = 0;
5004 		break;
5005 	}
5006 	case KVM_SET_VCPU_EVENTS: {
5007 		struct kvm_vcpu_events events;
5008 
5009 		r = -EFAULT;
5010 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5011 			break;
5012 
5013 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5014 		break;
5015 	}
5016 	case KVM_GET_DEBUGREGS: {
5017 		struct kvm_debugregs dbgregs;
5018 
5019 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5020 
5021 		r = -EFAULT;
5022 		if (copy_to_user(argp, &dbgregs,
5023 				 sizeof(struct kvm_debugregs)))
5024 			break;
5025 		r = 0;
5026 		break;
5027 	}
5028 	case KVM_SET_DEBUGREGS: {
5029 		struct kvm_debugregs dbgregs;
5030 
5031 		r = -EFAULT;
5032 		if (copy_from_user(&dbgregs, argp,
5033 				   sizeof(struct kvm_debugregs)))
5034 			break;
5035 
5036 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5037 		break;
5038 	}
5039 	case KVM_GET_XSAVE: {
5040 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5041 		r = -ENOMEM;
5042 		if (!u.xsave)
5043 			break;
5044 
5045 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5046 
5047 		r = -EFAULT;
5048 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5049 			break;
5050 		r = 0;
5051 		break;
5052 	}
5053 	case KVM_SET_XSAVE: {
5054 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5055 		if (IS_ERR(u.xsave)) {
5056 			r = PTR_ERR(u.xsave);
5057 			goto out_nofree;
5058 		}
5059 
5060 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5061 		break;
5062 	}
5063 	case KVM_GET_XCRS: {
5064 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5065 		r = -ENOMEM;
5066 		if (!u.xcrs)
5067 			break;
5068 
5069 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5070 
5071 		r = -EFAULT;
5072 		if (copy_to_user(argp, u.xcrs,
5073 				 sizeof(struct kvm_xcrs)))
5074 			break;
5075 		r = 0;
5076 		break;
5077 	}
5078 	case KVM_SET_XCRS: {
5079 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5080 		if (IS_ERR(u.xcrs)) {
5081 			r = PTR_ERR(u.xcrs);
5082 			goto out_nofree;
5083 		}
5084 
5085 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5086 		break;
5087 	}
5088 	case KVM_SET_TSC_KHZ: {
5089 		u32 user_tsc_khz;
5090 
5091 		r = -EINVAL;
5092 		user_tsc_khz = (u32)arg;
5093 
5094 		if (kvm_has_tsc_control &&
5095 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5096 			goto out;
5097 
5098 		if (user_tsc_khz == 0)
5099 			user_tsc_khz = tsc_khz;
5100 
5101 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5102 			r = 0;
5103 
5104 		goto out;
5105 	}
5106 	case KVM_GET_TSC_KHZ: {
5107 		r = vcpu->arch.virtual_tsc_khz;
5108 		goto out;
5109 	}
5110 	case KVM_KVMCLOCK_CTRL: {
5111 		r = kvm_set_guest_paused(vcpu);
5112 		goto out;
5113 	}
5114 	case KVM_ENABLE_CAP: {
5115 		struct kvm_enable_cap cap;
5116 
5117 		r = -EFAULT;
5118 		if (copy_from_user(&cap, argp, sizeof(cap)))
5119 			goto out;
5120 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5121 		break;
5122 	}
5123 	case KVM_GET_NESTED_STATE: {
5124 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5125 		u32 user_data_size;
5126 
5127 		r = -EINVAL;
5128 		if (!kvm_x86_ops.nested_ops->get_state)
5129 			break;
5130 
5131 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5132 		r = -EFAULT;
5133 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5134 			break;
5135 
5136 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5137 						     user_data_size);
5138 		if (r < 0)
5139 			break;
5140 
5141 		if (r > user_data_size) {
5142 			if (put_user(r, &user_kvm_nested_state->size))
5143 				r = -EFAULT;
5144 			else
5145 				r = -E2BIG;
5146 			break;
5147 		}
5148 
5149 		r = 0;
5150 		break;
5151 	}
5152 	case KVM_SET_NESTED_STATE: {
5153 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5154 		struct kvm_nested_state kvm_state;
5155 		int idx;
5156 
5157 		r = -EINVAL;
5158 		if (!kvm_x86_ops.nested_ops->set_state)
5159 			break;
5160 
5161 		r = -EFAULT;
5162 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5163 			break;
5164 
5165 		r = -EINVAL;
5166 		if (kvm_state.size < sizeof(kvm_state))
5167 			break;
5168 
5169 		if (kvm_state.flags &
5170 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5171 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5172 		      | KVM_STATE_NESTED_GIF_SET))
5173 			break;
5174 
5175 		/* nested_run_pending implies guest_mode.  */
5176 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5177 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5178 			break;
5179 
5180 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5181 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5182 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5183 		break;
5184 	}
5185 	case KVM_GET_SUPPORTED_HV_CPUID:
5186 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5187 		break;
5188 #ifdef CONFIG_KVM_XEN
5189 	case KVM_XEN_VCPU_GET_ATTR: {
5190 		struct kvm_xen_vcpu_attr xva;
5191 
5192 		r = -EFAULT;
5193 		if (copy_from_user(&xva, argp, sizeof(xva)))
5194 			goto out;
5195 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5196 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5197 			r = -EFAULT;
5198 		break;
5199 	}
5200 	case KVM_XEN_VCPU_SET_ATTR: {
5201 		struct kvm_xen_vcpu_attr xva;
5202 
5203 		r = -EFAULT;
5204 		if (copy_from_user(&xva, argp, sizeof(xva)))
5205 			goto out;
5206 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5207 		break;
5208 	}
5209 #endif
5210 	case KVM_GET_SREGS2: {
5211 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5212 		r = -ENOMEM;
5213 		if (!u.sregs2)
5214 			goto out;
5215 		__get_sregs2(vcpu, u.sregs2);
5216 		r = -EFAULT;
5217 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5218 			goto out;
5219 		r = 0;
5220 		break;
5221 	}
5222 	case KVM_SET_SREGS2: {
5223 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5224 		if (IS_ERR(u.sregs2)) {
5225 			r = PTR_ERR(u.sregs2);
5226 			u.sregs2 = NULL;
5227 			goto out;
5228 		}
5229 		r = __set_sregs2(vcpu, u.sregs2);
5230 		break;
5231 	}
5232 	default:
5233 		r = -EINVAL;
5234 	}
5235 out:
5236 	kfree(u.buffer);
5237 out_nofree:
5238 	vcpu_put(vcpu);
5239 	return r;
5240 }
5241 
5242 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5243 {
5244 	return VM_FAULT_SIGBUS;
5245 }
5246 
5247 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5248 {
5249 	int ret;
5250 
5251 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5252 		return -EINVAL;
5253 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5254 	return ret;
5255 }
5256 
5257 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5258 					      u64 ident_addr)
5259 {
5260 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5261 }
5262 
5263 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5264 					 unsigned long kvm_nr_mmu_pages)
5265 {
5266 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5267 		return -EINVAL;
5268 
5269 	mutex_lock(&kvm->slots_lock);
5270 
5271 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5272 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5273 
5274 	mutex_unlock(&kvm->slots_lock);
5275 	return 0;
5276 }
5277 
5278 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5279 {
5280 	return kvm->arch.n_max_mmu_pages;
5281 }
5282 
5283 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5284 {
5285 	struct kvm_pic *pic = kvm->arch.vpic;
5286 	int r;
5287 
5288 	r = 0;
5289 	switch (chip->chip_id) {
5290 	case KVM_IRQCHIP_PIC_MASTER:
5291 		memcpy(&chip->chip.pic, &pic->pics[0],
5292 			sizeof(struct kvm_pic_state));
5293 		break;
5294 	case KVM_IRQCHIP_PIC_SLAVE:
5295 		memcpy(&chip->chip.pic, &pic->pics[1],
5296 			sizeof(struct kvm_pic_state));
5297 		break;
5298 	case KVM_IRQCHIP_IOAPIC:
5299 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5300 		break;
5301 	default:
5302 		r = -EINVAL;
5303 		break;
5304 	}
5305 	return r;
5306 }
5307 
5308 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5309 {
5310 	struct kvm_pic *pic = kvm->arch.vpic;
5311 	int r;
5312 
5313 	r = 0;
5314 	switch (chip->chip_id) {
5315 	case KVM_IRQCHIP_PIC_MASTER:
5316 		spin_lock(&pic->lock);
5317 		memcpy(&pic->pics[0], &chip->chip.pic,
5318 			sizeof(struct kvm_pic_state));
5319 		spin_unlock(&pic->lock);
5320 		break;
5321 	case KVM_IRQCHIP_PIC_SLAVE:
5322 		spin_lock(&pic->lock);
5323 		memcpy(&pic->pics[1], &chip->chip.pic,
5324 			sizeof(struct kvm_pic_state));
5325 		spin_unlock(&pic->lock);
5326 		break;
5327 	case KVM_IRQCHIP_IOAPIC:
5328 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5329 		break;
5330 	default:
5331 		r = -EINVAL;
5332 		break;
5333 	}
5334 	kvm_pic_update_irq(pic);
5335 	return r;
5336 }
5337 
5338 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5339 {
5340 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5341 
5342 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5343 
5344 	mutex_lock(&kps->lock);
5345 	memcpy(ps, &kps->channels, sizeof(*ps));
5346 	mutex_unlock(&kps->lock);
5347 	return 0;
5348 }
5349 
5350 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5351 {
5352 	int i;
5353 	struct kvm_pit *pit = kvm->arch.vpit;
5354 
5355 	mutex_lock(&pit->pit_state.lock);
5356 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5357 	for (i = 0; i < 3; i++)
5358 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5359 	mutex_unlock(&pit->pit_state.lock);
5360 	return 0;
5361 }
5362 
5363 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5364 {
5365 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5366 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5367 		sizeof(ps->channels));
5368 	ps->flags = kvm->arch.vpit->pit_state.flags;
5369 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5370 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5371 	return 0;
5372 }
5373 
5374 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5375 {
5376 	int start = 0;
5377 	int i;
5378 	u32 prev_legacy, cur_legacy;
5379 	struct kvm_pit *pit = kvm->arch.vpit;
5380 
5381 	mutex_lock(&pit->pit_state.lock);
5382 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5383 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5384 	if (!prev_legacy && cur_legacy)
5385 		start = 1;
5386 	memcpy(&pit->pit_state.channels, &ps->channels,
5387 	       sizeof(pit->pit_state.channels));
5388 	pit->pit_state.flags = ps->flags;
5389 	for (i = 0; i < 3; i++)
5390 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5391 				   start && i == 0);
5392 	mutex_unlock(&pit->pit_state.lock);
5393 	return 0;
5394 }
5395 
5396 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5397 				 struct kvm_reinject_control *control)
5398 {
5399 	struct kvm_pit *pit = kvm->arch.vpit;
5400 
5401 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5402 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5403 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5404 	 */
5405 	mutex_lock(&pit->pit_state.lock);
5406 	kvm_pit_set_reinject(pit, control->pit_reinject);
5407 	mutex_unlock(&pit->pit_state.lock);
5408 
5409 	return 0;
5410 }
5411 
5412 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5413 {
5414 
5415 	/*
5416 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5417 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5418 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5419 	 * VM-Exit.
5420 	 */
5421 	struct kvm_vcpu *vcpu;
5422 	int i;
5423 
5424 	kvm_for_each_vcpu(i, vcpu, kvm)
5425 		kvm_vcpu_kick(vcpu);
5426 }
5427 
5428 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5429 			bool line_status)
5430 {
5431 	if (!irqchip_in_kernel(kvm))
5432 		return -ENXIO;
5433 
5434 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5435 					irq_event->irq, irq_event->level,
5436 					line_status);
5437 	return 0;
5438 }
5439 
5440 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5441 			    struct kvm_enable_cap *cap)
5442 {
5443 	int r;
5444 
5445 	if (cap->flags)
5446 		return -EINVAL;
5447 
5448 	switch (cap->cap) {
5449 	case KVM_CAP_DISABLE_QUIRKS:
5450 		kvm->arch.disabled_quirks = cap->args[0];
5451 		r = 0;
5452 		break;
5453 	case KVM_CAP_SPLIT_IRQCHIP: {
5454 		mutex_lock(&kvm->lock);
5455 		r = -EINVAL;
5456 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5457 			goto split_irqchip_unlock;
5458 		r = -EEXIST;
5459 		if (irqchip_in_kernel(kvm))
5460 			goto split_irqchip_unlock;
5461 		if (kvm->created_vcpus)
5462 			goto split_irqchip_unlock;
5463 		r = kvm_setup_empty_irq_routing(kvm);
5464 		if (r)
5465 			goto split_irqchip_unlock;
5466 		/* Pairs with irqchip_in_kernel. */
5467 		smp_wmb();
5468 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5469 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5470 		r = 0;
5471 split_irqchip_unlock:
5472 		mutex_unlock(&kvm->lock);
5473 		break;
5474 	}
5475 	case KVM_CAP_X2APIC_API:
5476 		r = -EINVAL;
5477 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5478 			break;
5479 
5480 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5481 			kvm->arch.x2apic_format = true;
5482 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5483 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5484 
5485 		r = 0;
5486 		break;
5487 	case KVM_CAP_X86_DISABLE_EXITS:
5488 		r = -EINVAL;
5489 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5490 			break;
5491 
5492 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5493 			kvm_can_mwait_in_guest())
5494 			kvm->arch.mwait_in_guest = true;
5495 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5496 			kvm->arch.hlt_in_guest = true;
5497 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5498 			kvm->arch.pause_in_guest = true;
5499 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5500 			kvm->arch.cstate_in_guest = true;
5501 		r = 0;
5502 		break;
5503 	case KVM_CAP_MSR_PLATFORM_INFO:
5504 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5505 		r = 0;
5506 		break;
5507 	case KVM_CAP_EXCEPTION_PAYLOAD:
5508 		kvm->arch.exception_payload_enabled = cap->args[0];
5509 		r = 0;
5510 		break;
5511 	case KVM_CAP_X86_USER_SPACE_MSR:
5512 		kvm->arch.user_space_msr_mask = cap->args[0];
5513 		r = 0;
5514 		break;
5515 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5516 		r = -EINVAL;
5517 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5518 			break;
5519 
5520 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5521 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5522 			break;
5523 
5524 		if (kvm_has_bus_lock_exit &&
5525 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5526 			kvm->arch.bus_lock_detection_enabled = true;
5527 		r = 0;
5528 		break;
5529 #ifdef CONFIG_X86_SGX_KVM
5530 	case KVM_CAP_SGX_ATTRIBUTE: {
5531 		unsigned long allowed_attributes = 0;
5532 
5533 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5534 		if (r)
5535 			break;
5536 
5537 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5538 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5539 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5540 			kvm->arch.sgx_provisioning_allowed = true;
5541 		else
5542 			r = -EINVAL;
5543 		break;
5544 	}
5545 #endif
5546 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5547 		r = -EINVAL;
5548 		if (kvm_x86_ops.vm_copy_enc_context_from)
5549 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5550 		return r;
5551 	case KVM_CAP_EXIT_HYPERCALL:
5552 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5553 			r = -EINVAL;
5554 			break;
5555 		}
5556 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5557 		r = 0;
5558 		break;
5559 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5560 		r = -EINVAL;
5561 		if (cap->args[0] & ~1)
5562 			break;
5563 		kvm->arch.exit_on_emulation_error = cap->args[0];
5564 		r = 0;
5565 		break;
5566 	default:
5567 		r = -EINVAL;
5568 		break;
5569 	}
5570 	return r;
5571 }
5572 
5573 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5574 {
5575 	struct kvm_x86_msr_filter *msr_filter;
5576 
5577 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5578 	if (!msr_filter)
5579 		return NULL;
5580 
5581 	msr_filter->default_allow = default_allow;
5582 	return msr_filter;
5583 }
5584 
5585 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5586 {
5587 	u32 i;
5588 
5589 	if (!msr_filter)
5590 		return;
5591 
5592 	for (i = 0; i < msr_filter->count; i++)
5593 		kfree(msr_filter->ranges[i].bitmap);
5594 
5595 	kfree(msr_filter);
5596 }
5597 
5598 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5599 			      struct kvm_msr_filter_range *user_range)
5600 {
5601 	unsigned long *bitmap = NULL;
5602 	size_t bitmap_size;
5603 
5604 	if (!user_range->nmsrs)
5605 		return 0;
5606 
5607 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5608 		return -EINVAL;
5609 
5610 	if (!user_range->flags)
5611 		return -EINVAL;
5612 
5613 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5614 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5615 		return -EINVAL;
5616 
5617 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5618 	if (IS_ERR(bitmap))
5619 		return PTR_ERR(bitmap);
5620 
5621 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5622 		.flags = user_range->flags,
5623 		.base = user_range->base,
5624 		.nmsrs = user_range->nmsrs,
5625 		.bitmap = bitmap,
5626 	};
5627 
5628 	msr_filter->count++;
5629 	return 0;
5630 }
5631 
5632 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5633 {
5634 	struct kvm_msr_filter __user *user_msr_filter = argp;
5635 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5636 	struct kvm_msr_filter filter;
5637 	bool default_allow;
5638 	bool empty = true;
5639 	int r = 0;
5640 	u32 i;
5641 
5642 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5643 		return -EFAULT;
5644 
5645 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5646 		empty &= !filter.ranges[i].nmsrs;
5647 
5648 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5649 	if (empty && !default_allow)
5650 		return -EINVAL;
5651 
5652 	new_filter = kvm_alloc_msr_filter(default_allow);
5653 	if (!new_filter)
5654 		return -ENOMEM;
5655 
5656 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5657 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5658 		if (r) {
5659 			kvm_free_msr_filter(new_filter);
5660 			return r;
5661 		}
5662 	}
5663 
5664 	mutex_lock(&kvm->lock);
5665 
5666 	/* The per-VM filter is protected by kvm->lock... */
5667 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5668 
5669 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5670 	synchronize_srcu(&kvm->srcu);
5671 
5672 	kvm_free_msr_filter(old_filter);
5673 
5674 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5675 	mutex_unlock(&kvm->lock);
5676 
5677 	return 0;
5678 }
5679 
5680 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5681 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5682 {
5683 	struct kvm_vcpu *vcpu;
5684 	int i, ret = 0;
5685 
5686 	mutex_lock(&kvm->lock);
5687 	kvm_for_each_vcpu(i, vcpu, kvm) {
5688 		if (!vcpu->arch.pv_time_enabled)
5689 			continue;
5690 
5691 		ret = kvm_set_guest_paused(vcpu);
5692 		if (ret) {
5693 			kvm_err("Failed to pause guest VCPU%d: %d\n",
5694 				vcpu->vcpu_id, ret);
5695 			break;
5696 		}
5697 	}
5698 	mutex_unlock(&kvm->lock);
5699 
5700 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
5701 }
5702 
5703 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5704 {
5705 	switch (state) {
5706 	case PM_HIBERNATION_PREPARE:
5707 	case PM_SUSPEND_PREPARE:
5708 		return kvm_arch_suspend_notifier(kvm);
5709 	}
5710 
5711 	return NOTIFY_DONE;
5712 }
5713 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5714 
5715 long kvm_arch_vm_ioctl(struct file *filp,
5716 		       unsigned int ioctl, unsigned long arg)
5717 {
5718 	struct kvm *kvm = filp->private_data;
5719 	void __user *argp = (void __user *)arg;
5720 	int r = -ENOTTY;
5721 	/*
5722 	 * This union makes it completely explicit to gcc-3.x
5723 	 * that these two variables' stack usage should be
5724 	 * combined, not added together.
5725 	 */
5726 	union {
5727 		struct kvm_pit_state ps;
5728 		struct kvm_pit_state2 ps2;
5729 		struct kvm_pit_config pit_config;
5730 	} u;
5731 
5732 	switch (ioctl) {
5733 	case KVM_SET_TSS_ADDR:
5734 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5735 		break;
5736 	case KVM_SET_IDENTITY_MAP_ADDR: {
5737 		u64 ident_addr;
5738 
5739 		mutex_lock(&kvm->lock);
5740 		r = -EINVAL;
5741 		if (kvm->created_vcpus)
5742 			goto set_identity_unlock;
5743 		r = -EFAULT;
5744 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5745 			goto set_identity_unlock;
5746 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5747 set_identity_unlock:
5748 		mutex_unlock(&kvm->lock);
5749 		break;
5750 	}
5751 	case KVM_SET_NR_MMU_PAGES:
5752 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5753 		break;
5754 	case KVM_GET_NR_MMU_PAGES:
5755 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5756 		break;
5757 	case KVM_CREATE_IRQCHIP: {
5758 		mutex_lock(&kvm->lock);
5759 
5760 		r = -EEXIST;
5761 		if (irqchip_in_kernel(kvm))
5762 			goto create_irqchip_unlock;
5763 
5764 		r = -EINVAL;
5765 		if (kvm->created_vcpus)
5766 			goto create_irqchip_unlock;
5767 
5768 		r = kvm_pic_init(kvm);
5769 		if (r)
5770 			goto create_irqchip_unlock;
5771 
5772 		r = kvm_ioapic_init(kvm);
5773 		if (r) {
5774 			kvm_pic_destroy(kvm);
5775 			goto create_irqchip_unlock;
5776 		}
5777 
5778 		r = kvm_setup_default_irq_routing(kvm);
5779 		if (r) {
5780 			kvm_ioapic_destroy(kvm);
5781 			kvm_pic_destroy(kvm);
5782 			goto create_irqchip_unlock;
5783 		}
5784 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5785 		smp_wmb();
5786 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5787 	create_irqchip_unlock:
5788 		mutex_unlock(&kvm->lock);
5789 		break;
5790 	}
5791 	case KVM_CREATE_PIT:
5792 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5793 		goto create_pit;
5794 	case KVM_CREATE_PIT2:
5795 		r = -EFAULT;
5796 		if (copy_from_user(&u.pit_config, argp,
5797 				   sizeof(struct kvm_pit_config)))
5798 			goto out;
5799 	create_pit:
5800 		mutex_lock(&kvm->lock);
5801 		r = -EEXIST;
5802 		if (kvm->arch.vpit)
5803 			goto create_pit_unlock;
5804 		r = -ENOMEM;
5805 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5806 		if (kvm->arch.vpit)
5807 			r = 0;
5808 	create_pit_unlock:
5809 		mutex_unlock(&kvm->lock);
5810 		break;
5811 	case KVM_GET_IRQCHIP: {
5812 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5813 		struct kvm_irqchip *chip;
5814 
5815 		chip = memdup_user(argp, sizeof(*chip));
5816 		if (IS_ERR(chip)) {
5817 			r = PTR_ERR(chip);
5818 			goto out;
5819 		}
5820 
5821 		r = -ENXIO;
5822 		if (!irqchip_kernel(kvm))
5823 			goto get_irqchip_out;
5824 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5825 		if (r)
5826 			goto get_irqchip_out;
5827 		r = -EFAULT;
5828 		if (copy_to_user(argp, chip, sizeof(*chip)))
5829 			goto get_irqchip_out;
5830 		r = 0;
5831 	get_irqchip_out:
5832 		kfree(chip);
5833 		break;
5834 	}
5835 	case KVM_SET_IRQCHIP: {
5836 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5837 		struct kvm_irqchip *chip;
5838 
5839 		chip = memdup_user(argp, sizeof(*chip));
5840 		if (IS_ERR(chip)) {
5841 			r = PTR_ERR(chip);
5842 			goto out;
5843 		}
5844 
5845 		r = -ENXIO;
5846 		if (!irqchip_kernel(kvm))
5847 			goto set_irqchip_out;
5848 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5849 	set_irqchip_out:
5850 		kfree(chip);
5851 		break;
5852 	}
5853 	case KVM_GET_PIT: {
5854 		r = -EFAULT;
5855 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5856 			goto out;
5857 		r = -ENXIO;
5858 		if (!kvm->arch.vpit)
5859 			goto out;
5860 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5861 		if (r)
5862 			goto out;
5863 		r = -EFAULT;
5864 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5865 			goto out;
5866 		r = 0;
5867 		break;
5868 	}
5869 	case KVM_SET_PIT: {
5870 		r = -EFAULT;
5871 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5872 			goto out;
5873 		mutex_lock(&kvm->lock);
5874 		r = -ENXIO;
5875 		if (!kvm->arch.vpit)
5876 			goto set_pit_out;
5877 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5878 set_pit_out:
5879 		mutex_unlock(&kvm->lock);
5880 		break;
5881 	}
5882 	case KVM_GET_PIT2: {
5883 		r = -ENXIO;
5884 		if (!kvm->arch.vpit)
5885 			goto out;
5886 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5887 		if (r)
5888 			goto out;
5889 		r = -EFAULT;
5890 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5891 			goto out;
5892 		r = 0;
5893 		break;
5894 	}
5895 	case KVM_SET_PIT2: {
5896 		r = -EFAULT;
5897 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5898 			goto out;
5899 		mutex_lock(&kvm->lock);
5900 		r = -ENXIO;
5901 		if (!kvm->arch.vpit)
5902 			goto set_pit2_out;
5903 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5904 set_pit2_out:
5905 		mutex_unlock(&kvm->lock);
5906 		break;
5907 	}
5908 	case KVM_REINJECT_CONTROL: {
5909 		struct kvm_reinject_control control;
5910 		r =  -EFAULT;
5911 		if (copy_from_user(&control, argp, sizeof(control)))
5912 			goto out;
5913 		r = -ENXIO;
5914 		if (!kvm->arch.vpit)
5915 			goto out;
5916 		r = kvm_vm_ioctl_reinject(kvm, &control);
5917 		break;
5918 	}
5919 	case KVM_SET_BOOT_CPU_ID:
5920 		r = 0;
5921 		mutex_lock(&kvm->lock);
5922 		if (kvm->created_vcpus)
5923 			r = -EBUSY;
5924 		else
5925 			kvm->arch.bsp_vcpu_id = arg;
5926 		mutex_unlock(&kvm->lock);
5927 		break;
5928 #ifdef CONFIG_KVM_XEN
5929 	case KVM_XEN_HVM_CONFIG: {
5930 		struct kvm_xen_hvm_config xhc;
5931 		r = -EFAULT;
5932 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5933 			goto out;
5934 		r = kvm_xen_hvm_config(kvm, &xhc);
5935 		break;
5936 	}
5937 	case KVM_XEN_HVM_GET_ATTR: {
5938 		struct kvm_xen_hvm_attr xha;
5939 
5940 		r = -EFAULT;
5941 		if (copy_from_user(&xha, argp, sizeof(xha)))
5942 			goto out;
5943 		r = kvm_xen_hvm_get_attr(kvm, &xha);
5944 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5945 			r = -EFAULT;
5946 		break;
5947 	}
5948 	case KVM_XEN_HVM_SET_ATTR: {
5949 		struct kvm_xen_hvm_attr xha;
5950 
5951 		r = -EFAULT;
5952 		if (copy_from_user(&xha, argp, sizeof(xha)))
5953 			goto out;
5954 		r = kvm_xen_hvm_set_attr(kvm, &xha);
5955 		break;
5956 	}
5957 #endif
5958 	case KVM_SET_CLOCK: {
5959 		struct kvm_arch *ka = &kvm->arch;
5960 		struct kvm_clock_data user_ns;
5961 		u64 now_ns;
5962 
5963 		r = -EFAULT;
5964 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5965 			goto out;
5966 
5967 		r = -EINVAL;
5968 		if (user_ns.flags)
5969 			goto out;
5970 
5971 		r = 0;
5972 		/*
5973 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5974 		 * kvm_gen_update_masterclock() can be cut down to locked
5975 		 * pvclock_update_vm_gtod_copy().
5976 		 */
5977 		kvm_gen_update_masterclock(kvm);
5978 
5979 		/*
5980 		 * This pairs with kvm_guest_time_update(): when masterclock is
5981 		 * in use, we use master_kernel_ns + kvmclock_offset to set
5982 		 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5983 		 * is slightly ahead) here we risk going negative on unsigned
5984 		 * 'system_time' when 'user_ns.clock' is very small.
5985 		 */
5986 		raw_spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5987 		if (kvm->arch.use_master_clock)
5988 			now_ns = ka->master_kernel_ns;
5989 		else
5990 			now_ns = get_kvmclock_base_ns();
5991 		ka->kvmclock_offset = user_ns.clock - now_ns;
5992 		raw_spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5993 
5994 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5995 		break;
5996 	}
5997 	case KVM_GET_CLOCK: {
5998 		struct kvm_clock_data user_ns;
5999 		u64 now_ns;
6000 
6001 		now_ns = get_kvmclock_ns(kvm);
6002 		user_ns.clock = now_ns;
6003 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6004 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6005 
6006 		r = -EFAULT;
6007 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6008 			goto out;
6009 		r = 0;
6010 		break;
6011 	}
6012 	case KVM_MEMORY_ENCRYPT_OP: {
6013 		r = -ENOTTY;
6014 		if (kvm_x86_ops.mem_enc_op)
6015 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6016 		break;
6017 	}
6018 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6019 		struct kvm_enc_region region;
6020 
6021 		r = -EFAULT;
6022 		if (copy_from_user(&region, argp, sizeof(region)))
6023 			goto out;
6024 
6025 		r = -ENOTTY;
6026 		if (kvm_x86_ops.mem_enc_reg_region)
6027 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6028 		break;
6029 	}
6030 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6031 		struct kvm_enc_region region;
6032 
6033 		r = -EFAULT;
6034 		if (copy_from_user(&region, argp, sizeof(region)))
6035 			goto out;
6036 
6037 		r = -ENOTTY;
6038 		if (kvm_x86_ops.mem_enc_unreg_region)
6039 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6040 		break;
6041 	}
6042 	case KVM_HYPERV_EVENTFD: {
6043 		struct kvm_hyperv_eventfd hvevfd;
6044 
6045 		r = -EFAULT;
6046 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6047 			goto out;
6048 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6049 		break;
6050 	}
6051 	case KVM_SET_PMU_EVENT_FILTER:
6052 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6053 		break;
6054 	case KVM_X86_SET_MSR_FILTER:
6055 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6056 		break;
6057 	default:
6058 		r = -ENOTTY;
6059 	}
6060 out:
6061 	return r;
6062 }
6063 
6064 static void kvm_init_msr_list(void)
6065 {
6066 	struct x86_pmu_capability x86_pmu;
6067 	u32 dummy[2];
6068 	unsigned i;
6069 
6070 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6071 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6072 
6073 	perf_get_x86_pmu_capability(&x86_pmu);
6074 
6075 	num_msrs_to_save = 0;
6076 	num_emulated_msrs = 0;
6077 	num_msr_based_features = 0;
6078 
6079 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6080 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6081 			continue;
6082 
6083 		/*
6084 		 * Even MSRs that are valid in the host may not be exposed
6085 		 * to the guests in some cases.
6086 		 */
6087 		switch (msrs_to_save_all[i]) {
6088 		case MSR_IA32_BNDCFGS:
6089 			if (!kvm_mpx_supported())
6090 				continue;
6091 			break;
6092 		case MSR_TSC_AUX:
6093 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6094 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6095 				continue;
6096 			break;
6097 		case MSR_IA32_UMWAIT_CONTROL:
6098 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6099 				continue;
6100 			break;
6101 		case MSR_IA32_RTIT_CTL:
6102 		case MSR_IA32_RTIT_STATUS:
6103 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6104 				continue;
6105 			break;
6106 		case MSR_IA32_RTIT_CR3_MATCH:
6107 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6108 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6109 				continue;
6110 			break;
6111 		case MSR_IA32_RTIT_OUTPUT_BASE:
6112 		case MSR_IA32_RTIT_OUTPUT_MASK:
6113 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6114 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6115 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6116 				continue;
6117 			break;
6118 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6119 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6120 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6121 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6122 				continue;
6123 			break;
6124 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6125 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6126 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6127 				continue;
6128 			break;
6129 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6130 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6131 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6132 				continue;
6133 			break;
6134 		default:
6135 			break;
6136 		}
6137 
6138 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6139 	}
6140 
6141 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6142 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6143 			continue;
6144 
6145 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6146 	}
6147 
6148 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6149 		struct kvm_msr_entry msr;
6150 
6151 		msr.index = msr_based_features_all[i];
6152 		if (kvm_get_msr_feature(&msr))
6153 			continue;
6154 
6155 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6156 	}
6157 }
6158 
6159 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6160 			   const void *v)
6161 {
6162 	int handled = 0;
6163 	int n;
6164 
6165 	do {
6166 		n = min(len, 8);
6167 		if (!(lapic_in_kernel(vcpu) &&
6168 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6169 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6170 			break;
6171 		handled += n;
6172 		addr += n;
6173 		len -= n;
6174 		v += n;
6175 	} while (len);
6176 
6177 	return handled;
6178 }
6179 
6180 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6181 {
6182 	int handled = 0;
6183 	int n;
6184 
6185 	do {
6186 		n = min(len, 8);
6187 		if (!(lapic_in_kernel(vcpu) &&
6188 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6189 					 addr, n, v))
6190 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6191 			break;
6192 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6193 		handled += n;
6194 		addr += n;
6195 		len -= n;
6196 		v += n;
6197 	} while (len);
6198 
6199 	return handled;
6200 }
6201 
6202 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6203 			struct kvm_segment *var, int seg)
6204 {
6205 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6206 }
6207 
6208 void kvm_get_segment(struct kvm_vcpu *vcpu,
6209 		     struct kvm_segment *var, int seg)
6210 {
6211 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6212 }
6213 
6214 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6215 			   struct x86_exception *exception)
6216 {
6217 	gpa_t t_gpa;
6218 
6219 	BUG_ON(!mmu_is_nested(vcpu));
6220 
6221 	/* NPT walks are always user-walks */
6222 	access |= PFERR_USER_MASK;
6223 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6224 
6225 	return t_gpa;
6226 }
6227 
6228 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6229 			      struct x86_exception *exception)
6230 {
6231 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6232 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6233 }
6234 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6235 
6236  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6237 				struct x86_exception *exception)
6238 {
6239 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6240 	access |= PFERR_FETCH_MASK;
6241 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6242 }
6243 
6244 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6245 			       struct x86_exception *exception)
6246 {
6247 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6248 	access |= PFERR_WRITE_MASK;
6249 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6250 }
6251 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6252 
6253 /* uses this to access any guest's mapped memory without checking CPL */
6254 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6255 				struct x86_exception *exception)
6256 {
6257 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6258 }
6259 
6260 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6261 				      struct kvm_vcpu *vcpu, u32 access,
6262 				      struct x86_exception *exception)
6263 {
6264 	void *data = val;
6265 	int r = X86EMUL_CONTINUE;
6266 
6267 	while (bytes) {
6268 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6269 							    exception);
6270 		unsigned offset = addr & (PAGE_SIZE-1);
6271 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6272 		int ret;
6273 
6274 		if (gpa == UNMAPPED_GVA)
6275 			return X86EMUL_PROPAGATE_FAULT;
6276 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6277 					       offset, toread);
6278 		if (ret < 0) {
6279 			r = X86EMUL_IO_NEEDED;
6280 			goto out;
6281 		}
6282 
6283 		bytes -= toread;
6284 		data += toread;
6285 		addr += toread;
6286 	}
6287 out:
6288 	return r;
6289 }
6290 
6291 /* used for instruction fetching */
6292 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6293 				gva_t addr, void *val, unsigned int bytes,
6294 				struct x86_exception *exception)
6295 {
6296 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6297 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6298 	unsigned offset;
6299 	int ret;
6300 
6301 	/* Inline kvm_read_guest_virt_helper for speed.  */
6302 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6303 						    exception);
6304 	if (unlikely(gpa == UNMAPPED_GVA))
6305 		return X86EMUL_PROPAGATE_FAULT;
6306 
6307 	offset = addr & (PAGE_SIZE-1);
6308 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6309 		bytes = (unsigned)PAGE_SIZE - offset;
6310 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6311 				       offset, bytes);
6312 	if (unlikely(ret < 0))
6313 		return X86EMUL_IO_NEEDED;
6314 
6315 	return X86EMUL_CONTINUE;
6316 }
6317 
6318 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6319 			       gva_t addr, void *val, unsigned int bytes,
6320 			       struct x86_exception *exception)
6321 {
6322 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6323 
6324 	/*
6325 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6326 	 * is returned, but our callers are not ready for that and they blindly
6327 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6328 	 * uninitialized kernel stack memory into cr2 and error code.
6329 	 */
6330 	memset(exception, 0, sizeof(*exception));
6331 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6332 					  exception);
6333 }
6334 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6335 
6336 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6337 			     gva_t addr, void *val, unsigned int bytes,
6338 			     struct x86_exception *exception, bool system)
6339 {
6340 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6341 	u32 access = 0;
6342 
6343 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6344 		access |= PFERR_USER_MASK;
6345 
6346 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6347 }
6348 
6349 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6350 		unsigned long addr, void *val, unsigned int bytes)
6351 {
6352 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6353 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6354 
6355 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6356 }
6357 
6358 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6359 				      struct kvm_vcpu *vcpu, u32 access,
6360 				      struct x86_exception *exception)
6361 {
6362 	void *data = val;
6363 	int r = X86EMUL_CONTINUE;
6364 
6365 	while (bytes) {
6366 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6367 							     access,
6368 							     exception);
6369 		unsigned offset = addr & (PAGE_SIZE-1);
6370 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6371 		int ret;
6372 
6373 		if (gpa == UNMAPPED_GVA)
6374 			return X86EMUL_PROPAGATE_FAULT;
6375 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6376 		if (ret < 0) {
6377 			r = X86EMUL_IO_NEEDED;
6378 			goto out;
6379 		}
6380 
6381 		bytes -= towrite;
6382 		data += towrite;
6383 		addr += towrite;
6384 	}
6385 out:
6386 	return r;
6387 }
6388 
6389 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6390 			      unsigned int bytes, struct x86_exception *exception,
6391 			      bool system)
6392 {
6393 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6394 	u32 access = PFERR_WRITE_MASK;
6395 
6396 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6397 		access |= PFERR_USER_MASK;
6398 
6399 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6400 					   access, exception);
6401 }
6402 
6403 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6404 				unsigned int bytes, struct x86_exception *exception)
6405 {
6406 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6407 	vcpu->arch.l1tf_flush_l1d = true;
6408 
6409 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6410 					   PFERR_WRITE_MASK, exception);
6411 }
6412 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6413 
6414 int handle_ud(struct kvm_vcpu *vcpu)
6415 {
6416 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6417 	int emul_type = EMULTYPE_TRAP_UD;
6418 	char sig[5]; /* ud2; .ascii "kvm" */
6419 	struct x86_exception e;
6420 
6421 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6422 		return 1;
6423 
6424 	if (force_emulation_prefix &&
6425 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6426 				sig, sizeof(sig), &e) == 0 &&
6427 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6428 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6429 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6430 	}
6431 
6432 	return kvm_emulate_instruction(vcpu, emul_type);
6433 }
6434 EXPORT_SYMBOL_GPL(handle_ud);
6435 
6436 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6437 			    gpa_t gpa, bool write)
6438 {
6439 	/* For APIC access vmexit */
6440 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6441 		return 1;
6442 
6443 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6444 		trace_vcpu_match_mmio(gva, gpa, write, true);
6445 		return 1;
6446 	}
6447 
6448 	return 0;
6449 }
6450 
6451 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6452 				gpa_t *gpa, struct x86_exception *exception,
6453 				bool write)
6454 {
6455 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6456 		| (write ? PFERR_WRITE_MASK : 0);
6457 
6458 	/*
6459 	 * currently PKRU is only applied to ept enabled guest so
6460 	 * there is no pkey in EPT page table for L1 guest or EPT
6461 	 * shadow page table for L2 guest.
6462 	 */
6463 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6464 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
6465 			      vcpu->arch.mmio_access, 0, access))) {
6466 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6467 					(gva & (PAGE_SIZE - 1));
6468 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6469 		return 1;
6470 	}
6471 
6472 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6473 
6474 	if (*gpa == UNMAPPED_GVA)
6475 		return -1;
6476 
6477 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6478 }
6479 
6480 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6481 			const void *val, int bytes)
6482 {
6483 	int ret;
6484 
6485 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6486 	if (ret < 0)
6487 		return 0;
6488 	kvm_page_track_write(vcpu, gpa, val, bytes);
6489 	return 1;
6490 }
6491 
6492 struct read_write_emulator_ops {
6493 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6494 				  int bytes);
6495 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6496 				  void *val, int bytes);
6497 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6498 			       int bytes, void *val);
6499 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6500 				    void *val, int bytes);
6501 	bool write;
6502 };
6503 
6504 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6505 {
6506 	if (vcpu->mmio_read_completed) {
6507 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6508 			       vcpu->mmio_fragments[0].gpa, val);
6509 		vcpu->mmio_read_completed = 0;
6510 		return 1;
6511 	}
6512 
6513 	return 0;
6514 }
6515 
6516 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6517 			void *val, int bytes)
6518 {
6519 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6520 }
6521 
6522 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6523 			 void *val, int bytes)
6524 {
6525 	return emulator_write_phys(vcpu, gpa, val, bytes);
6526 }
6527 
6528 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6529 {
6530 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6531 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6532 }
6533 
6534 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6535 			  void *val, int bytes)
6536 {
6537 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6538 	return X86EMUL_IO_NEEDED;
6539 }
6540 
6541 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6542 			   void *val, int bytes)
6543 {
6544 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6545 
6546 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6547 	return X86EMUL_CONTINUE;
6548 }
6549 
6550 static const struct read_write_emulator_ops read_emultor = {
6551 	.read_write_prepare = read_prepare,
6552 	.read_write_emulate = read_emulate,
6553 	.read_write_mmio = vcpu_mmio_read,
6554 	.read_write_exit_mmio = read_exit_mmio,
6555 };
6556 
6557 static const struct read_write_emulator_ops write_emultor = {
6558 	.read_write_emulate = write_emulate,
6559 	.read_write_mmio = write_mmio,
6560 	.read_write_exit_mmio = write_exit_mmio,
6561 	.write = true,
6562 };
6563 
6564 static int emulator_read_write_onepage(unsigned long addr, void *val,
6565 				       unsigned int bytes,
6566 				       struct x86_exception *exception,
6567 				       struct kvm_vcpu *vcpu,
6568 				       const struct read_write_emulator_ops *ops)
6569 {
6570 	gpa_t gpa;
6571 	int handled, ret;
6572 	bool write = ops->write;
6573 	struct kvm_mmio_fragment *frag;
6574 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6575 
6576 	/*
6577 	 * If the exit was due to a NPF we may already have a GPA.
6578 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6579 	 * Note, this cannot be used on string operations since string
6580 	 * operation using rep will only have the initial GPA from the NPF
6581 	 * occurred.
6582 	 */
6583 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6584 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6585 		gpa = ctxt->gpa_val;
6586 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6587 	} else {
6588 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6589 		if (ret < 0)
6590 			return X86EMUL_PROPAGATE_FAULT;
6591 	}
6592 
6593 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6594 		return X86EMUL_CONTINUE;
6595 
6596 	/*
6597 	 * Is this MMIO handled locally?
6598 	 */
6599 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6600 	if (handled == bytes)
6601 		return X86EMUL_CONTINUE;
6602 
6603 	gpa += handled;
6604 	bytes -= handled;
6605 	val += handled;
6606 
6607 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6608 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6609 	frag->gpa = gpa;
6610 	frag->data = val;
6611 	frag->len = bytes;
6612 	return X86EMUL_CONTINUE;
6613 }
6614 
6615 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6616 			unsigned long addr,
6617 			void *val, unsigned int bytes,
6618 			struct x86_exception *exception,
6619 			const struct read_write_emulator_ops *ops)
6620 {
6621 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6622 	gpa_t gpa;
6623 	int rc;
6624 
6625 	if (ops->read_write_prepare &&
6626 		  ops->read_write_prepare(vcpu, val, bytes))
6627 		return X86EMUL_CONTINUE;
6628 
6629 	vcpu->mmio_nr_fragments = 0;
6630 
6631 	/* Crossing a page boundary? */
6632 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6633 		int now;
6634 
6635 		now = -addr & ~PAGE_MASK;
6636 		rc = emulator_read_write_onepage(addr, val, now, exception,
6637 						 vcpu, ops);
6638 
6639 		if (rc != X86EMUL_CONTINUE)
6640 			return rc;
6641 		addr += now;
6642 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6643 			addr = (u32)addr;
6644 		val += now;
6645 		bytes -= now;
6646 	}
6647 
6648 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6649 					 vcpu, ops);
6650 	if (rc != X86EMUL_CONTINUE)
6651 		return rc;
6652 
6653 	if (!vcpu->mmio_nr_fragments)
6654 		return rc;
6655 
6656 	gpa = vcpu->mmio_fragments[0].gpa;
6657 
6658 	vcpu->mmio_needed = 1;
6659 	vcpu->mmio_cur_fragment = 0;
6660 
6661 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6662 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6663 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6664 	vcpu->run->mmio.phys_addr = gpa;
6665 
6666 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6667 }
6668 
6669 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6670 				  unsigned long addr,
6671 				  void *val,
6672 				  unsigned int bytes,
6673 				  struct x86_exception *exception)
6674 {
6675 	return emulator_read_write(ctxt, addr, val, bytes,
6676 				   exception, &read_emultor);
6677 }
6678 
6679 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6680 			    unsigned long addr,
6681 			    const void *val,
6682 			    unsigned int bytes,
6683 			    struct x86_exception *exception)
6684 {
6685 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6686 				   exception, &write_emultor);
6687 }
6688 
6689 #define CMPXCHG_TYPE(t, ptr, old, new) \
6690 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6691 
6692 #ifdef CONFIG_X86_64
6693 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6694 #else
6695 #  define CMPXCHG64(ptr, old, new) \
6696 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6697 #endif
6698 
6699 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6700 				     unsigned long addr,
6701 				     const void *old,
6702 				     const void *new,
6703 				     unsigned int bytes,
6704 				     struct x86_exception *exception)
6705 {
6706 	struct kvm_host_map map;
6707 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6708 	u64 page_line_mask;
6709 	gpa_t gpa;
6710 	char *kaddr;
6711 	bool exchanged;
6712 
6713 	/* guests cmpxchg8b have to be emulated atomically */
6714 	if (bytes > 8 || (bytes & (bytes - 1)))
6715 		goto emul_write;
6716 
6717 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6718 
6719 	if (gpa == UNMAPPED_GVA ||
6720 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6721 		goto emul_write;
6722 
6723 	/*
6724 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6725 	 * enabled in the host and the access splits a cache line.
6726 	 */
6727 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6728 		page_line_mask = ~(cache_line_size() - 1);
6729 	else
6730 		page_line_mask = PAGE_MASK;
6731 
6732 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6733 		goto emul_write;
6734 
6735 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6736 		goto emul_write;
6737 
6738 	kaddr = map.hva + offset_in_page(gpa);
6739 
6740 	switch (bytes) {
6741 	case 1:
6742 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6743 		break;
6744 	case 2:
6745 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6746 		break;
6747 	case 4:
6748 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6749 		break;
6750 	case 8:
6751 		exchanged = CMPXCHG64(kaddr, old, new);
6752 		break;
6753 	default:
6754 		BUG();
6755 	}
6756 
6757 	kvm_vcpu_unmap(vcpu, &map, true);
6758 
6759 	if (!exchanged)
6760 		return X86EMUL_CMPXCHG_FAILED;
6761 
6762 	kvm_page_track_write(vcpu, gpa, new, bytes);
6763 
6764 	return X86EMUL_CONTINUE;
6765 
6766 emul_write:
6767 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6768 
6769 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6770 }
6771 
6772 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6773 {
6774 	int r = 0, i;
6775 
6776 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6777 		if (vcpu->arch.pio.in)
6778 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6779 					    vcpu->arch.pio.size, pd);
6780 		else
6781 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6782 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6783 					     pd);
6784 		if (r)
6785 			break;
6786 		pd += vcpu->arch.pio.size;
6787 	}
6788 	return r;
6789 }
6790 
6791 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6792 			       unsigned short port,
6793 			       unsigned int count, bool in)
6794 {
6795 	vcpu->arch.pio.port = port;
6796 	vcpu->arch.pio.in = in;
6797 	vcpu->arch.pio.count  = count;
6798 	vcpu->arch.pio.size = size;
6799 
6800 	if (!kernel_pio(vcpu, vcpu->arch.pio_data))
6801 		return 1;
6802 
6803 	vcpu->run->exit_reason = KVM_EXIT_IO;
6804 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6805 	vcpu->run->io.size = size;
6806 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6807 	vcpu->run->io.count = count;
6808 	vcpu->run->io.port = port;
6809 
6810 	return 0;
6811 }
6812 
6813 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6814 			     unsigned short port, unsigned int count)
6815 {
6816 	WARN_ON(vcpu->arch.pio.count);
6817 	memset(vcpu->arch.pio_data, 0, size * count);
6818 	return emulator_pio_in_out(vcpu, size, port, count, true);
6819 }
6820 
6821 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
6822 {
6823 	int size = vcpu->arch.pio.size;
6824 	unsigned count = vcpu->arch.pio.count;
6825 	memcpy(val, vcpu->arch.pio_data, size * count);
6826 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
6827 	vcpu->arch.pio.count = 0;
6828 }
6829 
6830 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6831 			   unsigned short port, void *val, unsigned int count)
6832 {
6833 	if (vcpu->arch.pio.count) {
6834 		/* Complete previous iteration.  */
6835 	} else {
6836 		int r = __emulator_pio_in(vcpu, size, port, count);
6837 		if (!r)
6838 			return r;
6839 
6840 		/* Results already available, fall through.  */
6841 	}
6842 
6843 	WARN_ON(count != vcpu->arch.pio.count);
6844 	complete_emulator_pio_in(vcpu, val);
6845 	return 1;
6846 }
6847 
6848 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6849 				    int size, unsigned short port, void *val,
6850 				    unsigned int count)
6851 {
6852 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6853 
6854 }
6855 
6856 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6857 			    unsigned short port, const void *val,
6858 			    unsigned int count)
6859 {
6860 	int ret;
6861 
6862 	memcpy(vcpu->arch.pio_data, val, size * count);
6863 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6864 	ret = emulator_pio_in_out(vcpu, size, port, count, false);
6865 	if (ret)
6866                 vcpu->arch.pio.count = 0;
6867 
6868         return ret;
6869 }
6870 
6871 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6872 				     int size, unsigned short port,
6873 				     const void *val, unsigned int count)
6874 {
6875 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6876 }
6877 
6878 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6879 {
6880 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6881 }
6882 
6883 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6884 {
6885 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6886 }
6887 
6888 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6889 {
6890 	if (!need_emulate_wbinvd(vcpu))
6891 		return X86EMUL_CONTINUE;
6892 
6893 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
6894 		int cpu = get_cpu();
6895 
6896 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6897 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6898 				wbinvd_ipi, NULL, 1);
6899 		put_cpu();
6900 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6901 	} else
6902 		wbinvd();
6903 	return X86EMUL_CONTINUE;
6904 }
6905 
6906 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6907 {
6908 	kvm_emulate_wbinvd_noskip(vcpu);
6909 	return kvm_skip_emulated_instruction(vcpu);
6910 }
6911 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6912 
6913 
6914 
6915 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6916 {
6917 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6918 }
6919 
6920 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6921 			    unsigned long *dest)
6922 {
6923 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6924 }
6925 
6926 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6927 			   unsigned long value)
6928 {
6929 
6930 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6931 }
6932 
6933 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6934 {
6935 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6936 }
6937 
6938 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6939 {
6940 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6941 	unsigned long value;
6942 
6943 	switch (cr) {
6944 	case 0:
6945 		value = kvm_read_cr0(vcpu);
6946 		break;
6947 	case 2:
6948 		value = vcpu->arch.cr2;
6949 		break;
6950 	case 3:
6951 		value = kvm_read_cr3(vcpu);
6952 		break;
6953 	case 4:
6954 		value = kvm_read_cr4(vcpu);
6955 		break;
6956 	case 8:
6957 		value = kvm_get_cr8(vcpu);
6958 		break;
6959 	default:
6960 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6961 		return 0;
6962 	}
6963 
6964 	return value;
6965 }
6966 
6967 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6968 {
6969 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6970 	int res = 0;
6971 
6972 	switch (cr) {
6973 	case 0:
6974 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6975 		break;
6976 	case 2:
6977 		vcpu->arch.cr2 = val;
6978 		break;
6979 	case 3:
6980 		res = kvm_set_cr3(vcpu, val);
6981 		break;
6982 	case 4:
6983 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6984 		break;
6985 	case 8:
6986 		res = kvm_set_cr8(vcpu, val);
6987 		break;
6988 	default:
6989 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6990 		res = -1;
6991 	}
6992 
6993 	return res;
6994 }
6995 
6996 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6997 {
6998 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6999 }
7000 
7001 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7002 {
7003 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7004 }
7005 
7006 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7007 {
7008 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7009 }
7010 
7011 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7012 {
7013 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7014 }
7015 
7016 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7017 {
7018 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7019 }
7020 
7021 static unsigned long emulator_get_cached_segment_base(
7022 	struct x86_emulate_ctxt *ctxt, int seg)
7023 {
7024 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7025 }
7026 
7027 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7028 				 struct desc_struct *desc, u32 *base3,
7029 				 int seg)
7030 {
7031 	struct kvm_segment var;
7032 
7033 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7034 	*selector = var.selector;
7035 
7036 	if (var.unusable) {
7037 		memset(desc, 0, sizeof(*desc));
7038 		if (base3)
7039 			*base3 = 0;
7040 		return false;
7041 	}
7042 
7043 	if (var.g)
7044 		var.limit >>= 12;
7045 	set_desc_limit(desc, var.limit);
7046 	set_desc_base(desc, (unsigned long)var.base);
7047 #ifdef CONFIG_X86_64
7048 	if (base3)
7049 		*base3 = var.base >> 32;
7050 #endif
7051 	desc->type = var.type;
7052 	desc->s = var.s;
7053 	desc->dpl = var.dpl;
7054 	desc->p = var.present;
7055 	desc->avl = var.avl;
7056 	desc->l = var.l;
7057 	desc->d = var.db;
7058 	desc->g = var.g;
7059 
7060 	return true;
7061 }
7062 
7063 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7064 				 struct desc_struct *desc, u32 base3,
7065 				 int seg)
7066 {
7067 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7068 	struct kvm_segment var;
7069 
7070 	var.selector = selector;
7071 	var.base = get_desc_base(desc);
7072 #ifdef CONFIG_X86_64
7073 	var.base |= ((u64)base3) << 32;
7074 #endif
7075 	var.limit = get_desc_limit(desc);
7076 	if (desc->g)
7077 		var.limit = (var.limit << 12) | 0xfff;
7078 	var.type = desc->type;
7079 	var.dpl = desc->dpl;
7080 	var.db = desc->d;
7081 	var.s = desc->s;
7082 	var.l = desc->l;
7083 	var.g = desc->g;
7084 	var.avl = desc->avl;
7085 	var.present = desc->p;
7086 	var.unusable = !var.present;
7087 	var.padding = 0;
7088 
7089 	kvm_set_segment(vcpu, &var, seg);
7090 	return;
7091 }
7092 
7093 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7094 			    u32 msr_index, u64 *pdata)
7095 {
7096 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7097 	int r;
7098 
7099 	r = kvm_get_msr(vcpu, msr_index, pdata);
7100 
7101 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7102 		/* Bounce to user space */
7103 		return X86EMUL_IO_NEEDED;
7104 	}
7105 
7106 	return r;
7107 }
7108 
7109 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7110 			    u32 msr_index, u64 data)
7111 {
7112 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7113 	int r;
7114 
7115 	r = kvm_set_msr(vcpu, msr_index, data);
7116 
7117 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7118 		/* Bounce to user space */
7119 		return X86EMUL_IO_NEEDED;
7120 	}
7121 
7122 	return r;
7123 }
7124 
7125 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7126 {
7127 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7128 
7129 	return vcpu->arch.smbase;
7130 }
7131 
7132 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7133 {
7134 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7135 
7136 	vcpu->arch.smbase = smbase;
7137 }
7138 
7139 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7140 			      u32 pmc)
7141 {
7142 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7143 }
7144 
7145 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7146 			     u32 pmc, u64 *pdata)
7147 {
7148 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7149 }
7150 
7151 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7152 {
7153 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7154 }
7155 
7156 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7157 			      struct x86_instruction_info *info,
7158 			      enum x86_intercept_stage stage)
7159 {
7160 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7161 					    &ctxt->exception);
7162 }
7163 
7164 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7165 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7166 			      bool exact_only)
7167 {
7168 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7169 }
7170 
7171 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7172 {
7173 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7174 }
7175 
7176 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7177 {
7178 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7179 }
7180 
7181 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7182 {
7183 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7184 }
7185 
7186 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7187 {
7188 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7189 }
7190 
7191 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7192 {
7193 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7194 }
7195 
7196 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7197 {
7198 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7199 }
7200 
7201 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7202 {
7203 	return emul_to_vcpu(ctxt)->arch.hflags;
7204 }
7205 
7206 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7207 {
7208 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7209 
7210 	kvm_smm_changed(vcpu, false);
7211 }
7212 
7213 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7214 				  const char *smstate)
7215 {
7216 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7217 }
7218 
7219 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7220 {
7221 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7222 }
7223 
7224 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7225 {
7226 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7227 }
7228 
7229 static const struct x86_emulate_ops emulate_ops = {
7230 	.read_gpr            = emulator_read_gpr,
7231 	.write_gpr           = emulator_write_gpr,
7232 	.read_std            = emulator_read_std,
7233 	.write_std           = emulator_write_std,
7234 	.read_phys           = kvm_read_guest_phys_system,
7235 	.fetch               = kvm_fetch_guest_virt,
7236 	.read_emulated       = emulator_read_emulated,
7237 	.write_emulated      = emulator_write_emulated,
7238 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7239 	.invlpg              = emulator_invlpg,
7240 	.pio_in_emulated     = emulator_pio_in_emulated,
7241 	.pio_out_emulated    = emulator_pio_out_emulated,
7242 	.get_segment         = emulator_get_segment,
7243 	.set_segment         = emulator_set_segment,
7244 	.get_cached_segment_base = emulator_get_cached_segment_base,
7245 	.get_gdt             = emulator_get_gdt,
7246 	.get_idt	     = emulator_get_idt,
7247 	.set_gdt             = emulator_set_gdt,
7248 	.set_idt	     = emulator_set_idt,
7249 	.get_cr              = emulator_get_cr,
7250 	.set_cr              = emulator_set_cr,
7251 	.cpl                 = emulator_get_cpl,
7252 	.get_dr              = emulator_get_dr,
7253 	.set_dr              = emulator_set_dr,
7254 	.get_smbase          = emulator_get_smbase,
7255 	.set_smbase          = emulator_set_smbase,
7256 	.set_msr             = emulator_set_msr,
7257 	.get_msr             = emulator_get_msr,
7258 	.check_pmc	     = emulator_check_pmc,
7259 	.read_pmc            = emulator_read_pmc,
7260 	.halt                = emulator_halt,
7261 	.wbinvd              = emulator_wbinvd,
7262 	.fix_hypercall       = emulator_fix_hypercall,
7263 	.intercept           = emulator_intercept,
7264 	.get_cpuid           = emulator_get_cpuid,
7265 	.guest_has_long_mode = emulator_guest_has_long_mode,
7266 	.guest_has_movbe     = emulator_guest_has_movbe,
7267 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7268 	.set_nmi_mask        = emulator_set_nmi_mask,
7269 	.get_hflags          = emulator_get_hflags,
7270 	.exiting_smm         = emulator_exiting_smm,
7271 	.leave_smm           = emulator_leave_smm,
7272 	.triple_fault        = emulator_triple_fault,
7273 	.set_xcr             = emulator_set_xcr,
7274 };
7275 
7276 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7277 {
7278 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7279 	/*
7280 	 * an sti; sti; sequence only disable interrupts for the first
7281 	 * instruction. So, if the last instruction, be it emulated or
7282 	 * not, left the system with the INT_STI flag enabled, it
7283 	 * means that the last instruction is an sti. We should not
7284 	 * leave the flag on in this case. The same goes for mov ss
7285 	 */
7286 	if (int_shadow & mask)
7287 		mask = 0;
7288 	if (unlikely(int_shadow || mask)) {
7289 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7290 		if (!mask)
7291 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7292 	}
7293 }
7294 
7295 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7296 {
7297 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7298 	if (ctxt->exception.vector == PF_VECTOR)
7299 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7300 
7301 	if (ctxt->exception.error_code_valid)
7302 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7303 				      ctxt->exception.error_code);
7304 	else
7305 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7306 	return false;
7307 }
7308 
7309 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7310 {
7311 	struct x86_emulate_ctxt *ctxt;
7312 
7313 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7314 	if (!ctxt) {
7315 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7316 		return NULL;
7317 	}
7318 
7319 	ctxt->vcpu = vcpu;
7320 	ctxt->ops = &emulate_ops;
7321 	vcpu->arch.emulate_ctxt = ctxt;
7322 
7323 	return ctxt;
7324 }
7325 
7326 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7327 {
7328 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7329 	int cs_db, cs_l;
7330 
7331 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7332 
7333 	ctxt->gpa_available = false;
7334 	ctxt->eflags = kvm_get_rflags(vcpu);
7335 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7336 
7337 	ctxt->eip = kvm_rip_read(vcpu);
7338 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7339 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7340 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7341 		     cs_db				? X86EMUL_MODE_PROT32 :
7342 							  X86EMUL_MODE_PROT16;
7343 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7344 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7345 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7346 
7347 	ctxt->interruptibility = 0;
7348 	ctxt->have_exception = false;
7349 	ctxt->exception.vector = -1;
7350 	ctxt->perm_ok = false;
7351 
7352 	init_decode_cache(ctxt);
7353 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7354 }
7355 
7356 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7357 {
7358 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7359 	int ret;
7360 
7361 	init_emulate_ctxt(vcpu);
7362 
7363 	ctxt->op_bytes = 2;
7364 	ctxt->ad_bytes = 2;
7365 	ctxt->_eip = ctxt->eip + inc_eip;
7366 	ret = emulate_int_real(ctxt, irq);
7367 
7368 	if (ret != X86EMUL_CONTINUE) {
7369 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7370 	} else {
7371 		ctxt->eip = ctxt->_eip;
7372 		kvm_rip_write(vcpu, ctxt->eip);
7373 		kvm_set_rflags(vcpu, ctxt->eflags);
7374 	}
7375 }
7376 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7377 
7378 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7379 {
7380 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7381 	u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7382 	struct kvm_run *run = vcpu->run;
7383 
7384 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7385 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7386 	run->emulation_failure.ndata = 0;
7387 	run->emulation_failure.flags = 0;
7388 
7389 	if (insn_size) {
7390 		run->emulation_failure.ndata = 3;
7391 		run->emulation_failure.flags |=
7392 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7393 		run->emulation_failure.insn_size = insn_size;
7394 		memset(run->emulation_failure.insn_bytes, 0x90,
7395 		       sizeof(run->emulation_failure.insn_bytes));
7396 		memcpy(run->emulation_failure.insn_bytes,
7397 		       ctxt->fetch.data, insn_size);
7398 	}
7399 }
7400 
7401 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7402 {
7403 	struct kvm *kvm = vcpu->kvm;
7404 
7405 	++vcpu->stat.insn_emulation_fail;
7406 	trace_kvm_emulate_insn_failed(vcpu);
7407 
7408 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7409 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7410 		return 1;
7411 	}
7412 
7413 	if (kvm->arch.exit_on_emulation_error ||
7414 	    (emulation_type & EMULTYPE_SKIP)) {
7415 		prepare_emulation_failure_exit(vcpu);
7416 		return 0;
7417 	}
7418 
7419 	kvm_queue_exception(vcpu, UD_VECTOR);
7420 
7421 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7422 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7423 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7424 		vcpu->run->internal.ndata = 0;
7425 		return 0;
7426 	}
7427 
7428 	return 1;
7429 }
7430 
7431 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7432 				  bool write_fault_to_shadow_pgtable,
7433 				  int emulation_type)
7434 {
7435 	gpa_t gpa = cr2_or_gpa;
7436 	kvm_pfn_t pfn;
7437 
7438 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7439 		return false;
7440 
7441 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7442 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7443 		return false;
7444 
7445 	if (!vcpu->arch.mmu->direct_map) {
7446 		/*
7447 		 * Write permission should be allowed since only
7448 		 * write access need to be emulated.
7449 		 */
7450 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7451 
7452 		/*
7453 		 * If the mapping is invalid in guest, let cpu retry
7454 		 * it to generate fault.
7455 		 */
7456 		if (gpa == UNMAPPED_GVA)
7457 			return true;
7458 	}
7459 
7460 	/*
7461 	 * Do not retry the unhandleable instruction if it faults on the
7462 	 * readonly host memory, otherwise it will goto a infinite loop:
7463 	 * retry instruction -> write #PF -> emulation fail -> retry
7464 	 * instruction -> ...
7465 	 */
7466 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7467 
7468 	/*
7469 	 * If the instruction failed on the error pfn, it can not be fixed,
7470 	 * report the error to userspace.
7471 	 */
7472 	if (is_error_noslot_pfn(pfn))
7473 		return false;
7474 
7475 	kvm_release_pfn_clean(pfn);
7476 
7477 	/* The instructions are well-emulated on direct mmu. */
7478 	if (vcpu->arch.mmu->direct_map) {
7479 		unsigned int indirect_shadow_pages;
7480 
7481 		write_lock(&vcpu->kvm->mmu_lock);
7482 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7483 		write_unlock(&vcpu->kvm->mmu_lock);
7484 
7485 		if (indirect_shadow_pages)
7486 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7487 
7488 		return true;
7489 	}
7490 
7491 	/*
7492 	 * if emulation was due to access to shadowed page table
7493 	 * and it failed try to unshadow page and re-enter the
7494 	 * guest to let CPU execute the instruction.
7495 	 */
7496 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7497 
7498 	/*
7499 	 * If the access faults on its page table, it can not
7500 	 * be fixed by unprotecting shadow page and it should
7501 	 * be reported to userspace.
7502 	 */
7503 	return !write_fault_to_shadow_pgtable;
7504 }
7505 
7506 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7507 			      gpa_t cr2_or_gpa,  int emulation_type)
7508 {
7509 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7510 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7511 
7512 	last_retry_eip = vcpu->arch.last_retry_eip;
7513 	last_retry_addr = vcpu->arch.last_retry_addr;
7514 
7515 	/*
7516 	 * If the emulation is caused by #PF and it is non-page_table
7517 	 * writing instruction, it means the VM-EXIT is caused by shadow
7518 	 * page protected, we can zap the shadow page and retry this
7519 	 * instruction directly.
7520 	 *
7521 	 * Note: if the guest uses a non-page-table modifying instruction
7522 	 * on the PDE that points to the instruction, then we will unmap
7523 	 * the instruction and go to an infinite loop. So, we cache the
7524 	 * last retried eip and the last fault address, if we meet the eip
7525 	 * and the address again, we can break out of the potential infinite
7526 	 * loop.
7527 	 */
7528 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7529 
7530 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7531 		return false;
7532 
7533 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7534 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7535 		return false;
7536 
7537 	if (x86_page_table_writing_insn(ctxt))
7538 		return false;
7539 
7540 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7541 		return false;
7542 
7543 	vcpu->arch.last_retry_eip = ctxt->eip;
7544 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7545 
7546 	if (!vcpu->arch.mmu->direct_map)
7547 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7548 
7549 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7550 
7551 	return true;
7552 }
7553 
7554 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7555 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7556 
7557 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7558 {
7559 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7560 
7561 	if (entering_smm) {
7562 		vcpu->arch.hflags |= HF_SMM_MASK;
7563 	} else {
7564 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7565 
7566 		/* Process a latched INIT or SMI, if any.  */
7567 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7568 
7569 		/*
7570 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7571 		 * on SMM exit we still need to reload them from
7572 		 * guest memory
7573 		 */
7574 		vcpu->arch.pdptrs_from_userspace = false;
7575 	}
7576 
7577 	kvm_mmu_reset_context(vcpu);
7578 }
7579 
7580 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7581 				unsigned long *db)
7582 {
7583 	u32 dr6 = 0;
7584 	int i;
7585 	u32 enable, rwlen;
7586 
7587 	enable = dr7;
7588 	rwlen = dr7 >> 16;
7589 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7590 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7591 			dr6 |= (1 << i);
7592 	return dr6;
7593 }
7594 
7595 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7596 {
7597 	struct kvm_run *kvm_run = vcpu->run;
7598 
7599 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7600 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7601 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7602 		kvm_run->debug.arch.exception = DB_VECTOR;
7603 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7604 		return 0;
7605 	}
7606 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7607 	return 1;
7608 }
7609 
7610 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7611 {
7612 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7613 	int r;
7614 
7615 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7616 	if (unlikely(!r))
7617 		return 0;
7618 
7619 	/*
7620 	 * rflags is the old, "raw" value of the flags.  The new value has
7621 	 * not been saved yet.
7622 	 *
7623 	 * This is correct even for TF set by the guest, because "the
7624 	 * processor will not generate this exception after the instruction
7625 	 * that sets the TF flag".
7626 	 */
7627 	if (unlikely(rflags & X86_EFLAGS_TF))
7628 		r = kvm_vcpu_do_singlestep(vcpu);
7629 	return r;
7630 }
7631 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7632 
7633 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7634 {
7635 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7636 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7637 		struct kvm_run *kvm_run = vcpu->run;
7638 		unsigned long eip = kvm_get_linear_rip(vcpu);
7639 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7640 					   vcpu->arch.guest_debug_dr7,
7641 					   vcpu->arch.eff_db);
7642 
7643 		if (dr6 != 0) {
7644 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7645 			kvm_run->debug.arch.pc = eip;
7646 			kvm_run->debug.arch.exception = DB_VECTOR;
7647 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7648 			*r = 0;
7649 			return true;
7650 		}
7651 	}
7652 
7653 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7654 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7655 		unsigned long eip = kvm_get_linear_rip(vcpu);
7656 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7657 					   vcpu->arch.dr7,
7658 					   vcpu->arch.db);
7659 
7660 		if (dr6 != 0) {
7661 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7662 			*r = 1;
7663 			return true;
7664 		}
7665 	}
7666 
7667 	return false;
7668 }
7669 
7670 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7671 {
7672 	switch (ctxt->opcode_len) {
7673 	case 1:
7674 		switch (ctxt->b) {
7675 		case 0xe4:	/* IN */
7676 		case 0xe5:
7677 		case 0xec:
7678 		case 0xed:
7679 		case 0xe6:	/* OUT */
7680 		case 0xe7:
7681 		case 0xee:
7682 		case 0xef:
7683 		case 0x6c:	/* INS */
7684 		case 0x6d:
7685 		case 0x6e:	/* OUTS */
7686 		case 0x6f:
7687 			return true;
7688 		}
7689 		break;
7690 	case 2:
7691 		switch (ctxt->b) {
7692 		case 0x33:	/* RDPMC */
7693 			return true;
7694 		}
7695 		break;
7696 	}
7697 
7698 	return false;
7699 }
7700 
7701 /*
7702  * Decode to be emulated instruction. Return EMULATION_OK if success.
7703  */
7704 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7705 				    void *insn, int insn_len)
7706 {
7707 	int r = EMULATION_OK;
7708 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7709 
7710 	init_emulate_ctxt(vcpu);
7711 
7712 	/*
7713 	 * We will reenter on the same instruction since we do not set
7714 	 * complete_userspace_io. This does not handle watchpoints yet,
7715 	 * those would be handled in the emulate_ops.
7716 	 */
7717 	if (!(emulation_type & EMULTYPE_SKIP) &&
7718 	    kvm_vcpu_check_breakpoint(vcpu, &r))
7719 		return r;
7720 
7721 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7722 
7723 	trace_kvm_emulate_insn_start(vcpu);
7724 	++vcpu->stat.insn_emulation;
7725 
7726 	return r;
7727 }
7728 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7729 
7730 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7731 			    int emulation_type, void *insn, int insn_len)
7732 {
7733 	int r;
7734 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7735 	bool writeback = true;
7736 	bool write_fault_to_spt;
7737 
7738 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7739 		return 1;
7740 
7741 	vcpu->arch.l1tf_flush_l1d = true;
7742 
7743 	/*
7744 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7745 	 * never reused.
7746 	 */
7747 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7748 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7749 
7750 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7751 		kvm_clear_exception_queue(vcpu);
7752 
7753 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
7754 						    insn, insn_len);
7755 		if (r != EMULATION_OK)  {
7756 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7757 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7758 				kvm_queue_exception(vcpu, UD_VECTOR);
7759 				return 1;
7760 			}
7761 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7762 						  write_fault_to_spt,
7763 						  emulation_type))
7764 				return 1;
7765 			if (ctxt->have_exception) {
7766 				/*
7767 				 * #UD should result in just EMULATION_FAILED, and trap-like
7768 				 * exception should not be encountered during decode.
7769 				 */
7770 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7771 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7772 				inject_emulated_exception(vcpu);
7773 				return 1;
7774 			}
7775 			return handle_emulation_failure(vcpu, emulation_type);
7776 		}
7777 	}
7778 
7779 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7780 	    !is_vmware_backdoor_opcode(ctxt)) {
7781 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7782 		return 1;
7783 	}
7784 
7785 	/*
7786 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7787 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7788 	 * updating interruptibility state and injecting single-step #DBs.
7789 	 */
7790 	if (emulation_type & EMULTYPE_SKIP) {
7791 		kvm_rip_write(vcpu, ctxt->_eip);
7792 		if (ctxt->eflags & X86_EFLAGS_RF)
7793 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7794 		return 1;
7795 	}
7796 
7797 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7798 		return 1;
7799 
7800 	/* this is needed for vmware backdoor interface to work since it
7801 	   changes registers values  during IO operation */
7802 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7803 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7804 		emulator_invalidate_register_cache(ctxt);
7805 	}
7806 
7807 restart:
7808 	if (emulation_type & EMULTYPE_PF) {
7809 		/* Save the faulting GPA (cr2) in the address field */
7810 		ctxt->exception.address = cr2_or_gpa;
7811 
7812 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7813 		if (vcpu->arch.mmu->direct_map) {
7814 			ctxt->gpa_available = true;
7815 			ctxt->gpa_val = cr2_or_gpa;
7816 		}
7817 	} else {
7818 		/* Sanitize the address out of an abundance of paranoia. */
7819 		ctxt->exception.address = 0;
7820 	}
7821 
7822 	r = x86_emulate_insn(ctxt);
7823 
7824 	if (r == EMULATION_INTERCEPTED)
7825 		return 1;
7826 
7827 	if (r == EMULATION_FAILED) {
7828 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7829 					emulation_type))
7830 			return 1;
7831 
7832 		return handle_emulation_failure(vcpu, emulation_type);
7833 	}
7834 
7835 	if (ctxt->have_exception) {
7836 		r = 1;
7837 		if (inject_emulated_exception(vcpu))
7838 			return r;
7839 	} else if (vcpu->arch.pio.count) {
7840 		if (!vcpu->arch.pio.in) {
7841 			/* FIXME: return into emulator if single-stepping.  */
7842 			vcpu->arch.pio.count = 0;
7843 		} else {
7844 			writeback = false;
7845 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7846 		}
7847 		r = 0;
7848 	} else if (vcpu->mmio_needed) {
7849 		++vcpu->stat.mmio_exits;
7850 
7851 		if (!vcpu->mmio_is_write)
7852 			writeback = false;
7853 		r = 0;
7854 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7855 	} else if (r == EMULATION_RESTART)
7856 		goto restart;
7857 	else
7858 		r = 1;
7859 
7860 	if (writeback) {
7861 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7862 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7863 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7864 		if (!ctxt->have_exception ||
7865 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7866 			kvm_rip_write(vcpu, ctxt->eip);
7867 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7868 				r = kvm_vcpu_do_singlestep(vcpu);
7869 			if (kvm_x86_ops.update_emulated_instruction)
7870 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
7871 			__kvm_set_rflags(vcpu, ctxt->eflags);
7872 		}
7873 
7874 		/*
7875 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7876 		 * do nothing, and it will be requested again as soon as
7877 		 * the shadow expires.  But we still need to check here,
7878 		 * because POPF has no interrupt shadow.
7879 		 */
7880 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7881 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7882 	} else
7883 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7884 
7885 	return r;
7886 }
7887 
7888 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7889 {
7890 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7891 }
7892 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7893 
7894 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7895 					void *insn, int insn_len)
7896 {
7897 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7898 }
7899 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7900 
7901 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7902 {
7903 	vcpu->arch.pio.count = 0;
7904 	return 1;
7905 }
7906 
7907 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7908 {
7909 	vcpu->arch.pio.count = 0;
7910 
7911 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7912 		return 1;
7913 
7914 	return kvm_skip_emulated_instruction(vcpu);
7915 }
7916 
7917 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7918 			    unsigned short port)
7919 {
7920 	unsigned long val = kvm_rax_read(vcpu);
7921 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7922 
7923 	if (ret)
7924 		return ret;
7925 
7926 	/*
7927 	 * Workaround userspace that relies on old KVM behavior of %rip being
7928 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7929 	 */
7930 	if (port == 0x7e &&
7931 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7932 		vcpu->arch.complete_userspace_io =
7933 			complete_fast_pio_out_port_0x7e;
7934 		kvm_skip_emulated_instruction(vcpu);
7935 	} else {
7936 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7937 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7938 	}
7939 	return 0;
7940 }
7941 
7942 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7943 {
7944 	unsigned long val;
7945 
7946 	/* We should only ever be called with arch.pio.count equal to 1 */
7947 	BUG_ON(vcpu->arch.pio.count != 1);
7948 
7949 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7950 		vcpu->arch.pio.count = 0;
7951 		return 1;
7952 	}
7953 
7954 	/* For size less than 4 we merge, else we zero extend */
7955 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7956 
7957 	/*
7958 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7959 	 * the copy and tracing
7960 	 */
7961 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7962 	kvm_rax_write(vcpu, val);
7963 
7964 	return kvm_skip_emulated_instruction(vcpu);
7965 }
7966 
7967 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7968 			   unsigned short port)
7969 {
7970 	unsigned long val;
7971 	int ret;
7972 
7973 	/* For size less than 4 we merge, else we zero extend */
7974 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7975 
7976 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7977 	if (ret) {
7978 		kvm_rax_write(vcpu, val);
7979 		return ret;
7980 	}
7981 
7982 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7983 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7984 
7985 	return 0;
7986 }
7987 
7988 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7989 {
7990 	int ret;
7991 
7992 	if (in)
7993 		ret = kvm_fast_pio_in(vcpu, size, port);
7994 	else
7995 		ret = kvm_fast_pio_out(vcpu, size, port);
7996 	return ret && kvm_skip_emulated_instruction(vcpu);
7997 }
7998 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7999 
8000 static int kvmclock_cpu_down_prep(unsigned int cpu)
8001 {
8002 	__this_cpu_write(cpu_tsc_khz, 0);
8003 	return 0;
8004 }
8005 
8006 static void tsc_khz_changed(void *data)
8007 {
8008 	struct cpufreq_freqs *freq = data;
8009 	unsigned long khz = 0;
8010 
8011 	if (data)
8012 		khz = freq->new;
8013 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8014 		khz = cpufreq_quick_get(raw_smp_processor_id());
8015 	if (!khz)
8016 		khz = tsc_khz;
8017 	__this_cpu_write(cpu_tsc_khz, khz);
8018 }
8019 
8020 #ifdef CONFIG_X86_64
8021 static void kvm_hyperv_tsc_notifier(void)
8022 {
8023 	struct kvm *kvm;
8024 	struct kvm_vcpu *vcpu;
8025 	int cpu;
8026 	unsigned long flags;
8027 
8028 	mutex_lock(&kvm_lock);
8029 	list_for_each_entry(kvm, &vm_list, vm_list)
8030 		kvm_make_mclock_inprogress_request(kvm);
8031 
8032 	hyperv_stop_tsc_emulation();
8033 
8034 	/* TSC frequency always matches when on Hyper-V */
8035 	for_each_present_cpu(cpu)
8036 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8037 	kvm_max_guest_tsc_khz = tsc_khz;
8038 
8039 	list_for_each_entry(kvm, &vm_list, vm_list) {
8040 		struct kvm_arch *ka = &kvm->arch;
8041 
8042 		raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8043 		pvclock_update_vm_gtod_copy(kvm);
8044 		raw_spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8045 
8046 		kvm_for_each_vcpu(cpu, vcpu, kvm)
8047 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8048 
8049 		kvm_for_each_vcpu(cpu, vcpu, kvm)
8050 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8051 	}
8052 	mutex_unlock(&kvm_lock);
8053 }
8054 #endif
8055 
8056 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8057 {
8058 	struct kvm *kvm;
8059 	struct kvm_vcpu *vcpu;
8060 	int i, send_ipi = 0;
8061 
8062 	/*
8063 	 * We allow guests to temporarily run on slowing clocks,
8064 	 * provided we notify them after, or to run on accelerating
8065 	 * clocks, provided we notify them before.  Thus time never
8066 	 * goes backwards.
8067 	 *
8068 	 * However, we have a problem.  We can't atomically update
8069 	 * the frequency of a given CPU from this function; it is
8070 	 * merely a notifier, which can be called from any CPU.
8071 	 * Changing the TSC frequency at arbitrary points in time
8072 	 * requires a recomputation of local variables related to
8073 	 * the TSC for each VCPU.  We must flag these local variables
8074 	 * to be updated and be sure the update takes place with the
8075 	 * new frequency before any guests proceed.
8076 	 *
8077 	 * Unfortunately, the combination of hotplug CPU and frequency
8078 	 * change creates an intractable locking scenario; the order
8079 	 * of when these callouts happen is undefined with respect to
8080 	 * CPU hotplug, and they can race with each other.  As such,
8081 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8082 	 * undefined; you can actually have a CPU frequency change take
8083 	 * place in between the computation of X and the setting of the
8084 	 * variable.  To protect against this problem, all updates of
8085 	 * the per_cpu tsc_khz variable are done in an interrupt
8086 	 * protected IPI, and all callers wishing to update the value
8087 	 * must wait for a synchronous IPI to complete (which is trivial
8088 	 * if the caller is on the CPU already).  This establishes the
8089 	 * necessary total order on variable updates.
8090 	 *
8091 	 * Note that because a guest time update may take place
8092 	 * anytime after the setting of the VCPU's request bit, the
8093 	 * correct TSC value must be set before the request.  However,
8094 	 * to ensure the update actually makes it to any guest which
8095 	 * starts running in hardware virtualization between the set
8096 	 * and the acquisition of the spinlock, we must also ping the
8097 	 * CPU after setting the request bit.
8098 	 *
8099 	 */
8100 
8101 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8102 
8103 	mutex_lock(&kvm_lock);
8104 	list_for_each_entry(kvm, &vm_list, vm_list) {
8105 		kvm_for_each_vcpu(i, vcpu, kvm) {
8106 			if (vcpu->cpu != cpu)
8107 				continue;
8108 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8109 			if (vcpu->cpu != raw_smp_processor_id())
8110 				send_ipi = 1;
8111 		}
8112 	}
8113 	mutex_unlock(&kvm_lock);
8114 
8115 	if (freq->old < freq->new && send_ipi) {
8116 		/*
8117 		 * We upscale the frequency.  Must make the guest
8118 		 * doesn't see old kvmclock values while running with
8119 		 * the new frequency, otherwise we risk the guest sees
8120 		 * time go backwards.
8121 		 *
8122 		 * In case we update the frequency for another cpu
8123 		 * (which might be in guest context) send an interrupt
8124 		 * to kick the cpu out of guest context.  Next time
8125 		 * guest context is entered kvmclock will be updated,
8126 		 * so the guest will not see stale values.
8127 		 */
8128 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8129 	}
8130 }
8131 
8132 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8133 				     void *data)
8134 {
8135 	struct cpufreq_freqs *freq = data;
8136 	int cpu;
8137 
8138 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8139 		return 0;
8140 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8141 		return 0;
8142 
8143 	for_each_cpu(cpu, freq->policy->cpus)
8144 		__kvmclock_cpufreq_notifier(freq, cpu);
8145 
8146 	return 0;
8147 }
8148 
8149 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8150 	.notifier_call  = kvmclock_cpufreq_notifier
8151 };
8152 
8153 static int kvmclock_cpu_online(unsigned int cpu)
8154 {
8155 	tsc_khz_changed(NULL);
8156 	return 0;
8157 }
8158 
8159 static void kvm_timer_init(void)
8160 {
8161 	max_tsc_khz = tsc_khz;
8162 
8163 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8164 #ifdef CONFIG_CPU_FREQ
8165 		struct cpufreq_policy *policy;
8166 		int cpu;
8167 
8168 		cpu = get_cpu();
8169 		policy = cpufreq_cpu_get(cpu);
8170 		if (policy) {
8171 			if (policy->cpuinfo.max_freq)
8172 				max_tsc_khz = policy->cpuinfo.max_freq;
8173 			cpufreq_cpu_put(policy);
8174 		}
8175 		put_cpu();
8176 #endif
8177 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8178 					  CPUFREQ_TRANSITION_NOTIFIER);
8179 	}
8180 
8181 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8182 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8183 }
8184 
8185 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8186 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8187 
8188 int kvm_is_in_guest(void)
8189 {
8190 	return __this_cpu_read(current_vcpu) != NULL;
8191 }
8192 
8193 static int kvm_is_user_mode(void)
8194 {
8195 	int user_mode = 3;
8196 
8197 	if (__this_cpu_read(current_vcpu))
8198 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8199 
8200 	return user_mode != 0;
8201 }
8202 
8203 static unsigned long kvm_get_guest_ip(void)
8204 {
8205 	unsigned long ip = 0;
8206 
8207 	if (__this_cpu_read(current_vcpu))
8208 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8209 
8210 	return ip;
8211 }
8212 
8213 static void kvm_handle_intel_pt_intr(void)
8214 {
8215 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8216 
8217 	kvm_make_request(KVM_REQ_PMI, vcpu);
8218 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8219 			(unsigned long *)&vcpu->arch.pmu.global_status);
8220 }
8221 
8222 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8223 	.is_in_guest		= kvm_is_in_guest,
8224 	.is_user_mode		= kvm_is_user_mode,
8225 	.get_guest_ip		= kvm_get_guest_ip,
8226 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8227 };
8228 
8229 #ifdef CONFIG_X86_64
8230 static void pvclock_gtod_update_fn(struct work_struct *work)
8231 {
8232 	struct kvm *kvm;
8233 
8234 	struct kvm_vcpu *vcpu;
8235 	int i;
8236 
8237 	mutex_lock(&kvm_lock);
8238 	list_for_each_entry(kvm, &vm_list, vm_list)
8239 		kvm_for_each_vcpu(i, vcpu, kvm)
8240 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8241 	atomic_set(&kvm_guest_has_master_clock, 0);
8242 	mutex_unlock(&kvm_lock);
8243 }
8244 
8245 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8246 
8247 /*
8248  * Indirection to move queue_work() out of the tk_core.seq write held
8249  * region to prevent possible deadlocks against time accessors which
8250  * are invoked with work related locks held.
8251  */
8252 static void pvclock_irq_work_fn(struct irq_work *w)
8253 {
8254 	queue_work(system_long_wq, &pvclock_gtod_work);
8255 }
8256 
8257 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8258 
8259 /*
8260  * Notification about pvclock gtod data update.
8261  */
8262 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8263 			       void *priv)
8264 {
8265 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8266 	struct timekeeper *tk = priv;
8267 
8268 	update_pvclock_gtod(tk);
8269 
8270 	/*
8271 	 * Disable master clock if host does not trust, or does not use,
8272 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8273 	 * this is invoked with tk_core.seq write held.
8274 	 */
8275 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8276 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8277 		irq_work_queue(&pvclock_irq_work);
8278 	return 0;
8279 }
8280 
8281 static struct notifier_block pvclock_gtod_notifier = {
8282 	.notifier_call = pvclock_gtod_notify,
8283 };
8284 #endif
8285 
8286 int kvm_arch_init(void *opaque)
8287 {
8288 	struct kvm_x86_init_ops *ops = opaque;
8289 	int r;
8290 
8291 	if (kvm_x86_ops.hardware_enable) {
8292 		printk(KERN_ERR "kvm: already loaded the other module\n");
8293 		r = -EEXIST;
8294 		goto out;
8295 	}
8296 
8297 	if (!ops->cpu_has_kvm_support()) {
8298 		pr_err_ratelimited("kvm: no hardware support\n");
8299 		r = -EOPNOTSUPP;
8300 		goto out;
8301 	}
8302 	if (ops->disabled_by_bios()) {
8303 		pr_err_ratelimited("kvm: disabled by bios\n");
8304 		r = -EOPNOTSUPP;
8305 		goto out;
8306 	}
8307 
8308 	/*
8309 	 * KVM explicitly assumes that the guest has an FPU and
8310 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8311 	 * vCPU's FPU state as a fxregs_state struct.
8312 	 */
8313 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8314 		printk(KERN_ERR "kvm: inadequate fpu\n");
8315 		r = -EOPNOTSUPP;
8316 		goto out;
8317 	}
8318 
8319 	r = -ENOMEM;
8320 
8321 	x86_emulator_cache = kvm_alloc_emulator_cache();
8322 	if (!x86_emulator_cache) {
8323 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8324 		goto out;
8325 	}
8326 
8327 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8328 	if (!user_return_msrs) {
8329 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8330 		goto out_free_x86_emulator_cache;
8331 	}
8332 	kvm_nr_uret_msrs = 0;
8333 
8334 	r = kvm_mmu_module_init();
8335 	if (r)
8336 		goto out_free_percpu;
8337 
8338 	kvm_timer_init();
8339 
8340 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8341 
8342 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8343 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8344 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8345 	}
8346 
8347 	if (pi_inject_timer == -1)
8348 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8349 #ifdef CONFIG_X86_64
8350 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8351 
8352 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8353 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8354 #endif
8355 
8356 	return 0;
8357 
8358 out_free_percpu:
8359 	free_percpu(user_return_msrs);
8360 out_free_x86_emulator_cache:
8361 	kmem_cache_destroy(x86_emulator_cache);
8362 out:
8363 	return r;
8364 }
8365 
8366 void kvm_arch_exit(void)
8367 {
8368 #ifdef CONFIG_X86_64
8369 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8370 		clear_hv_tscchange_cb();
8371 #endif
8372 	kvm_lapic_exit();
8373 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8374 
8375 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8376 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8377 					    CPUFREQ_TRANSITION_NOTIFIER);
8378 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8379 #ifdef CONFIG_X86_64
8380 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8381 	irq_work_sync(&pvclock_irq_work);
8382 	cancel_work_sync(&pvclock_gtod_work);
8383 #endif
8384 	kvm_x86_ops.hardware_enable = NULL;
8385 	kvm_mmu_module_exit();
8386 	free_percpu(user_return_msrs);
8387 	kmem_cache_destroy(x86_emulator_cache);
8388 #ifdef CONFIG_KVM_XEN
8389 	static_key_deferred_flush(&kvm_xen_enabled);
8390 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8391 #endif
8392 }
8393 
8394 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8395 {
8396 	++vcpu->stat.halt_exits;
8397 	if (lapic_in_kernel(vcpu)) {
8398 		vcpu->arch.mp_state = state;
8399 		return 1;
8400 	} else {
8401 		vcpu->run->exit_reason = reason;
8402 		return 0;
8403 	}
8404 }
8405 
8406 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8407 {
8408 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8409 }
8410 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8411 
8412 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8413 {
8414 	int ret = kvm_skip_emulated_instruction(vcpu);
8415 	/*
8416 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8417 	 * KVM_EXIT_DEBUG here.
8418 	 */
8419 	return kvm_vcpu_halt(vcpu) && ret;
8420 }
8421 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8422 
8423 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8424 {
8425 	int ret = kvm_skip_emulated_instruction(vcpu);
8426 
8427 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8428 }
8429 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8430 
8431 #ifdef CONFIG_X86_64
8432 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8433 			        unsigned long clock_type)
8434 {
8435 	struct kvm_clock_pairing clock_pairing;
8436 	struct timespec64 ts;
8437 	u64 cycle;
8438 	int ret;
8439 
8440 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8441 		return -KVM_EOPNOTSUPP;
8442 
8443 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8444 		return -KVM_EOPNOTSUPP;
8445 
8446 	clock_pairing.sec = ts.tv_sec;
8447 	clock_pairing.nsec = ts.tv_nsec;
8448 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8449 	clock_pairing.flags = 0;
8450 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8451 
8452 	ret = 0;
8453 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8454 			    sizeof(struct kvm_clock_pairing)))
8455 		ret = -KVM_EFAULT;
8456 
8457 	return ret;
8458 }
8459 #endif
8460 
8461 /*
8462  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8463  *
8464  * @apicid - apicid of vcpu to be kicked.
8465  */
8466 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8467 {
8468 	struct kvm_lapic_irq lapic_irq;
8469 
8470 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8471 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8472 	lapic_irq.level = 0;
8473 	lapic_irq.dest_id = apicid;
8474 	lapic_irq.msi_redir_hint = false;
8475 
8476 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8477 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8478 }
8479 
8480 bool kvm_apicv_activated(struct kvm *kvm)
8481 {
8482 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8483 }
8484 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8485 
8486 static void kvm_apicv_init(struct kvm *kvm)
8487 {
8488 	mutex_init(&kvm->arch.apicv_update_lock);
8489 
8490 	if (enable_apicv)
8491 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8492 			  &kvm->arch.apicv_inhibit_reasons);
8493 	else
8494 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8495 			&kvm->arch.apicv_inhibit_reasons);
8496 }
8497 
8498 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8499 {
8500 	struct kvm_vcpu *target = NULL;
8501 	struct kvm_apic_map *map;
8502 
8503 	vcpu->stat.directed_yield_attempted++;
8504 
8505 	if (single_task_running())
8506 		goto no_yield;
8507 
8508 	rcu_read_lock();
8509 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8510 
8511 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8512 		target = map->phys_map[dest_id]->vcpu;
8513 
8514 	rcu_read_unlock();
8515 
8516 	if (!target || !READ_ONCE(target->ready))
8517 		goto no_yield;
8518 
8519 	/* Ignore requests to yield to self */
8520 	if (vcpu == target)
8521 		goto no_yield;
8522 
8523 	if (kvm_vcpu_yield_to(target) <= 0)
8524 		goto no_yield;
8525 
8526 	vcpu->stat.directed_yield_successful++;
8527 
8528 no_yield:
8529 	return;
8530 }
8531 
8532 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8533 {
8534 	u64 ret = vcpu->run->hypercall.ret;
8535 
8536 	if (!is_64_bit_mode(vcpu))
8537 		ret = (u32)ret;
8538 	kvm_rax_write(vcpu, ret);
8539 	++vcpu->stat.hypercalls;
8540 	return kvm_skip_emulated_instruction(vcpu);
8541 }
8542 
8543 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8544 {
8545 	unsigned long nr, a0, a1, a2, a3, ret;
8546 	int op_64_bit;
8547 
8548 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8549 		return kvm_xen_hypercall(vcpu);
8550 
8551 	if (kvm_hv_hypercall_enabled(vcpu))
8552 		return kvm_hv_hypercall(vcpu);
8553 
8554 	nr = kvm_rax_read(vcpu);
8555 	a0 = kvm_rbx_read(vcpu);
8556 	a1 = kvm_rcx_read(vcpu);
8557 	a2 = kvm_rdx_read(vcpu);
8558 	a3 = kvm_rsi_read(vcpu);
8559 
8560 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8561 
8562 	op_64_bit = is_64_bit_mode(vcpu);
8563 	if (!op_64_bit) {
8564 		nr &= 0xFFFFFFFF;
8565 		a0 &= 0xFFFFFFFF;
8566 		a1 &= 0xFFFFFFFF;
8567 		a2 &= 0xFFFFFFFF;
8568 		a3 &= 0xFFFFFFFF;
8569 	}
8570 
8571 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8572 		ret = -KVM_EPERM;
8573 		goto out;
8574 	}
8575 
8576 	ret = -KVM_ENOSYS;
8577 
8578 	switch (nr) {
8579 	case KVM_HC_VAPIC_POLL_IRQ:
8580 		ret = 0;
8581 		break;
8582 	case KVM_HC_KICK_CPU:
8583 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8584 			break;
8585 
8586 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8587 		kvm_sched_yield(vcpu, a1);
8588 		ret = 0;
8589 		break;
8590 #ifdef CONFIG_X86_64
8591 	case KVM_HC_CLOCK_PAIRING:
8592 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8593 		break;
8594 #endif
8595 	case KVM_HC_SEND_IPI:
8596 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8597 			break;
8598 
8599 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8600 		break;
8601 	case KVM_HC_SCHED_YIELD:
8602 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8603 			break;
8604 
8605 		kvm_sched_yield(vcpu, a0);
8606 		ret = 0;
8607 		break;
8608 	case KVM_HC_MAP_GPA_RANGE: {
8609 		u64 gpa = a0, npages = a1, attrs = a2;
8610 
8611 		ret = -KVM_ENOSYS;
8612 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8613 			break;
8614 
8615 		if (!PAGE_ALIGNED(gpa) || !npages ||
8616 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8617 			ret = -KVM_EINVAL;
8618 			break;
8619 		}
8620 
8621 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
8622 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
8623 		vcpu->run->hypercall.args[0]  = gpa;
8624 		vcpu->run->hypercall.args[1]  = npages;
8625 		vcpu->run->hypercall.args[2]  = attrs;
8626 		vcpu->run->hypercall.longmode = op_64_bit;
8627 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8628 		return 0;
8629 	}
8630 	default:
8631 		ret = -KVM_ENOSYS;
8632 		break;
8633 	}
8634 out:
8635 	if (!op_64_bit)
8636 		ret = (u32)ret;
8637 	kvm_rax_write(vcpu, ret);
8638 
8639 	++vcpu->stat.hypercalls;
8640 	return kvm_skip_emulated_instruction(vcpu);
8641 }
8642 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8643 
8644 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8645 {
8646 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8647 	char instruction[3];
8648 	unsigned long rip = kvm_rip_read(vcpu);
8649 
8650 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8651 
8652 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8653 		&ctxt->exception);
8654 }
8655 
8656 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8657 {
8658 	return vcpu->run->request_interrupt_window &&
8659 		likely(!pic_in_kernel(vcpu->kvm));
8660 }
8661 
8662 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8663 {
8664 	struct kvm_run *kvm_run = vcpu->run;
8665 
8666 	/*
8667 	 * if_flag is obsolete and useless, so do not bother
8668 	 * setting it for SEV-ES guests.  Userspace can just
8669 	 * use kvm_run->ready_for_interrupt_injection.
8670 	 */
8671 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8672 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8673 
8674 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8675 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8676 
8677 	/*
8678 	 * The call to kvm_ready_for_interrupt_injection() may end up in
8679 	 * kvm_xen_has_interrupt() which may require the srcu lock to be
8680 	 * held, to protect against changes in the vcpu_info address.
8681 	 */
8682 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8683 	kvm_run->ready_for_interrupt_injection =
8684 		pic_in_kernel(vcpu->kvm) ||
8685 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8686 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8687 
8688 	if (is_smm(vcpu))
8689 		kvm_run->flags |= KVM_RUN_X86_SMM;
8690 }
8691 
8692 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8693 {
8694 	int max_irr, tpr;
8695 
8696 	if (!kvm_x86_ops.update_cr8_intercept)
8697 		return;
8698 
8699 	if (!lapic_in_kernel(vcpu))
8700 		return;
8701 
8702 	if (vcpu->arch.apicv_active)
8703 		return;
8704 
8705 	if (!vcpu->arch.apic->vapic_addr)
8706 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8707 	else
8708 		max_irr = -1;
8709 
8710 	if (max_irr != -1)
8711 		max_irr >>= 4;
8712 
8713 	tpr = kvm_lapic_get_cr8(vcpu);
8714 
8715 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8716 }
8717 
8718 
8719 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8720 {
8721 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8722 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
8723 		return 1;
8724 	}
8725 
8726 	return kvm_x86_ops.nested_ops->check_events(vcpu);
8727 }
8728 
8729 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8730 {
8731 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8732 		vcpu->arch.exception.error_code = false;
8733 	static_call(kvm_x86_queue_exception)(vcpu);
8734 }
8735 
8736 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8737 {
8738 	int r;
8739 	bool can_inject = true;
8740 
8741 	/* try to reinject previous events if any */
8742 
8743 	if (vcpu->arch.exception.injected) {
8744 		kvm_inject_exception(vcpu);
8745 		can_inject = false;
8746 	}
8747 	/*
8748 	 * Do not inject an NMI or interrupt if there is a pending
8749 	 * exception.  Exceptions and interrupts are recognized at
8750 	 * instruction boundaries, i.e. the start of an instruction.
8751 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8752 	 * NMIs and interrupts, i.e. traps are recognized before an
8753 	 * NMI/interrupt that's pending on the same instruction.
8754 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8755 	 * priority, but are only generated (pended) during instruction
8756 	 * execution, i.e. a pending fault-like exception means the
8757 	 * fault occurred on the *previous* instruction and must be
8758 	 * serviced prior to recognizing any new events in order to
8759 	 * fully complete the previous instruction.
8760 	 */
8761 	else if (!vcpu->arch.exception.pending) {
8762 		if (vcpu->arch.nmi_injected) {
8763 			static_call(kvm_x86_set_nmi)(vcpu);
8764 			can_inject = false;
8765 		} else if (vcpu->arch.interrupt.injected) {
8766 			static_call(kvm_x86_set_irq)(vcpu);
8767 			can_inject = false;
8768 		}
8769 	}
8770 
8771 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8772 		     vcpu->arch.exception.pending);
8773 
8774 	/*
8775 	 * Call check_nested_events() even if we reinjected a previous event
8776 	 * in order for caller to determine if it should require immediate-exit
8777 	 * from L2 to L1 due to pending L1 events which require exit
8778 	 * from L2 to L1.
8779 	 */
8780 	if (is_guest_mode(vcpu)) {
8781 		r = kvm_check_nested_events(vcpu);
8782 		if (r < 0)
8783 			goto out;
8784 	}
8785 
8786 	/* try to inject new event if pending */
8787 	if (vcpu->arch.exception.pending) {
8788 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8789 					vcpu->arch.exception.has_error_code,
8790 					vcpu->arch.exception.error_code);
8791 
8792 		vcpu->arch.exception.pending = false;
8793 		vcpu->arch.exception.injected = true;
8794 
8795 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8796 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8797 					     X86_EFLAGS_RF);
8798 
8799 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8800 			kvm_deliver_exception_payload(vcpu);
8801 			if (vcpu->arch.dr7 & DR7_GD) {
8802 				vcpu->arch.dr7 &= ~DR7_GD;
8803 				kvm_update_dr7(vcpu);
8804 			}
8805 		}
8806 
8807 		kvm_inject_exception(vcpu);
8808 		can_inject = false;
8809 	}
8810 
8811 	/* Don't inject interrupts if the user asked to avoid doing so */
8812 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
8813 		return 0;
8814 
8815 	/*
8816 	 * Finally, inject interrupt events.  If an event cannot be injected
8817 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8818 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8819 	 * and can architecturally be injected, but we cannot do it right now:
8820 	 * an interrupt could have arrived just now and we have to inject it
8821 	 * as a vmexit, or there could already an event in the queue, which is
8822 	 * indicated by can_inject.  In that case we request an immediate exit
8823 	 * in order to make progress and get back here for another iteration.
8824 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8825 	 */
8826 	if (vcpu->arch.smi_pending) {
8827 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8828 		if (r < 0)
8829 			goto out;
8830 		if (r) {
8831 			vcpu->arch.smi_pending = false;
8832 			++vcpu->arch.smi_count;
8833 			enter_smm(vcpu);
8834 			can_inject = false;
8835 		} else
8836 			static_call(kvm_x86_enable_smi_window)(vcpu);
8837 	}
8838 
8839 	if (vcpu->arch.nmi_pending) {
8840 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8841 		if (r < 0)
8842 			goto out;
8843 		if (r) {
8844 			--vcpu->arch.nmi_pending;
8845 			vcpu->arch.nmi_injected = true;
8846 			static_call(kvm_x86_set_nmi)(vcpu);
8847 			can_inject = false;
8848 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8849 		}
8850 		if (vcpu->arch.nmi_pending)
8851 			static_call(kvm_x86_enable_nmi_window)(vcpu);
8852 	}
8853 
8854 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8855 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8856 		if (r < 0)
8857 			goto out;
8858 		if (r) {
8859 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8860 			static_call(kvm_x86_set_irq)(vcpu);
8861 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8862 		}
8863 		if (kvm_cpu_has_injectable_intr(vcpu))
8864 			static_call(kvm_x86_enable_irq_window)(vcpu);
8865 	}
8866 
8867 	if (is_guest_mode(vcpu) &&
8868 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8869 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8870 		*req_immediate_exit = true;
8871 
8872 	WARN_ON(vcpu->arch.exception.pending);
8873 	return 0;
8874 
8875 out:
8876 	if (r == -EBUSY) {
8877 		*req_immediate_exit = true;
8878 		r = 0;
8879 	}
8880 	return r;
8881 }
8882 
8883 static void process_nmi(struct kvm_vcpu *vcpu)
8884 {
8885 	unsigned limit = 2;
8886 
8887 	/*
8888 	 * x86 is limited to one NMI running, and one NMI pending after it.
8889 	 * If an NMI is already in progress, limit further NMIs to just one.
8890 	 * Otherwise, allow two (and we'll inject the first one immediately).
8891 	 */
8892 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8893 		limit = 1;
8894 
8895 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8896 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8897 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8898 }
8899 
8900 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8901 {
8902 	u32 flags = 0;
8903 	flags |= seg->g       << 23;
8904 	flags |= seg->db      << 22;
8905 	flags |= seg->l       << 21;
8906 	flags |= seg->avl     << 20;
8907 	flags |= seg->present << 15;
8908 	flags |= seg->dpl     << 13;
8909 	flags |= seg->s       << 12;
8910 	flags |= seg->type    << 8;
8911 	return flags;
8912 }
8913 
8914 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8915 {
8916 	struct kvm_segment seg;
8917 	int offset;
8918 
8919 	kvm_get_segment(vcpu, &seg, n);
8920 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8921 
8922 	if (n < 3)
8923 		offset = 0x7f84 + n * 12;
8924 	else
8925 		offset = 0x7f2c + (n - 3) * 12;
8926 
8927 	put_smstate(u32, buf, offset + 8, seg.base);
8928 	put_smstate(u32, buf, offset + 4, seg.limit);
8929 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8930 }
8931 
8932 #ifdef CONFIG_X86_64
8933 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8934 {
8935 	struct kvm_segment seg;
8936 	int offset;
8937 	u16 flags;
8938 
8939 	kvm_get_segment(vcpu, &seg, n);
8940 	offset = 0x7e00 + n * 16;
8941 
8942 	flags = enter_smm_get_segment_flags(&seg) >> 8;
8943 	put_smstate(u16, buf, offset, seg.selector);
8944 	put_smstate(u16, buf, offset + 2, flags);
8945 	put_smstate(u32, buf, offset + 4, seg.limit);
8946 	put_smstate(u64, buf, offset + 8, seg.base);
8947 }
8948 #endif
8949 
8950 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8951 {
8952 	struct desc_ptr dt;
8953 	struct kvm_segment seg;
8954 	unsigned long val;
8955 	int i;
8956 
8957 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8958 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8959 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8960 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8961 
8962 	for (i = 0; i < 8; i++)
8963 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8964 
8965 	kvm_get_dr(vcpu, 6, &val);
8966 	put_smstate(u32, buf, 0x7fcc, (u32)val);
8967 	kvm_get_dr(vcpu, 7, &val);
8968 	put_smstate(u32, buf, 0x7fc8, (u32)val);
8969 
8970 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8971 	put_smstate(u32, buf, 0x7fc4, seg.selector);
8972 	put_smstate(u32, buf, 0x7f64, seg.base);
8973 	put_smstate(u32, buf, 0x7f60, seg.limit);
8974 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8975 
8976 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8977 	put_smstate(u32, buf, 0x7fc0, seg.selector);
8978 	put_smstate(u32, buf, 0x7f80, seg.base);
8979 	put_smstate(u32, buf, 0x7f7c, seg.limit);
8980 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8981 
8982 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
8983 	put_smstate(u32, buf, 0x7f74, dt.address);
8984 	put_smstate(u32, buf, 0x7f70, dt.size);
8985 
8986 	static_call(kvm_x86_get_idt)(vcpu, &dt);
8987 	put_smstate(u32, buf, 0x7f58, dt.address);
8988 	put_smstate(u32, buf, 0x7f54, dt.size);
8989 
8990 	for (i = 0; i < 6; i++)
8991 		enter_smm_save_seg_32(vcpu, buf, i);
8992 
8993 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8994 
8995 	/* revision id */
8996 	put_smstate(u32, buf, 0x7efc, 0x00020000);
8997 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8998 }
8999 
9000 #ifdef CONFIG_X86_64
9001 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9002 {
9003 	struct desc_ptr dt;
9004 	struct kvm_segment seg;
9005 	unsigned long val;
9006 	int i;
9007 
9008 	for (i = 0; i < 16; i++)
9009 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9010 
9011 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9012 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9013 
9014 	kvm_get_dr(vcpu, 6, &val);
9015 	put_smstate(u64, buf, 0x7f68, val);
9016 	kvm_get_dr(vcpu, 7, &val);
9017 	put_smstate(u64, buf, 0x7f60, val);
9018 
9019 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9020 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9021 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9022 
9023 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9024 
9025 	/* revision id */
9026 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9027 
9028 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9029 
9030 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9031 	put_smstate(u16, buf, 0x7e90, seg.selector);
9032 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9033 	put_smstate(u32, buf, 0x7e94, seg.limit);
9034 	put_smstate(u64, buf, 0x7e98, seg.base);
9035 
9036 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9037 	put_smstate(u32, buf, 0x7e84, dt.size);
9038 	put_smstate(u64, buf, 0x7e88, dt.address);
9039 
9040 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9041 	put_smstate(u16, buf, 0x7e70, seg.selector);
9042 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9043 	put_smstate(u32, buf, 0x7e74, seg.limit);
9044 	put_smstate(u64, buf, 0x7e78, seg.base);
9045 
9046 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9047 	put_smstate(u32, buf, 0x7e64, dt.size);
9048 	put_smstate(u64, buf, 0x7e68, dt.address);
9049 
9050 	for (i = 0; i < 6; i++)
9051 		enter_smm_save_seg_64(vcpu, buf, i);
9052 }
9053 #endif
9054 
9055 static void enter_smm(struct kvm_vcpu *vcpu)
9056 {
9057 	struct kvm_segment cs, ds;
9058 	struct desc_ptr dt;
9059 	unsigned long cr0;
9060 	char buf[512];
9061 
9062 	memset(buf, 0, 512);
9063 #ifdef CONFIG_X86_64
9064 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9065 		enter_smm_save_state_64(vcpu, buf);
9066 	else
9067 #endif
9068 		enter_smm_save_state_32(vcpu, buf);
9069 
9070 	/*
9071 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9072 	 * state (e.g. leave guest mode) after we've saved the state into the
9073 	 * SMM state-save area.
9074 	 */
9075 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9076 
9077 	kvm_smm_changed(vcpu, true);
9078 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9079 
9080 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9081 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9082 	else
9083 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9084 
9085 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9086 	kvm_rip_write(vcpu, 0x8000);
9087 
9088 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9089 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9090 	vcpu->arch.cr0 = cr0;
9091 
9092 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9093 
9094 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9095 	dt.address = dt.size = 0;
9096 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9097 
9098 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9099 
9100 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9101 	cs.base = vcpu->arch.smbase;
9102 
9103 	ds.selector = 0;
9104 	ds.base = 0;
9105 
9106 	cs.limit    = ds.limit = 0xffffffff;
9107 	cs.type     = ds.type = 0x3;
9108 	cs.dpl      = ds.dpl = 0;
9109 	cs.db       = ds.db = 0;
9110 	cs.s        = ds.s = 1;
9111 	cs.l        = ds.l = 0;
9112 	cs.g        = ds.g = 1;
9113 	cs.avl      = ds.avl = 0;
9114 	cs.present  = ds.present = 1;
9115 	cs.unusable = ds.unusable = 0;
9116 	cs.padding  = ds.padding = 0;
9117 
9118 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9119 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9120 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9121 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9122 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9123 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9124 
9125 #ifdef CONFIG_X86_64
9126 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9127 		static_call(kvm_x86_set_efer)(vcpu, 0);
9128 #endif
9129 
9130 	kvm_update_cpuid_runtime(vcpu);
9131 	kvm_mmu_reset_context(vcpu);
9132 }
9133 
9134 static void process_smi(struct kvm_vcpu *vcpu)
9135 {
9136 	vcpu->arch.smi_pending = true;
9137 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9138 }
9139 
9140 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9141 				       unsigned long *vcpu_bitmap)
9142 {
9143 	cpumask_var_t cpus;
9144 
9145 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9146 
9147 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9148 				    NULL, vcpu_bitmap, cpus);
9149 
9150 	free_cpumask_var(cpus);
9151 }
9152 
9153 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9154 {
9155 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9156 }
9157 
9158 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9159 {
9160 	bool activate;
9161 
9162 	if (!lapic_in_kernel(vcpu))
9163 		return;
9164 
9165 	mutex_lock(&vcpu->kvm->arch.apicv_update_lock);
9166 
9167 	activate = kvm_apicv_activated(vcpu->kvm);
9168 	if (vcpu->arch.apicv_active == activate)
9169 		goto out;
9170 
9171 	vcpu->arch.apicv_active = activate;
9172 	kvm_apic_update_apicv(vcpu);
9173 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9174 
9175 	/*
9176 	 * When APICv gets disabled, we may still have injected interrupts
9177 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9178 	 * still active when the interrupt got accepted. Make sure
9179 	 * inject_pending_event() is called to check for that.
9180 	 */
9181 	if (!vcpu->arch.apicv_active)
9182 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9183 
9184 out:
9185 	mutex_unlock(&vcpu->kvm->arch.apicv_update_lock);
9186 }
9187 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9188 
9189 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9190 {
9191 	unsigned long old, new;
9192 
9193 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9194 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9195 		return;
9196 
9197 	old = new = kvm->arch.apicv_inhibit_reasons;
9198 
9199 	if (activate)
9200 		__clear_bit(bit, &new);
9201 	else
9202 		__set_bit(bit, &new);
9203 
9204 	if (!!old != !!new) {
9205 		trace_kvm_apicv_update_request(activate, bit);
9206 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9207 		kvm->arch.apicv_inhibit_reasons = new;
9208 		if (new) {
9209 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9210 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
9211 		}
9212 	} else
9213 		kvm->arch.apicv_inhibit_reasons = new;
9214 }
9215 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9216 
9217 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9218 {
9219 	mutex_lock(&kvm->arch.apicv_update_lock);
9220 	__kvm_request_apicv_update(kvm, activate, bit);
9221 	mutex_unlock(&kvm->arch.apicv_update_lock);
9222 }
9223 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9224 
9225 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9226 {
9227 	if (!kvm_apic_present(vcpu))
9228 		return;
9229 
9230 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9231 
9232 	if (irqchip_split(vcpu->kvm))
9233 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9234 	else {
9235 		if (vcpu->arch.apicv_active)
9236 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9237 		if (ioapic_in_kernel(vcpu->kvm))
9238 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9239 	}
9240 
9241 	if (is_guest_mode(vcpu))
9242 		vcpu->arch.load_eoi_exitmap_pending = true;
9243 	else
9244 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9245 }
9246 
9247 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9248 {
9249 	u64 eoi_exit_bitmap[4];
9250 
9251 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9252 		return;
9253 
9254 	if (to_hv_vcpu(vcpu))
9255 		bitmap_or((ulong *)eoi_exit_bitmap,
9256 			  vcpu->arch.ioapic_handled_vectors,
9257 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9258 
9259 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9260 }
9261 
9262 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9263 					    unsigned long start, unsigned long end)
9264 {
9265 	unsigned long apic_address;
9266 
9267 	/*
9268 	 * The physical address of apic access page is stored in the VMCS.
9269 	 * Update it when it becomes invalid.
9270 	 */
9271 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9272 	if (start <= apic_address && apic_address < end)
9273 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9274 }
9275 
9276 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9277 {
9278 	if (!lapic_in_kernel(vcpu))
9279 		return;
9280 
9281 	if (!kvm_x86_ops.set_apic_access_page_addr)
9282 		return;
9283 
9284 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9285 }
9286 
9287 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9288 {
9289 	smp_send_reschedule(vcpu->cpu);
9290 }
9291 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9292 
9293 /*
9294  * Returns 1 to let vcpu_run() continue the guest execution loop without
9295  * exiting to the userspace.  Otherwise, the value will be returned to the
9296  * userspace.
9297  */
9298 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9299 {
9300 	int r;
9301 	bool req_int_win =
9302 		dm_request_for_irq_injection(vcpu) &&
9303 		kvm_cpu_accept_dm_intr(vcpu);
9304 	fastpath_t exit_fastpath;
9305 
9306 	bool req_immediate_exit = false;
9307 
9308 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9309 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9310 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9311 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9312 		trace_kvm_dirty_ring_exit(vcpu);
9313 		r = 0;
9314 		goto out;
9315 	}
9316 
9317 	if (kvm_request_pending(vcpu)) {
9318 		if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9319 			r = -EIO;
9320 			goto out;
9321 		}
9322 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9323 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9324 				r = 0;
9325 				goto out;
9326 			}
9327 		}
9328 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9329 			kvm_mmu_unload(vcpu);
9330 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9331 			__kvm_migrate_timers(vcpu);
9332 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9333 			kvm_gen_update_masterclock(vcpu->kvm);
9334 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9335 			kvm_gen_kvmclock_update(vcpu);
9336 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9337 			r = kvm_guest_time_update(vcpu);
9338 			if (unlikely(r))
9339 				goto out;
9340 		}
9341 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9342 			kvm_mmu_sync_roots(vcpu);
9343 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9344 			kvm_mmu_load_pgd(vcpu);
9345 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9346 			kvm_vcpu_flush_tlb_all(vcpu);
9347 
9348 			/* Flushing all ASIDs flushes the current ASID... */
9349 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9350 		}
9351 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9352 			kvm_vcpu_flush_tlb_current(vcpu);
9353 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9354 			kvm_vcpu_flush_tlb_guest(vcpu);
9355 
9356 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9357 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9358 			r = 0;
9359 			goto out;
9360 		}
9361 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9362 			if (is_guest_mode(vcpu)) {
9363 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9364 			} else {
9365 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9366 				vcpu->mmio_needed = 0;
9367 				r = 0;
9368 				goto out;
9369 			}
9370 		}
9371 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9372 			/* Page is swapped out. Do synthetic halt */
9373 			vcpu->arch.apf.halted = true;
9374 			r = 1;
9375 			goto out;
9376 		}
9377 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9378 			record_steal_time(vcpu);
9379 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9380 			process_smi(vcpu);
9381 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9382 			process_nmi(vcpu);
9383 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9384 			kvm_pmu_handle_event(vcpu);
9385 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9386 			kvm_pmu_deliver_pmi(vcpu);
9387 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9388 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9389 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9390 				     vcpu->arch.ioapic_handled_vectors)) {
9391 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9392 				vcpu->run->eoi.vector =
9393 						vcpu->arch.pending_ioapic_eoi;
9394 				r = 0;
9395 				goto out;
9396 			}
9397 		}
9398 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9399 			vcpu_scan_ioapic(vcpu);
9400 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9401 			vcpu_load_eoi_exitmap(vcpu);
9402 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9403 			kvm_vcpu_reload_apic_access_page(vcpu);
9404 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9405 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9406 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9407 			r = 0;
9408 			goto out;
9409 		}
9410 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9411 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9412 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9413 			r = 0;
9414 			goto out;
9415 		}
9416 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9417 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9418 
9419 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9420 			vcpu->run->hyperv = hv_vcpu->exit;
9421 			r = 0;
9422 			goto out;
9423 		}
9424 
9425 		/*
9426 		 * KVM_REQ_HV_STIMER has to be processed after
9427 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9428 		 * depend on the guest clock being up-to-date
9429 		 */
9430 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9431 			kvm_hv_process_stimers(vcpu);
9432 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9433 			kvm_vcpu_update_apicv(vcpu);
9434 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9435 			kvm_check_async_pf_completion(vcpu);
9436 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9437 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9438 
9439 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9440 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9441 	}
9442 
9443 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9444 	    kvm_xen_has_interrupt(vcpu)) {
9445 		++vcpu->stat.req_event;
9446 		r = kvm_apic_accept_events(vcpu);
9447 		if (r < 0) {
9448 			r = 0;
9449 			goto out;
9450 		}
9451 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9452 			r = 1;
9453 			goto out;
9454 		}
9455 
9456 		r = inject_pending_event(vcpu, &req_immediate_exit);
9457 		if (r < 0) {
9458 			r = 0;
9459 			goto out;
9460 		}
9461 		if (req_int_win)
9462 			static_call(kvm_x86_enable_irq_window)(vcpu);
9463 
9464 		if (kvm_lapic_enabled(vcpu)) {
9465 			update_cr8_intercept(vcpu);
9466 			kvm_lapic_sync_to_vapic(vcpu);
9467 		}
9468 	}
9469 
9470 	r = kvm_mmu_reload(vcpu);
9471 	if (unlikely(r)) {
9472 		goto cancel_injection;
9473 	}
9474 
9475 	preempt_disable();
9476 
9477 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9478 
9479 	/*
9480 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9481 	 * IPI are then delayed after guest entry, which ensures that they
9482 	 * result in virtual interrupt delivery.
9483 	 */
9484 	local_irq_disable();
9485 	vcpu->mode = IN_GUEST_MODE;
9486 
9487 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9488 
9489 	/*
9490 	 * 1) We should set ->mode before checking ->requests.  Please see
9491 	 * the comment in kvm_vcpu_exiting_guest_mode().
9492 	 *
9493 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9494 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9495 	 * (see vmx_deliver_posted_interrupt).
9496 	 *
9497 	 * 3) This also orders the write to mode from any reads to the page
9498 	 * tables done while the VCPU is running.  Please see the comment
9499 	 * in kvm_flush_remote_tlbs.
9500 	 */
9501 	smp_mb__after_srcu_read_unlock();
9502 
9503 	/*
9504 	 * This handles the case where a posted interrupt was
9505 	 * notified with kvm_vcpu_kick.
9506 	 */
9507 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9508 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9509 
9510 	if (kvm_vcpu_exit_request(vcpu)) {
9511 		vcpu->mode = OUTSIDE_GUEST_MODE;
9512 		smp_wmb();
9513 		local_irq_enable();
9514 		preempt_enable();
9515 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9516 		r = 1;
9517 		goto cancel_injection;
9518 	}
9519 
9520 	if (req_immediate_exit) {
9521 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9522 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9523 	}
9524 
9525 	fpregs_assert_state_consistent();
9526 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9527 		switch_fpu_return();
9528 
9529 	if (unlikely(vcpu->arch.switch_db_regs)) {
9530 		set_debugreg(0, 7);
9531 		set_debugreg(vcpu->arch.eff_db[0], 0);
9532 		set_debugreg(vcpu->arch.eff_db[1], 1);
9533 		set_debugreg(vcpu->arch.eff_db[2], 2);
9534 		set_debugreg(vcpu->arch.eff_db[3], 3);
9535 	} else if (unlikely(hw_breakpoint_active())) {
9536 		set_debugreg(0, 7);
9537 	}
9538 
9539 	for (;;) {
9540 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9541 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9542 			break;
9543 
9544 		if (vcpu->arch.apicv_active)
9545 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9546 
9547 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9548 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9549 			break;
9550 		}
9551 	}
9552 
9553 	/*
9554 	 * Do this here before restoring debug registers on the host.  And
9555 	 * since we do this before handling the vmexit, a DR access vmexit
9556 	 * can (a) read the correct value of the debug registers, (b) set
9557 	 * KVM_DEBUGREG_WONT_EXIT again.
9558 	 */
9559 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9560 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9561 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9562 		kvm_update_dr0123(vcpu);
9563 		kvm_update_dr7(vcpu);
9564 	}
9565 
9566 	/*
9567 	 * If the guest has used debug registers, at least dr7
9568 	 * will be disabled while returning to the host.
9569 	 * If we don't have active breakpoints in the host, we don't
9570 	 * care about the messed up debug address registers. But if
9571 	 * we have some of them active, restore the old state.
9572 	 */
9573 	if (hw_breakpoint_active())
9574 		hw_breakpoint_restore();
9575 
9576 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9577 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9578 
9579 	vcpu->mode = OUTSIDE_GUEST_MODE;
9580 	smp_wmb();
9581 
9582 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9583 
9584 	/*
9585 	 * Consume any pending interrupts, including the possible source of
9586 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9587 	 * An instruction is required after local_irq_enable() to fully unblock
9588 	 * interrupts on processors that implement an interrupt shadow, the
9589 	 * stat.exits increment will do nicely.
9590 	 */
9591 	kvm_before_interrupt(vcpu);
9592 	local_irq_enable();
9593 	++vcpu->stat.exits;
9594 	local_irq_disable();
9595 	kvm_after_interrupt(vcpu);
9596 
9597 	/*
9598 	 * Wait until after servicing IRQs to account guest time so that any
9599 	 * ticks that occurred while running the guest are properly accounted
9600 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9601 	 * of accounting via context tracking, but the loss of accuracy is
9602 	 * acceptable for all known use cases.
9603 	 */
9604 	vtime_account_guest_exit();
9605 
9606 	if (lapic_in_kernel(vcpu)) {
9607 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9608 		if (delta != S64_MIN) {
9609 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9610 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9611 		}
9612 	}
9613 
9614 	local_irq_enable();
9615 	preempt_enable();
9616 
9617 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9618 
9619 	/*
9620 	 * Profile KVM exit RIPs:
9621 	 */
9622 	if (unlikely(prof_on == KVM_PROFILING)) {
9623 		unsigned long rip = kvm_rip_read(vcpu);
9624 		profile_hit(KVM_PROFILING, (void *)rip);
9625 	}
9626 
9627 	if (unlikely(vcpu->arch.tsc_always_catchup))
9628 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9629 
9630 	if (vcpu->arch.apic_attention)
9631 		kvm_lapic_sync_from_vapic(vcpu);
9632 
9633 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9634 	return r;
9635 
9636 cancel_injection:
9637 	if (req_immediate_exit)
9638 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9639 	static_call(kvm_x86_cancel_injection)(vcpu);
9640 	if (unlikely(vcpu->arch.apic_attention))
9641 		kvm_lapic_sync_from_vapic(vcpu);
9642 out:
9643 	return r;
9644 }
9645 
9646 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9647 {
9648 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9649 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9650 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9651 		kvm_vcpu_block(vcpu);
9652 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9653 
9654 		if (kvm_x86_ops.post_block)
9655 			static_call(kvm_x86_post_block)(vcpu);
9656 
9657 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9658 			return 1;
9659 	}
9660 
9661 	if (kvm_apic_accept_events(vcpu) < 0)
9662 		return 0;
9663 	switch(vcpu->arch.mp_state) {
9664 	case KVM_MP_STATE_HALTED:
9665 	case KVM_MP_STATE_AP_RESET_HOLD:
9666 		vcpu->arch.pv.pv_unhalted = false;
9667 		vcpu->arch.mp_state =
9668 			KVM_MP_STATE_RUNNABLE;
9669 		fallthrough;
9670 	case KVM_MP_STATE_RUNNABLE:
9671 		vcpu->arch.apf.halted = false;
9672 		break;
9673 	case KVM_MP_STATE_INIT_RECEIVED:
9674 		break;
9675 	default:
9676 		return -EINTR;
9677 	}
9678 	return 1;
9679 }
9680 
9681 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9682 {
9683 	if (is_guest_mode(vcpu))
9684 		kvm_check_nested_events(vcpu);
9685 
9686 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9687 		!vcpu->arch.apf.halted);
9688 }
9689 
9690 static int vcpu_run(struct kvm_vcpu *vcpu)
9691 {
9692 	int r;
9693 	struct kvm *kvm = vcpu->kvm;
9694 
9695 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9696 	vcpu->arch.l1tf_flush_l1d = true;
9697 
9698 	for (;;) {
9699 		if (kvm_vcpu_running(vcpu)) {
9700 			r = vcpu_enter_guest(vcpu);
9701 		} else {
9702 			r = vcpu_block(kvm, vcpu);
9703 		}
9704 
9705 		if (r <= 0)
9706 			break;
9707 
9708 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9709 		if (kvm_cpu_has_pending_timer(vcpu))
9710 			kvm_inject_pending_timer_irqs(vcpu);
9711 
9712 		if (dm_request_for_irq_injection(vcpu) &&
9713 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9714 			r = 0;
9715 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9716 			++vcpu->stat.request_irq_exits;
9717 			break;
9718 		}
9719 
9720 		if (__xfer_to_guest_mode_work_pending()) {
9721 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9722 			r = xfer_to_guest_mode_handle_work(vcpu);
9723 			if (r)
9724 				return r;
9725 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9726 		}
9727 	}
9728 
9729 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9730 
9731 	return r;
9732 }
9733 
9734 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9735 {
9736 	int r;
9737 
9738 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9739 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9740 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9741 	return r;
9742 }
9743 
9744 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9745 {
9746 	BUG_ON(!vcpu->arch.pio.count);
9747 
9748 	return complete_emulated_io(vcpu);
9749 }
9750 
9751 /*
9752  * Implements the following, as a state machine:
9753  *
9754  * read:
9755  *   for each fragment
9756  *     for each mmio piece in the fragment
9757  *       write gpa, len
9758  *       exit
9759  *       copy data
9760  *   execute insn
9761  *
9762  * write:
9763  *   for each fragment
9764  *     for each mmio piece in the fragment
9765  *       write gpa, len
9766  *       copy data
9767  *       exit
9768  */
9769 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9770 {
9771 	struct kvm_run *run = vcpu->run;
9772 	struct kvm_mmio_fragment *frag;
9773 	unsigned len;
9774 
9775 	BUG_ON(!vcpu->mmio_needed);
9776 
9777 	/* Complete previous fragment */
9778 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9779 	len = min(8u, frag->len);
9780 	if (!vcpu->mmio_is_write)
9781 		memcpy(frag->data, run->mmio.data, len);
9782 
9783 	if (frag->len <= 8) {
9784 		/* Switch to the next fragment. */
9785 		frag++;
9786 		vcpu->mmio_cur_fragment++;
9787 	} else {
9788 		/* Go forward to the next mmio piece. */
9789 		frag->data += len;
9790 		frag->gpa += len;
9791 		frag->len -= len;
9792 	}
9793 
9794 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9795 		vcpu->mmio_needed = 0;
9796 
9797 		/* FIXME: return into emulator if single-stepping.  */
9798 		if (vcpu->mmio_is_write)
9799 			return 1;
9800 		vcpu->mmio_read_completed = 1;
9801 		return complete_emulated_io(vcpu);
9802 	}
9803 
9804 	run->exit_reason = KVM_EXIT_MMIO;
9805 	run->mmio.phys_addr = frag->gpa;
9806 	if (vcpu->mmio_is_write)
9807 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9808 	run->mmio.len = min(8u, frag->len);
9809 	run->mmio.is_write = vcpu->mmio_is_write;
9810 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9811 	return 0;
9812 }
9813 
9814 /* Swap (qemu) user FPU context for the guest FPU context. */
9815 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9816 {
9817 	/*
9818 	 * Exclude PKRU from restore as restored separately in
9819 	 * kvm_x86_ops.run().
9820 	 */
9821 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
9822 	trace_kvm_fpu(1);
9823 }
9824 
9825 /* When vcpu_run ends, restore user space FPU context. */
9826 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9827 {
9828 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
9829 	++vcpu->stat.fpu_reload;
9830 	trace_kvm_fpu(0);
9831 }
9832 
9833 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9834 {
9835 	struct kvm_run *kvm_run = vcpu->run;
9836 	int r;
9837 
9838 	vcpu_load(vcpu);
9839 	kvm_sigset_activate(vcpu);
9840 	kvm_run->flags = 0;
9841 	kvm_load_guest_fpu(vcpu);
9842 
9843 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9844 		if (kvm_run->immediate_exit) {
9845 			r = -EINTR;
9846 			goto out;
9847 		}
9848 		kvm_vcpu_block(vcpu);
9849 		if (kvm_apic_accept_events(vcpu) < 0) {
9850 			r = 0;
9851 			goto out;
9852 		}
9853 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9854 		r = -EAGAIN;
9855 		if (signal_pending(current)) {
9856 			r = -EINTR;
9857 			kvm_run->exit_reason = KVM_EXIT_INTR;
9858 			++vcpu->stat.signal_exits;
9859 		}
9860 		goto out;
9861 	}
9862 
9863 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
9864 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
9865 		r = -EINVAL;
9866 		goto out;
9867 	}
9868 
9869 	if (kvm_run->kvm_dirty_regs) {
9870 		r = sync_regs(vcpu);
9871 		if (r != 0)
9872 			goto out;
9873 	}
9874 
9875 	/* re-sync apic's tpr */
9876 	if (!lapic_in_kernel(vcpu)) {
9877 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9878 			r = -EINVAL;
9879 			goto out;
9880 		}
9881 	}
9882 
9883 	if (unlikely(vcpu->arch.complete_userspace_io)) {
9884 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9885 		vcpu->arch.complete_userspace_io = NULL;
9886 		r = cui(vcpu);
9887 		if (r <= 0)
9888 			goto out;
9889 	} else
9890 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9891 
9892 	if (kvm_run->immediate_exit)
9893 		r = -EINTR;
9894 	else
9895 		r = vcpu_run(vcpu);
9896 
9897 out:
9898 	kvm_put_guest_fpu(vcpu);
9899 	if (kvm_run->kvm_valid_regs)
9900 		store_regs(vcpu);
9901 	post_kvm_run_save(vcpu);
9902 	kvm_sigset_deactivate(vcpu);
9903 
9904 	vcpu_put(vcpu);
9905 	return r;
9906 }
9907 
9908 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9909 {
9910 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9911 		/*
9912 		 * We are here if userspace calls get_regs() in the middle of
9913 		 * instruction emulation. Registers state needs to be copied
9914 		 * back from emulation context to vcpu. Userspace shouldn't do
9915 		 * that usually, but some bad designed PV devices (vmware
9916 		 * backdoor interface) need this to work
9917 		 */
9918 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9919 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9920 	}
9921 	regs->rax = kvm_rax_read(vcpu);
9922 	regs->rbx = kvm_rbx_read(vcpu);
9923 	regs->rcx = kvm_rcx_read(vcpu);
9924 	regs->rdx = kvm_rdx_read(vcpu);
9925 	regs->rsi = kvm_rsi_read(vcpu);
9926 	regs->rdi = kvm_rdi_read(vcpu);
9927 	regs->rsp = kvm_rsp_read(vcpu);
9928 	regs->rbp = kvm_rbp_read(vcpu);
9929 #ifdef CONFIG_X86_64
9930 	regs->r8 = kvm_r8_read(vcpu);
9931 	regs->r9 = kvm_r9_read(vcpu);
9932 	regs->r10 = kvm_r10_read(vcpu);
9933 	regs->r11 = kvm_r11_read(vcpu);
9934 	regs->r12 = kvm_r12_read(vcpu);
9935 	regs->r13 = kvm_r13_read(vcpu);
9936 	regs->r14 = kvm_r14_read(vcpu);
9937 	regs->r15 = kvm_r15_read(vcpu);
9938 #endif
9939 
9940 	regs->rip = kvm_rip_read(vcpu);
9941 	regs->rflags = kvm_get_rflags(vcpu);
9942 }
9943 
9944 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9945 {
9946 	vcpu_load(vcpu);
9947 	__get_regs(vcpu, regs);
9948 	vcpu_put(vcpu);
9949 	return 0;
9950 }
9951 
9952 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9953 {
9954 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9955 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9956 
9957 	kvm_rax_write(vcpu, regs->rax);
9958 	kvm_rbx_write(vcpu, regs->rbx);
9959 	kvm_rcx_write(vcpu, regs->rcx);
9960 	kvm_rdx_write(vcpu, regs->rdx);
9961 	kvm_rsi_write(vcpu, regs->rsi);
9962 	kvm_rdi_write(vcpu, regs->rdi);
9963 	kvm_rsp_write(vcpu, regs->rsp);
9964 	kvm_rbp_write(vcpu, regs->rbp);
9965 #ifdef CONFIG_X86_64
9966 	kvm_r8_write(vcpu, regs->r8);
9967 	kvm_r9_write(vcpu, regs->r9);
9968 	kvm_r10_write(vcpu, regs->r10);
9969 	kvm_r11_write(vcpu, regs->r11);
9970 	kvm_r12_write(vcpu, regs->r12);
9971 	kvm_r13_write(vcpu, regs->r13);
9972 	kvm_r14_write(vcpu, regs->r14);
9973 	kvm_r15_write(vcpu, regs->r15);
9974 #endif
9975 
9976 	kvm_rip_write(vcpu, regs->rip);
9977 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9978 
9979 	vcpu->arch.exception.pending = false;
9980 
9981 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9982 }
9983 
9984 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9985 {
9986 	vcpu_load(vcpu);
9987 	__set_regs(vcpu, regs);
9988 	vcpu_put(vcpu);
9989 	return 0;
9990 }
9991 
9992 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9993 {
9994 	struct kvm_segment cs;
9995 
9996 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9997 	*db = cs.db;
9998 	*l = cs.l;
9999 }
10000 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10001 
10002 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10003 {
10004 	struct desc_ptr dt;
10005 
10006 	if (vcpu->arch.guest_state_protected)
10007 		goto skip_protected_regs;
10008 
10009 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10010 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10011 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10012 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10013 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10014 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10015 
10016 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10017 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10018 
10019 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10020 	sregs->idt.limit = dt.size;
10021 	sregs->idt.base = dt.address;
10022 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10023 	sregs->gdt.limit = dt.size;
10024 	sregs->gdt.base = dt.address;
10025 
10026 	sregs->cr2 = vcpu->arch.cr2;
10027 	sregs->cr3 = kvm_read_cr3(vcpu);
10028 
10029 skip_protected_regs:
10030 	sregs->cr0 = kvm_read_cr0(vcpu);
10031 	sregs->cr4 = kvm_read_cr4(vcpu);
10032 	sregs->cr8 = kvm_get_cr8(vcpu);
10033 	sregs->efer = vcpu->arch.efer;
10034 	sregs->apic_base = kvm_get_apic_base(vcpu);
10035 }
10036 
10037 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10038 {
10039 	__get_sregs_common(vcpu, sregs);
10040 
10041 	if (vcpu->arch.guest_state_protected)
10042 		return;
10043 
10044 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10045 		set_bit(vcpu->arch.interrupt.nr,
10046 			(unsigned long *)sregs->interrupt_bitmap);
10047 }
10048 
10049 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10050 {
10051 	int i;
10052 
10053 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10054 
10055 	if (vcpu->arch.guest_state_protected)
10056 		return;
10057 
10058 	if (is_pae_paging(vcpu)) {
10059 		for (i = 0 ; i < 4 ; i++)
10060 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10061 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10062 	}
10063 }
10064 
10065 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10066 				  struct kvm_sregs *sregs)
10067 {
10068 	vcpu_load(vcpu);
10069 	__get_sregs(vcpu, sregs);
10070 	vcpu_put(vcpu);
10071 	return 0;
10072 }
10073 
10074 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10075 				    struct kvm_mp_state *mp_state)
10076 {
10077 	int r;
10078 
10079 	vcpu_load(vcpu);
10080 	if (kvm_mpx_supported())
10081 		kvm_load_guest_fpu(vcpu);
10082 
10083 	r = kvm_apic_accept_events(vcpu);
10084 	if (r < 0)
10085 		goto out;
10086 	r = 0;
10087 
10088 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10089 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10090 	    vcpu->arch.pv.pv_unhalted)
10091 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10092 	else
10093 		mp_state->mp_state = vcpu->arch.mp_state;
10094 
10095 out:
10096 	if (kvm_mpx_supported())
10097 		kvm_put_guest_fpu(vcpu);
10098 	vcpu_put(vcpu);
10099 	return r;
10100 }
10101 
10102 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10103 				    struct kvm_mp_state *mp_state)
10104 {
10105 	int ret = -EINVAL;
10106 
10107 	vcpu_load(vcpu);
10108 
10109 	if (!lapic_in_kernel(vcpu) &&
10110 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10111 		goto out;
10112 
10113 	/*
10114 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10115 	 * INIT state; latched init should be reported using
10116 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10117 	 */
10118 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10119 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10120 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10121 		goto out;
10122 
10123 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10124 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10125 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10126 	} else
10127 		vcpu->arch.mp_state = mp_state->mp_state;
10128 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10129 
10130 	ret = 0;
10131 out:
10132 	vcpu_put(vcpu);
10133 	return ret;
10134 }
10135 
10136 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10137 		    int reason, bool has_error_code, u32 error_code)
10138 {
10139 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10140 	int ret;
10141 
10142 	init_emulate_ctxt(vcpu);
10143 
10144 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10145 				   has_error_code, error_code);
10146 	if (ret) {
10147 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10148 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10149 		vcpu->run->internal.ndata = 0;
10150 		return 0;
10151 	}
10152 
10153 	kvm_rip_write(vcpu, ctxt->eip);
10154 	kvm_set_rflags(vcpu, ctxt->eflags);
10155 	return 1;
10156 }
10157 EXPORT_SYMBOL_GPL(kvm_task_switch);
10158 
10159 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10160 {
10161 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10162 		/*
10163 		 * When EFER.LME and CR0.PG are set, the processor is in
10164 		 * 64-bit mode (though maybe in a 32-bit code segment).
10165 		 * CR4.PAE and EFER.LMA must be set.
10166 		 */
10167 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10168 			return false;
10169 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10170 			return false;
10171 	} else {
10172 		/*
10173 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10174 		 * segment cannot be 64-bit.
10175 		 */
10176 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10177 			return false;
10178 	}
10179 
10180 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10181 }
10182 
10183 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10184 		int *mmu_reset_needed, bool update_pdptrs)
10185 {
10186 	struct msr_data apic_base_msr;
10187 	int idx;
10188 	struct desc_ptr dt;
10189 
10190 	if (!kvm_is_valid_sregs(vcpu, sregs))
10191 		return -EINVAL;
10192 
10193 	apic_base_msr.data = sregs->apic_base;
10194 	apic_base_msr.host_initiated = true;
10195 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10196 		return -EINVAL;
10197 
10198 	if (vcpu->arch.guest_state_protected)
10199 		return 0;
10200 
10201 	dt.size = sregs->idt.limit;
10202 	dt.address = sregs->idt.base;
10203 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10204 	dt.size = sregs->gdt.limit;
10205 	dt.address = sregs->gdt.base;
10206 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10207 
10208 	vcpu->arch.cr2 = sregs->cr2;
10209 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10210 	vcpu->arch.cr3 = sregs->cr3;
10211 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10212 
10213 	kvm_set_cr8(vcpu, sregs->cr8);
10214 
10215 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10216 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10217 
10218 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10219 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10220 	vcpu->arch.cr0 = sregs->cr0;
10221 
10222 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10223 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10224 
10225 	if (update_pdptrs) {
10226 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10227 		if (is_pae_paging(vcpu)) {
10228 			load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10229 			*mmu_reset_needed = 1;
10230 		}
10231 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10232 	}
10233 
10234 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10235 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10236 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10237 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10238 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10239 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10240 
10241 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10242 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10243 
10244 	update_cr8_intercept(vcpu);
10245 
10246 	/* Older userspace won't unhalt the vcpu on reset. */
10247 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10248 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10249 	    !is_protmode(vcpu))
10250 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10251 
10252 	return 0;
10253 }
10254 
10255 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10256 {
10257 	int pending_vec, max_bits;
10258 	int mmu_reset_needed = 0;
10259 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10260 
10261 	if (ret)
10262 		return ret;
10263 
10264 	if (mmu_reset_needed)
10265 		kvm_mmu_reset_context(vcpu);
10266 
10267 	max_bits = KVM_NR_INTERRUPTS;
10268 	pending_vec = find_first_bit(
10269 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10270 
10271 	if (pending_vec < max_bits) {
10272 		kvm_queue_interrupt(vcpu, pending_vec, false);
10273 		pr_debug("Set back pending irq %d\n", pending_vec);
10274 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10275 	}
10276 	return 0;
10277 }
10278 
10279 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10280 {
10281 	int mmu_reset_needed = 0;
10282 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10283 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10284 		!(sregs2->efer & EFER_LMA);
10285 	int i, ret;
10286 
10287 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10288 		return -EINVAL;
10289 
10290 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10291 		return -EINVAL;
10292 
10293 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10294 				 &mmu_reset_needed, !valid_pdptrs);
10295 	if (ret)
10296 		return ret;
10297 
10298 	if (valid_pdptrs) {
10299 		for (i = 0; i < 4 ; i++)
10300 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10301 
10302 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10303 		mmu_reset_needed = 1;
10304 		vcpu->arch.pdptrs_from_userspace = true;
10305 	}
10306 	if (mmu_reset_needed)
10307 		kvm_mmu_reset_context(vcpu);
10308 	return 0;
10309 }
10310 
10311 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10312 				  struct kvm_sregs *sregs)
10313 {
10314 	int ret;
10315 
10316 	vcpu_load(vcpu);
10317 	ret = __set_sregs(vcpu, sregs);
10318 	vcpu_put(vcpu);
10319 	return ret;
10320 }
10321 
10322 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10323 					struct kvm_guest_debug *dbg)
10324 {
10325 	unsigned long rflags;
10326 	int i, r;
10327 
10328 	if (vcpu->arch.guest_state_protected)
10329 		return -EINVAL;
10330 
10331 	vcpu_load(vcpu);
10332 
10333 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10334 		r = -EBUSY;
10335 		if (vcpu->arch.exception.pending)
10336 			goto out;
10337 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10338 			kvm_queue_exception(vcpu, DB_VECTOR);
10339 		else
10340 			kvm_queue_exception(vcpu, BP_VECTOR);
10341 	}
10342 
10343 	/*
10344 	 * Read rflags as long as potentially injected trace flags are still
10345 	 * filtered out.
10346 	 */
10347 	rflags = kvm_get_rflags(vcpu);
10348 
10349 	vcpu->guest_debug = dbg->control;
10350 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10351 		vcpu->guest_debug = 0;
10352 
10353 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10354 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10355 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10356 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10357 	} else {
10358 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10359 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10360 	}
10361 	kvm_update_dr7(vcpu);
10362 
10363 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10364 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10365 
10366 	/*
10367 	 * Trigger an rflags update that will inject or remove the trace
10368 	 * flags.
10369 	 */
10370 	kvm_set_rflags(vcpu, rflags);
10371 
10372 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10373 
10374 	r = 0;
10375 
10376 out:
10377 	vcpu_put(vcpu);
10378 	return r;
10379 }
10380 
10381 /*
10382  * Translate a guest virtual address to a guest physical address.
10383  */
10384 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10385 				    struct kvm_translation *tr)
10386 {
10387 	unsigned long vaddr = tr->linear_address;
10388 	gpa_t gpa;
10389 	int idx;
10390 
10391 	vcpu_load(vcpu);
10392 
10393 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10394 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10395 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10396 	tr->physical_address = gpa;
10397 	tr->valid = gpa != UNMAPPED_GVA;
10398 	tr->writeable = 1;
10399 	tr->usermode = 0;
10400 
10401 	vcpu_put(vcpu);
10402 	return 0;
10403 }
10404 
10405 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10406 {
10407 	struct fxregs_state *fxsave;
10408 
10409 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10410 		return 0;
10411 
10412 	vcpu_load(vcpu);
10413 
10414 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10415 	memcpy(fpu->fpr, fxsave->st_space, 128);
10416 	fpu->fcw = fxsave->cwd;
10417 	fpu->fsw = fxsave->swd;
10418 	fpu->ftwx = fxsave->twd;
10419 	fpu->last_opcode = fxsave->fop;
10420 	fpu->last_ip = fxsave->rip;
10421 	fpu->last_dp = fxsave->rdp;
10422 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10423 
10424 	vcpu_put(vcpu);
10425 	return 0;
10426 }
10427 
10428 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10429 {
10430 	struct fxregs_state *fxsave;
10431 
10432 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10433 		return 0;
10434 
10435 	vcpu_load(vcpu);
10436 
10437 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10438 
10439 	memcpy(fxsave->st_space, fpu->fpr, 128);
10440 	fxsave->cwd = fpu->fcw;
10441 	fxsave->swd = fpu->fsw;
10442 	fxsave->twd = fpu->ftwx;
10443 	fxsave->fop = fpu->last_opcode;
10444 	fxsave->rip = fpu->last_ip;
10445 	fxsave->rdp = fpu->last_dp;
10446 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10447 
10448 	vcpu_put(vcpu);
10449 	return 0;
10450 }
10451 
10452 static void store_regs(struct kvm_vcpu *vcpu)
10453 {
10454 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10455 
10456 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10457 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10458 
10459 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10460 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10461 
10462 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10463 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10464 				vcpu, &vcpu->run->s.regs.events);
10465 }
10466 
10467 static int sync_regs(struct kvm_vcpu *vcpu)
10468 {
10469 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10470 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10471 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10472 	}
10473 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10474 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10475 			return -EINVAL;
10476 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10477 	}
10478 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10479 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10480 				vcpu, &vcpu->run->s.regs.events))
10481 			return -EINVAL;
10482 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10483 	}
10484 
10485 	return 0;
10486 }
10487 
10488 static void fx_init(struct kvm_vcpu *vcpu)
10489 {
10490 	/*
10491 	 * Ensure guest xcr0 is valid for loading
10492 	 */
10493 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10494 
10495 	vcpu->arch.cr0 |= X86_CR0_ET;
10496 }
10497 
10498 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10499 {
10500 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10501 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10502 			     "guest TSC will not be reliable\n");
10503 
10504 	return 0;
10505 }
10506 
10507 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10508 {
10509 	struct page *page;
10510 	int r;
10511 
10512 	vcpu->arch.last_vmentry_cpu = -1;
10513 	vcpu->arch.regs_avail = ~0;
10514 	vcpu->arch.regs_dirty = ~0;
10515 
10516 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10517 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10518 	else
10519 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10520 
10521 	r = kvm_mmu_create(vcpu);
10522 	if (r < 0)
10523 		return r;
10524 
10525 	if (irqchip_in_kernel(vcpu->kvm)) {
10526 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10527 		if (r < 0)
10528 			goto fail_mmu_destroy;
10529 		if (kvm_apicv_activated(vcpu->kvm))
10530 			vcpu->arch.apicv_active = true;
10531 	} else
10532 		static_branch_inc(&kvm_has_noapic_vcpu);
10533 
10534 	r = -ENOMEM;
10535 
10536 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10537 	if (!page)
10538 		goto fail_free_lapic;
10539 	vcpu->arch.pio_data = page_address(page);
10540 
10541 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10542 				       GFP_KERNEL_ACCOUNT);
10543 	if (!vcpu->arch.mce_banks)
10544 		goto fail_free_pio_data;
10545 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10546 
10547 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10548 				GFP_KERNEL_ACCOUNT))
10549 		goto fail_free_mce_banks;
10550 
10551 	if (!alloc_emulate_ctxt(vcpu))
10552 		goto free_wbinvd_dirty_mask;
10553 
10554 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
10555 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10556 		goto free_emulate_ctxt;
10557 	}
10558 
10559 	fx_init(vcpu);
10560 
10561 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10562 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10563 
10564 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10565 
10566 	kvm_async_pf_hash_reset(vcpu);
10567 	kvm_pmu_init(vcpu);
10568 
10569 	vcpu->arch.pending_external_vector = -1;
10570 	vcpu->arch.preempted_in_kernel = false;
10571 
10572 #if IS_ENABLED(CONFIG_HYPERV)
10573 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
10574 #endif
10575 
10576 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10577 	if (r)
10578 		goto free_guest_fpu;
10579 
10580 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10581 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10582 	kvm_vcpu_mtrr_init(vcpu);
10583 	vcpu_load(vcpu);
10584 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10585 	kvm_vcpu_reset(vcpu, false);
10586 	kvm_init_mmu(vcpu);
10587 	vcpu_put(vcpu);
10588 	return 0;
10589 
10590 free_guest_fpu:
10591 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10592 free_emulate_ctxt:
10593 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10594 free_wbinvd_dirty_mask:
10595 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10596 fail_free_mce_banks:
10597 	kfree(vcpu->arch.mce_banks);
10598 fail_free_pio_data:
10599 	free_page((unsigned long)vcpu->arch.pio_data);
10600 fail_free_lapic:
10601 	kvm_free_lapic(vcpu);
10602 fail_mmu_destroy:
10603 	kvm_mmu_destroy(vcpu);
10604 	return r;
10605 }
10606 
10607 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10608 {
10609 	struct kvm *kvm = vcpu->kvm;
10610 
10611 	if (mutex_lock_killable(&vcpu->mutex))
10612 		return;
10613 	vcpu_load(vcpu);
10614 	kvm_synchronize_tsc(vcpu, 0);
10615 	vcpu_put(vcpu);
10616 
10617 	/* poll control enabled by default */
10618 	vcpu->arch.msr_kvm_poll_control = 1;
10619 
10620 	mutex_unlock(&vcpu->mutex);
10621 
10622 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10623 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10624 						KVMCLOCK_SYNC_PERIOD);
10625 }
10626 
10627 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10628 {
10629 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10630 	int idx;
10631 
10632 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10633 
10634 	kvmclock_reset(vcpu);
10635 
10636 	static_call(kvm_x86_vcpu_free)(vcpu);
10637 
10638 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10639 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10640 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10641 
10642 	kvm_hv_vcpu_uninit(vcpu);
10643 	kvm_pmu_destroy(vcpu);
10644 	kfree(vcpu->arch.mce_banks);
10645 	kvm_free_lapic(vcpu);
10646 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10647 	kvm_mmu_destroy(vcpu);
10648 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10649 	free_page((unsigned long)vcpu->arch.pio_data);
10650 	kvfree(vcpu->arch.cpuid_entries);
10651 	if (!lapic_in_kernel(vcpu))
10652 		static_branch_dec(&kvm_has_noapic_vcpu);
10653 }
10654 
10655 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10656 {
10657 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
10658 	unsigned long new_cr0;
10659 	u32 eax, dummy;
10660 
10661 	kvm_lapic_reset(vcpu, init_event);
10662 
10663 	vcpu->arch.hflags = 0;
10664 
10665 	vcpu->arch.smi_pending = 0;
10666 	vcpu->arch.smi_count = 0;
10667 	atomic_set(&vcpu->arch.nmi_queued, 0);
10668 	vcpu->arch.nmi_pending = 0;
10669 	vcpu->arch.nmi_injected = false;
10670 	kvm_clear_interrupt_queue(vcpu);
10671 	kvm_clear_exception_queue(vcpu);
10672 
10673 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10674 	kvm_update_dr0123(vcpu);
10675 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10676 	vcpu->arch.dr7 = DR7_FIXED_1;
10677 	kvm_update_dr7(vcpu);
10678 
10679 	vcpu->arch.cr2 = 0;
10680 
10681 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10682 	vcpu->arch.apf.msr_en_val = 0;
10683 	vcpu->arch.apf.msr_int_val = 0;
10684 	vcpu->arch.st.msr_val = 0;
10685 
10686 	kvmclock_reset(vcpu);
10687 
10688 	kvm_clear_async_pf_completion_queue(vcpu);
10689 	kvm_async_pf_hash_reset(vcpu);
10690 	vcpu->arch.apf.halted = false;
10691 
10692 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
10693 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
10694 
10695 		/*
10696 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10697 		 * called with loaded FPU and does not let userspace fix the state.
10698 		 */
10699 		if (init_event)
10700 			kvm_put_guest_fpu(vcpu);
10701 
10702 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
10703 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
10704 
10705 		if (init_event)
10706 			kvm_load_guest_fpu(vcpu);
10707 	}
10708 
10709 	if (!init_event) {
10710 		kvm_pmu_reset(vcpu);
10711 		vcpu->arch.smbase = 0x30000;
10712 
10713 		vcpu->arch.msr_misc_features_enables = 0;
10714 
10715 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10716 	}
10717 
10718 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10719 	vcpu->arch.regs_avail = ~0;
10720 	vcpu->arch.regs_dirty = ~0;
10721 
10722 	/*
10723 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
10724 	 * if no CPUID match is found.  Note, it's impossible to get a match at
10725 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
10726 	 * i.e. it'simpossible for kvm_cpuid() to find a valid entry on RESET.
10727 	 * But, go through the motions in case that's ever remedied.
10728 	 */
10729 	eax = 1;
10730 	if (!kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true))
10731 		eax = 0x600;
10732 	kvm_rdx_write(vcpu, eax);
10733 
10734 	vcpu->arch.ia32_xss = 0;
10735 
10736 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10737 
10738 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
10739 	kvm_rip_write(vcpu, 0xfff0);
10740 
10741 	vcpu->arch.cr3 = 0;
10742 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10743 
10744 	/*
10745 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
10746 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
10747 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
10748 	 */
10749 	new_cr0 = X86_CR0_ET;
10750 	if (init_event)
10751 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
10752 	else
10753 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
10754 
10755 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
10756 	static_call(kvm_x86_set_cr4)(vcpu, 0);
10757 	static_call(kvm_x86_set_efer)(vcpu, 0);
10758 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10759 
10760 	/*
10761 	 * Reset the MMU context if paging was enabled prior to INIT (which is
10762 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
10763 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
10764 	 * checked because it is unconditionally cleared on INIT and all other
10765 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
10766 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
10767 	 */
10768 	if (old_cr0 & X86_CR0_PG)
10769 		kvm_mmu_reset_context(vcpu);
10770 
10771 	/*
10772 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
10773 	 * APM states the TLBs are untouched by INIT, but it also states that
10774 	 * the TLBs are flushed on "External initialization of the processor."
10775 	 * Flush the guest TLB regardless of vendor, there is no meaningful
10776 	 * benefit in relying on the guest to flush the TLB immediately after
10777 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
10778 	 * performance perspective.
10779 	 */
10780 	if (init_event)
10781 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
10782 }
10783 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
10784 
10785 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10786 {
10787 	struct kvm_segment cs;
10788 
10789 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10790 	cs.selector = vector << 8;
10791 	cs.base = vector << 12;
10792 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10793 	kvm_rip_write(vcpu, 0);
10794 }
10795 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10796 
10797 int kvm_arch_hardware_enable(void)
10798 {
10799 	struct kvm *kvm;
10800 	struct kvm_vcpu *vcpu;
10801 	int i;
10802 	int ret;
10803 	u64 local_tsc;
10804 	u64 max_tsc = 0;
10805 	bool stable, backwards_tsc = false;
10806 
10807 	kvm_user_return_msr_cpu_online();
10808 	ret = static_call(kvm_x86_hardware_enable)();
10809 	if (ret != 0)
10810 		return ret;
10811 
10812 	local_tsc = rdtsc();
10813 	stable = !kvm_check_tsc_unstable();
10814 	list_for_each_entry(kvm, &vm_list, vm_list) {
10815 		kvm_for_each_vcpu(i, vcpu, kvm) {
10816 			if (!stable && vcpu->cpu == smp_processor_id())
10817 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10818 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10819 				backwards_tsc = true;
10820 				if (vcpu->arch.last_host_tsc > max_tsc)
10821 					max_tsc = vcpu->arch.last_host_tsc;
10822 			}
10823 		}
10824 	}
10825 
10826 	/*
10827 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10828 	 * platforms that reset TSC during suspend or hibernate actions, but
10829 	 * maintain synchronization.  We must compensate.  Fortunately, we can
10830 	 * detect that condition here, which happens early in CPU bringup,
10831 	 * before any KVM threads can be running.  Unfortunately, we can't
10832 	 * bring the TSCs fully up to date with real time, as we aren't yet far
10833 	 * enough into CPU bringup that we know how much real time has actually
10834 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10835 	 * variables that haven't been updated yet.
10836 	 *
10837 	 * So we simply find the maximum observed TSC above, then record the
10838 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
10839 	 * the adjustment will be applied.  Note that we accumulate
10840 	 * adjustments, in case multiple suspend cycles happen before some VCPU
10841 	 * gets a chance to run again.  In the event that no KVM threads get a
10842 	 * chance to run, we will miss the entire elapsed period, as we'll have
10843 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10844 	 * loose cycle time.  This isn't too big a deal, since the loss will be
10845 	 * uniform across all VCPUs (not to mention the scenario is extremely
10846 	 * unlikely). It is possible that a second hibernate recovery happens
10847 	 * much faster than a first, causing the observed TSC here to be
10848 	 * smaller; this would require additional padding adjustment, which is
10849 	 * why we set last_host_tsc to the local tsc observed here.
10850 	 *
10851 	 * N.B. - this code below runs only on platforms with reliable TSC,
10852 	 * as that is the only way backwards_tsc is set above.  Also note
10853 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10854 	 * have the same delta_cyc adjustment applied if backwards_tsc
10855 	 * is detected.  Note further, this adjustment is only done once,
10856 	 * as we reset last_host_tsc on all VCPUs to stop this from being
10857 	 * called multiple times (one for each physical CPU bringup).
10858 	 *
10859 	 * Platforms with unreliable TSCs don't have to deal with this, they
10860 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
10861 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
10862 	 * guarantee that they stay in perfect synchronization.
10863 	 */
10864 	if (backwards_tsc) {
10865 		u64 delta_cyc = max_tsc - local_tsc;
10866 		list_for_each_entry(kvm, &vm_list, vm_list) {
10867 			kvm->arch.backwards_tsc_observed = true;
10868 			kvm_for_each_vcpu(i, vcpu, kvm) {
10869 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
10870 				vcpu->arch.last_host_tsc = local_tsc;
10871 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10872 			}
10873 
10874 			/*
10875 			 * We have to disable TSC offset matching.. if you were
10876 			 * booting a VM while issuing an S4 host suspend....
10877 			 * you may have some problem.  Solving this issue is
10878 			 * left as an exercise to the reader.
10879 			 */
10880 			kvm->arch.last_tsc_nsec = 0;
10881 			kvm->arch.last_tsc_write = 0;
10882 		}
10883 
10884 	}
10885 	return 0;
10886 }
10887 
10888 void kvm_arch_hardware_disable(void)
10889 {
10890 	static_call(kvm_x86_hardware_disable)();
10891 	drop_user_return_notifiers();
10892 }
10893 
10894 int kvm_arch_hardware_setup(void *opaque)
10895 {
10896 	struct kvm_x86_init_ops *ops = opaque;
10897 	int r;
10898 
10899 	rdmsrl_safe(MSR_EFER, &host_efer);
10900 
10901 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10902 		rdmsrl(MSR_IA32_XSS, host_xss);
10903 
10904 	r = ops->hardware_setup();
10905 	if (r != 0)
10906 		return r;
10907 
10908 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10909 	kvm_ops_static_call_update();
10910 
10911 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10912 		supported_xss = 0;
10913 
10914 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10915 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10916 #undef __kvm_cpu_cap_has
10917 
10918 	if (kvm_has_tsc_control) {
10919 		/*
10920 		 * Make sure the user can only configure tsc_khz values that
10921 		 * fit into a signed integer.
10922 		 * A min value is not calculated because it will always
10923 		 * be 1 on all machines.
10924 		 */
10925 		u64 max = min(0x7fffffffULL,
10926 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10927 		kvm_max_guest_tsc_khz = max;
10928 
10929 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10930 	}
10931 
10932 	kvm_init_msr_list();
10933 	return 0;
10934 }
10935 
10936 void kvm_arch_hardware_unsetup(void)
10937 {
10938 	static_call(kvm_x86_hardware_unsetup)();
10939 }
10940 
10941 int kvm_arch_check_processor_compat(void *opaque)
10942 {
10943 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10944 	struct kvm_x86_init_ops *ops = opaque;
10945 
10946 	WARN_ON(!irqs_disabled());
10947 
10948 	if (__cr4_reserved_bits(cpu_has, c) !=
10949 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10950 		return -EIO;
10951 
10952 	return ops->check_processor_compatibility();
10953 }
10954 
10955 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10956 {
10957 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10958 }
10959 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10960 
10961 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10962 {
10963 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10964 }
10965 
10966 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10967 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10968 
10969 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10970 {
10971 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10972 
10973 	vcpu->arch.l1tf_flush_l1d = true;
10974 	if (pmu->version && unlikely(pmu->event_count)) {
10975 		pmu->need_cleanup = true;
10976 		kvm_make_request(KVM_REQ_PMU, vcpu);
10977 	}
10978 	static_call(kvm_x86_sched_in)(vcpu, cpu);
10979 }
10980 
10981 void kvm_arch_free_vm(struct kvm *kvm)
10982 {
10983 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
10984 	vfree(kvm);
10985 }
10986 
10987 
10988 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10989 {
10990 	int ret;
10991 
10992 	if (type)
10993 		return -EINVAL;
10994 
10995 	ret = kvm_page_track_init(kvm);
10996 	if (ret)
10997 		return ret;
10998 
10999 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11000 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11001 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11002 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11003 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11004 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11005 
11006 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11007 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11008 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11009 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11010 		&kvm->arch.irq_sources_bitmap);
11011 
11012 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11013 	mutex_init(&kvm->arch.apic_map_lock);
11014 	raw_spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11015 
11016 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11017 	pvclock_update_vm_gtod_copy(kvm);
11018 
11019 	kvm->arch.guest_can_read_msr_platform_info = true;
11020 
11021 #if IS_ENABLED(CONFIG_HYPERV)
11022 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11023 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11024 #endif
11025 
11026 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11027 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11028 
11029 	kvm_apicv_init(kvm);
11030 	kvm_hv_init_vm(kvm);
11031 	kvm_mmu_init_vm(kvm);
11032 	kvm_xen_init_vm(kvm);
11033 
11034 	return static_call(kvm_x86_vm_init)(kvm);
11035 }
11036 
11037 int kvm_arch_post_init_vm(struct kvm *kvm)
11038 {
11039 	return kvm_mmu_post_init_vm(kvm);
11040 }
11041 
11042 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11043 {
11044 	vcpu_load(vcpu);
11045 	kvm_mmu_unload(vcpu);
11046 	vcpu_put(vcpu);
11047 }
11048 
11049 static void kvm_free_vcpus(struct kvm *kvm)
11050 {
11051 	unsigned int i;
11052 	struct kvm_vcpu *vcpu;
11053 
11054 	/*
11055 	 * Unpin any mmu pages first.
11056 	 */
11057 	kvm_for_each_vcpu(i, vcpu, kvm) {
11058 		kvm_clear_async_pf_completion_queue(vcpu);
11059 		kvm_unload_vcpu_mmu(vcpu);
11060 	}
11061 	kvm_for_each_vcpu(i, vcpu, kvm)
11062 		kvm_vcpu_destroy(vcpu);
11063 
11064 	mutex_lock(&kvm->lock);
11065 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11066 		kvm->vcpus[i] = NULL;
11067 
11068 	atomic_set(&kvm->online_vcpus, 0);
11069 	mutex_unlock(&kvm->lock);
11070 }
11071 
11072 void kvm_arch_sync_events(struct kvm *kvm)
11073 {
11074 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11075 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11076 	kvm_free_pit(kvm);
11077 }
11078 
11079 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11080 
11081 /**
11082  * __x86_set_memory_region: Setup KVM internal memory slot
11083  *
11084  * @kvm: the kvm pointer to the VM.
11085  * @id: the slot ID to setup.
11086  * @gpa: the GPA to install the slot (unused when @size == 0).
11087  * @size: the size of the slot. Set to zero to uninstall a slot.
11088  *
11089  * This function helps to setup a KVM internal memory slot.  Specify
11090  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11091  * slot.  The return code can be one of the following:
11092  *
11093  *   HVA:           on success (uninstall will return a bogus HVA)
11094  *   -errno:        on error
11095  *
11096  * The caller should always use IS_ERR() to check the return value
11097  * before use.  Note, the KVM internal memory slots are guaranteed to
11098  * remain valid and unchanged until the VM is destroyed, i.e., the
11099  * GPA->HVA translation will not change.  However, the HVA is a user
11100  * address, i.e. its accessibility is not guaranteed, and must be
11101  * accessed via __copy_{to,from}_user().
11102  */
11103 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11104 				      u32 size)
11105 {
11106 	int i, r;
11107 	unsigned long hva, old_npages;
11108 	struct kvm_memslots *slots = kvm_memslots(kvm);
11109 	struct kvm_memory_slot *slot;
11110 
11111 	/* Called with kvm->slots_lock held.  */
11112 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11113 		return ERR_PTR_USR(-EINVAL);
11114 
11115 	slot = id_to_memslot(slots, id);
11116 	if (size) {
11117 		if (slot && slot->npages)
11118 			return ERR_PTR_USR(-EEXIST);
11119 
11120 		/*
11121 		 * MAP_SHARED to prevent internal slot pages from being moved
11122 		 * by fork()/COW.
11123 		 */
11124 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11125 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11126 		if (IS_ERR((void *)hva))
11127 			return (void __user *)hva;
11128 	} else {
11129 		if (!slot || !slot->npages)
11130 			return NULL;
11131 
11132 		old_npages = slot->npages;
11133 		hva = slot->userspace_addr;
11134 	}
11135 
11136 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11137 		struct kvm_userspace_memory_region m;
11138 
11139 		m.slot = id | (i << 16);
11140 		m.flags = 0;
11141 		m.guest_phys_addr = gpa;
11142 		m.userspace_addr = hva;
11143 		m.memory_size = size;
11144 		r = __kvm_set_memory_region(kvm, &m);
11145 		if (r < 0)
11146 			return ERR_PTR_USR(r);
11147 	}
11148 
11149 	if (!size)
11150 		vm_munmap(hva, old_npages * PAGE_SIZE);
11151 
11152 	return (void __user *)hva;
11153 }
11154 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11155 
11156 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11157 {
11158 	kvm_mmu_pre_destroy_vm(kvm);
11159 }
11160 
11161 void kvm_arch_destroy_vm(struct kvm *kvm)
11162 {
11163 	if (current->mm == kvm->mm) {
11164 		/*
11165 		 * Free memory regions allocated on behalf of userspace,
11166 		 * unless the the memory map has changed due to process exit
11167 		 * or fd copying.
11168 		 */
11169 		mutex_lock(&kvm->slots_lock);
11170 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11171 					0, 0);
11172 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11173 					0, 0);
11174 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11175 		mutex_unlock(&kvm->slots_lock);
11176 	}
11177 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11178 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11179 	kvm_pic_destroy(kvm);
11180 	kvm_ioapic_destroy(kvm);
11181 	kvm_free_vcpus(kvm);
11182 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11183 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11184 	kvm_mmu_uninit_vm(kvm);
11185 	kvm_page_track_cleanup(kvm);
11186 	kvm_xen_destroy_vm(kvm);
11187 	kvm_hv_destroy_vm(kvm);
11188 }
11189 
11190 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11191 {
11192 	int i;
11193 
11194 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11195 		kvfree(slot->arch.rmap[i]);
11196 		slot->arch.rmap[i] = NULL;
11197 	}
11198 }
11199 
11200 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11201 {
11202 	int i;
11203 
11204 	memslot_rmap_free(slot);
11205 
11206 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11207 		kvfree(slot->arch.lpage_info[i - 1]);
11208 		slot->arch.lpage_info[i - 1] = NULL;
11209 	}
11210 
11211 	kvm_page_track_free_memslot(slot);
11212 }
11213 
11214 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11215 			      unsigned long npages)
11216 {
11217 	const int sz = sizeof(*slot->arch.rmap[0]);
11218 	int i;
11219 
11220 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11221 		int level = i + 1;
11222 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11223 
11224 		if (slot->arch.rmap[i])
11225 			continue;
11226 
11227 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11228 		if (!slot->arch.rmap[i]) {
11229 			memslot_rmap_free(slot);
11230 			return -ENOMEM;
11231 		}
11232 	}
11233 
11234 	return 0;
11235 }
11236 
11237 int alloc_all_memslots_rmaps(struct kvm *kvm)
11238 {
11239 	struct kvm_memslots *slots;
11240 	struct kvm_memory_slot *slot;
11241 	int r, i;
11242 
11243 	/*
11244 	 * Check if memslots alreday have rmaps early before acquiring
11245 	 * the slots_arch_lock below.
11246 	 */
11247 	if (kvm_memslots_have_rmaps(kvm))
11248 		return 0;
11249 
11250 	mutex_lock(&kvm->slots_arch_lock);
11251 
11252 	/*
11253 	 * Read memslots_have_rmaps again, under the slots arch lock,
11254 	 * before allocating the rmaps
11255 	 */
11256 	if (kvm_memslots_have_rmaps(kvm)) {
11257 		mutex_unlock(&kvm->slots_arch_lock);
11258 		return 0;
11259 	}
11260 
11261 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11262 		slots = __kvm_memslots(kvm, i);
11263 		kvm_for_each_memslot(slot, slots) {
11264 			r = memslot_rmap_alloc(slot, slot->npages);
11265 			if (r) {
11266 				mutex_unlock(&kvm->slots_arch_lock);
11267 				return r;
11268 			}
11269 		}
11270 	}
11271 
11272 	/*
11273 	 * Ensure that memslots_have_rmaps becomes true strictly after
11274 	 * all the rmap pointers are set.
11275 	 */
11276 	smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11277 	mutex_unlock(&kvm->slots_arch_lock);
11278 	return 0;
11279 }
11280 
11281 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11282 				      struct kvm_memory_slot *slot,
11283 				      unsigned long npages)
11284 {
11285 	int i, r;
11286 
11287 	/*
11288 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11289 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11290 	 * the new memslot is successful.
11291 	 */
11292 	memset(&slot->arch, 0, sizeof(slot->arch));
11293 
11294 	if (kvm_memslots_have_rmaps(kvm)) {
11295 		r = memslot_rmap_alloc(slot, npages);
11296 		if (r)
11297 			return r;
11298 	}
11299 
11300 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11301 		struct kvm_lpage_info *linfo;
11302 		unsigned long ugfn;
11303 		int lpages;
11304 		int level = i + 1;
11305 
11306 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11307 
11308 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11309 		if (!linfo)
11310 			goto out_free;
11311 
11312 		slot->arch.lpage_info[i - 1] = linfo;
11313 
11314 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11315 			linfo[0].disallow_lpage = 1;
11316 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11317 			linfo[lpages - 1].disallow_lpage = 1;
11318 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11319 		/*
11320 		 * If the gfn and userspace address are not aligned wrt each
11321 		 * other, disable large page support for this slot.
11322 		 */
11323 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11324 			unsigned long j;
11325 
11326 			for (j = 0; j < lpages; ++j)
11327 				linfo[j].disallow_lpage = 1;
11328 		}
11329 	}
11330 
11331 	if (kvm_page_track_create_memslot(slot, npages))
11332 		goto out_free;
11333 
11334 	return 0;
11335 
11336 out_free:
11337 	memslot_rmap_free(slot);
11338 
11339 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11340 		kvfree(slot->arch.lpage_info[i - 1]);
11341 		slot->arch.lpage_info[i - 1] = NULL;
11342 	}
11343 	return -ENOMEM;
11344 }
11345 
11346 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11347 {
11348 	struct kvm_vcpu *vcpu;
11349 	int i;
11350 
11351 	/*
11352 	 * memslots->generation has been incremented.
11353 	 * mmio generation may have reached its maximum value.
11354 	 */
11355 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11356 
11357 	/* Force re-initialization of steal_time cache */
11358 	kvm_for_each_vcpu(i, vcpu, kvm)
11359 		kvm_vcpu_kick(vcpu);
11360 }
11361 
11362 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11363 				struct kvm_memory_slot *memslot,
11364 				const struct kvm_userspace_memory_region *mem,
11365 				enum kvm_mr_change change)
11366 {
11367 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11368 		return kvm_alloc_memslot_metadata(kvm, memslot,
11369 						  mem->memory_size >> PAGE_SHIFT);
11370 	return 0;
11371 }
11372 
11373 
11374 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11375 {
11376 	struct kvm_arch *ka = &kvm->arch;
11377 
11378 	if (!kvm_x86_ops.cpu_dirty_log_size)
11379 		return;
11380 
11381 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11382 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11383 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11384 
11385 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11386 }
11387 
11388 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11389 				     struct kvm_memory_slot *old,
11390 				     const struct kvm_memory_slot *new,
11391 				     enum kvm_mr_change change)
11392 {
11393 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11394 
11395 	/*
11396 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11397 	 * applies to all operations.
11398 	 */
11399 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11400 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11401 
11402 	/*
11403 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11404 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11405 	 *
11406 	 * For a memslot with dirty logging disabled:
11407 	 * CREATE:      No dirty mappings will already exist.
11408 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11409 	 *		kvm_arch_flush_shadow_memslot()
11410 	 *
11411 	 * For a memslot with dirty logging enabled:
11412 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11413 	 *		and no dirty bits to clear.
11414 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11415 	 *		kvm_arch_flush_shadow_memslot().
11416 	 */
11417 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11418 		return;
11419 
11420 	/*
11421 	 * READONLY and non-flags changes were filtered out above, and the only
11422 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11423 	 * logging isn't being toggled on or off.
11424 	 */
11425 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11426 		return;
11427 
11428 	if (!log_dirty_pages) {
11429 		/*
11430 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11431 		 * large sptes have to be split.  If live migration succeeds,
11432 		 * the guest in the source machine will be destroyed and large
11433 		 * sptes will be created in the destination.  However, if the
11434 		 * guest continues to run in the source machine (for example if
11435 		 * live migration fails), small sptes will remain around and
11436 		 * cause bad performance.
11437 		 *
11438 		 * Scan sptes if dirty logging has been stopped, dropping those
11439 		 * which can be collapsed into a single large-page spte.  Later
11440 		 * page faults will create the large-page sptes.
11441 		 */
11442 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11443 	} else {
11444 		/*
11445 		 * Initially-all-set does not require write protecting any page,
11446 		 * because they're all assumed to be dirty.
11447 		 */
11448 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11449 			return;
11450 
11451 		if (kvm_x86_ops.cpu_dirty_log_size) {
11452 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11453 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11454 		} else {
11455 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11456 		}
11457 	}
11458 }
11459 
11460 void kvm_arch_commit_memory_region(struct kvm *kvm,
11461 				const struct kvm_userspace_memory_region *mem,
11462 				struct kvm_memory_slot *old,
11463 				const struct kvm_memory_slot *new,
11464 				enum kvm_mr_change change)
11465 {
11466 	if (!kvm->arch.n_requested_mmu_pages)
11467 		kvm_mmu_change_mmu_pages(kvm,
11468 				kvm_mmu_calculate_default_mmu_pages(kvm));
11469 
11470 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
11471 
11472 	/* Free the arrays associated with the old memslot. */
11473 	if (change == KVM_MR_MOVE)
11474 		kvm_arch_free_memslot(kvm, old);
11475 }
11476 
11477 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11478 {
11479 	kvm_mmu_zap_all(kvm);
11480 }
11481 
11482 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11483 				   struct kvm_memory_slot *slot)
11484 {
11485 	kvm_page_track_flush_slot(kvm, slot);
11486 }
11487 
11488 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11489 {
11490 	return (is_guest_mode(vcpu) &&
11491 			kvm_x86_ops.guest_apic_has_interrupt &&
11492 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11493 }
11494 
11495 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11496 {
11497 	if (!list_empty_careful(&vcpu->async_pf.done))
11498 		return true;
11499 
11500 	if (kvm_apic_has_events(vcpu))
11501 		return true;
11502 
11503 	if (vcpu->arch.pv.pv_unhalted)
11504 		return true;
11505 
11506 	if (vcpu->arch.exception.pending)
11507 		return true;
11508 
11509 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11510 	    (vcpu->arch.nmi_pending &&
11511 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11512 		return true;
11513 
11514 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11515 	    (vcpu->arch.smi_pending &&
11516 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11517 		return true;
11518 
11519 	if (kvm_arch_interrupt_allowed(vcpu) &&
11520 	    (kvm_cpu_has_interrupt(vcpu) ||
11521 	    kvm_guest_apic_has_interrupt(vcpu)))
11522 		return true;
11523 
11524 	if (kvm_hv_has_stimer_pending(vcpu))
11525 		return true;
11526 
11527 	if (is_guest_mode(vcpu) &&
11528 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11529 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11530 		return true;
11531 
11532 	return false;
11533 }
11534 
11535 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11536 {
11537 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11538 }
11539 
11540 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11541 {
11542 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11543 		return true;
11544 
11545 	return false;
11546 }
11547 
11548 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11549 {
11550 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11551 		return true;
11552 
11553 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11554 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11555 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11556 		return true;
11557 
11558 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11559 }
11560 
11561 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11562 {
11563 	if (vcpu->arch.guest_state_protected)
11564 		return true;
11565 
11566 	return vcpu->arch.preempted_in_kernel;
11567 }
11568 
11569 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11570 {
11571 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11572 }
11573 
11574 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11575 {
11576 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11577 }
11578 
11579 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11580 {
11581 	/* Can't read the RIP when guest state is protected, just return 0 */
11582 	if (vcpu->arch.guest_state_protected)
11583 		return 0;
11584 
11585 	if (is_64_bit_mode(vcpu))
11586 		return kvm_rip_read(vcpu);
11587 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11588 		     kvm_rip_read(vcpu));
11589 }
11590 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11591 
11592 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11593 {
11594 	return kvm_get_linear_rip(vcpu) == linear_rip;
11595 }
11596 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11597 
11598 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11599 {
11600 	unsigned long rflags;
11601 
11602 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11603 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11604 		rflags &= ~X86_EFLAGS_TF;
11605 	return rflags;
11606 }
11607 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11608 
11609 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11610 {
11611 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11612 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11613 		rflags |= X86_EFLAGS_TF;
11614 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11615 }
11616 
11617 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11618 {
11619 	__kvm_set_rflags(vcpu, rflags);
11620 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11621 }
11622 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11623 
11624 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11625 {
11626 	int r;
11627 
11628 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11629 	      work->wakeup_all)
11630 		return;
11631 
11632 	r = kvm_mmu_reload(vcpu);
11633 	if (unlikely(r))
11634 		return;
11635 
11636 	if (!vcpu->arch.mmu->direct_map &&
11637 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11638 		return;
11639 
11640 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11641 }
11642 
11643 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11644 {
11645 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11646 
11647 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11648 }
11649 
11650 static inline u32 kvm_async_pf_next_probe(u32 key)
11651 {
11652 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11653 }
11654 
11655 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11656 {
11657 	u32 key = kvm_async_pf_hash_fn(gfn);
11658 
11659 	while (vcpu->arch.apf.gfns[key] != ~0)
11660 		key = kvm_async_pf_next_probe(key);
11661 
11662 	vcpu->arch.apf.gfns[key] = gfn;
11663 }
11664 
11665 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11666 {
11667 	int i;
11668 	u32 key = kvm_async_pf_hash_fn(gfn);
11669 
11670 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11671 		     (vcpu->arch.apf.gfns[key] != gfn &&
11672 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11673 		key = kvm_async_pf_next_probe(key);
11674 
11675 	return key;
11676 }
11677 
11678 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11679 {
11680 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11681 }
11682 
11683 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11684 {
11685 	u32 i, j, k;
11686 
11687 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11688 
11689 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11690 		return;
11691 
11692 	while (true) {
11693 		vcpu->arch.apf.gfns[i] = ~0;
11694 		do {
11695 			j = kvm_async_pf_next_probe(j);
11696 			if (vcpu->arch.apf.gfns[j] == ~0)
11697 				return;
11698 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11699 			/*
11700 			 * k lies cyclically in ]i,j]
11701 			 * |    i.k.j |
11702 			 * |....j i.k.| or  |.k..j i...|
11703 			 */
11704 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11705 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11706 		i = j;
11707 	}
11708 }
11709 
11710 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11711 {
11712 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11713 
11714 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11715 				      sizeof(reason));
11716 }
11717 
11718 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11719 {
11720 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11721 
11722 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11723 					     &token, offset, sizeof(token));
11724 }
11725 
11726 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11727 {
11728 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11729 	u32 val;
11730 
11731 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11732 					 &val, offset, sizeof(val)))
11733 		return false;
11734 
11735 	return !val;
11736 }
11737 
11738 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11739 {
11740 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11741 		return false;
11742 
11743 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11744 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11745 		return false;
11746 
11747 	return true;
11748 }
11749 
11750 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11751 {
11752 	if (unlikely(!lapic_in_kernel(vcpu) ||
11753 		     kvm_event_needs_reinjection(vcpu) ||
11754 		     vcpu->arch.exception.pending))
11755 		return false;
11756 
11757 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11758 		return false;
11759 
11760 	/*
11761 	 * If interrupts are off we cannot even use an artificial
11762 	 * halt state.
11763 	 */
11764 	return kvm_arch_interrupt_allowed(vcpu);
11765 }
11766 
11767 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11768 				     struct kvm_async_pf *work)
11769 {
11770 	struct x86_exception fault;
11771 
11772 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11773 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11774 
11775 	if (kvm_can_deliver_async_pf(vcpu) &&
11776 	    !apf_put_user_notpresent(vcpu)) {
11777 		fault.vector = PF_VECTOR;
11778 		fault.error_code_valid = true;
11779 		fault.error_code = 0;
11780 		fault.nested_page_fault = false;
11781 		fault.address = work->arch.token;
11782 		fault.async_page_fault = true;
11783 		kvm_inject_page_fault(vcpu, &fault);
11784 		return true;
11785 	} else {
11786 		/*
11787 		 * It is not possible to deliver a paravirtualized asynchronous
11788 		 * page fault, but putting the guest in an artificial halt state
11789 		 * can be beneficial nevertheless: if an interrupt arrives, we
11790 		 * can deliver it timely and perhaps the guest will schedule
11791 		 * another process.  When the instruction that triggered a page
11792 		 * fault is retried, hopefully the page will be ready in the host.
11793 		 */
11794 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11795 		return false;
11796 	}
11797 }
11798 
11799 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11800 				 struct kvm_async_pf *work)
11801 {
11802 	struct kvm_lapic_irq irq = {
11803 		.delivery_mode = APIC_DM_FIXED,
11804 		.vector = vcpu->arch.apf.vec
11805 	};
11806 
11807 	if (work->wakeup_all)
11808 		work->arch.token = ~0; /* broadcast wakeup */
11809 	else
11810 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11811 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11812 
11813 	if ((work->wakeup_all || work->notpresent_injected) &&
11814 	    kvm_pv_async_pf_enabled(vcpu) &&
11815 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11816 		vcpu->arch.apf.pageready_pending = true;
11817 		kvm_apic_set_irq(vcpu, &irq, NULL);
11818 	}
11819 
11820 	vcpu->arch.apf.halted = false;
11821 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11822 }
11823 
11824 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11825 {
11826 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11827 	if (!vcpu->arch.apf.pageready_pending)
11828 		kvm_vcpu_kick(vcpu);
11829 }
11830 
11831 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11832 {
11833 	if (!kvm_pv_async_pf_enabled(vcpu))
11834 		return true;
11835 	else
11836 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11837 }
11838 
11839 void kvm_arch_start_assignment(struct kvm *kvm)
11840 {
11841 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11842 		static_call_cond(kvm_x86_start_assignment)(kvm);
11843 }
11844 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11845 
11846 void kvm_arch_end_assignment(struct kvm *kvm)
11847 {
11848 	atomic_dec(&kvm->arch.assigned_device_count);
11849 }
11850 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11851 
11852 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11853 {
11854 	return atomic_read(&kvm->arch.assigned_device_count);
11855 }
11856 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11857 
11858 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11859 {
11860 	atomic_inc(&kvm->arch.noncoherent_dma_count);
11861 }
11862 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11863 
11864 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11865 {
11866 	atomic_dec(&kvm->arch.noncoherent_dma_count);
11867 }
11868 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11869 
11870 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11871 {
11872 	return atomic_read(&kvm->arch.noncoherent_dma_count);
11873 }
11874 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11875 
11876 bool kvm_arch_has_irq_bypass(void)
11877 {
11878 	return true;
11879 }
11880 
11881 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11882 				      struct irq_bypass_producer *prod)
11883 {
11884 	struct kvm_kernel_irqfd *irqfd =
11885 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11886 	int ret;
11887 
11888 	irqfd->producer = prod;
11889 	kvm_arch_start_assignment(irqfd->kvm);
11890 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11891 					 prod->irq, irqfd->gsi, 1);
11892 
11893 	if (ret)
11894 		kvm_arch_end_assignment(irqfd->kvm);
11895 
11896 	return ret;
11897 }
11898 
11899 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11900 				      struct irq_bypass_producer *prod)
11901 {
11902 	int ret;
11903 	struct kvm_kernel_irqfd *irqfd =
11904 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11905 
11906 	WARN_ON(irqfd->producer != prod);
11907 	irqfd->producer = NULL;
11908 
11909 	/*
11910 	 * When producer of consumer is unregistered, we change back to
11911 	 * remapped mode, so we can re-use the current implementation
11912 	 * when the irq is masked/disabled or the consumer side (KVM
11913 	 * int this case doesn't want to receive the interrupts.
11914 	*/
11915 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11916 	if (ret)
11917 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11918 		       " fails: %d\n", irqfd->consumer.token, ret);
11919 
11920 	kvm_arch_end_assignment(irqfd->kvm);
11921 }
11922 
11923 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11924 				   uint32_t guest_irq, bool set)
11925 {
11926 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11927 }
11928 
11929 bool kvm_vector_hashing_enabled(void)
11930 {
11931 	return vector_hashing;
11932 }
11933 
11934 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11935 {
11936 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11937 }
11938 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11939 
11940 
11941 int kvm_spec_ctrl_test_value(u64 value)
11942 {
11943 	/*
11944 	 * test that setting IA32_SPEC_CTRL to given value
11945 	 * is allowed by the host processor
11946 	 */
11947 
11948 	u64 saved_value;
11949 	unsigned long flags;
11950 	int ret = 0;
11951 
11952 	local_irq_save(flags);
11953 
11954 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11955 		ret = 1;
11956 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11957 		ret = 1;
11958 	else
11959 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11960 
11961 	local_irq_restore(flags);
11962 
11963 	return ret;
11964 }
11965 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11966 
11967 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11968 {
11969 	struct x86_exception fault;
11970 	u32 access = error_code &
11971 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11972 
11973 	if (!(error_code & PFERR_PRESENT_MASK) ||
11974 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11975 		/*
11976 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11977 		 * tables probably do not match the TLB.  Just proceed
11978 		 * with the error code that the processor gave.
11979 		 */
11980 		fault.vector = PF_VECTOR;
11981 		fault.error_code_valid = true;
11982 		fault.error_code = error_code;
11983 		fault.nested_page_fault = false;
11984 		fault.address = gva;
11985 	}
11986 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11987 }
11988 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11989 
11990 /*
11991  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11992  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11993  * indicates whether exit to userspace is needed.
11994  */
11995 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11996 			      struct x86_exception *e)
11997 {
11998 	if (r == X86EMUL_PROPAGATE_FAULT) {
11999 		kvm_inject_emulated_page_fault(vcpu, e);
12000 		return 1;
12001 	}
12002 
12003 	/*
12004 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12005 	 * while handling a VMX instruction KVM could've handled the request
12006 	 * correctly by exiting to userspace and performing I/O but there
12007 	 * doesn't seem to be a real use-case behind such requests, just return
12008 	 * KVM_EXIT_INTERNAL_ERROR for now.
12009 	 */
12010 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12011 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12012 	vcpu->run->internal.ndata = 0;
12013 
12014 	return 0;
12015 }
12016 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12017 
12018 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12019 {
12020 	bool pcid_enabled;
12021 	struct x86_exception e;
12022 	struct {
12023 		u64 pcid;
12024 		u64 gla;
12025 	} operand;
12026 	int r;
12027 
12028 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12029 	if (r != X86EMUL_CONTINUE)
12030 		return kvm_handle_memory_failure(vcpu, r, &e);
12031 
12032 	if (operand.pcid >> 12 != 0) {
12033 		kvm_inject_gp(vcpu, 0);
12034 		return 1;
12035 	}
12036 
12037 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12038 
12039 	switch (type) {
12040 	case INVPCID_TYPE_INDIV_ADDR:
12041 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12042 		    is_noncanonical_address(operand.gla, vcpu)) {
12043 			kvm_inject_gp(vcpu, 0);
12044 			return 1;
12045 		}
12046 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12047 		return kvm_skip_emulated_instruction(vcpu);
12048 
12049 	case INVPCID_TYPE_SINGLE_CTXT:
12050 		if (!pcid_enabled && (operand.pcid != 0)) {
12051 			kvm_inject_gp(vcpu, 0);
12052 			return 1;
12053 		}
12054 
12055 		kvm_invalidate_pcid(vcpu, operand.pcid);
12056 		return kvm_skip_emulated_instruction(vcpu);
12057 
12058 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12059 		/*
12060 		 * Currently, KVM doesn't mark global entries in the shadow
12061 		 * page tables, so a non-global flush just degenerates to a
12062 		 * global flush. If needed, we could optimize this later by
12063 		 * keeping track of global entries in shadow page tables.
12064 		 */
12065 
12066 		fallthrough;
12067 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12068 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12069 		return kvm_skip_emulated_instruction(vcpu);
12070 
12071 	default:
12072 		BUG(); /* We have already checked above that type <= 3 */
12073 	}
12074 }
12075 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12076 
12077 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12078 {
12079 	struct kvm_run *run = vcpu->run;
12080 	struct kvm_mmio_fragment *frag;
12081 	unsigned int len;
12082 
12083 	BUG_ON(!vcpu->mmio_needed);
12084 
12085 	/* Complete previous fragment */
12086 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12087 	len = min(8u, frag->len);
12088 	if (!vcpu->mmio_is_write)
12089 		memcpy(frag->data, run->mmio.data, len);
12090 
12091 	if (frag->len <= 8) {
12092 		/* Switch to the next fragment. */
12093 		frag++;
12094 		vcpu->mmio_cur_fragment++;
12095 	} else {
12096 		/* Go forward to the next mmio piece. */
12097 		frag->data += len;
12098 		frag->gpa += len;
12099 		frag->len -= len;
12100 	}
12101 
12102 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12103 		vcpu->mmio_needed = 0;
12104 
12105 		// VMG change, at this point, we're always done
12106 		// RIP has already been advanced
12107 		return 1;
12108 	}
12109 
12110 	// More MMIO is needed
12111 	run->mmio.phys_addr = frag->gpa;
12112 	run->mmio.len = min(8u, frag->len);
12113 	run->mmio.is_write = vcpu->mmio_is_write;
12114 	if (run->mmio.is_write)
12115 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12116 	run->exit_reason = KVM_EXIT_MMIO;
12117 
12118 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12119 
12120 	return 0;
12121 }
12122 
12123 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12124 			  void *data)
12125 {
12126 	int handled;
12127 	struct kvm_mmio_fragment *frag;
12128 
12129 	if (!data)
12130 		return -EINVAL;
12131 
12132 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12133 	if (handled == bytes)
12134 		return 1;
12135 
12136 	bytes -= handled;
12137 	gpa += handled;
12138 	data += handled;
12139 
12140 	/*TODO: Check if need to increment number of frags */
12141 	frag = vcpu->mmio_fragments;
12142 	vcpu->mmio_nr_fragments = 1;
12143 	frag->len = bytes;
12144 	frag->gpa = gpa;
12145 	frag->data = data;
12146 
12147 	vcpu->mmio_needed = 1;
12148 	vcpu->mmio_cur_fragment = 0;
12149 
12150 	vcpu->run->mmio.phys_addr = gpa;
12151 	vcpu->run->mmio.len = min(8u, frag->len);
12152 	vcpu->run->mmio.is_write = 1;
12153 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12154 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12155 
12156 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12157 
12158 	return 0;
12159 }
12160 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12161 
12162 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12163 			 void *data)
12164 {
12165 	int handled;
12166 	struct kvm_mmio_fragment *frag;
12167 
12168 	if (!data)
12169 		return -EINVAL;
12170 
12171 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12172 	if (handled == bytes)
12173 		return 1;
12174 
12175 	bytes -= handled;
12176 	gpa += handled;
12177 	data += handled;
12178 
12179 	/*TODO: Check if need to increment number of frags */
12180 	frag = vcpu->mmio_fragments;
12181 	vcpu->mmio_nr_fragments = 1;
12182 	frag->len = bytes;
12183 	frag->gpa = gpa;
12184 	frag->data = data;
12185 
12186 	vcpu->mmio_needed = 1;
12187 	vcpu->mmio_cur_fragment = 0;
12188 
12189 	vcpu->run->mmio.phys_addr = gpa;
12190 	vcpu->run->mmio.len = min(8u, frag->len);
12191 	vcpu->run->mmio.is_write = 0;
12192 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12193 
12194 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12195 
12196 	return 0;
12197 }
12198 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12199 
12200 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12201 			   unsigned int port);
12202 
12203 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12204 {
12205 	int size = vcpu->arch.pio.size;
12206 	int port = vcpu->arch.pio.port;
12207 
12208 	vcpu->arch.pio.count = 0;
12209 	if (vcpu->arch.sev_pio_count)
12210 		return kvm_sev_es_outs(vcpu, size, port);
12211 	return 1;
12212 }
12213 
12214 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12215 			   unsigned int port)
12216 {
12217 	for (;;) {
12218 		unsigned int count =
12219 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12220 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12221 
12222 		/* memcpy done already by emulator_pio_out.  */
12223 		vcpu->arch.sev_pio_count -= count;
12224 		vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12225 		if (!ret)
12226 			break;
12227 
12228 		/* Emulation done by the kernel.  */
12229 		if (!vcpu->arch.sev_pio_count)
12230 			return 1;
12231 	}
12232 
12233 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12234 	return 0;
12235 }
12236 
12237 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12238 			  unsigned int port);
12239 
12240 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12241 {
12242 	unsigned count = vcpu->arch.pio.count;
12243 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12244 	vcpu->arch.sev_pio_count -= count;
12245 	vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12246 }
12247 
12248 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12249 {
12250 	int size = vcpu->arch.pio.size;
12251 	int port = vcpu->arch.pio.port;
12252 
12253 	advance_sev_es_emulated_ins(vcpu);
12254 	if (vcpu->arch.sev_pio_count)
12255 		return kvm_sev_es_ins(vcpu, size, port);
12256 	return 1;
12257 }
12258 
12259 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12260 			  unsigned int port)
12261 {
12262 	for (;;) {
12263 		unsigned int count =
12264 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12265 		if (!__emulator_pio_in(vcpu, size, port, count))
12266 			break;
12267 
12268 		/* Emulation done by the kernel.  */
12269 		advance_sev_es_emulated_ins(vcpu);
12270 		if (!vcpu->arch.sev_pio_count)
12271 			return 1;
12272 	}
12273 
12274 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12275 	return 0;
12276 }
12277 
12278 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12279 			 unsigned int port, void *data,  unsigned int count,
12280 			 int in)
12281 {
12282 	vcpu->arch.sev_pio_data = data;
12283 	vcpu->arch.sev_pio_count = count;
12284 	return in ? kvm_sev_es_ins(vcpu, size, port)
12285 		  : kvm_sev_es_outs(vcpu, size, port);
12286 }
12287 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12288 
12289 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12290 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12291 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12292 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12293 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12294 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12295 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12296 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12297 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12298 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12299 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12300 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12301 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12302 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12303 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12304 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12305 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12306 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12307 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12308 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12309 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12310 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12311 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12312 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12313 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12314 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12315 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12316