xref: /linux/arch/x86/kvm/x86.c (revision e0c1b49f5b674cca7b10549c53b3791d0bbc90a8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
62 
63 #include <trace/events/kvm.h>
64 
65 #include <asm/debugreg.h>
66 #include <asm/msr.h>
67 #include <asm/desc.h>
68 #include <asm/mce.h>
69 #include <asm/pkru.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/api.h>
72 #include <asm/fpu/xcr.h>
73 #include <asm/fpu/xstate.h>
74 #include <asm/pvclock.h>
75 #include <asm/div64.h>
76 #include <asm/irq_remapping.h>
77 #include <asm/mshyperv.h>
78 #include <asm/hypervisor.h>
79 #include <asm/tlbflush.h>
80 #include <asm/intel_pt.h>
81 #include <asm/emulate_prefix.h>
82 #include <asm/sgx.h>
83 #include <clocksource/hyperv_timer.h>
84 
85 #define CREATE_TRACE_POINTS
86 #include "trace.h"
87 
88 #define MAX_IO_MSRS 256
89 #define KVM_MAX_MCE_BANKS 32
90 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
91 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
92 
93 #define emul_to_vcpu(ctxt) \
94 	((struct kvm_vcpu *)(ctxt)->vcpu)
95 
96 /* EFER defaults:
97  * - enable syscall per default because its emulated by KVM
98  * - enable LME and LMA per default on 64 bit KVM
99  */
100 #ifdef CONFIG_X86_64
101 static
102 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
103 #else
104 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 #endif
106 
107 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
108 
109 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
110 
111 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
112                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
113 
114 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
115 static void process_nmi(struct kvm_vcpu *vcpu);
116 static void process_smi(struct kvm_vcpu *vcpu);
117 static void enter_smm(struct kvm_vcpu *vcpu);
118 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
119 static void store_regs(struct kvm_vcpu *vcpu);
120 static int sync_regs(struct kvm_vcpu *vcpu);
121 
122 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
123 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
124 
125 struct kvm_x86_ops kvm_x86_ops __read_mostly;
126 EXPORT_SYMBOL_GPL(kvm_x86_ops);
127 
128 #define KVM_X86_OP(func)					     \
129 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
130 				*(((struct kvm_x86_ops *)0)->func));
131 #define KVM_X86_OP_NULL KVM_X86_OP
132 #include <asm/kvm-x86-ops.h>
133 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
134 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
135 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
136 
137 static bool __read_mostly ignore_msrs = 0;
138 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
139 
140 bool __read_mostly report_ignored_msrs = true;
141 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(report_ignored_msrs);
143 
144 unsigned int min_timer_period_us = 200;
145 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
146 
147 static bool __read_mostly kvmclock_periodic_sync = true;
148 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
149 
150 bool __read_mostly kvm_has_tsc_control;
151 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
152 u32  __read_mostly kvm_max_guest_tsc_khz;
153 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
154 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
155 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
156 u64  __read_mostly kvm_max_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
158 u64 __read_mostly kvm_default_tsc_scaling_ratio;
159 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
160 bool __read_mostly kvm_has_bus_lock_exit;
161 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
162 
163 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
164 static u32 __read_mostly tsc_tolerance_ppm = 250;
165 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
166 
167 /*
168  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
169  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
170  * advancement entirely.  Any other value is used as-is and disables adaptive
171  * tuning, i.e. allows privileged userspace to set an exact advancement time.
172  */
173 static int __read_mostly lapic_timer_advance_ns = -1;
174 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
175 
176 static bool __read_mostly vector_hashing = true;
177 module_param(vector_hashing, bool, S_IRUGO);
178 
179 bool __read_mostly enable_vmware_backdoor = false;
180 module_param(enable_vmware_backdoor, bool, S_IRUGO);
181 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
182 
183 static bool __read_mostly force_emulation_prefix = false;
184 module_param(force_emulation_prefix, bool, S_IRUGO);
185 
186 int __read_mostly pi_inject_timer = -1;
187 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
188 
189 /*
190  * Restoring the host value for MSRs that are only consumed when running in
191  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
192  * returns to userspace, i.e. the kernel can run with the guest's value.
193  */
194 #define KVM_MAX_NR_USER_RETURN_MSRS 16
195 
196 struct kvm_user_return_msrs {
197 	struct user_return_notifier urn;
198 	bool registered;
199 	struct kvm_user_return_msr_values {
200 		u64 host;
201 		u64 curr;
202 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
203 };
204 
205 u32 __read_mostly kvm_nr_uret_msrs;
206 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
207 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
208 static struct kvm_user_return_msrs __percpu *user_return_msrs;
209 
210 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
211 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
212 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
213 				| XFEATURE_MASK_PKRU)
214 
215 u64 __read_mostly host_efer;
216 EXPORT_SYMBOL_GPL(host_efer);
217 
218 bool __read_mostly allow_smaller_maxphyaddr = 0;
219 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
220 
221 bool __read_mostly enable_apicv = true;
222 EXPORT_SYMBOL_GPL(enable_apicv);
223 
224 u64 __read_mostly host_xss;
225 EXPORT_SYMBOL_GPL(host_xss);
226 u64 __read_mostly supported_xss;
227 EXPORT_SYMBOL_GPL(supported_xss);
228 
229 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
230 	KVM_GENERIC_VM_STATS(),
231 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
232 	STATS_DESC_COUNTER(VM, mmu_pte_write),
233 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
234 	STATS_DESC_COUNTER(VM, mmu_flooded),
235 	STATS_DESC_COUNTER(VM, mmu_recycled),
236 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
237 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
238 	STATS_DESC_ICOUNTER(VM, pages_4k),
239 	STATS_DESC_ICOUNTER(VM, pages_2m),
240 	STATS_DESC_ICOUNTER(VM, pages_1g),
241 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
242 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
243 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
244 };
245 
246 const struct kvm_stats_header kvm_vm_stats_header = {
247 	.name_size = KVM_STATS_NAME_SIZE,
248 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
249 	.id_offset = sizeof(struct kvm_stats_header),
250 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
251 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
252 		       sizeof(kvm_vm_stats_desc),
253 };
254 
255 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
256 	KVM_GENERIC_VCPU_STATS(),
257 	STATS_DESC_COUNTER(VCPU, pf_fixed),
258 	STATS_DESC_COUNTER(VCPU, pf_guest),
259 	STATS_DESC_COUNTER(VCPU, tlb_flush),
260 	STATS_DESC_COUNTER(VCPU, invlpg),
261 	STATS_DESC_COUNTER(VCPU, exits),
262 	STATS_DESC_COUNTER(VCPU, io_exits),
263 	STATS_DESC_COUNTER(VCPU, mmio_exits),
264 	STATS_DESC_COUNTER(VCPU, signal_exits),
265 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
266 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
267 	STATS_DESC_COUNTER(VCPU, l1d_flush),
268 	STATS_DESC_COUNTER(VCPU, halt_exits),
269 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
270 	STATS_DESC_COUNTER(VCPU, irq_exits),
271 	STATS_DESC_COUNTER(VCPU, host_state_reload),
272 	STATS_DESC_COUNTER(VCPU, fpu_reload),
273 	STATS_DESC_COUNTER(VCPU, insn_emulation),
274 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
275 	STATS_DESC_COUNTER(VCPU, hypercalls),
276 	STATS_DESC_COUNTER(VCPU, irq_injections),
277 	STATS_DESC_COUNTER(VCPU, nmi_injections),
278 	STATS_DESC_COUNTER(VCPU, req_event),
279 	STATS_DESC_COUNTER(VCPU, nested_run),
280 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
281 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
282 	STATS_DESC_ICOUNTER(VCPU, guest_mode)
283 };
284 
285 const struct kvm_stats_header kvm_vcpu_stats_header = {
286 	.name_size = KVM_STATS_NAME_SIZE,
287 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
288 	.id_offset = sizeof(struct kvm_stats_header),
289 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
290 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
291 		       sizeof(kvm_vcpu_stats_desc),
292 };
293 
294 u64 __read_mostly host_xcr0;
295 u64 __read_mostly supported_xcr0;
296 EXPORT_SYMBOL_GPL(supported_xcr0);
297 
298 static struct kmem_cache *x86_emulator_cache;
299 
300 /*
301  * When called, it means the previous get/set msr reached an invalid msr.
302  * Return true if we want to ignore/silent this failed msr access.
303  */
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
305 {
306 	const char *op = write ? "wrmsr" : "rdmsr";
307 
308 	if (ignore_msrs) {
309 		if (report_ignored_msrs)
310 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
311 				      op, msr, data);
312 		/* Mask the error */
313 		return true;
314 	} else {
315 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
316 				      op, msr, data);
317 		return false;
318 	}
319 }
320 
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
322 {
323 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 	unsigned int size = sizeof(struct x86_emulate_ctxt);
325 
326 	return kmem_cache_create_usercopy("x86_emulator", size,
327 					  __alignof__(struct x86_emulate_ctxt),
328 					  SLAB_ACCOUNT, useroffset,
329 					  size - useroffset, NULL);
330 }
331 
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
333 
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
335 {
336 	int i;
337 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 		vcpu->arch.apf.gfns[i] = ~0;
339 }
340 
341 static void kvm_on_user_return(struct user_return_notifier *urn)
342 {
343 	unsigned slot;
344 	struct kvm_user_return_msrs *msrs
345 		= container_of(urn, struct kvm_user_return_msrs, urn);
346 	struct kvm_user_return_msr_values *values;
347 	unsigned long flags;
348 
349 	/*
350 	 * Disabling irqs at this point since the following code could be
351 	 * interrupted and executed through kvm_arch_hardware_disable()
352 	 */
353 	local_irq_save(flags);
354 	if (msrs->registered) {
355 		msrs->registered = false;
356 		user_return_notifier_unregister(urn);
357 	}
358 	local_irq_restore(flags);
359 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 		values = &msrs->values[slot];
361 		if (values->host != values->curr) {
362 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 			values->curr = values->host;
364 		}
365 	}
366 }
367 
368 static int kvm_probe_user_return_msr(u32 msr)
369 {
370 	u64 val;
371 	int ret;
372 
373 	preempt_disable();
374 	ret = rdmsrl_safe(msr, &val);
375 	if (ret)
376 		goto out;
377 	ret = wrmsrl_safe(msr, val);
378 out:
379 	preempt_enable();
380 	return ret;
381 }
382 
383 int kvm_add_user_return_msr(u32 msr)
384 {
385 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
386 
387 	if (kvm_probe_user_return_msr(msr))
388 		return -1;
389 
390 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 	return kvm_nr_uret_msrs++;
392 }
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
394 
395 int kvm_find_user_return_msr(u32 msr)
396 {
397 	int i;
398 
399 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 		if (kvm_uret_msrs_list[i] == msr)
401 			return i;
402 	}
403 	return -1;
404 }
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
406 
407 static void kvm_user_return_msr_cpu_online(void)
408 {
409 	unsigned int cpu = smp_processor_id();
410 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
411 	u64 value;
412 	int i;
413 
414 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 		msrs->values[i].host = value;
417 		msrs->values[i].curr = value;
418 	}
419 }
420 
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
422 {
423 	unsigned int cpu = smp_processor_id();
424 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 	int err;
426 
427 	value = (value & mask) | (msrs->values[slot].host & ~mask);
428 	if (value == msrs->values[slot].curr)
429 		return 0;
430 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
431 	if (err)
432 		return 1;
433 
434 	msrs->values[slot].curr = value;
435 	if (!msrs->registered) {
436 		msrs->urn.on_user_return = kvm_on_user_return;
437 		user_return_notifier_register(&msrs->urn);
438 		msrs->registered = true;
439 	}
440 	return 0;
441 }
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
443 
444 static void drop_user_return_notifiers(void)
445 {
446 	unsigned int cpu = smp_processor_id();
447 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
448 
449 	if (msrs->registered)
450 		kvm_on_user_return(&msrs->urn);
451 }
452 
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
454 {
455 	return vcpu->arch.apic_base;
456 }
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
458 
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
460 {
461 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
462 }
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
464 
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
466 {
467 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
471 
472 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
473 		return 1;
474 	if (!msr_info->host_initiated) {
475 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
476 			return 1;
477 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
478 			return 1;
479 	}
480 
481 	kvm_lapic_set_base(vcpu, msr_info->data);
482 	kvm_recalculate_apic_map(vcpu->kvm);
483 	return 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
486 
487 /*
488  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
489  *
490  * Hardware virtualization extension instructions may fault if a reboot turns
491  * off virtualization while processes are running.  Usually after catching the
492  * fault we just panic; during reboot instead the instruction is ignored.
493  */
494 noinstr void kvm_spurious_fault(void)
495 {
496 	/* Fault while not rebooting.  We want the trace. */
497 	BUG_ON(!kvm_rebooting);
498 }
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
500 
501 #define EXCPT_BENIGN		0
502 #define EXCPT_CONTRIBUTORY	1
503 #define EXCPT_PF		2
504 
505 static int exception_class(int vector)
506 {
507 	switch (vector) {
508 	case PF_VECTOR:
509 		return EXCPT_PF;
510 	case DE_VECTOR:
511 	case TS_VECTOR:
512 	case NP_VECTOR:
513 	case SS_VECTOR:
514 	case GP_VECTOR:
515 		return EXCPT_CONTRIBUTORY;
516 	default:
517 		break;
518 	}
519 	return EXCPT_BENIGN;
520 }
521 
522 #define EXCPT_FAULT		0
523 #define EXCPT_TRAP		1
524 #define EXCPT_ABORT		2
525 #define EXCPT_INTERRUPT		3
526 
527 static int exception_type(int vector)
528 {
529 	unsigned int mask;
530 
531 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 		return EXCPT_INTERRUPT;
533 
534 	mask = 1 << vector;
535 
536 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
537 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
538 		return EXCPT_TRAP;
539 
540 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
541 		return EXCPT_ABORT;
542 
543 	/* Reserved exceptions will result in fault */
544 	return EXCPT_FAULT;
545 }
546 
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
548 {
549 	unsigned nr = vcpu->arch.exception.nr;
550 	bool has_payload = vcpu->arch.exception.has_payload;
551 	unsigned long payload = vcpu->arch.exception.payload;
552 
553 	if (!has_payload)
554 		return;
555 
556 	switch (nr) {
557 	case DB_VECTOR:
558 		/*
559 		 * "Certain debug exceptions may clear bit 0-3.  The
560 		 * remaining contents of the DR6 register are never
561 		 * cleared by the processor".
562 		 */
563 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
564 		/*
565 		 * In order to reflect the #DB exception payload in guest
566 		 * dr6, three components need to be considered: active low
567 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
568 		 * DR6_BS and DR6_BT)
569 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 		 * In the target guest dr6:
571 		 * FIXED_1 bits should always be set.
572 		 * Active low bits should be cleared if 1-setting in payload.
573 		 * Active high bits should be set if 1-setting in payload.
574 		 *
575 		 * Note, the payload is compatible with the pending debug
576 		 * exceptions/exit qualification under VMX, that active_low bits
577 		 * are active high in payload.
578 		 * So they need to be flipped for DR6.
579 		 */
580 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 		vcpu->arch.dr6 |= payload;
582 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
583 
584 		/*
585 		 * The #DB payload is defined as compatible with the 'pending
586 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 		 * defined in the 'pending debug exceptions' field (enabled
588 		 * breakpoint), it is reserved and must be zero in DR6.
589 		 */
590 		vcpu->arch.dr6 &= ~BIT(12);
591 		break;
592 	case PF_VECTOR:
593 		vcpu->arch.cr2 = payload;
594 		break;
595 	}
596 
597 	vcpu->arch.exception.has_payload = false;
598 	vcpu->arch.exception.payload = 0;
599 }
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
601 
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 		unsigned nr, bool has_error, u32 error_code,
604 	        bool has_payload, unsigned long payload, bool reinject)
605 {
606 	u32 prev_nr;
607 	int class1, class2;
608 
609 	kvm_make_request(KVM_REQ_EVENT, vcpu);
610 
611 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
612 	queue:
613 		if (reinject) {
614 			/*
615 			 * On vmentry, vcpu->arch.exception.pending is only
616 			 * true if an event injection was blocked by
617 			 * nested_run_pending.  In that case, however,
618 			 * vcpu_enter_guest requests an immediate exit,
619 			 * and the guest shouldn't proceed far enough to
620 			 * need reinjection.
621 			 */
622 			WARN_ON_ONCE(vcpu->arch.exception.pending);
623 			vcpu->arch.exception.injected = true;
624 			if (WARN_ON_ONCE(has_payload)) {
625 				/*
626 				 * A reinjected event has already
627 				 * delivered its payload.
628 				 */
629 				has_payload = false;
630 				payload = 0;
631 			}
632 		} else {
633 			vcpu->arch.exception.pending = true;
634 			vcpu->arch.exception.injected = false;
635 		}
636 		vcpu->arch.exception.has_error_code = has_error;
637 		vcpu->arch.exception.nr = nr;
638 		vcpu->arch.exception.error_code = error_code;
639 		vcpu->arch.exception.has_payload = has_payload;
640 		vcpu->arch.exception.payload = payload;
641 		if (!is_guest_mode(vcpu))
642 			kvm_deliver_exception_payload(vcpu);
643 		return;
644 	}
645 
646 	/* to check exception */
647 	prev_nr = vcpu->arch.exception.nr;
648 	if (prev_nr == DF_VECTOR) {
649 		/* triple fault -> shutdown */
650 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
651 		return;
652 	}
653 	class1 = exception_class(prev_nr);
654 	class2 = exception_class(nr);
655 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
657 		/*
658 		 * Generate double fault per SDM Table 5-5.  Set
659 		 * exception.pending = true so that the double fault
660 		 * can trigger a nested vmexit.
661 		 */
662 		vcpu->arch.exception.pending = true;
663 		vcpu->arch.exception.injected = false;
664 		vcpu->arch.exception.has_error_code = true;
665 		vcpu->arch.exception.nr = DF_VECTOR;
666 		vcpu->arch.exception.error_code = 0;
667 		vcpu->arch.exception.has_payload = false;
668 		vcpu->arch.exception.payload = 0;
669 	} else
670 		/* replace previous exception with a new one in a hope
671 		   that instruction re-execution will regenerate lost
672 		   exception */
673 		goto queue;
674 }
675 
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
677 {
678 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
679 }
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
681 
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
683 {
684 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
685 }
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
687 
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 			   unsigned long payload)
690 {
691 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
692 }
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
694 
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 				    u32 error_code, unsigned long payload)
697 {
698 	kvm_multiple_exception(vcpu, nr, true, error_code,
699 			       true, payload, false);
700 }
701 
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
703 {
704 	if (err)
705 		kvm_inject_gp(vcpu, 0);
706 	else
707 		return kvm_skip_emulated_instruction(vcpu);
708 
709 	return 1;
710 }
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
712 
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
714 {
715 	++vcpu->stat.pf_guest;
716 	vcpu->arch.exception.nested_apf =
717 		is_guest_mode(vcpu) && fault->async_page_fault;
718 	if (vcpu->arch.exception.nested_apf) {
719 		vcpu->arch.apf.nested_apf_token = fault->address;
720 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
721 	} else {
722 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
723 					fault->address);
724 	}
725 }
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
727 
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 				    struct x86_exception *fault)
730 {
731 	struct kvm_mmu *fault_mmu;
732 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
733 
734 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
735 					       vcpu->arch.walk_mmu;
736 
737 	/*
738 	 * Invalidate the TLB entry for the faulting address, if it exists,
739 	 * else the access will fault indefinitely (and to emulate hardware).
740 	 */
741 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 	    !(fault->error_code & PFERR_RSVD_MASK))
743 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 				       fault_mmu->root_hpa);
745 
746 	fault_mmu->inject_page_fault(vcpu, fault);
747 	return fault->nested_page_fault;
748 }
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
750 
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
752 {
753 	atomic_inc(&vcpu->arch.nmi_queued);
754 	kvm_make_request(KVM_REQ_NMI, vcpu);
755 }
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
757 
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
759 {
760 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
761 }
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
763 
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
765 {
766 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
767 }
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
769 
770 /*
771  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
772  * a #GP and return false.
773  */
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
775 {
776 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
777 		return true;
778 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
779 	return false;
780 }
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
782 
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
784 {
785 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 		return true;
787 
788 	kvm_queue_exception(vcpu, UD_VECTOR);
789 	return false;
790 }
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
792 
793 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
794 {
795 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
796 }
797 
798 /*
799  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
800  */
801 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
802 {
803 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
804 	gpa_t real_gpa;
805 	int i;
806 	int ret;
807 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
808 
809 	/*
810 	 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
811 	 * to an L1 GPA.
812 	 */
813 	real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(pdpt_gfn),
814 				      PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
815 	if (real_gpa == UNMAPPED_GVA)
816 		return 0;
817 
818 	/* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
819 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
820 				       cr3 & GENMASK(11, 5), sizeof(pdpte));
821 	if (ret < 0)
822 		return 0;
823 
824 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
825 		if ((pdpte[i] & PT_PRESENT_MASK) &&
826 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
827 			return 0;
828 		}
829 	}
830 
831 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
832 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
833 	vcpu->arch.pdptrs_from_userspace = false;
834 
835 	return 1;
836 }
837 EXPORT_SYMBOL_GPL(load_pdptrs);
838 
839 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
840 {
841 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
842 		kvm_clear_async_pf_completion_queue(vcpu);
843 		kvm_async_pf_hash_reset(vcpu);
844 	}
845 
846 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
847 		kvm_mmu_reset_context(vcpu);
848 
849 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
850 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
851 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
852 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
853 }
854 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
855 
856 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
857 {
858 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
859 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
860 
861 	cr0 |= X86_CR0_ET;
862 
863 #ifdef CONFIG_X86_64
864 	if (cr0 & 0xffffffff00000000UL)
865 		return 1;
866 #endif
867 
868 	cr0 &= ~CR0_RESERVED_BITS;
869 
870 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
871 		return 1;
872 
873 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
874 		return 1;
875 
876 #ifdef CONFIG_X86_64
877 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
878 	    (cr0 & X86_CR0_PG)) {
879 		int cs_db, cs_l;
880 
881 		if (!is_pae(vcpu))
882 			return 1;
883 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
884 		if (cs_l)
885 			return 1;
886 	}
887 #endif
888 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
889 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
890 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
891 		return 1;
892 
893 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
894 		return 1;
895 
896 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
897 
898 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
899 
900 	return 0;
901 }
902 EXPORT_SYMBOL_GPL(kvm_set_cr0);
903 
904 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
905 {
906 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
907 }
908 EXPORT_SYMBOL_GPL(kvm_lmsw);
909 
910 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
911 {
912 	if (vcpu->arch.guest_state_protected)
913 		return;
914 
915 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
916 
917 		if (vcpu->arch.xcr0 != host_xcr0)
918 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
919 
920 		if (vcpu->arch.xsaves_enabled &&
921 		    vcpu->arch.ia32_xss != host_xss)
922 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
923 	}
924 
925 	if (static_cpu_has(X86_FEATURE_PKU) &&
926 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
927 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
928 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
929 		write_pkru(vcpu->arch.pkru);
930 }
931 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
932 
933 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
934 {
935 	if (vcpu->arch.guest_state_protected)
936 		return;
937 
938 	if (static_cpu_has(X86_FEATURE_PKU) &&
939 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
940 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
941 		vcpu->arch.pkru = rdpkru();
942 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
943 			write_pkru(vcpu->arch.host_pkru);
944 	}
945 
946 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
947 
948 		if (vcpu->arch.xcr0 != host_xcr0)
949 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
950 
951 		if (vcpu->arch.xsaves_enabled &&
952 		    vcpu->arch.ia32_xss != host_xss)
953 			wrmsrl(MSR_IA32_XSS, host_xss);
954 	}
955 
956 }
957 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
958 
959 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
960 {
961 	u64 xcr0 = xcr;
962 	u64 old_xcr0 = vcpu->arch.xcr0;
963 	u64 valid_bits;
964 
965 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
966 	if (index != XCR_XFEATURE_ENABLED_MASK)
967 		return 1;
968 	if (!(xcr0 & XFEATURE_MASK_FP))
969 		return 1;
970 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
971 		return 1;
972 
973 	/*
974 	 * Do not allow the guest to set bits that we do not support
975 	 * saving.  However, xcr0 bit 0 is always set, even if the
976 	 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
977 	 */
978 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
979 	if (xcr0 & ~valid_bits)
980 		return 1;
981 
982 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
983 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
984 		return 1;
985 
986 	if (xcr0 & XFEATURE_MASK_AVX512) {
987 		if (!(xcr0 & XFEATURE_MASK_YMM))
988 			return 1;
989 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
990 			return 1;
991 	}
992 	vcpu->arch.xcr0 = xcr0;
993 
994 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
995 		kvm_update_cpuid_runtime(vcpu);
996 	return 0;
997 }
998 
999 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1000 {
1001 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1002 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1003 		kvm_inject_gp(vcpu, 0);
1004 		return 1;
1005 	}
1006 
1007 	return kvm_skip_emulated_instruction(vcpu);
1008 }
1009 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1010 
1011 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1012 {
1013 	if (cr4 & cr4_reserved_bits)
1014 		return false;
1015 
1016 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1017 		return false;
1018 
1019 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1022 
1023 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1024 {
1025 	/*
1026 	 * If any role bit is changed, the MMU needs to be reset.
1027 	 *
1028 	 * If CR4.PCIDE is changed 1 -> 0, the guest TLB must be flushed.
1029 	 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1030 	 * according to the SDM; however, stale prev_roots could be reused
1031 	 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1032 	 * free them all.  KVM_REQ_MMU_RELOAD is fit for the both cases; it
1033 	 * is slow, but changing CR4.PCIDE is a rare case.
1034 	 *
1035 	 * If CR4.PGE is changed, the guest TLB must be flushed.
1036 	 *
1037 	 * Note: resetting MMU is a superset of KVM_REQ_MMU_RELOAD and
1038 	 * KVM_REQ_MMU_RELOAD is a superset of KVM_REQ_TLB_FLUSH_GUEST, hence
1039 	 * the usage of "else if".
1040 	 */
1041 	if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1042 		kvm_mmu_reset_context(vcpu);
1043 	else if ((cr4 ^ old_cr4) & X86_CR4_PCIDE)
1044 		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
1045 	else if ((cr4 ^ old_cr4) & X86_CR4_PGE)
1046 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1049 
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1051 {
1052 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1054 				   X86_CR4_SMEP;
1055 
1056 	if (!kvm_is_valid_cr4(vcpu, cr4))
1057 		return 1;
1058 
1059 	if (is_long_mode(vcpu)) {
1060 		if (!(cr4 & X86_CR4_PAE))
1061 			return 1;
1062 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1063 			return 1;
1064 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1066 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 				   kvm_read_cr3(vcpu)))
1068 		return 1;
1069 
1070 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1072 			return 1;
1073 
1074 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1076 			return 1;
1077 	}
1078 
1079 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1080 
1081 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1082 
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1086 
1087 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1088 {
1089 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1090 	unsigned long roots_to_free = 0;
1091 	int i;
1092 
1093 	/*
1094 	 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1095 	 * this is reachable when running EPT=1 and unrestricted_guest=0,  and
1096 	 * also via the emulator.  KVM's TDP page tables are not in the scope of
1097 	 * the invalidation, but the guest's TLB entries need to be flushed as
1098 	 * the CPU may have cached entries in its TLB for the target PCID.
1099 	 */
1100 	if (unlikely(tdp_enabled)) {
1101 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1102 		return;
1103 	}
1104 
1105 	/*
1106 	 * If neither the current CR3 nor any of the prev_roots use the given
1107 	 * PCID, then nothing needs to be done here because a resync will
1108 	 * happen anyway before switching to any other CR3.
1109 	 */
1110 	if (kvm_get_active_pcid(vcpu) == pcid) {
1111 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1112 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1113 	}
1114 
1115 	/*
1116 	 * If PCID is disabled, there is no need to free prev_roots even if the
1117 	 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1118 	 * with PCIDE=0.
1119 	 */
1120 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1121 		return;
1122 
1123 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1124 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1125 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1126 
1127 	kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1128 }
1129 
1130 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1131 {
1132 	bool skip_tlb_flush = false;
1133 	unsigned long pcid = 0;
1134 #ifdef CONFIG_X86_64
1135 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1136 
1137 	if (pcid_enabled) {
1138 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1139 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1140 		pcid = cr3 & X86_CR3_PCID_MASK;
1141 	}
1142 #endif
1143 
1144 	/* PDPTRs are always reloaded for PAE paging. */
1145 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1146 		goto handle_tlb_flush;
1147 
1148 	/*
1149 	 * Do not condition the GPA check on long mode, this helper is used to
1150 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1151 	 * the current vCPU mode is accurate.
1152 	 */
1153 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1154 		return 1;
1155 
1156 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1157 		return 1;
1158 
1159 	if (cr3 != kvm_read_cr3(vcpu))
1160 		kvm_mmu_new_pgd(vcpu, cr3);
1161 
1162 	vcpu->arch.cr3 = cr3;
1163 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1164 
1165 handle_tlb_flush:
1166 	/*
1167 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1168 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1169 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1170 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1171 	 * i.e. only PCID=0 can be relevant.
1172 	 */
1173 	if (!skip_tlb_flush)
1174 		kvm_invalidate_pcid(vcpu, pcid);
1175 
1176 	return 0;
1177 }
1178 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1179 
1180 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1181 {
1182 	if (cr8 & CR8_RESERVED_BITS)
1183 		return 1;
1184 	if (lapic_in_kernel(vcpu))
1185 		kvm_lapic_set_tpr(vcpu, cr8);
1186 	else
1187 		vcpu->arch.cr8 = cr8;
1188 	return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1191 
1192 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1193 {
1194 	if (lapic_in_kernel(vcpu))
1195 		return kvm_lapic_get_cr8(vcpu);
1196 	else
1197 		return vcpu->arch.cr8;
1198 }
1199 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1200 
1201 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1202 {
1203 	int i;
1204 
1205 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1206 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1207 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1208 	}
1209 }
1210 
1211 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1212 {
1213 	unsigned long dr7;
1214 
1215 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1216 		dr7 = vcpu->arch.guest_debug_dr7;
1217 	else
1218 		dr7 = vcpu->arch.dr7;
1219 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1220 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1221 	if (dr7 & DR7_BP_EN_MASK)
1222 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1225 
1226 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1227 {
1228 	u64 fixed = DR6_FIXED_1;
1229 
1230 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1231 		fixed |= DR6_RTM;
1232 
1233 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1234 		fixed |= DR6_BUS_LOCK;
1235 	return fixed;
1236 }
1237 
1238 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1239 {
1240 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1241 
1242 	switch (dr) {
1243 	case 0 ... 3:
1244 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1245 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1246 			vcpu->arch.eff_db[dr] = val;
1247 		break;
1248 	case 4:
1249 	case 6:
1250 		if (!kvm_dr6_valid(val))
1251 			return 1; /* #GP */
1252 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1253 		break;
1254 	case 5:
1255 	default: /* 7 */
1256 		if (!kvm_dr7_valid(val))
1257 			return 1; /* #GP */
1258 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1259 		kvm_update_dr7(vcpu);
1260 		break;
1261 	}
1262 
1263 	return 0;
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_set_dr);
1266 
1267 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1268 {
1269 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1270 
1271 	switch (dr) {
1272 	case 0 ... 3:
1273 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1274 		break;
1275 	case 4:
1276 	case 6:
1277 		*val = vcpu->arch.dr6;
1278 		break;
1279 	case 5:
1280 	default: /* 7 */
1281 		*val = vcpu->arch.dr7;
1282 		break;
1283 	}
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_get_dr);
1286 
1287 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1288 {
1289 	u32 ecx = kvm_rcx_read(vcpu);
1290 	u64 data;
1291 
1292 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1293 		kvm_inject_gp(vcpu, 0);
1294 		return 1;
1295 	}
1296 
1297 	kvm_rax_write(vcpu, (u32)data);
1298 	kvm_rdx_write(vcpu, data >> 32);
1299 	return kvm_skip_emulated_instruction(vcpu);
1300 }
1301 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1302 
1303 /*
1304  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1305  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1306  *
1307  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1308  * extract the supported MSRs from the related const lists.
1309  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1310  * capabilities of the host cpu. This capabilities test skips MSRs that are
1311  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1312  * may depend on host virtualization features rather than host cpu features.
1313  */
1314 
1315 static const u32 msrs_to_save_all[] = {
1316 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1317 	MSR_STAR,
1318 #ifdef CONFIG_X86_64
1319 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1320 #endif
1321 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1322 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1323 	MSR_IA32_SPEC_CTRL,
1324 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1325 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1326 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1327 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1328 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1329 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1330 	MSR_IA32_UMWAIT_CONTROL,
1331 
1332 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1333 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1334 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1335 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1336 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1337 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1338 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1339 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1340 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1341 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1342 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1343 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1344 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1345 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1346 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1347 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1348 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1349 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1350 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1351 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1352 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1353 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1354 
1355 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1356 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1357 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1358 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1359 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1360 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1361 };
1362 
1363 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1364 static unsigned num_msrs_to_save;
1365 
1366 static const u32 emulated_msrs_all[] = {
1367 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1368 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1369 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1370 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1371 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1372 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1373 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1374 	HV_X64_MSR_RESET,
1375 	HV_X64_MSR_VP_INDEX,
1376 	HV_X64_MSR_VP_RUNTIME,
1377 	HV_X64_MSR_SCONTROL,
1378 	HV_X64_MSR_STIMER0_CONFIG,
1379 	HV_X64_MSR_VP_ASSIST_PAGE,
1380 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1381 	HV_X64_MSR_TSC_EMULATION_STATUS,
1382 	HV_X64_MSR_SYNDBG_OPTIONS,
1383 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1384 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1385 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1386 
1387 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1388 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1389 
1390 	MSR_IA32_TSC_ADJUST,
1391 	MSR_IA32_TSC_DEADLINE,
1392 	MSR_IA32_ARCH_CAPABILITIES,
1393 	MSR_IA32_PERF_CAPABILITIES,
1394 	MSR_IA32_MISC_ENABLE,
1395 	MSR_IA32_MCG_STATUS,
1396 	MSR_IA32_MCG_CTL,
1397 	MSR_IA32_MCG_EXT_CTL,
1398 	MSR_IA32_SMBASE,
1399 	MSR_SMI_COUNT,
1400 	MSR_PLATFORM_INFO,
1401 	MSR_MISC_FEATURES_ENABLES,
1402 	MSR_AMD64_VIRT_SPEC_CTRL,
1403 	MSR_AMD64_TSC_RATIO,
1404 	MSR_IA32_POWER_CTL,
1405 	MSR_IA32_UCODE_REV,
1406 
1407 	/*
1408 	 * The following list leaves out MSRs whose values are determined
1409 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1410 	 * We always support the "true" VMX control MSRs, even if the host
1411 	 * processor does not, so I am putting these registers here rather
1412 	 * than in msrs_to_save_all.
1413 	 */
1414 	MSR_IA32_VMX_BASIC,
1415 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1416 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1417 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1418 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1419 	MSR_IA32_VMX_MISC,
1420 	MSR_IA32_VMX_CR0_FIXED0,
1421 	MSR_IA32_VMX_CR4_FIXED0,
1422 	MSR_IA32_VMX_VMCS_ENUM,
1423 	MSR_IA32_VMX_PROCBASED_CTLS2,
1424 	MSR_IA32_VMX_EPT_VPID_CAP,
1425 	MSR_IA32_VMX_VMFUNC,
1426 
1427 	MSR_K7_HWCR,
1428 	MSR_KVM_POLL_CONTROL,
1429 };
1430 
1431 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1432 static unsigned num_emulated_msrs;
1433 
1434 /*
1435  * List of msr numbers which are used to expose MSR-based features that
1436  * can be used by a hypervisor to validate requested CPU features.
1437  */
1438 static const u32 msr_based_features_all[] = {
1439 	MSR_IA32_VMX_BASIC,
1440 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1441 	MSR_IA32_VMX_PINBASED_CTLS,
1442 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1443 	MSR_IA32_VMX_PROCBASED_CTLS,
1444 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1445 	MSR_IA32_VMX_EXIT_CTLS,
1446 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1447 	MSR_IA32_VMX_ENTRY_CTLS,
1448 	MSR_IA32_VMX_MISC,
1449 	MSR_IA32_VMX_CR0_FIXED0,
1450 	MSR_IA32_VMX_CR0_FIXED1,
1451 	MSR_IA32_VMX_CR4_FIXED0,
1452 	MSR_IA32_VMX_CR4_FIXED1,
1453 	MSR_IA32_VMX_VMCS_ENUM,
1454 	MSR_IA32_VMX_PROCBASED_CTLS2,
1455 	MSR_IA32_VMX_EPT_VPID_CAP,
1456 	MSR_IA32_VMX_VMFUNC,
1457 
1458 	MSR_F10H_DECFG,
1459 	MSR_IA32_UCODE_REV,
1460 	MSR_IA32_ARCH_CAPABILITIES,
1461 	MSR_IA32_PERF_CAPABILITIES,
1462 };
1463 
1464 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1465 static unsigned int num_msr_based_features;
1466 
1467 static u64 kvm_get_arch_capabilities(void)
1468 {
1469 	u64 data = 0;
1470 
1471 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1472 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1473 
1474 	/*
1475 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1476 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1477 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1478 	 * L1 guests, so it need not worry about its own (L2) guests.
1479 	 */
1480 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1481 
1482 	/*
1483 	 * If we're doing cache flushes (either "always" or "cond")
1484 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1485 	 * If an outer hypervisor is doing the cache flush for us
1486 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1487 	 * capability to the guest too, and if EPT is disabled we're not
1488 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1489 	 * require a nested hypervisor to do a flush of its own.
1490 	 */
1491 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1492 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1493 
1494 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1495 		data |= ARCH_CAP_RDCL_NO;
1496 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1497 		data |= ARCH_CAP_SSB_NO;
1498 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1499 		data |= ARCH_CAP_MDS_NO;
1500 
1501 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1502 		/*
1503 		 * If RTM=0 because the kernel has disabled TSX, the host might
1504 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1505 		 * and therefore knows that there cannot be TAA) but keep
1506 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1507 		 * and we want to allow migrating those guests to tsx=off hosts.
1508 		 */
1509 		data &= ~ARCH_CAP_TAA_NO;
1510 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1511 		data |= ARCH_CAP_TAA_NO;
1512 	} else {
1513 		/*
1514 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1515 		 * host so the guest can choose between disabling TSX or
1516 		 * using VERW to clear CPU buffers.
1517 		 */
1518 	}
1519 
1520 	return data;
1521 }
1522 
1523 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1524 {
1525 	switch (msr->index) {
1526 	case MSR_IA32_ARCH_CAPABILITIES:
1527 		msr->data = kvm_get_arch_capabilities();
1528 		break;
1529 	case MSR_IA32_UCODE_REV:
1530 		rdmsrl_safe(msr->index, &msr->data);
1531 		break;
1532 	default:
1533 		return static_call(kvm_x86_get_msr_feature)(msr);
1534 	}
1535 	return 0;
1536 }
1537 
1538 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1539 {
1540 	struct kvm_msr_entry msr;
1541 	int r;
1542 
1543 	msr.index = index;
1544 	r = kvm_get_msr_feature(&msr);
1545 
1546 	if (r == KVM_MSR_RET_INVALID) {
1547 		/* Unconditionally clear the output for simplicity */
1548 		*data = 0;
1549 		if (kvm_msr_ignored_check(index, 0, false))
1550 			r = 0;
1551 	}
1552 
1553 	if (r)
1554 		return r;
1555 
1556 	*data = msr.data;
1557 
1558 	return 0;
1559 }
1560 
1561 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1562 {
1563 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1564 		return false;
1565 
1566 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1567 		return false;
1568 
1569 	if (efer & (EFER_LME | EFER_LMA) &&
1570 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1571 		return false;
1572 
1573 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1574 		return false;
1575 
1576 	return true;
1577 
1578 }
1579 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1580 {
1581 	if (efer & efer_reserved_bits)
1582 		return false;
1583 
1584 	return __kvm_valid_efer(vcpu, efer);
1585 }
1586 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1587 
1588 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1589 {
1590 	u64 old_efer = vcpu->arch.efer;
1591 	u64 efer = msr_info->data;
1592 	int r;
1593 
1594 	if (efer & efer_reserved_bits)
1595 		return 1;
1596 
1597 	if (!msr_info->host_initiated) {
1598 		if (!__kvm_valid_efer(vcpu, efer))
1599 			return 1;
1600 
1601 		if (is_paging(vcpu) &&
1602 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1603 			return 1;
1604 	}
1605 
1606 	efer &= ~EFER_LMA;
1607 	efer |= vcpu->arch.efer & EFER_LMA;
1608 
1609 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1610 	if (r) {
1611 		WARN_ON(r > 0);
1612 		return r;
1613 	}
1614 
1615 	/* Update reserved bits */
1616 	if ((efer ^ old_efer) & EFER_NX)
1617 		kvm_mmu_reset_context(vcpu);
1618 
1619 	return 0;
1620 }
1621 
1622 void kvm_enable_efer_bits(u64 mask)
1623 {
1624        efer_reserved_bits &= ~mask;
1625 }
1626 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1627 
1628 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1629 {
1630 	struct kvm_x86_msr_filter *msr_filter;
1631 	struct msr_bitmap_range *ranges;
1632 	struct kvm *kvm = vcpu->kvm;
1633 	bool allowed;
1634 	int idx;
1635 	u32 i;
1636 
1637 	/* x2APIC MSRs do not support filtering. */
1638 	if (index >= 0x800 && index <= 0x8ff)
1639 		return true;
1640 
1641 	idx = srcu_read_lock(&kvm->srcu);
1642 
1643 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1644 	if (!msr_filter) {
1645 		allowed = true;
1646 		goto out;
1647 	}
1648 
1649 	allowed = msr_filter->default_allow;
1650 	ranges = msr_filter->ranges;
1651 
1652 	for (i = 0; i < msr_filter->count; i++) {
1653 		u32 start = ranges[i].base;
1654 		u32 end = start + ranges[i].nmsrs;
1655 		u32 flags = ranges[i].flags;
1656 		unsigned long *bitmap = ranges[i].bitmap;
1657 
1658 		if ((index >= start) && (index < end) && (flags & type)) {
1659 			allowed = !!test_bit(index - start, bitmap);
1660 			break;
1661 		}
1662 	}
1663 
1664 out:
1665 	srcu_read_unlock(&kvm->srcu, idx);
1666 
1667 	return allowed;
1668 }
1669 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1670 
1671 /*
1672  * Write @data into the MSR specified by @index.  Select MSR specific fault
1673  * checks are bypassed if @host_initiated is %true.
1674  * Returns 0 on success, non-0 otherwise.
1675  * Assumes vcpu_load() was already called.
1676  */
1677 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1678 			 bool host_initiated)
1679 {
1680 	struct msr_data msr;
1681 
1682 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1683 		return KVM_MSR_RET_FILTERED;
1684 
1685 	switch (index) {
1686 	case MSR_FS_BASE:
1687 	case MSR_GS_BASE:
1688 	case MSR_KERNEL_GS_BASE:
1689 	case MSR_CSTAR:
1690 	case MSR_LSTAR:
1691 		if (is_noncanonical_address(data, vcpu))
1692 			return 1;
1693 		break;
1694 	case MSR_IA32_SYSENTER_EIP:
1695 	case MSR_IA32_SYSENTER_ESP:
1696 		/*
1697 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1698 		 * non-canonical address is written on Intel but not on
1699 		 * AMD (which ignores the top 32-bits, because it does
1700 		 * not implement 64-bit SYSENTER).
1701 		 *
1702 		 * 64-bit code should hence be able to write a non-canonical
1703 		 * value on AMD.  Making the address canonical ensures that
1704 		 * vmentry does not fail on Intel after writing a non-canonical
1705 		 * value, and that something deterministic happens if the guest
1706 		 * invokes 64-bit SYSENTER.
1707 		 */
1708 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1709 		break;
1710 	case MSR_TSC_AUX:
1711 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1712 			return 1;
1713 
1714 		if (!host_initiated &&
1715 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1716 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1717 			return 1;
1718 
1719 		/*
1720 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1721 		 * incomplete and conflicting architectural behavior.  Current
1722 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1723 		 * reserved and always read as zeros.  Enforce Intel's reserved
1724 		 * bits check if and only if the guest CPU is Intel, and clear
1725 		 * the bits in all other cases.  This ensures cross-vendor
1726 		 * migration will provide consistent behavior for the guest.
1727 		 */
1728 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1729 			return 1;
1730 
1731 		data = (u32)data;
1732 		break;
1733 	}
1734 
1735 	msr.data = data;
1736 	msr.index = index;
1737 	msr.host_initiated = host_initiated;
1738 
1739 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1740 }
1741 
1742 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1743 				     u32 index, u64 data, bool host_initiated)
1744 {
1745 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1746 
1747 	if (ret == KVM_MSR_RET_INVALID)
1748 		if (kvm_msr_ignored_check(index, data, true))
1749 			ret = 0;
1750 
1751 	return ret;
1752 }
1753 
1754 /*
1755  * Read the MSR specified by @index into @data.  Select MSR specific fault
1756  * checks are bypassed if @host_initiated is %true.
1757  * Returns 0 on success, non-0 otherwise.
1758  * Assumes vcpu_load() was already called.
1759  */
1760 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1761 		  bool host_initiated)
1762 {
1763 	struct msr_data msr;
1764 	int ret;
1765 
1766 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1767 		return KVM_MSR_RET_FILTERED;
1768 
1769 	switch (index) {
1770 	case MSR_TSC_AUX:
1771 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1772 			return 1;
1773 
1774 		if (!host_initiated &&
1775 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1776 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1777 			return 1;
1778 		break;
1779 	}
1780 
1781 	msr.index = index;
1782 	msr.host_initiated = host_initiated;
1783 
1784 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1785 	if (!ret)
1786 		*data = msr.data;
1787 	return ret;
1788 }
1789 
1790 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1791 				     u32 index, u64 *data, bool host_initiated)
1792 {
1793 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1794 
1795 	if (ret == KVM_MSR_RET_INVALID) {
1796 		/* Unconditionally clear *data for simplicity */
1797 		*data = 0;
1798 		if (kvm_msr_ignored_check(index, 0, false))
1799 			ret = 0;
1800 	}
1801 
1802 	return ret;
1803 }
1804 
1805 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1806 {
1807 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1808 }
1809 EXPORT_SYMBOL_GPL(kvm_get_msr);
1810 
1811 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1812 {
1813 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1814 }
1815 EXPORT_SYMBOL_GPL(kvm_set_msr);
1816 
1817 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1818 {
1819 	int err = vcpu->run->msr.error;
1820 	if (!err) {
1821 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1822 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1823 	}
1824 
1825 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1826 }
1827 
1828 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1829 {
1830 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1831 }
1832 
1833 static u64 kvm_msr_reason(int r)
1834 {
1835 	switch (r) {
1836 	case KVM_MSR_RET_INVALID:
1837 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1838 	case KVM_MSR_RET_FILTERED:
1839 		return KVM_MSR_EXIT_REASON_FILTER;
1840 	default:
1841 		return KVM_MSR_EXIT_REASON_INVAL;
1842 	}
1843 }
1844 
1845 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1846 			      u32 exit_reason, u64 data,
1847 			      int (*completion)(struct kvm_vcpu *vcpu),
1848 			      int r)
1849 {
1850 	u64 msr_reason = kvm_msr_reason(r);
1851 
1852 	/* Check if the user wanted to know about this MSR fault */
1853 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1854 		return 0;
1855 
1856 	vcpu->run->exit_reason = exit_reason;
1857 	vcpu->run->msr.error = 0;
1858 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1859 	vcpu->run->msr.reason = msr_reason;
1860 	vcpu->run->msr.index = index;
1861 	vcpu->run->msr.data = data;
1862 	vcpu->arch.complete_userspace_io = completion;
1863 
1864 	return 1;
1865 }
1866 
1867 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1868 {
1869 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1870 				   complete_emulated_rdmsr, r);
1871 }
1872 
1873 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1874 {
1875 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1876 				   complete_emulated_wrmsr, r);
1877 }
1878 
1879 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1880 {
1881 	u32 ecx = kvm_rcx_read(vcpu);
1882 	u64 data;
1883 	int r;
1884 
1885 	r = kvm_get_msr(vcpu, ecx, &data);
1886 
1887 	/* MSR read failed? See if we should ask user space */
1888 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1889 		/* Bounce to user space */
1890 		return 0;
1891 	}
1892 
1893 	if (!r) {
1894 		trace_kvm_msr_read(ecx, data);
1895 
1896 		kvm_rax_write(vcpu, data & -1u);
1897 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1898 	} else {
1899 		trace_kvm_msr_read_ex(ecx);
1900 	}
1901 
1902 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1905 
1906 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1907 {
1908 	u32 ecx = kvm_rcx_read(vcpu);
1909 	u64 data = kvm_read_edx_eax(vcpu);
1910 	int r;
1911 
1912 	r = kvm_set_msr(vcpu, ecx, data);
1913 
1914 	/* MSR write failed? See if we should ask user space */
1915 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1916 		/* Bounce to user space */
1917 		return 0;
1918 
1919 	/* Signal all other negative errors to userspace */
1920 	if (r < 0)
1921 		return r;
1922 
1923 	if (!r)
1924 		trace_kvm_msr_write(ecx, data);
1925 	else
1926 		trace_kvm_msr_write_ex(ecx, data);
1927 
1928 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1929 }
1930 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1931 
1932 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1933 {
1934 	return kvm_skip_emulated_instruction(vcpu);
1935 }
1936 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1937 
1938 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1939 {
1940 	/* Treat an INVD instruction as a NOP and just skip it. */
1941 	return kvm_emulate_as_nop(vcpu);
1942 }
1943 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1944 
1945 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1946 {
1947 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1948 	return kvm_emulate_as_nop(vcpu);
1949 }
1950 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1951 
1952 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1953 {
1954 	kvm_queue_exception(vcpu, UD_VECTOR);
1955 	return 1;
1956 }
1957 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1958 
1959 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1960 {
1961 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1962 	return kvm_emulate_as_nop(vcpu);
1963 }
1964 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1965 
1966 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1967 {
1968 	xfer_to_guest_mode_prepare();
1969 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1970 		xfer_to_guest_mode_work_pending();
1971 }
1972 
1973 /*
1974  * The fast path for frequent and performance sensitive wrmsr emulation,
1975  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1976  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1977  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1978  * other cases which must be called after interrupts are enabled on the host.
1979  */
1980 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1981 {
1982 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1983 		return 1;
1984 
1985 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1986 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1987 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1988 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1989 
1990 		data &= ~(1 << 12);
1991 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1992 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1993 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1994 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1995 		return 0;
1996 	}
1997 
1998 	return 1;
1999 }
2000 
2001 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2002 {
2003 	if (!kvm_can_use_hv_timer(vcpu))
2004 		return 1;
2005 
2006 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
2007 	return 0;
2008 }
2009 
2010 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2011 {
2012 	u32 msr = kvm_rcx_read(vcpu);
2013 	u64 data;
2014 	fastpath_t ret = EXIT_FASTPATH_NONE;
2015 
2016 	switch (msr) {
2017 	case APIC_BASE_MSR + (APIC_ICR >> 4):
2018 		data = kvm_read_edx_eax(vcpu);
2019 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2020 			kvm_skip_emulated_instruction(vcpu);
2021 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2022 		}
2023 		break;
2024 	case MSR_IA32_TSC_DEADLINE:
2025 		data = kvm_read_edx_eax(vcpu);
2026 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2027 			kvm_skip_emulated_instruction(vcpu);
2028 			ret = EXIT_FASTPATH_REENTER_GUEST;
2029 		}
2030 		break;
2031 	default:
2032 		break;
2033 	}
2034 
2035 	if (ret != EXIT_FASTPATH_NONE)
2036 		trace_kvm_msr_write(msr, data);
2037 
2038 	return ret;
2039 }
2040 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2041 
2042 /*
2043  * Adapt set_msr() to msr_io()'s calling convention
2044  */
2045 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2046 {
2047 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2048 }
2049 
2050 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2051 {
2052 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2053 }
2054 
2055 #ifdef CONFIG_X86_64
2056 struct pvclock_clock {
2057 	int vclock_mode;
2058 	u64 cycle_last;
2059 	u64 mask;
2060 	u32 mult;
2061 	u32 shift;
2062 	u64 base_cycles;
2063 	u64 offset;
2064 };
2065 
2066 struct pvclock_gtod_data {
2067 	seqcount_t	seq;
2068 
2069 	struct pvclock_clock clock; /* extract of a clocksource struct */
2070 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2071 
2072 	ktime_t		offs_boot;
2073 	u64		wall_time_sec;
2074 };
2075 
2076 static struct pvclock_gtod_data pvclock_gtod_data;
2077 
2078 static void update_pvclock_gtod(struct timekeeper *tk)
2079 {
2080 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2081 
2082 	write_seqcount_begin(&vdata->seq);
2083 
2084 	/* copy pvclock gtod data */
2085 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2086 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2087 	vdata->clock.mask		= tk->tkr_mono.mask;
2088 	vdata->clock.mult		= tk->tkr_mono.mult;
2089 	vdata->clock.shift		= tk->tkr_mono.shift;
2090 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2091 	vdata->clock.offset		= tk->tkr_mono.base;
2092 
2093 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2094 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2095 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2096 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2097 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2098 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2099 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2100 
2101 	vdata->wall_time_sec            = tk->xtime_sec;
2102 
2103 	vdata->offs_boot		= tk->offs_boot;
2104 
2105 	write_seqcount_end(&vdata->seq);
2106 }
2107 
2108 static s64 get_kvmclock_base_ns(void)
2109 {
2110 	/* Count up from boot time, but with the frequency of the raw clock.  */
2111 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2112 }
2113 #else
2114 static s64 get_kvmclock_base_ns(void)
2115 {
2116 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2117 	return ktime_get_boottime_ns();
2118 }
2119 #endif
2120 
2121 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2122 {
2123 	int version;
2124 	int r;
2125 	struct pvclock_wall_clock wc;
2126 	u32 wc_sec_hi;
2127 	u64 wall_nsec;
2128 
2129 	if (!wall_clock)
2130 		return;
2131 
2132 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2133 	if (r)
2134 		return;
2135 
2136 	if (version & 1)
2137 		++version;  /* first time write, random junk */
2138 
2139 	++version;
2140 
2141 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2142 		return;
2143 
2144 	/*
2145 	 * The guest calculates current wall clock time by adding
2146 	 * system time (updated by kvm_guest_time_update below) to the
2147 	 * wall clock specified here.  We do the reverse here.
2148 	 */
2149 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2150 
2151 	wc.nsec = do_div(wall_nsec, 1000000000);
2152 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2153 	wc.version = version;
2154 
2155 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2156 
2157 	if (sec_hi_ofs) {
2158 		wc_sec_hi = wall_nsec >> 32;
2159 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2160 				&wc_sec_hi, sizeof(wc_sec_hi));
2161 	}
2162 
2163 	version++;
2164 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2165 }
2166 
2167 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2168 				  bool old_msr, bool host_initiated)
2169 {
2170 	struct kvm_arch *ka = &vcpu->kvm->arch;
2171 
2172 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2173 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2174 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2175 
2176 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2177 	}
2178 
2179 	vcpu->arch.time = system_time;
2180 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2181 
2182 	/* we verify if the enable bit is set... */
2183 	vcpu->arch.pv_time_enabled = false;
2184 	if (!(system_time & 1))
2185 		return;
2186 
2187 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2188 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2189 				       sizeof(struct pvclock_vcpu_time_info)))
2190 		vcpu->arch.pv_time_enabled = true;
2191 
2192 	return;
2193 }
2194 
2195 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2196 {
2197 	do_shl32_div32(dividend, divisor);
2198 	return dividend;
2199 }
2200 
2201 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2202 			       s8 *pshift, u32 *pmultiplier)
2203 {
2204 	uint64_t scaled64;
2205 	int32_t  shift = 0;
2206 	uint64_t tps64;
2207 	uint32_t tps32;
2208 
2209 	tps64 = base_hz;
2210 	scaled64 = scaled_hz;
2211 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2212 		tps64 >>= 1;
2213 		shift--;
2214 	}
2215 
2216 	tps32 = (uint32_t)tps64;
2217 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2218 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2219 			scaled64 >>= 1;
2220 		else
2221 			tps32 <<= 1;
2222 		shift++;
2223 	}
2224 
2225 	*pshift = shift;
2226 	*pmultiplier = div_frac(scaled64, tps32);
2227 }
2228 
2229 #ifdef CONFIG_X86_64
2230 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2231 #endif
2232 
2233 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2234 static unsigned long max_tsc_khz;
2235 
2236 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2237 {
2238 	u64 v = (u64)khz * (1000000 + ppm);
2239 	do_div(v, 1000000);
2240 	return v;
2241 }
2242 
2243 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2244 
2245 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2246 {
2247 	u64 ratio;
2248 
2249 	/* Guest TSC same frequency as host TSC? */
2250 	if (!scale) {
2251 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2252 		return 0;
2253 	}
2254 
2255 	/* TSC scaling supported? */
2256 	if (!kvm_has_tsc_control) {
2257 		if (user_tsc_khz > tsc_khz) {
2258 			vcpu->arch.tsc_catchup = 1;
2259 			vcpu->arch.tsc_always_catchup = 1;
2260 			return 0;
2261 		} else {
2262 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2263 			return -1;
2264 		}
2265 	}
2266 
2267 	/* TSC scaling required  - calculate ratio */
2268 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2269 				user_tsc_khz, tsc_khz);
2270 
2271 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2272 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2273 			            user_tsc_khz);
2274 		return -1;
2275 	}
2276 
2277 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2278 	return 0;
2279 }
2280 
2281 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2282 {
2283 	u32 thresh_lo, thresh_hi;
2284 	int use_scaling = 0;
2285 
2286 	/* tsc_khz can be zero if TSC calibration fails */
2287 	if (user_tsc_khz == 0) {
2288 		/* set tsc_scaling_ratio to a safe value */
2289 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2290 		return -1;
2291 	}
2292 
2293 	/* Compute a scale to convert nanoseconds in TSC cycles */
2294 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2295 			   &vcpu->arch.virtual_tsc_shift,
2296 			   &vcpu->arch.virtual_tsc_mult);
2297 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2298 
2299 	/*
2300 	 * Compute the variation in TSC rate which is acceptable
2301 	 * within the range of tolerance and decide if the
2302 	 * rate being applied is within that bounds of the hardware
2303 	 * rate.  If so, no scaling or compensation need be done.
2304 	 */
2305 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2306 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2307 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2308 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2309 		use_scaling = 1;
2310 	}
2311 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2312 }
2313 
2314 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2315 {
2316 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2317 				      vcpu->arch.virtual_tsc_mult,
2318 				      vcpu->arch.virtual_tsc_shift);
2319 	tsc += vcpu->arch.this_tsc_write;
2320 	return tsc;
2321 }
2322 
2323 static inline int gtod_is_based_on_tsc(int mode)
2324 {
2325 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2326 }
2327 
2328 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2329 {
2330 #ifdef CONFIG_X86_64
2331 	bool vcpus_matched;
2332 	struct kvm_arch *ka = &vcpu->kvm->arch;
2333 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2334 
2335 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2336 			 atomic_read(&vcpu->kvm->online_vcpus));
2337 
2338 	/*
2339 	 * Once the masterclock is enabled, always perform request in
2340 	 * order to update it.
2341 	 *
2342 	 * In order to enable masterclock, the host clocksource must be TSC
2343 	 * and the vcpus need to have matched TSCs.  When that happens,
2344 	 * perform request to enable masterclock.
2345 	 */
2346 	if (ka->use_master_clock ||
2347 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2348 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2349 
2350 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2351 			    atomic_read(&vcpu->kvm->online_vcpus),
2352 		            ka->use_master_clock, gtod->clock.vclock_mode);
2353 #endif
2354 }
2355 
2356 /*
2357  * Multiply tsc by a fixed point number represented by ratio.
2358  *
2359  * The most significant 64-N bits (mult) of ratio represent the
2360  * integral part of the fixed point number; the remaining N bits
2361  * (frac) represent the fractional part, ie. ratio represents a fixed
2362  * point number (mult + frac * 2^(-N)).
2363  *
2364  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2365  */
2366 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2367 {
2368 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2369 }
2370 
2371 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2372 {
2373 	u64 _tsc = tsc;
2374 
2375 	if (ratio != kvm_default_tsc_scaling_ratio)
2376 		_tsc = __scale_tsc(ratio, tsc);
2377 
2378 	return _tsc;
2379 }
2380 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2381 
2382 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2383 {
2384 	u64 tsc;
2385 
2386 	tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2387 
2388 	return target_tsc - tsc;
2389 }
2390 
2391 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2392 {
2393 	return vcpu->arch.l1_tsc_offset +
2394 		kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2395 }
2396 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2397 
2398 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2399 {
2400 	u64 nested_offset;
2401 
2402 	if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2403 		nested_offset = l1_offset;
2404 	else
2405 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2406 						kvm_tsc_scaling_ratio_frac_bits);
2407 
2408 	nested_offset += l2_offset;
2409 	return nested_offset;
2410 }
2411 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2412 
2413 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2414 {
2415 	if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2416 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2417 				       kvm_tsc_scaling_ratio_frac_bits);
2418 
2419 	return l1_multiplier;
2420 }
2421 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2422 
2423 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2424 {
2425 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2426 				   vcpu->arch.l1_tsc_offset,
2427 				   l1_offset);
2428 
2429 	vcpu->arch.l1_tsc_offset = l1_offset;
2430 
2431 	/*
2432 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2433 	 * according to the spec this should set L1's TSC (as opposed to
2434 	 * setting L1's offset for L2).
2435 	 */
2436 	if (is_guest_mode(vcpu))
2437 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2438 			l1_offset,
2439 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2440 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2441 	else
2442 		vcpu->arch.tsc_offset = l1_offset;
2443 
2444 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2445 }
2446 
2447 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2448 {
2449 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2450 
2451 	/* Userspace is changing the multiplier while L2 is active */
2452 	if (is_guest_mode(vcpu))
2453 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2454 			l1_multiplier,
2455 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2456 	else
2457 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2458 
2459 	if (kvm_has_tsc_control)
2460 		static_call(kvm_x86_write_tsc_multiplier)(
2461 			vcpu, vcpu->arch.tsc_scaling_ratio);
2462 }
2463 
2464 static inline bool kvm_check_tsc_unstable(void)
2465 {
2466 #ifdef CONFIG_X86_64
2467 	/*
2468 	 * TSC is marked unstable when we're running on Hyper-V,
2469 	 * 'TSC page' clocksource is good.
2470 	 */
2471 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2472 		return false;
2473 #endif
2474 	return check_tsc_unstable();
2475 }
2476 
2477 /*
2478  * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2479  * offset for the vcpu and tracks the TSC matching generation that the vcpu
2480  * participates in.
2481  */
2482 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2483 				  u64 ns, bool matched)
2484 {
2485 	struct kvm *kvm = vcpu->kvm;
2486 
2487 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2488 
2489 	/*
2490 	 * We also track th most recent recorded KHZ, write and time to
2491 	 * allow the matching interval to be extended at each write.
2492 	 */
2493 	kvm->arch.last_tsc_nsec = ns;
2494 	kvm->arch.last_tsc_write = tsc;
2495 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2496 	kvm->arch.last_tsc_offset = offset;
2497 
2498 	vcpu->arch.last_guest_tsc = tsc;
2499 
2500 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2501 
2502 	if (!matched) {
2503 		/*
2504 		 * We split periods of matched TSC writes into generations.
2505 		 * For each generation, we track the original measured
2506 		 * nanosecond time, offset, and write, so if TSCs are in
2507 		 * sync, we can match exact offset, and if not, we can match
2508 		 * exact software computation in compute_guest_tsc()
2509 		 *
2510 		 * These values are tracked in kvm->arch.cur_xxx variables.
2511 		 */
2512 		kvm->arch.cur_tsc_generation++;
2513 		kvm->arch.cur_tsc_nsec = ns;
2514 		kvm->arch.cur_tsc_write = tsc;
2515 		kvm->arch.cur_tsc_offset = offset;
2516 		kvm->arch.nr_vcpus_matched_tsc = 0;
2517 	} else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2518 		kvm->arch.nr_vcpus_matched_tsc++;
2519 	}
2520 
2521 	/* Keep track of which generation this VCPU has synchronized to */
2522 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2523 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2524 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2525 
2526 	kvm_track_tsc_matching(vcpu);
2527 }
2528 
2529 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2530 {
2531 	struct kvm *kvm = vcpu->kvm;
2532 	u64 offset, ns, elapsed;
2533 	unsigned long flags;
2534 	bool matched = false;
2535 	bool synchronizing = false;
2536 
2537 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2538 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2539 	ns = get_kvmclock_base_ns();
2540 	elapsed = ns - kvm->arch.last_tsc_nsec;
2541 
2542 	if (vcpu->arch.virtual_tsc_khz) {
2543 		if (data == 0) {
2544 			/*
2545 			 * detection of vcpu initialization -- need to sync
2546 			 * with other vCPUs. This particularly helps to keep
2547 			 * kvm_clock stable after CPU hotplug
2548 			 */
2549 			synchronizing = true;
2550 		} else {
2551 			u64 tsc_exp = kvm->arch.last_tsc_write +
2552 						nsec_to_cycles(vcpu, elapsed);
2553 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2554 			/*
2555 			 * Special case: TSC write with a small delta (1 second)
2556 			 * of virtual cycle time against real time is
2557 			 * interpreted as an attempt to synchronize the CPU.
2558 			 */
2559 			synchronizing = data < tsc_exp + tsc_hz &&
2560 					data + tsc_hz > tsc_exp;
2561 		}
2562 	}
2563 
2564 	/*
2565 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2566 	 * TSC, we add elapsed time in this computation.  We could let the
2567 	 * compensation code attempt to catch up if we fall behind, but
2568 	 * it's better to try to match offsets from the beginning.
2569          */
2570 	if (synchronizing &&
2571 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2572 		if (!kvm_check_tsc_unstable()) {
2573 			offset = kvm->arch.cur_tsc_offset;
2574 		} else {
2575 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2576 			data += delta;
2577 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2578 		}
2579 		matched = true;
2580 	}
2581 
2582 	__kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2583 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2584 }
2585 
2586 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2587 					   s64 adjustment)
2588 {
2589 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2590 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2591 }
2592 
2593 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2594 {
2595 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2596 		WARN_ON(adjustment < 0);
2597 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2598 				   vcpu->arch.l1_tsc_scaling_ratio);
2599 	adjust_tsc_offset_guest(vcpu, adjustment);
2600 }
2601 
2602 #ifdef CONFIG_X86_64
2603 
2604 static u64 read_tsc(void)
2605 {
2606 	u64 ret = (u64)rdtsc_ordered();
2607 	u64 last = pvclock_gtod_data.clock.cycle_last;
2608 
2609 	if (likely(ret >= last))
2610 		return ret;
2611 
2612 	/*
2613 	 * GCC likes to generate cmov here, but this branch is extremely
2614 	 * predictable (it's just a function of time and the likely is
2615 	 * very likely) and there's a data dependence, so force GCC
2616 	 * to generate a branch instead.  I don't barrier() because
2617 	 * we don't actually need a barrier, and if this function
2618 	 * ever gets inlined it will generate worse code.
2619 	 */
2620 	asm volatile ("");
2621 	return last;
2622 }
2623 
2624 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2625 			  int *mode)
2626 {
2627 	long v;
2628 	u64 tsc_pg_val;
2629 
2630 	switch (clock->vclock_mode) {
2631 	case VDSO_CLOCKMODE_HVCLOCK:
2632 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2633 						  tsc_timestamp);
2634 		if (tsc_pg_val != U64_MAX) {
2635 			/* TSC page valid */
2636 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2637 			v = (tsc_pg_val - clock->cycle_last) &
2638 				clock->mask;
2639 		} else {
2640 			/* TSC page invalid */
2641 			*mode = VDSO_CLOCKMODE_NONE;
2642 		}
2643 		break;
2644 	case VDSO_CLOCKMODE_TSC:
2645 		*mode = VDSO_CLOCKMODE_TSC;
2646 		*tsc_timestamp = read_tsc();
2647 		v = (*tsc_timestamp - clock->cycle_last) &
2648 			clock->mask;
2649 		break;
2650 	default:
2651 		*mode = VDSO_CLOCKMODE_NONE;
2652 	}
2653 
2654 	if (*mode == VDSO_CLOCKMODE_NONE)
2655 		*tsc_timestamp = v = 0;
2656 
2657 	return v * clock->mult;
2658 }
2659 
2660 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2661 {
2662 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2663 	unsigned long seq;
2664 	int mode;
2665 	u64 ns;
2666 
2667 	do {
2668 		seq = read_seqcount_begin(&gtod->seq);
2669 		ns = gtod->raw_clock.base_cycles;
2670 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2671 		ns >>= gtod->raw_clock.shift;
2672 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2673 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2674 	*t = ns;
2675 
2676 	return mode;
2677 }
2678 
2679 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2680 {
2681 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2682 	unsigned long seq;
2683 	int mode;
2684 	u64 ns;
2685 
2686 	do {
2687 		seq = read_seqcount_begin(&gtod->seq);
2688 		ts->tv_sec = gtod->wall_time_sec;
2689 		ns = gtod->clock.base_cycles;
2690 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2691 		ns >>= gtod->clock.shift;
2692 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2693 
2694 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2695 	ts->tv_nsec = ns;
2696 
2697 	return mode;
2698 }
2699 
2700 /* returns true if host is using TSC based clocksource */
2701 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2702 {
2703 	/* checked again under seqlock below */
2704 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2705 		return false;
2706 
2707 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2708 						      tsc_timestamp));
2709 }
2710 
2711 /* returns true if host is using TSC based clocksource */
2712 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2713 					   u64 *tsc_timestamp)
2714 {
2715 	/* checked again under seqlock below */
2716 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2717 		return false;
2718 
2719 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2720 }
2721 #endif
2722 
2723 /*
2724  *
2725  * Assuming a stable TSC across physical CPUS, and a stable TSC
2726  * across virtual CPUs, the following condition is possible.
2727  * Each numbered line represents an event visible to both
2728  * CPUs at the next numbered event.
2729  *
2730  * "timespecX" represents host monotonic time. "tscX" represents
2731  * RDTSC value.
2732  *
2733  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2734  *
2735  * 1.  read timespec0,tsc0
2736  * 2.					| timespec1 = timespec0 + N
2737  * 					| tsc1 = tsc0 + M
2738  * 3. transition to guest		| transition to guest
2739  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2740  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2741  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2742  *
2743  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2744  *
2745  * 	- ret0 < ret1
2746  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2747  *		...
2748  *	- 0 < N - M => M < N
2749  *
2750  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2751  * always the case (the difference between two distinct xtime instances
2752  * might be smaller then the difference between corresponding TSC reads,
2753  * when updating guest vcpus pvclock areas).
2754  *
2755  * To avoid that problem, do not allow visibility of distinct
2756  * system_timestamp/tsc_timestamp values simultaneously: use a master
2757  * copy of host monotonic time values. Update that master copy
2758  * in lockstep.
2759  *
2760  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2761  *
2762  */
2763 
2764 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2765 {
2766 #ifdef CONFIG_X86_64
2767 	struct kvm_arch *ka = &kvm->arch;
2768 	int vclock_mode;
2769 	bool host_tsc_clocksource, vcpus_matched;
2770 
2771 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2772 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2773 			atomic_read(&kvm->online_vcpus));
2774 
2775 	/*
2776 	 * If the host uses TSC clock, then passthrough TSC as stable
2777 	 * to the guest.
2778 	 */
2779 	host_tsc_clocksource = kvm_get_time_and_clockread(
2780 					&ka->master_kernel_ns,
2781 					&ka->master_cycle_now);
2782 
2783 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2784 				&& !ka->backwards_tsc_observed
2785 				&& !ka->boot_vcpu_runs_old_kvmclock;
2786 
2787 	if (ka->use_master_clock)
2788 		atomic_set(&kvm_guest_has_master_clock, 1);
2789 
2790 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2791 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2792 					vcpus_matched);
2793 #endif
2794 }
2795 
2796 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2797 {
2798 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2799 }
2800 
2801 static void __kvm_start_pvclock_update(struct kvm *kvm)
2802 {
2803 	raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2804 	write_seqcount_begin(&kvm->arch.pvclock_sc);
2805 }
2806 
2807 static void kvm_start_pvclock_update(struct kvm *kvm)
2808 {
2809 	kvm_make_mclock_inprogress_request(kvm);
2810 
2811 	/* no guest entries from this point */
2812 	__kvm_start_pvclock_update(kvm);
2813 }
2814 
2815 static void kvm_end_pvclock_update(struct kvm *kvm)
2816 {
2817 	struct kvm_arch *ka = &kvm->arch;
2818 	struct kvm_vcpu *vcpu;
2819 	int i;
2820 
2821 	write_seqcount_end(&ka->pvclock_sc);
2822 	raw_spin_unlock_irq(&ka->tsc_write_lock);
2823 	kvm_for_each_vcpu(i, vcpu, kvm)
2824 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2825 
2826 	/* guest entries allowed */
2827 	kvm_for_each_vcpu(i, vcpu, kvm)
2828 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2829 }
2830 
2831 static void kvm_update_masterclock(struct kvm *kvm)
2832 {
2833 	kvm_hv_invalidate_tsc_page(kvm);
2834 	kvm_start_pvclock_update(kvm);
2835 	pvclock_update_vm_gtod_copy(kvm);
2836 	kvm_end_pvclock_update(kvm);
2837 }
2838 
2839 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc.  */
2840 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2841 {
2842 	struct kvm_arch *ka = &kvm->arch;
2843 	struct pvclock_vcpu_time_info hv_clock;
2844 
2845 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2846 	get_cpu();
2847 
2848 	data->flags = 0;
2849 	if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2850 #ifdef CONFIG_X86_64
2851 		struct timespec64 ts;
2852 
2853 		if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2854 			data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2855 			data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2856 		} else
2857 #endif
2858 		data->host_tsc = rdtsc();
2859 
2860 		data->flags |= KVM_CLOCK_TSC_STABLE;
2861 		hv_clock.tsc_timestamp = ka->master_cycle_now;
2862 		hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2863 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2864 				   &hv_clock.tsc_shift,
2865 				   &hv_clock.tsc_to_system_mul);
2866 		data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2867 	} else {
2868 		data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2869 	}
2870 
2871 	put_cpu();
2872 }
2873 
2874 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2875 {
2876 	struct kvm_arch *ka = &kvm->arch;
2877 	unsigned seq;
2878 
2879 	do {
2880 		seq = read_seqcount_begin(&ka->pvclock_sc);
2881 		__get_kvmclock(kvm, data);
2882 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2883 }
2884 
2885 u64 get_kvmclock_ns(struct kvm *kvm)
2886 {
2887 	struct kvm_clock_data data;
2888 
2889 	get_kvmclock(kvm, &data);
2890 	return data.clock;
2891 }
2892 
2893 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2894 				   struct gfn_to_hva_cache *cache,
2895 				   unsigned int offset)
2896 {
2897 	struct kvm_vcpu_arch *vcpu = &v->arch;
2898 	struct pvclock_vcpu_time_info guest_hv_clock;
2899 
2900 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2901 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2902 		return;
2903 
2904 	/* This VCPU is paused, but it's legal for a guest to read another
2905 	 * VCPU's kvmclock, so we really have to follow the specification where
2906 	 * it says that version is odd if data is being modified, and even after
2907 	 * it is consistent.
2908 	 *
2909 	 * Version field updates must be kept separate.  This is because
2910 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2911 	 * writes within a string instruction are weakly ordered.  So there
2912 	 * are three writes overall.
2913 	 *
2914 	 * As a small optimization, only write the version field in the first
2915 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2916 	 * version field is the first in the struct.
2917 	 */
2918 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2919 
2920 	if (guest_hv_clock.version & 1)
2921 		++guest_hv_clock.version;  /* first time write, random junk */
2922 
2923 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2924 	kvm_write_guest_offset_cached(v->kvm, cache,
2925 				      &vcpu->hv_clock, offset,
2926 				      sizeof(vcpu->hv_clock.version));
2927 
2928 	smp_wmb();
2929 
2930 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2931 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2932 
2933 	if (vcpu->pvclock_set_guest_stopped_request) {
2934 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2935 		vcpu->pvclock_set_guest_stopped_request = false;
2936 	}
2937 
2938 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2939 
2940 	kvm_write_guest_offset_cached(v->kvm, cache,
2941 				      &vcpu->hv_clock, offset,
2942 				      sizeof(vcpu->hv_clock));
2943 
2944 	smp_wmb();
2945 
2946 	vcpu->hv_clock.version++;
2947 	kvm_write_guest_offset_cached(v->kvm, cache,
2948 				     &vcpu->hv_clock, offset,
2949 				     sizeof(vcpu->hv_clock.version));
2950 }
2951 
2952 static int kvm_guest_time_update(struct kvm_vcpu *v)
2953 {
2954 	unsigned long flags, tgt_tsc_khz;
2955 	unsigned seq;
2956 	struct kvm_vcpu_arch *vcpu = &v->arch;
2957 	struct kvm_arch *ka = &v->kvm->arch;
2958 	s64 kernel_ns;
2959 	u64 tsc_timestamp, host_tsc;
2960 	u8 pvclock_flags;
2961 	bool use_master_clock;
2962 
2963 	kernel_ns = 0;
2964 	host_tsc = 0;
2965 
2966 	/*
2967 	 * If the host uses TSC clock, then passthrough TSC as stable
2968 	 * to the guest.
2969 	 */
2970 	do {
2971 		seq = read_seqcount_begin(&ka->pvclock_sc);
2972 		use_master_clock = ka->use_master_clock;
2973 		if (use_master_clock) {
2974 			host_tsc = ka->master_cycle_now;
2975 			kernel_ns = ka->master_kernel_ns;
2976 		}
2977 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2978 
2979 	/* Keep irq disabled to prevent changes to the clock */
2980 	local_irq_save(flags);
2981 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2982 	if (unlikely(tgt_tsc_khz == 0)) {
2983 		local_irq_restore(flags);
2984 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2985 		return 1;
2986 	}
2987 	if (!use_master_clock) {
2988 		host_tsc = rdtsc();
2989 		kernel_ns = get_kvmclock_base_ns();
2990 	}
2991 
2992 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2993 
2994 	/*
2995 	 * We may have to catch up the TSC to match elapsed wall clock
2996 	 * time for two reasons, even if kvmclock is used.
2997 	 *   1) CPU could have been running below the maximum TSC rate
2998 	 *   2) Broken TSC compensation resets the base at each VCPU
2999 	 *      entry to avoid unknown leaps of TSC even when running
3000 	 *      again on the same CPU.  This may cause apparent elapsed
3001 	 *      time to disappear, and the guest to stand still or run
3002 	 *	very slowly.
3003 	 */
3004 	if (vcpu->tsc_catchup) {
3005 		u64 tsc = compute_guest_tsc(v, kernel_ns);
3006 		if (tsc > tsc_timestamp) {
3007 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3008 			tsc_timestamp = tsc;
3009 		}
3010 	}
3011 
3012 	local_irq_restore(flags);
3013 
3014 	/* With all the info we got, fill in the values */
3015 
3016 	if (kvm_has_tsc_control)
3017 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3018 					    v->arch.l1_tsc_scaling_ratio);
3019 
3020 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3021 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3022 				   &vcpu->hv_clock.tsc_shift,
3023 				   &vcpu->hv_clock.tsc_to_system_mul);
3024 		vcpu->hw_tsc_khz = tgt_tsc_khz;
3025 	}
3026 
3027 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3028 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3029 	vcpu->last_guest_tsc = tsc_timestamp;
3030 
3031 	/* If the host uses TSC clocksource, then it is stable */
3032 	pvclock_flags = 0;
3033 	if (use_master_clock)
3034 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3035 
3036 	vcpu->hv_clock.flags = pvclock_flags;
3037 
3038 	if (vcpu->pv_time_enabled)
3039 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3040 	if (vcpu->xen.vcpu_info_set)
3041 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3042 				       offsetof(struct compat_vcpu_info, time));
3043 	if (vcpu->xen.vcpu_time_info_set)
3044 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3045 	if (!v->vcpu_idx)
3046 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3047 	return 0;
3048 }
3049 
3050 /*
3051  * kvmclock updates which are isolated to a given vcpu, such as
3052  * vcpu->cpu migration, should not allow system_timestamp from
3053  * the rest of the vcpus to remain static. Otherwise ntp frequency
3054  * correction applies to one vcpu's system_timestamp but not
3055  * the others.
3056  *
3057  * So in those cases, request a kvmclock update for all vcpus.
3058  * We need to rate-limit these requests though, as they can
3059  * considerably slow guests that have a large number of vcpus.
3060  * The time for a remote vcpu to update its kvmclock is bound
3061  * by the delay we use to rate-limit the updates.
3062  */
3063 
3064 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3065 
3066 static void kvmclock_update_fn(struct work_struct *work)
3067 {
3068 	int i;
3069 	struct delayed_work *dwork = to_delayed_work(work);
3070 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3071 					   kvmclock_update_work);
3072 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3073 	struct kvm_vcpu *vcpu;
3074 
3075 	kvm_for_each_vcpu(i, vcpu, kvm) {
3076 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3077 		kvm_vcpu_kick(vcpu);
3078 	}
3079 }
3080 
3081 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3082 {
3083 	struct kvm *kvm = v->kvm;
3084 
3085 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3086 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3087 					KVMCLOCK_UPDATE_DELAY);
3088 }
3089 
3090 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3091 
3092 static void kvmclock_sync_fn(struct work_struct *work)
3093 {
3094 	struct delayed_work *dwork = to_delayed_work(work);
3095 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3096 					   kvmclock_sync_work);
3097 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3098 
3099 	if (!kvmclock_periodic_sync)
3100 		return;
3101 
3102 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3103 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3104 					KVMCLOCK_SYNC_PERIOD);
3105 }
3106 
3107 /*
3108  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3109  */
3110 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3111 {
3112 	/* McStatusWrEn enabled? */
3113 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3114 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3115 
3116 	return false;
3117 }
3118 
3119 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3120 {
3121 	u64 mcg_cap = vcpu->arch.mcg_cap;
3122 	unsigned bank_num = mcg_cap & 0xff;
3123 	u32 msr = msr_info->index;
3124 	u64 data = msr_info->data;
3125 
3126 	switch (msr) {
3127 	case MSR_IA32_MCG_STATUS:
3128 		vcpu->arch.mcg_status = data;
3129 		break;
3130 	case MSR_IA32_MCG_CTL:
3131 		if (!(mcg_cap & MCG_CTL_P) &&
3132 		    (data || !msr_info->host_initiated))
3133 			return 1;
3134 		if (data != 0 && data != ~(u64)0)
3135 			return 1;
3136 		vcpu->arch.mcg_ctl = data;
3137 		break;
3138 	default:
3139 		if (msr >= MSR_IA32_MC0_CTL &&
3140 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3141 			u32 offset = array_index_nospec(
3142 				msr - MSR_IA32_MC0_CTL,
3143 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3144 
3145 			/* only 0 or all 1s can be written to IA32_MCi_CTL
3146 			 * some Linux kernels though clear bit 10 in bank 4 to
3147 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3148 			 * this to avoid an uncatched #GP in the guest
3149 			 */
3150 			if ((offset & 0x3) == 0 &&
3151 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
3152 				return -1;
3153 
3154 			/* MCi_STATUS */
3155 			if (!msr_info->host_initiated &&
3156 			    (offset & 0x3) == 1 && data != 0) {
3157 				if (!can_set_mci_status(vcpu))
3158 					return -1;
3159 			}
3160 
3161 			vcpu->arch.mce_banks[offset] = data;
3162 			break;
3163 		}
3164 		return 1;
3165 	}
3166 	return 0;
3167 }
3168 
3169 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3170 {
3171 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3172 
3173 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3174 }
3175 
3176 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3177 {
3178 	gpa_t gpa = data & ~0x3f;
3179 
3180 	/* Bits 4:5 are reserved, Should be zero */
3181 	if (data & 0x30)
3182 		return 1;
3183 
3184 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3185 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3186 		return 1;
3187 
3188 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3189 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3190 		return 1;
3191 
3192 	if (!lapic_in_kernel(vcpu))
3193 		return data ? 1 : 0;
3194 
3195 	vcpu->arch.apf.msr_en_val = data;
3196 
3197 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3198 		kvm_clear_async_pf_completion_queue(vcpu);
3199 		kvm_async_pf_hash_reset(vcpu);
3200 		return 0;
3201 	}
3202 
3203 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3204 					sizeof(u64)))
3205 		return 1;
3206 
3207 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3208 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3209 
3210 	kvm_async_pf_wakeup_all(vcpu);
3211 
3212 	return 0;
3213 }
3214 
3215 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3216 {
3217 	/* Bits 8-63 are reserved */
3218 	if (data >> 8)
3219 		return 1;
3220 
3221 	if (!lapic_in_kernel(vcpu))
3222 		return 1;
3223 
3224 	vcpu->arch.apf.msr_int_val = data;
3225 
3226 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3227 
3228 	return 0;
3229 }
3230 
3231 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3232 {
3233 	vcpu->arch.pv_time_enabled = false;
3234 	vcpu->arch.time = 0;
3235 }
3236 
3237 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3238 {
3239 	++vcpu->stat.tlb_flush;
3240 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3241 }
3242 
3243 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3244 {
3245 	++vcpu->stat.tlb_flush;
3246 
3247 	if (!tdp_enabled) {
3248 		/*
3249 		 * A TLB flush on behalf of the guest is equivalent to
3250 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3251 		 * a forced sync of the shadow page tables.  Ensure all the
3252 		 * roots are synced and the guest TLB in hardware is clean.
3253 		 */
3254 		kvm_mmu_sync_roots(vcpu);
3255 		kvm_mmu_sync_prev_roots(vcpu);
3256 	}
3257 
3258 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3259 }
3260 
3261 static void record_steal_time(struct kvm_vcpu *vcpu)
3262 {
3263 	struct kvm_host_map map;
3264 	struct kvm_steal_time *st;
3265 
3266 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3267 		kvm_xen_runstate_set_running(vcpu);
3268 		return;
3269 	}
3270 
3271 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3272 		return;
3273 
3274 	/* -EAGAIN is returned in atomic context so we can just return. */
3275 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3276 			&map, &vcpu->arch.st.cache, false))
3277 		return;
3278 
3279 	st = map.hva +
3280 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3281 
3282 	/*
3283 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3284 	 * expensive IPIs.
3285 	 */
3286 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3287 		u8 st_preempted = xchg(&st->preempted, 0);
3288 
3289 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3290 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3291 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3292 			kvm_vcpu_flush_tlb_guest(vcpu);
3293 	} else {
3294 		st->preempted = 0;
3295 	}
3296 
3297 	vcpu->arch.st.preempted = 0;
3298 
3299 	if (st->version & 1)
3300 		st->version += 1;  /* first time write, random junk */
3301 
3302 	st->version += 1;
3303 
3304 	smp_wmb();
3305 
3306 	st->steal += current->sched_info.run_delay -
3307 		vcpu->arch.st.last_steal;
3308 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3309 
3310 	smp_wmb();
3311 
3312 	st->version += 1;
3313 
3314 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3315 }
3316 
3317 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3318 {
3319 	bool pr = false;
3320 	u32 msr = msr_info->index;
3321 	u64 data = msr_info->data;
3322 
3323 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3324 		return kvm_xen_write_hypercall_page(vcpu, data);
3325 
3326 	switch (msr) {
3327 	case MSR_AMD64_NB_CFG:
3328 	case MSR_IA32_UCODE_WRITE:
3329 	case MSR_VM_HSAVE_PA:
3330 	case MSR_AMD64_PATCH_LOADER:
3331 	case MSR_AMD64_BU_CFG2:
3332 	case MSR_AMD64_DC_CFG:
3333 	case MSR_F15H_EX_CFG:
3334 		break;
3335 
3336 	case MSR_IA32_UCODE_REV:
3337 		if (msr_info->host_initiated)
3338 			vcpu->arch.microcode_version = data;
3339 		break;
3340 	case MSR_IA32_ARCH_CAPABILITIES:
3341 		if (!msr_info->host_initiated)
3342 			return 1;
3343 		vcpu->arch.arch_capabilities = data;
3344 		break;
3345 	case MSR_IA32_PERF_CAPABILITIES: {
3346 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3347 
3348 		if (!msr_info->host_initiated)
3349 			return 1;
3350 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3351 			return 1;
3352 		if (data & ~msr_ent.data)
3353 			return 1;
3354 
3355 		vcpu->arch.perf_capabilities = data;
3356 
3357 		return 0;
3358 		}
3359 	case MSR_EFER:
3360 		return set_efer(vcpu, msr_info);
3361 	case MSR_K7_HWCR:
3362 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3363 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3364 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3365 
3366 		/* Handle McStatusWrEn */
3367 		if (data == BIT_ULL(18)) {
3368 			vcpu->arch.msr_hwcr = data;
3369 		} else if (data != 0) {
3370 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3371 				    data);
3372 			return 1;
3373 		}
3374 		break;
3375 	case MSR_FAM10H_MMIO_CONF_BASE:
3376 		if (data != 0) {
3377 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3378 				    "0x%llx\n", data);
3379 			return 1;
3380 		}
3381 		break;
3382 	case 0x200 ... 0x2ff:
3383 		return kvm_mtrr_set_msr(vcpu, msr, data);
3384 	case MSR_IA32_APICBASE:
3385 		return kvm_set_apic_base(vcpu, msr_info);
3386 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3387 		return kvm_x2apic_msr_write(vcpu, msr, data);
3388 	case MSR_IA32_TSC_DEADLINE:
3389 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3390 		break;
3391 	case MSR_IA32_TSC_ADJUST:
3392 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3393 			if (!msr_info->host_initiated) {
3394 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3395 				adjust_tsc_offset_guest(vcpu, adj);
3396 				/* Before back to guest, tsc_timestamp must be adjusted
3397 				 * as well, otherwise guest's percpu pvclock time could jump.
3398 				 */
3399 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3400 			}
3401 			vcpu->arch.ia32_tsc_adjust_msr = data;
3402 		}
3403 		break;
3404 	case MSR_IA32_MISC_ENABLE:
3405 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3406 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3407 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3408 				return 1;
3409 			vcpu->arch.ia32_misc_enable_msr = data;
3410 			kvm_update_cpuid_runtime(vcpu);
3411 		} else {
3412 			vcpu->arch.ia32_misc_enable_msr = data;
3413 		}
3414 		break;
3415 	case MSR_IA32_SMBASE:
3416 		if (!msr_info->host_initiated)
3417 			return 1;
3418 		vcpu->arch.smbase = data;
3419 		break;
3420 	case MSR_IA32_POWER_CTL:
3421 		vcpu->arch.msr_ia32_power_ctl = data;
3422 		break;
3423 	case MSR_IA32_TSC:
3424 		if (msr_info->host_initiated) {
3425 			kvm_synchronize_tsc(vcpu, data);
3426 		} else {
3427 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3428 			adjust_tsc_offset_guest(vcpu, adj);
3429 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3430 		}
3431 		break;
3432 	case MSR_IA32_XSS:
3433 		if (!msr_info->host_initiated &&
3434 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3435 			return 1;
3436 		/*
3437 		 * KVM supports exposing PT to the guest, but does not support
3438 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3439 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3440 		 */
3441 		if (data & ~supported_xss)
3442 			return 1;
3443 		vcpu->arch.ia32_xss = data;
3444 		break;
3445 	case MSR_SMI_COUNT:
3446 		if (!msr_info->host_initiated)
3447 			return 1;
3448 		vcpu->arch.smi_count = data;
3449 		break;
3450 	case MSR_KVM_WALL_CLOCK_NEW:
3451 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3452 			return 1;
3453 
3454 		vcpu->kvm->arch.wall_clock = data;
3455 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3456 		break;
3457 	case MSR_KVM_WALL_CLOCK:
3458 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3459 			return 1;
3460 
3461 		vcpu->kvm->arch.wall_clock = data;
3462 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3463 		break;
3464 	case MSR_KVM_SYSTEM_TIME_NEW:
3465 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3466 			return 1;
3467 
3468 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3469 		break;
3470 	case MSR_KVM_SYSTEM_TIME:
3471 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3472 			return 1;
3473 
3474 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3475 		break;
3476 	case MSR_KVM_ASYNC_PF_EN:
3477 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3478 			return 1;
3479 
3480 		if (kvm_pv_enable_async_pf(vcpu, data))
3481 			return 1;
3482 		break;
3483 	case MSR_KVM_ASYNC_PF_INT:
3484 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3485 			return 1;
3486 
3487 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3488 			return 1;
3489 		break;
3490 	case MSR_KVM_ASYNC_PF_ACK:
3491 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3492 			return 1;
3493 		if (data & 0x1) {
3494 			vcpu->arch.apf.pageready_pending = false;
3495 			kvm_check_async_pf_completion(vcpu);
3496 		}
3497 		break;
3498 	case MSR_KVM_STEAL_TIME:
3499 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3500 			return 1;
3501 
3502 		if (unlikely(!sched_info_on()))
3503 			return 1;
3504 
3505 		if (data & KVM_STEAL_RESERVED_MASK)
3506 			return 1;
3507 
3508 		vcpu->arch.st.msr_val = data;
3509 
3510 		if (!(data & KVM_MSR_ENABLED))
3511 			break;
3512 
3513 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3514 
3515 		break;
3516 	case MSR_KVM_PV_EOI_EN:
3517 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3518 			return 1;
3519 
3520 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3521 			return 1;
3522 		break;
3523 
3524 	case MSR_KVM_POLL_CONTROL:
3525 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3526 			return 1;
3527 
3528 		/* only enable bit supported */
3529 		if (data & (-1ULL << 1))
3530 			return 1;
3531 
3532 		vcpu->arch.msr_kvm_poll_control = data;
3533 		break;
3534 
3535 	case MSR_IA32_MCG_CTL:
3536 	case MSR_IA32_MCG_STATUS:
3537 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3538 		return set_msr_mce(vcpu, msr_info);
3539 
3540 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3541 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3542 		pr = true;
3543 		fallthrough;
3544 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3545 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3546 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3547 			return kvm_pmu_set_msr(vcpu, msr_info);
3548 
3549 		if (pr || data != 0)
3550 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3551 				    "0x%x data 0x%llx\n", msr, data);
3552 		break;
3553 	case MSR_K7_CLK_CTL:
3554 		/*
3555 		 * Ignore all writes to this no longer documented MSR.
3556 		 * Writes are only relevant for old K7 processors,
3557 		 * all pre-dating SVM, but a recommended workaround from
3558 		 * AMD for these chips. It is possible to specify the
3559 		 * affected processor models on the command line, hence
3560 		 * the need to ignore the workaround.
3561 		 */
3562 		break;
3563 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3564 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3565 	case HV_X64_MSR_SYNDBG_OPTIONS:
3566 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3567 	case HV_X64_MSR_CRASH_CTL:
3568 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3569 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3570 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3571 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3572 		return kvm_hv_set_msr_common(vcpu, msr, data,
3573 					     msr_info->host_initiated);
3574 	case MSR_IA32_BBL_CR_CTL3:
3575 		/* Drop writes to this legacy MSR -- see rdmsr
3576 		 * counterpart for further detail.
3577 		 */
3578 		if (report_ignored_msrs)
3579 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3580 				msr, data);
3581 		break;
3582 	case MSR_AMD64_OSVW_ID_LENGTH:
3583 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3584 			return 1;
3585 		vcpu->arch.osvw.length = data;
3586 		break;
3587 	case MSR_AMD64_OSVW_STATUS:
3588 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3589 			return 1;
3590 		vcpu->arch.osvw.status = data;
3591 		break;
3592 	case MSR_PLATFORM_INFO:
3593 		if (!msr_info->host_initiated ||
3594 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3595 		     cpuid_fault_enabled(vcpu)))
3596 			return 1;
3597 		vcpu->arch.msr_platform_info = data;
3598 		break;
3599 	case MSR_MISC_FEATURES_ENABLES:
3600 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3601 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3602 		     !supports_cpuid_fault(vcpu)))
3603 			return 1;
3604 		vcpu->arch.msr_misc_features_enables = data;
3605 		break;
3606 	default:
3607 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3608 			return kvm_pmu_set_msr(vcpu, msr_info);
3609 		return KVM_MSR_RET_INVALID;
3610 	}
3611 	return 0;
3612 }
3613 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3614 
3615 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3616 {
3617 	u64 data;
3618 	u64 mcg_cap = vcpu->arch.mcg_cap;
3619 	unsigned bank_num = mcg_cap & 0xff;
3620 
3621 	switch (msr) {
3622 	case MSR_IA32_P5_MC_ADDR:
3623 	case MSR_IA32_P5_MC_TYPE:
3624 		data = 0;
3625 		break;
3626 	case MSR_IA32_MCG_CAP:
3627 		data = vcpu->arch.mcg_cap;
3628 		break;
3629 	case MSR_IA32_MCG_CTL:
3630 		if (!(mcg_cap & MCG_CTL_P) && !host)
3631 			return 1;
3632 		data = vcpu->arch.mcg_ctl;
3633 		break;
3634 	case MSR_IA32_MCG_STATUS:
3635 		data = vcpu->arch.mcg_status;
3636 		break;
3637 	default:
3638 		if (msr >= MSR_IA32_MC0_CTL &&
3639 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3640 			u32 offset = array_index_nospec(
3641 				msr - MSR_IA32_MC0_CTL,
3642 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3643 
3644 			data = vcpu->arch.mce_banks[offset];
3645 			break;
3646 		}
3647 		return 1;
3648 	}
3649 	*pdata = data;
3650 	return 0;
3651 }
3652 
3653 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3654 {
3655 	switch (msr_info->index) {
3656 	case MSR_IA32_PLATFORM_ID:
3657 	case MSR_IA32_EBL_CR_POWERON:
3658 	case MSR_IA32_LASTBRANCHFROMIP:
3659 	case MSR_IA32_LASTBRANCHTOIP:
3660 	case MSR_IA32_LASTINTFROMIP:
3661 	case MSR_IA32_LASTINTTOIP:
3662 	case MSR_AMD64_SYSCFG:
3663 	case MSR_K8_TSEG_ADDR:
3664 	case MSR_K8_TSEG_MASK:
3665 	case MSR_VM_HSAVE_PA:
3666 	case MSR_K8_INT_PENDING_MSG:
3667 	case MSR_AMD64_NB_CFG:
3668 	case MSR_FAM10H_MMIO_CONF_BASE:
3669 	case MSR_AMD64_BU_CFG2:
3670 	case MSR_IA32_PERF_CTL:
3671 	case MSR_AMD64_DC_CFG:
3672 	case MSR_F15H_EX_CFG:
3673 	/*
3674 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3675 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3676 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3677 	 * so for existing CPU-specific MSRs.
3678 	 */
3679 	case MSR_RAPL_POWER_UNIT:
3680 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3681 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3682 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3683 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3684 		msr_info->data = 0;
3685 		break;
3686 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3687 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3688 			return kvm_pmu_get_msr(vcpu, msr_info);
3689 		if (!msr_info->host_initiated)
3690 			return 1;
3691 		msr_info->data = 0;
3692 		break;
3693 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3694 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3695 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3696 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3697 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3698 			return kvm_pmu_get_msr(vcpu, msr_info);
3699 		msr_info->data = 0;
3700 		break;
3701 	case MSR_IA32_UCODE_REV:
3702 		msr_info->data = vcpu->arch.microcode_version;
3703 		break;
3704 	case MSR_IA32_ARCH_CAPABILITIES:
3705 		if (!msr_info->host_initiated &&
3706 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3707 			return 1;
3708 		msr_info->data = vcpu->arch.arch_capabilities;
3709 		break;
3710 	case MSR_IA32_PERF_CAPABILITIES:
3711 		if (!msr_info->host_initiated &&
3712 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3713 			return 1;
3714 		msr_info->data = vcpu->arch.perf_capabilities;
3715 		break;
3716 	case MSR_IA32_POWER_CTL:
3717 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3718 		break;
3719 	case MSR_IA32_TSC: {
3720 		/*
3721 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3722 		 * even when not intercepted. AMD manual doesn't explicitly
3723 		 * state this but appears to behave the same.
3724 		 *
3725 		 * On userspace reads and writes, however, we unconditionally
3726 		 * return L1's TSC value to ensure backwards-compatible
3727 		 * behavior for migration.
3728 		 */
3729 		u64 offset, ratio;
3730 
3731 		if (msr_info->host_initiated) {
3732 			offset = vcpu->arch.l1_tsc_offset;
3733 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3734 		} else {
3735 			offset = vcpu->arch.tsc_offset;
3736 			ratio = vcpu->arch.tsc_scaling_ratio;
3737 		}
3738 
3739 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3740 		break;
3741 	}
3742 	case MSR_MTRRcap:
3743 	case 0x200 ... 0x2ff:
3744 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3745 	case 0xcd: /* fsb frequency */
3746 		msr_info->data = 3;
3747 		break;
3748 		/*
3749 		 * MSR_EBC_FREQUENCY_ID
3750 		 * Conservative value valid for even the basic CPU models.
3751 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3752 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3753 		 * and 266MHz for model 3, or 4. Set Core Clock
3754 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3755 		 * 31:24) even though these are only valid for CPU
3756 		 * models > 2, however guests may end up dividing or
3757 		 * multiplying by zero otherwise.
3758 		 */
3759 	case MSR_EBC_FREQUENCY_ID:
3760 		msr_info->data = 1 << 24;
3761 		break;
3762 	case MSR_IA32_APICBASE:
3763 		msr_info->data = kvm_get_apic_base(vcpu);
3764 		break;
3765 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3766 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3767 	case MSR_IA32_TSC_DEADLINE:
3768 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3769 		break;
3770 	case MSR_IA32_TSC_ADJUST:
3771 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3772 		break;
3773 	case MSR_IA32_MISC_ENABLE:
3774 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3775 		break;
3776 	case MSR_IA32_SMBASE:
3777 		if (!msr_info->host_initiated)
3778 			return 1;
3779 		msr_info->data = vcpu->arch.smbase;
3780 		break;
3781 	case MSR_SMI_COUNT:
3782 		msr_info->data = vcpu->arch.smi_count;
3783 		break;
3784 	case MSR_IA32_PERF_STATUS:
3785 		/* TSC increment by tick */
3786 		msr_info->data = 1000ULL;
3787 		/* CPU multiplier */
3788 		msr_info->data |= (((uint64_t)4ULL) << 40);
3789 		break;
3790 	case MSR_EFER:
3791 		msr_info->data = vcpu->arch.efer;
3792 		break;
3793 	case MSR_KVM_WALL_CLOCK:
3794 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3795 			return 1;
3796 
3797 		msr_info->data = vcpu->kvm->arch.wall_clock;
3798 		break;
3799 	case MSR_KVM_WALL_CLOCK_NEW:
3800 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3801 			return 1;
3802 
3803 		msr_info->data = vcpu->kvm->arch.wall_clock;
3804 		break;
3805 	case MSR_KVM_SYSTEM_TIME:
3806 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3807 			return 1;
3808 
3809 		msr_info->data = vcpu->arch.time;
3810 		break;
3811 	case MSR_KVM_SYSTEM_TIME_NEW:
3812 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3813 			return 1;
3814 
3815 		msr_info->data = vcpu->arch.time;
3816 		break;
3817 	case MSR_KVM_ASYNC_PF_EN:
3818 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3819 			return 1;
3820 
3821 		msr_info->data = vcpu->arch.apf.msr_en_val;
3822 		break;
3823 	case MSR_KVM_ASYNC_PF_INT:
3824 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3825 			return 1;
3826 
3827 		msr_info->data = vcpu->arch.apf.msr_int_val;
3828 		break;
3829 	case MSR_KVM_ASYNC_PF_ACK:
3830 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3831 			return 1;
3832 
3833 		msr_info->data = 0;
3834 		break;
3835 	case MSR_KVM_STEAL_TIME:
3836 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3837 			return 1;
3838 
3839 		msr_info->data = vcpu->arch.st.msr_val;
3840 		break;
3841 	case MSR_KVM_PV_EOI_EN:
3842 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3843 			return 1;
3844 
3845 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3846 		break;
3847 	case MSR_KVM_POLL_CONTROL:
3848 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3849 			return 1;
3850 
3851 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3852 		break;
3853 	case MSR_IA32_P5_MC_ADDR:
3854 	case MSR_IA32_P5_MC_TYPE:
3855 	case MSR_IA32_MCG_CAP:
3856 	case MSR_IA32_MCG_CTL:
3857 	case MSR_IA32_MCG_STATUS:
3858 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3859 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3860 				   msr_info->host_initiated);
3861 	case MSR_IA32_XSS:
3862 		if (!msr_info->host_initiated &&
3863 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3864 			return 1;
3865 		msr_info->data = vcpu->arch.ia32_xss;
3866 		break;
3867 	case MSR_K7_CLK_CTL:
3868 		/*
3869 		 * Provide expected ramp-up count for K7. All other
3870 		 * are set to zero, indicating minimum divisors for
3871 		 * every field.
3872 		 *
3873 		 * This prevents guest kernels on AMD host with CPU
3874 		 * type 6, model 8 and higher from exploding due to
3875 		 * the rdmsr failing.
3876 		 */
3877 		msr_info->data = 0x20000000;
3878 		break;
3879 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3880 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3881 	case HV_X64_MSR_SYNDBG_OPTIONS:
3882 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3883 	case HV_X64_MSR_CRASH_CTL:
3884 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3885 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3886 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3887 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3888 		return kvm_hv_get_msr_common(vcpu,
3889 					     msr_info->index, &msr_info->data,
3890 					     msr_info->host_initiated);
3891 	case MSR_IA32_BBL_CR_CTL3:
3892 		/* This legacy MSR exists but isn't fully documented in current
3893 		 * silicon.  It is however accessed by winxp in very narrow
3894 		 * scenarios where it sets bit #19, itself documented as
3895 		 * a "reserved" bit.  Best effort attempt to source coherent
3896 		 * read data here should the balance of the register be
3897 		 * interpreted by the guest:
3898 		 *
3899 		 * L2 cache control register 3: 64GB range, 256KB size,
3900 		 * enabled, latency 0x1, configured
3901 		 */
3902 		msr_info->data = 0xbe702111;
3903 		break;
3904 	case MSR_AMD64_OSVW_ID_LENGTH:
3905 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3906 			return 1;
3907 		msr_info->data = vcpu->arch.osvw.length;
3908 		break;
3909 	case MSR_AMD64_OSVW_STATUS:
3910 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3911 			return 1;
3912 		msr_info->data = vcpu->arch.osvw.status;
3913 		break;
3914 	case MSR_PLATFORM_INFO:
3915 		if (!msr_info->host_initiated &&
3916 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3917 			return 1;
3918 		msr_info->data = vcpu->arch.msr_platform_info;
3919 		break;
3920 	case MSR_MISC_FEATURES_ENABLES:
3921 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3922 		break;
3923 	case MSR_K7_HWCR:
3924 		msr_info->data = vcpu->arch.msr_hwcr;
3925 		break;
3926 	default:
3927 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3928 			return kvm_pmu_get_msr(vcpu, msr_info);
3929 		return KVM_MSR_RET_INVALID;
3930 	}
3931 	return 0;
3932 }
3933 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3934 
3935 /*
3936  * Read or write a bunch of msrs. All parameters are kernel addresses.
3937  *
3938  * @return number of msrs set successfully.
3939  */
3940 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3941 		    struct kvm_msr_entry *entries,
3942 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3943 				  unsigned index, u64 *data))
3944 {
3945 	int i;
3946 
3947 	for (i = 0; i < msrs->nmsrs; ++i)
3948 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3949 			break;
3950 
3951 	return i;
3952 }
3953 
3954 /*
3955  * Read or write a bunch of msrs. Parameters are user addresses.
3956  *
3957  * @return number of msrs set successfully.
3958  */
3959 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3960 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3961 				unsigned index, u64 *data),
3962 		  int writeback)
3963 {
3964 	struct kvm_msrs msrs;
3965 	struct kvm_msr_entry *entries;
3966 	int r, n;
3967 	unsigned size;
3968 
3969 	r = -EFAULT;
3970 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3971 		goto out;
3972 
3973 	r = -E2BIG;
3974 	if (msrs.nmsrs >= MAX_IO_MSRS)
3975 		goto out;
3976 
3977 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3978 	entries = memdup_user(user_msrs->entries, size);
3979 	if (IS_ERR(entries)) {
3980 		r = PTR_ERR(entries);
3981 		goto out;
3982 	}
3983 
3984 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3985 	if (r < 0)
3986 		goto out_free;
3987 
3988 	r = -EFAULT;
3989 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3990 		goto out_free;
3991 
3992 	r = n;
3993 
3994 out_free:
3995 	kfree(entries);
3996 out:
3997 	return r;
3998 }
3999 
4000 static inline bool kvm_can_mwait_in_guest(void)
4001 {
4002 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
4003 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
4004 		boot_cpu_has(X86_FEATURE_ARAT);
4005 }
4006 
4007 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4008 					    struct kvm_cpuid2 __user *cpuid_arg)
4009 {
4010 	struct kvm_cpuid2 cpuid;
4011 	int r;
4012 
4013 	r = -EFAULT;
4014 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4015 		return r;
4016 
4017 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4018 	if (r)
4019 		return r;
4020 
4021 	r = -EFAULT;
4022 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4023 		return r;
4024 
4025 	return 0;
4026 }
4027 
4028 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4029 {
4030 	int r = 0;
4031 
4032 	switch (ext) {
4033 	case KVM_CAP_IRQCHIP:
4034 	case KVM_CAP_HLT:
4035 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4036 	case KVM_CAP_SET_TSS_ADDR:
4037 	case KVM_CAP_EXT_CPUID:
4038 	case KVM_CAP_EXT_EMUL_CPUID:
4039 	case KVM_CAP_CLOCKSOURCE:
4040 	case KVM_CAP_PIT:
4041 	case KVM_CAP_NOP_IO_DELAY:
4042 	case KVM_CAP_MP_STATE:
4043 	case KVM_CAP_SYNC_MMU:
4044 	case KVM_CAP_USER_NMI:
4045 	case KVM_CAP_REINJECT_CONTROL:
4046 	case KVM_CAP_IRQ_INJECT_STATUS:
4047 	case KVM_CAP_IOEVENTFD:
4048 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
4049 	case KVM_CAP_PIT2:
4050 	case KVM_CAP_PIT_STATE2:
4051 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4052 	case KVM_CAP_VCPU_EVENTS:
4053 	case KVM_CAP_HYPERV:
4054 	case KVM_CAP_HYPERV_VAPIC:
4055 	case KVM_CAP_HYPERV_SPIN:
4056 	case KVM_CAP_HYPERV_SYNIC:
4057 	case KVM_CAP_HYPERV_SYNIC2:
4058 	case KVM_CAP_HYPERV_VP_INDEX:
4059 	case KVM_CAP_HYPERV_EVENTFD:
4060 	case KVM_CAP_HYPERV_TLBFLUSH:
4061 	case KVM_CAP_HYPERV_SEND_IPI:
4062 	case KVM_CAP_HYPERV_CPUID:
4063 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4064 	case KVM_CAP_SYS_HYPERV_CPUID:
4065 	case KVM_CAP_PCI_SEGMENT:
4066 	case KVM_CAP_DEBUGREGS:
4067 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4068 	case KVM_CAP_XSAVE:
4069 	case KVM_CAP_ASYNC_PF:
4070 	case KVM_CAP_ASYNC_PF_INT:
4071 	case KVM_CAP_GET_TSC_KHZ:
4072 	case KVM_CAP_KVMCLOCK_CTRL:
4073 	case KVM_CAP_READONLY_MEM:
4074 	case KVM_CAP_HYPERV_TIME:
4075 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4076 	case KVM_CAP_TSC_DEADLINE_TIMER:
4077 	case KVM_CAP_DISABLE_QUIRKS:
4078 	case KVM_CAP_SET_BOOT_CPU_ID:
4079  	case KVM_CAP_SPLIT_IRQCHIP:
4080 	case KVM_CAP_IMMEDIATE_EXIT:
4081 	case KVM_CAP_PMU_EVENT_FILTER:
4082 	case KVM_CAP_GET_MSR_FEATURES:
4083 	case KVM_CAP_MSR_PLATFORM_INFO:
4084 	case KVM_CAP_EXCEPTION_PAYLOAD:
4085 	case KVM_CAP_SET_GUEST_DEBUG:
4086 	case KVM_CAP_LAST_CPU:
4087 	case KVM_CAP_X86_USER_SPACE_MSR:
4088 	case KVM_CAP_X86_MSR_FILTER:
4089 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4090 #ifdef CONFIG_X86_SGX_KVM
4091 	case KVM_CAP_SGX_ATTRIBUTE:
4092 #endif
4093 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4094 	case KVM_CAP_SREGS2:
4095 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4096 	case KVM_CAP_VCPU_ATTRIBUTES:
4097 		r = 1;
4098 		break;
4099 	case KVM_CAP_EXIT_HYPERCALL:
4100 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4101 		break;
4102 	case KVM_CAP_SET_GUEST_DEBUG2:
4103 		return KVM_GUESTDBG_VALID_MASK;
4104 #ifdef CONFIG_KVM_XEN
4105 	case KVM_CAP_XEN_HVM:
4106 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4107 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4108 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
4109 		if (sched_info_on())
4110 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4111 		break;
4112 #endif
4113 	case KVM_CAP_SYNC_REGS:
4114 		r = KVM_SYNC_X86_VALID_FIELDS;
4115 		break;
4116 	case KVM_CAP_ADJUST_CLOCK:
4117 		r = KVM_CLOCK_VALID_FLAGS;
4118 		break;
4119 	case KVM_CAP_X86_DISABLE_EXITS:
4120 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4121 		      KVM_X86_DISABLE_EXITS_CSTATE;
4122 		if(kvm_can_mwait_in_guest())
4123 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4124 		break;
4125 	case KVM_CAP_X86_SMM:
4126 		/* SMBASE is usually relocated above 1M on modern chipsets,
4127 		 * and SMM handlers might indeed rely on 4G segment limits,
4128 		 * so do not report SMM to be available if real mode is
4129 		 * emulated via vm86 mode.  Still, do not go to great lengths
4130 		 * to avoid userspace's usage of the feature, because it is a
4131 		 * fringe case that is not enabled except via specific settings
4132 		 * of the module parameters.
4133 		 */
4134 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4135 		break;
4136 	case KVM_CAP_VAPIC:
4137 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4138 		break;
4139 	case KVM_CAP_NR_VCPUS:
4140 		r = KVM_SOFT_MAX_VCPUS;
4141 		break;
4142 	case KVM_CAP_MAX_VCPUS:
4143 		r = KVM_MAX_VCPUS;
4144 		break;
4145 	case KVM_CAP_MAX_VCPU_ID:
4146 		r = KVM_MAX_VCPU_IDS;
4147 		break;
4148 	case KVM_CAP_PV_MMU:	/* obsolete */
4149 		r = 0;
4150 		break;
4151 	case KVM_CAP_MCE:
4152 		r = KVM_MAX_MCE_BANKS;
4153 		break;
4154 	case KVM_CAP_XCRS:
4155 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4156 		break;
4157 	case KVM_CAP_TSC_CONTROL:
4158 		r = kvm_has_tsc_control;
4159 		break;
4160 	case KVM_CAP_X2APIC_API:
4161 		r = KVM_X2APIC_API_VALID_FLAGS;
4162 		break;
4163 	case KVM_CAP_NESTED_STATE:
4164 		r = kvm_x86_ops.nested_ops->get_state ?
4165 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4166 		break;
4167 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4168 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4169 		break;
4170 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4171 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4172 		break;
4173 	case KVM_CAP_SMALLER_MAXPHYADDR:
4174 		r = (int) allow_smaller_maxphyaddr;
4175 		break;
4176 	case KVM_CAP_STEAL_TIME:
4177 		r = sched_info_on();
4178 		break;
4179 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4180 		if (kvm_has_bus_lock_exit)
4181 			r = KVM_BUS_LOCK_DETECTION_OFF |
4182 			    KVM_BUS_LOCK_DETECTION_EXIT;
4183 		else
4184 			r = 0;
4185 		break;
4186 	default:
4187 		break;
4188 	}
4189 	return r;
4190 
4191 }
4192 
4193 long kvm_arch_dev_ioctl(struct file *filp,
4194 			unsigned int ioctl, unsigned long arg)
4195 {
4196 	void __user *argp = (void __user *)arg;
4197 	long r;
4198 
4199 	switch (ioctl) {
4200 	case KVM_GET_MSR_INDEX_LIST: {
4201 		struct kvm_msr_list __user *user_msr_list = argp;
4202 		struct kvm_msr_list msr_list;
4203 		unsigned n;
4204 
4205 		r = -EFAULT;
4206 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4207 			goto out;
4208 		n = msr_list.nmsrs;
4209 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4210 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4211 			goto out;
4212 		r = -E2BIG;
4213 		if (n < msr_list.nmsrs)
4214 			goto out;
4215 		r = -EFAULT;
4216 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4217 				 num_msrs_to_save * sizeof(u32)))
4218 			goto out;
4219 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4220 				 &emulated_msrs,
4221 				 num_emulated_msrs * sizeof(u32)))
4222 			goto out;
4223 		r = 0;
4224 		break;
4225 	}
4226 	case KVM_GET_SUPPORTED_CPUID:
4227 	case KVM_GET_EMULATED_CPUID: {
4228 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4229 		struct kvm_cpuid2 cpuid;
4230 
4231 		r = -EFAULT;
4232 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4233 			goto out;
4234 
4235 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4236 					    ioctl);
4237 		if (r)
4238 			goto out;
4239 
4240 		r = -EFAULT;
4241 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4242 			goto out;
4243 		r = 0;
4244 		break;
4245 	}
4246 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4247 		r = -EFAULT;
4248 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4249 				 sizeof(kvm_mce_cap_supported)))
4250 			goto out;
4251 		r = 0;
4252 		break;
4253 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4254 		struct kvm_msr_list __user *user_msr_list = argp;
4255 		struct kvm_msr_list msr_list;
4256 		unsigned int n;
4257 
4258 		r = -EFAULT;
4259 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4260 			goto out;
4261 		n = msr_list.nmsrs;
4262 		msr_list.nmsrs = num_msr_based_features;
4263 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4264 			goto out;
4265 		r = -E2BIG;
4266 		if (n < msr_list.nmsrs)
4267 			goto out;
4268 		r = -EFAULT;
4269 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4270 				 num_msr_based_features * sizeof(u32)))
4271 			goto out;
4272 		r = 0;
4273 		break;
4274 	}
4275 	case KVM_GET_MSRS:
4276 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4277 		break;
4278 	case KVM_GET_SUPPORTED_HV_CPUID:
4279 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4280 		break;
4281 	default:
4282 		r = -EINVAL;
4283 		break;
4284 	}
4285 out:
4286 	return r;
4287 }
4288 
4289 static void wbinvd_ipi(void *garbage)
4290 {
4291 	wbinvd();
4292 }
4293 
4294 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4295 {
4296 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4297 }
4298 
4299 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4300 {
4301 	/* Address WBINVD may be executed by guest */
4302 	if (need_emulate_wbinvd(vcpu)) {
4303 		if (static_call(kvm_x86_has_wbinvd_exit)())
4304 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4305 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4306 			smp_call_function_single(vcpu->cpu,
4307 					wbinvd_ipi, NULL, 1);
4308 	}
4309 
4310 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4311 
4312 	/* Save host pkru register if supported */
4313 	vcpu->arch.host_pkru = read_pkru();
4314 
4315 	/* Apply any externally detected TSC adjustments (due to suspend) */
4316 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4317 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4318 		vcpu->arch.tsc_offset_adjustment = 0;
4319 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4320 	}
4321 
4322 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4323 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4324 				rdtsc() - vcpu->arch.last_host_tsc;
4325 		if (tsc_delta < 0)
4326 			mark_tsc_unstable("KVM discovered backwards TSC");
4327 
4328 		if (kvm_check_tsc_unstable()) {
4329 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4330 						vcpu->arch.last_guest_tsc);
4331 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4332 			vcpu->arch.tsc_catchup = 1;
4333 		}
4334 
4335 		if (kvm_lapic_hv_timer_in_use(vcpu))
4336 			kvm_lapic_restart_hv_timer(vcpu);
4337 
4338 		/*
4339 		 * On a host with synchronized TSC, there is no need to update
4340 		 * kvmclock on vcpu->cpu migration
4341 		 */
4342 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4343 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4344 		if (vcpu->cpu != cpu)
4345 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4346 		vcpu->cpu = cpu;
4347 	}
4348 
4349 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4350 }
4351 
4352 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4353 {
4354 	struct kvm_host_map map;
4355 	struct kvm_steal_time *st;
4356 
4357 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4358 		return;
4359 
4360 	if (vcpu->arch.st.preempted)
4361 		return;
4362 
4363 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4364 			&vcpu->arch.st.cache, true))
4365 		return;
4366 
4367 	st = map.hva +
4368 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4369 
4370 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4371 
4372 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4373 }
4374 
4375 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4376 {
4377 	int idx;
4378 
4379 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4380 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4381 
4382 	/*
4383 	 * Take the srcu lock as memslots will be accessed to check the gfn
4384 	 * cache generation against the memslots generation.
4385 	 */
4386 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4387 	if (kvm_xen_msr_enabled(vcpu->kvm))
4388 		kvm_xen_runstate_set_preempted(vcpu);
4389 	else
4390 		kvm_steal_time_set_preempted(vcpu);
4391 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4392 
4393 	static_call(kvm_x86_vcpu_put)(vcpu);
4394 	vcpu->arch.last_host_tsc = rdtsc();
4395 }
4396 
4397 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4398 				    struct kvm_lapic_state *s)
4399 {
4400 	if (vcpu->arch.apicv_active)
4401 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4402 
4403 	return kvm_apic_get_state(vcpu, s);
4404 }
4405 
4406 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4407 				    struct kvm_lapic_state *s)
4408 {
4409 	int r;
4410 
4411 	r = kvm_apic_set_state(vcpu, s);
4412 	if (r)
4413 		return r;
4414 	update_cr8_intercept(vcpu);
4415 
4416 	return 0;
4417 }
4418 
4419 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4420 {
4421 	/*
4422 	 * We can accept userspace's request for interrupt injection
4423 	 * as long as we have a place to store the interrupt number.
4424 	 * The actual injection will happen when the CPU is able to
4425 	 * deliver the interrupt.
4426 	 */
4427 	if (kvm_cpu_has_extint(vcpu))
4428 		return false;
4429 
4430 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4431 	return (!lapic_in_kernel(vcpu) ||
4432 		kvm_apic_accept_pic_intr(vcpu));
4433 }
4434 
4435 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4436 {
4437 	/*
4438 	 * Do not cause an interrupt window exit if an exception
4439 	 * is pending or an event needs reinjection; userspace
4440 	 * might want to inject the interrupt manually using KVM_SET_REGS
4441 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4442 	 * instruction boundary and with no events half-injected.
4443 	 */
4444 	return (kvm_arch_interrupt_allowed(vcpu) &&
4445 		kvm_cpu_accept_dm_intr(vcpu) &&
4446 		!kvm_event_needs_reinjection(vcpu) &&
4447 		!vcpu->arch.exception.pending);
4448 }
4449 
4450 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4451 				    struct kvm_interrupt *irq)
4452 {
4453 	if (irq->irq >= KVM_NR_INTERRUPTS)
4454 		return -EINVAL;
4455 
4456 	if (!irqchip_in_kernel(vcpu->kvm)) {
4457 		kvm_queue_interrupt(vcpu, irq->irq, false);
4458 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4459 		return 0;
4460 	}
4461 
4462 	/*
4463 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4464 	 * fail for in-kernel 8259.
4465 	 */
4466 	if (pic_in_kernel(vcpu->kvm))
4467 		return -ENXIO;
4468 
4469 	if (vcpu->arch.pending_external_vector != -1)
4470 		return -EEXIST;
4471 
4472 	vcpu->arch.pending_external_vector = irq->irq;
4473 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4474 	return 0;
4475 }
4476 
4477 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4478 {
4479 	kvm_inject_nmi(vcpu);
4480 
4481 	return 0;
4482 }
4483 
4484 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4485 {
4486 	kvm_make_request(KVM_REQ_SMI, vcpu);
4487 
4488 	return 0;
4489 }
4490 
4491 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4492 					   struct kvm_tpr_access_ctl *tac)
4493 {
4494 	if (tac->flags)
4495 		return -EINVAL;
4496 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4497 	return 0;
4498 }
4499 
4500 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4501 					u64 mcg_cap)
4502 {
4503 	int r;
4504 	unsigned bank_num = mcg_cap & 0xff, bank;
4505 
4506 	r = -EINVAL;
4507 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4508 		goto out;
4509 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4510 		goto out;
4511 	r = 0;
4512 	vcpu->arch.mcg_cap = mcg_cap;
4513 	/* Init IA32_MCG_CTL to all 1s */
4514 	if (mcg_cap & MCG_CTL_P)
4515 		vcpu->arch.mcg_ctl = ~(u64)0;
4516 	/* Init IA32_MCi_CTL to all 1s */
4517 	for (bank = 0; bank < bank_num; bank++)
4518 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4519 
4520 	static_call(kvm_x86_setup_mce)(vcpu);
4521 out:
4522 	return r;
4523 }
4524 
4525 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4526 				      struct kvm_x86_mce *mce)
4527 {
4528 	u64 mcg_cap = vcpu->arch.mcg_cap;
4529 	unsigned bank_num = mcg_cap & 0xff;
4530 	u64 *banks = vcpu->arch.mce_banks;
4531 
4532 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4533 		return -EINVAL;
4534 	/*
4535 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4536 	 * reporting is disabled
4537 	 */
4538 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4539 	    vcpu->arch.mcg_ctl != ~(u64)0)
4540 		return 0;
4541 	banks += 4 * mce->bank;
4542 	/*
4543 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4544 	 * reporting is disabled for the bank
4545 	 */
4546 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4547 		return 0;
4548 	if (mce->status & MCI_STATUS_UC) {
4549 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4550 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4551 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4552 			return 0;
4553 		}
4554 		if (banks[1] & MCI_STATUS_VAL)
4555 			mce->status |= MCI_STATUS_OVER;
4556 		banks[2] = mce->addr;
4557 		banks[3] = mce->misc;
4558 		vcpu->arch.mcg_status = mce->mcg_status;
4559 		banks[1] = mce->status;
4560 		kvm_queue_exception(vcpu, MC_VECTOR);
4561 	} else if (!(banks[1] & MCI_STATUS_VAL)
4562 		   || !(banks[1] & MCI_STATUS_UC)) {
4563 		if (banks[1] & MCI_STATUS_VAL)
4564 			mce->status |= MCI_STATUS_OVER;
4565 		banks[2] = mce->addr;
4566 		banks[3] = mce->misc;
4567 		banks[1] = mce->status;
4568 	} else
4569 		banks[1] |= MCI_STATUS_OVER;
4570 	return 0;
4571 }
4572 
4573 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4574 					       struct kvm_vcpu_events *events)
4575 {
4576 	process_nmi(vcpu);
4577 
4578 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4579 		process_smi(vcpu);
4580 
4581 	/*
4582 	 * In guest mode, payload delivery should be deferred,
4583 	 * so that the L1 hypervisor can intercept #PF before
4584 	 * CR2 is modified (or intercept #DB before DR6 is
4585 	 * modified under nVMX). Unless the per-VM capability,
4586 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4587 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4588 	 * opportunistically defer the exception payload, deliver it if the
4589 	 * capability hasn't been requested before processing a
4590 	 * KVM_GET_VCPU_EVENTS.
4591 	 */
4592 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4593 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4594 		kvm_deliver_exception_payload(vcpu);
4595 
4596 	/*
4597 	 * The API doesn't provide the instruction length for software
4598 	 * exceptions, so don't report them. As long as the guest RIP
4599 	 * isn't advanced, we should expect to encounter the exception
4600 	 * again.
4601 	 */
4602 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4603 		events->exception.injected = 0;
4604 		events->exception.pending = 0;
4605 	} else {
4606 		events->exception.injected = vcpu->arch.exception.injected;
4607 		events->exception.pending = vcpu->arch.exception.pending;
4608 		/*
4609 		 * For ABI compatibility, deliberately conflate
4610 		 * pending and injected exceptions when
4611 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4612 		 */
4613 		if (!vcpu->kvm->arch.exception_payload_enabled)
4614 			events->exception.injected |=
4615 				vcpu->arch.exception.pending;
4616 	}
4617 	events->exception.nr = vcpu->arch.exception.nr;
4618 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4619 	events->exception.error_code = vcpu->arch.exception.error_code;
4620 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4621 	events->exception_payload = vcpu->arch.exception.payload;
4622 
4623 	events->interrupt.injected =
4624 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4625 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4626 	events->interrupt.soft = 0;
4627 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4628 
4629 	events->nmi.injected = vcpu->arch.nmi_injected;
4630 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4631 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4632 	events->nmi.pad = 0;
4633 
4634 	events->sipi_vector = 0; /* never valid when reporting to user space */
4635 
4636 	events->smi.smm = is_smm(vcpu);
4637 	events->smi.pending = vcpu->arch.smi_pending;
4638 	events->smi.smm_inside_nmi =
4639 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4640 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4641 
4642 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4643 			 | KVM_VCPUEVENT_VALID_SHADOW
4644 			 | KVM_VCPUEVENT_VALID_SMM);
4645 	if (vcpu->kvm->arch.exception_payload_enabled)
4646 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4647 
4648 	memset(&events->reserved, 0, sizeof(events->reserved));
4649 }
4650 
4651 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4652 
4653 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4654 					      struct kvm_vcpu_events *events)
4655 {
4656 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4657 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4658 			      | KVM_VCPUEVENT_VALID_SHADOW
4659 			      | KVM_VCPUEVENT_VALID_SMM
4660 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4661 		return -EINVAL;
4662 
4663 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4664 		if (!vcpu->kvm->arch.exception_payload_enabled)
4665 			return -EINVAL;
4666 		if (events->exception.pending)
4667 			events->exception.injected = 0;
4668 		else
4669 			events->exception_has_payload = 0;
4670 	} else {
4671 		events->exception.pending = 0;
4672 		events->exception_has_payload = 0;
4673 	}
4674 
4675 	if ((events->exception.injected || events->exception.pending) &&
4676 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4677 		return -EINVAL;
4678 
4679 	/* INITs are latched while in SMM */
4680 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4681 	    (events->smi.smm || events->smi.pending) &&
4682 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4683 		return -EINVAL;
4684 
4685 	process_nmi(vcpu);
4686 	vcpu->arch.exception.injected = events->exception.injected;
4687 	vcpu->arch.exception.pending = events->exception.pending;
4688 	vcpu->arch.exception.nr = events->exception.nr;
4689 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4690 	vcpu->arch.exception.error_code = events->exception.error_code;
4691 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4692 	vcpu->arch.exception.payload = events->exception_payload;
4693 
4694 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4695 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4696 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4697 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4698 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4699 						events->interrupt.shadow);
4700 
4701 	vcpu->arch.nmi_injected = events->nmi.injected;
4702 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4703 		vcpu->arch.nmi_pending = events->nmi.pending;
4704 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4705 
4706 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4707 	    lapic_in_kernel(vcpu))
4708 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4709 
4710 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4711 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4712 			kvm_smm_changed(vcpu, events->smi.smm);
4713 
4714 		vcpu->arch.smi_pending = events->smi.pending;
4715 
4716 		if (events->smi.smm) {
4717 			if (events->smi.smm_inside_nmi)
4718 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4719 			else
4720 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4721 		}
4722 
4723 		if (lapic_in_kernel(vcpu)) {
4724 			if (events->smi.latched_init)
4725 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4726 			else
4727 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4728 		}
4729 	}
4730 
4731 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4732 
4733 	return 0;
4734 }
4735 
4736 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4737 					     struct kvm_debugregs *dbgregs)
4738 {
4739 	unsigned long val;
4740 
4741 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4742 	kvm_get_dr(vcpu, 6, &val);
4743 	dbgregs->dr6 = val;
4744 	dbgregs->dr7 = vcpu->arch.dr7;
4745 	dbgregs->flags = 0;
4746 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4747 }
4748 
4749 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4750 					    struct kvm_debugregs *dbgregs)
4751 {
4752 	if (dbgregs->flags)
4753 		return -EINVAL;
4754 
4755 	if (!kvm_dr6_valid(dbgregs->dr6))
4756 		return -EINVAL;
4757 	if (!kvm_dr7_valid(dbgregs->dr7))
4758 		return -EINVAL;
4759 
4760 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4761 	kvm_update_dr0123(vcpu);
4762 	vcpu->arch.dr6 = dbgregs->dr6;
4763 	vcpu->arch.dr7 = dbgregs->dr7;
4764 	kvm_update_dr7(vcpu);
4765 
4766 	return 0;
4767 }
4768 
4769 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4770 					 struct kvm_xsave *guest_xsave)
4771 {
4772 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4773 		return;
4774 
4775 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4776 				       guest_xsave->region,
4777 				       sizeof(guest_xsave->region),
4778 				       vcpu->arch.pkru);
4779 }
4780 
4781 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4782 					struct kvm_xsave *guest_xsave)
4783 {
4784 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4785 		return 0;
4786 
4787 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
4788 					      guest_xsave->region,
4789 					      supported_xcr0, &vcpu->arch.pkru);
4790 }
4791 
4792 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4793 					struct kvm_xcrs *guest_xcrs)
4794 {
4795 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4796 		guest_xcrs->nr_xcrs = 0;
4797 		return;
4798 	}
4799 
4800 	guest_xcrs->nr_xcrs = 1;
4801 	guest_xcrs->flags = 0;
4802 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4803 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4804 }
4805 
4806 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4807 				       struct kvm_xcrs *guest_xcrs)
4808 {
4809 	int i, r = 0;
4810 
4811 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4812 		return -EINVAL;
4813 
4814 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4815 		return -EINVAL;
4816 
4817 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4818 		/* Only support XCR0 currently */
4819 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4820 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4821 				guest_xcrs->xcrs[i].value);
4822 			break;
4823 		}
4824 	if (r)
4825 		r = -EINVAL;
4826 	return r;
4827 }
4828 
4829 /*
4830  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4831  * stopped by the hypervisor.  This function will be called from the host only.
4832  * EINVAL is returned when the host attempts to set the flag for a guest that
4833  * does not support pv clocks.
4834  */
4835 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4836 {
4837 	if (!vcpu->arch.pv_time_enabled)
4838 		return -EINVAL;
4839 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4840 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4841 	return 0;
4842 }
4843 
4844 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
4845 				 struct kvm_device_attr *attr)
4846 {
4847 	int r;
4848 
4849 	switch (attr->attr) {
4850 	case KVM_VCPU_TSC_OFFSET:
4851 		r = 0;
4852 		break;
4853 	default:
4854 		r = -ENXIO;
4855 	}
4856 
4857 	return r;
4858 }
4859 
4860 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
4861 				 struct kvm_device_attr *attr)
4862 {
4863 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4864 	int r;
4865 
4866 	if ((u64)(unsigned long)uaddr != attr->addr)
4867 		return -EFAULT;
4868 
4869 	switch (attr->attr) {
4870 	case KVM_VCPU_TSC_OFFSET:
4871 		r = -EFAULT;
4872 		if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
4873 			break;
4874 		r = 0;
4875 		break;
4876 	default:
4877 		r = -ENXIO;
4878 	}
4879 
4880 	return r;
4881 }
4882 
4883 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
4884 				 struct kvm_device_attr *attr)
4885 {
4886 	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4887 	struct kvm *kvm = vcpu->kvm;
4888 	int r;
4889 
4890 	if ((u64)(unsigned long)uaddr != attr->addr)
4891 		return -EFAULT;
4892 
4893 	switch (attr->attr) {
4894 	case KVM_VCPU_TSC_OFFSET: {
4895 		u64 offset, tsc, ns;
4896 		unsigned long flags;
4897 		bool matched;
4898 
4899 		r = -EFAULT;
4900 		if (get_user(offset, uaddr))
4901 			break;
4902 
4903 		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
4904 
4905 		matched = (vcpu->arch.virtual_tsc_khz &&
4906 			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
4907 			   kvm->arch.last_tsc_offset == offset);
4908 
4909 		tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
4910 		ns = get_kvmclock_base_ns();
4911 
4912 		__kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
4913 		raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
4914 
4915 		r = 0;
4916 		break;
4917 	}
4918 	default:
4919 		r = -ENXIO;
4920 	}
4921 
4922 	return r;
4923 }
4924 
4925 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
4926 				      unsigned int ioctl,
4927 				      void __user *argp)
4928 {
4929 	struct kvm_device_attr attr;
4930 	int r;
4931 
4932 	if (copy_from_user(&attr, argp, sizeof(attr)))
4933 		return -EFAULT;
4934 
4935 	if (attr.group != KVM_VCPU_TSC_CTRL)
4936 		return -ENXIO;
4937 
4938 	switch (ioctl) {
4939 	case KVM_HAS_DEVICE_ATTR:
4940 		r = kvm_arch_tsc_has_attr(vcpu, &attr);
4941 		break;
4942 	case KVM_GET_DEVICE_ATTR:
4943 		r = kvm_arch_tsc_get_attr(vcpu, &attr);
4944 		break;
4945 	case KVM_SET_DEVICE_ATTR:
4946 		r = kvm_arch_tsc_set_attr(vcpu, &attr);
4947 		break;
4948 	}
4949 
4950 	return r;
4951 }
4952 
4953 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4954 				     struct kvm_enable_cap *cap)
4955 {
4956 	int r;
4957 	uint16_t vmcs_version;
4958 	void __user *user_ptr;
4959 
4960 	if (cap->flags)
4961 		return -EINVAL;
4962 
4963 	switch (cap->cap) {
4964 	case KVM_CAP_HYPERV_SYNIC2:
4965 		if (cap->args[0])
4966 			return -EINVAL;
4967 		fallthrough;
4968 
4969 	case KVM_CAP_HYPERV_SYNIC:
4970 		if (!irqchip_in_kernel(vcpu->kvm))
4971 			return -EINVAL;
4972 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4973 					     KVM_CAP_HYPERV_SYNIC2);
4974 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4975 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4976 			return -ENOTTY;
4977 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4978 		if (!r) {
4979 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4980 			if (copy_to_user(user_ptr, &vmcs_version,
4981 					 sizeof(vmcs_version)))
4982 				r = -EFAULT;
4983 		}
4984 		return r;
4985 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4986 		if (!kvm_x86_ops.enable_direct_tlbflush)
4987 			return -ENOTTY;
4988 
4989 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4990 
4991 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4992 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4993 
4994 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4995 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4996 		if (vcpu->arch.pv_cpuid.enforce)
4997 			kvm_update_pv_runtime(vcpu);
4998 
4999 		return 0;
5000 	default:
5001 		return -EINVAL;
5002 	}
5003 }
5004 
5005 long kvm_arch_vcpu_ioctl(struct file *filp,
5006 			 unsigned int ioctl, unsigned long arg)
5007 {
5008 	struct kvm_vcpu *vcpu = filp->private_data;
5009 	void __user *argp = (void __user *)arg;
5010 	int r;
5011 	union {
5012 		struct kvm_sregs2 *sregs2;
5013 		struct kvm_lapic_state *lapic;
5014 		struct kvm_xsave *xsave;
5015 		struct kvm_xcrs *xcrs;
5016 		void *buffer;
5017 	} u;
5018 
5019 	vcpu_load(vcpu);
5020 
5021 	u.buffer = NULL;
5022 	switch (ioctl) {
5023 	case KVM_GET_LAPIC: {
5024 		r = -EINVAL;
5025 		if (!lapic_in_kernel(vcpu))
5026 			goto out;
5027 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5028 				GFP_KERNEL_ACCOUNT);
5029 
5030 		r = -ENOMEM;
5031 		if (!u.lapic)
5032 			goto out;
5033 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5034 		if (r)
5035 			goto out;
5036 		r = -EFAULT;
5037 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5038 			goto out;
5039 		r = 0;
5040 		break;
5041 	}
5042 	case KVM_SET_LAPIC: {
5043 		r = -EINVAL;
5044 		if (!lapic_in_kernel(vcpu))
5045 			goto out;
5046 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
5047 		if (IS_ERR(u.lapic)) {
5048 			r = PTR_ERR(u.lapic);
5049 			goto out_nofree;
5050 		}
5051 
5052 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5053 		break;
5054 	}
5055 	case KVM_INTERRUPT: {
5056 		struct kvm_interrupt irq;
5057 
5058 		r = -EFAULT;
5059 		if (copy_from_user(&irq, argp, sizeof(irq)))
5060 			goto out;
5061 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5062 		break;
5063 	}
5064 	case KVM_NMI: {
5065 		r = kvm_vcpu_ioctl_nmi(vcpu);
5066 		break;
5067 	}
5068 	case KVM_SMI: {
5069 		r = kvm_vcpu_ioctl_smi(vcpu);
5070 		break;
5071 	}
5072 	case KVM_SET_CPUID: {
5073 		struct kvm_cpuid __user *cpuid_arg = argp;
5074 		struct kvm_cpuid cpuid;
5075 
5076 		r = -EFAULT;
5077 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5078 			goto out;
5079 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5080 		break;
5081 	}
5082 	case KVM_SET_CPUID2: {
5083 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5084 		struct kvm_cpuid2 cpuid;
5085 
5086 		r = -EFAULT;
5087 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5088 			goto out;
5089 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5090 					      cpuid_arg->entries);
5091 		break;
5092 	}
5093 	case KVM_GET_CPUID2: {
5094 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5095 		struct kvm_cpuid2 cpuid;
5096 
5097 		r = -EFAULT;
5098 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5099 			goto out;
5100 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5101 					      cpuid_arg->entries);
5102 		if (r)
5103 			goto out;
5104 		r = -EFAULT;
5105 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5106 			goto out;
5107 		r = 0;
5108 		break;
5109 	}
5110 	case KVM_GET_MSRS: {
5111 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5112 		r = msr_io(vcpu, argp, do_get_msr, 1);
5113 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5114 		break;
5115 	}
5116 	case KVM_SET_MSRS: {
5117 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5118 		r = msr_io(vcpu, argp, do_set_msr, 0);
5119 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5120 		break;
5121 	}
5122 	case KVM_TPR_ACCESS_REPORTING: {
5123 		struct kvm_tpr_access_ctl tac;
5124 
5125 		r = -EFAULT;
5126 		if (copy_from_user(&tac, argp, sizeof(tac)))
5127 			goto out;
5128 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5129 		if (r)
5130 			goto out;
5131 		r = -EFAULT;
5132 		if (copy_to_user(argp, &tac, sizeof(tac)))
5133 			goto out;
5134 		r = 0;
5135 		break;
5136 	};
5137 	case KVM_SET_VAPIC_ADDR: {
5138 		struct kvm_vapic_addr va;
5139 		int idx;
5140 
5141 		r = -EINVAL;
5142 		if (!lapic_in_kernel(vcpu))
5143 			goto out;
5144 		r = -EFAULT;
5145 		if (copy_from_user(&va, argp, sizeof(va)))
5146 			goto out;
5147 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5148 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5149 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5150 		break;
5151 	}
5152 	case KVM_X86_SETUP_MCE: {
5153 		u64 mcg_cap;
5154 
5155 		r = -EFAULT;
5156 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5157 			goto out;
5158 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5159 		break;
5160 	}
5161 	case KVM_X86_SET_MCE: {
5162 		struct kvm_x86_mce mce;
5163 
5164 		r = -EFAULT;
5165 		if (copy_from_user(&mce, argp, sizeof(mce)))
5166 			goto out;
5167 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5168 		break;
5169 	}
5170 	case KVM_GET_VCPU_EVENTS: {
5171 		struct kvm_vcpu_events events;
5172 
5173 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5174 
5175 		r = -EFAULT;
5176 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5177 			break;
5178 		r = 0;
5179 		break;
5180 	}
5181 	case KVM_SET_VCPU_EVENTS: {
5182 		struct kvm_vcpu_events events;
5183 
5184 		r = -EFAULT;
5185 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5186 			break;
5187 
5188 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5189 		break;
5190 	}
5191 	case KVM_GET_DEBUGREGS: {
5192 		struct kvm_debugregs dbgregs;
5193 
5194 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5195 
5196 		r = -EFAULT;
5197 		if (copy_to_user(argp, &dbgregs,
5198 				 sizeof(struct kvm_debugregs)))
5199 			break;
5200 		r = 0;
5201 		break;
5202 	}
5203 	case KVM_SET_DEBUGREGS: {
5204 		struct kvm_debugregs dbgregs;
5205 
5206 		r = -EFAULT;
5207 		if (copy_from_user(&dbgregs, argp,
5208 				   sizeof(struct kvm_debugregs)))
5209 			break;
5210 
5211 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5212 		break;
5213 	}
5214 	case KVM_GET_XSAVE: {
5215 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5216 		r = -ENOMEM;
5217 		if (!u.xsave)
5218 			break;
5219 
5220 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5221 
5222 		r = -EFAULT;
5223 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5224 			break;
5225 		r = 0;
5226 		break;
5227 	}
5228 	case KVM_SET_XSAVE: {
5229 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5230 		if (IS_ERR(u.xsave)) {
5231 			r = PTR_ERR(u.xsave);
5232 			goto out_nofree;
5233 		}
5234 
5235 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5236 		break;
5237 	}
5238 	case KVM_GET_XCRS: {
5239 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5240 		r = -ENOMEM;
5241 		if (!u.xcrs)
5242 			break;
5243 
5244 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5245 
5246 		r = -EFAULT;
5247 		if (copy_to_user(argp, u.xcrs,
5248 				 sizeof(struct kvm_xcrs)))
5249 			break;
5250 		r = 0;
5251 		break;
5252 	}
5253 	case KVM_SET_XCRS: {
5254 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5255 		if (IS_ERR(u.xcrs)) {
5256 			r = PTR_ERR(u.xcrs);
5257 			goto out_nofree;
5258 		}
5259 
5260 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5261 		break;
5262 	}
5263 	case KVM_SET_TSC_KHZ: {
5264 		u32 user_tsc_khz;
5265 
5266 		r = -EINVAL;
5267 		user_tsc_khz = (u32)arg;
5268 
5269 		if (kvm_has_tsc_control &&
5270 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5271 			goto out;
5272 
5273 		if (user_tsc_khz == 0)
5274 			user_tsc_khz = tsc_khz;
5275 
5276 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5277 			r = 0;
5278 
5279 		goto out;
5280 	}
5281 	case KVM_GET_TSC_KHZ: {
5282 		r = vcpu->arch.virtual_tsc_khz;
5283 		goto out;
5284 	}
5285 	case KVM_KVMCLOCK_CTRL: {
5286 		r = kvm_set_guest_paused(vcpu);
5287 		goto out;
5288 	}
5289 	case KVM_ENABLE_CAP: {
5290 		struct kvm_enable_cap cap;
5291 
5292 		r = -EFAULT;
5293 		if (copy_from_user(&cap, argp, sizeof(cap)))
5294 			goto out;
5295 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5296 		break;
5297 	}
5298 	case KVM_GET_NESTED_STATE: {
5299 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5300 		u32 user_data_size;
5301 
5302 		r = -EINVAL;
5303 		if (!kvm_x86_ops.nested_ops->get_state)
5304 			break;
5305 
5306 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5307 		r = -EFAULT;
5308 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5309 			break;
5310 
5311 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5312 						     user_data_size);
5313 		if (r < 0)
5314 			break;
5315 
5316 		if (r > user_data_size) {
5317 			if (put_user(r, &user_kvm_nested_state->size))
5318 				r = -EFAULT;
5319 			else
5320 				r = -E2BIG;
5321 			break;
5322 		}
5323 
5324 		r = 0;
5325 		break;
5326 	}
5327 	case KVM_SET_NESTED_STATE: {
5328 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5329 		struct kvm_nested_state kvm_state;
5330 		int idx;
5331 
5332 		r = -EINVAL;
5333 		if (!kvm_x86_ops.nested_ops->set_state)
5334 			break;
5335 
5336 		r = -EFAULT;
5337 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5338 			break;
5339 
5340 		r = -EINVAL;
5341 		if (kvm_state.size < sizeof(kvm_state))
5342 			break;
5343 
5344 		if (kvm_state.flags &
5345 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5346 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5347 		      | KVM_STATE_NESTED_GIF_SET))
5348 			break;
5349 
5350 		/* nested_run_pending implies guest_mode.  */
5351 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5352 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5353 			break;
5354 
5355 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5356 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5357 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5358 		break;
5359 	}
5360 	case KVM_GET_SUPPORTED_HV_CPUID:
5361 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5362 		break;
5363 #ifdef CONFIG_KVM_XEN
5364 	case KVM_XEN_VCPU_GET_ATTR: {
5365 		struct kvm_xen_vcpu_attr xva;
5366 
5367 		r = -EFAULT;
5368 		if (copy_from_user(&xva, argp, sizeof(xva)))
5369 			goto out;
5370 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5371 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5372 			r = -EFAULT;
5373 		break;
5374 	}
5375 	case KVM_XEN_VCPU_SET_ATTR: {
5376 		struct kvm_xen_vcpu_attr xva;
5377 
5378 		r = -EFAULT;
5379 		if (copy_from_user(&xva, argp, sizeof(xva)))
5380 			goto out;
5381 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5382 		break;
5383 	}
5384 #endif
5385 	case KVM_GET_SREGS2: {
5386 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5387 		r = -ENOMEM;
5388 		if (!u.sregs2)
5389 			goto out;
5390 		__get_sregs2(vcpu, u.sregs2);
5391 		r = -EFAULT;
5392 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5393 			goto out;
5394 		r = 0;
5395 		break;
5396 	}
5397 	case KVM_SET_SREGS2: {
5398 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5399 		if (IS_ERR(u.sregs2)) {
5400 			r = PTR_ERR(u.sregs2);
5401 			u.sregs2 = NULL;
5402 			goto out;
5403 		}
5404 		r = __set_sregs2(vcpu, u.sregs2);
5405 		break;
5406 	}
5407 	case KVM_HAS_DEVICE_ATTR:
5408 	case KVM_GET_DEVICE_ATTR:
5409 	case KVM_SET_DEVICE_ATTR:
5410 		r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5411 		break;
5412 	default:
5413 		r = -EINVAL;
5414 	}
5415 out:
5416 	kfree(u.buffer);
5417 out_nofree:
5418 	vcpu_put(vcpu);
5419 	return r;
5420 }
5421 
5422 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5423 {
5424 	return VM_FAULT_SIGBUS;
5425 }
5426 
5427 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5428 {
5429 	int ret;
5430 
5431 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5432 		return -EINVAL;
5433 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5434 	return ret;
5435 }
5436 
5437 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5438 					      u64 ident_addr)
5439 {
5440 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5441 }
5442 
5443 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5444 					 unsigned long kvm_nr_mmu_pages)
5445 {
5446 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5447 		return -EINVAL;
5448 
5449 	mutex_lock(&kvm->slots_lock);
5450 
5451 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5452 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5453 
5454 	mutex_unlock(&kvm->slots_lock);
5455 	return 0;
5456 }
5457 
5458 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5459 {
5460 	return kvm->arch.n_max_mmu_pages;
5461 }
5462 
5463 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5464 {
5465 	struct kvm_pic *pic = kvm->arch.vpic;
5466 	int r;
5467 
5468 	r = 0;
5469 	switch (chip->chip_id) {
5470 	case KVM_IRQCHIP_PIC_MASTER:
5471 		memcpy(&chip->chip.pic, &pic->pics[0],
5472 			sizeof(struct kvm_pic_state));
5473 		break;
5474 	case KVM_IRQCHIP_PIC_SLAVE:
5475 		memcpy(&chip->chip.pic, &pic->pics[1],
5476 			sizeof(struct kvm_pic_state));
5477 		break;
5478 	case KVM_IRQCHIP_IOAPIC:
5479 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5480 		break;
5481 	default:
5482 		r = -EINVAL;
5483 		break;
5484 	}
5485 	return r;
5486 }
5487 
5488 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5489 {
5490 	struct kvm_pic *pic = kvm->arch.vpic;
5491 	int r;
5492 
5493 	r = 0;
5494 	switch (chip->chip_id) {
5495 	case KVM_IRQCHIP_PIC_MASTER:
5496 		spin_lock(&pic->lock);
5497 		memcpy(&pic->pics[0], &chip->chip.pic,
5498 			sizeof(struct kvm_pic_state));
5499 		spin_unlock(&pic->lock);
5500 		break;
5501 	case KVM_IRQCHIP_PIC_SLAVE:
5502 		spin_lock(&pic->lock);
5503 		memcpy(&pic->pics[1], &chip->chip.pic,
5504 			sizeof(struct kvm_pic_state));
5505 		spin_unlock(&pic->lock);
5506 		break;
5507 	case KVM_IRQCHIP_IOAPIC:
5508 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5509 		break;
5510 	default:
5511 		r = -EINVAL;
5512 		break;
5513 	}
5514 	kvm_pic_update_irq(pic);
5515 	return r;
5516 }
5517 
5518 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5519 {
5520 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5521 
5522 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5523 
5524 	mutex_lock(&kps->lock);
5525 	memcpy(ps, &kps->channels, sizeof(*ps));
5526 	mutex_unlock(&kps->lock);
5527 	return 0;
5528 }
5529 
5530 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5531 {
5532 	int i;
5533 	struct kvm_pit *pit = kvm->arch.vpit;
5534 
5535 	mutex_lock(&pit->pit_state.lock);
5536 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5537 	for (i = 0; i < 3; i++)
5538 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5539 	mutex_unlock(&pit->pit_state.lock);
5540 	return 0;
5541 }
5542 
5543 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5544 {
5545 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5546 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5547 		sizeof(ps->channels));
5548 	ps->flags = kvm->arch.vpit->pit_state.flags;
5549 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5550 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5551 	return 0;
5552 }
5553 
5554 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5555 {
5556 	int start = 0;
5557 	int i;
5558 	u32 prev_legacy, cur_legacy;
5559 	struct kvm_pit *pit = kvm->arch.vpit;
5560 
5561 	mutex_lock(&pit->pit_state.lock);
5562 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5563 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5564 	if (!prev_legacy && cur_legacy)
5565 		start = 1;
5566 	memcpy(&pit->pit_state.channels, &ps->channels,
5567 	       sizeof(pit->pit_state.channels));
5568 	pit->pit_state.flags = ps->flags;
5569 	for (i = 0; i < 3; i++)
5570 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5571 				   start && i == 0);
5572 	mutex_unlock(&pit->pit_state.lock);
5573 	return 0;
5574 }
5575 
5576 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5577 				 struct kvm_reinject_control *control)
5578 {
5579 	struct kvm_pit *pit = kvm->arch.vpit;
5580 
5581 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5582 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5583 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5584 	 */
5585 	mutex_lock(&pit->pit_state.lock);
5586 	kvm_pit_set_reinject(pit, control->pit_reinject);
5587 	mutex_unlock(&pit->pit_state.lock);
5588 
5589 	return 0;
5590 }
5591 
5592 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5593 {
5594 
5595 	/*
5596 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5597 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5598 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5599 	 * VM-Exit.
5600 	 */
5601 	struct kvm_vcpu *vcpu;
5602 	int i;
5603 
5604 	kvm_for_each_vcpu(i, vcpu, kvm)
5605 		kvm_vcpu_kick(vcpu);
5606 }
5607 
5608 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5609 			bool line_status)
5610 {
5611 	if (!irqchip_in_kernel(kvm))
5612 		return -ENXIO;
5613 
5614 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5615 					irq_event->irq, irq_event->level,
5616 					line_status);
5617 	return 0;
5618 }
5619 
5620 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5621 			    struct kvm_enable_cap *cap)
5622 {
5623 	int r;
5624 
5625 	if (cap->flags)
5626 		return -EINVAL;
5627 
5628 	switch (cap->cap) {
5629 	case KVM_CAP_DISABLE_QUIRKS:
5630 		kvm->arch.disabled_quirks = cap->args[0];
5631 		r = 0;
5632 		break;
5633 	case KVM_CAP_SPLIT_IRQCHIP: {
5634 		mutex_lock(&kvm->lock);
5635 		r = -EINVAL;
5636 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5637 			goto split_irqchip_unlock;
5638 		r = -EEXIST;
5639 		if (irqchip_in_kernel(kvm))
5640 			goto split_irqchip_unlock;
5641 		if (kvm->created_vcpus)
5642 			goto split_irqchip_unlock;
5643 		r = kvm_setup_empty_irq_routing(kvm);
5644 		if (r)
5645 			goto split_irqchip_unlock;
5646 		/* Pairs with irqchip_in_kernel. */
5647 		smp_wmb();
5648 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5649 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5650 		r = 0;
5651 split_irqchip_unlock:
5652 		mutex_unlock(&kvm->lock);
5653 		break;
5654 	}
5655 	case KVM_CAP_X2APIC_API:
5656 		r = -EINVAL;
5657 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5658 			break;
5659 
5660 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5661 			kvm->arch.x2apic_format = true;
5662 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5663 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5664 
5665 		r = 0;
5666 		break;
5667 	case KVM_CAP_X86_DISABLE_EXITS:
5668 		r = -EINVAL;
5669 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5670 			break;
5671 
5672 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5673 			kvm_can_mwait_in_guest())
5674 			kvm->arch.mwait_in_guest = true;
5675 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5676 			kvm->arch.hlt_in_guest = true;
5677 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5678 			kvm->arch.pause_in_guest = true;
5679 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5680 			kvm->arch.cstate_in_guest = true;
5681 		r = 0;
5682 		break;
5683 	case KVM_CAP_MSR_PLATFORM_INFO:
5684 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5685 		r = 0;
5686 		break;
5687 	case KVM_CAP_EXCEPTION_PAYLOAD:
5688 		kvm->arch.exception_payload_enabled = cap->args[0];
5689 		r = 0;
5690 		break;
5691 	case KVM_CAP_X86_USER_SPACE_MSR:
5692 		kvm->arch.user_space_msr_mask = cap->args[0];
5693 		r = 0;
5694 		break;
5695 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5696 		r = -EINVAL;
5697 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5698 			break;
5699 
5700 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5701 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5702 			break;
5703 
5704 		if (kvm_has_bus_lock_exit &&
5705 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5706 			kvm->arch.bus_lock_detection_enabled = true;
5707 		r = 0;
5708 		break;
5709 #ifdef CONFIG_X86_SGX_KVM
5710 	case KVM_CAP_SGX_ATTRIBUTE: {
5711 		unsigned long allowed_attributes = 0;
5712 
5713 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5714 		if (r)
5715 			break;
5716 
5717 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5718 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5719 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5720 			kvm->arch.sgx_provisioning_allowed = true;
5721 		else
5722 			r = -EINVAL;
5723 		break;
5724 	}
5725 #endif
5726 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5727 		r = -EINVAL;
5728 		if (kvm_x86_ops.vm_copy_enc_context_from)
5729 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5730 		return r;
5731 	case KVM_CAP_EXIT_HYPERCALL:
5732 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5733 			r = -EINVAL;
5734 			break;
5735 		}
5736 		kvm->arch.hypercall_exit_enabled = cap->args[0];
5737 		r = 0;
5738 		break;
5739 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5740 		r = -EINVAL;
5741 		if (cap->args[0] & ~1)
5742 			break;
5743 		kvm->arch.exit_on_emulation_error = cap->args[0];
5744 		r = 0;
5745 		break;
5746 	default:
5747 		r = -EINVAL;
5748 		break;
5749 	}
5750 	return r;
5751 }
5752 
5753 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5754 {
5755 	struct kvm_x86_msr_filter *msr_filter;
5756 
5757 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5758 	if (!msr_filter)
5759 		return NULL;
5760 
5761 	msr_filter->default_allow = default_allow;
5762 	return msr_filter;
5763 }
5764 
5765 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5766 {
5767 	u32 i;
5768 
5769 	if (!msr_filter)
5770 		return;
5771 
5772 	for (i = 0; i < msr_filter->count; i++)
5773 		kfree(msr_filter->ranges[i].bitmap);
5774 
5775 	kfree(msr_filter);
5776 }
5777 
5778 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5779 			      struct kvm_msr_filter_range *user_range)
5780 {
5781 	unsigned long *bitmap = NULL;
5782 	size_t bitmap_size;
5783 
5784 	if (!user_range->nmsrs)
5785 		return 0;
5786 
5787 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5788 		return -EINVAL;
5789 
5790 	if (!user_range->flags)
5791 		return -EINVAL;
5792 
5793 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5794 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5795 		return -EINVAL;
5796 
5797 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5798 	if (IS_ERR(bitmap))
5799 		return PTR_ERR(bitmap);
5800 
5801 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5802 		.flags = user_range->flags,
5803 		.base = user_range->base,
5804 		.nmsrs = user_range->nmsrs,
5805 		.bitmap = bitmap,
5806 	};
5807 
5808 	msr_filter->count++;
5809 	return 0;
5810 }
5811 
5812 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5813 {
5814 	struct kvm_msr_filter __user *user_msr_filter = argp;
5815 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5816 	struct kvm_msr_filter filter;
5817 	bool default_allow;
5818 	bool empty = true;
5819 	int r = 0;
5820 	u32 i;
5821 
5822 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5823 		return -EFAULT;
5824 
5825 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5826 		empty &= !filter.ranges[i].nmsrs;
5827 
5828 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5829 	if (empty && !default_allow)
5830 		return -EINVAL;
5831 
5832 	new_filter = kvm_alloc_msr_filter(default_allow);
5833 	if (!new_filter)
5834 		return -ENOMEM;
5835 
5836 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5837 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5838 		if (r) {
5839 			kvm_free_msr_filter(new_filter);
5840 			return r;
5841 		}
5842 	}
5843 
5844 	mutex_lock(&kvm->lock);
5845 
5846 	/* The per-VM filter is protected by kvm->lock... */
5847 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5848 
5849 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5850 	synchronize_srcu(&kvm->srcu);
5851 
5852 	kvm_free_msr_filter(old_filter);
5853 
5854 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5855 	mutex_unlock(&kvm->lock);
5856 
5857 	return 0;
5858 }
5859 
5860 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5861 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5862 {
5863 	struct kvm_vcpu *vcpu;
5864 	int i, ret = 0;
5865 
5866 	mutex_lock(&kvm->lock);
5867 	kvm_for_each_vcpu(i, vcpu, kvm) {
5868 		if (!vcpu->arch.pv_time_enabled)
5869 			continue;
5870 
5871 		ret = kvm_set_guest_paused(vcpu);
5872 		if (ret) {
5873 			kvm_err("Failed to pause guest VCPU%d: %d\n",
5874 				vcpu->vcpu_id, ret);
5875 			break;
5876 		}
5877 	}
5878 	mutex_unlock(&kvm->lock);
5879 
5880 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
5881 }
5882 
5883 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5884 {
5885 	switch (state) {
5886 	case PM_HIBERNATION_PREPARE:
5887 	case PM_SUSPEND_PREPARE:
5888 		return kvm_arch_suspend_notifier(kvm);
5889 	}
5890 
5891 	return NOTIFY_DONE;
5892 }
5893 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5894 
5895 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
5896 {
5897 	struct kvm_clock_data data = { 0 };
5898 
5899 	get_kvmclock(kvm, &data);
5900 	if (copy_to_user(argp, &data, sizeof(data)))
5901 		return -EFAULT;
5902 
5903 	return 0;
5904 }
5905 
5906 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
5907 {
5908 	struct kvm_arch *ka = &kvm->arch;
5909 	struct kvm_clock_data data;
5910 	u64 now_raw_ns;
5911 
5912 	if (copy_from_user(&data, argp, sizeof(data)))
5913 		return -EFAULT;
5914 
5915 	/*
5916 	 * Only KVM_CLOCK_REALTIME is used, but allow passing the
5917 	 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
5918 	 */
5919 	if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
5920 		return -EINVAL;
5921 
5922 	kvm_hv_invalidate_tsc_page(kvm);
5923 	kvm_start_pvclock_update(kvm);
5924 	pvclock_update_vm_gtod_copy(kvm);
5925 
5926 	/*
5927 	 * This pairs with kvm_guest_time_update(): when masterclock is
5928 	 * in use, we use master_kernel_ns + kvmclock_offset to set
5929 	 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5930 	 * is slightly ahead) here we risk going negative on unsigned
5931 	 * 'system_time' when 'data.clock' is very small.
5932 	 */
5933 	if (data.flags & KVM_CLOCK_REALTIME) {
5934 		u64 now_real_ns = ktime_get_real_ns();
5935 
5936 		/*
5937 		 * Avoid stepping the kvmclock backwards.
5938 		 */
5939 		if (now_real_ns > data.realtime)
5940 			data.clock += now_real_ns - data.realtime;
5941 	}
5942 
5943 	if (ka->use_master_clock)
5944 		now_raw_ns = ka->master_kernel_ns;
5945 	else
5946 		now_raw_ns = get_kvmclock_base_ns();
5947 	ka->kvmclock_offset = data.clock - now_raw_ns;
5948 	kvm_end_pvclock_update(kvm);
5949 	return 0;
5950 }
5951 
5952 long kvm_arch_vm_ioctl(struct file *filp,
5953 		       unsigned int ioctl, unsigned long arg)
5954 {
5955 	struct kvm *kvm = filp->private_data;
5956 	void __user *argp = (void __user *)arg;
5957 	int r = -ENOTTY;
5958 	/*
5959 	 * This union makes it completely explicit to gcc-3.x
5960 	 * that these two variables' stack usage should be
5961 	 * combined, not added together.
5962 	 */
5963 	union {
5964 		struct kvm_pit_state ps;
5965 		struct kvm_pit_state2 ps2;
5966 		struct kvm_pit_config pit_config;
5967 	} u;
5968 
5969 	switch (ioctl) {
5970 	case KVM_SET_TSS_ADDR:
5971 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5972 		break;
5973 	case KVM_SET_IDENTITY_MAP_ADDR: {
5974 		u64 ident_addr;
5975 
5976 		mutex_lock(&kvm->lock);
5977 		r = -EINVAL;
5978 		if (kvm->created_vcpus)
5979 			goto set_identity_unlock;
5980 		r = -EFAULT;
5981 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5982 			goto set_identity_unlock;
5983 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5984 set_identity_unlock:
5985 		mutex_unlock(&kvm->lock);
5986 		break;
5987 	}
5988 	case KVM_SET_NR_MMU_PAGES:
5989 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5990 		break;
5991 	case KVM_GET_NR_MMU_PAGES:
5992 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5993 		break;
5994 	case KVM_CREATE_IRQCHIP: {
5995 		mutex_lock(&kvm->lock);
5996 
5997 		r = -EEXIST;
5998 		if (irqchip_in_kernel(kvm))
5999 			goto create_irqchip_unlock;
6000 
6001 		r = -EINVAL;
6002 		if (kvm->created_vcpus)
6003 			goto create_irqchip_unlock;
6004 
6005 		r = kvm_pic_init(kvm);
6006 		if (r)
6007 			goto create_irqchip_unlock;
6008 
6009 		r = kvm_ioapic_init(kvm);
6010 		if (r) {
6011 			kvm_pic_destroy(kvm);
6012 			goto create_irqchip_unlock;
6013 		}
6014 
6015 		r = kvm_setup_default_irq_routing(kvm);
6016 		if (r) {
6017 			kvm_ioapic_destroy(kvm);
6018 			kvm_pic_destroy(kvm);
6019 			goto create_irqchip_unlock;
6020 		}
6021 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6022 		smp_wmb();
6023 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6024 	create_irqchip_unlock:
6025 		mutex_unlock(&kvm->lock);
6026 		break;
6027 	}
6028 	case KVM_CREATE_PIT:
6029 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6030 		goto create_pit;
6031 	case KVM_CREATE_PIT2:
6032 		r = -EFAULT;
6033 		if (copy_from_user(&u.pit_config, argp,
6034 				   sizeof(struct kvm_pit_config)))
6035 			goto out;
6036 	create_pit:
6037 		mutex_lock(&kvm->lock);
6038 		r = -EEXIST;
6039 		if (kvm->arch.vpit)
6040 			goto create_pit_unlock;
6041 		r = -ENOMEM;
6042 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6043 		if (kvm->arch.vpit)
6044 			r = 0;
6045 	create_pit_unlock:
6046 		mutex_unlock(&kvm->lock);
6047 		break;
6048 	case KVM_GET_IRQCHIP: {
6049 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6050 		struct kvm_irqchip *chip;
6051 
6052 		chip = memdup_user(argp, sizeof(*chip));
6053 		if (IS_ERR(chip)) {
6054 			r = PTR_ERR(chip);
6055 			goto out;
6056 		}
6057 
6058 		r = -ENXIO;
6059 		if (!irqchip_kernel(kvm))
6060 			goto get_irqchip_out;
6061 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6062 		if (r)
6063 			goto get_irqchip_out;
6064 		r = -EFAULT;
6065 		if (copy_to_user(argp, chip, sizeof(*chip)))
6066 			goto get_irqchip_out;
6067 		r = 0;
6068 	get_irqchip_out:
6069 		kfree(chip);
6070 		break;
6071 	}
6072 	case KVM_SET_IRQCHIP: {
6073 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6074 		struct kvm_irqchip *chip;
6075 
6076 		chip = memdup_user(argp, sizeof(*chip));
6077 		if (IS_ERR(chip)) {
6078 			r = PTR_ERR(chip);
6079 			goto out;
6080 		}
6081 
6082 		r = -ENXIO;
6083 		if (!irqchip_kernel(kvm))
6084 			goto set_irqchip_out;
6085 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6086 	set_irqchip_out:
6087 		kfree(chip);
6088 		break;
6089 	}
6090 	case KVM_GET_PIT: {
6091 		r = -EFAULT;
6092 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6093 			goto out;
6094 		r = -ENXIO;
6095 		if (!kvm->arch.vpit)
6096 			goto out;
6097 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6098 		if (r)
6099 			goto out;
6100 		r = -EFAULT;
6101 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6102 			goto out;
6103 		r = 0;
6104 		break;
6105 	}
6106 	case KVM_SET_PIT: {
6107 		r = -EFAULT;
6108 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6109 			goto out;
6110 		mutex_lock(&kvm->lock);
6111 		r = -ENXIO;
6112 		if (!kvm->arch.vpit)
6113 			goto set_pit_out;
6114 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6115 set_pit_out:
6116 		mutex_unlock(&kvm->lock);
6117 		break;
6118 	}
6119 	case KVM_GET_PIT2: {
6120 		r = -ENXIO;
6121 		if (!kvm->arch.vpit)
6122 			goto out;
6123 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6124 		if (r)
6125 			goto out;
6126 		r = -EFAULT;
6127 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6128 			goto out;
6129 		r = 0;
6130 		break;
6131 	}
6132 	case KVM_SET_PIT2: {
6133 		r = -EFAULT;
6134 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6135 			goto out;
6136 		mutex_lock(&kvm->lock);
6137 		r = -ENXIO;
6138 		if (!kvm->arch.vpit)
6139 			goto set_pit2_out;
6140 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6141 set_pit2_out:
6142 		mutex_unlock(&kvm->lock);
6143 		break;
6144 	}
6145 	case KVM_REINJECT_CONTROL: {
6146 		struct kvm_reinject_control control;
6147 		r =  -EFAULT;
6148 		if (copy_from_user(&control, argp, sizeof(control)))
6149 			goto out;
6150 		r = -ENXIO;
6151 		if (!kvm->arch.vpit)
6152 			goto out;
6153 		r = kvm_vm_ioctl_reinject(kvm, &control);
6154 		break;
6155 	}
6156 	case KVM_SET_BOOT_CPU_ID:
6157 		r = 0;
6158 		mutex_lock(&kvm->lock);
6159 		if (kvm->created_vcpus)
6160 			r = -EBUSY;
6161 		else
6162 			kvm->arch.bsp_vcpu_id = arg;
6163 		mutex_unlock(&kvm->lock);
6164 		break;
6165 #ifdef CONFIG_KVM_XEN
6166 	case KVM_XEN_HVM_CONFIG: {
6167 		struct kvm_xen_hvm_config xhc;
6168 		r = -EFAULT;
6169 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6170 			goto out;
6171 		r = kvm_xen_hvm_config(kvm, &xhc);
6172 		break;
6173 	}
6174 	case KVM_XEN_HVM_GET_ATTR: {
6175 		struct kvm_xen_hvm_attr xha;
6176 
6177 		r = -EFAULT;
6178 		if (copy_from_user(&xha, argp, sizeof(xha)))
6179 			goto out;
6180 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6181 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6182 			r = -EFAULT;
6183 		break;
6184 	}
6185 	case KVM_XEN_HVM_SET_ATTR: {
6186 		struct kvm_xen_hvm_attr xha;
6187 
6188 		r = -EFAULT;
6189 		if (copy_from_user(&xha, argp, sizeof(xha)))
6190 			goto out;
6191 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6192 		break;
6193 	}
6194 #endif
6195 	case KVM_SET_CLOCK:
6196 		r = kvm_vm_ioctl_set_clock(kvm, argp);
6197 		break;
6198 	case KVM_GET_CLOCK:
6199 		r = kvm_vm_ioctl_get_clock(kvm, argp);
6200 		break;
6201 	case KVM_MEMORY_ENCRYPT_OP: {
6202 		r = -ENOTTY;
6203 		if (kvm_x86_ops.mem_enc_op)
6204 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6205 		break;
6206 	}
6207 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6208 		struct kvm_enc_region region;
6209 
6210 		r = -EFAULT;
6211 		if (copy_from_user(&region, argp, sizeof(region)))
6212 			goto out;
6213 
6214 		r = -ENOTTY;
6215 		if (kvm_x86_ops.mem_enc_reg_region)
6216 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
6217 		break;
6218 	}
6219 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6220 		struct kvm_enc_region region;
6221 
6222 		r = -EFAULT;
6223 		if (copy_from_user(&region, argp, sizeof(region)))
6224 			goto out;
6225 
6226 		r = -ENOTTY;
6227 		if (kvm_x86_ops.mem_enc_unreg_region)
6228 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
6229 		break;
6230 	}
6231 	case KVM_HYPERV_EVENTFD: {
6232 		struct kvm_hyperv_eventfd hvevfd;
6233 
6234 		r = -EFAULT;
6235 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6236 			goto out;
6237 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6238 		break;
6239 	}
6240 	case KVM_SET_PMU_EVENT_FILTER:
6241 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6242 		break;
6243 	case KVM_X86_SET_MSR_FILTER:
6244 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6245 		break;
6246 	default:
6247 		r = -ENOTTY;
6248 	}
6249 out:
6250 	return r;
6251 }
6252 
6253 static void kvm_init_msr_list(void)
6254 {
6255 	struct x86_pmu_capability x86_pmu;
6256 	u32 dummy[2];
6257 	unsigned i;
6258 
6259 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6260 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6261 
6262 	perf_get_x86_pmu_capability(&x86_pmu);
6263 
6264 	num_msrs_to_save = 0;
6265 	num_emulated_msrs = 0;
6266 	num_msr_based_features = 0;
6267 
6268 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6269 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6270 			continue;
6271 
6272 		/*
6273 		 * Even MSRs that are valid in the host may not be exposed
6274 		 * to the guests in some cases.
6275 		 */
6276 		switch (msrs_to_save_all[i]) {
6277 		case MSR_IA32_BNDCFGS:
6278 			if (!kvm_mpx_supported())
6279 				continue;
6280 			break;
6281 		case MSR_TSC_AUX:
6282 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6283 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6284 				continue;
6285 			break;
6286 		case MSR_IA32_UMWAIT_CONTROL:
6287 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6288 				continue;
6289 			break;
6290 		case MSR_IA32_RTIT_CTL:
6291 		case MSR_IA32_RTIT_STATUS:
6292 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6293 				continue;
6294 			break;
6295 		case MSR_IA32_RTIT_CR3_MATCH:
6296 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6297 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6298 				continue;
6299 			break;
6300 		case MSR_IA32_RTIT_OUTPUT_BASE:
6301 		case MSR_IA32_RTIT_OUTPUT_MASK:
6302 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6303 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6304 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6305 				continue;
6306 			break;
6307 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6308 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6309 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6310 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6311 				continue;
6312 			break;
6313 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6314 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6315 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6316 				continue;
6317 			break;
6318 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6319 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6320 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6321 				continue;
6322 			break;
6323 		default:
6324 			break;
6325 		}
6326 
6327 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6328 	}
6329 
6330 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6331 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6332 			continue;
6333 
6334 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6335 	}
6336 
6337 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6338 		struct kvm_msr_entry msr;
6339 
6340 		msr.index = msr_based_features_all[i];
6341 		if (kvm_get_msr_feature(&msr))
6342 			continue;
6343 
6344 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6345 	}
6346 }
6347 
6348 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6349 			   const void *v)
6350 {
6351 	int handled = 0;
6352 	int n;
6353 
6354 	do {
6355 		n = min(len, 8);
6356 		if (!(lapic_in_kernel(vcpu) &&
6357 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6358 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6359 			break;
6360 		handled += n;
6361 		addr += n;
6362 		len -= n;
6363 		v += n;
6364 	} while (len);
6365 
6366 	return handled;
6367 }
6368 
6369 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6370 {
6371 	int handled = 0;
6372 	int n;
6373 
6374 	do {
6375 		n = min(len, 8);
6376 		if (!(lapic_in_kernel(vcpu) &&
6377 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6378 					 addr, n, v))
6379 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6380 			break;
6381 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6382 		handled += n;
6383 		addr += n;
6384 		len -= n;
6385 		v += n;
6386 	} while (len);
6387 
6388 	return handled;
6389 }
6390 
6391 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6392 			struct kvm_segment *var, int seg)
6393 {
6394 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6395 }
6396 
6397 void kvm_get_segment(struct kvm_vcpu *vcpu,
6398 		     struct kvm_segment *var, int seg)
6399 {
6400 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6401 }
6402 
6403 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6404 			   struct x86_exception *exception)
6405 {
6406 	gpa_t t_gpa;
6407 
6408 	BUG_ON(!mmu_is_nested(vcpu));
6409 
6410 	/* NPT walks are always user-walks */
6411 	access |= PFERR_USER_MASK;
6412 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6413 
6414 	return t_gpa;
6415 }
6416 
6417 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6418 			      struct x86_exception *exception)
6419 {
6420 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6421 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6422 }
6423 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6424 
6425  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6426 				struct x86_exception *exception)
6427 {
6428 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6429 	access |= PFERR_FETCH_MASK;
6430 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6431 }
6432 
6433 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6434 			       struct x86_exception *exception)
6435 {
6436 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6437 	access |= PFERR_WRITE_MASK;
6438 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6439 }
6440 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6441 
6442 /* uses this to access any guest's mapped memory without checking CPL */
6443 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6444 				struct x86_exception *exception)
6445 {
6446 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6447 }
6448 
6449 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6450 				      struct kvm_vcpu *vcpu, u32 access,
6451 				      struct x86_exception *exception)
6452 {
6453 	void *data = val;
6454 	int r = X86EMUL_CONTINUE;
6455 
6456 	while (bytes) {
6457 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6458 							    exception);
6459 		unsigned offset = addr & (PAGE_SIZE-1);
6460 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6461 		int ret;
6462 
6463 		if (gpa == UNMAPPED_GVA)
6464 			return X86EMUL_PROPAGATE_FAULT;
6465 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6466 					       offset, toread);
6467 		if (ret < 0) {
6468 			r = X86EMUL_IO_NEEDED;
6469 			goto out;
6470 		}
6471 
6472 		bytes -= toread;
6473 		data += toread;
6474 		addr += toread;
6475 	}
6476 out:
6477 	return r;
6478 }
6479 
6480 /* used for instruction fetching */
6481 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6482 				gva_t addr, void *val, unsigned int bytes,
6483 				struct x86_exception *exception)
6484 {
6485 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6486 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6487 	unsigned offset;
6488 	int ret;
6489 
6490 	/* Inline kvm_read_guest_virt_helper for speed.  */
6491 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6492 						    exception);
6493 	if (unlikely(gpa == UNMAPPED_GVA))
6494 		return X86EMUL_PROPAGATE_FAULT;
6495 
6496 	offset = addr & (PAGE_SIZE-1);
6497 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6498 		bytes = (unsigned)PAGE_SIZE - offset;
6499 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6500 				       offset, bytes);
6501 	if (unlikely(ret < 0))
6502 		return X86EMUL_IO_NEEDED;
6503 
6504 	return X86EMUL_CONTINUE;
6505 }
6506 
6507 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6508 			       gva_t addr, void *val, unsigned int bytes,
6509 			       struct x86_exception *exception)
6510 {
6511 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6512 
6513 	/*
6514 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6515 	 * is returned, but our callers are not ready for that and they blindly
6516 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6517 	 * uninitialized kernel stack memory into cr2 and error code.
6518 	 */
6519 	memset(exception, 0, sizeof(*exception));
6520 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6521 					  exception);
6522 }
6523 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6524 
6525 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6526 			     gva_t addr, void *val, unsigned int bytes,
6527 			     struct x86_exception *exception, bool system)
6528 {
6529 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6530 	u32 access = 0;
6531 
6532 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6533 		access |= PFERR_USER_MASK;
6534 
6535 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6536 }
6537 
6538 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6539 		unsigned long addr, void *val, unsigned int bytes)
6540 {
6541 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6542 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6543 
6544 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6545 }
6546 
6547 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6548 				      struct kvm_vcpu *vcpu, u32 access,
6549 				      struct x86_exception *exception)
6550 {
6551 	void *data = val;
6552 	int r = X86EMUL_CONTINUE;
6553 
6554 	while (bytes) {
6555 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6556 							     access,
6557 							     exception);
6558 		unsigned offset = addr & (PAGE_SIZE-1);
6559 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6560 		int ret;
6561 
6562 		if (gpa == UNMAPPED_GVA)
6563 			return X86EMUL_PROPAGATE_FAULT;
6564 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6565 		if (ret < 0) {
6566 			r = X86EMUL_IO_NEEDED;
6567 			goto out;
6568 		}
6569 
6570 		bytes -= towrite;
6571 		data += towrite;
6572 		addr += towrite;
6573 	}
6574 out:
6575 	return r;
6576 }
6577 
6578 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6579 			      unsigned int bytes, struct x86_exception *exception,
6580 			      bool system)
6581 {
6582 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6583 	u32 access = PFERR_WRITE_MASK;
6584 
6585 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6586 		access |= PFERR_USER_MASK;
6587 
6588 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6589 					   access, exception);
6590 }
6591 
6592 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6593 				unsigned int bytes, struct x86_exception *exception)
6594 {
6595 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6596 	vcpu->arch.l1tf_flush_l1d = true;
6597 
6598 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6599 					   PFERR_WRITE_MASK, exception);
6600 }
6601 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6602 
6603 int handle_ud(struct kvm_vcpu *vcpu)
6604 {
6605 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6606 	int emul_type = EMULTYPE_TRAP_UD;
6607 	char sig[5]; /* ud2; .ascii "kvm" */
6608 	struct x86_exception e;
6609 
6610 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6611 		return 1;
6612 
6613 	if (force_emulation_prefix &&
6614 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6615 				sig, sizeof(sig), &e) == 0 &&
6616 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6617 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6618 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6619 	}
6620 
6621 	return kvm_emulate_instruction(vcpu, emul_type);
6622 }
6623 EXPORT_SYMBOL_GPL(handle_ud);
6624 
6625 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6626 			    gpa_t gpa, bool write)
6627 {
6628 	/* For APIC access vmexit */
6629 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6630 		return 1;
6631 
6632 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6633 		trace_vcpu_match_mmio(gva, gpa, write, true);
6634 		return 1;
6635 	}
6636 
6637 	return 0;
6638 }
6639 
6640 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6641 				gpa_t *gpa, struct x86_exception *exception,
6642 				bool write)
6643 {
6644 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6645 		| (write ? PFERR_WRITE_MASK : 0);
6646 
6647 	/*
6648 	 * currently PKRU is only applied to ept enabled guest so
6649 	 * there is no pkey in EPT page table for L1 guest or EPT
6650 	 * shadow page table for L2 guest.
6651 	 */
6652 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6653 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
6654 			      vcpu->arch.mmio_access, 0, access))) {
6655 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6656 					(gva & (PAGE_SIZE - 1));
6657 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6658 		return 1;
6659 	}
6660 
6661 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6662 
6663 	if (*gpa == UNMAPPED_GVA)
6664 		return -1;
6665 
6666 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6667 }
6668 
6669 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6670 			const void *val, int bytes)
6671 {
6672 	int ret;
6673 
6674 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6675 	if (ret < 0)
6676 		return 0;
6677 	kvm_page_track_write(vcpu, gpa, val, bytes);
6678 	return 1;
6679 }
6680 
6681 struct read_write_emulator_ops {
6682 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6683 				  int bytes);
6684 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6685 				  void *val, int bytes);
6686 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6687 			       int bytes, void *val);
6688 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6689 				    void *val, int bytes);
6690 	bool write;
6691 };
6692 
6693 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6694 {
6695 	if (vcpu->mmio_read_completed) {
6696 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6697 			       vcpu->mmio_fragments[0].gpa, val);
6698 		vcpu->mmio_read_completed = 0;
6699 		return 1;
6700 	}
6701 
6702 	return 0;
6703 }
6704 
6705 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6706 			void *val, int bytes)
6707 {
6708 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6709 }
6710 
6711 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6712 			 void *val, int bytes)
6713 {
6714 	return emulator_write_phys(vcpu, gpa, val, bytes);
6715 }
6716 
6717 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6718 {
6719 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6720 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6721 }
6722 
6723 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6724 			  void *val, int bytes)
6725 {
6726 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6727 	return X86EMUL_IO_NEEDED;
6728 }
6729 
6730 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6731 			   void *val, int bytes)
6732 {
6733 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6734 
6735 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6736 	return X86EMUL_CONTINUE;
6737 }
6738 
6739 static const struct read_write_emulator_ops read_emultor = {
6740 	.read_write_prepare = read_prepare,
6741 	.read_write_emulate = read_emulate,
6742 	.read_write_mmio = vcpu_mmio_read,
6743 	.read_write_exit_mmio = read_exit_mmio,
6744 };
6745 
6746 static const struct read_write_emulator_ops write_emultor = {
6747 	.read_write_emulate = write_emulate,
6748 	.read_write_mmio = write_mmio,
6749 	.read_write_exit_mmio = write_exit_mmio,
6750 	.write = true,
6751 };
6752 
6753 static int emulator_read_write_onepage(unsigned long addr, void *val,
6754 				       unsigned int bytes,
6755 				       struct x86_exception *exception,
6756 				       struct kvm_vcpu *vcpu,
6757 				       const struct read_write_emulator_ops *ops)
6758 {
6759 	gpa_t gpa;
6760 	int handled, ret;
6761 	bool write = ops->write;
6762 	struct kvm_mmio_fragment *frag;
6763 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6764 
6765 	/*
6766 	 * If the exit was due to a NPF we may already have a GPA.
6767 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6768 	 * Note, this cannot be used on string operations since string
6769 	 * operation using rep will only have the initial GPA from the NPF
6770 	 * occurred.
6771 	 */
6772 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6773 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6774 		gpa = ctxt->gpa_val;
6775 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6776 	} else {
6777 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6778 		if (ret < 0)
6779 			return X86EMUL_PROPAGATE_FAULT;
6780 	}
6781 
6782 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6783 		return X86EMUL_CONTINUE;
6784 
6785 	/*
6786 	 * Is this MMIO handled locally?
6787 	 */
6788 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6789 	if (handled == bytes)
6790 		return X86EMUL_CONTINUE;
6791 
6792 	gpa += handled;
6793 	bytes -= handled;
6794 	val += handled;
6795 
6796 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6797 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6798 	frag->gpa = gpa;
6799 	frag->data = val;
6800 	frag->len = bytes;
6801 	return X86EMUL_CONTINUE;
6802 }
6803 
6804 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6805 			unsigned long addr,
6806 			void *val, unsigned int bytes,
6807 			struct x86_exception *exception,
6808 			const struct read_write_emulator_ops *ops)
6809 {
6810 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6811 	gpa_t gpa;
6812 	int rc;
6813 
6814 	if (ops->read_write_prepare &&
6815 		  ops->read_write_prepare(vcpu, val, bytes))
6816 		return X86EMUL_CONTINUE;
6817 
6818 	vcpu->mmio_nr_fragments = 0;
6819 
6820 	/* Crossing a page boundary? */
6821 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6822 		int now;
6823 
6824 		now = -addr & ~PAGE_MASK;
6825 		rc = emulator_read_write_onepage(addr, val, now, exception,
6826 						 vcpu, ops);
6827 
6828 		if (rc != X86EMUL_CONTINUE)
6829 			return rc;
6830 		addr += now;
6831 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6832 			addr = (u32)addr;
6833 		val += now;
6834 		bytes -= now;
6835 	}
6836 
6837 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6838 					 vcpu, ops);
6839 	if (rc != X86EMUL_CONTINUE)
6840 		return rc;
6841 
6842 	if (!vcpu->mmio_nr_fragments)
6843 		return rc;
6844 
6845 	gpa = vcpu->mmio_fragments[0].gpa;
6846 
6847 	vcpu->mmio_needed = 1;
6848 	vcpu->mmio_cur_fragment = 0;
6849 
6850 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6851 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6852 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6853 	vcpu->run->mmio.phys_addr = gpa;
6854 
6855 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6856 }
6857 
6858 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6859 				  unsigned long addr,
6860 				  void *val,
6861 				  unsigned int bytes,
6862 				  struct x86_exception *exception)
6863 {
6864 	return emulator_read_write(ctxt, addr, val, bytes,
6865 				   exception, &read_emultor);
6866 }
6867 
6868 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6869 			    unsigned long addr,
6870 			    const void *val,
6871 			    unsigned int bytes,
6872 			    struct x86_exception *exception)
6873 {
6874 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6875 				   exception, &write_emultor);
6876 }
6877 
6878 #define CMPXCHG_TYPE(t, ptr, old, new) \
6879 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6880 
6881 #ifdef CONFIG_X86_64
6882 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6883 #else
6884 #  define CMPXCHG64(ptr, old, new) \
6885 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6886 #endif
6887 
6888 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6889 				     unsigned long addr,
6890 				     const void *old,
6891 				     const void *new,
6892 				     unsigned int bytes,
6893 				     struct x86_exception *exception)
6894 {
6895 	struct kvm_host_map map;
6896 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6897 	u64 page_line_mask;
6898 	gpa_t gpa;
6899 	char *kaddr;
6900 	bool exchanged;
6901 
6902 	/* guests cmpxchg8b have to be emulated atomically */
6903 	if (bytes > 8 || (bytes & (bytes - 1)))
6904 		goto emul_write;
6905 
6906 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6907 
6908 	if (gpa == UNMAPPED_GVA ||
6909 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6910 		goto emul_write;
6911 
6912 	/*
6913 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6914 	 * enabled in the host and the access splits a cache line.
6915 	 */
6916 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6917 		page_line_mask = ~(cache_line_size() - 1);
6918 	else
6919 		page_line_mask = PAGE_MASK;
6920 
6921 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6922 		goto emul_write;
6923 
6924 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6925 		goto emul_write;
6926 
6927 	kaddr = map.hva + offset_in_page(gpa);
6928 
6929 	switch (bytes) {
6930 	case 1:
6931 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6932 		break;
6933 	case 2:
6934 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6935 		break;
6936 	case 4:
6937 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6938 		break;
6939 	case 8:
6940 		exchanged = CMPXCHG64(kaddr, old, new);
6941 		break;
6942 	default:
6943 		BUG();
6944 	}
6945 
6946 	kvm_vcpu_unmap(vcpu, &map, true);
6947 
6948 	if (!exchanged)
6949 		return X86EMUL_CMPXCHG_FAILED;
6950 
6951 	kvm_page_track_write(vcpu, gpa, new, bytes);
6952 
6953 	return X86EMUL_CONTINUE;
6954 
6955 emul_write:
6956 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6957 
6958 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6959 }
6960 
6961 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6962 {
6963 	int r = 0, i;
6964 
6965 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6966 		if (vcpu->arch.pio.in)
6967 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6968 					    vcpu->arch.pio.size, pd);
6969 		else
6970 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6971 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6972 					     pd);
6973 		if (r)
6974 			break;
6975 		pd += vcpu->arch.pio.size;
6976 	}
6977 	return r;
6978 }
6979 
6980 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6981 			       unsigned short port,
6982 			       unsigned int count, bool in)
6983 {
6984 	vcpu->arch.pio.port = port;
6985 	vcpu->arch.pio.in = in;
6986 	vcpu->arch.pio.count  = count;
6987 	vcpu->arch.pio.size = size;
6988 
6989 	if (!kernel_pio(vcpu, vcpu->arch.pio_data))
6990 		return 1;
6991 
6992 	vcpu->run->exit_reason = KVM_EXIT_IO;
6993 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6994 	vcpu->run->io.size = size;
6995 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6996 	vcpu->run->io.count = count;
6997 	vcpu->run->io.port = port;
6998 
6999 	return 0;
7000 }
7001 
7002 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7003 			     unsigned short port, unsigned int count)
7004 {
7005 	WARN_ON(vcpu->arch.pio.count);
7006 	memset(vcpu->arch.pio_data, 0, size * count);
7007 	return emulator_pio_in_out(vcpu, size, port, count, true);
7008 }
7009 
7010 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7011 {
7012 	int size = vcpu->arch.pio.size;
7013 	unsigned count = vcpu->arch.pio.count;
7014 	memcpy(val, vcpu->arch.pio_data, size * count);
7015 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7016 	vcpu->arch.pio.count = 0;
7017 }
7018 
7019 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7020 			   unsigned short port, void *val, unsigned int count)
7021 {
7022 	if (vcpu->arch.pio.count) {
7023 		/* Complete previous iteration.  */
7024 	} else {
7025 		int r = __emulator_pio_in(vcpu, size, port, count);
7026 		if (!r)
7027 			return r;
7028 
7029 		/* Results already available, fall through.  */
7030 	}
7031 
7032 	WARN_ON(count != vcpu->arch.pio.count);
7033 	complete_emulator_pio_in(vcpu, val);
7034 	return 1;
7035 }
7036 
7037 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7038 				    int size, unsigned short port, void *val,
7039 				    unsigned int count)
7040 {
7041 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7042 
7043 }
7044 
7045 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7046 			    unsigned short port, const void *val,
7047 			    unsigned int count)
7048 {
7049 	int ret;
7050 
7051 	memcpy(vcpu->arch.pio_data, val, size * count);
7052 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7053 	ret = emulator_pio_in_out(vcpu, size, port, count, false);
7054 	if (ret)
7055                 vcpu->arch.pio.count = 0;
7056 
7057         return ret;
7058 }
7059 
7060 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7061 				     int size, unsigned short port,
7062 				     const void *val, unsigned int count)
7063 {
7064 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7065 }
7066 
7067 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7068 {
7069 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7070 }
7071 
7072 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7073 {
7074 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7075 }
7076 
7077 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7078 {
7079 	if (!need_emulate_wbinvd(vcpu))
7080 		return X86EMUL_CONTINUE;
7081 
7082 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
7083 		int cpu = get_cpu();
7084 
7085 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7086 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7087 				wbinvd_ipi, NULL, 1);
7088 		put_cpu();
7089 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7090 	} else
7091 		wbinvd();
7092 	return X86EMUL_CONTINUE;
7093 }
7094 
7095 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7096 {
7097 	kvm_emulate_wbinvd_noskip(vcpu);
7098 	return kvm_skip_emulated_instruction(vcpu);
7099 }
7100 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7101 
7102 
7103 
7104 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7105 {
7106 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7107 }
7108 
7109 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7110 			    unsigned long *dest)
7111 {
7112 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7113 }
7114 
7115 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7116 			   unsigned long value)
7117 {
7118 
7119 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7120 }
7121 
7122 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7123 {
7124 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7125 }
7126 
7127 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7128 {
7129 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7130 	unsigned long value;
7131 
7132 	switch (cr) {
7133 	case 0:
7134 		value = kvm_read_cr0(vcpu);
7135 		break;
7136 	case 2:
7137 		value = vcpu->arch.cr2;
7138 		break;
7139 	case 3:
7140 		value = kvm_read_cr3(vcpu);
7141 		break;
7142 	case 4:
7143 		value = kvm_read_cr4(vcpu);
7144 		break;
7145 	case 8:
7146 		value = kvm_get_cr8(vcpu);
7147 		break;
7148 	default:
7149 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7150 		return 0;
7151 	}
7152 
7153 	return value;
7154 }
7155 
7156 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7157 {
7158 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7159 	int res = 0;
7160 
7161 	switch (cr) {
7162 	case 0:
7163 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7164 		break;
7165 	case 2:
7166 		vcpu->arch.cr2 = val;
7167 		break;
7168 	case 3:
7169 		res = kvm_set_cr3(vcpu, val);
7170 		break;
7171 	case 4:
7172 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7173 		break;
7174 	case 8:
7175 		res = kvm_set_cr8(vcpu, val);
7176 		break;
7177 	default:
7178 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7179 		res = -1;
7180 	}
7181 
7182 	return res;
7183 }
7184 
7185 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7186 {
7187 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7188 }
7189 
7190 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7191 {
7192 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7193 }
7194 
7195 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7196 {
7197 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7198 }
7199 
7200 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7201 {
7202 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7203 }
7204 
7205 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7206 {
7207 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7208 }
7209 
7210 static unsigned long emulator_get_cached_segment_base(
7211 	struct x86_emulate_ctxt *ctxt, int seg)
7212 {
7213 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7214 }
7215 
7216 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7217 				 struct desc_struct *desc, u32 *base3,
7218 				 int seg)
7219 {
7220 	struct kvm_segment var;
7221 
7222 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7223 	*selector = var.selector;
7224 
7225 	if (var.unusable) {
7226 		memset(desc, 0, sizeof(*desc));
7227 		if (base3)
7228 			*base3 = 0;
7229 		return false;
7230 	}
7231 
7232 	if (var.g)
7233 		var.limit >>= 12;
7234 	set_desc_limit(desc, var.limit);
7235 	set_desc_base(desc, (unsigned long)var.base);
7236 #ifdef CONFIG_X86_64
7237 	if (base3)
7238 		*base3 = var.base >> 32;
7239 #endif
7240 	desc->type = var.type;
7241 	desc->s = var.s;
7242 	desc->dpl = var.dpl;
7243 	desc->p = var.present;
7244 	desc->avl = var.avl;
7245 	desc->l = var.l;
7246 	desc->d = var.db;
7247 	desc->g = var.g;
7248 
7249 	return true;
7250 }
7251 
7252 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7253 				 struct desc_struct *desc, u32 base3,
7254 				 int seg)
7255 {
7256 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7257 	struct kvm_segment var;
7258 
7259 	var.selector = selector;
7260 	var.base = get_desc_base(desc);
7261 #ifdef CONFIG_X86_64
7262 	var.base |= ((u64)base3) << 32;
7263 #endif
7264 	var.limit = get_desc_limit(desc);
7265 	if (desc->g)
7266 		var.limit = (var.limit << 12) | 0xfff;
7267 	var.type = desc->type;
7268 	var.dpl = desc->dpl;
7269 	var.db = desc->d;
7270 	var.s = desc->s;
7271 	var.l = desc->l;
7272 	var.g = desc->g;
7273 	var.avl = desc->avl;
7274 	var.present = desc->p;
7275 	var.unusable = !var.present;
7276 	var.padding = 0;
7277 
7278 	kvm_set_segment(vcpu, &var, seg);
7279 	return;
7280 }
7281 
7282 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7283 			    u32 msr_index, u64 *pdata)
7284 {
7285 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7286 	int r;
7287 
7288 	r = kvm_get_msr(vcpu, msr_index, pdata);
7289 
7290 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7291 		/* Bounce to user space */
7292 		return X86EMUL_IO_NEEDED;
7293 	}
7294 
7295 	return r;
7296 }
7297 
7298 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7299 			    u32 msr_index, u64 data)
7300 {
7301 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7302 	int r;
7303 
7304 	r = kvm_set_msr(vcpu, msr_index, data);
7305 
7306 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7307 		/* Bounce to user space */
7308 		return X86EMUL_IO_NEEDED;
7309 	}
7310 
7311 	return r;
7312 }
7313 
7314 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7315 {
7316 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7317 
7318 	return vcpu->arch.smbase;
7319 }
7320 
7321 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7322 {
7323 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7324 
7325 	vcpu->arch.smbase = smbase;
7326 }
7327 
7328 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7329 			      u32 pmc)
7330 {
7331 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7332 }
7333 
7334 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7335 			     u32 pmc, u64 *pdata)
7336 {
7337 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7338 }
7339 
7340 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7341 {
7342 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7343 }
7344 
7345 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7346 			      struct x86_instruction_info *info,
7347 			      enum x86_intercept_stage stage)
7348 {
7349 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7350 					    &ctxt->exception);
7351 }
7352 
7353 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7354 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7355 			      bool exact_only)
7356 {
7357 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7358 }
7359 
7360 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7361 {
7362 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7363 }
7364 
7365 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7366 {
7367 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7368 }
7369 
7370 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7371 {
7372 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7373 }
7374 
7375 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7376 {
7377 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7378 }
7379 
7380 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7381 {
7382 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7383 }
7384 
7385 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7386 {
7387 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7388 }
7389 
7390 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7391 {
7392 	return emul_to_vcpu(ctxt)->arch.hflags;
7393 }
7394 
7395 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7396 {
7397 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7398 
7399 	kvm_smm_changed(vcpu, false);
7400 }
7401 
7402 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7403 				  const char *smstate)
7404 {
7405 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7406 }
7407 
7408 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7409 {
7410 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7411 }
7412 
7413 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7414 {
7415 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7416 }
7417 
7418 static const struct x86_emulate_ops emulate_ops = {
7419 	.read_gpr            = emulator_read_gpr,
7420 	.write_gpr           = emulator_write_gpr,
7421 	.read_std            = emulator_read_std,
7422 	.write_std           = emulator_write_std,
7423 	.read_phys           = kvm_read_guest_phys_system,
7424 	.fetch               = kvm_fetch_guest_virt,
7425 	.read_emulated       = emulator_read_emulated,
7426 	.write_emulated      = emulator_write_emulated,
7427 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7428 	.invlpg              = emulator_invlpg,
7429 	.pio_in_emulated     = emulator_pio_in_emulated,
7430 	.pio_out_emulated    = emulator_pio_out_emulated,
7431 	.get_segment         = emulator_get_segment,
7432 	.set_segment         = emulator_set_segment,
7433 	.get_cached_segment_base = emulator_get_cached_segment_base,
7434 	.get_gdt             = emulator_get_gdt,
7435 	.get_idt	     = emulator_get_idt,
7436 	.set_gdt             = emulator_set_gdt,
7437 	.set_idt	     = emulator_set_idt,
7438 	.get_cr              = emulator_get_cr,
7439 	.set_cr              = emulator_set_cr,
7440 	.cpl                 = emulator_get_cpl,
7441 	.get_dr              = emulator_get_dr,
7442 	.set_dr              = emulator_set_dr,
7443 	.get_smbase          = emulator_get_smbase,
7444 	.set_smbase          = emulator_set_smbase,
7445 	.set_msr             = emulator_set_msr,
7446 	.get_msr             = emulator_get_msr,
7447 	.check_pmc	     = emulator_check_pmc,
7448 	.read_pmc            = emulator_read_pmc,
7449 	.halt                = emulator_halt,
7450 	.wbinvd              = emulator_wbinvd,
7451 	.fix_hypercall       = emulator_fix_hypercall,
7452 	.intercept           = emulator_intercept,
7453 	.get_cpuid           = emulator_get_cpuid,
7454 	.guest_has_long_mode = emulator_guest_has_long_mode,
7455 	.guest_has_movbe     = emulator_guest_has_movbe,
7456 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7457 	.set_nmi_mask        = emulator_set_nmi_mask,
7458 	.get_hflags          = emulator_get_hflags,
7459 	.exiting_smm         = emulator_exiting_smm,
7460 	.leave_smm           = emulator_leave_smm,
7461 	.triple_fault        = emulator_triple_fault,
7462 	.set_xcr             = emulator_set_xcr,
7463 };
7464 
7465 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7466 {
7467 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7468 	/*
7469 	 * an sti; sti; sequence only disable interrupts for the first
7470 	 * instruction. So, if the last instruction, be it emulated or
7471 	 * not, left the system with the INT_STI flag enabled, it
7472 	 * means that the last instruction is an sti. We should not
7473 	 * leave the flag on in this case. The same goes for mov ss
7474 	 */
7475 	if (int_shadow & mask)
7476 		mask = 0;
7477 	if (unlikely(int_shadow || mask)) {
7478 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7479 		if (!mask)
7480 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7481 	}
7482 }
7483 
7484 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7485 {
7486 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7487 	if (ctxt->exception.vector == PF_VECTOR)
7488 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7489 
7490 	if (ctxt->exception.error_code_valid)
7491 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7492 				      ctxt->exception.error_code);
7493 	else
7494 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7495 	return false;
7496 }
7497 
7498 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7499 {
7500 	struct x86_emulate_ctxt *ctxt;
7501 
7502 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7503 	if (!ctxt) {
7504 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7505 		return NULL;
7506 	}
7507 
7508 	ctxt->vcpu = vcpu;
7509 	ctxt->ops = &emulate_ops;
7510 	vcpu->arch.emulate_ctxt = ctxt;
7511 
7512 	return ctxt;
7513 }
7514 
7515 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7516 {
7517 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7518 	int cs_db, cs_l;
7519 
7520 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7521 
7522 	ctxt->gpa_available = false;
7523 	ctxt->eflags = kvm_get_rflags(vcpu);
7524 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7525 
7526 	ctxt->eip = kvm_rip_read(vcpu);
7527 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7528 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7529 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7530 		     cs_db				? X86EMUL_MODE_PROT32 :
7531 							  X86EMUL_MODE_PROT16;
7532 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7533 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7534 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7535 
7536 	ctxt->interruptibility = 0;
7537 	ctxt->have_exception = false;
7538 	ctxt->exception.vector = -1;
7539 	ctxt->perm_ok = false;
7540 
7541 	init_decode_cache(ctxt);
7542 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7543 }
7544 
7545 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7546 {
7547 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7548 	int ret;
7549 
7550 	init_emulate_ctxt(vcpu);
7551 
7552 	ctxt->op_bytes = 2;
7553 	ctxt->ad_bytes = 2;
7554 	ctxt->_eip = ctxt->eip + inc_eip;
7555 	ret = emulate_int_real(ctxt, irq);
7556 
7557 	if (ret != X86EMUL_CONTINUE) {
7558 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7559 	} else {
7560 		ctxt->eip = ctxt->_eip;
7561 		kvm_rip_write(vcpu, ctxt->eip);
7562 		kvm_set_rflags(vcpu, ctxt->eflags);
7563 	}
7564 }
7565 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7566 
7567 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7568 					   u8 ndata, u8 *insn_bytes, u8 insn_size)
7569 {
7570 	struct kvm_run *run = vcpu->run;
7571 	u64 info[5];
7572 	u8 info_start;
7573 
7574 	/*
7575 	 * Zero the whole array used to retrieve the exit info, as casting to
7576 	 * u32 for select entries will leave some chunks uninitialized.
7577 	 */
7578 	memset(&info, 0, sizeof(info));
7579 
7580 	static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
7581 					   &info[2], (u32 *)&info[3],
7582 					   (u32 *)&info[4]);
7583 
7584 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7585 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7586 
7587 	/*
7588 	 * There's currently space for 13 entries, but 5 are used for the exit
7589 	 * reason and info.  Restrict to 4 to reduce the maintenance burden
7590 	 * when expanding kvm_run.emulation_failure in the future.
7591 	 */
7592 	if (WARN_ON_ONCE(ndata > 4))
7593 		ndata = 4;
7594 
7595 	/* Always include the flags as a 'data' entry. */
7596 	info_start = 1;
7597 	run->emulation_failure.flags = 0;
7598 
7599 	if (insn_size) {
7600 		BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
7601 			      sizeof(run->emulation_failure.insn_bytes) != 16));
7602 		info_start += 2;
7603 		run->emulation_failure.flags |=
7604 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7605 		run->emulation_failure.insn_size = insn_size;
7606 		memset(run->emulation_failure.insn_bytes, 0x90,
7607 		       sizeof(run->emulation_failure.insn_bytes));
7608 		memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
7609 	}
7610 
7611 	memcpy(&run->internal.data[info_start], info, sizeof(info));
7612 	memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
7613 	       ndata * sizeof(data[0]));
7614 
7615 	run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
7616 }
7617 
7618 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
7619 {
7620 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7621 
7622 	prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
7623 				       ctxt->fetch.end - ctxt->fetch.data);
7624 }
7625 
7626 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7627 					  u8 ndata)
7628 {
7629 	prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
7630 }
7631 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
7632 
7633 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7634 {
7635 	__kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
7636 }
7637 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
7638 
7639 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7640 {
7641 	struct kvm *kvm = vcpu->kvm;
7642 
7643 	++vcpu->stat.insn_emulation_fail;
7644 	trace_kvm_emulate_insn_failed(vcpu);
7645 
7646 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7647 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7648 		return 1;
7649 	}
7650 
7651 	if (kvm->arch.exit_on_emulation_error ||
7652 	    (emulation_type & EMULTYPE_SKIP)) {
7653 		prepare_emulation_ctxt_failure_exit(vcpu);
7654 		return 0;
7655 	}
7656 
7657 	kvm_queue_exception(vcpu, UD_VECTOR);
7658 
7659 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7660 		prepare_emulation_ctxt_failure_exit(vcpu);
7661 		return 0;
7662 	}
7663 
7664 	return 1;
7665 }
7666 
7667 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7668 				  bool write_fault_to_shadow_pgtable,
7669 				  int emulation_type)
7670 {
7671 	gpa_t gpa = cr2_or_gpa;
7672 	kvm_pfn_t pfn;
7673 
7674 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7675 		return false;
7676 
7677 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7678 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7679 		return false;
7680 
7681 	if (!vcpu->arch.mmu->direct_map) {
7682 		/*
7683 		 * Write permission should be allowed since only
7684 		 * write access need to be emulated.
7685 		 */
7686 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7687 
7688 		/*
7689 		 * If the mapping is invalid in guest, let cpu retry
7690 		 * it to generate fault.
7691 		 */
7692 		if (gpa == UNMAPPED_GVA)
7693 			return true;
7694 	}
7695 
7696 	/*
7697 	 * Do not retry the unhandleable instruction if it faults on the
7698 	 * readonly host memory, otherwise it will goto a infinite loop:
7699 	 * retry instruction -> write #PF -> emulation fail -> retry
7700 	 * instruction -> ...
7701 	 */
7702 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7703 
7704 	/*
7705 	 * If the instruction failed on the error pfn, it can not be fixed,
7706 	 * report the error to userspace.
7707 	 */
7708 	if (is_error_noslot_pfn(pfn))
7709 		return false;
7710 
7711 	kvm_release_pfn_clean(pfn);
7712 
7713 	/* The instructions are well-emulated on direct mmu. */
7714 	if (vcpu->arch.mmu->direct_map) {
7715 		unsigned int indirect_shadow_pages;
7716 
7717 		write_lock(&vcpu->kvm->mmu_lock);
7718 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7719 		write_unlock(&vcpu->kvm->mmu_lock);
7720 
7721 		if (indirect_shadow_pages)
7722 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7723 
7724 		return true;
7725 	}
7726 
7727 	/*
7728 	 * if emulation was due to access to shadowed page table
7729 	 * and it failed try to unshadow page and re-enter the
7730 	 * guest to let CPU execute the instruction.
7731 	 */
7732 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7733 
7734 	/*
7735 	 * If the access faults on its page table, it can not
7736 	 * be fixed by unprotecting shadow page and it should
7737 	 * be reported to userspace.
7738 	 */
7739 	return !write_fault_to_shadow_pgtable;
7740 }
7741 
7742 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7743 			      gpa_t cr2_or_gpa,  int emulation_type)
7744 {
7745 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7746 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7747 
7748 	last_retry_eip = vcpu->arch.last_retry_eip;
7749 	last_retry_addr = vcpu->arch.last_retry_addr;
7750 
7751 	/*
7752 	 * If the emulation is caused by #PF and it is non-page_table
7753 	 * writing instruction, it means the VM-EXIT is caused by shadow
7754 	 * page protected, we can zap the shadow page and retry this
7755 	 * instruction directly.
7756 	 *
7757 	 * Note: if the guest uses a non-page-table modifying instruction
7758 	 * on the PDE that points to the instruction, then we will unmap
7759 	 * the instruction and go to an infinite loop. So, we cache the
7760 	 * last retried eip and the last fault address, if we meet the eip
7761 	 * and the address again, we can break out of the potential infinite
7762 	 * loop.
7763 	 */
7764 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7765 
7766 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7767 		return false;
7768 
7769 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7770 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7771 		return false;
7772 
7773 	if (x86_page_table_writing_insn(ctxt))
7774 		return false;
7775 
7776 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7777 		return false;
7778 
7779 	vcpu->arch.last_retry_eip = ctxt->eip;
7780 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7781 
7782 	if (!vcpu->arch.mmu->direct_map)
7783 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7784 
7785 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7786 
7787 	return true;
7788 }
7789 
7790 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7791 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7792 
7793 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7794 {
7795 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7796 
7797 	if (entering_smm) {
7798 		vcpu->arch.hflags |= HF_SMM_MASK;
7799 	} else {
7800 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7801 
7802 		/* Process a latched INIT or SMI, if any.  */
7803 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7804 
7805 		/*
7806 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7807 		 * on SMM exit we still need to reload them from
7808 		 * guest memory
7809 		 */
7810 		vcpu->arch.pdptrs_from_userspace = false;
7811 	}
7812 
7813 	kvm_mmu_reset_context(vcpu);
7814 }
7815 
7816 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7817 				unsigned long *db)
7818 {
7819 	u32 dr6 = 0;
7820 	int i;
7821 	u32 enable, rwlen;
7822 
7823 	enable = dr7;
7824 	rwlen = dr7 >> 16;
7825 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7826 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7827 			dr6 |= (1 << i);
7828 	return dr6;
7829 }
7830 
7831 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7832 {
7833 	struct kvm_run *kvm_run = vcpu->run;
7834 
7835 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7836 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7837 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7838 		kvm_run->debug.arch.exception = DB_VECTOR;
7839 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7840 		return 0;
7841 	}
7842 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7843 	return 1;
7844 }
7845 
7846 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7847 {
7848 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7849 	int r;
7850 
7851 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7852 	if (unlikely(!r))
7853 		return 0;
7854 
7855 	/*
7856 	 * rflags is the old, "raw" value of the flags.  The new value has
7857 	 * not been saved yet.
7858 	 *
7859 	 * This is correct even for TF set by the guest, because "the
7860 	 * processor will not generate this exception after the instruction
7861 	 * that sets the TF flag".
7862 	 */
7863 	if (unlikely(rflags & X86_EFLAGS_TF))
7864 		r = kvm_vcpu_do_singlestep(vcpu);
7865 	return r;
7866 }
7867 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7868 
7869 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7870 {
7871 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7872 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7873 		struct kvm_run *kvm_run = vcpu->run;
7874 		unsigned long eip = kvm_get_linear_rip(vcpu);
7875 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7876 					   vcpu->arch.guest_debug_dr7,
7877 					   vcpu->arch.eff_db);
7878 
7879 		if (dr6 != 0) {
7880 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7881 			kvm_run->debug.arch.pc = eip;
7882 			kvm_run->debug.arch.exception = DB_VECTOR;
7883 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7884 			*r = 0;
7885 			return true;
7886 		}
7887 	}
7888 
7889 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7890 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7891 		unsigned long eip = kvm_get_linear_rip(vcpu);
7892 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7893 					   vcpu->arch.dr7,
7894 					   vcpu->arch.db);
7895 
7896 		if (dr6 != 0) {
7897 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7898 			*r = 1;
7899 			return true;
7900 		}
7901 	}
7902 
7903 	return false;
7904 }
7905 
7906 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7907 {
7908 	switch (ctxt->opcode_len) {
7909 	case 1:
7910 		switch (ctxt->b) {
7911 		case 0xe4:	/* IN */
7912 		case 0xe5:
7913 		case 0xec:
7914 		case 0xed:
7915 		case 0xe6:	/* OUT */
7916 		case 0xe7:
7917 		case 0xee:
7918 		case 0xef:
7919 		case 0x6c:	/* INS */
7920 		case 0x6d:
7921 		case 0x6e:	/* OUTS */
7922 		case 0x6f:
7923 			return true;
7924 		}
7925 		break;
7926 	case 2:
7927 		switch (ctxt->b) {
7928 		case 0x33:	/* RDPMC */
7929 			return true;
7930 		}
7931 		break;
7932 	}
7933 
7934 	return false;
7935 }
7936 
7937 /*
7938  * Decode to be emulated instruction. Return EMULATION_OK if success.
7939  */
7940 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7941 				    void *insn, int insn_len)
7942 {
7943 	int r = EMULATION_OK;
7944 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7945 
7946 	init_emulate_ctxt(vcpu);
7947 
7948 	/*
7949 	 * We will reenter on the same instruction since we do not set
7950 	 * complete_userspace_io. This does not handle watchpoints yet,
7951 	 * those would be handled in the emulate_ops.
7952 	 */
7953 	if (!(emulation_type & EMULTYPE_SKIP) &&
7954 	    kvm_vcpu_check_breakpoint(vcpu, &r))
7955 		return r;
7956 
7957 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7958 
7959 	trace_kvm_emulate_insn_start(vcpu);
7960 	++vcpu->stat.insn_emulation;
7961 
7962 	return r;
7963 }
7964 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7965 
7966 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7967 			    int emulation_type, void *insn, int insn_len)
7968 {
7969 	int r;
7970 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7971 	bool writeback = true;
7972 	bool write_fault_to_spt;
7973 
7974 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7975 		return 1;
7976 
7977 	vcpu->arch.l1tf_flush_l1d = true;
7978 
7979 	/*
7980 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7981 	 * never reused.
7982 	 */
7983 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7984 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7985 
7986 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7987 		kvm_clear_exception_queue(vcpu);
7988 
7989 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
7990 						    insn, insn_len);
7991 		if (r != EMULATION_OK)  {
7992 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7993 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7994 				kvm_queue_exception(vcpu, UD_VECTOR);
7995 				return 1;
7996 			}
7997 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7998 						  write_fault_to_spt,
7999 						  emulation_type))
8000 				return 1;
8001 			if (ctxt->have_exception) {
8002 				/*
8003 				 * #UD should result in just EMULATION_FAILED, and trap-like
8004 				 * exception should not be encountered during decode.
8005 				 */
8006 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8007 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8008 				inject_emulated_exception(vcpu);
8009 				return 1;
8010 			}
8011 			return handle_emulation_failure(vcpu, emulation_type);
8012 		}
8013 	}
8014 
8015 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8016 	    !is_vmware_backdoor_opcode(ctxt)) {
8017 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8018 		return 1;
8019 	}
8020 
8021 	/*
8022 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
8023 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
8024 	 * updating interruptibility state and injecting single-step #DBs.
8025 	 */
8026 	if (emulation_type & EMULTYPE_SKIP) {
8027 		kvm_rip_write(vcpu, ctxt->_eip);
8028 		if (ctxt->eflags & X86_EFLAGS_RF)
8029 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8030 		return 1;
8031 	}
8032 
8033 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8034 		return 1;
8035 
8036 	/* this is needed for vmware backdoor interface to work since it
8037 	   changes registers values  during IO operation */
8038 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8039 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8040 		emulator_invalidate_register_cache(ctxt);
8041 	}
8042 
8043 restart:
8044 	if (emulation_type & EMULTYPE_PF) {
8045 		/* Save the faulting GPA (cr2) in the address field */
8046 		ctxt->exception.address = cr2_or_gpa;
8047 
8048 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
8049 		if (vcpu->arch.mmu->direct_map) {
8050 			ctxt->gpa_available = true;
8051 			ctxt->gpa_val = cr2_or_gpa;
8052 		}
8053 	} else {
8054 		/* Sanitize the address out of an abundance of paranoia. */
8055 		ctxt->exception.address = 0;
8056 	}
8057 
8058 	r = x86_emulate_insn(ctxt);
8059 
8060 	if (r == EMULATION_INTERCEPTED)
8061 		return 1;
8062 
8063 	if (r == EMULATION_FAILED) {
8064 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8065 					emulation_type))
8066 			return 1;
8067 
8068 		return handle_emulation_failure(vcpu, emulation_type);
8069 	}
8070 
8071 	if (ctxt->have_exception) {
8072 		r = 1;
8073 		if (inject_emulated_exception(vcpu))
8074 			return r;
8075 	} else if (vcpu->arch.pio.count) {
8076 		if (!vcpu->arch.pio.in) {
8077 			/* FIXME: return into emulator if single-stepping.  */
8078 			vcpu->arch.pio.count = 0;
8079 		} else {
8080 			writeback = false;
8081 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
8082 		}
8083 		r = 0;
8084 	} else if (vcpu->mmio_needed) {
8085 		++vcpu->stat.mmio_exits;
8086 
8087 		if (!vcpu->mmio_is_write)
8088 			writeback = false;
8089 		r = 0;
8090 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8091 	} else if (r == EMULATION_RESTART)
8092 		goto restart;
8093 	else
8094 		r = 1;
8095 
8096 	if (writeback) {
8097 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8098 		toggle_interruptibility(vcpu, ctxt->interruptibility);
8099 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8100 		if (!ctxt->have_exception ||
8101 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8102 			kvm_rip_write(vcpu, ctxt->eip);
8103 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8104 				r = kvm_vcpu_do_singlestep(vcpu);
8105 			if (kvm_x86_ops.update_emulated_instruction)
8106 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
8107 			__kvm_set_rflags(vcpu, ctxt->eflags);
8108 		}
8109 
8110 		/*
8111 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8112 		 * do nothing, and it will be requested again as soon as
8113 		 * the shadow expires.  But we still need to check here,
8114 		 * because POPF has no interrupt shadow.
8115 		 */
8116 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8117 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8118 	} else
8119 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8120 
8121 	return r;
8122 }
8123 
8124 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8125 {
8126 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8127 }
8128 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8129 
8130 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8131 					void *insn, int insn_len)
8132 {
8133 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8134 }
8135 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8136 
8137 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8138 {
8139 	vcpu->arch.pio.count = 0;
8140 	return 1;
8141 }
8142 
8143 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8144 {
8145 	vcpu->arch.pio.count = 0;
8146 
8147 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8148 		return 1;
8149 
8150 	return kvm_skip_emulated_instruction(vcpu);
8151 }
8152 
8153 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8154 			    unsigned short port)
8155 {
8156 	unsigned long val = kvm_rax_read(vcpu);
8157 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8158 
8159 	if (ret)
8160 		return ret;
8161 
8162 	/*
8163 	 * Workaround userspace that relies on old KVM behavior of %rip being
8164 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8165 	 */
8166 	if (port == 0x7e &&
8167 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8168 		vcpu->arch.complete_userspace_io =
8169 			complete_fast_pio_out_port_0x7e;
8170 		kvm_skip_emulated_instruction(vcpu);
8171 	} else {
8172 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8173 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8174 	}
8175 	return 0;
8176 }
8177 
8178 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8179 {
8180 	unsigned long val;
8181 
8182 	/* We should only ever be called with arch.pio.count equal to 1 */
8183 	BUG_ON(vcpu->arch.pio.count != 1);
8184 
8185 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8186 		vcpu->arch.pio.count = 0;
8187 		return 1;
8188 	}
8189 
8190 	/* For size less than 4 we merge, else we zero extend */
8191 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8192 
8193 	/*
8194 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8195 	 * the copy and tracing
8196 	 */
8197 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8198 	kvm_rax_write(vcpu, val);
8199 
8200 	return kvm_skip_emulated_instruction(vcpu);
8201 }
8202 
8203 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8204 			   unsigned short port)
8205 {
8206 	unsigned long val;
8207 	int ret;
8208 
8209 	/* For size less than 4 we merge, else we zero extend */
8210 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8211 
8212 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8213 	if (ret) {
8214 		kvm_rax_write(vcpu, val);
8215 		return ret;
8216 	}
8217 
8218 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8219 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8220 
8221 	return 0;
8222 }
8223 
8224 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8225 {
8226 	int ret;
8227 
8228 	if (in)
8229 		ret = kvm_fast_pio_in(vcpu, size, port);
8230 	else
8231 		ret = kvm_fast_pio_out(vcpu, size, port);
8232 	return ret && kvm_skip_emulated_instruction(vcpu);
8233 }
8234 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8235 
8236 static int kvmclock_cpu_down_prep(unsigned int cpu)
8237 {
8238 	__this_cpu_write(cpu_tsc_khz, 0);
8239 	return 0;
8240 }
8241 
8242 static void tsc_khz_changed(void *data)
8243 {
8244 	struct cpufreq_freqs *freq = data;
8245 	unsigned long khz = 0;
8246 
8247 	if (data)
8248 		khz = freq->new;
8249 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8250 		khz = cpufreq_quick_get(raw_smp_processor_id());
8251 	if (!khz)
8252 		khz = tsc_khz;
8253 	__this_cpu_write(cpu_tsc_khz, khz);
8254 }
8255 
8256 #ifdef CONFIG_X86_64
8257 static void kvm_hyperv_tsc_notifier(void)
8258 {
8259 	struct kvm *kvm;
8260 	int cpu;
8261 
8262 	mutex_lock(&kvm_lock);
8263 	list_for_each_entry(kvm, &vm_list, vm_list)
8264 		kvm_make_mclock_inprogress_request(kvm);
8265 
8266 	/* no guest entries from this point */
8267 	hyperv_stop_tsc_emulation();
8268 
8269 	/* TSC frequency always matches when on Hyper-V */
8270 	for_each_present_cpu(cpu)
8271 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8272 	kvm_max_guest_tsc_khz = tsc_khz;
8273 
8274 	list_for_each_entry(kvm, &vm_list, vm_list) {
8275 		__kvm_start_pvclock_update(kvm);
8276 		pvclock_update_vm_gtod_copy(kvm);
8277 		kvm_end_pvclock_update(kvm);
8278 	}
8279 
8280 	mutex_unlock(&kvm_lock);
8281 }
8282 #endif
8283 
8284 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8285 {
8286 	struct kvm *kvm;
8287 	struct kvm_vcpu *vcpu;
8288 	int i, send_ipi = 0;
8289 
8290 	/*
8291 	 * We allow guests to temporarily run on slowing clocks,
8292 	 * provided we notify them after, or to run on accelerating
8293 	 * clocks, provided we notify them before.  Thus time never
8294 	 * goes backwards.
8295 	 *
8296 	 * However, we have a problem.  We can't atomically update
8297 	 * the frequency of a given CPU from this function; it is
8298 	 * merely a notifier, which can be called from any CPU.
8299 	 * Changing the TSC frequency at arbitrary points in time
8300 	 * requires a recomputation of local variables related to
8301 	 * the TSC for each VCPU.  We must flag these local variables
8302 	 * to be updated and be sure the update takes place with the
8303 	 * new frequency before any guests proceed.
8304 	 *
8305 	 * Unfortunately, the combination of hotplug CPU and frequency
8306 	 * change creates an intractable locking scenario; the order
8307 	 * of when these callouts happen is undefined with respect to
8308 	 * CPU hotplug, and they can race with each other.  As such,
8309 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8310 	 * undefined; you can actually have a CPU frequency change take
8311 	 * place in between the computation of X and the setting of the
8312 	 * variable.  To protect against this problem, all updates of
8313 	 * the per_cpu tsc_khz variable are done in an interrupt
8314 	 * protected IPI, and all callers wishing to update the value
8315 	 * must wait for a synchronous IPI to complete (which is trivial
8316 	 * if the caller is on the CPU already).  This establishes the
8317 	 * necessary total order on variable updates.
8318 	 *
8319 	 * Note that because a guest time update may take place
8320 	 * anytime after the setting of the VCPU's request bit, the
8321 	 * correct TSC value must be set before the request.  However,
8322 	 * to ensure the update actually makes it to any guest which
8323 	 * starts running in hardware virtualization between the set
8324 	 * and the acquisition of the spinlock, we must also ping the
8325 	 * CPU after setting the request bit.
8326 	 *
8327 	 */
8328 
8329 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8330 
8331 	mutex_lock(&kvm_lock);
8332 	list_for_each_entry(kvm, &vm_list, vm_list) {
8333 		kvm_for_each_vcpu(i, vcpu, kvm) {
8334 			if (vcpu->cpu != cpu)
8335 				continue;
8336 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8337 			if (vcpu->cpu != raw_smp_processor_id())
8338 				send_ipi = 1;
8339 		}
8340 	}
8341 	mutex_unlock(&kvm_lock);
8342 
8343 	if (freq->old < freq->new && send_ipi) {
8344 		/*
8345 		 * We upscale the frequency.  Must make the guest
8346 		 * doesn't see old kvmclock values while running with
8347 		 * the new frequency, otherwise we risk the guest sees
8348 		 * time go backwards.
8349 		 *
8350 		 * In case we update the frequency for another cpu
8351 		 * (which might be in guest context) send an interrupt
8352 		 * to kick the cpu out of guest context.  Next time
8353 		 * guest context is entered kvmclock will be updated,
8354 		 * so the guest will not see stale values.
8355 		 */
8356 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8357 	}
8358 }
8359 
8360 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8361 				     void *data)
8362 {
8363 	struct cpufreq_freqs *freq = data;
8364 	int cpu;
8365 
8366 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8367 		return 0;
8368 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8369 		return 0;
8370 
8371 	for_each_cpu(cpu, freq->policy->cpus)
8372 		__kvmclock_cpufreq_notifier(freq, cpu);
8373 
8374 	return 0;
8375 }
8376 
8377 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8378 	.notifier_call  = kvmclock_cpufreq_notifier
8379 };
8380 
8381 static int kvmclock_cpu_online(unsigned int cpu)
8382 {
8383 	tsc_khz_changed(NULL);
8384 	return 0;
8385 }
8386 
8387 static void kvm_timer_init(void)
8388 {
8389 	max_tsc_khz = tsc_khz;
8390 
8391 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8392 #ifdef CONFIG_CPU_FREQ
8393 		struct cpufreq_policy *policy;
8394 		int cpu;
8395 
8396 		cpu = get_cpu();
8397 		policy = cpufreq_cpu_get(cpu);
8398 		if (policy) {
8399 			if (policy->cpuinfo.max_freq)
8400 				max_tsc_khz = policy->cpuinfo.max_freq;
8401 			cpufreq_cpu_put(policy);
8402 		}
8403 		put_cpu();
8404 #endif
8405 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8406 					  CPUFREQ_TRANSITION_NOTIFIER);
8407 	}
8408 
8409 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8410 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8411 }
8412 
8413 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8414 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8415 
8416 int kvm_is_in_guest(void)
8417 {
8418 	return __this_cpu_read(current_vcpu) != NULL;
8419 }
8420 
8421 static int kvm_is_user_mode(void)
8422 {
8423 	int user_mode = 3;
8424 
8425 	if (__this_cpu_read(current_vcpu))
8426 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8427 
8428 	return user_mode != 0;
8429 }
8430 
8431 static unsigned long kvm_get_guest_ip(void)
8432 {
8433 	unsigned long ip = 0;
8434 
8435 	if (__this_cpu_read(current_vcpu))
8436 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8437 
8438 	return ip;
8439 }
8440 
8441 static void kvm_handle_intel_pt_intr(void)
8442 {
8443 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8444 
8445 	kvm_make_request(KVM_REQ_PMI, vcpu);
8446 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8447 			(unsigned long *)&vcpu->arch.pmu.global_status);
8448 }
8449 
8450 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8451 	.is_in_guest		= kvm_is_in_guest,
8452 	.is_user_mode		= kvm_is_user_mode,
8453 	.get_guest_ip		= kvm_get_guest_ip,
8454 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8455 };
8456 
8457 #ifdef CONFIG_X86_64
8458 static void pvclock_gtod_update_fn(struct work_struct *work)
8459 {
8460 	struct kvm *kvm;
8461 
8462 	struct kvm_vcpu *vcpu;
8463 	int i;
8464 
8465 	mutex_lock(&kvm_lock);
8466 	list_for_each_entry(kvm, &vm_list, vm_list)
8467 		kvm_for_each_vcpu(i, vcpu, kvm)
8468 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8469 	atomic_set(&kvm_guest_has_master_clock, 0);
8470 	mutex_unlock(&kvm_lock);
8471 }
8472 
8473 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8474 
8475 /*
8476  * Indirection to move queue_work() out of the tk_core.seq write held
8477  * region to prevent possible deadlocks against time accessors which
8478  * are invoked with work related locks held.
8479  */
8480 static void pvclock_irq_work_fn(struct irq_work *w)
8481 {
8482 	queue_work(system_long_wq, &pvclock_gtod_work);
8483 }
8484 
8485 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8486 
8487 /*
8488  * Notification about pvclock gtod data update.
8489  */
8490 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8491 			       void *priv)
8492 {
8493 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8494 	struct timekeeper *tk = priv;
8495 
8496 	update_pvclock_gtod(tk);
8497 
8498 	/*
8499 	 * Disable master clock if host does not trust, or does not use,
8500 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8501 	 * this is invoked with tk_core.seq write held.
8502 	 */
8503 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8504 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8505 		irq_work_queue(&pvclock_irq_work);
8506 	return 0;
8507 }
8508 
8509 static struct notifier_block pvclock_gtod_notifier = {
8510 	.notifier_call = pvclock_gtod_notify,
8511 };
8512 #endif
8513 
8514 int kvm_arch_init(void *opaque)
8515 {
8516 	struct kvm_x86_init_ops *ops = opaque;
8517 	int r;
8518 
8519 	if (kvm_x86_ops.hardware_enable) {
8520 		pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8521 		r = -EEXIST;
8522 		goto out;
8523 	}
8524 
8525 	if (!ops->cpu_has_kvm_support()) {
8526 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8527 				   ops->runtime_ops->name);
8528 		r = -EOPNOTSUPP;
8529 		goto out;
8530 	}
8531 	if (ops->disabled_by_bios()) {
8532 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8533 				   ops->runtime_ops->name);
8534 		r = -EOPNOTSUPP;
8535 		goto out;
8536 	}
8537 
8538 	/*
8539 	 * KVM explicitly assumes that the guest has an FPU and
8540 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8541 	 * vCPU's FPU state as a fxregs_state struct.
8542 	 */
8543 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8544 		printk(KERN_ERR "kvm: inadequate fpu\n");
8545 		r = -EOPNOTSUPP;
8546 		goto out;
8547 	}
8548 
8549 	r = -ENOMEM;
8550 
8551 	x86_emulator_cache = kvm_alloc_emulator_cache();
8552 	if (!x86_emulator_cache) {
8553 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8554 		goto out;
8555 	}
8556 
8557 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8558 	if (!user_return_msrs) {
8559 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8560 		goto out_free_x86_emulator_cache;
8561 	}
8562 	kvm_nr_uret_msrs = 0;
8563 
8564 	r = kvm_mmu_module_init();
8565 	if (r)
8566 		goto out_free_percpu;
8567 
8568 	kvm_timer_init();
8569 
8570 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8571 
8572 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8573 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8574 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8575 	}
8576 
8577 	if (pi_inject_timer == -1)
8578 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8579 #ifdef CONFIG_X86_64
8580 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8581 
8582 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8583 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8584 #endif
8585 
8586 	return 0;
8587 
8588 out_free_percpu:
8589 	free_percpu(user_return_msrs);
8590 out_free_x86_emulator_cache:
8591 	kmem_cache_destroy(x86_emulator_cache);
8592 out:
8593 	return r;
8594 }
8595 
8596 void kvm_arch_exit(void)
8597 {
8598 #ifdef CONFIG_X86_64
8599 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8600 		clear_hv_tscchange_cb();
8601 #endif
8602 	kvm_lapic_exit();
8603 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8604 
8605 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8606 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8607 					    CPUFREQ_TRANSITION_NOTIFIER);
8608 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8609 #ifdef CONFIG_X86_64
8610 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8611 	irq_work_sync(&pvclock_irq_work);
8612 	cancel_work_sync(&pvclock_gtod_work);
8613 #endif
8614 	kvm_x86_ops.hardware_enable = NULL;
8615 	kvm_mmu_module_exit();
8616 	free_percpu(user_return_msrs);
8617 	kmem_cache_destroy(x86_emulator_cache);
8618 #ifdef CONFIG_KVM_XEN
8619 	static_key_deferred_flush(&kvm_xen_enabled);
8620 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8621 #endif
8622 }
8623 
8624 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8625 {
8626 	++vcpu->stat.halt_exits;
8627 	if (lapic_in_kernel(vcpu)) {
8628 		vcpu->arch.mp_state = state;
8629 		return 1;
8630 	} else {
8631 		vcpu->run->exit_reason = reason;
8632 		return 0;
8633 	}
8634 }
8635 
8636 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8637 {
8638 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8639 }
8640 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8641 
8642 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8643 {
8644 	int ret = kvm_skip_emulated_instruction(vcpu);
8645 	/*
8646 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8647 	 * KVM_EXIT_DEBUG here.
8648 	 */
8649 	return kvm_vcpu_halt(vcpu) && ret;
8650 }
8651 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8652 
8653 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8654 {
8655 	int ret = kvm_skip_emulated_instruction(vcpu);
8656 
8657 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8658 }
8659 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8660 
8661 #ifdef CONFIG_X86_64
8662 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8663 			        unsigned long clock_type)
8664 {
8665 	struct kvm_clock_pairing clock_pairing;
8666 	struct timespec64 ts;
8667 	u64 cycle;
8668 	int ret;
8669 
8670 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8671 		return -KVM_EOPNOTSUPP;
8672 
8673 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8674 		return -KVM_EOPNOTSUPP;
8675 
8676 	clock_pairing.sec = ts.tv_sec;
8677 	clock_pairing.nsec = ts.tv_nsec;
8678 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8679 	clock_pairing.flags = 0;
8680 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8681 
8682 	ret = 0;
8683 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8684 			    sizeof(struct kvm_clock_pairing)))
8685 		ret = -KVM_EFAULT;
8686 
8687 	return ret;
8688 }
8689 #endif
8690 
8691 /*
8692  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8693  *
8694  * @apicid - apicid of vcpu to be kicked.
8695  */
8696 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8697 {
8698 	struct kvm_lapic_irq lapic_irq;
8699 
8700 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8701 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8702 	lapic_irq.level = 0;
8703 	lapic_irq.dest_id = apicid;
8704 	lapic_irq.msi_redir_hint = false;
8705 
8706 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8707 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8708 }
8709 
8710 bool kvm_apicv_activated(struct kvm *kvm)
8711 {
8712 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8713 }
8714 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8715 
8716 static void kvm_apicv_init(struct kvm *kvm)
8717 {
8718 	init_rwsem(&kvm->arch.apicv_update_lock);
8719 
8720 	if (enable_apicv)
8721 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8722 			  &kvm->arch.apicv_inhibit_reasons);
8723 	else
8724 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8725 			&kvm->arch.apicv_inhibit_reasons);
8726 }
8727 
8728 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8729 {
8730 	struct kvm_vcpu *target = NULL;
8731 	struct kvm_apic_map *map;
8732 
8733 	vcpu->stat.directed_yield_attempted++;
8734 
8735 	if (single_task_running())
8736 		goto no_yield;
8737 
8738 	rcu_read_lock();
8739 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8740 
8741 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8742 		target = map->phys_map[dest_id]->vcpu;
8743 
8744 	rcu_read_unlock();
8745 
8746 	if (!target || !READ_ONCE(target->ready))
8747 		goto no_yield;
8748 
8749 	/* Ignore requests to yield to self */
8750 	if (vcpu == target)
8751 		goto no_yield;
8752 
8753 	if (kvm_vcpu_yield_to(target) <= 0)
8754 		goto no_yield;
8755 
8756 	vcpu->stat.directed_yield_successful++;
8757 
8758 no_yield:
8759 	return;
8760 }
8761 
8762 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8763 {
8764 	u64 ret = vcpu->run->hypercall.ret;
8765 
8766 	if (!is_64_bit_mode(vcpu))
8767 		ret = (u32)ret;
8768 	kvm_rax_write(vcpu, ret);
8769 	++vcpu->stat.hypercalls;
8770 	return kvm_skip_emulated_instruction(vcpu);
8771 }
8772 
8773 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8774 {
8775 	unsigned long nr, a0, a1, a2, a3, ret;
8776 	int op_64_bit;
8777 
8778 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8779 		return kvm_xen_hypercall(vcpu);
8780 
8781 	if (kvm_hv_hypercall_enabled(vcpu))
8782 		return kvm_hv_hypercall(vcpu);
8783 
8784 	nr = kvm_rax_read(vcpu);
8785 	a0 = kvm_rbx_read(vcpu);
8786 	a1 = kvm_rcx_read(vcpu);
8787 	a2 = kvm_rdx_read(vcpu);
8788 	a3 = kvm_rsi_read(vcpu);
8789 
8790 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8791 
8792 	op_64_bit = is_64_bit_mode(vcpu);
8793 	if (!op_64_bit) {
8794 		nr &= 0xFFFFFFFF;
8795 		a0 &= 0xFFFFFFFF;
8796 		a1 &= 0xFFFFFFFF;
8797 		a2 &= 0xFFFFFFFF;
8798 		a3 &= 0xFFFFFFFF;
8799 	}
8800 
8801 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8802 		ret = -KVM_EPERM;
8803 		goto out;
8804 	}
8805 
8806 	ret = -KVM_ENOSYS;
8807 
8808 	switch (nr) {
8809 	case KVM_HC_VAPIC_POLL_IRQ:
8810 		ret = 0;
8811 		break;
8812 	case KVM_HC_KICK_CPU:
8813 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8814 			break;
8815 
8816 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8817 		kvm_sched_yield(vcpu, a1);
8818 		ret = 0;
8819 		break;
8820 #ifdef CONFIG_X86_64
8821 	case KVM_HC_CLOCK_PAIRING:
8822 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8823 		break;
8824 #endif
8825 	case KVM_HC_SEND_IPI:
8826 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8827 			break;
8828 
8829 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8830 		break;
8831 	case KVM_HC_SCHED_YIELD:
8832 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8833 			break;
8834 
8835 		kvm_sched_yield(vcpu, a0);
8836 		ret = 0;
8837 		break;
8838 	case KVM_HC_MAP_GPA_RANGE: {
8839 		u64 gpa = a0, npages = a1, attrs = a2;
8840 
8841 		ret = -KVM_ENOSYS;
8842 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8843 			break;
8844 
8845 		if (!PAGE_ALIGNED(gpa) || !npages ||
8846 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8847 			ret = -KVM_EINVAL;
8848 			break;
8849 		}
8850 
8851 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
8852 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
8853 		vcpu->run->hypercall.args[0]  = gpa;
8854 		vcpu->run->hypercall.args[1]  = npages;
8855 		vcpu->run->hypercall.args[2]  = attrs;
8856 		vcpu->run->hypercall.longmode = op_64_bit;
8857 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8858 		return 0;
8859 	}
8860 	default:
8861 		ret = -KVM_ENOSYS;
8862 		break;
8863 	}
8864 out:
8865 	if (!op_64_bit)
8866 		ret = (u32)ret;
8867 	kvm_rax_write(vcpu, ret);
8868 
8869 	++vcpu->stat.hypercalls;
8870 	return kvm_skip_emulated_instruction(vcpu);
8871 }
8872 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8873 
8874 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8875 {
8876 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8877 	char instruction[3];
8878 	unsigned long rip = kvm_rip_read(vcpu);
8879 
8880 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8881 
8882 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8883 		&ctxt->exception);
8884 }
8885 
8886 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8887 {
8888 	return vcpu->run->request_interrupt_window &&
8889 		likely(!pic_in_kernel(vcpu->kvm));
8890 }
8891 
8892 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8893 {
8894 	struct kvm_run *kvm_run = vcpu->run;
8895 
8896 	/*
8897 	 * if_flag is obsolete and useless, so do not bother
8898 	 * setting it for SEV-ES guests.  Userspace can just
8899 	 * use kvm_run->ready_for_interrupt_injection.
8900 	 */
8901 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8902 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8903 
8904 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8905 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8906 
8907 	/*
8908 	 * The call to kvm_ready_for_interrupt_injection() may end up in
8909 	 * kvm_xen_has_interrupt() which may require the srcu lock to be
8910 	 * held, to protect against changes in the vcpu_info address.
8911 	 */
8912 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8913 	kvm_run->ready_for_interrupt_injection =
8914 		pic_in_kernel(vcpu->kvm) ||
8915 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8916 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8917 
8918 	if (is_smm(vcpu))
8919 		kvm_run->flags |= KVM_RUN_X86_SMM;
8920 }
8921 
8922 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8923 {
8924 	int max_irr, tpr;
8925 
8926 	if (!kvm_x86_ops.update_cr8_intercept)
8927 		return;
8928 
8929 	if (!lapic_in_kernel(vcpu))
8930 		return;
8931 
8932 	if (vcpu->arch.apicv_active)
8933 		return;
8934 
8935 	if (!vcpu->arch.apic->vapic_addr)
8936 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8937 	else
8938 		max_irr = -1;
8939 
8940 	if (max_irr != -1)
8941 		max_irr >>= 4;
8942 
8943 	tpr = kvm_lapic_get_cr8(vcpu);
8944 
8945 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8946 }
8947 
8948 
8949 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8950 {
8951 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8952 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
8953 		return 1;
8954 	}
8955 
8956 	return kvm_x86_ops.nested_ops->check_events(vcpu);
8957 }
8958 
8959 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8960 {
8961 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8962 		vcpu->arch.exception.error_code = false;
8963 	static_call(kvm_x86_queue_exception)(vcpu);
8964 }
8965 
8966 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8967 {
8968 	int r;
8969 	bool can_inject = true;
8970 
8971 	/* try to reinject previous events if any */
8972 
8973 	if (vcpu->arch.exception.injected) {
8974 		kvm_inject_exception(vcpu);
8975 		can_inject = false;
8976 	}
8977 	/*
8978 	 * Do not inject an NMI or interrupt if there is a pending
8979 	 * exception.  Exceptions and interrupts are recognized at
8980 	 * instruction boundaries, i.e. the start of an instruction.
8981 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8982 	 * NMIs and interrupts, i.e. traps are recognized before an
8983 	 * NMI/interrupt that's pending on the same instruction.
8984 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8985 	 * priority, but are only generated (pended) during instruction
8986 	 * execution, i.e. a pending fault-like exception means the
8987 	 * fault occurred on the *previous* instruction and must be
8988 	 * serviced prior to recognizing any new events in order to
8989 	 * fully complete the previous instruction.
8990 	 */
8991 	else if (!vcpu->arch.exception.pending) {
8992 		if (vcpu->arch.nmi_injected) {
8993 			static_call(kvm_x86_set_nmi)(vcpu);
8994 			can_inject = false;
8995 		} else if (vcpu->arch.interrupt.injected) {
8996 			static_call(kvm_x86_set_irq)(vcpu);
8997 			can_inject = false;
8998 		}
8999 	}
9000 
9001 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
9002 		     vcpu->arch.exception.pending);
9003 
9004 	/*
9005 	 * Call check_nested_events() even if we reinjected a previous event
9006 	 * in order for caller to determine if it should require immediate-exit
9007 	 * from L2 to L1 due to pending L1 events which require exit
9008 	 * from L2 to L1.
9009 	 */
9010 	if (is_guest_mode(vcpu)) {
9011 		r = kvm_check_nested_events(vcpu);
9012 		if (r < 0)
9013 			goto out;
9014 	}
9015 
9016 	/* try to inject new event if pending */
9017 	if (vcpu->arch.exception.pending) {
9018 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
9019 					vcpu->arch.exception.has_error_code,
9020 					vcpu->arch.exception.error_code);
9021 
9022 		vcpu->arch.exception.pending = false;
9023 		vcpu->arch.exception.injected = true;
9024 
9025 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9026 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9027 					     X86_EFLAGS_RF);
9028 
9029 		if (vcpu->arch.exception.nr == DB_VECTOR) {
9030 			kvm_deliver_exception_payload(vcpu);
9031 			if (vcpu->arch.dr7 & DR7_GD) {
9032 				vcpu->arch.dr7 &= ~DR7_GD;
9033 				kvm_update_dr7(vcpu);
9034 			}
9035 		}
9036 
9037 		kvm_inject_exception(vcpu);
9038 		can_inject = false;
9039 	}
9040 
9041 	/* Don't inject interrupts if the user asked to avoid doing so */
9042 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9043 		return 0;
9044 
9045 	/*
9046 	 * Finally, inject interrupt events.  If an event cannot be injected
9047 	 * due to architectural conditions (e.g. IF=0) a window-open exit
9048 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
9049 	 * and can architecturally be injected, but we cannot do it right now:
9050 	 * an interrupt could have arrived just now and we have to inject it
9051 	 * as a vmexit, or there could already an event in the queue, which is
9052 	 * indicated by can_inject.  In that case we request an immediate exit
9053 	 * in order to make progress and get back here for another iteration.
9054 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9055 	 */
9056 	if (vcpu->arch.smi_pending) {
9057 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9058 		if (r < 0)
9059 			goto out;
9060 		if (r) {
9061 			vcpu->arch.smi_pending = false;
9062 			++vcpu->arch.smi_count;
9063 			enter_smm(vcpu);
9064 			can_inject = false;
9065 		} else
9066 			static_call(kvm_x86_enable_smi_window)(vcpu);
9067 	}
9068 
9069 	if (vcpu->arch.nmi_pending) {
9070 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9071 		if (r < 0)
9072 			goto out;
9073 		if (r) {
9074 			--vcpu->arch.nmi_pending;
9075 			vcpu->arch.nmi_injected = true;
9076 			static_call(kvm_x86_set_nmi)(vcpu);
9077 			can_inject = false;
9078 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9079 		}
9080 		if (vcpu->arch.nmi_pending)
9081 			static_call(kvm_x86_enable_nmi_window)(vcpu);
9082 	}
9083 
9084 	if (kvm_cpu_has_injectable_intr(vcpu)) {
9085 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9086 		if (r < 0)
9087 			goto out;
9088 		if (r) {
9089 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9090 			static_call(kvm_x86_set_irq)(vcpu);
9091 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9092 		}
9093 		if (kvm_cpu_has_injectable_intr(vcpu))
9094 			static_call(kvm_x86_enable_irq_window)(vcpu);
9095 	}
9096 
9097 	if (is_guest_mode(vcpu) &&
9098 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
9099 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9100 		*req_immediate_exit = true;
9101 
9102 	WARN_ON(vcpu->arch.exception.pending);
9103 	return 0;
9104 
9105 out:
9106 	if (r == -EBUSY) {
9107 		*req_immediate_exit = true;
9108 		r = 0;
9109 	}
9110 	return r;
9111 }
9112 
9113 static void process_nmi(struct kvm_vcpu *vcpu)
9114 {
9115 	unsigned limit = 2;
9116 
9117 	/*
9118 	 * x86 is limited to one NMI running, and one NMI pending after it.
9119 	 * If an NMI is already in progress, limit further NMIs to just one.
9120 	 * Otherwise, allow two (and we'll inject the first one immediately).
9121 	 */
9122 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9123 		limit = 1;
9124 
9125 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9126 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9127 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9128 }
9129 
9130 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9131 {
9132 	u32 flags = 0;
9133 	flags |= seg->g       << 23;
9134 	flags |= seg->db      << 22;
9135 	flags |= seg->l       << 21;
9136 	flags |= seg->avl     << 20;
9137 	flags |= seg->present << 15;
9138 	flags |= seg->dpl     << 13;
9139 	flags |= seg->s       << 12;
9140 	flags |= seg->type    << 8;
9141 	return flags;
9142 }
9143 
9144 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9145 {
9146 	struct kvm_segment seg;
9147 	int offset;
9148 
9149 	kvm_get_segment(vcpu, &seg, n);
9150 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9151 
9152 	if (n < 3)
9153 		offset = 0x7f84 + n * 12;
9154 	else
9155 		offset = 0x7f2c + (n - 3) * 12;
9156 
9157 	put_smstate(u32, buf, offset + 8, seg.base);
9158 	put_smstate(u32, buf, offset + 4, seg.limit);
9159 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9160 }
9161 
9162 #ifdef CONFIG_X86_64
9163 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9164 {
9165 	struct kvm_segment seg;
9166 	int offset;
9167 	u16 flags;
9168 
9169 	kvm_get_segment(vcpu, &seg, n);
9170 	offset = 0x7e00 + n * 16;
9171 
9172 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9173 	put_smstate(u16, buf, offset, seg.selector);
9174 	put_smstate(u16, buf, offset + 2, flags);
9175 	put_smstate(u32, buf, offset + 4, seg.limit);
9176 	put_smstate(u64, buf, offset + 8, seg.base);
9177 }
9178 #endif
9179 
9180 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9181 {
9182 	struct desc_ptr dt;
9183 	struct kvm_segment seg;
9184 	unsigned long val;
9185 	int i;
9186 
9187 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9188 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9189 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9190 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9191 
9192 	for (i = 0; i < 8; i++)
9193 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9194 
9195 	kvm_get_dr(vcpu, 6, &val);
9196 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9197 	kvm_get_dr(vcpu, 7, &val);
9198 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9199 
9200 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9201 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9202 	put_smstate(u32, buf, 0x7f64, seg.base);
9203 	put_smstate(u32, buf, 0x7f60, seg.limit);
9204 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9205 
9206 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9207 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9208 	put_smstate(u32, buf, 0x7f80, seg.base);
9209 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9210 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9211 
9212 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9213 	put_smstate(u32, buf, 0x7f74, dt.address);
9214 	put_smstate(u32, buf, 0x7f70, dt.size);
9215 
9216 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9217 	put_smstate(u32, buf, 0x7f58, dt.address);
9218 	put_smstate(u32, buf, 0x7f54, dt.size);
9219 
9220 	for (i = 0; i < 6; i++)
9221 		enter_smm_save_seg_32(vcpu, buf, i);
9222 
9223 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9224 
9225 	/* revision id */
9226 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9227 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9228 }
9229 
9230 #ifdef CONFIG_X86_64
9231 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9232 {
9233 	struct desc_ptr dt;
9234 	struct kvm_segment seg;
9235 	unsigned long val;
9236 	int i;
9237 
9238 	for (i = 0; i < 16; i++)
9239 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9240 
9241 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9242 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9243 
9244 	kvm_get_dr(vcpu, 6, &val);
9245 	put_smstate(u64, buf, 0x7f68, val);
9246 	kvm_get_dr(vcpu, 7, &val);
9247 	put_smstate(u64, buf, 0x7f60, val);
9248 
9249 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9250 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9251 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9252 
9253 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9254 
9255 	/* revision id */
9256 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9257 
9258 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9259 
9260 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9261 	put_smstate(u16, buf, 0x7e90, seg.selector);
9262 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9263 	put_smstate(u32, buf, 0x7e94, seg.limit);
9264 	put_smstate(u64, buf, 0x7e98, seg.base);
9265 
9266 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9267 	put_smstate(u32, buf, 0x7e84, dt.size);
9268 	put_smstate(u64, buf, 0x7e88, dt.address);
9269 
9270 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9271 	put_smstate(u16, buf, 0x7e70, seg.selector);
9272 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9273 	put_smstate(u32, buf, 0x7e74, seg.limit);
9274 	put_smstate(u64, buf, 0x7e78, seg.base);
9275 
9276 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9277 	put_smstate(u32, buf, 0x7e64, dt.size);
9278 	put_smstate(u64, buf, 0x7e68, dt.address);
9279 
9280 	for (i = 0; i < 6; i++)
9281 		enter_smm_save_seg_64(vcpu, buf, i);
9282 }
9283 #endif
9284 
9285 static void enter_smm(struct kvm_vcpu *vcpu)
9286 {
9287 	struct kvm_segment cs, ds;
9288 	struct desc_ptr dt;
9289 	unsigned long cr0;
9290 	char buf[512];
9291 
9292 	memset(buf, 0, 512);
9293 #ifdef CONFIG_X86_64
9294 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9295 		enter_smm_save_state_64(vcpu, buf);
9296 	else
9297 #endif
9298 		enter_smm_save_state_32(vcpu, buf);
9299 
9300 	/*
9301 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9302 	 * state (e.g. leave guest mode) after we've saved the state into the
9303 	 * SMM state-save area.
9304 	 */
9305 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9306 
9307 	kvm_smm_changed(vcpu, true);
9308 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9309 
9310 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9311 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9312 	else
9313 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9314 
9315 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9316 	kvm_rip_write(vcpu, 0x8000);
9317 
9318 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9319 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
9320 	vcpu->arch.cr0 = cr0;
9321 
9322 	static_call(kvm_x86_set_cr4)(vcpu, 0);
9323 
9324 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
9325 	dt.address = dt.size = 0;
9326 	static_call(kvm_x86_set_idt)(vcpu, &dt);
9327 
9328 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9329 
9330 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9331 	cs.base = vcpu->arch.smbase;
9332 
9333 	ds.selector = 0;
9334 	ds.base = 0;
9335 
9336 	cs.limit    = ds.limit = 0xffffffff;
9337 	cs.type     = ds.type = 0x3;
9338 	cs.dpl      = ds.dpl = 0;
9339 	cs.db       = ds.db = 0;
9340 	cs.s        = ds.s = 1;
9341 	cs.l        = ds.l = 0;
9342 	cs.g        = ds.g = 1;
9343 	cs.avl      = ds.avl = 0;
9344 	cs.present  = ds.present = 1;
9345 	cs.unusable = ds.unusable = 0;
9346 	cs.padding  = ds.padding = 0;
9347 
9348 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9349 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9350 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9351 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9352 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9353 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9354 
9355 #ifdef CONFIG_X86_64
9356 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9357 		static_call(kvm_x86_set_efer)(vcpu, 0);
9358 #endif
9359 
9360 	kvm_update_cpuid_runtime(vcpu);
9361 	kvm_mmu_reset_context(vcpu);
9362 }
9363 
9364 static void process_smi(struct kvm_vcpu *vcpu)
9365 {
9366 	vcpu->arch.smi_pending = true;
9367 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9368 }
9369 
9370 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9371 				       unsigned long *vcpu_bitmap)
9372 {
9373 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9374 }
9375 
9376 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9377 {
9378 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9379 }
9380 
9381 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9382 {
9383 	bool activate;
9384 
9385 	if (!lapic_in_kernel(vcpu))
9386 		return;
9387 
9388 	down_read(&vcpu->kvm->arch.apicv_update_lock);
9389 
9390 	activate = kvm_apicv_activated(vcpu->kvm);
9391 	if (vcpu->arch.apicv_active == activate)
9392 		goto out;
9393 
9394 	vcpu->arch.apicv_active = activate;
9395 	kvm_apic_update_apicv(vcpu);
9396 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9397 
9398 	/*
9399 	 * When APICv gets disabled, we may still have injected interrupts
9400 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9401 	 * still active when the interrupt got accepted. Make sure
9402 	 * inject_pending_event() is called to check for that.
9403 	 */
9404 	if (!vcpu->arch.apicv_active)
9405 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9406 
9407 out:
9408 	up_read(&vcpu->kvm->arch.apicv_update_lock);
9409 }
9410 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9411 
9412 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9413 {
9414 	unsigned long old, new;
9415 
9416 	lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9417 
9418 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9419 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9420 		return;
9421 
9422 	old = new = kvm->arch.apicv_inhibit_reasons;
9423 
9424 	if (activate)
9425 		__clear_bit(bit, &new);
9426 	else
9427 		__set_bit(bit, &new);
9428 
9429 	if (!!old != !!new) {
9430 		trace_kvm_apicv_update_request(activate, bit);
9431 		/*
9432 		 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9433 		 * false positives in the sanity check WARN in svm_vcpu_run().
9434 		 * This task will wait for all vCPUs to ack the kick IRQ before
9435 		 * updating apicv_inhibit_reasons, and all other vCPUs will
9436 		 * block on acquiring apicv_update_lock so that vCPUs can't
9437 		 * redo svm_vcpu_run() without seeing the new inhibit state.
9438 		 *
9439 		 * Note, holding apicv_update_lock and taking it in the read
9440 		 * side (handling the request) also prevents other vCPUs from
9441 		 * servicing the request with a stale apicv_inhibit_reasons.
9442 		 */
9443 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9444 		kvm->arch.apicv_inhibit_reasons = new;
9445 		if (new) {
9446 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9447 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
9448 		}
9449 	} else
9450 		kvm->arch.apicv_inhibit_reasons = new;
9451 }
9452 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9453 
9454 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9455 {
9456 	down_write(&kvm->arch.apicv_update_lock);
9457 	__kvm_request_apicv_update(kvm, activate, bit);
9458 	up_write(&kvm->arch.apicv_update_lock);
9459 }
9460 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9461 
9462 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9463 {
9464 	if (!kvm_apic_present(vcpu))
9465 		return;
9466 
9467 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9468 
9469 	if (irqchip_split(vcpu->kvm))
9470 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9471 	else {
9472 		if (vcpu->arch.apicv_active)
9473 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9474 		if (ioapic_in_kernel(vcpu->kvm))
9475 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9476 	}
9477 
9478 	if (is_guest_mode(vcpu))
9479 		vcpu->arch.load_eoi_exitmap_pending = true;
9480 	else
9481 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9482 }
9483 
9484 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9485 {
9486 	u64 eoi_exit_bitmap[4];
9487 
9488 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9489 		return;
9490 
9491 	if (to_hv_vcpu(vcpu))
9492 		bitmap_or((ulong *)eoi_exit_bitmap,
9493 			  vcpu->arch.ioapic_handled_vectors,
9494 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9495 
9496 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9497 }
9498 
9499 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9500 					    unsigned long start, unsigned long end)
9501 {
9502 	unsigned long apic_address;
9503 
9504 	/*
9505 	 * The physical address of apic access page is stored in the VMCS.
9506 	 * Update it when it becomes invalid.
9507 	 */
9508 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9509 	if (start <= apic_address && apic_address < end)
9510 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9511 }
9512 
9513 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9514 {
9515 	if (!lapic_in_kernel(vcpu))
9516 		return;
9517 
9518 	if (!kvm_x86_ops.set_apic_access_page_addr)
9519 		return;
9520 
9521 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9522 }
9523 
9524 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9525 {
9526 	smp_send_reschedule(vcpu->cpu);
9527 }
9528 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9529 
9530 /*
9531  * Returns 1 to let vcpu_run() continue the guest execution loop without
9532  * exiting to the userspace.  Otherwise, the value will be returned to the
9533  * userspace.
9534  */
9535 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9536 {
9537 	int r;
9538 	bool req_int_win =
9539 		dm_request_for_irq_injection(vcpu) &&
9540 		kvm_cpu_accept_dm_intr(vcpu);
9541 	fastpath_t exit_fastpath;
9542 
9543 	bool req_immediate_exit = false;
9544 
9545 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9546 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9547 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9548 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9549 		trace_kvm_dirty_ring_exit(vcpu);
9550 		r = 0;
9551 		goto out;
9552 	}
9553 
9554 	if (kvm_request_pending(vcpu)) {
9555 		if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9556 			r = -EIO;
9557 			goto out;
9558 		}
9559 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9560 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9561 				r = 0;
9562 				goto out;
9563 			}
9564 		}
9565 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9566 			kvm_mmu_unload(vcpu);
9567 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9568 			__kvm_migrate_timers(vcpu);
9569 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9570 			kvm_update_masterclock(vcpu->kvm);
9571 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9572 			kvm_gen_kvmclock_update(vcpu);
9573 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9574 			r = kvm_guest_time_update(vcpu);
9575 			if (unlikely(r))
9576 				goto out;
9577 		}
9578 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9579 			kvm_mmu_sync_roots(vcpu);
9580 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9581 			kvm_mmu_load_pgd(vcpu);
9582 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9583 			kvm_vcpu_flush_tlb_all(vcpu);
9584 
9585 			/* Flushing all ASIDs flushes the current ASID... */
9586 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9587 		}
9588 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9589 			kvm_vcpu_flush_tlb_current(vcpu);
9590 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9591 			kvm_vcpu_flush_tlb_guest(vcpu);
9592 
9593 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9594 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9595 			r = 0;
9596 			goto out;
9597 		}
9598 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9599 			if (is_guest_mode(vcpu)) {
9600 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9601 			} else {
9602 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9603 				vcpu->mmio_needed = 0;
9604 				r = 0;
9605 				goto out;
9606 			}
9607 		}
9608 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9609 			/* Page is swapped out. Do synthetic halt */
9610 			vcpu->arch.apf.halted = true;
9611 			r = 1;
9612 			goto out;
9613 		}
9614 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9615 			record_steal_time(vcpu);
9616 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9617 			process_smi(vcpu);
9618 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9619 			process_nmi(vcpu);
9620 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9621 			kvm_pmu_handle_event(vcpu);
9622 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9623 			kvm_pmu_deliver_pmi(vcpu);
9624 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9625 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9626 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9627 				     vcpu->arch.ioapic_handled_vectors)) {
9628 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9629 				vcpu->run->eoi.vector =
9630 						vcpu->arch.pending_ioapic_eoi;
9631 				r = 0;
9632 				goto out;
9633 			}
9634 		}
9635 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9636 			vcpu_scan_ioapic(vcpu);
9637 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9638 			vcpu_load_eoi_exitmap(vcpu);
9639 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9640 			kvm_vcpu_reload_apic_access_page(vcpu);
9641 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9642 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9643 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9644 			r = 0;
9645 			goto out;
9646 		}
9647 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9648 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9649 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9650 			r = 0;
9651 			goto out;
9652 		}
9653 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9654 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9655 
9656 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9657 			vcpu->run->hyperv = hv_vcpu->exit;
9658 			r = 0;
9659 			goto out;
9660 		}
9661 
9662 		/*
9663 		 * KVM_REQ_HV_STIMER has to be processed after
9664 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9665 		 * depend on the guest clock being up-to-date
9666 		 */
9667 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9668 			kvm_hv_process_stimers(vcpu);
9669 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9670 			kvm_vcpu_update_apicv(vcpu);
9671 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9672 			kvm_check_async_pf_completion(vcpu);
9673 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9674 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9675 
9676 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9677 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9678 	}
9679 
9680 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9681 	    kvm_xen_has_interrupt(vcpu)) {
9682 		++vcpu->stat.req_event;
9683 		r = kvm_apic_accept_events(vcpu);
9684 		if (r < 0) {
9685 			r = 0;
9686 			goto out;
9687 		}
9688 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9689 			r = 1;
9690 			goto out;
9691 		}
9692 
9693 		r = inject_pending_event(vcpu, &req_immediate_exit);
9694 		if (r < 0) {
9695 			r = 0;
9696 			goto out;
9697 		}
9698 		if (req_int_win)
9699 			static_call(kvm_x86_enable_irq_window)(vcpu);
9700 
9701 		if (kvm_lapic_enabled(vcpu)) {
9702 			update_cr8_intercept(vcpu);
9703 			kvm_lapic_sync_to_vapic(vcpu);
9704 		}
9705 	}
9706 
9707 	r = kvm_mmu_reload(vcpu);
9708 	if (unlikely(r)) {
9709 		goto cancel_injection;
9710 	}
9711 
9712 	preempt_disable();
9713 
9714 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9715 
9716 	/*
9717 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9718 	 * IPI are then delayed after guest entry, which ensures that they
9719 	 * result in virtual interrupt delivery.
9720 	 */
9721 	local_irq_disable();
9722 	vcpu->mode = IN_GUEST_MODE;
9723 
9724 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9725 
9726 	/*
9727 	 * 1) We should set ->mode before checking ->requests.  Please see
9728 	 * the comment in kvm_vcpu_exiting_guest_mode().
9729 	 *
9730 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9731 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9732 	 * (see vmx_deliver_posted_interrupt).
9733 	 *
9734 	 * 3) This also orders the write to mode from any reads to the page
9735 	 * tables done while the VCPU is running.  Please see the comment
9736 	 * in kvm_flush_remote_tlbs.
9737 	 */
9738 	smp_mb__after_srcu_read_unlock();
9739 
9740 	/*
9741 	 * This handles the case where a posted interrupt was
9742 	 * notified with kvm_vcpu_kick.
9743 	 */
9744 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9745 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9746 
9747 	if (kvm_vcpu_exit_request(vcpu)) {
9748 		vcpu->mode = OUTSIDE_GUEST_MODE;
9749 		smp_wmb();
9750 		local_irq_enable();
9751 		preempt_enable();
9752 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9753 		r = 1;
9754 		goto cancel_injection;
9755 	}
9756 
9757 	if (req_immediate_exit) {
9758 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9759 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9760 	}
9761 
9762 	fpregs_assert_state_consistent();
9763 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9764 		switch_fpu_return();
9765 
9766 	if (unlikely(vcpu->arch.switch_db_regs)) {
9767 		set_debugreg(0, 7);
9768 		set_debugreg(vcpu->arch.eff_db[0], 0);
9769 		set_debugreg(vcpu->arch.eff_db[1], 1);
9770 		set_debugreg(vcpu->arch.eff_db[2], 2);
9771 		set_debugreg(vcpu->arch.eff_db[3], 3);
9772 	} else if (unlikely(hw_breakpoint_active())) {
9773 		set_debugreg(0, 7);
9774 	}
9775 
9776 	for (;;) {
9777 		/*
9778 		 * Assert that vCPU vs. VM APICv state is consistent.  An APICv
9779 		 * update must kick and wait for all vCPUs before toggling the
9780 		 * per-VM state, and responsing vCPUs must wait for the update
9781 		 * to complete before servicing KVM_REQ_APICV_UPDATE.
9782 		 */
9783 		WARN_ON_ONCE(kvm_apicv_activated(vcpu->kvm) != kvm_vcpu_apicv_active(vcpu));
9784 
9785 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9786 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9787 			break;
9788 
9789 		if (vcpu->arch.apicv_active)
9790 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9791 
9792 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9793 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9794 			break;
9795 		}
9796 	}
9797 
9798 	/*
9799 	 * Do this here before restoring debug registers on the host.  And
9800 	 * since we do this before handling the vmexit, a DR access vmexit
9801 	 * can (a) read the correct value of the debug registers, (b) set
9802 	 * KVM_DEBUGREG_WONT_EXIT again.
9803 	 */
9804 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9805 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9806 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9807 		kvm_update_dr0123(vcpu);
9808 		kvm_update_dr7(vcpu);
9809 	}
9810 
9811 	/*
9812 	 * If the guest has used debug registers, at least dr7
9813 	 * will be disabled while returning to the host.
9814 	 * If we don't have active breakpoints in the host, we don't
9815 	 * care about the messed up debug address registers. But if
9816 	 * we have some of them active, restore the old state.
9817 	 */
9818 	if (hw_breakpoint_active())
9819 		hw_breakpoint_restore();
9820 
9821 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9822 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9823 
9824 	vcpu->mode = OUTSIDE_GUEST_MODE;
9825 	smp_wmb();
9826 
9827 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9828 
9829 	/*
9830 	 * Consume any pending interrupts, including the possible source of
9831 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9832 	 * An instruction is required after local_irq_enable() to fully unblock
9833 	 * interrupts on processors that implement an interrupt shadow, the
9834 	 * stat.exits increment will do nicely.
9835 	 */
9836 	kvm_before_interrupt(vcpu);
9837 	local_irq_enable();
9838 	++vcpu->stat.exits;
9839 	local_irq_disable();
9840 	kvm_after_interrupt(vcpu);
9841 
9842 	/*
9843 	 * Wait until after servicing IRQs to account guest time so that any
9844 	 * ticks that occurred while running the guest are properly accounted
9845 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9846 	 * of accounting via context tracking, but the loss of accuracy is
9847 	 * acceptable for all known use cases.
9848 	 */
9849 	vtime_account_guest_exit();
9850 
9851 	if (lapic_in_kernel(vcpu)) {
9852 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9853 		if (delta != S64_MIN) {
9854 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9855 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9856 		}
9857 	}
9858 
9859 	local_irq_enable();
9860 	preempt_enable();
9861 
9862 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9863 
9864 	/*
9865 	 * Profile KVM exit RIPs:
9866 	 */
9867 	if (unlikely(prof_on == KVM_PROFILING)) {
9868 		unsigned long rip = kvm_rip_read(vcpu);
9869 		profile_hit(KVM_PROFILING, (void *)rip);
9870 	}
9871 
9872 	if (unlikely(vcpu->arch.tsc_always_catchup))
9873 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9874 
9875 	if (vcpu->arch.apic_attention)
9876 		kvm_lapic_sync_from_vapic(vcpu);
9877 
9878 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9879 	return r;
9880 
9881 cancel_injection:
9882 	if (req_immediate_exit)
9883 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9884 	static_call(kvm_x86_cancel_injection)(vcpu);
9885 	if (unlikely(vcpu->arch.apic_attention))
9886 		kvm_lapic_sync_from_vapic(vcpu);
9887 out:
9888 	return r;
9889 }
9890 
9891 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9892 {
9893 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9894 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9895 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9896 		kvm_vcpu_block(vcpu);
9897 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9898 
9899 		if (kvm_x86_ops.post_block)
9900 			static_call(kvm_x86_post_block)(vcpu);
9901 
9902 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9903 			return 1;
9904 	}
9905 
9906 	if (kvm_apic_accept_events(vcpu) < 0)
9907 		return 0;
9908 	switch(vcpu->arch.mp_state) {
9909 	case KVM_MP_STATE_HALTED:
9910 	case KVM_MP_STATE_AP_RESET_HOLD:
9911 		vcpu->arch.pv.pv_unhalted = false;
9912 		vcpu->arch.mp_state =
9913 			KVM_MP_STATE_RUNNABLE;
9914 		fallthrough;
9915 	case KVM_MP_STATE_RUNNABLE:
9916 		vcpu->arch.apf.halted = false;
9917 		break;
9918 	case KVM_MP_STATE_INIT_RECEIVED:
9919 		break;
9920 	default:
9921 		return -EINTR;
9922 	}
9923 	return 1;
9924 }
9925 
9926 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9927 {
9928 	if (is_guest_mode(vcpu))
9929 		kvm_check_nested_events(vcpu);
9930 
9931 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9932 		!vcpu->arch.apf.halted);
9933 }
9934 
9935 static int vcpu_run(struct kvm_vcpu *vcpu)
9936 {
9937 	int r;
9938 	struct kvm *kvm = vcpu->kvm;
9939 
9940 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9941 	vcpu->arch.l1tf_flush_l1d = true;
9942 
9943 	for (;;) {
9944 		if (kvm_vcpu_running(vcpu)) {
9945 			r = vcpu_enter_guest(vcpu);
9946 		} else {
9947 			r = vcpu_block(kvm, vcpu);
9948 		}
9949 
9950 		if (r <= 0)
9951 			break;
9952 
9953 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9954 		if (kvm_cpu_has_pending_timer(vcpu))
9955 			kvm_inject_pending_timer_irqs(vcpu);
9956 
9957 		if (dm_request_for_irq_injection(vcpu) &&
9958 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9959 			r = 0;
9960 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9961 			++vcpu->stat.request_irq_exits;
9962 			break;
9963 		}
9964 
9965 		if (__xfer_to_guest_mode_work_pending()) {
9966 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9967 			r = xfer_to_guest_mode_handle_work(vcpu);
9968 			if (r)
9969 				return r;
9970 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9971 		}
9972 	}
9973 
9974 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9975 
9976 	return r;
9977 }
9978 
9979 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9980 {
9981 	int r;
9982 
9983 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9984 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9985 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9986 	return r;
9987 }
9988 
9989 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9990 {
9991 	BUG_ON(!vcpu->arch.pio.count);
9992 
9993 	return complete_emulated_io(vcpu);
9994 }
9995 
9996 /*
9997  * Implements the following, as a state machine:
9998  *
9999  * read:
10000  *   for each fragment
10001  *     for each mmio piece in the fragment
10002  *       write gpa, len
10003  *       exit
10004  *       copy data
10005  *   execute insn
10006  *
10007  * write:
10008  *   for each fragment
10009  *     for each mmio piece in the fragment
10010  *       write gpa, len
10011  *       copy data
10012  *       exit
10013  */
10014 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10015 {
10016 	struct kvm_run *run = vcpu->run;
10017 	struct kvm_mmio_fragment *frag;
10018 	unsigned len;
10019 
10020 	BUG_ON(!vcpu->mmio_needed);
10021 
10022 	/* Complete previous fragment */
10023 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10024 	len = min(8u, frag->len);
10025 	if (!vcpu->mmio_is_write)
10026 		memcpy(frag->data, run->mmio.data, len);
10027 
10028 	if (frag->len <= 8) {
10029 		/* Switch to the next fragment. */
10030 		frag++;
10031 		vcpu->mmio_cur_fragment++;
10032 	} else {
10033 		/* Go forward to the next mmio piece. */
10034 		frag->data += len;
10035 		frag->gpa += len;
10036 		frag->len -= len;
10037 	}
10038 
10039 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10040 		vcpu->mmio_needed = 0;
10041 
10042 		/* FIXME: return into emulator if single-stepping.  */
10043 		if (vcpu->mmio_is_write)
10044 			return 1;
10045 		vcpu->mmio_read_completed = 1;
10046 		return complete_emulated_io(vcpu);
10047 	}
10048 
10049 	run->exit_reason = KVM_EXIT_MMIO;
10050 	run->mmio.phys_addr = frag->gpa;
10051 	if (vcpu->mmio_is_write)
10052 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10053 	run->mmio.len = min(8u, frag->len);
10054 	run->mmio.is_write = vcpu->mmio_is_write;
10055 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10056 	return 0;
10057 }
10058 
10059 /* Swap (qemu) user FPU context for the guest FPU context. */
10060 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10061 {
10062 	/*
10063 	 * Exclude PKRU from restore as restored separately in
10064 	 * kvm_x86_ops.run().
10065 	 */
10066 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10067 	trace_kvm_fpu(1);
10068 }
10069 
10070 /* When vcpu_run ends, restore user space FPU context. */
10071 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10072 {
10073 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10074 	++vcpu->stat.fpu_reload;
10075 	trace_kvm_fpu(0);
10076 }
10077 
10078 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10079 {
10080 	struct kvm_run *kvm_run = vcpu->run;
10081 	int r;
10082 
10083 	vcpu_load(vcpu);
10084 	kvm_sigset_activate(vcpu);
10085 	kvm_run->flags = 0;
10086 	kvm_load_guest_fpu(vcpu);
10087 
10088 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10089 		if (kvm_run->immediate_exit) {
10090 			r = -EINTR;
10091 			goto out;
10092 		}
10093 		kvm_vcpu_block(vcpu);
10094 		if (kvm_apic_accept_events(vcpu) < 0) {
10095 			r = 0;
10096 			goto out;
10097 		}
10098 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10099 		r = -EAGAIN;
10100 		if (signal_pending(current)) {
10101 			r = -EINTR;
10102 			kvm_run->exit_reason = KVM_EXIT_INTR;
10103 			++vcpu->stat.signal_exits;
10104 		}
10105 		goto out;
10106 	}
10107 
10108 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10109 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10110 		r = -EINVAL;
10111 		goto out;
10112 	}
10113 
10114 	if (kvm_run->kvm_dirty_regs) {
10115 		r = sync_regs(vcpu);
10116 		if (r != 0)
10117 			goto out;
10118 	}
10119 
10120 	/* re-sync apic's tpr */
10121 	if (!lapic_in_kernel(vcpu)) {
10122 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10123 			r = -EINVAL;
10124 			goto out;
10125 		}
10126 	}
10127 
10128 	if (unlikely(vcpu->arch.complete_userspace_io)) {
10129 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10130 		vcpu->arch.complete_userspace_io = NULL;
10131 		r = cui(vcpu);
10132 		if (r <= 0)
10133 			goto out;
10134 	} else
10135 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10136 
10137 	if (kvm_run->immediate_exit)
10138 		r = -EINTR;
10139 	else
10140 		r = vcpu_run(vcpu);
10141 
10142 out:
10143 	kvm_put_guest_fpu(vcpu);
10144 	if (kvm_run->kvm_valid_regs)
10145 		store_regs(vcpu);
10146 	post_kvm_run_save(vcpu);
10147 	kvm_sigset_deactivate(vcpu);
10148 
10149 	vcpu_put(vcpu);
10150 	return r;
10151 }
10152 
10153 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10154 {
10155 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10156 		/*
10157 		 * We are here if userspace calls get_regs() in the middle of
10158 		 * instruction emulation. Registers state needs to be copied
10159 		 * back from emulation context to vcpu. Userspace shouldn't do
10160 		 * that usually, but some bad designed PV devices (vmware
10161 		 * backdoor interface) need this to work
10162 		 */
10163 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10164 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10165 	}
10166 	regs->rax = kvm_rax_read(vcpu);
10167 	regs->rbx = kvm_rbx_read(vcpu);
10168 	regs->rcx = kvm_rcx_read(vcpu);
10169 	regs->rdx = kvm_rdx_read(vcpu);
10170 	regs->rsi = kvm_rsi_read(vcpu);
10171 	regs->rdi = kvm_rdi_read(vcpu);
10172 	regs->rsp = kvm_rsp_read(vcpu);
10173 	regs->rbp = kvm_rbp_read(vcpu);
10174 #ifdef CONFIG_X86_64
10175 	regs->r8 = kvm_r8_read(vcpu);
10176 	regs->r9 = kvm_r9_read(vcpu);
10177 	regs->r10 = kvm_r10_read(vcpu);
10178 	regs->r11 = kvm_r11_read(vcpu);
10179 	regs->r12 = kvm_r12_read(vcpu);
10180 	regs->r13 = kvm_r13_read(vcpu);
10181 	regs->r14 = kvm_r14_read(vcpu);
10182 	regs->r15 = kvm_r15_read(vcpu);
10183 #endif
10184 
10185 	regs->rip = kvm_rip_read(vcpu);
10186 	regs->rflags = kvm_get_rflags(vcpu);
10187 }
10188 
10189 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10190 {
10191 	vcpu_load(vcpu);
10192 	__get_regs(vcpu, regs);
10193 	vcpu_put(vcpu);
10194 	return 0;
10195 }
10196 
10197 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10198 {
10199 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10200 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10201 
10202 	kvm_rax_write(vcpu, regs->rax);
10203 	kvm_rbx_write(vcpu, regs->rbx);
10204 	kvm_rcx_write(vcpu, regs->rcx);
10205 	kvm_rdx_write(vcpu, regs->rdx);
10206 	kvm_rsi_write(vcpu, regs->rsi);
10207 	kvm_rdi_write(vcpu, regs->rdi);
10208 	kvm_rsp_write(vcpu, regs->rsp);
10209 	kvm_rbp_write(vcpu, regs->rbp);
10210 #ifdef CONFIG_X86_64
10211 	kvm_r8_write(vcpu, regs->r8);
10212 	kvm_r9_write(vcpu, regs->r9);
10213 	kvm_r10_write(vcpu, regs->r10);
10214 	kvm_r11_write(vcpu, regs->r11);
10215 	kvm_r12_write(vcpu, regs->r12);
10216 	kvm_r13_write(vcpu, regs->r13);
10217 	kvm_r14_write(vcpu, regs->r14);
10218 	kvm_r15_write(vcpu, regs->r15);
10219 #endif
10220 
10221 	kvm_rip_write(vcpu, regs->rip);
10222 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10223 
10224 	vcpu->arch.exception.pending = false;
10225 
10226 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10227 }
10228 
10229 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10230 {
10231 	vcpu_load(vcpu);
10232 	__set_regs(vcpu, regs);
10233 	vcpu_put(vcpu);
10234 	return 0;
10235 }
10236 
10237 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10238 {
10239 	struct kvm_segment cs;
10240 
10241 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10242 	*db = cs.db;
10243 	*l = cs.l;
10244 }
10245 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10246 
10247 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10248 {
10249 	struct desc_ptr dt;
10250 
10251 	if (vcpu->arch.guest_state_protected)
10252 		goto skip_protected_regs;
10253 
10254 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10255 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10256 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10257 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10258 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10259 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10260 
10261 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10262 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10263 
10264 	static_call(kvm_x86_get_idt)(vcpu, &dt);
10265 	sregs->idt.limit = dt.size;
10266 	sregs->idt.base = dt.address;
10267 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
10268 	sregs->gdt.limit = dt.size;
10269 	sregs->gdt.base = dt.address;
10270 
10271 	sregs->cr2 = vcpu->arch.cr2;
10272 	sregs->cr3 = kvm_read_cr3(vcpu);
10273 
10274 skip_protected_regs:
10275 	sregs->cr0 = kvm_read_cr0(vcpu);
10276 	sregs->cr4 = kvm_read_cr4(vcpu);
10277 	sregs->cr8 = kvm_get_cr8(vcpu);
10278 	sregs->efer = vcpu->arch.efer;
10279 	sregs->apic_base = kvm_get_apic_base(vcpu);
10280 }
10281 
10282 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10283 {
10284 	__get_sregs_common(vcpu, sregs);
10285 
10286 	if (vcpu->arch.guest_state_protected)
10287 		return;
10288 
10289 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10290 		set_bit(vcpu->arch.interrupt.nr,
10291 			(unsigned long *)sregs->interrupt_bitmap);
10292 }
10293 
10294 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10295 {
10296 	int i;
10297 
10298 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10299 
10300 	if (vcpu->arch.guest_state_protected)
10301 		return;
10302 
10303 	if (is_pae_paging(vcpu)) {
10304 		for (i = 0 ; i < 4 ; i++)
10305 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10306 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10307 	}
10308 }
10309 
10310 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10311 				  struct kvm_sregs *sregs)
10312 {
10313 	vcpu_load(vcpu);
10314 	__get_sregs(vcpu, sregs);
10315 	vcpu_put(vcpu);
10316 	return 0;
10317 }
10318 
10319 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10320 				    struct kvm_mp_state *mp_state)
10321 {
10322 	int r;
10323 
10324 	vcpu_load(vcpu);
10325 	if (kvm_mpx_supported())
10326 		kvm_load_guest_fpu(vcpu);
10327 
10328 	r = kvm_apic_accept_events(vcpu);
10329 	if (r < 0)
10330 		goto out;
10331 	r = 0;
10332 
10333 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10334 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10335 	    vcpu->arch.pv.pv_unhalted)
10336 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10337 	else
10338 		mp_state->mp_state = vcpu->arch.mp_state;
10339 
10340 out:
10341 	if (kvm_mpx_supported())
10342 		kvm_put_guest_fpu(vcpu);
10343 	vcpu_put(vcpu);
10344 	return r;
10345 }
10346 
10347 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10348 				    struct kvm_mp_state *mp_state)
10349 {
10350 	int ret = -EINVAL;
10351 
10352 	vcpu_load(vcpu);
10353 
10354 	if (!lapic_in_kernel(vcpu) &&
10355 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10356 		goto out;
10357 
10358 	/*
10359 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10360 	 * INIT state; latched init should be reported using
10361 	 * KVM_SET_VCPU_EVENTS, so reject it here.
10362 	 */
10363 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10364 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10365 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10366 		goto out;
10367 
10368 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10369 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10370 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10371 	} else
10372 		vcpu->arch.mp_state = mp_state->mp_state;
10373 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10374 
10375 	ret = 0;
10376 out:
10377 	vcpu_put(vcpu);
10378 	return ret;
10379 }
10380 
10381 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10382 		    int reason, bool has_error_code, u32 error_code)
10383 {
10384 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10385 	int ret;
10386 
10387 	init_emulate_ctxt(vcpu);
10388 
10389 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10390 				   has_error_code, error_code);
10391 	if (ret) {
10392 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10393 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10394 		vcpu->run->internal.ndata = 0;
10395 		return 0;
10396 	}
10397 
10398 	kvm_rip_write(vcpu, ctxt->eip);
10399 	kvm_set_rflags(vcpu, ctxt->eflags);
10400 	return 1;
10401 }
10402 EXPORT_SYMBOL_GPL(kvm_task_switch);
10403 
10404 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10405 {
10406 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10407 		/*
10408 		 * When EFER.LME and CR0.PG are set, the processor is in
10409 		 * 64-bit mode (though maybe in a 32-bit code segment).
10410 		 * CR4.PAE and EFER.LMA must be set.
10411 		 */
10412 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10413 			return false;
10414 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10415 			return false;
10416 	} else {
10417 		/*
10418 		 * Not in 64-bit mode: EFER.LMA is clear and the code
10419 		 * segment cannot be 64-bit.
10420 		 */
10421 		if (sregs->efer & EFER_LMA || sregs->cs.l)
10422 			return false;
10423 	}
10424 
10425 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
10426 }
10427 
10428 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10429 		int *mmu_reset_needed, bool update_pdptrs)
10430 {
10431 	struct msr_data apic_base_msr;
10432 	int idx;
10433 	struct desc_ptr dt;
10434 
10435 	if (!kvm_is_valid_sregs(vcpu, sregs))
10436 		return -EINVAL;
10437 
10438 	apic_base_msr.data = sregs->apic_base;
10439 	apic_base_msr.host_initiated = true;
10440 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10441 		return -EINVAL;
10442 
10443 	if (vcpu->arch.guest_state_protected)
10444 		return 0;
10445 
10446 	dt.size = sregs->idt.limit;
10447 	dt.address = sregs->idt.base;
10448 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10449 	dt.size = sregs->gdt.limit;
10450 	dt.address = sregs->gdt.base;
10451 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10452 
10453 	vcpu->arch.cr2 = sregs->cr2;
10454 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10455 	vcpu->arch.cr3 = sregs->cr3;
10456 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10457 
10458 	kvm_set_cr8(vcpu, sregs->cr8);
10459 
10460 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10461 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10462 
10463 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10464 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10465 	vcpu->arch.cr0 = sregs->cr0;
10466 
10467 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10468 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10469 
10470 	if (update_pdptrs) {
10471 		idx = srcu_read_lock(&vcpu->kvm->srcu);
10472 		if (is_pae_paging(vcpu)) {
10473 			load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10474 			*mmu_reset_needed = 1;
10475 		}
10476 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
10477 	}
10478 
10479 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10480 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10481 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10482 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10483 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10484 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10485 
10486 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10487 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10488 
10489 	update_cr8_intercept(vcpu);
10490 
10491 	/* Older userspace won't unhalt the vcpu on reset. */
10492 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10493 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10494 	    !is_protmode(vcpu))
10495 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10496 
10497 	return 0;
10498 }
10499 
10500 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10501 {
10502 	int pending_vec, max_bits;
10503 	int mmu_reset_needed = 0;
10504 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10505 
10506 	if (ret)
10507 		return ret;
10508 
10509 	if (mmu_reset_needed)
10510 		kvm_mmu_reset_context(vcpu);
10511 
10512 	max_bits = KVM_NR_INTERRUPTS;
10513 	pending_vec = find_first_bit(
10514 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10515 
10516 	if (pending_vec < max_bits) {
10517 		kvm_queue_interrupt(vcpu, pending_vec, false);
10518 		pr_debug("Set back pending irq %d\n", pending_vec);
10519 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10520 	}
10521 	return 0;
10522 }
10523 
10524 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10525 {
10526 	int mmu_reset_needed = 0;
10527 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10528 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10529 		!(sregs2->efer & EFER_LMA);
10530 	int i, ret;
10531 
10532 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10533 		return -EINVAL;
10534 
10535 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10536 		return -EINVAL;
10537 
10538 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10539 				 &mmu_reset_needed, !valid_pdptrs);
10540 	if (ret)
10541 		return ret;
10542 
10543 	if (valid_pdptrs) {
10544 		for (i = 0; i < 4 ; i++)
10545 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10546 
10547 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10548 		mmu_reset_needed = 1;
10549 		vcpu->arch.pdptrs_from_userspace = true;
10550 	}
10551 	if (mmu_reset_needed)
10552 		kvm_mmu_reset_context(vcpu);
10553 	return 0;
10554 }
10555 
10556 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10557 				  struct kvm_sregs *sregs)
10558 {
10559 	int ret;
10560 
10561 	vcpu_load(vcpu);
10562 	ret = __set_sregs(vcpu, sregs);
10563 	vcpu_put(vcpu);
10564 	return ret;
10565 }
10566 
10567 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10568 					struct kvm_guest_debug *dbg)
10569 {
10570 	unsigned long rflags;
10571 	int i, r;
10572 
10573 	if (vcpu->arch.guest_state_protected)
10574 		return -EINVAL;
10575 
10576 	vcpu_load(vcpu);
10577 
10578 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10579 		r = -EBUSY;
10580 		if (vcpu->arch.exception.pending)
10581 			goto out;
10582 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10583 			kvm_queue_exception(vcpu, DB_VECTOR);
10584 		else
10585 			kvm_queue_exception(vcpu, BP_VECTOR);
10586 	}
10587 
10588 	/*
10589 	 * Read rflags as long as potentially injected trace flags are still
10590 	 * filtered out.
10591 	 */
10592 	rflags = kvm_get_rflags(vcpu);
10593 
10594 	vcpu->guest_debug = dbg->control;
10595 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10596 		vcpu->guest_debug = 0;
10597 
10598 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10599 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10600 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10601 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10602 	} else {
10603 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10604 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10605 	}
10606 	kvm_update_dr7(vcpu);
10607 
10608 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10609 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10610 
10611 	/*
10612 	 * Trigger an rflags update that will inject or remove the trace
10613 	 * flags.
10614 	 */
10615 	kvm_set_rflags(vcpu, rflags);
10616 
10617 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10618 
10619 	r = 0;
10620 
10621 out:
10622 	vcpu_put(vcpu);
10623 	return r;
10624 }
10625 
10626 /*
10627  * Translate a guest virtual address to a guest physical address.
10628  */
10629 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10630 				    struct kvm_translation *tr)
10631 {
10632 	unsigned long vaddr = tr->linear_address;
10633 	gpa_t gpa;
10634 	int idx;
10635 
10636 	vcpu_load(vcpu);
10637 
10638 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10639 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10640 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10641 	tr->physical_address = gpa;
10642 	tr->valid = gpa != UNMAPPED_GVA;
10643 	tr->writeable = 1;
10644 	tr->usermode = 0;
10645 
10646 	vcpu_put(vcpu);
10647 	return 0;
10648 }
10649 
10650 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10651 {
10652 	struct fxregs_state *fxsave;
10653 
10654 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10655 		return 0;
10656 
10657 	vcpu_load(vcpu);
10658 
10659 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10660 	memcpy(fpu->fpr, fxsave->st_space, 128);
10661 	fpu->fcw = fxsave->cwd;
10662 	fpu->fsw = fxsave->swd;
10663 	fpu->ftwx = fxsave->twd;
10664 	fpu->last_opcode = fxsave->fop;
10665 	fpu->last_ip = fxsave->rip;
10666 	fpu->last_dp = fxsave->rdp;
10667 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10668 
10669 	vcpu_put(vcpu);
10670 	return 0;
10671 }
10672 
10673 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10674 {
10675 	struct fxregs_state *fxsave;
10676 
10677 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10678 		return 0;
10679 
10680 	vcpu_load(vcpu);
10681 
10682 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10683 
10684 	memcpy(fxsave->st_space, fpu->fpr, 128);
10685 	fxsave->cwd = fpu->fcw;
10686 	fxsave->swd = fpu->fsw;
10687 	fxsave->twd = fpu->ftwx;
10688 	fxsave->fop = fpu->last_opcode;
10689 	fxsave->rip = fpu->last_ip;
10690 	fxsave->rdp = fpu->last_dp;
10691 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10692 
10693 	vcpu_put(vcpu);
10694 	return 0;
10695 }
10696 
10697 static void store_regs(struct kvm_vcpu *vcpu)
10698 {
10699 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10700 
10701 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10702 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10703 
10704 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10705 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10706 
10707 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10708 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10709 				vcpu, &vcpu->run->s.regs.events);
10710 }
10711 
10712 static int sync_regs(struct kvm_vcpu *vcpu)
10713 {
10714 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10715 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10716 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10717 	}
10718 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10719 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10720 			return -EINVAL;
10721 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10722 	}
10723 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10724 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10725 				vcpu, &vcpu->run->s.regs.events))
10726 			return -EINVAL;
10727 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10728 	}
10729 
10730 	return 0;
10731 }
10732 
10733 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10734 {
10735 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10736 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10737 			     "guest TSC will not be reliable\n");
10738 
10739 	return 0;
10740 }
10741 
10742 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10743 {
10744 	struct page *page;
10745 	int r;
10746 
10747 	vcpu->arch.last_vmentry_cpu = -1;
10748 	vcpu->arch.regs_avail = ~0;
10749 	vcpu->arch.regs_dirty = ~0;
10750 
10751 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10752 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10753 	else
10754 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10755 
10756 	r = kvm_mmu_create(vcpu);
10757 	if (r < 0)
10758 		return r;
10759 
10760 	if (irqchip_in_kernel(vcpu->kvm)) {
10761 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10762 		if (r < 0)
10763 			goto fail_mmu_destroy;
10764 		if (kvm_apicv_activated(vcpu->kvm))
10765 			vcpu->arch.apicv_active = true;
10766 	} else
10767 		static_branch_inc(&kvm_has_noapic_vcpu);
10768 
10769 	r = -ENOMEM;
10770 
10771 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10772 	if (!page)
10773 		goto fail_free_lapic;
10774 	vcpu->arch.pio_data = page_address(page);
10775 
10776 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10777 				       GFP_KERNEL_ACCOUNT);
10778 	if (!vcpu->arch.mce_banks)
10779 		goto fail_free_pio_data;
10780 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10781 
10782 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10783 				GFP_KERNEL_ACCOUNT))
10784 		goto fail_free_mce_banks;
10785 
10786 	if (!alloc_emulate_ctxt(vcpu))
10787 		goto free_wbinvd_dirty_mask;
10788 
10789 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
10790 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10791 		goto free_emulate_ctxt;
10792 	}
10793 
10794 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10795 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10796 
10797 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10798 
10799 	kvm_async_pf_hash_reset(vcpu);
10800 	kvm_pmu_init(vcpu);
10801 
10802 	vcpu->arch.pending_external_vector = -1;
10803 	vcpu->arch.preempted_in_kernel = false;
10804 
10805 #if IS_ENABLED(CONFIG_HYPERV)
10806 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
10807 #endif
10808 
10809 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10810 	if (r)
10811 		goto free_guest_fpu;
10812 
10813 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10814 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10815 	kvm_vcpu_mtrr_init(vcpu);
10816 	vcpu_load(vcpu);
10817 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10818 	kvm_vcpu_reset(vcpu, false);
10819 	kvm_init_mmu(vcpu);
10820 	vcpu_put(vcpu);
10821 	return 0;
10822 
10823 free_guest_fpu:
10824 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10825 free_emulate_ctxt:
10826 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10827 free_wbinvd_dirty_mask:
10828 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10829 fail_free_mce_banks:
10830 	kfree(vcpu->arch.mce_banks);
10831 fail_free_pio_data:
10832 	free_page((unsigned long)vcpu->arch.pio_data);
10833 fail_free_lapic:
10834 	kvm_free_lapic(vcpu);
10835 fail_mmu_destroy:
10836 	kvm_mmu_destroy(vcpu);
10837 	return r;
10838 }
10839 
10840 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10841 {
10842 	struct kvm *kvm = vcpu->kvm;
10843 
10844 	if (mutex_lock_killable(&vcpu->mutex))
10845 		return;
10846 	vcpu_load(vcpu);
10847 	kvm_synchronize_tsc(vcpu, 0);
10848 	vcpu_put(vcpu);
10849 
10850 	/* poll control enabled by default */
10851 	vcpu->arch.msr_kvm_poll_control = 1;
10852 
10853 	mutex_unlock(&vcpu->mutex);
10854 
10855 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10856 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10857 						KVMCLOCK_SYNC_PERIOD);
10858 }
10859 
10860 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10861 {
10862 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10863 	int idx;
10864 
10865 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10866 
10867 	kvmclock_reset(vcpu);
10868 
10869 	static_call(kvm_x86_vcpu_free)(vcpu);
10870 
10871 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10872 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10873 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
10874 
10875 	kvm_hv_vcpu_uninit(vcpu);
10876 	kvm_pmu_destroy(vcpu);
10877 	kfree(vcpu->arch.mce_banks);
10878 	kvm_free_lapic(vcpu);
10879 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10880 	kvm_mmu_destroy(vcpu);
10881 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10882 	free_page((unsigned long)vcpu->arch.pio_data);
10883 	kvfree(vcpu->arch.cpuid_entries);
10884 	if (!lapic_in_kernel(vcpu))
10885 		static_branch_dec(&kvm_has_noapic_vcpu);
10886 }
10887 
10888 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10889 {
10890 	struct kvm_cpuid_entry2 *cpuid_0x1;
10891 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
10892 	unsigned long new_cr0;
10893 
10894 	/*
10895 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
10896 	 * to handle side effects.  RESET emulation hits those flows and relies
10897 	 * on emulated/virtualized registers, including those that are loaded
10898 	 * into hardware, to be zeroed at vCPU creation.  Use CRs as a sentinel
10899 	 * to detect improper or missing initialization.
10900 	 */
10901 	WARN_ON_ONCE(!init_event &&
10902 		     (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
10903 
10904 	kvm_lapic_reset(vcpu, init_event);
10905 
10906 	vcpu->arch.hflags = 0;
10907 
10908 	vcpu->arch.smi_pending = 0;
10909 	vcpu->arch.smi_count = 0;
10910 	atomic_set(&vcpu->arch.nmi_queued, 0);
10911 	vcpu->arch.nmi_pending = 0;
10912 	vcpu->arch.nmi_injected = false;
10913 	kvm_clear_interrupt_queue(vcpu);
10914 	kvm_clear_exception_queue(vcpu);
10915 
10916 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10917 	kvm_update_dr0123(vcpu);
10918 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10919 	vcpu->arch.dr7 = DR7_FIXED_1;
10920 	kvm_update_dr7(vcpu);
10921 
10922 	vcpu->arch.cr2 = 0;
10923 
10924 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10925 	vcpu->arch.apf.msr_en_val = 0;
10926 	vcpu->arch.apf.msr_int_val = 0;
10927 	vcpu->arch.st.msr_val = 0;
10928 
10929 	kvmclock_reset(vcpu);
10930 
10931 	kvm_clear_async_pf_completion_queue(vcpu);
10932 	kvm_async_pf_hash_reset(vcpu);
10933 	vcpu->arch.apf.halted = false;
10934 
10935 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
10936 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
10937 
10938 		/*
10939 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10940 		 * called with loaded FPU and does not let userspace fix the state.
10941 		 */
10942 		if (init_event)
10943 			kvm_put_guest_fpu(vcpu);
10944 
10945 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
10946 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
10947 
10948 		if (init_event)
10949 			kvm_load_guest_fpu(vcpu);
10950 	}
10951 
10952 	if (!init_event) {
10953 		kvm_pmu_reset(vcpu);
10954 		vcpu->arch.smbase = 0x30000;
10955 
10956 		vcpu->arch.msr_misc_features_enables = 0;
10957 
10958 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10959 	}
10960 
10961 	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
10962 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10963 	kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
10964 
10965 	/*
10966 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
10967 	 * if no CPUID match is found.  Note, it's impossible to get a match at
10968 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
10969 	 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
10970 	 * on RESET.  But, go through the motions in case that's ever remedied.
10971 	 */
10972 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
10973 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
10974 
10975 	vcpu->arch.ia32_xss = 0;
10976 
10977 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10978 
10979 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
10980 	kvm_rip_write(vcpu, 0xfff0);
10981 
10982 	vcpu->arch.cr3 = 0;
10983 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10984 
10985 	/*
10986 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
10987 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
10988 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
10989 	 */
10990 	new_cr0 = X86_CR0_ET;
10991 	if (init_event)
10992 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
10993 	else
10994 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
10995 
10996 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
10997 	static_call(kvm_x86_set_cr4)(vcpu, 0);
10998 	static_call(kvm_x86_set_efer)(vcpu, 0);
10999 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11000 
11001 	/*
11002 	 * Reset the MMU context if paging was enabled prior to INIT (which is
11003 	 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET).  Unlike the
11004 	 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11005 	 * checked because it is unconditionally cleared on INIT and all other
11006 	 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11007 	 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11008 	 */
11009 	if (old_cr0 & X86_CR0_PG)
11010 		kvm_mmu_reset_context(vcpu);
11011 
11012 	/*
11013 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
11014 	 * APM states the TLBs are untouched by INIT, but it also states that
11015 	 * the TLBs are flushed on "External initialization of the processor."
11016 	 * Flush the guest TLB regardless of vendor, there is no meaningful
11017 	 * benefit in relying on the guest to flush the TLB immediately after
11018 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
11019 	 * performance perspective.
11020 	 */
11021 	if (init_event)
11022 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11023 }
11024 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11025 
11026 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11027 {
11028 	struct kvm_segment cs;
11029 
11030 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11031 	cs.selector = vector << 8;
11032 	cs.base = vector << 12;
11033 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11034 	kvm_rip_write(vcpu, 0);
11035 }
11036 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11037 
11038 int kvm_arch_hardware_enable(void)
11039 {
11040 	struct kvm *kvm;
11041 	struct kvm_vcpu *vcpu;
11042 	int i;
11043 	int ret;
11044 	u64 local_tsc;
11045 	u64 max_tsc = 0;
11046 	bool stable, backwards_tsc = false;
11047 
11048 	kvm_user_return_msr_cpu_online();
11049 	ret = static_call(kvm_x86_hardware_enable)();
11050 	if (ret != 0)
11051 		return ret;
11052 
11053 	local_tsc = rdtsc();
11054 	stable = !kvm_check_tsc_unstable();
11055 	list_for_each_entry(kvm, &vm_list, vm_list) {
11056 		kvm_for_each_vcpu(i, vcpu, kvm) {
11057 			if (!stable && vcpu->cpu == smp_processor_id())
11058 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11059 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11060 				backwards_tsc = true;
11061 				if (vcpu->arch.last_host_tsc > max_tsc)
11062 					max_tsc = vcpu->arch.last_host_tsc;
11063 			}
11064 		}
11065 	}
11066 
11067 	/*
11068 	 * Sometimes, even reliable TSCs go backwards.  This happens on
11069 	 * platforms that reset TSC during suspend or hibernate actions, but
11070 	 * maintain synchronization.  We must compensate.  Fortunately, we can
11071 	 * detect that condition here, which happens early in CPU bringup,
11072 	 * before any KVM threads can be running.  Unfortunately, we can't
11073 	 * bring the TSCs fully up to date with real time, as we aren't yet far
11074 	 * enough into CPU bringup that we know how much real time has actually
11075 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11076 	 * variables that haven't been updated yet.
11077 	 *
11078 	 * So we simply find the maximum observed TSC above, then record the
11079 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
11080 	 * the adjustment will be applied.  Note that we accumulate
11081 	 * adjustments, in case multiple suspend cycles happen before some VCPU
11082 	 * gets a chance to run again.  In the event that no KVM threads get a
11083 	 * chance to run, we will miss the entire elapsed period, as we'll have
11084 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11085 	 * loose cycle time.  This isn't too big a deal, since the loss will be
11086 	 * uniform across all VCPUs (not to mention the scenario is extremely
11087 	 * unlikely). It is possible that a second hibernate recovery happens
11088 	 * much faster than a first, causing the observed TSC here to be
11089 	 * smaller; this would require additional padding adjustment, which is
11090 	 * why we set last_host_tsc to the local tsc observed here.
11091 	 *
11092 	 * N.B. - this code below runs only on platforms with reliable TSC,
11093 	 * as that is the only way backwards_tsc is set above.  Also note
11094 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11095 	 * have the same delta_cyc adjustment applied if backwards_tsc
11096 	 * is detected.  Note further, this adjustment is only done once,
11097 	 * as we reset last_host_tsc on all VCPUs to stop this from being
11098 	 * called multiple times (one for each physical CPU bringup).
11099 	 *
11100 	 * Platforms with unreliable TSCs don't have to deal with this, they
11101 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
11102 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
11103 	 * guarantee that they stay in perfect synchronization.
11104 	 */
11105 	if (backwards_tsc) {
11106 		u64 delta_cyc = max_tsc - local_tsc;
11107 		list_for_each_entry(kvm, &vm_list, vm_list) {
11108 			kvm->arch.backwards_tsc_observed = true;
11109 			kvm_for_each_vcpu(i, vcpu, kvm) {
11110 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
11111 				vcpu->arch.last_host_tsc = local_tsc;
11112 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11113 			}
11114 
11115 			/*
11116 			 * We have to disable TSC offset matching.. if you were
11117 			 * booting a VM while issuing an S4 host suspend....
11118 			 * you may have some problem.  Solving this issue is
11119 			 * left as an exercise to the reader.
11120 			 */
11121 			kvm->arch.last_tsc_nsec = 0;
11122 			kvm->arch.last_tsc_write = 0;
11123 		}
11124 
11125 	}
11126 	return 0;
11127 }
11128 
11129 void kvm_arch_hardware_disable(void)
11130 {
11131 	static_call(kvm_x86_hardware_disable)();
11132 	drop_user_return_notifiers();
11133 }
11134 
11135 int kvm_arch_hardware_setup(void *opaque)
11136 {
11137 	struct kvm_x86_init_ops *ops = opaque;
11138 	int r;
11139 
11140 	rdmsrl_safe(MSR_EFER, &host_efer);
11141 
11142 	if (boot_cpu_has(X86_FEATURE_XSAVES))
11143 		rdmsrl(MSR_IA32_XSS, host_xss);
11144 
11145 	r = ops->hardware_setup();
11146 	if (r != 0)
11147 		return r;
11148 
11149 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11150 	kvm_ops_static_call_update();
11151 
11152 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11153 		supported_xss = 0;
11154 
11155 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11156 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11157 #undef __kvm_cpu_cap_has
11158 
11159 	if (kvm_has_tsc_control) {
11160 		/*
11161 		 * Make sure the user can only configure tsc_khz values that
11162 		 * fit into a signed integer.
11163 		 * A min value is not calculated because it will always
11164 		 * be 1 on all machines.
11165 		 */
11166 		u64 max = min(0x7fffffffULL,
11167 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11168 		kvm_max_guest_tsc_khz = max;
11169 
11170 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11171 	}
11172 
11173 	kvm_init_msr_list();
11174 	return 0;
11175 }
11176 
11177 void kvm_arch_hardware_unsetup(void)
11178 {
11179 	static_call(kvm_x86_hardware_unsetup)();
11180 }
11181 
11182 int kvm_arch_check_processor_compat(void *opaque)
11183 {
11184 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11185 	struct kvm_x86_init_ops *ops = opaque;
11186 
11187 	WARN_ON(!irqs_disabled());
11188 
11189 	if (__cr4_reserved_bits(cpu_has, c) !=
11190 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11191 		return -EIO;
11192 
11193 	return ops->check_processor_compatibility();
11194 }
11195 
11196 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11197 {
11198 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11199 }
11200 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11201 
11202 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11203 {
11204 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11205 }
11206 
11207 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11208 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11209 
11210 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11211 {
11212 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11213 
11214 	vcpu->arch.l1tf_flush_l1d = true;
11215 	if (pmu->version && unlikely(pmu->event_count)) {
11216 		pmu->need_cleanup = true;
11217 		kvm_make_request(KVM_REQ_PMU, vcpu);
11218 	}
11219 	static_call(kvm_x86_sched_in)(vcpu, cpu);
11220 }
11221 
11222 void kvm_arch_free_vm(struct kvm *kvm)
11223 {
11224 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
11225 	__kvm_arch_free_vm(kvm);
11226 }
11227 
11228 
11229 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11230 {
11231 	int ret;
11232 	unsigned long flags;
11233 
11234 	if (type)
11235 		return -EINVAL;
11236 
11237 	ret = kvm_page_track_init(kvm);
11238 	if (ret)
11239 		return ret;
11240 
11241 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11242 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11243 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11244 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11245 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11246 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11247 
11248 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11249 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11250 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11251 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11252 		&kvm->arch.irq_sources_bitmap);
11253 
11254 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11255 	mutex_init(&kvm->arch.apic_map_lock);
11256 	seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11257 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11258 
11259 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11260 	pvclock_update_vm_gtod_copy(kvm);
11261 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11262 
11263 	kvm->arch.guest_can_read_msr_platform_info = true;
11264 
11265 #if IS_ENABLED(CONFIG_HYPERV)
11266 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11267 	kvm->arch.hv_root_tdp = INVALID_PAGE;
11268 #endif
11269 
11270 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11271 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11272 
11273 	kvm_apicv_init(kvm);
11274 	kvm_hv_init_vm(kvm);
11275 	kvm_mmu_init_vm(kvm);
11276 	kvm_xen_init_vm(kvm);
11277 
11278 	return static_call(kvm_x86_vm_init)(kvm);
11279 }
11280 
11281 int kvm_arch_post_init_vm(struct kvm *kvm)
11282 {
11283 	return kvm_mmu_post_init_vm(kvm);
11284 }
11285 
11286 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11287 {
11288 	vcpu_load(vcpu);
11289 	kvm_mmu_unload(vcpu);
11290 	vcpu_put(vcpu);
11291 }
11292 
11293 static void kvm_free_vcpus(struct kvm *kvm)
11294 {
11295 	unsigned int i;
11296 	struct kvm_vcpu *vcpu;
11297 
11298 	/*
11299 	 * Unpin any mmu pages first.
11300 	 */
11301 	kvm_for_each_vcpu(i, vcpu, kvm) {
11302 		kvm_clear_async_pf_completion_queue(vcpu);
11303 		kvm_unload_vcpu_mmu(vcpu);
11304 	}
11305 	kvm_for_each_vcpu(i, vcpu, kvm)
11306 		kvm_vcpu_destroy(vcpu);
11307 
11308 	mutex_lock(&kvm->lock);
11309 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11310 		kvm->vcpus[i] = NULL;
11311 
11312 	atomic_set(&kvm->online_vcpus, 0);
11313 	mutex_unlock(&kvm->lock);
11314 }
11315 
11316 void kvm_arch_sync_events(struct kvm *kvm)
11317 {
11318 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11319 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11320 	kvm_free_pit(kvm);
11321 }
11322 
11323 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
11324 
11325 /**
11326  * __x86_set_memory_region: Setup KVM internal memory slot
11327  *
11328  * @kvm: the kvm pointer to the VM.
11329  * @id: the slot ID to setup.
11330  * @gpa: the GPA to install the slot (unused when @size == 0).
11331  * @size: the size of the slot. Set to zero to uninstall a slot.
11332  *
11333  * This function helps to setup a KVM internal memory slot.  Specify
11334  * @size > 0 to install a new slot, while @size == 0 to uninstall a
11335  * slot.  The return code can be one of the following:
11336  *
11337  *   HVA:           on success (uninstall will return a bogus HVA)
11338  *   -errno:        on error
11339  *
11340  * The caller should always use IS_ERR() to check the return value
11341  * before use.  Note, the KVM internal memory slots are guaranteed to
11342  * remain valid and unchanged until the VM is destroyed, i.e., the
11343  * GPA->HVA translation will not change.  However, the HVA is a user
11344  * address, i.e. its accessibility is not guaranteed, and must be
11345  * accessed via __copy_{to,from}_user().
11346  */
11347 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11348 				      u32 size)
11349 {
11350 	int i, r;
11351 	unsigned long hva, old_npages;
11352 	struct kvm_memslots *slots = kvm_memslots(kvm);
11353 	struct kvm_memory_slot *slot;
11354 
11355 	/* Called with kvm->slots_lock held.  */
11356 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11357 		return ERR_PTR_USR(-EINVAL);
11358 
11359 	slot = id_to_memslot(slots, id);
11360 	if (size) {
11361 		if (slot && slot->npages)
11362 			return ERR_PTR_USR(-EEXIST);
11363 
11364 		/*
11365 		 * MAP_SHARED to prevent internal slot pages from being moved
11366 		 * by fork()/COW.
11367 		 */
11368 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11369 			      MAP_SHARED | MAP_ANONYMOUS, 0);
11370 		if (IS_ERR((void *)hva))
11371 			return (void __user *)hva;
11372 	} else {
11373 		if (!slot || !slot->npages)
11374 			return NULL;
11375 
11376 		old_npages = slot->npages;
11377 		hva = slot->userspace_addr;
11378 	}
11379 
11380 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11381 		struct kvm_userspace_memory_region m;
11382 
11383 		m.slot = id | (i << 16);
11384 		m.flags = 0;
11385 		m.guest_phys_addr = gpa;
11386 		m.userspace_addr = hva;
11387 		m.memory_size = size;
11388 		r = __kvm_set_memory_region(kvm, &m);
11389 		if (r < 0)
11390 			return ERR_PTR_USR(r);
11391 	}
11392 
11393 	if (!size)
11394 		vm_munmap(hva, old_npages * PAGE_SIZE);
11395 
11396 	return (void __user *)hva;
11397 }
11398 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11399 
11400 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11401 {
11402 	kvm_mmu_pre_destroy_vm(kvm);
11403 }
11404 
11405 void kvm_arch_destroy_vm(struct kvm *kvm)
11406 {
11407 	if (current->mm == kvm->mm) {
11408 		/*
11409 		 * Free memory regions allocated on behalf of userspace,
11410 		 * unless the the memory map has changed due to process exit
11411 		 * or fd copying.
11412 		 */
11413 		mutex_lock(&kvm->slots_lock);
11414 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11415 					0, 0);
11416 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11417 					0, 0);
11418 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11419 		mutex_unlock(&kvm->slots_lock);
11420 	}
11421 	static_call_cond(kvm_x86_vm_destroy)(kvm);
11422 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11423 	kvm_pic_destroy(kvm);
11424 	kvm_ioapic_destroy(kvm);
11425 	kvm_free_vcpus(kvm);
11426 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11427 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11428 	kvm_mmu_uninit_vm(kvm);
11429 	kvm_page_track_cleanup(kvm);
11430 	kvm_xen_destroy_vm(kvm);
11431 	kvm_hv_destroy_vm(kvm);
11432 }
11433 
11434 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11435 {
11436 	int i;
11437 
11438 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11439 		kvfree(slot->arch.rmap[i]);
11440 		slot->arch.rmap[i] = NULL;
11441 	}
11442 }
11443 
11444 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11445 {
11446 	int i;
11447 
11448 	memslot_rmap_free(slot);
11449 
11450 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11451 		kvfree(slot->arch.lpage_info[i - 1]);
11452 		slot->arch.lpage_info[i - 1] = NULL;
11453 	}
11454 
11455 	kvm_page_track_free_memslot(slot);
11456 }
11457 
11458 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11459 {
11460 	const int sz = sizeof(*slot->arch.rmap[0]);
11461 	int i;
11462 
11463 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11464 		int level = i + 1;
11465 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11466 
11467 		if (slot->arch.rmap[i])
11468 			continue;
11469 
11470 		slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11471 		if (!slot->arch.rmap[i]) {
11472 			memslot_rmap_free(slot);
11473 			return -ENOMEM;
11474 		}
11475 	}
11476 
11477 	return 0;
11478 }
11479 
11480 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11481 				      struct kvm_memory_slot *slot,
11482 				      unsigned long npages)
11483 {
11484 	int i, r;
11485 
11486 	/*
11487 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
11488 	 * old arrays will be freed by __kvm_set_memory_region() if installing
11489 	 * the new memslot is successful.
11490 	 */
11491 	memset(&slot->arch, 0, sizeof(slot->arch));
11492 
11493 	if (kvm_memslots_have_rmaps(kvm)) {
11494 		r = memslot_rmap_alloc(slot, npages);
11495 		if (r)
11496 			return r;
11497 	}
11498 
11499 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11500 		struct kvm_lpage_info *linfo;
11501 		unsigned long ugfn;
11502 		int lpages;
11503 		int level = i + 1;
11504 
11505 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11506 
11507 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11508 		if (!linfo)
11509 			goto out_free;
11510 
11511 		slot->arch.lpage_info[i - 1] = linfo;
11512 
11513 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11514 			linfo[0].disallow_lpage = 1;
11515 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11516 			linfo[lpages - 1].disallow_lpage = 1;
11517 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
11518 		/*
11519 		 * If the gfn and userspace address are not aligned wrt each
11520 		 * other, disable large page support for this slot.
11521 		 */
11522 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11523 			unsigned long j;
11524 
11525 			for (j = 0; j < lpages; ++j)
11526 				linfo[j].disallow_lpage = 1;
11527 		}
11528 	}
11529 
11530 	if (kvm_page_track_create_memslot(kvm, slot, npages))
11531 		goto out_free;
11532 
11533 	return 0;
11534 
11535 out_free:
11536 	memslot_rmap_free(slot);
11537 
11538 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11539 		kvfree(slot->arch.lpage_info[i - 1]);
11540 		slot->arch.lpage_info[i - 1] = NULL;
11541 	}
11542 	return -ENOMEM;
11543 }
11544 
11545 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11546 {
11547 	struct kvm_vcpu *vcpu;
11548 	int i;
11549 
11550 	/*
11551 	 * memslots->generation has been incremented.
11552 	 * mmio generation may have reached its maximum value.
11553 	 */
11554 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11555 
11556 	/* Force re-initialization of steal_time cache */
11557 	kvm_for_each_vcpu(i, vcpu, kvm)
11558 		kvm_vcpu_kick(vcpu);
11559 }
11560 
11561 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11562 				struct kvm_memory_slot *memslot,
11563 				const struct kvm_userspace_memory_region *mem,
11564 				enum kvm_mr_change change)
11565 {
11566 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11567 		return kvm_alloc_memslot_metadata(kvm, memslot,
11568 						  mem->memory_size >> PAGE_SHIFT);
11569 	return 0;
11570 }
11571 
11572 
11573 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11574 {
11575 	struct kvm_arch *ka = &kvm->arch;
11576 
11577 	if (!kvm_x86_ops.cpu_dirty_log_size)
11578 		return;
11579 
11580 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11581 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11582 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11583 
11584 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11585 }
11586 
11587 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11588 				     struct kvm_memory_slot *old,
11589 				     const struct kvm_memory_slot *new,
11590 				     enum kvm_mr_change change)
11591 {
11592 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11593 
11594 	/*
11595 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11596 	 * applies to all operations.
11597 	 */
11598 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11599 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11600 
11601 	/*
11602 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11603 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11604 	 *
11605 	 * For a memslot with dirty logging disabled:
11606 	 * CREATE:      No dirty mappings will already exist.
11607 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11608 	 *		kvm_arch_flush_shadow_memslot()
11609 	 *
11610 	 * For a memslot with dirty logging enabled:
11611 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11612 	 *		and no dirty bits to clear.
11613 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11614 	 *		kvm_arch_flush_shadow_memslot().
11615 	 */
11616 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11617 		return;
11618 
11619 	/*
11620 	 * READONLY and non-flags changes were filtered out above, and the only
11621 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11622 	 * logging isn't being toggled on or off.
11623 	 */
11624 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11625 		return;
11626 
11627 	if (!log_dirty_pages) {
11628 		/*
11629 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11630 		 * large sptes have to be split.  If live migration succeeds,
11631 		 * the guest in the source machine will be destroyed and large
11632 		 * sptes will be created in the destination.  However, if the
11633 		 * guest continues to run in the source machine (for example if
11634 		 * live migration fails), small sptes will remain around and
11635 		 * cause bad performance.
11636 		 *
11637 		 * Scan sptes if dirty logging has been stopped, dropping those
11638 		 * which can be collapsed into a single large-page spte.  Later
11639 		 * page faults will create the large-page sptes.
11640 		 */
11641 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11642 	} else {
11643 		/*
11644 		 * Initially-all-set does not require write protecting any page,
11645 		 * because they're all assumed to be dirty.
11646 		 */
11647 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11648 			return;
11649 
11650 		if (kvm_x86_ops.cpu_dirty_log_size) {
11651 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11652 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11653 		} else {
11654 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11655 		}
11656 	}
11657 }
11658 
11659 void kvm_arch_commit_memory_region(struct kvm *kvm,
11660 				const struct kvm_userspace_memory_region *mem,
11661 				struct kvm_memory_slot *old,
11662 				const struct kvm_memory_slot *new,
11663 				enum kvm_mr_change change)
11664 {
11665 	if (!kvm->arch.n_requested_mmu_pages)
11666 		kvm_mmu_change_mmu_pages(kvm,
11667 				kvm_mmu_calculate_default_mmu_pages(kvm));
11668 
11669 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
11670 
11671 	/* Free the arrays associated with the old memslot. */
11672 	if (change == KVM_MR_MOVE)
11673 		kvm_arch_free_memslot(kvm, old);
11674 }
11675 
11676 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11677 {
11678 	kvm_mmu_zap_all(kvm);
11679 }
11680 
11681 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11682 				   struct kvm_memory_slot *slot)
11683 {
11684 	kvm_page_track_flush_slot(kvm, slot);
11685 }
11686 
11687 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11688 {
11689 	return (is_guest_mode(vcpu) &&
11690 			kvm_x86_ops.guest_apic_has_interrupt &&
11691 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11692 }
11693 
11694 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11695 {
11696 	if (!list_empty_careful(&vcpu->async_pf.done))
11697 		return true;
11698 
11699 	if (kvm_apic_has_events(vcpu))
11700 		return true;
11701 
11702 	if (vcpu->arch.pv.pv_unhalted)
11703 		return true;
11704 
11705 	if (vcpu->arch.exception.pending)
11706 		return true;
11707 
11708 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11709 	    (vcpu->arch.nmi_pending &&
11710 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11711 		return true;
11712 
11713 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11714 	    (vcpu->arch.smi_pending &&
11715 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11716 		return true;
11717 
11718 	if (kvm_arch_interrupt_allowed(vcpu) &&
11719 	    (kvm_cpu_has_interrupt(vcpu) ||
11720 	    kvm_guest_apic_has_interrupt(vcpu)))
11721 		return true;
11722 
11723 	if (kvm_hv_has_stimer_pending(vcpu))
11724 		return true;
11725 
11726 	if (is_guest_mode(vcpu) &&
11727 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11728 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11729 		return true;
11730 
11731 	return false;
11732 }
11733 
11734 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11735 {
11736 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11737 }
11738 
11739 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11740 {
11741 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11742 		return true;
11743 
11744 	return false;
11745 }
11746 
11747 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11748 {
11749 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11750 		return true;
11751 
11752 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11753 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11754 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11755 		return true;
11756 
11757 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11758 }
11759 
11760 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11761 {
11762 	if (vcpu->arch.guest_state_protected)
11763 		return true;
11764 
11765 	return vcpu->arch.preempted_in_kernel;
11766 }
11767 
11768 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11769 {
11770 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11771 }
11772 
11773 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11774 {
11775 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11776 }
11777 
11778 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11779 {
11780 	/* Can't read the RIP when guest state is protected, just return 0 */
11781 	if (vcpu->arch.guest_state_protected)
11782 		return 0;
11783 
11784 	if (is_64_bit_mode(vcpu))
11785 		return kvm_rip_read(vcpu);
11786 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11787 		     kvm_rip_read(vcpu));
11788 }
11789 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11790 
11791 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11792 {
11793 	return kvm_get_linear_rip(vcpu) == linear_rip;
11794 }
11795 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11796 
11797 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11798 {
11799 	unsigned long rflags;
11800 
11801 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11802 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11803 		rflags &= ~X86_EFLAGS_TF;
11804 	return rflags;
11805 }
11806 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11807 
11808 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11809 {
11810 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11811 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11812 		rflags |= X86_EFLAGS_TF;
11813 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11814 }
11815 
11816 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11817 {
11818 	__kvm_set_rflags(vcpu, rflags);
11819 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11820 }
11821 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11822 
11823 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11824 {
11825 	int r;
11826 
11827 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11828 	      work->wakeup_all)
11829 		return;
11830 
11831 	r = kvm_mmu_reload(vcpu);
11832 	if (unlikely(r))
11833 		return;
11834 
11835 	if (!vcpu->arch.mmu->direct_map &&
11836 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11837 		return;
11838 
11839 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11840 }
11841 
11842 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11843 {
11844 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11845 
11846 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11847 }
11848 
11849 static inline u32 kvm_async_pf_next_probe(u32 key)
11850 {
11851 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11852 }
11853 
11854 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11855 {
11856 	u32 key = kvm_async_pf_hash_fn(gfn);
11857 
11858 	while (vcpu->arch.apf.gfns[key] != ~0)
11859 		key = kvm_async_pf_next_probe(key);
11860 
11861 	vcpu->arch.apf.gfns[key] = gfn;
11862 }
11863 
11864 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11865 {
11866 	int i;
11867 	u32 key = kvm_async_pf_hash_fn(gfn);
11868 
11869 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11870 		     (vcpu->arch.apf.gfns[key] != gfn &&
11871 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11872 		key = kvm_async_pf_next_probe(key);
11873 
11874 	return key;
11875 }
11876 
11877 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11878 {
11879 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11880 }
11881 
11882 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11883 {
11884 	u32 i, j, k;
11885 
11886 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11887 
11888 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11889 		return;
11890 
11891 	while (true) {
11892 		vcpu->arch.apf.gfns[i] = ~0;
11893 		do {
11894 			j = kvm_async_pf_next_probe(j);
11895 			if (vcpu->arch.apf.gfns[j] == ~0)
11896 				return;
11897 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11898 			/*
11899 			 * k lies cyclically in ]i,j]
11900 			 * |    i.k.j |
11901 			 * |....j i.k.| or  |.k..j i...|
11902 			 */
11903 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11904 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11905 		i = j;
11906 	}
11907 }
11908 
11909 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11910 {
11911 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11912 
11913 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11914 				      sizeof(reason));
11915 }
11916 
11917 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11918 {
11919 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11920 
11921 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11922 					     &token, offset, sizeof(token));
11923 }
11924 
11925 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11926 {
11927 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11928 	u32 val;
11929 
11930 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11931 					 &val, offset, sizeof(val)))
11932 		return false;
11933 
11934 	return !val;
11935 }
11936 
11937 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11938 {
11939 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11940 		return false;
11941 
11942 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11943 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11944 		return false;
11945 
11946 	return true;
11947 }
11948 
11949 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11950 {
11951 	if (unlikely(!lapic_in_kernel(vcpu) ||
11952 		     kvm_event_needs_reinjection(vcpu) ||
11953 		     vcpu->arch.exception.pending))
11954 		return false;
11955 
11956 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11957 		return false;
11958 
11959 	/*
11960 	 * If interrupts are off we cannot even use an artificial
11961 	 * halt state.
11962 	 */
11963 	return kvm_arch_interrupt_allowed(vcpu);
11964 }
11965 
11966 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11967 				     struct kvm_async_pf *work)
11968 {
11969 	struct x86_exception fault;
11970 
11971 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11972 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11973 
11974 	if (kvm_can_deliver_async_pf(vcpu) &&
11975 	    !apf_put_user_notpresent(vcpu)) {
11976 		fault.vector = PF_VECTOR;
11977 		fault.error_code_valid = true;
11978 		fault.error_code = 0;
11979 		fault.nested_page_fault = false;
11980 		fault.address = work->arch.token;
11981 		fault.async_page_fault = true;
11982 		kvm_inject_page_fault(vcpu, &fault);
11983 		return true;
11984 	} else {
11985 		/*
11986 		 * It is not possible to deliver a paravirtualized asynchronous
11987 		 * page fault, but putting the guest in an artificial halt state
11988 		 * can be beneficial nevertheless: if an interrupt arrives, we
11989 		 * can deliver it timely and perhaps the guest will schedule
11990 		 * another process.  When the instruction that triggered a page
11991 		 * fault is retried, hopefully the page will be ready in the host.
11992 		 */
11993 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11994 		return false;
11995 	}
11996 }
11997 
11998 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11999 				 struct kvm_async_pf *work)
12000 {
12001 	struct kvm_lapic_irq irq = {
12002 		.delivery_mode = APIC_DM_FIXED,
12003 		.vector = vcpu->arch.apf.vec
12004 	};
12005 
12006 	if (work->wakeup_all)
12007 		work->arch.token = ~0; /* broadcast wakeup */
12008 	else
12009 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12010 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12011 
12012 	if ((work->wakeup_all || work->notpresent_injected) &&
12013 	    kvm_pv_async_pf_enabled(vcpu) &&
12014 	    !apf_put_user_ready(vcpu, work->arch.token)) {
12015 		vcpu->arch.apf.pageready_pending = true;
12016 		kvm_apic_set_irq(vcpu, &irq, NULL);
12017 	}
12018 
12019 	vcpu->arch.apf.halted = false;
12020 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12021 }
12022 
12023 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12024 {
12025 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
12026 	if (!vcpu->arch.apf.pageready_pending)
12027 		kvm_vcpu_kick(vcpu);
12028 }
12029 
12030 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12031 {
12032 	if (!kvm_pv_async_pf_enabled(vcpu))
12033 		return true;
12034 	else
12035 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12036 }
12037 
12038 void kvm_arch_start_assignment(struct kvm *kvm)
12039 {
12040 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12041 		static_call_cond(kvm_x86_start_assignment)(kvm);
12042 }
12043 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12044 
12045 void kvm_arch_end_assignment(struct kvm *kvm)
12046 {
12047 	atomic_dec(&kvm->arch.assigned_device_count);
12048 }
12049 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12050 
12051 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12052 {
12053 	return atomic_read(&kvm->arch.assigned_device_count);
12054 }
12055 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12056 
12057 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12058 {
12059 	atomic_inc(&kvm->arch.noncoherent_dma_count);
12060 }
12061 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12062 
12063 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12064 {
12065 	atomic_dec(&kvm->arch.noncoherent_dma_count);
12066 }
12067 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12068 
12069 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12070 {
12071 	return atomic_read(&kvm->arch.noncoherent_dma_count);
12072 }
12073 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12074 
12075 bool kvm_arch_has_irq_bypass(void)
12076 {
12077 	return true;
12078 }
12079 
12080 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12081 				      struct irq_bypass_producer *prod)
12082 {
12083 	struct kvm_kernel_irqfd *irqfd =
12084 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12085 	int ret;
12086 
12087 	irqfd->producer = prod;
12088 	kvm_arch_start_assignment(irqfd->kvm);
12089 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12090 					 prod->irq, irqfd->gsi, 1);
12091 
12092 	if (ret)
12093 		kvm_arch_end_assignment(irqfd->kvm);
12094 
12095 	return ret;
12096 }
12097 
12098 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12099 				      struct irq_bypass_producer *prod)
12100 {
12101 	int ret;
12102 	struct kvm_kernel_irqfd *irqfd =
12103 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12104 
12105 	WARN_ON(irqfd->producer != prod);
12106 	irqfd->producer = NULL;
12107 
12108 	/*
12109 	 * When producer of consumer is unregistered, we change back to
12110 	 * remapped mode, so we can re-use the current implementation
12111 	 * when the irq is masked/disabled or the consumer side (KVM
12112 	 * int this case doesn't want to receive the interrupts.
12113 	*/
12114 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12115 	if (ret)
12116 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12117 		       " fails: %d\n", irqfd->consumer.token, ret);
12118 
12119 	kvm_arch_end_assignment(irqfd->kvm);
12120 }
12121 
12122 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12123 				   uint32_t guest_irq, bool set)
12124 {
12125 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12126 }
12127 
12128 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12129 				  struct kvm_kernel_irq_routing_entry *new)
12130 {
12131 	if (new->type != KVM_IRQ_ROUTING_MSI)
12132 		return true;
12133 
12134 	return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12135 }
12136 
12137 bool kvm_vector_hashing_enabled(void)
12138 {
12139 	return vector_hashing;
12140 }
12141 
12142 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12143 {
12144 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12145 }
12146 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12147 
12148 
12149 int kvm_spec_ctrl_test_value(u64 value)
12150 {
12151 	/*
12152 	 * test that setting IA32_SPEC_CTRL to given value
12153 	 * is allowed by the host processor
12154 	 */
12155 
12156 	u64 saved_value;
12157 	unsigned long flags;
12158 	int ret = 0;
12159 
12160 	local_irq_save(flags);
12161 
12162 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12163 		ret = 1;
12164 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12165 		ret = 1;
12166 	else
12167 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12168 
12169 	local_irq_restore(flags);
12170 
12171 	return ret;
12172 }
12173 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12174 
12175 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12176 {
12177 	struct x86_exception fault;
12178 	u32 access = error_code &
12179 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12180 
12181 	if (!(error_code & PFERR_PRESENT_MASK) ||
12182 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12183 		/*
12184 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12185 		 * tables probably do not match the TLB.  Just proceed
12186 		 * with the error code that the processor gave.
12187 		 */
12188 		fault.vector = PF_VECTOR;
12189 		fault.error_code_valid = true;
12190 		fault.error_code = error_code;
12191 		fault.nested_page_fault = false;
12192 		fault.address = gva;
12193 	}
12194 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12195 }
12196 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12197 
12198 /*
12199  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12200  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12201  * indicates whether exit to userspace is needed.
12202  */
12203 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12204 			      struct x86_exception *e)
12205 {
12206 	if (r == X86EMUL_PROPAGATE_FAULT) {
12207 		kvm_inject_emulated_page_fault(vcpu, e);
12208 		return 1;
12209 	}
12210 
12211 	/*
12212 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12213 	 * while handling a VMX instruction KVM could've handled the request
12214 	 * correctly by exiting to userspace and performing I/O but there
12215 	 * doesn't seem to be a real use-case behind such requests, just return
12216 	 * KVM_EXIT_INTERNAL_ERROR for now.
12217 	 */
12218 	kvm_prepare_emulation_failure_exit(vcpu);
12219 
12220 	return 0;
12221 }
12222 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12223 
12224 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12225 {
12226 	bool pcid_enabled;
12227 	struct x86_exception e;
12228 	struct {
12229 		u64 pcid;
12230 		u64 gla;
12231 	} operand;
12232 	int r;
12233 
12234 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12235 	if (r != X86EMUL_CONTINUE)
12236 		return kvm_handle_memory_failure(vcpu, r, &e);
12237 
12238 	if (operand.pcid >> 12 != 0) {
12239 		kvm_inject_gp(vcpu, 0);
12240 		return 1;
12241 	}
12242 
12243 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12244 
12245 	switch (type) {
12246 	case INVPCID_TYPE_INDIV_ADDR:
12247 		if ((!pcid_enabled && (operand.pcid != 0)) ||
12248 		    is_noncanonical_address(operand.gla, vcpu)) {
12249 			kvm_inject_gp(vcpu, 0);
12250 			return 1;
12251 		}
12252 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12253 		return kvm_skip_emulated_instruction(vcpu);
12254 
12255 	case INVPCID_TYPE_SINGLE_CTXT:
12256 		if (!pcid_enabled && (operand.pcid != 0)) {
12257 			kvm_inject_gp(vcpu, 0);
12258 			return 1;
12259 		}
12260 
12261 		kvm_invalidate_pcid(vcpu, operand.pcid);
12262 		return kvm_skip_emulated_instruction(vcpu);
12263 
12264 	case INVPCID_TYPE_ALL_NON_GLOBAL:
12265 		/*
12266 		 * Currently, KVM doesn't mark global entries in the shadow
12267 		 * page tables, so a non-global flush just degenerates to a
12268 		 * global flush. If needed, we could optimize this later by
12269 		 * keeping track of global entries in shadow page tables.
12270 		 */
12271 
12272 		fallthrough;
12273 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
12274 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12275 		return kvm_skip_emulated_instruction(vcpu);
12276 
12277 	default:
12278 		BUG(); /* We have already checked above that type <= 3 */
12279 	}
12280 }
12281 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12282 
12283 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12284 {
12285 	struct kvm_run *run = vcpu->run;
12286 	struct kvm_mmio_fragment *frag;
12287 	unsigned int len;
12288 
12289 	BUG_ON(!vcpu->mmio_needed);
12290 
12291 	/* Complete previous fragment */
12292 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12293 	len = min(8u, frag->len);
12294 	if (!vcpu->mmio_is_write)
12295 		memcpy(frag->data, run->mmio.data, len);
12296 
12297 	if (frag->len <= 8) {
12298 		/* Switch to the next fragment. */
12299 		frag++;
12300 		vcpu->mmio_cur_fragment++;
12301 	} else {
12302 		/* Go forward to the next mmio piece. */
12303 		frag->data += len;
12304 		frag->gpa += len;
12305 		frag->len -= len;
12306 	}
12307 
12308 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12309 		vcpu->mmio_needed = 0;
12310 
12311 		// VMG change, at this point, we're always done
12312 		// RIP has already been advanced
12313 		return 1;
12314 	}
12315 
12316 	// More MMIO is needed
12317 	run->mmio.phys_addr = frag->gpa;
12318 	run->mmio.len = min(8u, frag->len);
12319 	run->mmio.is_write = vcpu->mmio_is_write;
12320 	if (run->mmio.is_write)
12321 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12322 	run->exit_reason = KVM_EXIT_MMIO;
12323 
12324 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12325 
12326 	return 0;
12327 }
12328 
12329 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12330 			  void *data)
12331 {
12332 	int handled;
12333 	struct kvm_mmio_fragment *frag;
12334 
12335 	if (!data)
12336 		return -EINVAL;
12337 
12338 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12339 	if (handled == bytes)
12340 		return 1;
12341 
12342 	bytes -= handled;
12343 	gpa += handled;
12344 	data += handled;
12345 
12346 	/*TODO: Check if need to increment number of frags */
12347 	frag = vcpu->mmio_fragments;
12348 	vcpu->mmio_nr_fragments = 1;
12349 	frag->len = bytes;
12350 	frag->gpa = gpa;
12351 	frag->data = data;
12352 
12353 	vcpu->mmio_needed = 1;
12354 	vcpu->mmio_cur_fragment = 0;
12355 
12356 	vcpu->run->mmio.phys_addr = gpa;
12357 	vcpu->run->mmio.len = min(8u, frag->len);
12358 	vcpu->run->mmio.is_write = 1;
12359 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12360 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12361 
12362 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12363 
12364 	return 0;
12365 }
12366 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12367 
12368 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12369 			 void *data)
12370 {
12371 	int handled;
12372 	struct kvm_mmio_fragment *frag;
12373 
12374 	if (!data)
12375 		return -EINVAL;
12376 
12377 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12378 	if (handled == bytes)
12379 		return 1;
12380 
12381 	bytes -= handled;
12382 	gpa += handled;
12383 	data += handled;
12384 
12385 	/*TODO: Check if need to increment number of frags */
12386 	frag = vcpu->mmio_fragments;
12387 	vcpu->mmio_nr_fragments = 1;
12388 	frag->len = bytes;
12389 	frag->gpa = gpa;
12390 	frag->data = data;
12391 
12392 	vcpu->mmio_needed = 1;
12393 	vcpu->mmio_cur_fragment = 0;
12394 
12395 	vcpu->run->mmio.phys_addr = gpa;
12396 	vcpu->run->mmio.len = min(8u, frag->len);
12397 	vcpu->run->mmio.is_write = 0;
12398 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
12399 
12400 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12401 
12402 	return 0;
12403 }
12404 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12405 
12406 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12407 			   unsigned int port);
12408 
12409 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12410 {
12411 	int size = vcpu->arch.pio.size;
12412 	int port = vcpu->arch.pio.port;
12413 
12414 	vcpu->arch.pio.count = 0;
12415 	if (vcpu->arch.sev_pio_count)
12416 		return kvm_sev_es_outs(vcpu, size, port);
12417 	return 1;
12418 }
12419 
12420 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12421 			   unsigned int port)
12422 {
12423 	for (;;) {
12424 		unsigned int count =
12425 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12426 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12427 
12428 		/* memcpy done already by emulator_pio_out.  */
12429 		vcpu->arch.sev_pio_count -= count;
12430 		vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12431 		if (!ret)
12432 			break;
12433 
12434 		/* Emulation done by the kernel.  */
12435 		if (!vcpu->arch.sev_pio_count)
12436 			return 1;
12437 	}
12438 
12439 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12440 	return 0;
12441 }
12442 
12443 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12444 			  unsigned int port);
12445 
12446 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12447 {
12448 	unsigned count = vcpu->arch.pio.count;
12449 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12450 	vcpu->arch.sev_pio_count -= count;
12451 	vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12452 }
12453 
12454 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12455 {
12456 	int size = vcpu->arch.pio.size;
12457 	int port = vcpu->arch.pio.port;
12458 
12459 	advance_sev_es_emulated_ins(vcpu);
12460 	if (vcpu->arch.sev_pio_count)
12461 		return kvm_sev_es_ins(vcpu, size, port);
12462 	return 1;
12463 }
12464 
12465 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12466 			  unsigned int port)
12467 {
12468 	for (;;) {
12469 		unsigned int count =
12470 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12471 		if (!__emulator_pio_in(vcpu, size, port, count))
12472 			break;
12473 
12474 		/* Emulation done by the kernel.  */
12475 		advance_sev_es_emulated_ins(vcpu);
12476 		if (!vcpu->arch.sev_pio_count)
12477 			return 1;
12478 	}
12479 
12480 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12481 	return 0;
12482 }
12483 
12484 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12485 			 unsigned int port, void *data,  unsigned int count,
12486 			 int in)
12487 {
12488 	vcpu->arch.sev_pio_data = data;
12489 	vcpu->arch.sev_pio_count = count;
12490 	return in ? kvm_sev_es_ins(vcpu, size, port)
12491 		  : kvm_sev_es_outs(vcpu, size, port);
12492 }
12493 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12494 
12495 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12496 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12497 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12498 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12499 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12500 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12501 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12502 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12503 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12504 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12505 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12506 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12507 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12508 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12509 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12510 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12511 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12512 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12513 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12514 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12515 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12516 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12517 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12518 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12519 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12520 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12521 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12522