xref: /openbmc/linux/arch/x86/kvm/x86.c (revision dff03381)
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/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
60 #include <linux/suspend.h>
61 
62 #include <trace/events/kvm.h>
63 
64 #include <asm/debugreg.h>
65 #include <asm/msr.h>
66 #include <asm/desc.h>
67 #include <asm/mce.h>
68 #include <asm/pkru.h>
69 #include <linux/kernel_stat.h>
70 #include <asm/fpu/api.h>
71 #include <asm/fpu/xcr.h>
72 #include <asm/fpu/xstate.h>
73 #include <asm/pvclock.h>
74 #include <asm/div64.h>
75 #include <asm/irq_remapping.h>
76 #include <asm/mshyperv.h>
77 #include <asm/hypervisor.h>
78 #include <asm/tlbflush.h>
79 #include <asm/intel_pt.h>
80 #include <asm/emulate_prefix.h>
81 #include <asm/sgx.h>
82 #include <clocksource/hyperv_timer.h>
83 
84 #define CREATE_TRACE_POINTS
85 #include "trace.h"
86 
87 #define MAX_IO_MSRS 256
88 #define KVM_MAX_MCE_BANKS 32
89 
90 struct kvm_caps kvm_caps __read_mostly = {
91 	.supported_mce_cap = MCG_CTL_P | MCG_SER_P,
92 };
93 EXPORT_SYMBOL_GPL(kvm_caps);
94 
95 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
96 
97 #define emul_to_vcpu(ctxt) \
98 	((struct kvm_vcpu *)(ctxt)->vcpu)
99 
100 /* EFER defaults:
101  * - enable syscall per default because its emulated by KVM
102  * - enable LME and LMA per default on 64 bit KVM
103  */
104 #ifdef CONFIG_X86_64
105 static
106 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
107 #else
108 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
109 #endif
110 
111 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
112 
113 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
114 
115 #define KVM_CAP_PMU_VALID_MASK KVM_PMU_CAP_DISABLE
116 
117 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
118                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
119 
120 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
121 static void process_nmi(struct kvm_vcpu *vcpu);
122 static void process_smi(struct kvm_vcpu *vcpu);
123 static void enter_smm(struct kvm_vcpu *vcpu);
124 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
125 static void store_regs(struct kvm_vcpu *vcpu);
126 static int sync_regs(struct kvm_vcpu *vcpu);
127 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
128 
129 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
130 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
131 
132 struct kvm_x86_ops kvm_x86_ops __read_mostly;
133 
134 #define KVM_X86_OP(func)					     \
135 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
136 				*(((struct kvm_x86_ops *)0)->func));
137 #define KVM_X86_OP_OPTIONAL KVM_X86_OP
138 #define KVM_X86_OP_OPTIONAL_RET0 KVM_X86_OP
139 #include <asm/kvm-x86-ops.h>
140 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
141 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
142 
143 static bool __read_mostly ignore_msrs = 0;
144 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
145 
146 bool __read_mostly report_ignored_msrs = true;
147 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
148 EXPORT_SYMBOL_GPL(report_ignored_msrs);
149 
150 unsigned int min_timer_period_us = 200;
151 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
152 
153 static bool __read_mostly kvmclock_periodic_sync = true;
154 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
155 
156 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
157 static u32 __read_mostly tsc_tolerance_ppm = 250;
158 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
159 
160 /*
161  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
162  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
163  * advancement entirely.  Any other value is used as-is and disables adaptive
164  * tuning, i.e. allows privileged userspace to set an exact advancement time.
165  */
166 static int __read_mostly lapic_timer_advance_ns = -1;
167 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
168 
169 static bool __read_mostly vector_hashing = true;
170 module_param(vector_hashing, bool, S_IRUGO);
171 
172 bool __read_mostly enable_vmware_backdoor = false;
173 module_param(enable_vmware_backdoor, bool, S_IRUGO);
174 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
175 
176 static bool __read_mostly force_emulation_prefix = false;
177 module_param(force_emulation_prefix, bool, S_IRUGO);
178 
179 int __read_mostly pi_inject_timer = -1;
180 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
181 
182 /* Enable/disable PMU virtualization */
183 bool __read_mostly enable_pmu = true;
184 EXPORT_SYMBOL_GPL(enable_pmu);
185 module_param(enable_pmu, bool, 0444);
186 
187 bool __read_mostly eager_page_split = true;
188 module_param(eager_page_split, bool, 0644);
189 
190 /*
191  * Restoring the host value for MSRs that are only consumed when running in
192  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
193  * returns to userspace, i.e. the kernel can run with the guest's value.
194  */
195 #define KVM_MAX_NR_USER_RETURN_MSRS 16
196 
197 struct kvm_user_return_msrs {
198 	struct user_return_notifier urn;
199 	bool registered;
200 	struct kvm_user_return_msr_values {
201 		u64 host;
202 		u64 curr;
203 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
204 };
205 
206 u32 __read_mostly kvm_nr_uret_msrs;
207 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
208 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
209 static struct kvm_user_return_msrs __percpu *user_return_msrs;
210 
211 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
212 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
213 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
214 				| XFEATURE_MASK_PKRU | XFEATURE_MASK_XTILE)
215 
216 u64 __read_mostly host_efer;
217 EXPORT_SYMBOL_GPL(host_efer);
218 
219 bool __read_mostly allow_smaller_maxphyaddr = 0;
220 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
221 
222 bool __read_mostly enable_apicv = true;
223 EXPORT_SYMBOL_GPL(enable_apicv);
224 
225 u64 __read_mostly host_xss;
226 EXPORT_SYMBOL_GPL(host_xss);
227 
228 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
229 	KVM_GENERIC_VM_STATS(),
230 	STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
231 	STATS_DESC_COUNTER(VM, mmu_pte_write),
232 	STATS_DESC_COUNTER(VM, mmu_pde_zapped),
233 	STATS_DESC_COUNTER(VM, mmu_flooded),
234 	STATS_DESC_COUNTER(VM, mmu_recycled),
235 	STATS_DESC_COUNTER(VM, mmu_cache_miss),
236 	STATS_DESC_ICOUNTER(VM, mmu_unsync),
237 	STATS_DESC_ICOUNTER(VM, pages_4k),
238 	STATS_DESC_ICOUNTER(VM, pages_2m),
239 	STATS_DESC_ICOUNTER(VM, pages_1g),
240 	STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
241 	STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
242 	STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
243 };
244 
245 const struct kvm_stats_header kvm_vm_stats_header = {
246 	.name_size = KVM_STATS_NAME_SIZE,
247 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
248 	.id_offset = sizeof(struct kvm_stats_header),
249 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
250 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
251 		       sizeof(kvm_vm_stats_desc),
252 };
253 
254 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
255 	KVM_GENERIC_VCPU_STATS(),
256 	STATS_DESC_COUNTER(VCPU, pf_taken),
257 	STATS_DESC_COUNTER(VCPU, pf_fixed),
258 	STATS_DESC_COUNTER(VCPU, pf_emulate),
259 	STATS_DESC_COUNTER(VCPU, pf_spurious),
260 	STATS_DESC_COUNTER(VCPU, pf_fast),
261 	STATS_DESC_COUNTER(VCPU, pf_mmio_spte_created),
262 	STATS_DESC_COUNTER(VCPU, pf_guest),
263 	STATS_DESC_COUNTER(VCPU, tlb_flush),
264 	STATS_DESC_COUNTER(VCPU, invlpg),
265 	STATS_DESC_COUNTER(VCPU, exits),
266 	STATS_DESC_COUNTER(VCPU, io_exits),
267 	STATS_DESC_COUNTER(VCPU, mmio_exits),
268 	STATS_DESC_COUNTER(VCPU, signal_exits),
269 	STATS_DESC_COUNTER(VCPU, irq_window_exits),
270 	STATS_DESC_COUNTER(VCPU, nmi_window_exits),
271 	STATS_DESC_COUNTER(VCPU, l1d_flush),
272 	STATS_DESC_COUNTER(VCPU, halt_exits),
273 	STATS_DESC_COUNTER(VCPU, request_irq_exits),
274 	STATS_DESC_COUNTER(VCPU, irq_exits),
275 	STATS_DESC_COUNTER(VCPU, host_state_reload),
276 	STATS_DESC_COUNTER(VCPU, fpu_reload),
277 	STATS_DESC_COUNTER(VCPU, insn_emulation),
278 	STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
279 	STATS_DESC_COUNTER(VCPU, hypercalls),
280 	STATS_DESC_COUNTER(VCPU, irq_injections),
281 	STATS_DESC_COUNTER(VCPU, nmi_injections),
282 	STATS_DESC_COUNTER(VCPU, req_event),
283 	STATS_DESC_COUNTER(VCPU, nested_run),
284 	STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
285 	STATS_DESC_COUNTER(VCPU, directed_yield_successful),
286 	STATS_DESC_COUNTER(VCPU, preemption_reported),
287 	STATS_DESC_COUNTER(VCPU, preemption_other),
288 	STATS_DESC_IBOOLEAN(VCPU, guest_mode),
289 	STATS_DESC_COUNTER(VCPU, notify_window_exits),
290 };
291 
292 const struct kvm_stats_header kvm_vcpu_stats_header = {
293 	.name_size = KVM_STATS_NAME_SIZE,
294 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
295 	.id_offset = sizeof(struct kvm_stats_header),
296 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
297 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
298 		       sizeof(kvm_vcpu_stats_desc),
299 };
300 
301 u64 __read_mostly host_xcr0;
302 
303 static struct kmem_cache *x86_emulator_cache;
304 
305 /*
306  * When called, it means the previous get/set msr reached an invalid msr.
307  * Return true if we want to ignore/silent this failed msr access.
308  */
309 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
310 {
311 	const char *op = write ? "wrmsr" : "rdmsr";
312 
313 	if (ignore_msrs) {
314 		if (report_ignored_msrs)
315 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
316 				      op, msr, data);
317 		/* Mask the error */
318 		return true;
319 	} else {
320 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
321 				      op, msr, data);
322 		return false;
323 	}
324 }
325 
326 static struct kmem_cache *kvm_alloc_emulator_cache(void)
327 {
328 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
329 	unsigned int size = sizeof(struct x86_emulate_ctxt);
330 
331 	return kmem_cache_create_usercopy("x86_emulator", size,
332 					  __alignof__(struct x86_emulate_ctxt),
333 					  SLAB_ACCOUNT, useroffset,
334 					  size - useroffset, NULL);
335 }
336 
337 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
338 
339 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
340 {
341 	int i;
342 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
343 		vcpu->arch.apf.gfns[i] = ~0;
344 }
345 
346 static void kvm_on_user_return(struct user_return_notifier *urn)
347 {
348 	unsigned slot;
349 	struct kvm_user_return_msrs *msrs
350 		= container_of(urn, struct kvm_user_return_msrs, urn);
351 	struct kvm_user_return_msr_values *values;
352 	unsigned long flags;
353 
354 	/*
355 	 * Disabling irqs at this point since the following code could be
356 	 * interrupted and executed through kvm_arch_hardware_disable()
357 	 */
358 	local_irq_save(flags);
359 	if (msrs->registered) {
360 		msrs->registered = false;
361 		user_return_notifier_unregister(urn);
362 	}
363 	local_irq_restore(flags);
364 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
365 		values = &msrs->values[slot];
366 		if (values->host != values->curr) {
367 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
368 			values->curr = values->host;
369 		}
370 	}
371 }
372 
373 static int kvm_probe_user_return_msr(u32 msr)
374 {
375 	u64 val;
376 	int ret;
377 
378 	preempt_disable();
379 	ret = rdmsrl_safe(msr, &val);
380 	if (ret)
381 		goto out;
382 	ret = wrmsrl_safe(msr, val);
383 out:
384 	preempt_enable();
385 	return ret;
386 }
387 
388 int kvm_add_user_return_msr(u32 msr)
389 {
390 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
391 
392 	if (kvm_probe_user_return_msr(msr))
393 		return -1;
394 
395 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
396 	return kvm_nr_uret_msrs++;
397 }
398 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
399 
400 int kvm_find_user_return_msr(u32 msr)
401 {
402 	int i;
403 
404 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
405 		if (kvm_uret_msrs_list[i] == msr)
406 			return i;
407 	}
408 	return -1;
409 }
410 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
411 
412 static void kvm_user_return_msr_cpu_online(void)
413 {
414 	unsigned int cpu = smp_processor_id();
415 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
416 	u64 value;
417 	int i;
418 
419 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
420 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
421 		msrs->values[i].host = value;
422 		msrs->values[i].curr = value;
423 	}
424 }
425 
426 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
427 {
428 	unsigned int cpu = smp_processor_id();
429 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
430 	int err;
431 
432 	value = (value & mask) | (msrs->values[slot].host & ~mask);
433 	if (value == msrs->values[slot].curr)
434 		return 0;
435 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
436 	if (err)
437 		return 1;
438 
439 	msrs->values[slot].curr = value;
440 	if (!msrs->registered) {
441 		msrs->urn.on_user_return = kvm_on_user_return;
442 		user_return_notifier_register(&msrs->urn);
443 		msrs->registered = true;
444 	}
445 	return 0;
446 }
447 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
448 
449 static void drop_user_return_notifiers(void)
450 {
451 	unsigned int cpu = smp_processor_id();
452 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
453 
454 	if (msrs->registered)
455 		kvm_on_user_return(&msrs->urn);
456 }
457 
458 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
459 {
460 	return vcpu->arch.apic_base;
461 }
462 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
463 
464 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
465 {
466 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
467 }
468 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
469 
470 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
471 {
472 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
473 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
474 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
475 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
476 
477 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
478 		return 1;
479 	if (!msr_info->host_initiated) {
480 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
481 			return 1;
482 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
483 			return 1;
484 	}
485 
486 	kvm_lapic_set_base(vcpu, msr_info->data);
487 	kvm_recalculate_apic_map(vcpu->kvm);
488 	return 0;
489 }
490 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
491 
492 /*
493  * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
494  *
495  * Hardware virtualization extension instructions may fault if a reboot turns
496  * off virtualization while processes are running.  Usually after catching the
497  * fault we just panic; during reboot instead the instruction is ignored.
498  */
499 noinstr void kvm_spurious_fault(void)
500 {
501 	/* Fault while not rebooting.  We want the trace. */
502 	BUG_ON(!kvm_rebooting);
503 }
504 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
505 
506 #define EXCPT_BENIGN		0
507 #define EXCPT_CONTRIBUTORY	1
508 #define EXCPT_PF		2
509 
510 static int exception_class(int vector)
511 {
512 	switch (vector) {
513 	case PF_VECTOR:
514 		return EXCPT_PF;
515 	case DE_VECTOR:
516 	case TS_VECTOR:
517 	case NP_VECTOR:
518 	case SS_VECTOR:
519 	case GP_VECTOR:
520 		return EXCPT_CONTRIBUTORY;
521 	default:
522 		break;
523 	}
524 	return EXCPT_BENIGN;
525 }
526 
527 #define EXCPT_FAULT		0
528 #define EXCPT_TRAP		1
529 #define EXCPT_ABORT		2
530 #define EXCPT_INTERRUPT		3
531 
532 static int exception_type(int vector)
533 {
534 	unsigned int mask;
535 
536 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
537 		return EXCPT_INTERRUPT;
538 
539 	mask = 1 << vector;
540 
541 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
542 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
543 		return EXCPT_TRAP;
544 
545 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
546 		return EXCPT_ABORT;
547 
548 	/* Reserved exceptions will result in fault */
549 	return EXCPT_FAULT;
550 }
551 
552 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
553 {
554 	unsigned nr = vcpu->arch.exception.nr;
555 	bool has_payload = vcpu->arch.exception.has_payload;
556 	unsigned long payload = vcpu->arch.exception.payload;
557 
558 	if (!has_payload)
559 		return;
560 
561 	switch (nr) {
562 	case DB_VECTOR:
563 		/*
564 		 * "Certain debug exceptions may clear bit 0-3.  The
565 		 * remaining contents of the DR6 register are never
566 		 * cleared by the processor".
567 		 */
568 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
569 		/*
570 		 * In order to reflect the #DB exception payload in guest
571 		 * dr6, three components need to be considered: active low
572 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
573 		 * DR6_BS and DR6_BT)
574 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
575 		 * In the target guest dr6:
576 		 * FIXED_1 bits should always be set.
577 		 * Active low bits should be cleared if 1-setting in payload.
578 		 * Active high bits should be set if 1-setting in payload.
579 		 *
580 		 * Note, the payload is compatible with the pending debug
581 		 * exceptions/exit qualification under VMX, that active_low bits
582 		 * are active high in payload.
583 		 * So they need to be flipped for DR6.
584 		 */
585 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
586 		vcpu->arch.dr6 |= payload;
587 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
588 
589 		/*
590 		 * The #DB payload is defined as compatible with the 'pending
591 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
592 		 * defined in the 'pending debug exceptions' field (enabled
593 		 * breakpoint), it is reserved and must be zero in DR6.
594 		 */
595 		vcpu->arch.dr6 &= ~BIT(12);
596 		break;
597 	case PF_VECTOR:
598 		vcpu->arch.cr2 = payload;
599 		break;
600 	}
601 
602 	vcpu->arch.exception.has_payload = false;
603 	vcpu->arch.exception.payload = 0;
604 }
605 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
606 
607 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
608 		unsigned nr, bool has_error, u32 error_code,
609 	        bool has_payload, unsigned long payload, bool reinject)
610 {
611 	u32 prev_nr;
612 	int class1, class2;
613 
614 	kvm_make_request(KVM_REQ_EVENT, vcpu);
615 
616 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
617 	queue:
618 		if (reinject) {
619 			/*
620 			 * On vmentry, vcpu->arch.exception.pending is only
621 			 * true if an event injection was blocked by
622 			 * nested_run_pending.  In that case, however,
623 			 * vcpu_enter_guest requests an immediate exit,
624 			 * and the guest shouldn't proceed far enough to
625 			 * need reinjection.
626 			 */
627 			WARN_ON_ONCE(vcpu->arch.exception.pending);
628 			vcpu->arch.exception.injected = true;
629 			if (WARN_ON_ONCE(has_payload)) {
630 				/*
631 				 * A reinjected event has already
632 				 * delivered its payload.
633 				 */
634 				has_payload = false;
635 				payload = 0;
636 			}
637 		} else {
638 			vcpu->arch.exception.pending = true;
639 			vcpu->arch.exception.injected = false;
640 		}
641 		vcpu->arch.exception.has_error_code = has_error;
642 		vcpu->arch.exception.nr = nr;
643 		vcpu->arch.exception.error_code = error_code;
644 		vcpu->arch.exception.has_payload = has_payload;
645 		vcpu->arch.exception.payload = payload;
646 		if (!is_guest_mode(vcpu))
647 			kvm_deliver_exception_payload(vcpu);
648 		return;
649 	}
650 
651 	/* to check exception */
652 	prev_nr = vcpu->arch.exception.nr;
653 	if (prev_nr == DF_VECTOR) {
654 		/* triple fault -> shutdown */
655 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
656 		return;
657 	}
658 	class1 = exception_class(prev_nr);
659 	class2 = exception_class(nr);
660 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
661 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
662 		/*
663 		 * Generate double fault per SDM Table 5-5.  Set
664 		 * exception.pending = true so that the double fault
665 		 * can trigger a nested vmexit.
666 		 */
667 		vcpu->arch.exception.pending = true;
668 		vcpu->arch.exception.injected = false;
669 		vcpu->arch.exception.has_error_code = true;
670 		vcpu->arch.exception.nr = DF_VECTOR;
671 		vcpu->arch.exception.error_code = 0;
672 		vcpu->arch.exception.has_payload = false;
673 		vcpu->arch.exception.payload = 0;
674 	} else
675 		/* replace previous exception with a new one in a hope
676 		   that instruction re-execution will regenerate lost
677 		   exception */
678 		goto queue;
679 }
680 
681 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
682 {
683 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
684 }
685 EXPORT_SYMBOL_GPL(kvm_queue_exception);
686 
687 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
688 {
689 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
690 }
691 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
692 
693 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
694 			   unsigned long payload)
695 {
696 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
697 }
698 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
699 
700 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
701 				    u32 error_code, unsigned long payload)
702 {
703 	kvm_multiple_exception(vcpu, nr, true, error_code,
704 			       true, payload, false);
705 }
706 
707 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
708 {
709 	if (err)
710 		kvm_inject_gp(vcpu, 0);
711 	else
712 		return kvm_skip_emulated_instruction(vcpu);
713 
714 	return 1;
715 }
716 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
717 
718 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
719 {
720 	if (err) {
721 		kvm_inject_gp(vcpu, 0);
722 		return 1;
723 	}
724 
725 	return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
726 				       EMULTYPE_COMPLETE_USER_EXIT);
727 }
728 
729 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
730 {
731 	++vcpu->stat.pf_guest;
732 	vcpu->arch.exception.nested_apf =
733 		is_guest_mode(vcpu) && fault->async_page_fault;
734 	if (vcpu->arch.exception.nested_apf) {
735 		vcpu->arch.apf.nested_apf_token = fault->address;
736 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
737 	} else {
738 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
739 					fault->address);
740 	}
741 }
742 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
743 
744 /* Returns true if the page fault was immediately morphed into a VM-Exit. */
745 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
746 				    struct x86_exception *fault)
747 {
748 	struct kvm_mmu *fault_mmu;
749 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
750 
751 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
752 					       vcpu->arch.walk_mmu;
753 
754 	/*
755 	 * Invalidate the TLB entry for the faulting address, if it exists,
756 	 * else the access will fault indefinitely (and to emulate hardware).
757 	 */
758 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
759 	    !(fault->error_code & PFERR_RSVD_MASK))
760 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
761 				       fault_mmu->root.hpa);
762 
763 	/*
764 	 * A workaround for KVM's bad exception handling.  If KVM injected an
765 	 * exception into L2, and L2 encountered a #PF while vectoring the
766 	 * injected exception, manually check to see if L1 wants to intercept
767 	 * #PF, otherwise queuing the #PF will lead to #DF or a lost exception.
768 	 * In all other cases, defer the check to nested_ops->check_events(),
769 	 * which will correctly handle priority (this does not).  Note, other
770 	 * exceptions, e.g. #GP, are theoretically affected, #PF is simply the
771 	 * most problematic, e.g. when L0 and L1 are both intercepting #PF for
772 	 * shadow paging.
773 	 *
774 	 * TODO: Rewrite exception handling to track injected and pending
775 	 *       (VM-Exit) exceptions separately.
776 	 */
777 	if (unlikely(vcpu->arch.exception.injected && is_guest_mode(vcpu)) &&
778 	    kvm_x86_ops.nested_ops->handle_page_fault_workaround(vcpu, fault))
779 		return true;
780 
781 	fault_mmu->inject_page_fault(vcpu, fault);
782 	return false;
783 }
784 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
785 
786 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
787 {
788 	atomic_inc(&vcpu->arch.nmi_queued);
789 	kvm_make_request(KVM_REQ_NMI, vcpu);
790 }
791 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
792 
793 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
794 {
795 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
796 }
797 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
798 
799 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
800 {
801 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
802 }
803 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
804 
805 /*
806  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
807  * a #GP and return false.
808  */
809 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
810 {
811 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
812 		return true;
813 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
814 	return false;
815 }
816 EXPORT_SYMBOL_GPL(kvm_require_cpl);
817 
818 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
819 {
820 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
821 		return true;
822 
823 	kvm_queue_exception(vcpu, UD_VECTOR);
824 	return false;
825 }
826 EXPORT_SYMBOL_GPL(kvm_require_dr);
827 
828 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
829 {
830 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
831 }
832 
833 /*
834  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
835  */
836 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
837 {
838 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
839 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
840 	gpa_t real_gpa;
841 	int i;
842 	int ret;
843 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
844 
845 	/*
846 	 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
847 	 * to an L1 GPA.
848 	 */
849 	real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
850 				     PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
851 	if (real_gpa == INVALID_GPA)
852 		return 0;
853 
854 	/* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
855 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
856 				       cr3 & GENMASK(11, 5), sizeof(pdpte));
857 	if (ret < 0)
858 		return 0;
859 
860 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
861 		if ((pdpte[i] & PT_PRESENT_MASK) &&
862 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
863 			return 0;
864 		}
865 	}
866 
867 	/*
868 	 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
869 	 * Shadow page roots need to be reconstructed instead.
870 	 */
871 	if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
872 		kvm_mmu_free_roots(vcpu->kvm, mmu, KVM_MMU_ROOT_CURRENT);
873 
874 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
875 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
876 	kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
877 	vcpu->arch.pdptrs_from_userspace = false;
878 
879 	return 1;
880 }
881 EXPORT_SYMBOL_GPL(load_pdptrs);
882 
883 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
884 {
885 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
886 		kvm_clear_async_pf_completion_queue(vcpu);
887 		kvm_async_pf_hash_reset(vcpu);
888 
889 		/*
890 		 * Clearing CR0.PG is defined to flush the TLB from the guest's
891 		 * perspective.
892 		 */
893 		if (!(cr0 & X86_CR0_PG))
894 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
895 	}
896 
897 	if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
898 		kvm_mmu_reset_context(vcpu);
899 
900 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
901 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
902 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
903 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
904 }
905 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
906 
907 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
908 {
909 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
910 
911 	cr0 |= X86_CR0_ET;
912 
913 #ifdef CONFIG_X86_64
914 	if (cr0 & 0xffffffff00000000UL)
915 		return 1;
916 #endif
917 
918 	cr0 &= ~CR0_RESERVED_BITS;
919 
920 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
921 		return 1;
922 
923 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
924 		return 1;
925 
926 #ifdef CONFIG_X86_64
927 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
928 	    (cr0 & X86_CR0_PG)) {
929 		int cs_db, cs_l;
930 
931 		if (!is_pae(vcpu))
932 			return 1;
933 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
934 		if (cs_l)
935 			return 1;
936 	}
937 #endif
938 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
939 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
940 	    !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
941 		return 1;
942 
943 	if (!(cr0 & X86_CR0_PG) &&
944 	    (is_64_bit_mode(vcpu) || kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)))
945 		return 1;
946 
947 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
948 
949 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
950 
951 	return 0;
952 }
953 EXPORT_SYMBOL_GPL(kvm_set_cr0);
954 
955 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
956 {
957 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
958 }
959 EXPORT_SYMBOL_GPL(kvm_lmsw);
960 
961 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
962 {
963 	if (vcpu->arch.guest_state_protected)
964 		return;
965 
966 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
967 
968 		if (vcpu->arch.xcr0 != host_xcr0)
969 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
970 
971 		if (vcpu->arch.xsaves_enabled &&
972 		    vcpu->arch.ia32_xss != host_xss)
973 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
974 	}
975 
976 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
977 	if (static_cpu_has(X86_FEATURE_PKU) &&
978 	    vcpu->arch.pkru != vcpu->arch.host_pkru &&
979 	    ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
980 	     kvm_read_cr4_bits(vcpu, X86_CR4_PKE)))
981 		write_pkru(vcpu->arch.pkru);
982 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
983 }
984 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
985 
986 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
987 {
988 	if (vcpu->arch.guest_state_protected)
989 		return;
990 
991 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
992 	if (static_cpu_has(X86_FEATURE_PKU) &&
993 	    ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
994 	     kvm_read_cr4_bits(vcpu, X86_CR4_PKE))) {
995 		vcpu->arch.pkru = rdpkru();
996 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
997 			write_pkru(vcpu->arch.host_pkru);
998 	}
999 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
1000 
1001 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
1002 
1003 		if (vcpu->arch.xcr0 != host_xcr0)
1004 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
1005 
1006 		if (vcpu->arch.xsaves_enabled &&
1007 		    vcpu->arch.ia32_xss != host_xss)
1008 			wrmsrl(MSR_IA32_XSS, host_xss);
1009 	}
1010 
1011 }
1012 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
1013 
1014 static inline u64 kvm_guest_supported_xcr0(struct kvm_vcpu *vcpu)
1015 {
1016 	return vcpu->arch.guest_fpu.fpstate->user_xfeatures;
1017 }
1018 
1019 #ifdef CONFIG_X86_64
1020 static inline u64 kvm_guest_supported_xfd(struct kvm_vcpu *vcpu)
1021 {
1022 	return kvm_guest_supported_xcr0(vcpu) & XFEATURE_MASK_USER_DYNAMIC;
1023 }
1024 #endif
1025 
1026 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
1027 {
1028 	u64 xcr0 = xcr;
1029 	u64 old_xcr0 = vcpu->arch.xcr0;
1030 	u64 valid_bits;
1031 
1032 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
1033 	if (index != XCR_XFEATURE_ENABLED_MASK)
1034 		return 1;
1035 	if (!(xcr0 & XFEATURE_MASK_FP))
1036 		return 1;
1037 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
1038 		return 1;
1039 
1040 	/*
1041 	 * Do not allow the guest to set bits that we do not support
1042 	 * saving.  However, xcr0 bit 0 is always set, even if the
1043 	 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
1044 	 */
1045 	valid_bits = kvm_guest_supported_xcr0(vcpu) | XFEATURE_MASK_FP;
1046 	if (xcr0 & ~valid_bits)
1047 		return 1;
1048 
1049 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1050 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1051 		return 1;
1052 
1053 	if (xcr0 & XFEATURE_MASK_AVX512) {
1054 		if (!(xcr0 & XFEATURE_MASK_YMM))
1055 			return 1;
1056 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1057 			return 1;
1058 	}
1059 
1060 	if ((xcr0 & XFEATURE_MASK_XTILE) &&
1061 	    ((xcr0 & XFEATURE_MASK_XTILE) != XFEATURE_MASK_XTILE))
1062 		return 1;
1063 
1064 	vcpu->arch.xcr0 = xcr0;
1065 
1066 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1067 		kvm_update_cpuid_runtime(vcpu);
1068 	return 0;
1069 }
1070 
1071 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1072 {
1073 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1074 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1075 		kvm_inject_gp(vcpu, 0);
1076 		return 1;
1077 	}
1078 
1079 	return kvm_skip_emulated_instruction(vcpu);
1080 }
1081 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1082 
1083 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1084 {
1085 	if (cr4 & cr4_reserved_bits)
1086 		return false;
1087 
1088 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1089 		return false;
1090 
1091 	return true;
1092 }
1093 EXPORT_SYMBOL_GPL(__kvm_is_valid_cr4);
1094 
1095 static bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1096 {
1097 	return __kvm_is_valid_cr4(vcpu, cr4) &&
1098 	       static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1099 }
1100 
1101 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1102 {
1103 	if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1104 		kvm_mmu_reset_context(vcpu);
1105 
1106 	/*
1107 	 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1108 	 * according to the SDM; however, stale prev_roots could be reused
1109 	 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1110 	 * free them all.  This is *not* a superset of KVM_REQ_TLB_FLUSH_GUEST
1111 	 * or KVM_REQ_TLB_FLUSH_CURRENT, because the hardware TLB is not flushed,
1112 	 * so fall through.
1113 	 */
1114 	if (!tdp_enabled &&
1115 	    (cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE))
1116 		kvm_mmu_unload(vcpu);
1117 
1118 	/*
1119 	 * The TLB has to be flushed for all PCIDs if any of the following
1120 	 * (architecturally required) changes happen:
1121 	 * - CR4.PCIDE is changed from 1 to 0
1122 	 * - CR4.PGE is toggled
1123 	 *
1124 	 * This is a superset of KVM_REQ_TLB_FLUSH_CURRENT.
1125 	 */
1126 	if (((cr4 ^ old_cr4) & X86_CR4_PGE) ||
1127 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1128 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1129 
1130 	/*
1131 	 * The TLB has to be flushed for the current PCID if any of the
1132 	 * following (architecturally required) changes happen:
1133 	 * - CR4.SMEP is changed from 0 to 1
1134 	 * - CR4.PAE is toggled
1135 	 */
1136 	else if (((cr4 ^ old_cr4) & X86_CR4_PAE) ||
1137 		 ((cr4 & X86_CR4_SMEP) && !(old_cr4 & X86_CR4_SMEP)))
1138 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1139 
1140 }
1141 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1142 
1143 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1144 {
1145 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1146 
1147 	if (!kvm_is_valid_cr4(vcpu, cr4))
1148 		return 1;
1149 
1150 	if (is_long_mode(vcpu)) {
1151 		if (!(cr4 & X86_CR4_PAE))
1152 			return 1;
1153 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1154 			return 1;
1155 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1156 		   && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1157 		   && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1158 		return 1;
1159 
1160 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1161 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1162 			return 1;
1163 
1164 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1165 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1166 			return 1;
1167 	}
1168 
1169 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1170 
1171 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1172 
1173 	return 0;
1174 }
1175 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1176 
1177 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1178 {
1179 	struct kvm_mmu *mmu = vcpu->arch.mmu;
1180 	unsigned long roots_to_free = 0;
1181 	int i;
1182 
1183 	/*
1184 	 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1185 	 * this is reachable when running EPT=1 and unrestricted_guest=0,  and
1186 	 * also via the emulator.  KVM's TDP page tables are not in the scope of
1187 	 * the invalidation, but the guest's TLB entries need to be flushed as
1188 	 * the CPU may have cached entries in its TLB for the target PCID.
1189 	 */
1190 	if (unlikely(tdp_enabled)) {
1191 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1192 		return;
1193 	}
1194 
1195 	/*
1196 	 * If neither the current CR3 nor any of the prev_roots use the given
1197 	 * PCID, then nothing needs to be done here because a resync will
1198 	 * happen anyway before switching to any other CR3.
1199 	 */
1200 	if (kvm_get_active_pcid(vcpu) == pcid) {
1201 		kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1202 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1203 	}
1204 
1205 	/*
1206 	 * If PCID is disabled, there is no need to free prev_roots even if the
1207 	 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1208 	 * with PCIDE=0.
1209 	 */
1210 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1211 		return;
1212 
1213 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1214 		if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1215 			roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1216 
1217 	kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
1218 }
1219 
1220 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1221 {
1222 	bool skip_tlb_flush = false;
1223 	unsigned long pcid = 0;
1224 #ifdef CONFIG_X86_64
1225 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1226 
1227 	if (pcid_enabled) {
1228 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1229 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1230 		pcid = cr3 & X86_CR3_PCID_MASK;
1231 	}
1232 #endif
1233 
1234 	/* PDPTRs are always reloaded for PAE paging. */
1235 	if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1236 		goto handle_tlb_flush;
1237 
1238 	/*
1239 	 * Do not condition the GPA check on long mode, this helper is used to
1240 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1241 	 * the current vCPU mode is accurate.
1242 	 */
1243 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1244 		return 1;
1245 
1246 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1247 		return 1;
1248 
1249 	if (cr3 != kvm_read_cr3(vcpu))
1250 		kvm_mmu_new_pgd(vcpu, cr3);
1251 
1252 	vcpu->arch.cr3 = cr3;
1253 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1254 	/* Do not call post_set_cr3, we do not get here for confidential guests.  */
1255 
1256 handle_tlb_flush:
1257 	/*
1258 	 * A load of CR3 that flushes the TLB flushes only the current PCID,
1259 	 * even if PCID is disabled, in which case PCID=0 is flushed.  It's a
1260 	 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1261 	 * and it's impossible to use a non-zero PCID when PCID is disabled,
1262 	 * i.e. only PCID=0 can be relevant.
1263 	 */
1264 	if (!skip_tlb_flush)
1265 		kvm_invalidate_pcid(vcpu, pcid);
1266 
1267 	return 0;
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1270 
1271 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1272 {
1273 	if (cr8 & CR8_RESERVED_BITS)
1274 		return 1;
1275 	if (lapic_in_kernel(vcpu))
1276 		kvm_lapic_set_tpr(vcpu, cr8);
1277 	else
1278 		vcpu->arch.cr8 = cr8;
1279 	return 0;
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1282 
1283 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1284 {
1285 	if (lapic_in_kernel(vcpu))
1286 		return kvm_lapic_get_cr8(vcpu);
1287 	else
1288 		return vcpu->arch.cr8;
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1291 
1292 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1293 {
1294 	int i;
1295 
1296 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1297 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1298 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1299 	}
1300 }
1301 
1302 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1303 {
1304 	unsigned long dr7;
1305 
1306 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1307 		dr7 = vcpu->arch.guest_debug_dr7;
1308 	else
1309 		dr7 = vcpu->arch.dr7;
1310 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1311 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1312 	if (dr7 & DR7_BP_EN_MASK)
1313 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1314 }
1315 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1316 
1317 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1318 {
1319 	u64 fixed = DR6_FIXED_1;
1320 
1321 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1322 		fixed |= DR6_RTM;
1323 
1324 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1325 		fixed |= DR6_BUS_LOCK;
1326 	return fixed;
1327 }
1328 
1329 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1330 {
1331 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1332 
1333 	switch (dr) {
1334 	case 0 ... 3:
1335 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1336 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1337 			vcpu->arch.eff_db[dr] = val;
1338 		break;
1339 	case 4:
1340 	case 6:
1341 		if (!kvm_dr6_valid(val))
1342 			return 1; /* #GP */
1343 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1344 		break;
1345 	case 5:
1346 	default: /* 7 */
1347 		if (!kvm_dr7_valid(val))
1348 			return 1; /* #GP */
1349 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1350 		kvm_update_dr7(vcpu);
1351 		break;
1352 	}
1353 
1354 	return 0;
1355 }
1356 EXPORT_SYMBOL_GPL(kvm_set_dr);
1357 
1358 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1359 {
1360 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1361 
1362 	switch (dr) {
1363 	case 0 ... 3:
1364 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1365 		break;
1366 	case 4:
1367 	case 6:
1368 		*val = vcpu->arch.dr6;
1369 		break;
1370 	case 5:
1371 	default: /* 7 */
1372 		*val = vcpu->arch.dr7;
1373 		break;
1374 	}
1375 }
1376 EXPORT_SYMBOL_GPL(kvm_get_dr);
1377 
1378 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1379 {
1380 	u32 ecx = kvm_rcx_read(vcpu);
1381 	u64 data;
1382 
1383 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1384 		kvm_inject_gp(vcpu, 0);
1385 		return 1;
1386 	}
1387 
1388 	kvm_rax_write(vcpu, (u32)data);
1389 	kvm_rdx_write(vcpu, data >> 32);
1390 	return kvm_skip_emulated_instruction(vcpu);
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1393 
1394 /*
1395  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1396  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1397  *
1398  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1399  * extract the supported MSRs from the related const lists.
1400  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1401  * capabilities of the host cpu. This capabilities test skips MSRs that are
1402  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1403  * may depend on host virtualization features rather than host cpu features.
1404  */
1405 
1406 static const u32 msrs_to_save_all[] = {
1407 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1408 	MSR_STAR,
1409 #ifdef CONFIG_X86_64
1410 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1411 #endif
1412 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1413 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1414 	MSR_IA32_SPEC_CTRL,
1415 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1416 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1417 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1418 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1419 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1420 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1421 	MSR_IA32_UMWAIT_CONTROL,
1422 
1423 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1424 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1425 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1426 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1427 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1428 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1429 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1430 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1431 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1432 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1433 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1434 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1435 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1436 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1437 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1438 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1439 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1440 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1441 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1442 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1443 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1444 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1445 	MSR_IA32_PEBS_ENABLE, MSR_IA32_DS_AREA, MSR_PEBS_DATA_CFG,
1446 
1447 	MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1448 	MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1449 	MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1450 	MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1451 	MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1452 	MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1453 	MSR_IA32_XFD, MSR_IA32_XFD_ERR,
1454 };
1455 
1456 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1457 static unsigned num_msrs_to_save;
1458 
1459 static const u32 emulated_msrs_all[] = {
1460 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1461 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1462 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1463 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1464 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1465 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1466 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1467 	HV_X64_MSR_RESET,
1468 	HV_X64_MSR_VP_INDEX,
1469 	HV_X64_MSR_VP_RUNTIME,
1470 	HV_X64_MSR_SCONTROL,
1471 	HV_X64_MSR_STIMER0_CONFIG,
1472 	HV_X64_MSR_VP_ASSIST_PAGE,
1473 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1474 	HV_X64_MSR_TSC_EMULATION_STATUS,
1475 	HV_X64_MSR_SYNDBG_OPTIONS,
1476 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1477 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1478 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1479 
1480 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1481 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1482 
1483 	MSR_IA32_TSC_ADJUST,
1484 	MSR_IA32_TSC_DEADLINE,
1485 	MSR_IA32_ARCH_CAPABILITIES,
1486 	MSR_IA32_PERF_CAPABILITIES,
1487 	MSR_IA32_MISC_ENABLE,
1488 	MSR_IA32_MCG_STATUS,
1489 	MSR_IA32_MCG_CTL,
1490 	MSR_IA32_MCG_EXT_CTL,
1491 	MSR_IA32_SMBASE,
1492 	MSR_SMI_COUNT,
1493 	MSR_PLATFORM_INFO,
1494 	MSR_MISC_FEATURES_ENABLES,
1495 	MSR_AMD64_VIRT_SPEC_CTRL,
1496 	MSR_AMD64_TSC_RATIO,
1497 	MSR_IA32_POWER_CTL,
1498 	MSR_IA32_UCODE_REV,
1499 
1500 	/*
1501 	 * The following list leaves out MSRs whose values are determined
1502 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1503 	 * We always support the "true" VMX control MSRs, even if the host
1504 	 * processor does not, so I am putting these registers here rather
1505 	 * than in msrs_to_save_all.
1506 	 */
1507 	MSR_IA32_VMX_BASIC,
1508 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1509 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1510 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1511 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1512 	MSR_IA32_VMX_MISC,
1513 	MSR_IA32_VMX_CR0_FIXED0,
1514 	MSR_IA32_VMX_CR4_FIXED0,
1515 	MSR_IA32_VMX_VMCS_ENUM,
1516 	MSR_IA32_VMX_PROCBASED_CTLS2,
1517 	MSR_IA32_VMX_EPT_VPID_CAP,
1518 	MSR_IA32_VMX_VMFUNC,
1519 
1520 	MSR_K7_HWCR,
1521 	MSR_KVM_POLL_CONTROL,
1522 };
1523 
1524 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1525 static unsigned num_emulated_msrs;
1526 
1527 /*
1528  * List of msr numbers which are used to expose MSR-based features that
1529  * can be used by a hypervisor to validate requested CPU features.
1530  */
1531 static const u32 msr_based_features_all[] = {
1532 	MSR_IA32_VMX_BASIC,
1533 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1534 	MSR_IA32_VMX_PINBASED_CTLS,
1535 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1536 	MSR_IA32_VMX_PROCBASED_CTLS,
1537 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1538 	MSR_IA32_VMX_EXIT_CTLS,
1539 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1540 	MSR_IA32_VMX_ENTRY_CTLS,
1541 	MSR_IA32_VMX_MISC,
1542 	MSR_IA32_VMX_CR0_FIXED0,
1543 	MSR_IA32_VMX_CR0_FIXED1,
1544 	MSR_IA32_VMX_CR4_FIXED0,
1545 	MSR_IA32_VMX_CR4_FIXED1,
1546 	MSR_IA32_VMX_VMCS_ENUM,
1547 	MSR_IA32_VMX_PROCBASED_CTLS2,
1548 	MSR_IA32_VMX_EPT_VPID_CAP,
1549 	MSR_IA32_VMX_VMFUNC,
1550 
1551 	MSR_F10H_DECFG,
1552 	MSR_IA32_UCODE_REV,
1553 	MSR_IA32_ARCH_CAPABILITIES,
1554 	MSR_IA32_PERF_CAPABILITIES,
1555 };
1556 
1557 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1558 static unsigned int num_msr_based_features;
1559 
1560 static u64 kvm_get_arch_capabilities(void)
1561 {
1562 	u64 data = 0;
1563 
1564 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1565 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1566 
1567 	/*
1568 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1569 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1570 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1571 	 * L1 guests, so it need not worry about its own (L2) guests.
1572 	 */
1573 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1574 
1575 	/*
1576 	 * If we're doing cache flushes (either "always" or "cond")
1577 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1578 	 * If an outer hypervisor is doing the cache flush for us
1579 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1580 	 * capability to the guest too, and if EPT is disabled we're not
1581 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1582 	 * require a nested hypervisor to do a flush of its own.
1583 	 */
1584 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1585 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1586 
1587 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1588 		data |= ARCH_CAP_RDCL_NO;
1589 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1590 		data |= ARCH_CAP_SSB_NO;
1591 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1592 		data |= ARCH_CAP_MDS_NO;
1593 
1594 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1595 		/*
1596 		 * If RTM=0 because the kernel has disabled TSX, the host might
1597 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1598 		 * and therefore knows that there cannot be TAA) but keep
1599 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1600 		 * and we want to allow migrating those guests to tsx=off hosts.
1601 		 */
1602 		data &= ~ARCH_CAP_TAA_NO;
1603 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1604 		data |= ARCH_CAP_TAA_NO;
1605 	} else {
1606 		/*
1607 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1608 		 * host so the guest can choose between disabling TSX or
1609 		 * using VERW to clear CPU buffers.
1610 		 */
1611 	}
1612 
1613 	/* Guests don't need to know "Fill buffer clear control" exists */
1614 	data &= ~ARCH_CAP_FB_CLEAR_CTRL;
1615 
1616 	return data;
1617 }
1618 
1619 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1620 {
1621 	switch (msr->index) {
1622 	case MSR_IA32_ARCH_CAPABILITIES:
1623 		msr->data = kvm_get_arch_capabilities();
1624 		break;
1625 	case MSR_IA32_UCODE_REV:
1626 		rdmsrl_safe(msr->index, &msr->data);
1627 		break;
1628 	default:
1629 		return static_call(kvm_x86_get_msr_feature)(msr);
1630 	}
1631 	return 0;
1632 }
1633 
1634 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1635 {
1636 	struct kvm_msr_entry msr;
1637 	int r;
1638 
1639 	msr.index = index;
1640 	r = kvm_get_msr_feature(&msr);
1641 
1642 	if (r == KVM_MSR_RET_INVALID) {
1643 		/* Unconditionally clear the output for simplicity */
1644 		*data = 0;
1645 		if (kvm_msr_ignored_check(index, 0, false))
1646 			r = 0;
1647 	}
1648 
1649 	if (r)
1650 		return r;
1651 
1652 	*data = msr.data;
1653 
1654 	return 0;
1655 }
1656 
1657 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1658 {
1659 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1660 		return false;
1661 
1662 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1663 		return false;
1664 
1665 	if (efer & (EFER_LME | EFER_LMA) &&
1666 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1667 		return false;
1668 
1669 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1670 		return false;
1671 
1672 	return true;
1673 
1674 }
1675 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1676 {
1677 	if (efer & efer_reserved_bits)
1678 		return false;
1679 
1680 	return __kvm_valid_efer(vcpu, efer);
1681 }
1682 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1683 
1684 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1685 {
1686 	u64 old_efer = vcpu->arch.efer;
1687 	u64 efer = msr_info->data;
1688 	int r;
1689 
1690 	if (efer & efer_reserved_bits)
1691 		return 1;
1692 
1693 	if (!msr_info->host_initiated) {
1694 		if (!__kvm_valid_efer(vcpu, efer))
1695 			return 1;
1696 
1697 		if (is_paging(vcpu) &&
1698 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1699 			return 1;
1700 	}
1701 
1702 	efer &= ~EFER_LMA;
1703 	efer |= vcpu->arch.efer & EFER_LMA;
1704 
1705 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1706 	if (r) {
1707 		WARN_ON(r > 0);
1708 		return r;
1709 	}
1710 
1711 	if ((efer ^ old_efer) & KVM_MMU_EFER_ROLE_BITS)
1712 		kvm_mmu_reset_context(vcpu);
1713 
1714 	return 0;
1715 }
1716 
1717 void kvm_enable_efer_bits(u64 mask)
1718 {
1719        efer_reserved_bits &= ~mask;
1720 }
1721 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1722 
1723 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1724 {
1725 	struct kvm_x86_msr_filter *msr_filter;
1726 	struct msr_bitmap_range *ranges;
1727 	struct kvm *kvm = vcpu->kvm;
1728 	bool allowed;
1729 	int idx;
1730 	u32 i;
1731 
1732 	/* x2APIC MSRs do not support filtering. */
1733 	if (index >= 0x800 && index <= 0x8ff)
1734 		return true;
1735 
1736 	idx = srcu_read_lock(&kvm->srcu);
1737 
1738 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1739 	if (!msr_filter) {
1740 		allowed = true;
1741 		goto out;
1742 	}
1743 
1744 	allowed = msr_filter->default_allow;
1745 	ranges = msr_filter->ranges;
1746 
1747 	for (i = 0; i < msr_filter->count; i++) {
1748 		u32 start = ranges[i].base;
1749 		u32 end = start + ranges[i].nmsrs;
1750 		u32 flags = ranges[i].flags;
1751 		unsigned long *bitmap = ranges[i].bitmap;
1752 
1753 		if ((index >= start) && (index < end) && (flags & type)) {
1754 			allowed = !!test_bit(index - start, bitmap);
1755 			break;
1756 		}
1757 	}
1758 
1759 out:
1760 	srcu_read_unlock(&kvm->srcu, idx);
1761 
1762 	return allowed;
1763 }
1764 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1765 
1766 /*
1767  * Write @data into the MSR specified by @index.  Select MSR specific fault
1768  * checks are bypassed if @host_initiated is %true.
1769  * Returns 0 on success, non-0 otherwise.
1770  * Assumes vcpu_load() was already called.
1771  */
1772 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1773 			 bool host_initiated)
1774 {
1775 	struct msr_data msr;
1776 
1777 	switch (index) {
1778 	case MSR_FS_BASE:
1779 	case MSR_GS_BASE:
1780 	case MSR_KERNEL_GS_BASE:
1781 	case MSR_CSTAR:
1782 	case MSR_LSTAR:
1783 		if (is_noncanonical_address(data, vcpu))
1784 			return 1;
1785 		break;
1786 	case MSR_IA32_SYSENTER_EIP:
1787 	case MSR_IA32_SYSENTER_ESP:
1788 		/*
1789 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1790 		 * non-canonical address is written on Intel but not on
1791 		 * AMD (which ignores the top 32-bits, because it does
1792 		 * not implement 64-bit SYSENTER).
1793 		 *
1794 		 * 64-bit code should hence be able to write a non-canonical
1795 		 * value on AMD.  Making the address canonical ensures that
1796 		 * vmentry does not fail on Intel after writing a non-canonical
1797 		 * value, and that something deterministic happens if the guest
1798 		 * invokes 64-bit SYSENTER.
1799 		 */
1800 		data = __canonical_address(data, vcpu_virt_addr_bits(vcpu));
1801 		break;
1802 	case MSR_TSC_AUX:
1803 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1804 			return 1;
1805 
1806 		if (!host_initiated &&
1807 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1808 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1809 			return 1;
1810 
1811 		/*
1812 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1813 		 * incomplete and conflicting architectural behavior.  Current
1814 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1815 		 * reserved and always read as zeros.  Enforce Intel's reserved
1816 		 * bits check if and only if the guest CPU is Intel, and clear
1817 		 * the bits in all other cases.  This ensures cross-vendor
1818 		 * migration will provide consistent behavior for the guest.
1819 		 */
1820 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1821 			return 1;
1822 
1823 		data = (u32)data;
1824 		break;
1825 	}
1826 
1827 	msr.data = data;
1828 	msr.index = index;
1829 	msr.host_initiated = host_initiated;
1830 
1831 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1832 }
1833 
1834 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1835 				     u32 index, u64 data, bool host_initiated)
1836 {
1837 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1838 
1839 	if (ret == KVM_MSR_RET_INVALID)
1840 		if (kvm_msr_ignored_check(index, data, true))
1841 			ret = 0;
1842 
1843 	return ret;
1844 }
1845 
1846 /*
1847  * Read the MSR specified by @index into @data.  Select MSR specific fault
1848  * checks are bypassed if @host_initiated is %true.
1849  * Returns 0 on success, non-0 otherwise.
1850  * Assumes vcpu_load() was already called.
1851  */
1852 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1853 		  bool host_initiated)
1854 {
1855 	struct msr_data msr;
1856 	int ret;
1857 
1858 	switch (index) {
1859 	case MSR_TSC_AUX:
1860 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1861 			return 1;
1862 
1863 		if (!host_initiated &&
1864 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1865 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1866 			return 1;
1867 		break;
1868 	}
1869 
1870 	msr.index = index;
1871 	msr.host_initiated = host_initiated;
1872 
1873 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1874 	if (!ret)
1875 		*data = msr.data;
1876 	return ret;
1877 }
1878 
1879 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1880 				     u32 index, u64 *data, bool host_initiated)
1881 {
1882 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1883 
1884 	if (ret == KVM_MSR_RET_INVALID) {
1885 		/* Unconditionally clear *data for simplicity */
1886 		*data = 0;
1887 		if (kvm_msr_ignored_check(index, 0, false))
1888 			ret = 0;
1889 	}
1890 
1891 	return ret;
1892 }
1893 
1894 static int kvm_get_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1895 {
1896 	if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1897 		return KVM_MSR_RET_FILTERED;
1898 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1899 }
1900 
1901 static int kvm_set_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 data)
1902 {
1903 	if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1904 		return KVM_MSR_RET_FILTERED;
1905 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1906 }
1907 
1908 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1909 {
1910 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1911 }
1912 EXPORT_SYMBOL_GPL(kvm_get_msr);
1913 
1914 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1915 {
1916 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1917 }
1918 EXPORT_SYMBOL_GPL(kvm_set_msr);
1919 
1920 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1921 {
1922 	if (!vcpu->run->msr.error) {
1923 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1924 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1925 	}
1926 }
1927 
1928 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1929 {
1930 	return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
1931 }
1932 
1933 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1934 {
1935 	complete_userspace_rdmsr(vcpu);
1936 	return complete_emulated_msr_access(vcpu);
1937 }
1938 
1939 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
1940 {
1941 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1942 }
1943 
1944 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
1945 {
1946 	complete_userspace_rdmsr(vcpu);
1947 	return complete_fast_msr_access(vcpu);
1948 }
1949 
1950 static u64 kvm_msr_reason(int r)
1951 {
1952 	switch (r) {
1953 	case KVM_MSR_RET_INVALID:
1954 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1955 	case KVM_MSR_RET_FILTERED:
1956 		return KVM_MSR_EXIT_REASON_FILTER;
1957 	default:
1958 		return KVM_MSR_EXIT_REASON_INVAL;
1959 	}
1960 }
1961 
1962 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1963 			      u32 exit_reason, u64 data,
1964 			      int (*completion)(struct kvm_vcpu *vcpu),
1965 			      int r)
1966 {
1967 	u64 msr_reason = kvm_msr_reason(r);
1968 
1969 	/* Check if the user wanted to know about this MSR fault */
1970 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1971 		return 0;
1972 
1973 	vcpu->run->exit_reason = exit_reason;
1974 	vcpu->run->msr.error = 0;
1975 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1976 	vcpu->run->msr.reason = msr_reason;
1977 	vcpu->run->msr.index = index;
1978 	vcpu->run->msr.data = data;
1979 	vcpu->arch.complete_userspace_io = completion;
1980 
1981 	return 1;
1982 }
1983 
1984 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1985 {
1986 	u32 ecx = kvm_rcx_read(vcpu);
1987 	u64 data;
1988 	int r;
1989 
1990 	r = kvm_get_msr_with_filter(vcpu, ecx, &data);
1991 
1992 	if (!r) {
1993 		trace_kvm_msr_read(ecx, data);
1994 
1995 		kvm_rax_write(vcpu, data & -1u);
1996 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1997 	} else {
1998 		/* MSR read failed? See if we should ask user space */
1999 		if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
2000 				       complete_fast_rdmsr, r))
2001 			return 0;
2002 		trace_kvm_msr_read_ex(ecx);
2003 	}
2004 
2005 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2006 }
2007 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
2008 
2009 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
2010 {
2011 	u32 ecx = kvm_rcx_read(vcpu);
2012 	u64 data = kvm_read_edx_eax(vcpu);
2013 	int r;
2014 
2015 	r = kvm_set_msr_with_filter(vcpu, ecx, data);
2016 
2017 	if (!r) {
2018 		trace_kvm_msr_write(ecx, data);
2019 	} else {
2020 		/* MSR write failed? See if we should ask user space */
2021 		if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
2022 				       complete_fast_msr_access, r))
2023 			return 0;
2024 		/* Signal all other negative errors to userspace */
2025 		if (r < 0)
2026 			return r;
2027 		trace_kvm_msr_write_ex(ecx, data);
2028 	}
2029 
2030 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2031 }
2032 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
2033 
2034 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
2035 {
2036 	return kvm_skip_emulated_instruction(vcpu);
2037 }
2038 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
2039 
2040 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
2041 {
2042 	/* Treat an INVD instruction as a NOP and just skip it. */
2043 	return kvm_emulate_as_nop(vcpu);
2044 }
2045 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
2046 
2047 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
2048 {
2049 	kvm_queue_exception(vcpu, UD_VECTOR);
2050 	return 1;
2051 }
2052 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
2053 
2054 
2055 static int kvm_emulate_monitor_mwait(struct kvm_vcpu *vcpu, const char *insn)
2056 {
2057 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MWAIT_NEVER_UD_FAULTS) &&
2058 	    !guest_cpuid_has(vcpu, X86_FEATURE_MWAIT))
2059 		return kvm_handle_invalid_op(vcpu);
2060 
2061 	pr_warn_once("kvm: %s instruction emulated as NOP!\n", insn);
2062 	return kvm_emulate_as_nop(vcpu);
2063 }
2064 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
2065 {
2066 	return kvm_emulate_monitor_mwait(vcpu, "MWAIT");
2067 }
2068 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
2069 
2070 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
2071 {
2072 	return kvm_emulate_monitor_mwait(vcpu, "MONITOR");
2073 }
2074 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
2075 
2076 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
2077 {
2078 	xfer_to_guest_mode_prepare();
2079 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
2080 		xfer_to_guest_mode_work_pending();
2081 }
2082 
2083 /*
2084  * The fast path for frequent and performance sensitive wrmsr emulation,
2085  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2086  * the latency of virtual IPI by avoiding the expensive bits of transitioning
2087  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2088  * other cases which must be called after interrupts are enabled on the host.
2089  */
2090 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2091 {
2092 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2093 		return 1;
2094 
2095 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2096 	    ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2097 	    ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2098 	    ((u32)(data >> 32) != X2APIC_BROADCAST))
2099 		return kvm_x2apic_icr_write(vcpu->arch.apic, data);
2100 
2101 	return 1;
2102 }
2103 
2104 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2105 {
2106 	if (!kvm_can_use_hv_timer(vcpu))
2107 		return 1;
2108 
2109 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
2110 	return 0;
2111 }
2112 
2113 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2114 {
2115 	u32 msr = kvm_rcx_read(vcpu);
2116 	u64 data;
2117 	fastpath_t ret = EXIT_FASTPATH_NONE;
2118 
2119 	switch (msr) {
2120 	case APIC_BASE_MSR + (APIC_ICR >> 4):
2121 		data = kvm_read_edx_eax(vcpu);
2122 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2123 			kvm_skip_emulated_instruction(vcpu);
2124 			ret = EXIT_FASTPATH_EXIT_HANDLED;
2125 		}
2126 		break;
2127 	case MSR_IA32_TSC_DEADLINE:
2128 		data = kvm_read_edx_eax(vcpu);
2129 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2130 			kvm_skip_emulated_instruction(vcpu);
2131 			ret = EXIT_FASTPATH_REENTER_GUEST;
2132 		}
2133 		break;
2134 	default:
2135 		break;
2136 	}
2137 
2138 	if (ret != EXIT_FASTPATH_NONE)
2139 		trace_kvm_msr_write(msr, data);
2140 
2141 	return ret;
2142 }
2143 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2144 
2145 /*
2146  * Adapt set_msr() to msr_io()'s calling convention
2147  */
2148 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2149 {
2150 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
2151 }
2152 
2153 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2154 {
2155 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2156 }
2157 
2158 #ifdef CONFIG_X86_64
2159 struct pvclock_clock {
2160 	int vclock_mode;
2161 	u64 cycle_last;
2162 	u64 mask;
2163 	u32 mult;
2164 	u32 shift;
2165 	u64 base_cycles;
2166 	u64 offset;
2167 };
2168 
2169 struct pvclock_gtod_data {
2170 	seqcount_t	seq;
2171 
2172 	struct pvclock_clock clock; /* extract of a clocksource struct */
2173 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2174 
2175 	ktime_t		offs_boot;
2176 	u64		wall_time_sec;
2177 };
2178 
2179 static struct pvclock_gtod_data pvclock_gtod_data;
2180 
2181 static void update_pvclock_gtod(struct timekeeper *tk)
2182 {
2183 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2184 
2185 	write_seqcount_begin(&vdata->seq);
2186 
2187 	/* copy pvclock gtod data */
2188 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2189 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2190 	vdata->clock.mask		= tk->tkr_mono.mask;
2191 	vdata->clock.mult		= tk->tkr_mono.mult;
2192 	vdata->clock.shift		= tk->tkr_mono.shift;
2193 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2194 	vdata->clock.offset		= tk->tkr_mono.base;
2195 
2196 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2197 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2198 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2199 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2200 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2201 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2202 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2203 
2204 	vdata->wall_time_sec            = tk->xtime_sec;
2205 
2206 	vdata->offs_boot		= tk->offs_boot;
2207 
2208 	write_seqcount_end(&vdata->seq);
2209 }
2210 
2211 static s64 get_kvmclock_base_ns(void)
2212 {
2213 	/* Count up from boot time, but with the frequency of the raw clock.  */
2214 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2215 }
2216 #else
2217 static s64 get_kvmclock_base_ns(void)
2218 {
2219 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2220 	return ktime_get_boottime_ns();
2221 }
2222 #endif
2223 
2224 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2225 {
2226 	int version;
2227 	int r;
2228 	struct pvclock_wall_clock wc;
2229 	u32 wc_sec_hi;
2230 	u64 wall_nsec;
2231 
2232 	if (!wall_clock)
2233 		return;
2234 
2235 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2236 	if (r)
2237 		return;
2238 
2239 	if (version & 1)
2240 		++version;  /* first time write, random junk */
2241 
2242 	++version;
2243 
2244 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2245 		return;
2246 
2247 	/*
2248 	 * The guest calculates current wall clock time by adding
2249 	 * system time (updated by kvm_guest_time_update below) to the
2250 	 * wall clock specified here.  We do the reverse here.
2251 	 */
2252 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2253 
2254 	wc.nsec = do_div(wall_nsec, 1000000000);
2255 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2256 	wc.version = version;
2257 
2258 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2259 
2260 	if (sec_hi_ofs) {
2261 		wc_sec_hi = wall_nsec >> 32;
2262 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2263 				&wc_sec_hi, sizeof(wc_sec_hi));
2264 	}
2265 
2266 	version++;
2267 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2268 }
2269 
2270 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2271 				  bool old_msr, bool host_initiated)
2272 {
2273 	struct kvm_arch *ka = &vcpu->kvm->arch;
2274 
2275 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2276 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2277 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2278 
2279 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2280 	}
2281 
2282 	vcpu->arch.time = system_time;
2283 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2284 
2285 	/* we verify if the enable bit is set... */
2286 	if (system_time & 1) {
2287 		kvm_gfn_to_pfn_cache_init(vcpu->kvm, &vcpu->arch.pv_time, vcpu,
2288 					  KVM_HOST_USES_PFN, system_time & ~1ULL,
2289 					  sizeof(struct pvclock_vcpu_time_info));
2290 	} else {
2291 		kvm_gfn_to_pfn_cache_destroy(vcpu->kvm, &vcpu->arch.pv_time);
2292 	}
2293 
2294 	return;
2295 }
2296 
2297 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2298 {
2299 	do_shl32_div32(dividend, divisor);
2300 	return dividend;
2301 }
2302 
2303 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2304 			       s8 *pshift, u32 *pmultiplier)
2305 {
2306 	uint64_t scaled64;
2307 	int32_t  shift = 0;
2308 	uint64_t tps64;
2309 	uint32_t tps32;
2310 
2311 	tps64 = base_hz;
2312 	scaled64 = scaled_hz;
2313 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2314 		tps64 >>= 1;
2315 		shift--;
2316 	}
2317 
2318 	tps32 = (uint32_t)tps64;
2319 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2320 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2321 			scaled64 >>= 1;
2322 		else
2323 			tps32 <<= 1;
2324 		shift++;
2325 	}
2326 
2327 	*pshift = shift;
2328 	*pmultiplier = div_frac(scaled64, tps32);
2329 }
2330 
2331 #ifdef CONFIG_X86_64
2332 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2333 #endif
2334 
2335 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2336 static unsigned long max_tsc_khz;
2337 
2338 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2339 {
2340 	u64 v = (u64)khz * (1000000 + ppm);
2341 	do_div(v, 1000000);
2342 	return v;
2343 }
2344 
2345 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2346 
2347 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2348 {
2349 	u64 ratio;
2350 
2351 	/* Guest TSC same frequency as host TSC? */
2352 	if (!scale) {
2353 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2354 		return 0;
2355 	}
2356 
2357 	/* TSC scaling supported? */
2358 	if (!kvm_caps.has_tsc_control) {
2359 		if (user_tsc_khz > tsc_khz) {
2360 			vcpu->arch.tsc_catchup = 1;
2361 			vcpu->arch.tsc_always_catchup = 1;
2362 			return 0;
2363 		} else {
2364 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2365 			return -1;
2366 		}
2367 	}
2368 
2369 	/* TSC scaling required  - calculate ratio */
2370 	ratio = mul_u64_u32_div(1ULL << kvm_caps.tsc_scaling_ratio_frac_bits,
2371 				user_tsc_khz, tsc_khz);
2372 
2373 	if (ratio == 0 || ratio >= kvm_caps.max_tsc_scaling_ratio) {
2374 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2375 			            user_tsc_khz);
2376 		return -1;
2377 	}
2378 
2379 	kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2380 	return 0;
2381 }
2382 
2383 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2384 {
2385 	u32 thresh_lo, thresh_hi;
2386 	int use_scaling = 0;
2387 
2388 	/* tsc_khz can be zero if TSC calibration fails */
2389 	if (user_tsc_khz == 0) {
2390 		/* set tsc_scaling_ratio to a safe value */
2391 		kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2392 		return -1;
2393 	}
2394 
2395 	/* Compute a scale to convert nanoseconds in TSC cycles */
2396 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2397 			   &vcpu->arch.virtual_tsc_shift,
2398 			   &vcpu->arch.virtual_tsc_mult);
2399 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2400 
2401 	/*
2402 	 * Compute the variation in TSC rate which is acceptable
2403 	 * within the range of tolerance and decide if the
2404 	 * rate being applied is within that bounds of the hardware
2405 	 * rate.  If so, no scaling or compensation need be done.
2406 	 */
2407 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2408 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2409 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2410 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2411 		use_scaling = 1;
2412 	}
2413 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2414 }
2415 
2416 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2417 {
2418 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2419 				      vcpu->arch.virtual_tsc_mult,
2420 				      vcpu->arch.virtual_tsc_shift);
2421 	tsc += vcpu->arch.this_tsc_write;
2422 	return tsc;
2423 }
2424 
2425 #ifdef CONFIG_X86_64
2426 static inline int gtod_is_based_on_tsc(int mode)
2427 {
2428 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2429 }
2430 #endif
2431 
2432 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2433 {
2434 #ifdef CONFIG_X86_64
2435 	bool vcpus_matched;
2436 	struct kvm_arch *ka = &vcpu->kvm->arch;
2437 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2438 
2439 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2440 			 atomic_read(&vcpu->kvm->online_vcpus));
2441 
2442 	/*
2443 	 * Once the masterclock is enabled, always perform request in
2444 	 * order to update it.
2445 	 *
2446 	 * In order to enable masterclock, the host clocksource must be TSC
2447 	 * and the vcpus need to have matched TSCs.  When that happens,
2448 	 * perform request to enable masterclock.
2449 	 */
2450 	if (ka->use_master_clock ||
2451 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2452 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2453 
2454 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2455 			    atomic_read(&vcpu->kvm->online_vcpus),
2456 		            ka->use_master_clock, gtod->clock.vclock_mode);
2457 #endif
2458 }
2459 
2460 /*
2461  * Multiply tsc by a fixed point number represented by ratio.
2462  *
2463  * The most significant 64-N bits (mult) of ratio represent the
2464  * integral part of the fixed point number; the remaining N bits
2465  * (frac) represent the fractional part, ie. ratio represents a fixed
2466  * point number (mult + frac * 2^(-N)).
2467  *
2468  * N equals to kvm_caps.tsc_scaling_ratio_frac_bits.
2469  */
2470 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2471 {
2472 	return mul_u64_u64_shr(tsc, ratio, kvm_caps.tsc_scaling_ratio_frac_bits);
2473 }
2474 
2475 u64 kvm_scale_tsc(u64 tsc, u64 ratio)
2476 {
2477 	u64 _tsc = tsc;
2478 
2479 	if (ratio != kvm_caps.default_tsc_scaling_ratio)
2480 		_tsc = __scale_tsc(ratio, tsc);
2481 
2482 	return _tsc;
2483 }
2484 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2485 
2486 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2487 {
2488 	u64 tsc;
2489 
2490 	tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2491 
2492 	return target_tsc - tsc;
2493 }
2494 
2495 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2496 {
2497 	return vcpu->arch.l1_tsc_offset +
2498 		kvm_scale_tsc(host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2499 }
2500 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2501 
2502 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2503 {
2504 	u64 nested_offset;
2505 
2506 	if (l2_multiplier == kvm_caps.default_tsc_scaling_ratio)
2507 		nested_offset = l1_offset;
2508 	else
2509 		nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2510 						kvm_caps.tsc_scaling_ratio_frac_bits);
2511 
2512 	nested_offset += l2_offset;
2513 	return nested_offset;
2514 }
2515 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2516 
2517 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2518 {
2519 	if (l2_multiplier != kvm_caps.default_tsc_scaling_ratio)
2520 		return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2521 				       kvm_caps.tsc_scaling_ratio_frac_bits);
2522 
2523 	return l1_multiplier;
2524 }
2525 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2526 
2527 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2528 {
2529 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2530 				   vcpu->arch.l1_tsc_offset,
2531 				   l1_offset);
2532 
2533 	vcpu->arch.l1_tsc_offset = l1_offset;
2534 
2535 	/*
2536 	 * If we are here because L1 chose not to trap WRMSR to TSC then
2537 	 * according to the spec this should set L1's TSC (as opposed to
2538 	 * setting L1's offset for L2).
2539 	 */
2540 	if (is_guest_mode(vcpu))
2541 		vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2542 			l1_offset,
2543 			static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2544 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2545 	else
2546 		vcpu->arch.tsc_offset = l1_offset;
2547 
2548 	static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2549 }
2550 
2551 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2552 {
2553 	vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2554 
2555 	/* Userspace is changing the multiplier while L2 is active */
2556 	if (is_guest_mode(vcpu))
2557 		vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2558 			l1_multiplier,
2559 			static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2560 	else
2561 		vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2562 
2563 	if (kvm_caps.has_tsc_control)
2564 		static_call(kvm_x86_write_tsc_multiplier)(
2565 			vcpu, vcpu->arch.tsc_scaling_ratio);
2566 }
2567 
2568 static inline bool kvm_check_tsc_unstable(void)
2569 {
2570 #ifdef CONFIG_X86_64
2571 	/*
2572 	 * TSC is marked unstable when we're running on Hyper-V,
2573 	 * 'TSC page' clocksource is good.
2574 	 */
2575 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2576 		return false;
2577 #endif
2578 	return check_tsc_unstable();
2579 }
2580 
2581 /*
2582  * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2583  * offset for the vcpu and tracks the TSC matching generation that the vcpu
2584  * participates in.
2585  */
2586 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2587 				  u64 ns, bool matched)
2588 {
2589 	struct kvm *kvm = vcpu->kvm;
2590 
2591 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2592 
2593 	/*
2594 	 * We also track th most recent recorded KHZ, write and time to
2595 	 * allow the matching interval to be extended at each write.
2596 	 */
2597 	kvm->arch.last_tsc_nsec = ns;
2598 	kvm->arch.last_tsc_write = tsc;
2599 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2600 	kvm->arch.last_tsc_offset = offset;
2601 
2602 	vcpu->arch.last_guest_tsc = tsc;
2603 
2604 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2605 
2606 	if (!matched) {
2607 		/*
2608 		 * We split periods of matched TSC writes into generations.
2609 		 * For each generation, we track the original measured
2610 		 * nanosecond time, offset, and write, so if TSCs are in
2611 		 * sync, we can match exact offset, and if not, we can match
2612 		 * exact software computation in compute_guest_tsc()
2613 		 *
2614 		 * These values are tracked in kvm->arch.cur_xxx variables.
2615 		 */
2616 		kvm->arch.cur_tsc_generation++;
2617 		kvm->arch.cur_tsc_nsec = ns;
2618 		kvm->arch.cur_tsc_write = tsc;
2619 		kvm->arch.cur_tsc_offset = offset;
2620 		kvm->arch.nr_vcpus_matched_tsc = 0;
2621 	} else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2622 		kvm->arch.nr_vcpus_matched_tsc++;
2623 	}
2624 
2625 	/* Keep track of which generation this VCPU has synchronized to */
2626 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2627 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2628 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2629 
2630 	kvm_track_tsc_matching(vcpu);
2631 }
2632 
2633 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2634 {
2635 	struct kvm *kvm = vcpu->kvm;
2636 	u64 offset, ns, elapsed;
2637 	unsigned long flags;
2638 	bool matched = false;
2639 	bool synchronizing = false;
2640 
2641 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2642 	offset = kvm_compute_l1_tsc_offset(vcpu, data);
2643 	ns = get_kvmclock_base_ns();
2644 	elapsed = ns - kvm->arch.last_tsc_nsec;
2645 
2646 	if (vcpu->arch.virtual_tsc_khz) {
2647 		if (data == 0) {
2648 			/*
2649 			 * detection of vcpu initialization -- need to sync
2650 			 * with other vCPUs. This particularly helps to keep
2651 			 * kvm_clock stable after CPU hotplug
2652 			 */
2653 			synchronizing = true;
2654 		} else {
2655 			u64 tsc_exp = kvm->arch.last_tsc_write +
2656 						nsec_to_cycles(vcpu, elapsed);
2657 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2658 			/*
2659 			 * Special case: TSC write with a small delta (1 second)
2660 			 * of virtual cycle time against real time is
2661 			 * interpreted as an attempt to synchronize the CPU.
2662 			 */
2663 			synchronizing = data < tsc_exp + tsc_hz &&
2664 					data + tsc_hz > tsc_exp;
2665 		}
2666 	}
2667 
2668 	/*
2669 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2670 	 * TSC, we add elapsed time in this computation.  We could let the
2671 	 * compensation code attempt to catch up if we fall behind, but
2672 	 * it's better to try to match offsets from the beginning.
2673          */
2674 	if (synchronizing &&
2675 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2676 		if (!kvm_check_tsc_unstable()) {
2677 			offset = kvm->arch.cur_tsc_offset;
2678 		} else {
2679 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2680 			data += delta;
2681 			offset = kvm_compute_l1_tsc_offset(vcpu, data);
2682 		}
2683 		matched = true;
2684 	}
2685 
2686 	__kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2687 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2688 }
2689 
2690 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2691 					   s64 adjustment)
2692 {
2693 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2694 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2695 }
2696 
2697 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2698 {
2699 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio)
2700 		WARN_ON(adjustment < 0);
2701 	adjustment = kvm_scale_tsc((u64) adjustment,
2702 				   vcpu->arch.l1_tsc_scaling_ratio);
2703 	adjust_tsc_offset_guest(vcpu, adjustment);
2704 }
2705 
2706 #ifdef CONFIG_X86_64
2707 
2708 static u64 read_tsc(void)
2709 {
2710 	u64 ret = (u64)rdtsc_ordered();
2711 	u64 last = pvclock_gtod_data.clock.cycle_last;
2712 
2713 	if (likely(ret >= last))
2714 		return ret;
2715 
2716 	/*
2717 	 * GCC likes to generate cmov here, but this branch is extremely
2718 	 * predictable (it's just a function of time and the likely is
2719 	 * very likely) and there's a data dependence, so force GCC
2720 	 * to generate a branch instead.  I don't barrier() because
2721 	 * we don't actually need a barrier, and if this function
2722 	 * ever gets inlined it will generate worse code.
2723 	 */
2724 	asm volatile ("");
2725 	return last;
2726 }
2727 
2728 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2729 			  int *mode)
2730 {
2731 	long v;
2732 	u64 tsc_pg_val;
2733 
2734 	switch (clock->vclock_mode) {
2735 	case VDSO_CLOCKMODE_HVCLOCK:
2736 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2737 						  tsc_timestamp);
2738 		if (tsc_pg_val != U64_MAX) {
2739 			/* TSC page valid */
2740 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2741 			v = (tsc_pg_val - clock->cycle_last) &
2742 				clock->mask;
2743 		} else {
2744 			/* TSC page invalid */
2745 			*mode = VDSO_CLOCKMODE_NONE;
2746 		}
2747 		break;
2748 	case VDSO_CLOCKMODE_TSC:
2749 		*mode = VDSO_CLOCKMODE_TSC;
2750 		*tsc_timestamp = read_tsc();
2751 		v = (*tsc_timestamp - clock->cycle_last) &
2752 			clock->mask;
2753 		break;
2754 	default:
2755 		*mode = VDSO_CLOCKMODE_NONE;
2756 	}
2757 
2758 	if (*mode == VDSO_CLOCKMODE_NONE)
2759 		*tsc_timestamp = v = 0;
2760 
2761 	return v * clock->mult;
2762 }
2763 
2764 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2765 {
2766 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2767 	unsigned long seq;
2768 	int mode;
2769 	u64 ns;
2770 
2771 	do {
2772 		seq = read_seqcount_begin(&gtod->seq);
2773 		ns = gtod->raw_clock.base_cycles;
2774 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2775 		ns >>= gtod->raw_clock.shift;
2776 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2777 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2778 	*t = ns;
2779 
2780 	return mode;
2781 }
2782 
2783 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2784 {
2785 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2786 	unsigned long seq;
2787 	int mode;
2788 	u64 ns;
2789 
2790 	do {
2791 		seq = read_seqcount_begin(&gtod->seq);
2792 		ts->tv_sec = gtod->wall_time_sec;
2793 		ns = gtod->clock.base_cycles;
2794 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2795 		ns >>= gtod->clock.shift;
2796 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2797 
2798 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2799 	ts->tv_nsec = ns;
2800 
2801 	return mode;
2802 }
2803 
2804 /* returns true if host is using TSC based clocksource */
2805 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2806 {
2807 	/* checked again under seqlock below */
2808 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2809 		return false;
2810 
2811 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2812 						      tsc_timestamp));
2813 }
2814 
2815 /* returns true if host is using TSC based clocksource */
2816 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2817 					   u64 *tsc_timestamp)
2818 {
2819 	/* checked again under seqlock below */
2820 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2821 		return false;
2822 
2823 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2824 }
2825 #endif
2826 
2827 /*
2828  *
2829  * Assuming a stable TSC across physical CPUS, and a stable TSC
2830  * across virtual CPUs, the following condition is possible.
2831  * Each numbered line represents an event visible to both
2832  * CPUs at the next numbered event.
2833  *
2834  * "timespecX" represents host monotonic time. "tscX" represents
2835  * RDTSC value.
2836  *
2837  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2838  *
2839  * 1.  read timespec0,tsc0
2840  * 2.					| timespec1 = timespec0 + N
2841  * 					| tsc1 = tsc0 + M
2842  * 3. transition to guest		| transition to guest
2843  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2844  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2845  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2846  *
2847  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2848  *
2849  * 	- ret0 < ret1
2850  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2851  *		...
2852  *	- 0 < N - M => M < N
2853  *
2854  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2855  * always the case (the difference between two distinct xtime instances
2856  * might be smaller then the difference between corresponding TSC reads,
2857  * when updating guest vcpus pvclock areas).
2858  *
2859  * To avoid that problem, do not allow visibility of distinct
2860  * system_timestamp/tsc_timestamp values simultaneously: use a master
2861  * copy of host monotonic time values. Update that master copy
2862  * in lockstep.
2863  *
2864  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2865  *
2866  */
2867 
2868 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2869 {
2870 #ifdef CONFIG_X86_64
2871 	struct kvm_arch *ka = &kvm->arch;
2872 	int vclock_mode;
2873 	bool host_tsc_clocksource, vcpus_matched;
2874 
2875 	lockdep_assert_held(&kvm->arch.tsc_write_lock);
2876 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2877 			atomic_read(&kvm->online_vcpus));
2878 
2879 	/*
2880 	 * If the host uses TSC clock, then passthrough TSC as stable
2881 	 * to the guest.
2882 	 */
2883 	host_tsc_clocksource = kvm_get_time_and_clockread(
2884 					&ka->master_kernel_ns,
2885 					&ka->master_cycle_now);
2886 
2887 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2888 				&& !ka->backwards_tsc_observed
2889 				&& !ka->boot_vcpu_runs_old_kvmclock;
2890 
2891 	if (ka->use_master_clock)
2892 		atomic_set(&kvm_guest_has_master_clock, 1);
2893 
2894 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2895 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2896 					vcpus_matched);
2897 #endif
2898 }
2899 
2900 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2901 {
2902 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2903 }
2904 
2905 static void __kvm_start_pvclock_update(struct kvm *kvm)
2906 {
2907 	raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2908 	write_seqcount_begin(&kvm->arch.pvclock_sc);
2909 }
2910 
2911 static void kvm_start_pvclock_update(struct kvm *kvm)
2912 {
2913 	kvm_make_mclock_inprogress_request(kvm);
2914 
2915 	/* no guest entries from this point */
2916 	__kvm_start_pvclock_update(kvm);
2917 }
2918 
2919 static void kvm_end_pvclock_update(struct kvm *kvm)
2920 {
2921 	struct kvm_arch *ka = &kvm->arch;
2922 	struct kvm_vcpu *vcpu;
2923 	unsigned long i;
2924 
2925 	write_seqcount_end(&ka->pvclock_sc);
2926 	raw_spin_unlock_irq(&ka->tsc_write_lock);
2927 	kvm_for_each_vcpu(i, vcpu, kvm)
2928 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2929 
2930 	/* guest entries allowed */
2931 	kvm_for_each_vcpu(i, vcpu, kvm)
2932 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2933 }
2934 
2935 static void kvm_update_masterclock(struct kvm *kvm)
2936 {
2937 	kvm_hv_request_tsc_page_update(kvm);
2938 	kvm_start_pvclock_update(kvm);
2939 	pvclock_update_vm_gtod_copy(kvm);
2940 	kvm_end_pvclock_update(kvm);
2941 }
2942 
2943 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc.  */
2944 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2945 {
2946 	struct kvm_arch *ka = &kvm->arch;
2947 	struct pvclock_vcpu_time_info hv_clock;
2948 
2949 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2950 	get_cpu();
2951 
2952 	data->flags = 0;
2953 	if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2954 #ifdef CONFIG_X86_64
2955 		struct timespec64 ts;
2956 
2957 		if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2958 			data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2959 			data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2960 		} else
2961 #endif
2962 		data->host_tsc = rdtsc();
2963 
2964 		data->flags |= KVM_CLOCK_TSC_STABLE;
2965 		hv_clock.tsc_timestamp = ka->master_cycle_now;
2966 		hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2967 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2968 				   &hv_clock.tsc_shift,
2969 				   &hv_clock.tsc_to_system_mul);
2970 		data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2971 	} else {
2972 		data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2973 	}
2974 
2975 	put_cpu();
2976 }
2977 
2978 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2979 {
2980 	struct kvm_arch *ka = &kvm->arch;
2981 	unsigned seq;
2982 
2983 	do {
2984 		seq = read_seqcount_begin(&ka->pvclock_sc);
2985 		__get_kvmclock(kvm, data);
2986 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
2987 }
2988 
2989 u64 get_kvmclock_ns(struct kvm *kvm)
2990 {
2991 	struct kvm_clock_data data;
2992 
2993 	get_kvmclock(kvm, &data);
2994 	return data.clock;
2995 }
2996 
2997 static void kvm_setup_guest_pvclock(struct kvm_vcpu *v,
2998 				    struct gfn_to_pfn_cache *gpc,
2999 				    unsigned int offset)
3000 {
3001 	struct kvm_vcpu_arch *vcpu = &v->arch;
3002 	struct pvclock_vcpu_time_info *guest_hv_clock;
3003 	unsigned long flags;
3004 
3005 	read_lock_irqsave(&gpc->lock, flags);
3006 	while (!kvm_gfn_to_pfn_cache_check(v->kvm, gpc, gpc->gpa,
3007 					   offset + sizeof(*guest_hv_clock))) {
3008 		read_unlock_irqrestore(&gpc->lock, flags);
3009 
3010 		if (kvm_gfn_to_pfn_cache_refresh(v->kvm, gpc, gpc->gpa,
3011 						 offset + sizeof(*guest_hv_clock)))
3012 			return;
3013 
3014 		read_lock_irqsave(&gpc->lock, flags);
3015 	}
3016 
3017 	guest_hv_clock = (void *)(gpc->khva + offset);
3018 
3019 	/*
3020 	 * This VCPU is paused, but it's legal for a guest to read another
3021 	 * VCPU's kvmclock, so we really have to follow the specification where
3022 	 * it says that version is odd if data is being modified, and even after
3023 	 * it is consistent.
3024 	 */
3025 
3026 	guest_hv_clock->version = vcpu->hv_clock.version = (guest_hv_clock->version + 1) | 1;
3027 	smp_wmb();
3028 
3029 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
3030 	vcpu->hv_clock.flags |= (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
3031 
3032 	if (vcpu->pvclock_set_guest_stopped_request) {
3033 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
3034 		vcpu->pvclock_set_guest_stopped_request = false;
3035 	}
3036 
3037 	memcpy(guest_hv_clock, &vcpu->hv_clock, sizeof(*guest_hv_clock));
3038 	smp_wmb();
3039 
3040 	guest_hv_clock->version = ++vcpu->hv_clock.version;
3041 
3042 	mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
3043 	read_unlock_irqrestore(&gpc->lock, flags);
3044 
3045 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
3046 }
3047 
3048 static int kvm_guest_time_update(struct kvm_vcpu *v)
3049 {
3050 	unsigned long flags, tgt_tsc_khz;
3051 	unsigned seq;
3052 	struct kvm_vcpu_arch *vcpu = &v->arch;
3053 	struct kvm_arch *ka = &v->kvm->arch;
3054 	s64 kernel_ns;
3055 	u64 tsc_timestamp, host_tsc;
3056 	u8 pvclock_flags;
3057 	bool use_master_clock;
3058 
3059 	kernel_ns = 0;
3060 	host_tsc = 0;
3061 
3062 	/*
3063 	 * If the host uses TSC clock, then passthrough TSC as stable
3064 	 * to the guest.
3065 	 */
3066 	do {
3067 		seq = read_seqcount_begin(&ka->pvclock_sc);
3068 		use_master_clock = ka->use_master_clock;
3069 		if (use_master_clock) {
3070 			host_tsc = ka->master_cycle_now;
3071 			kernel_ns = ka->master_kernel_ns;
3072 		}
3073 	} while (read_seqcount_retry(&ka->pvclock_sc, seq));
3074 
3075 	/* Keep irq disabled to prevent changes to the clock */
3076 	local_irq_save(flags);
3077 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
3078 	if (unlikely(tgt_tsc_khz == 0)) {
3079 		local_irq_restore(flags);
3080 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3081 		return 1;
3082 	}
3083 	if (!use_master_clock) {
3084 		host_tsc = rdtsc();
3085 		kernel_ns = get_kvmclock_base_ns();
3086 	}
3087 
3088 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3089 
3090 	/*
3091 	 * We may have to catch up the TSC to match elapsed wall clock
3092 	 * time for two reasons, even if kvmclock is used.
3093 	 *   1) CPU could have been running below the maximum TSC rate
3094 	 *   2) Broken TSC compensation resets the base at each VCPU
3095 	 *      entry to avoid unknown leaps of TSC even when running
3096 	 *      again on the same CPU.  This may cause apparent elapsed
3097 	 *      time to disappear, and the guest to stand still or run
3098 	 *	very slowly.
3099 	 */
3100 	if (vcpu->tsc_catchup) {
3101 		u64 tsc = compute_guest_tsc(v, kernel_ns);
3102 		if (tsc > tsc_timestamp) {
3103 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3104 			tsc_timestamp = tsc;
3105 		}
3106 	}
3107 
3108 	local_irq_restore(flags);
3109 
3110 	/* With all the info we got, fill in the values */
3111 
3112 	if (kvm_caps.has_tsc_control)
3113 		tgt_tsc_khz = kvm_scale_tsc(tgt_tsc_khz,
3114 					    v->arch.l1_tsc_scaling_ratio);
3115 
3116 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3117 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3118 				   &vcpu->hv_clock.tsc_shift,
3119 				   &vcpu->hv_clock.tsc_to_system_mul);
3120 		vcpu->hw_tsc_khz = tgt_tsc_khz;
3121 	}
3122 
3123 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3124 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3125 	vcpu->last_guest_tsc = tsc_timestamp;
3126 
3127 	/* If the host uses TSC clocksource, then it is stable */
3128 	pvclock_flags = 0;
3129 	if (use_master_clock)
3130 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3131 
3132 	vcpu->hv_clock.flags = pvclock_flags;
3133 
3134 	if (vcpu->pv_time.active)
3135 		kvm_setup_guest_pvclock(v, &vcpu->pv_time, 0);
3136 	if (vcpu->xen.vcpu_info_cache.active)
3137 		kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_info_cache,
3138 					offsetof(struct compat_vcpu_info, time));
3139 	if (vcpu->xen.vcpu_time_info_cache.active)
3140 		kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_time_info_cache, 0);
3141 	kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3142 	return 0;
3143 }
3144 
3145 /*
3146  * kvmclock updates which are isolated to a given vcpu, such as
3147  * vcpu->cpu migration, should not allow system_timestamp from
3148  * the rest of the vcpus to remain static. Otherwise ntp frequency
3149  * correction applies to one vcpu's system_timestamp but not
3150  * the others.
3151  *
3152  * So in those cases, request a kvmclock update for all vcpus.
3153  * We need to rate-limit these requests though, as they can
3154  * considerably slow guests that have a large number of vcpus.
3155  * The time for a remote vcpu to update its kvmclock is bound
3156  * by the delay we use to rate-limit the updates.
3157  */
3158 
3159 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3160 
3161 static void kvmclock_update_fn(struct work_struct *work)
3162 {
3163 	unsigned long i;
3164 	struct delayed_work *dwork = to_delayed_work(work);
3165 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3166 					   kvmclock_update_work);
3167 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3168 	struct kvm_vcpu *vcpu;
3169 
3170 	kvm_for_each_vcpu(i, vcpu, kvm) {
3171 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3172 		kvm_vcpu_kick(vcpu);
3173 	}
3174 }
3175 
3176 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3177 {
3178 	struct kvm *kvm = v->kvm;
3179 
3180 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3181 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3182 					KVMCLOCK_UPDATE_DELAY);
3183 }
3184 
3185 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3186 
3187 static void kvmclock_sync_fn(struct work_struct *work)
3188 {
3189 	struct delayed_work *dwork = to_delayed_work(work);
3190 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3191 					   kvmclock_sync_work);
3192 	struct kvm *kvm = container_of(ka, struct kvm, arch);
3193 
3194 	if (!kvmclock_periodic_sync)
3195 		return;
3196 
3197 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3198 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3199 					KVMCLOCK_SYNC_PERIOD);
3200 }
3201 
3202 /* These helpers are safe iff @msr is known to be an MCx bank MSR. */
3203 static bool is_mci_control_msr(u32 msr)
3204 {
3205 	return (msr & 3) == 0;
3206 }
3207 static bool is_mci_status_msr(u32 msr)
3208 {
3209 	return (msr & 3) == 1;
3210 }
3211 
3212 /*
3213  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3214  */
3215 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3216 {
3217 	/* McStatusWrEn enabled? */
3218 	if (guest_cpuid_is_amd_or_hygon(vcpu))
3219 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3220 
3221 	return false;
3222 }
3223 
3224 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3225 {
3226 	u64 mcg_cap = vcpu->arch.mcg_cap;
3227 	unsigned bank_num = mcg_cap & 0xff;
3228 	u32 msr = msr_info->index;
3229 	u64 data = msr_info->data;
3230 	u32 offset, last_msr;
3231 
3232 	switch (msr) {
3233 	case MSR_IA32_MCG_STATUS:
3234 		vcpu->arch.mcg_status = data;
3235 		break;
3236 	case MSR_IA32_MCG_CTL:
3237 		if (!(mcg_cap & MCG_CTL_P) &&
3238 		    (data || !msr_info->host_initiated))
3239 			return 1;
3240 		if (data != 0 && data != ~(u64)0)
3241 			return 1;
3242 		vcpu->arch.mcg_ctl = data;
3243 		break;
3244 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3245 		last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3246 		if (msr > last_msr)
3247 			return 1;
3248 
3249 		if (!(mcg_cap & MCG_CMCI_P) && (data || !msr_info->host_initiated))
3250 			return 1;
3251 		/* An attempt to write a 1 to a reserved bit raises #GP */
3252 		if (data & ~(MCI_CTL2_CMCI_EN | MCI_CTL2_CMCI_THRESHOLD_MASK))
3253 			return 1;
3254 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3255 					    last_msr + 1 - MSR_IA32_MC0_CTL2);
3256 		vcpu->arch.mci_ctl2_banks[offset] = data;
3257 		break;
3258 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3259 		last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3260 		if (msr > last_msr)
3261 			return 1;
3262 
3263 		/*
3264 		 * Only 0 or all 1s can be written to IA32_MCi_CTL, all other
3265 		 * values are architecturally undefined.  But, some Linux
3266 		 * kernels clear bit 10 in bank 4 to workaround a BIOS/GART TLB
3267 		 * issue on AMD K8s, allow bit 10 to be clear when setting all
3268 		 * other bits in order to avoid an uncaught #GP in the guest.
3269 		 *
3270 		 * UNIXWARE clears bit 0 of MC1_CTL to ignore correctable,
3271 		 * single-bit ECC data errors.
3272 		 */
3273 		if (is_mci_control_msr(msr) &&
3274 		    data != 0 && (data | (1 << 10) | 1) != ~(u64)0)
3275 			return 1;
3276 
3277 		/*
3278 		 * All CPUs allow writing 0 to MCi_STATUS MSRs to clear the MSR.
3279 		 * AMD-based CPUs allow non-zero values, but if and only if
3280 		 * HWCR[McStatusWrEn] is set.
3281 		 */
3282 		if (!msr_info->host_initiated && is_mci_status_msr(msr) &&
3283 		    data != 0 && !can_set_mci_status(vcpu))
3284 			return 1;
3285 
3286 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3287 					    last_msr + 1 - MSR_IA32_MC0_CTL);
3288 		vcpu->arch.mce_banks[offset] = data;
3289 		break;
3290 	default:
3291 		return 1;
3292 	}
3293 	return 0;
3294 }
3295 
3296 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3297 {
3298 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3299 
3300 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3301 }
3302 
3303 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3304 {
3305 	gpa_t gpa = data & ~0x3f;
3306 
3307 	/* Bits 4:5 are reserved, Should be zero */
3308 	if (data & 0x30)
3309 		return 1;
3310 
3311 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3312 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3313 		return 1;
3314 
3315 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3316 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3317 		return 1;
3318 
3319 	if (!lapic_in_kernel(vcpu))
3320 		return data ? 1 : 0;
3321 
3322 	vcpu->arch.apf.msr_en_val = data;
3323 
3324 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3325 		kvm_clear_async_pf_completion_queue(vcpu);
3326 		kvm_async_pf_hash_reset(vcpu);
3327 		return 0;
3328 	}
3329 
3330 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3331 					sizeof(u64)))
3332 		return 1;
3333 
3334 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3335 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3336 
3337 	kvm_async_pf_wakeup_all(vcpu);
3338 
3339 	return 0;
3340 }
3341 
3342 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3343 {
3344 	/* Bits 8-63 are reserved */
3345 	if (data >> 8)
3346 		return 1;
3347 
3348 	if (!lapic_in_kernel(vcpu))
3349 		return 1;
3350 
3351 	vcpu->arch.apf.msr_int_val = data;
3352 
3353 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3354 
3355 	return 0;
3356 }
3357 
3358 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3359 {
3360 	kvm_gfn_to_pfn_cache_destroy(vcpu->kvm, &vcpu->arch.pv_time);
3361 	vcpu->arch.time = 0;
3362 }
3363 
3364 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3365 {
3366 	++vcpu->stat.tlb_flush;
3367 	static_call(kvm_x86_flush_tlb_all)(vcpu);
3368 }
3369 
3370 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3371 {
3372 	++vcpu->stat.tlb_flush;
3373 
3374 	if (!tdp_enabled) {
3375 		/*
3376 		 * A TLB flush on behalf of the guest is equivalent to
3377 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3378 		 * a forced sync of the shadow page tables.  Ensure all the
3379 		 * roots are synced and the guest TLB in hardware is clean.
3380 		 */
3381 		kvm_mmu_sync_roots(vcpu);
3382 		kvm_mmu_sync_prev_roots(vcpu);
3383 	}
3384 
3385 	static_call(kvm_x86_flush_tlb_guest)(vcpu);
3386 }
3387 
3388 
3389 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3390 {
3391 	++vcpu->stat.tlb_flush;
3392 	static_call(kvm_x86_flush_tlb_current)(vcpu);
3393 }
3394 
3395 /*
3396  * Service "local" TLB flush requests, which are specific to the current MMU
3397  * context.  In addition to the generic event handling in vcpu_enter_guest(),
3398  * TLB flushes that are targeted at an MMU context also need to be serviced
3399  * prior before nested VM-Enter/VM-Exit.
3400  */
3401 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3402 {
3403 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3404 		kvm_vcpu_flush_tlb_current(vcpu);
3405 
3406 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3407 		kvm_vcpu_flush_tlb_guest(vcpu);
3408 }
3409 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3410 
3411 static void record_steal_time(struct kvm_vcpu *vcpu)
3412 {
3413 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3414 	struct kvm_steal_time __user *st;
3415 	struct kvm_memslots *slots;
3416 	u64 steal;
3417 	u32 version;
3418 
3419 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3420 		kvm_xen_runstate_set_running(vcpu);
3421 		return;
3422 	}
3423 
3424 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3425 		return;
3426 
3427 	if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3428 		return;
3429 
3430 	slots = kvm_memslots(vcpu->kvm);
3431 
3432 	if (unlikely(slots->generation != ghc->generation ||
3433 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3434 		gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3435 
3436 		/* We rely on the fact that it fits in a single page. */
3437 		BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3438 
3439 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3440 		    kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3441 			return;
3442 	}
3443 
3444 	st = (struct kvm_steal_time __user *)ghc->hva;
3445 	/*
3446 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3447 	 * expensive IPIs.
3448 	 */
3449 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3450 		u8 st_preempted = 0;
3451 		int err = -EFAULT;
3452 
3453 		if (!user_access_begin(st, sizeof(*st)))
3454 			return;
3455 
3456 		asm volatile("1: xchgb %0, %2\n"
3457 			     "xor %1, %1\n"
3458 			     "2:\n"
3459 			     _ASM_EXTABLE_UA(1b, 2b)
3460 			     : "+q" (st_preempted),
3461 			       "+&r" (err),
3462 			       "+m" (st->preempted));
3463 		if (err)
3464 			goto out;
3465 
3466 		user_access_end();
3467 
3468 		vcpu->arch.st.preempted = 0;
3469 
3470 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3471 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3472 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3473 			kvm_vcpu_flush_tlb_guest(vcpu);
3474 
3475 		if (!user_access_begin(st, sizeof(*st)))
3476 			goto dirty;
3477 	} else {
3478 		if (!user_access_begin(st, sizeof(*st)))
3479 			return;
3480 
3481 		unsafe_put_user(0, &st->preempted, out);
3482 		vcpu->arch.st.preempted = 0;
3483 	}
3484 
3485 	unsafe_get_user(version, &st->version, out);
3486 	if (version & 1)
3487 		version += 1;  /* first time write, random junk */
3488 
3489 	version += 1;
3490 	unsafe_put_user(version, &st->version, out);
3491 
3492 	smp_wmb();
3493 
3494 	unsafe_get_user(steal, &st->steal, out);
3495 	steal += current->sched_info.run_delay -
3496 		vcpu->arch.st.last_steal;
3497 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3498 	unsafe_put_user(steal, &st->steal, out);
3499 
3500 	version += 1;
3501 	unsafe_put_user(version, &st->version, out);
3502 
3503  out:
3504 	user_access_end();
3505  dirty:
3506 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3507 }
3508 
3509 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3510 {
3511 	bool pr = false;
3512 	u32 msr = msr_info->index;
3513 	u64 data = msr_info->data;
3514 
3515 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3516 		return kvm_xen_write_hypercall_page(vcpu, data);
3517 
3518 	switch (msr) {
3519 	case MSR_AMD64_NB_CFG:
3520 	case MSR_IA32_UCODE_WRITE:
3521 	case MSR_VM_HSAVE_PA:
3522 	case MSR_AMD64_PATCH_LOADER:
3523 	case MSR_AMD64_BU_CFG2:
3524 	case MSR_AMD64_DC_CFG:
3525 	case MSR_F15H_EX_CFG:
3526 		break;
3527 
3528 	case MSR_IA32_UCODE_REV:
3529 		if (msr_info->host_initiated)
3530 			vcpu->arch.microcode_version = data;
3531 		break;
3532 	case MSR_IA32_ARCH_CAPABILITIES:
3533 		if (!msr_info->host_initiated)
3534 			return 1;
3535 		vcpu->arch.arch_capabilities = data;
3536 		break;
3537 	case MSR_IA32_PERF_CAPABILITIES: {
3538 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3539 
3540 		if (!msr_info->host_initiated)
3541 			return 1;
3542 		if (kvm_get_msr_feature(&msr_ent))
3543 			return 1;
3544 		if (data & ~msr_ent.data)
3545 			return 1;
3546 
3547 		vcpu->arch.perf_capabilities = data;
3548 
3549 		return 0;
3550 		}
3551 	case MSR_EFER:
3552 		return set_efer(vcpu, msr_info);
3553 	case MSR_K7_HWCR:
3554 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3555 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3556 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3557 
3558 		/* Handle McStatusWrEn */
3559 		if (data == BIT_ULL(18)) {
3560 			vcpu->arch.msr_hwcr = data;
3561 		} else if (data != 0) {
3562 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3563 				    data);
3564 			return 1;
3565 		}
3566 		break;
3567 	case MSR_FAM10H_MMIO_CONF_BASE:
3568 		if (data != 0) {
3569 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3570 				    "0x%llx\n", data);
3571 			return 1;
3572 		}
3573 		break;
3574 	case 0x200 ... MSR_IA32_MC0_CTL2 - 1:
3575 	case MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) ... 0x2ff:
3576 		return kvm_mtrr_set_msr(vcpu, msr, data);
3577 	case MSR_IA32_APICBASE:
3578 		return kvm_set_apic_base(vcpu, msr_info);
3579 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3580 		return kvm_x2apic_msr_write(vcpu, msr, data);
3581 	case MSR_IA32_TSC_DEADLINE:
3582 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3583 		break;
3584 	case MSR_IA32_TSC_ADJUST:
3585 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3586 			if (!msr_info->host_initiated) {
3587 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3588 				adjust_tsc_offset_guest(vcpu, adj);
3589 				/* Before back to guest, tsc_timestamp must be adjusted
3590 				 * as well, otherwise guest's percpu pvclock time could jump.
3591 				 */
3592 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3593 			}
3594 			vcpu->arch.ia32_tsc_adjust_msr = data;
3595 		}
3596 		break;
3597 	case MSR_IA32_MISC_ENABLE: {
3598 		u64 old_val = vcpu->arch.ia32_misc_enable_msr;
3599 
3600 		if (!msr_info->host_initiated) {
3601 			/* RO bits */
3602 			if ((old_val ^ data) & MSR_IA32_MISC_ENABLE_PMU_RO_MASK)
3603 				return 1;
3604 
3605 			/* R bits, i.e. writes are ignored, but don't fault. */
3606 			data = data & ~MSR_IA32_MISC_ENABLE_EMON;
3607 			data |= old_val & MSR_IA32_MISC_ENABLE_EMON;
3608 		}
3609 
3610 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3611 		    ((old_val ^ data)  & MSR_IA32_MISC_ENABLE_MWAIT)) {
3612 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3613 				return 1;
3614 			vcpu->arch.ia32_misc_enable_msr = data;
3615 			kvm_update_cpuid_runtime(vcpu);
3616 		} else {
3617 			vcpu->arch.ia32_misc_enable_msr = data;
3618 		}
3619 		break;
3620 	}
3621 	case MSR_IA32_SMBASE:
3622 		if (!msr_info->host_initiated)
3623 			return 1;
3624 		vcpu->arch.smbase = data;
3625 		break;
3626 	case MSR_IA32_POWER_CTL:
3627 		vcpu->arch.msr_ia32_power_ctl = data;
3628 		break;
3629 	case MSR_IA32_TSC:
3630 		if (msr_info->host_initiated) {
3631 			kvm_synchronize_tsc(vcpu, data);
3632 		} else {
3633 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3634 			adjust_tsc_offset_guest(vcpu, adj);
3635 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3636 		}
3637 		break;
3638 	case MSR_IA32_XSS:
3639 		if (!msr_info->host_initiated &&
3640 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3641 			return 1;
3642 		/*
3643 		 * KVM supports exposing PT to the guest, but does not support
3644 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3645 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3646 		 */
3647 		if (data & ~kvm_caps.supported_xss)
3648 			return 1;
3649 		vcpu->arch.ia32_xss = data;
3650 		kvm_update_cpuid_runtime(vcpu);
3651 		break;
3652 	case MSR_SMI_COUNT:
3653 		if (!msr_info->host_initiated)
3654 			return 1;
3655 		vcpu->arch.smi_count = data;
3656 		break;
3657 	case MSR_KVM_WALL_CLOCK_NEW:
3658 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3659 			return 1;
3660 
3661 		vcpu->kvm->arch.wall_clock = data;
3662 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3663 		break;
3664 	case MSR_KVM_WALL_CLOCK:
3665 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3666 			return 1;
3667 
3668 		vcpu->kvm->arch.wall_clock = data;
3669 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3670 		break;
3671 	case MSR_KVM_SYSTEM_TIME_NEW:
3672 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3673 			return 1;
3674 
3675 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3676 		break;
3677 	case MSR_KVM_SYSTEM_TIME:
3678 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3679 			return 1;
3680 
3681 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3682 		break;
3683 	case MSR_KVM_ASYNC_PF_EN:
3684 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3685 			return 1;
3686 
3687 		if (kvm_pv_enable_async_pf(vcpu, data))
3688 			return 1;
3689 		break;
3690 	case MSR_KVM_ASYNC_PF_INT:
3691 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3692 			return 1;
3693 
3694 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3695 			return 1;
3696 		break;
3697 	case MSR_KVM_ASYNC_PF_ACK:
3698 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3699 			return 1;
3700 		if (data & 0x1) {
3701 			vcpu->arch.apf.pageready_pending = false;
3702 			kvm_check_async_pf_completion(vcpu);
3703 		}
3704 		break;
3705 	case MSR_KVM_STEAL_TIME:
3706 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3707 			return 1;
3708 
3709 		if (unlikely(!sched_info_on()))
3710 			return 1;
3711 
3712 		if (data & KVM_STEAL_RESERVED_MASK)
3713 			return 1;
3714 
3715 		vcpu->arch.st.msr_val = data;
3716 
3717 		if (!(data & KVM_MSR_ENABLED))
3718 			break;
3719 
3720 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3721 
3722 		break;
3723 	case MSR_KVM_PV_EOI_EN:
3724 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3725 			return 1;
3726 
3727 		if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3728 			return 1;
3729 		break;
3730 
3731 	case MSR_KVM_POLL_CONTROL:
3732 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3733 			return 1;
3734 
3735 		/* only enable bit supported */
3736 		if (data & (-1ULL << 1))
3737 			return 1;
3738 
3739 		vcpu->arch.msr_kvm_poll_control = data;
3740 		break;
3741 
3742 	case MSR_IA32_MCG_CTL:
3743 	case MSR_IA32_MCG_STATUS:
3744 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3745 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3746 		return set_msr_mce(vcpu, msr_info);
3747 
3748 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3749 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3750 		pr = true;
3751 		fallthrough;
3752 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3753 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3754 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3755 			return kvm_pmu_set_msr(vcpu, msr_info);
3756 
3757 		if (pr || data != 0)
3758 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3759 				    "0x%x data 0x%llx\n", msr, data);
3760 		break;
3761 	case MSR_K7_CLK_CTL:
3762 		/*
3763 		 * Ignore all writes to this no longer documented MSR.
3764 		 * Writes are only relevant for old K7 processors,
3765 		 * all pre-dating SVM, but a recommended workaround from
3766 		 * AMD for these chips. It is possible to specify the
3767 		 * affected processor models on the command line, hence
3768 		 * the need to ignore the workaround.
3769 		 */
3770 		break;
3771 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3772 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3773 	case HV_X64_MSR_SYNDBG_OPTIONS:
3774 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3775 	case HV_X64_MSR_CRASH_CTL:
3776 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3777 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3778 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3779 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3780 		return kvm_hv_set_msr_common(vcpu, msr, data,
3781 					     msr_info->host_initiated);
3782 	case MSR_IA32_BBL_CR_CTL3:
3783 		/* Drop writes to this legacy MSR -- see rdmsr
3784 		 * counterpart for further detail.
3785 		 */
3786 		if (report_ignored_msrs)
3787 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3788 				msr, data);
3789 		break;
3790 	case MSR_AMD64_OSVW_ID_LENGTH:
3791 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3792 			return 1;
3793 		vcpu->arch.osvw.length = data;
3794 		break;
3795 	case MSR_AMD64_OSVW_STATUS:
3796 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3797 			return 1;
3798 		vcpu->arch.osvw.status = data;
3799 		break;
3800 	case MSR_PLATFORM_INFO:
3801 		if (!msr_info->host_initiated ||
3802 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3803 		     cpuid_fault_enabled(vcpu)))
3804 			return 1;
3805 		vcpu->arch.msr_platform_info = data;
3806 		break;
3807 	case MSR_MISC_FEATURES_ENABLES:
3808 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3809 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3810 		     !supports_cpuid_fault(vcpu)))
3811 			return 1;
3812 		vcpu->arch.msr_misc_features_enables = data;
3813 		break;
3814 #ifdef CONFIG_X86_64
3815 	case MSR_IA32_XFD:
3816 		if (!msr_info->host_initiated &&
3817 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3818 			return 1;
3819 
3820 		if (data & ~kvm_guest_supported_xfd(vcpu))
3821 			return 1;
3822 
3823 		fpu_update_guest_xfd(&vcpu->arch.guest_fpu, data);
3824 		break;
3825 	case MSR_IA32_XFD_ERR:
3826 		if (!msr_info->host_initiated &&
3827 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3828 			return 1;
3829 
3830 		if (data & ~kvm_guest_supported_xfd(vcpu))
3831 			return 1;
3832 
3833 		vcpu->arch.guest_fpu.xfd_err = data;
3834 		break;
3835 #endif
3836 	case MSR_IA32_PEBS_ENABLE:
3837 	case MSR_IA32_DS_AREA:
3838 	case MSR_PEBS_DATA_CFG:
3839 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3840 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3841 			return kvm_pmu_set_msr(vcpu, msr_info);
3842 		/*
3843 		 * Userspace is allowed to write '0' to MSRs that KVM reports
3844 		 * as to-be-saved, even if an MSRs isn't fully supported.
3845 		 */
3846 		return !msr_info->host_initiated || data;
3847 	default:
3848 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3849 			return kvm_pmu_set_msr(vcpu, msr_info);
3850 		return KVM_MSR_RET_INVALID;
3851 	}
3852 	return 0;
3853 }
3854 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3855 
3856 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3857 {
3858 	u64 data;
3859 	u64 mcg_cap = vcpu->arch.mcg_cap;
3860 	unsigned bank_num = mcg_cap & 0xff;
3861 	u32 offset, last_msr;
3862 
3863 	switch (msr) {
3864 	case MSR_IA32_P5_MC_ADDR:
3865 	case MSR_IA32_P5_MC_TYPE:
3866 		data = 0;
3867 		break;
3868 	case MSR_IA32_MCG_CAP:
3869 		data = vcpu->arch.mcg_cap;
3870 		break;
3871 	case MSR_IA32_MCG_CTL:
3872 		if (!(mcg_cap & MCG_CTL_P) && !host)
3873 			return 1;
3874 		data = vcpu->arch.mcg_ctl;
3875 		break;
3876 	case MSR_IA32_MCG_STATUS:
3877 		data = vcpu->arch.mcg_status;
3878 		break;
3879 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3880 		last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3881 		if (msr > last_msr)
3882 			return 1;
3883 
3884 		if (!(mcg_cap & MCG_CMCI_P) && !host)
3885 			return 1;
3886 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3887 					    last_msr + 1 - MSR_IA32_MC0_CTL2);
3888 		data = vcpu->arch.mci_ctl2_banks[offset];
3889 		break;
3890 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3891 		last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3892 		if (msr > last_msr)
3893 			return 1;
3894 
3895 		offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3896 					    last_msr + 1 - MSR_IA32_MC0_CTL);
3897 		data = vcpu->arch.mce_banks[offset];
3898 		break;
3899 	default:
3900 		return 1;
3901 	}
3902 	*pdata = data;
3903 	return 0;
3904 }
3905 
3906 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3907 {
3908 	switch (msr_info->index) {
3909 	case MSR_IA32_PLATFORM_ID:
3910 	case MSR_IA32_EBL_CR_POWERON:
3911 	case MSR_IA32_LASTBRANCHFROMIP:
3912 	case MSR_IA32_LASTBRANCHTOIP:
3913 	case MSR_IA32_LASTINTFROMIP:
3914 	case MSR_IA32_LASTINTTOIP:
3915 	case MSR_AMD64_SYSCFG:
3916 	case MSR_K8_TSEG_ADDR:
3917 	case MSR_K8_TSEG_MASK:
3918 	case MSR_VM_HSAVE_PA:
3919 	case MSR_K8_INT_PENDING_MSG:
3920 	case MSR_AMD64_NB_CFG:
3921 	case MSR_FAM10H_MMIO_CONF_BASE:
3922 	case MSR_AMD64_BU_CFG2:
3923 	case MSR_IA32_PERF_CTL:
3924 	case MSR_AMD64_DC_CFG:
3925 	case MSR_F15H_EX_CFG:
3926 	/*
3927 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3928 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3929 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3930 	 * so for existing CPU-specific MSRs.
3931 	 */
3932 	case MSR_RAPL_POWER_UNIT:
3933 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3934 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3935 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3936 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3937 		msr_info->data = 0;
3938 		break;
3939 	case MSR_IA32_PEBS_ENABLE:
3940 	case MSR_IA32_DS_AREA:
3941 	case MSR_PEBS_DATA_CFG:
3942 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3943 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3944 			return kvm_pmu_get_msr(vcpu, msr_info);
3945 		/*
3946 		 * Userspace is allowed to read MSRs that KVM reports as
3947 		 * to-be-saved, even if an MSR isn't fully supported.
3948 		 */
3949 		if (!msr_info->host_initiated)
3950 			return 1;
3951 		msr_info->data = 0;
3952 		break;
3953 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3954 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3955 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3956 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3957 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3958 			return kvm_pmu_get_msr(vcpu, msr_info);
3959 		msr_info->data = 0;
3960 		break;
3961 	case MSR_IA32_UCODE_REV:
3962 		msr_info->data = vcpu->arch.microcode_version;
3963 		break;
3964 	case MSR_IA32_ARCH_CAPABILITIES:
3965 		if (!msr_info->host_initiated &&
3966 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3967 			return 1;
3968 		msr_info->data = vcpu->arch.arch_capabilities;
3969 		break;
3970 	case MSR_IA32_PERF_CAPABILITIES:
3971 		if (!msr_info->host_initiated &&
3972 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3973 			return 1;
3974 		msr_info->data = vcpu->arch.perf_capabilities;
3975 		break;
3976 	case MSR_IA32_POWER_CTL:
3977 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3978 		break;
3979 	case MSR_IA32_TSC: {
3980 		/*
3981 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3982 		 * even when not intercepted. AMD manual doesn't explicitly
3983 		 * state this but appears to behave the same.
3984 		 *
3985 		 * On userspace reads and writes, however, we unconditionally
3986 		 * return L1's TSC value to ensure backwards-compatible
3987 		 * behavior for migration.
3988 		 */
3989 		u64 offset, ratio;
3990 
3991 		if (msr_info->host_initiated) {
3992 			offset = vcpu->arch.l1_tsc_offset;
3993 			ratio = vcpu->arch.l1_tsc_scaling_ratio;
3994 		} else {
3995 			offset = vcpu->arch.tsc_offset;
3996 			ratio = vcpu->arch.tsc_scaling_ratio;
3997 		}
3998 
3999 		msr_info->data = kvm_scale_tsc(rdtsc(), ratio) + offset;
4000 		break;
4001 	}
4002 	case MSR_MTRRcap:
4003 	case 0x200 ... MSR_IA32_MC0_CTL2 - 1:
4004 	case MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) ... 0x2ff:
4005 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
4006 	case 0xcd: /* fsb frequency */
4007 		msr_info->data = 3;
4008 		break;
4009 		/*
4010 		 * MSR_EBC_FREQUENCY_ID
4011 		 * Conservative value valid for even the basic CPU models.
4012 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
4013 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
4014 		 * and 266MHz for model 3, or 4. Set Core Clock
4015 		 * Frequency to System Bus Frequency Ratio to 1 (bits
4016 		 * 31:24) even though these are only valid for CPU
4017 		 * models > 2, however guests may end up dividing or
4018 		 * multiplying by zero otherwise.
4019 		 */
4020 	case MSR_EBC_FREQUENCY_ID:
4021 		msr_info->data = 1 << 24;
4022 		break;
4023 	case MSR_IA32_APICBASE:
4024 		msr_info->data = kvm_get_apic_base(vcpu);
4025 		break;
4026 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
4027 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
4028 	case MSR_IA32_TSC_DEADLINE:
4029 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
4030 		break;
4031 	case MSR_IA32_TSC_ADJUST:
4032 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
4033 		break;
4034 	case MSR_IA32_MISC_ENABLE:
4035 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
4036 		break;
4037 	case MSR_IA32_SMBASE:
4038 		if (!msr_info->host_initiated)
4039 			return 1;
4040 		msr_info->data = vcpu->arch.smbase;
4041 		break;
4042 	case MSR_SMI_COUNT:
4043 		msr_info->data = vcpu->arch.smi_count;
4044 		break;
4045 	case MSR_IA32_PERF_STATUS:
4046 		/* TSC increment by tick */
4047 		msr_info->data = 1000ULL;
4048 		/* CPU multiplier */
4049 		msr_info->data |= (((uint64_t)4ULL) << 40);
4050 		break;
4051 	case MSR_EFER:
4052 		msr_info->data = vcpu->arch.efer;
4053 		break;
4054 	case MSR_KVM_WALL_CLOCK:
4055 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4056 			return 1;
4057 
4058 		msr_info->data = vcpu->kvm->arch.wall_clock;
4059 		break;
4060 	case MSR_KVM_WALL_CLOCK_NEW:
4061 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4062 			return 1;
4063 
4064 		msr_info->data = vcpu->kvm->arch.wall_clock;
4065 		break;
4066 	case MSR_KVM_SYSTEM_TIME:
4067 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4068 			return 1;
4069 
4070 		msr_info->data = vcpu->arch.time;
4071 		break;
4072 	case MSR_KVM_SYSTEM_TIME_NEW:
4073 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4074 			return 1;
4075 
4076 		msr_info->data = vcpu->arch.time;
4077 		break;
4078 	case MSR_KVM_ASYNC_PF_EN:
4079 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
4080 			return 1;
4081 
4082 		msr_info->data = vcpu->arch.apf.msr_en_val;
4083 		break;
4084 	case MSR_KVM_ASYNC_PF_INT:
4085 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4086 			return 1;
4087 
4088 		msr_info->data = vcpu->arch.apf.msr_int_val;
4089 		break;
4090 	case MSR_KVM_ASYNC_PF_ACK:
4091 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4092 			return 1;
4093 
4094 		msr_info->data = 0;
4095 		break;
4096 	case MSR_KVM_STEAL_TIME:
4097 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
4098 			return 1;
4099 
4100 		msr_info->data = vcpu->arch.st.msr_val;
4101 		break;
4102 	case MSR_KVM_PV_EOI_EN:
4103 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
4104 			return 1;
4105 
4106 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
4107 		break;
4108 	case MSR_KVM_POLL_CONTROL:
4109 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
4110 			return 1;
4111 
4112 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
4113 		break;
4114 	case MSR_IA32_P5_MC_ADDR:
4115 	case MSR_IA32_P5_MC_TYPE:
4116 	case MSR_IA32_MCG_CAP:
4117 	case MSR_IA32_MCG_CTL:
4118 	case MSR_IA32_MCG_STATUS:
4119 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
4120 	case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
4121 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
4122 				   msr_info->host_initiated);
4123 	case MSR_IA32_XSS:
4124 		if (!msr_info->host_initiated &&
4125 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
4126 			return 1;
4127 		msr_info->data = vcpu->arch.ia32_xss;
4128 		break;
4129 	case MSR_K7_CLK_CTL:
4130 		/*
4131 		 * Provide expected ramp-up count for K7. All other
4132 		 * are set to zero, indicating minimum divisors for
4133 		 * every field.
4134 		 *
4135 		 * This prevents guest kernels on AMD host with CPU
4136 		 * type 6, model 8 and higher from exploding due to
4137 		 * the rdmsr failing.
4138 		 */
4139 		msr_info->data = 0x20000000;
4140 		break;
4141 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
4142 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
4143 	case HV_X64_MSR_SYNDBG_OPTIONS:
4144 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
4145 	case HV_X64_MSR_CRASH_CTL:
4146 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
4147 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
4148 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
4149 	case HV_X64_MSR_TSC_EMULATION_STATUS:
4150 		return kvm_hv_get_msr_common(vcpu,
4151 					     msr_info->index, &msr_info->data,
4152 					     msr_info->host_initiated);
4153 	case MSR_IA32_BBL_CR_CTL3:
4154 		/* This legacy MSR exists but isn't fully documented in current
4155 		 * silicon.  It is however accessed by winxp in very narrow
4156 		 * scenarios where it sets bit #19, itself documented as
4157 		 * a "reserved" bit.  Best effort attempt to source coherent
4158 		 * read data here should the balance of the register be
4159 		 * interpreted by the guest:
4160 		 *
4161 		 * L2 cache control register 3: 64GB range, 256KB size,
4162 		 * enabled, latency 0x1, configured
4163 		 */
4164 		msr_info->data = 0xbe702111;
4165 		break;
4166 	case MSR_AMD64_OSVW_ID_LENGTH:
4167 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4168 			return 1;
4169 		msr_info->data = vcpu->arch.osvw.length;
4170 		break;
4171 	case MSR_AMD64_OSVW_STATUS:
4172 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4173 			return 1;
4174 		msr_info->data = vcpu->arch.osvw.status;
4175 		break;
4176 	case MSR_PLATFORM_INFO:
4177 		if (!msr_info->host_initiated &&
4178 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4179 			return 1;
4180 		msr_info->data = vcpu->arch.msr_platform_info;
4181 		break;
4182 	case MSR_MISC_FEATURES_ENABLES:
4183 		msr_info->data = vcpu->arch.msr_misc_features_enables;
4184 		break;
4185 	case MSR_K7_HWCR:
4186 		msr_info->data = vcpu->arch.msr_hwcr;
4187 		break;
4188 #ifdef CONFIG_X86_64
4189 	case MSR_IA32_XFD:
4190 		if (!msr_info->host_initiated &&
4191 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4192 			return 1;
4193 
4194 		msr_info->data = vcpu->arch.guest_fpu.fpstate->xfd;
4195 		break;
4196 	case MSR_IA32_XFD_ERR:
4197 		if (!msr_info->host_initiated &&
4198 		    !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4199 			return 1;
4200 
4201 		msr_info->data = vcpu->arch.guest_fpu.xfd_err;
4202 		break;
4203 #endif
4204 	default:
4205 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4206 			return kvm_pmu_get_msr(vcpu, msr_info);
4207 		return KVM_MSR_RET_INVALID;
4208 	}
4209 	return 0;
4210 }
4211 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4212 
4213 /*
4214  * Read or write a bunch of msrs. All parameters are kernel addresses.
4215  *
4216  * @return number of msrs set successfully.
4217  */
4218 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4219 		    struct kvm_msr_entry *entries,
4220 		    int (*do_msr)(struct kvm_vcpu *vcpu,
4221 				  unsigned index, u64 *data))
4222 {
4223 	int i;
4224 
4225 	for (i = 0; i < msrs->nmsrs; ++i)
4226 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
4227 			break;
4228 
4229 	return i;
4230 }
4231 
4232 /*
4233  * Read or write a bunch of msrs. Parameters are user addresses.
4234  *
4235  * @return number of msrs set successfully.
4236  */
4237 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4238 		  int (*do_msr)(struct kvm_vcpu *vcpu,
4239 				unsigned index, u64 *data),
4240 		  int writeback)
4241 {
4242 	struct kvm_msrs msrs;
4243 	struct kvm_msr_entry *entries;
4244 	int r, n;
4245 	unsigned size;
4246 
4247 	r = -EFAULT;
4248 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4249 		goto out;
4250 
4251 	r = -E2BIG;
4252 	if (msrs.nmsrs >= MAX_IO_MSRS)
4253 		goto out;
4254 
4255 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4256 	entries = memdup_user(user_msrs->entries, size);
4257 	if (IS_ERR(entries)) {
4258 		r = PTR_ERR(entries);
4259 		goto out;
4260 	}
4261 
4262 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4263 	if (r < 0)
4264 		goto out_free;
4265 
4266 	r = -EFAULT;
4267 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
4268 		goto out_free;
4269 
4270 	r = n;
4271 
4272 out_free:
4273 	kfree(entries);
4274 out:
4275 	return r;
4276 }
4277 
4278 static inline bool kvm_can_mwait_in_guest(void)
4279 {
4280 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
4281 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
4282 		boot_cpu_has(X86_FEATURE_ARAT);
4283 }
4284 
4285 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4286 					    struct kvm_cpuid2 __user *cpuid_arg)
4287 {
4288 	struct kvm_cpuid2 cpuid;
4289 	int r;
4290 
4291 	r = -EFAULT;
4292 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4293 		return r;
4294 
4295 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4296 	if (r)
4297 		return r;
4298 
4299 	r = -EFAULT;
4300 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4301 		return r;
4302 
4303 	return 0;
4304 }
4305 
4306 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4307 {
4308 	int r = 0;
4309 
4310 	switch (ext) {
4311 	case KVM_CAP_IRQCHIP:
4312 	case KVM_CAP_HLT:
4313 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4314 	case KVM_CAP_SET_TSS_ADDR:
4315 	case KVM_CAP_EXT_CPUID:
4316 	case KVM_CAP_EXT_EMUL_CPUID:
4317 	case KVM_CAP_CLOCKSOURCE:
4318 	case KVM_CAP_PIT:
4319 	case KVM_CAP_NOP_IO_DELAY:
4320 	case KVM_CAP_MP_STATE:
4321 	case KVM_CAP_SYNC_MMU:
4322 	case KVM_CAP_USER_NMI:
4323 	case KVM_CAP_REINJECT_CONTROL:
4324 	case KVM_CAP_IRQ_INJECT_STATUS:
4325 	case KVM_CAP_IOEVENTFD:
4326 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
4327 	case KVM_CAP_PIT2:
4328 	case KVM_CAP_PIT_STATE2:
4329 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4330 	case KVM_CAP_VCPU_EVENTS:
4331 	case KVM_CAP_HYPERV:
4332 	case KVM_CAP_HYPERV_VAPIC:
4333 	case KVM_CAP_HYPERV_SPIN:
4334 	case KVM_CAP_HYPERV_SYNIC:
4335 	case KVM_CAP_HYPERV_SYNIC2:
4336 	case KVM_CAP_HYPERV_VP_INDEX:
4337 	case KVM_CAP_HYPERV_EVENTFD:
4338 	case KVM_CAP_HYPERV_TLBFLUSH:
4339 	case KVM_CAP_HYPERV_SEND_IPI:
4340 	case KVM_CAP_HYPERV_CPUID:
4341 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
4342 	case KVM_CAP_SYS_HYPERV_CPUID:
4343 	case KVM_CAP_PCI_SEGMENT:
4344 	case KVM_CAP_DEBUGREGS:
4345 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
4346 	case KVM_CAP_XSAVE:
4347 	case KVM_CAP_ASYNC_PF:
4348 	case KVM_CAP_ASYNC_PF_INT:
4349 	case KVM_CAP_GET_TSC_KHZ:
4350 	case KVM_CAP_KVMCLOCK_CTRL:
4351 	case KVM_CAP_READONLY_MEM:
4352 	case KVM_CAP_HYPERV_TIME:
4353 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4354 	case KVM_CAP_TSC_DEADLINE_TIMER:
4355 	case KVM_CAP_DISABLE_QUIRKS:
4356 	case KVM_CAP_SET_BOOT_CPU_ID:
4357  	case KVM_CAP_SPLIT_IRQCHIP:
4358 	case KVM_CAP_IMMEDIATE_EXIT:
4359 	case KVM_CAP_PMU_EVENT_FILTER:
4360 	case KVM_CAP_GET_MSR_FEATURES:
4361 	case KVM_CAP_MSR_PLATFORM_INFO:
4362 	case KVM_CAP_EXCEPTION_PAYLOAD:
4363 	case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
4364 	case KVM_CAP_SET_GUEST_DEBUG:
4365 	case KVM_CAP_LAST_CPU:
4366 	case KVM_CAP_X86_USER_SPACE_MSR:
4367 	case KVM_CAP_X86_MSR_FILTER:
4368 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4369 #ifdef CONFIG_X86_SGX_KVM
4370 	case KVM_CAP_SGX_ATTRIBUTE:
4371 #endif
4372 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4373 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4374 	case KVM_CAP_SREGS2:
4375 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4376 	case KVM_CAP_VCPU_ATTRIBUTES:
4377 	case KVM_CAP_SYS_ATTRIBUTES:
4378 	case KVM_CAP_VAPIC:
4379 	case KVM_CAP_ENABLE_CAP:
4380 	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
4381 		r = 1;
4382 		break;
4383 	case KVM_CAP_EXIT_HYPERCALL:
4384 		r = KVM_EXIT_HYPERCALL_VALID_MASK;
4385 		break;
4386 	case KVM_CAP_SET_GUEST_DEBUG2:
4387 		return KVM_GUESTDBG_VALID_MASK;
4388 #ifdef CONFIG_KVM_XEN
4389 	case KVM_CAP_XEN_HVM:
4390 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4391 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4392 		    KVM_XEN_HVM_CONFIG_SHARED_INFO |
4393 		    KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL |
4394 		    KVM_XEN_HVM_CONFIG_EVTCHN_SEND;
4395 		if (sched_info_on())
4396 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4397 		break;
4398 #endif
4399 	case KVM_CAP_SYNC_REGS:
4400 		r = KVM_SYNC_X86_VALID_FIELDS;
4401 		break;
4402 	case KVM_CAP_ADJUST_CLOCK:
4403 		r = KVM_CLOCK_VALID_FLAGS;
4404 		break;
4405 	case KVM_CAP_X86_DISABLE_EXITS:
4406 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4407 		      KVM_X86_DISABLE_EXITS_CSTATE;
4408 		if(kvm_can_mwait_in_guest())
4409 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
4410 		break;
4411 	case KVM_CAP_X86_SMM:
4412 		/* SMBASE is usually relocated above 1M on modern chipsets,
4413 		 * and SMM handlers might indeed rely on 4G segment limits,
4414 		 * so do not report SMM to be available if real mode is
4415 		 * emulated via vm86 mode.  Still, do not go to great lengths
4416 		 * to avoid userspace's usage of the feature, because it is a
4417 		 * fringe case that is not enabled except via specific settings
4418 		 * of the module parameters.
4419 		 */
4420 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4421 		break;
4422 	case KVM_CAP_NR_VCPUS:
4423 		r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4424 		break;
4425 	case KVM_CAP_MAX_VCPUS:
4426 		r = KVM_MAX_VCPUS;
4427 		break;
4428 	case KVM_CAP_MAX_VCPU_ID:
4429 		r = KVM_MAX_VCPU_IDS;
4430 		break;
4431 	case KVM_CAP_PV_MMU:	/* obsolete */
4432 		r = 0;
4433 		break;
4434 	case KVM_CAP_MCE:
4435 		r = KVM_MAX_MCE_BANKS;
4436 		break;
4437 	case KVM_CAP_XCRS:
4438 		r = boot_cpu_has(X86_FEATURE_XSAVE);
4439 		break;
4440 	case KVM_CAP_TSC_CONTROL:
4441 	case KVM_CAP_VM_TSC_CONTROL:
4442 		r = kvm_caps.has_tsc_control;
4443 		break;
4444 	case KVM_CAP_X2APIC_API:
4445 		r = KVM_X2APIC_API_VALID_FLAGS;
4446 		break;
4447 	case KVM_CAP_NESTED_STATE:
4448 		r = kvm_x86_ops.nested_ops->get_state ?
4449 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4450 		break;
4451 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4452 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4453 		break;
4454 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4455 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4456 		break;
4457 	case KVM_CAP_SMALLER_MAXPHYADDR:
4458 		r = (int) allow_smaller_maxphyaddr;
4459 		break;
4460 	case KVM_CAP_STEAL_TIME:
4461 		r = sched_info_on();
4462 		break;
4463 	case KVM_CAP_X86_BUS_LOCK_EXIT:
4464 		if (kvm_caps.has_bus_lock_exit)
4465 			r = KVM_BUS_LOCK_DETECTION_OFF |
4466 			    KVM_BUS_LOCK_DETECTION_EXIT;
4467 		else
4468 			r = 0;
4469 		break;
4470 	case KVM_CAP_XSAVE2: {
4471 		u64 guest_perm = xstate_get_guest_group_perm();
4472 
4473 		r = xstate_required_size(kvm_caps.supported_xcr0 & guest_perm, false);
4474 		if (r < sizeof(struct kvm_xsave))
4475 			r = sizeof(struct kvm_xsave);
4476 		break;
4477 	}
4478 	case KVM_CAP_PMU_CAPABILITY:
4479 		r = enable_pmu ? KVM_CAP_PMU_VALID_MASK : 0;
4480 		break;
4481 	case KVM_CAP_DISABLE_QUIRKS2:
4482 		r = KVM_X86_VALID_QUIRKS;
4483 		break;
4484 	case KVM_CAP_X86_NOTIFY_VMEXIT:
4485 		r = kvm_caps.has_notify_vmexit;
4486 		break;
4487 	default:
4488 		break;
4489 	}
4490 	return r;
4491 }
4492 
4493 static inline void __user *kvm_get_attr_addr(struct kvm_device_attr *attr)
4494 {
4495 	void __user *uaddr = (void __user*)(unsigned long)attr->addr;
4496 
4497 	if ((u64)(unsigned long)uaddr != attr->addr)
4498 		return ERR_PTR_USR(-EFAULT);
4499 	return uaddr;
4500 }
4501 
4502 static int kvm_x86_dev_get_attr(struct kvm_device_attr *attr)
4503 {
4504 	u64 __user *uaddr = kvm_get_attr_addr(attr);
4505 
4506 	if (attr->group)
4507 		return -ENXIO;
4508 
4509 	if (IS_ERR(uaddr))
4510 		return PTR_ERR(uaddr);
4511 
4512 	switch (attr->attr) {
4513 	case KVM_X86_XCOMP_GUEST_SUPP:
4514 		if (put_user(kvm_caps.supported_xcr0, uaddr))
4515 			return -EFAULT;
4516 		return 0;
4517 	default:
4518 		return -ENXIO;
4519 		break;
4520 	}
4521 }
4522 
4523 static int kvm_x86_dev_has_attr(struct kvm_device_attr *attr)
4524 {
4525 	if (attr->group)
4526 		return -ENXIO;
4527 
4528 	switch (attr->attr) {
4529 	case KVM_X86_XCOMP_GUEST_SUPP:
4530 		return 0;
4531 	default:
4532 		return -ENXIO;
4533 	}
4534 }
4535 
4536 long kvm_arch_dev_ioctl(struct file *filp,
4537 			unsigned int ioctl, unsigned long arg)
4538 {
4539 	void __user *argp = (void __user *)arg;
4540 	long r;
4541 
4542 	switch (ioctl) {
4543 	case KVM_GET_MSR_INDEX_LIST: {
4544 		struct kvm_msr_list __user *user_msr_list = argp;
4545 		struct kvm_msr_list msr_list;
4546 		unsigned n;
4547 
4548 		r = -EFAULT;
4549 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4550 			goto out;
4551 		n = msr_list.nmsrs;
4552 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4553 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4554 			goto out;
4555 		r = -E2BIG;
4556 		if (n < msr_list.nmsrs)
4557 			goto out;
4558 		r = -EFAULT;
4559 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4560 				 num_msrs_to_save * sizeof(u32)))
4561 			goto out;
4562 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4563 				 &emulated_msrs,
4564 				 num_emulated_msrs * sizeof(u32)))
4565 			goto out;
4566 		r = 0;
4567 		break;
4568 	}
4569 	case KVM_GET_SUPPORTED_CPUID:
4570 	case KVM_GET_EMULATED_CPUID: {
4571 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4572 		struct kvm_cpuid2 cpuid;
4573 
4574 		r = -EFAULT;
4575 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4576 			goto out;
4577 
4578 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4579 					    ioctl);
4580 		if (r)
4581 			goto out;
4582 
4583 		r = -EFAULT;
4584 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4585 			goto out;
4586 		r = 0;
4587 		break;
4588 	}
4589 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4590 		r = -EFAULT;
4591 		if (copy_to_user(argp, &kvm_caps.supported_mce_cap,
4592 				 sizeof(kvm_caps.supported_mce_cap)))
4593 			goto out;
4594 		r = 0;
4595 		break;
4596 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4597 		struct kvm_msr_list __user *user_msr_list = argp;
4598 		struct kvm_msr_list msr_list;
4599 		unsigned int n;
4600 
4601 		r = -EFAULT;
4602 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4603 			goto out;
4604 		n = msr_list.nmsrs;
4605 		msr_list.nmsrs = num_msr_based_features;
4606 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4607 			goto out;
4608 		r = -E2BIG;
4609 		if (n < msr_list.nmsrs)
4610 			goto out;
4611 		r = -EFAULT;
4612 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4613 				 num_msr_based_features * sizeof(u32)))
4614 			goto out;
4615 		r = 0;
4616 		break;
4617 	}
4618 	case KVM_GET_MSRS:
4619 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4620 		break;
4621 	case KVM_GET_SUPPORTED_HV_CPUID:
4622 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4623 		break;
4624 	case KVM_GET_DEVICE_ATTR: {
4625 		struct kvm_device_attr attr;
4626 		r = -EFAULT;
4627 		if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4628 			break;
4629 		r = kvm_x86_dev_get_attr(&attr);
4630 		break;
4631 	}
4632 	case KVM_HAS_DEVICE_ATTR: {
4633 		struct kvm_device_attr attr;
4634 		r = -EFAULT;
4635 		if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4636 			break;
4637 		r = kvm_x86_dev_has_attr(&attr);
4638 		break;
4639 	}
4640 	default:
4641 		r = -EINVAL;
4642 		break;
4643 	}
4644 out:
4645 	return r;
4646 }
4647 
4648 static void wbinvd_ipi(void *garbage)
4649 {
4650 	wbinvd();
4651 }
4652 
4653 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4654 {
4655 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4656 }
4657 
4658 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4659 {
4660 	/* Address WBINVD may be executed by guest */
4661 	if (need_emulate_wbinvd(vcpu)) {
4662 		if (static_call(kvm_x86_has_wbinvd_exit)())
4663 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4664 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4665 			smp_call_function_single(vcpu->cpu,
4666 					wbinvd_ipi, NULL, 1);
4667 	}
4668 
4669 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4670 
4671 	/* Save host pkru register if supported */
4672 	vcpu->arch.host_pkru = read_pkru();
4673 
4674 	/* Apply any externally detected TSC adjustments (due to suspend) */
4675 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4676 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4677 		vcpu->arch.tsc_offset_adjustment = 0;
4678 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4679 	}
4680 
4681 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4682 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4683 				rdtsc() - vcpu->arch.last_host_tsc;
4684 		if (tsc_delta < 0)
4685 			mark_tsc_unstable("KVM discovered backwards TSC");
4686 
4687 		if (kvm_check_tsc_unstable()) {
4688 			u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4689 						vcpu->arch.last_guest_tsc);
4690 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4691 			vcpu->arch.tsc_catchup = 1;
4692 		}
4693 
4694 		if (kvm_lapic_hv_timer_in_use(vcpu))
4695 			kvm_lapic_restart_hv_timer(vcpu);
4696 
4697 		/*
4698 		 * On a host with synchronized TSC, there is no need to update
4699 		 * kvmclock on vcpu->cpu migration
4700 		 */
4701 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4702 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4703 		if (vcpu->cpu != cpu)
4704 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4705 		vcpu->cpu = cpu;
4706 	}
4707 
4708 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4709 }
4710 
4711 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4712 {
4713 	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4714 	struct kvm_steal_time __user *st;
4715 	struct kvm_memslots *slots;
4716 	static const u8 preempted = KVM_VCPU_PREEMPTED;
4717 
4718 	/*
4719 	 * The vCPU can be marked preempted if and only if the VM-Exit was on
4720 	 * an instruction boundary and will not trigger guest emulation of any
4721 	 * kind (see vcpu_run).  Vendor specific code controls (conservatively)
4722 	 * when this is true, for example allowing the vCPU to be marked
4723 	 * preempted if and only if the VM-Exit was due to a host interrupt.
4724 	 */
4725 	if (!vcpu->arch.at_instruction_boundary) {
4726 		vcpu->stat.preemption_other++;
4727 		return;
4728 	}
4729 
4730 	vcpu->stat.preemption_reported++;
4731 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4732 		return;
4733 
4734 	if (vcpu->arch.st.preempted)
4735 		return;
4736 
4737 	/* This happens on process exit */
4738 	if (unlikely(current->mm != vcpu->kvm->mm))
4739 		return;
4740 
4741 	slots = kvm_memslots(vcpu->kvm);
4742 
4743 	if (unlikely(slots->generation != ghc->generation ||
4744 		     kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4745 		return;
4746 
4747 	st = (struct kvm_steal_time __user *)ghc->hva;
4748 	BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4749 
4750 	if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4751 		vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4752 
4753 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4754 }
4755 
4756 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4757 {
4758 	int idx;
4759 
4760 	if (vcpu->preempted) {
4761 		if (!vcpu->arch.guest_state_protected)
4762 			vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4763 
4764 		/*
4765 		 * Take the srcu lock as memslots will be accessed to check the gfn
4766 		 * cache generation against the memslots generation.
4767 		 */
4768 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4769 		if (kvm_xen_msr_enabled(vcpu->kvm))
4770 			kvm_xen_runstate_set_preempted(vcpu);
4771 		else
4772 			kvm_steal_time_set_preempted(vcpu);
4773 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4774 	}
4775 
4776 	static_call(kvm_x86_vcpu_put)(vcpu);
4777 	vcpu->arch.last_host_tsc = rdtsc();
4778 }
4779 
4780 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4781 				    struct kvm_lapic_state *s)
4782 {
4783 	static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4784 
4785 	return kvm_apic_get_state(vcpu, s);
4786 }
4787 
4788 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4789 				    struct kvm_lapic_state *s)
4790 {
4791 	int r;
4792 
4793 	r = kvm_apic_set_state(vcpu, s);
4794 	if (r)
4795 		return r;
4796 	update_cr8_intercept(vcpu);
4797 
4798 	return 0;
4799 }
4800 
4801 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4802 {
4803 	/*
4804 	 * We can accept userspace's request for interrupt injection
4805 	 * as long as we have a place to store the interrupt number.
4806 	 * The actual injection will happen when the CPU is able to
4807 	 * deliver the interrupt.
4808 	 */
4809 	if (kvm_cpu_has_extint(vcpu))
4810 		return false;
4811 
4812 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4813 	return (!lapic_in_kernel(vcpu) ||
4814 		kvm_apic_accept_pic_intr(vcpu));
4815 }
4816 
4817 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4818 {
4819 	/*
4820 	 * Do not cause an interrupt window exit if an exception
4821 	 * is pending or an event needs reinjection; userspace
4822 	 * might want to inject the interrupt manually using KVM_SET_REGS
4823 	 * or KVM_SET_SREGS.  For that to work, we must be at an
4824 	 * instruction boundary and with no events half-injected.
4825 	 */
4826 	return (kvm_arch_interrupt_allowed(vcpu) &&
4827 		kvm_cpu_accept_dm_intr(vcpu) &&
4828 		!kvm_event_needs_reinjection(vcpu) &&
4829 		!vcpu->arch.exception.pending);
4830 }
4831 
4832 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4833 				    struct kvm_interrupt *irq)
4834 {
4835 	if (irq->irq >= KVM_NR_INTERRUPTS)
4836 		return -EINVAL;
4837 
4838 	if (!irqchip_in_kernel(vcpu->kvm)) {
4839 		kvm_queue_interrupt(vcpu, irq->irq, false);
4840 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4841 		return 0;
4842 	}
4843 
4844 	/*
4845 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4846 	 * fail for in-kernel 8259.
4847 	 */
4848 	if (pic_in_kernel(vcpu->kvm))
4849 		return -ENXIO;
4850 
4851 	if (vcpu->arch.pending_external_vector != -1)
4852 		return -EEXIST;
4853 
4854 	vcpu->arch.pending_external_vector = irq->irq;
4855 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4856 	return 0;
4857 }
4858 
4859 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4860 {
4861 	kvm_inject_nmi(vcpu);
4862 
4863 	return 0;
4864 }
4865 
4866 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4867 {
4868 	kvm_make_request(KVM_REQ_SMI, vcpu);
4869 
4870 	return 0;
4871 }
4872 
4873 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4874 					   struct kvm_tpr_access_ctl *tac)
4875 {
4876 	if (tac->flags)
4877 		return -EINVAL;
4878 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4879 	return 0;
4880 }
4881 
4882 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4883 					u64 mcg_cap)
4884 {
4885 	int r;
4886 	unsigned bank_num = mcg_cap & 0xff, bank;
4887 
4888 	r = -EINVAL;
4889 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4890 		goto out;
4891 	if (mcg_cap & ~(kvm_caps.supported_mce_cap | 0xff | 0xff0000))
4892 		goto out;
4893 	r = 0;
4894 	vcpu->arch.mcg_cap = mcg_cap;
4895 	/* Init IA32_MCG_CTL to all 1s */
4896 	if (mcg_cap & MCG_CTL_P)
4897 		vcpu->arch.mcg_ctl = ~(u64)0;
4898 	/* Init IA32_MCi_CTL to all 1s, IA32_MCi_CTL2 to all 0s */
4899 	for (bank = 0; bank < bank_num; bank++) {
4900 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4901 		if (mcg_cap & MCG_CMCI_P)
4902 			vcpu->arch.mci_ctl2_banks[bank] = 0;
4903 	}
4904 
4905 	kvm_apic_after_set_mcg_cap(vcpu);
4906 
4907 	static_call(kvm_x86_setup_mce)(vcpu);
4908 out:
4909 	return r;
4910 }
4911 
4912 /*
4913  * Validate this is an UCNA (uncorrectable no action) error by checking the
4914  * MCG_STATUS and MCi_STATUS registers:
4915  * - none of the bits for Machine Check Exceptions are set
4916  * - both the VAL (valid) and UC (uncorrectable) bits are set
4917  * MCI_STATUS_PCC - Processor Context Corrupted
4918  * MCI_STATUS_S - Signaled as a Machine Check Exception
4919  * MCI_STATUS_AR - Software recoverable Action Required
4920  */
4921 static bool is_ucna(struct kvm_x86_mce *mce)
4922 {
4923 	return	!mce->mcg_status &&
4924 		!(mce->status & (MCI_STATUS_PCC | MCI_STATUS_S | MCI_STATUS_AR)) &&
4925 		(mce->status & MCI_STATUS_VAL) &&
4926 		(mce->status & MCI_STATUS_UC);
4927 }
4928 
4929 static int kvm_vcpu_x86_set_ucna(struct kvm_vcpu *vcpu, struct kvm_x86_mce *mce, u64* banks)
4930 {
4931 	u64 mcg_cap = vcpu->arch.mcg_cap;
4932 
4933 	banks[1] = mce->status;
4934 	banks[2] = mce->addr;
4935 	banks[3] = mce->misc;
4936 	vcpu->arch.mcg_status = mce->mcg_status;
4937 
4938 	if (!(mcg_cap & MCG_CMCI_P) ||
4939 	    !(vcpu->arch.mci_ctl2_banks[mce->bank] & MCI_CTL2_CMCI_EN))
4940 		return 0;
4941 
4942 	if (lapic_in_kernel(vcpu))
4943 		kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTCMCI);
4944 
4945 	return 0;
4946 }
4947 
4948 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4949 				      struct kvm_x86_mce *mce)
4950 {
4951 	u64 mcg_cap = vcpu->arch.mcg_cap;
4952 	unsigned bank_num = mcg_cap & 0xff;
4953 	u64 *banks = vcpu->arch.mce_banks;
4954 
4955 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4956 		return -EINVAL;
4957 
4958 	banks += array_index_nospec(4 * mce->bank, 4 * bank_num);
4959 
4960 	if (is_ucna(mce))
4961 		return kvm_vcpu_x86_set_ucna(vcpu, mce, banks);
4962 
4963 	/*
4964 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4965 	 * reporting is disabled
4966 	 */
4967 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4968 	    vcpu->arch.mcg_ctl != ~(u64)0)
4969 		return 0;
4970 	/*
4971 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4972 	 * reporting is disabled for the bank
4973 	 */
4974 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4975 		return 0;
4976 	if (mce->status & MCI_STATUS_UC) {
4977 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4978 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4979 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4980 			return 0;
4981 		}
4982 		if (banks[1] & MCI_STATUS_VAL)
4983 			mce->status |= MCI_STATUS_OVER;
4984 		banks[2] = mce->addr;
4985 		banks[3] = mce->misc;
4986 		vcpu->arch.mcg_status = mce->mcg_status;
4987 		banks[1] = mce->status;
4988 		kvm_queue_exception(vcpu, MC_VECTOR);
4989 	} else if (!(banks[1] & MCI_STATUS_VAL)
4990 		   || !(banks[1] & MCI_STATUS_UC)) {
4991 		if (banks[1] & MCI_STATUS_VAL)
4992 			mce->status |= MCI_STATUS_OVER;
4993 		banks[2] = mce->addr;
4994 		banks[3] = mce->misc;
4995 		banks[1] = mce->status;
4996 	} else
4997 		banks[1] |= MCI_STATUS_OVER;
4998 	return 0;
4999 }
5000 
5001 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
5002 					       struct kvm_vcpu_events *events)
5003 {
5004 	process_nmi(vcpu);
5005 
5006 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
5007 		process_smi(vcpu);
5008 
5009 	/*
5010 	 * In guest mode, payload delivery should be deferred,
5011 	 * so that the L1 hypervisor can intercept #PF before
5012 	 * CR2 is modified (or intercept #DB before DR6 is
5013 	 * modified under nVMX). Unless the per-VM capability,
5014 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
5015 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
5016 	 * opportunistically defer the exception payload, deliver it if the
5017 	 * capability hasn't been requested before processing a
5018 	 * KVM_GET_VCPU_EVENTS.
5019 	 */
5020 	if (!vcpu->kvm->arch.exception_payload_enabled &&
5021 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
5022 		kvm_deliver_exception_payload(vcpu);
5023 
5024 	/*
5025 	 * The API doesn't provide the instruction length for software
5026 	 * exceptions, so don't report them. As long as the guest RIP
5027 	 * isn't advanced, we should expect to encounter the exception
5028 	 * again.
5029 	 */
5030 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
5031 		events->exception.injected = 0;
5032 		events->exception.pending = 0;
5033 	} else {
5034 		events->exception.injected = vcpu->arch.exception.injected;
5035 		events->exception.pending = vcpu->arch.exception.pending;
5036 		/*
5037 		 * For ABI compatibility, deliberately conflate
5038 		 * pending and injected exceptions when
5039 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
5040 		 */
5041 		if (!vcpu->kvm->arch.exception_payload_enabled)
5042 			events->exception.injected |=
5043 				vcpu->arch.exception.pending;
5044 	}
5045 	events->exception.nr = vcpu->arch.exception.nr;
5046 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
5047 	events->exception.error_code = vcpu->arch.exception.error_code;
5048 	events->exception_has_payload = vcpu->arch.exception.has_payload;
5049 	events->exception_payload = vcpu->arch.exception.payload;
5050 
5051 	events->interrupt.injected =
5052 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
5053 	events->interrupt.nr = vcpu->arch.interrupt.nr;
5054 	events->interrupt.soft = 0;
5055 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
5056 
5057 	events->nmi.injected = vcpu->arch.nmi_injected;
5058 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
5059 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
5060 	events->nmi.pad = 0;
5061 
5062 	events->sipi_vector = 0; /* never valid when reporting to user space */
5063 
5064 	events->smi.smm = is_smm(vcpu);
5065 	events->smi.pending = vcpu->arch.smi_pending;
5066 	events->smi.smm_inside_nmi =
5067 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
5068 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
5069 
5070 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
5071 			 | KVM_VCPUEVENT_VALID_SHADOW
5072 			 | KVM_VCPUEVENT_VALID_SMM);
5073 	if (vcpu->kvm->arch.exception_payload_enabled)
5074 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
5075 	if (vcpu->kvm->arch.triple_fault_event) {
5076 		events->triple_fault.pending = kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5077 		events->flags |= KVM_VCPUEVENT_VALID_TRIPLE_FAULT;
5078 	}
5079 
5080 	memset(&events->reserved, 0, sizeof(events->reserved));
5081 }
5082 
5083 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
5084 
5085 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
5086 					      struct kvm_vcpu_events *events)
5087 {
5088 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
5089 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
5090 			      | KVM_VCPUEVENT_VALID_SHADOW
5091 			      | KVM_VCPUEVENT_VALID_SMM
5092 			      | KVM_VCPUEVENT_VALID_PAYLOAD
5093 			      | KVM_VCPUEVENT_VALID_TRIPLE_FAULT))
5094 		return -EINVAL;
5095 
5096 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
5097 		if (!vcpu->kvm->arch.exception_payload_enabled)
5098 			return -EINVAL;
5099 		if (events->exception.pending)
5100 			events->exception.injected = 0;
5101 		else
5102 			events->exception_has_payload = 0;
5103 	} else {
5104 		events->exception.pending = 0;
5105 		events->exception_has_payload = 0;
5106 	}
5107 
5108 	if ((events->exception.injected || events->exception.pending) &&
5109 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
5110 		return -EINVAL;
5111 
5112 	/* INITs are latched while in SMM */
5113 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
5114 	    (events->smi.smm || events->smi.pending) &&
5115 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
5116 		return -EINVAL;
5117 
5118 	process_nmi(vcpu);
5119 	vcpu->arch.exception.injected = events->exception.injected;
5120 	vcpu->arch.exception.pending = events->exception.pending;
5121 	vcpu->arch.exception.nr = events->exception.nr;
5122 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
5123 	vcpu->arch.exception.error_code = events->exception.error_code;
5124 	vcpu->arch.exception.has_payload = events->exception_has_payload;
5125 	vcpu->arch.exception.payload = events->exception_payload;
5126 
5127 	vcpu->arch.interrupt.injected = events->interrupt.injected;
5128 	vcpu->arch.interrupt.nr = events->interrupt.nr;
5129 	vcpu->arch.interrupt.soft = events->interrupt.soft;
5130 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
5131 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
5132 						events->interrupt.shadow);
5133 
5134 	vcpu->arch.nmi_injected = events->nmi.injected;
5135 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
5136 		vcpu->arch.nmi_pending = events->nmi.pending;
5137 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
5138 
5139 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
5140 	    lapic_in_kernel(vcpu))
5141 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
5142 
5143 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
5144 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
5145 			kvm_x86_ops.nested_ops->leave_nested(vcpu);
5146 			kvm_smm_changed(vcpu, events->smi.smm);
5147 		}
5148 
5149 		vcpu->arch.smi_pending = events->smi.pending;
5150 
5151 		if (events->smi.smm) {
5152 			if (events->smi.smm_inside_nmi)
5153 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
5154 			else
5155 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
5156 		}
5157 
5158 		if (lapic_in_kernel(vcpu)) {
5159 			if (events->smi.latched_init)
5160 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5161 			else
5162 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5163 		}
5164 	}
5165 
5166 	if (events->flags & KVM_VCPUEVENT_VALID_TRIPLE_FAULT) {
5167 		if (!vcpu->kvm->arch.triple_fault_event)
5168 			return -EINVAL;
5169 		if (events->triple_fault.pending)
5170 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5171 		else
5172 			kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5173 	}
5174 
5175 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5176 
5177 	return 0;
5178 }
5179 
5180 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
5181 					     struct kvm_debugregs *dbgregs)
5182 {
5183 	unsigned long val;
5184 
5185 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
5186 	kvm_get_dr(vcpu, 6, &val);
5187 	dbgregs->dr6 = val;
5188 	dbgregs->dr7 = vcpu->arch.dr7;
5189 	dbgregs->flags = 0;
5190 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
5191 }
5192 
5193 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
5194 					    struct kvm_debugregs *dbgregs)
5195 {
5196 	if (dbgregs->flags)
5197 		return -EINVAL;
5198 
5199 	if (!kvm_dr6_valid(dbgregs->dr6))
5200 		return -EINVAL;
5201 	if (!kvm_dr7_valid(dbgregs->dr7))
5202 		return -EINVAL;
5203 
5204 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
5205 	kvm_update_dr0123(vcpu);
5206 	vcpu->arch.dr6 = dbgregs->dr6;
5207 	vcpu->arch.dr7 = dbgregs->dr7;
5208 	kvm_update_dr7(vcpu);
5209 
5210 	return 0;
5211 }
5212 
5213 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
5214 					 struct kvm_xsave *guest_xsave)
5215 {
5216 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5217 		return;
5218 
5219 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5220 				       guest_xsave->region,
5221 				       sizeof(guest_xsave->region),
5222 				       vcpu->arch.pkru);
5223 }
5224 
5225 static void kvm_vcpu_ioctl_x86_get_xsave2(struct kvm_vcpu *vcpu,
5226 					  u8 *state, unsigned int size)
5227 {
5228 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5229 		return;
5230 
5231 	fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5232 				       state, size, vcpu->arch.pkru);
5233 }
5234 
5235 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
5236 					struct kvm_xsave *guest_xsave)
5237 {
5238 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5239 		return 0;
5240 
5241 	return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
5242 					      guest_xsave->region,
5243 					      kvm_caps.supported_xcr0,
5244 					      &vcpu->arch.pkru);
5245 }
5246 
5247 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
5248 					struct kvm_xcrs *guest_xcrs)
5249 {
5250 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
5251 		guest_xcrs->nr_xcrs = 0;
5252 		return;
5253 	}
5254 
5255 	guest_xcrs->nr_xcrs = 1;
5256 	guest_xcrs->flags = 0;
5257 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
5258 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
5259 }
5260 
5261 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
5262 				       struct kvm_xcrs *guest_xcrs)
5263 {
5264 	int i, r = 0;
5265 
5266 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
5267 		return -EINVAL;
5268 
5269 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
5270 		return -EINVAL;
5271 
5272 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
5273 		/* Only support XCR0 currently */
5274 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
5275 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
5276 				guest_xcrs->xcrs[i].value);
5277 			break;
5278 		}
5279 	if (r)
5280 		r = -EINVAL;
5281 	return r;
5282 }
5283 
5284 /*
5285  * kvm_set_guest_paused() indicates to the guest kernel that it has been
5286  * stopped by the hypervisor.  This function will be called from the host only.
5287  * EINVAL is returned when the host attempts to set the flag for a guest that
5288  * does not support pv clocks.
5289  */
5290 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5291 {
5292 	if (!vcpu->arch.pv_time.active)
5293 		return -EINVAL;
5294 	vcpu->arch.pvclock_set_guest_stopped_request = true;
5295 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5296 	return 0;
5297 }
5298 
5299 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5300 				 struct kvm_device_attr *attr)
5301 {
5302 	int r;
5303 
5304 	switch (attr->attr) {
5305 	case KVM_VCPU_TSC_OFFSET:
5306 		r = 0;
5307 		break;
5308 	default:
5309 		r = -ENXIO;
5310 	}
5311 
5312 	return r;
5313 }
5314 
5315 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5316 				 struct kvm_device_attr *attr)
5317 {
5318 	u64 __user *uaddr = kvm_get_attr_addr(attr);
5319 	int r;
5320 
5321 	if (IS_ERR(uaddr))
5322 		return PTR_ERR(uaddr);
5323 
5324 	switch (attr->attr) {
5325 	case KVM_VCPU_TSC_OFFSET:
5326 		r = -EFAULT;
5327 		if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5328 			break;
5329 		r = 0;
5330 		break;
5331 	default:
5332 		r = -ENXIO;
5333 	}
5334 
5335 	return r;
5336 }
5337 
5338 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5339 				 struct kvm_device_attr *attr)
5340 {
5341 	u64 __user *uaddr = kvm_get_attr_addr(attr);
5342 	struct kvm *kvm = vcpu->kvm;
5343 	int r;
5344 
5345 	if (IS_ERR(uaddr))
5346 		return PTR_ERR(uaddr);
5347 
5348 	switch (attr->attr) {
5349 	case KVM_VCPU_TSC_OFFSET: {
5350 		u64 offset, tsc, ns;
5351 		unsigned long flags;
5352 		bool matched;
5353 
5354 		r = -EFAULT;
5355 		if (get_user(offset, uaddr))
5356 			break;
5357 
5358 		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5359 
5360 		matched = (vcpu->arch.virtual_tsc_khz &&
5361 			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5362 			   kvm->arch.last_tsc_offset == offset);
5363 
5364 		tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5365 		ns = get_kvmclock_base_ns();
5366 
5367 		__kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5368 		raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5369 
5370 		r = 0;
5371 		break;
5372 	}
5373 	default:
5374 		r = -ENXIO;
5375 	}
5376 
5377 	return r;
5378 }
5379 
5380 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5381 				      unsigned int ioctl,
5382 				      void __user *argp)
5383 {
5384 	struct kvm_device_attr attr;
5385 	int r;
5386 
5387 	if (copy_from_user(&attr, argp, sizeof(attr)))
5388 		return -EFAULT;
5389 
5390 	if (attr.group != KVM_VCPU_TSC_CTRL)
5391 		return -ENXIO;
5392 
5393 	switch (ioctl) {
5394 	case KVM_HAS_DEVICE_ATTR:
5395 		r = kvm_arch_tsc_has_attr(vcpu, &attr);
5396 		break;
5397 	case KVM_GET_DEVICE_ATTR:
5398 		r = kvm_arch_tsc_get_attr(vcpu, &attr);
5399 		break;
5400 	case KVM_SET_DEVICE_ATTR:
5401 		r = kvm_arch_tsc_set_attr(vcpu, &attr);
5402 		break;
5403 	}
5404 
5405 	return r;
5406 }
5407 
5408 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5409 				     struct kvm_enable_cap *cap)
5410 {
5411 	int r;
5412 	uint16_t vmcs_version;
5413 	void __user *user_ptr;
5414 
5415 	if (cap->flags)
5416 		return -EINVAL;
5417 
5418 	switch (cap->cap) {
5419 	case KVM_CAP_HYPERV_SYNIC2:
5420 		if (cap->args[0])
5421 			return -EINVAL;
5422 		fallthrough;
5423 
5424 	case KVM_CAP_HYPERV_SYNIC:
5425 		if (!irqchip_in_kernel(vcpu->kvm))
5426 			return -EINVAL;
5427 		return kvm_hv_activate_synic(vcpu, cap->cap ==
5428 					     KVM_CAP_HYPERV_SYNIC2);
5429 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5430 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
5431 			return -ENOTTY;
5432 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5433 		if (!r) {
5434 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
5435 			if (copy_to_user(user_ptr, &vmcs_version,
5436 					 sizeof(vmcs_version)))
5437 				r = -EFAULT;
5438 		}
5439 		return r;
5440 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5441 		if (!kvm_x86_ops.enable_direct_tlbflush)
5442 			return -ENOTTY;
5443 
5444 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5445 
5446 	case KVM_CAP_HYPERV_ENFORCE_CPUID:
5447 		return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5448 
5449 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5450 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
5451 		if (vcpu->arch.pv_cpuid.enforce)
5452 			kvm_update_pv_runtime(vcpu);
5453 
5454 		return 0;
5455 	default:
5456 		return -EINVAL;
5457 	}
5458 }
5459 
5460 long kvm_arch_vcpu_ioctl(struct file *filp,
5461 			 unsigned int ioctl, unsigned long arg)
5462 {
5463 	struct kvm_vcpu *vcpu = filp->private_data;
5464 	void __user *argp = (void __user *)arg;
5465 	int r;
5466 	union {
5467 		struct kvm_sregs2 *sregs2;
5468 		struct kvm_lapic_state *lapic;
5469 		struct kvm_xsave *xsave;
5470 		struct kvm_xcrs *xcrs;
5471 		void *buffer;
5472 	} u;
5473 
5474 	vcpu_load(vcpu);
5475 
5476 	u.buffer = NULL;
5477 	switch (ioctl) {
5478 	case KVM_GET_LAPIC: {
5479 		r = -EINVAL;
5480 		if (!lapic_in_kernel(vcpu))
5481 			goto out;
5482 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5483 				GFP_KERNEL_ACCOUNT);
5484 
5485 		r = -ENOMEM;
5486 		if (!u.lapic)
5487 			goto out;
5488 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5489 		if (r)
5490 			goto out;
5491 		r = -EFAULT;
5492 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5493 			goto out;
5494 		r = 0;
5495 		break;
5496 	}
5497 	case KVM_SET_LAPIC: {
5498 		r = -EINVAL;
5499 		if (!lapic_in_kernel(vcpu))
5500 			goto out;
5501 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
5502 		if (IS_ERR(u.lapic)) {
5503 			r = PTR_ERR(u.lapic);
5504 			goto out_nofree;
5505 		}
5506 
5507 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5508 		break;
5509 	}
5510 	case KVM_INTERRUPT: {
5511 		struct kvm_interrupt irq;
5512 
5513 		r = -EFAULT;
5514 		if (copy_from_user(&irq, argp, sizeof(irq)))
5515 			goto out;
5516 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5517 		break;
5518 	}
5519 	case KVM_NMI: {
5520 		r = kvm_vcpu_ioctl_nmi(vcpu);
5521 		break;
5522 	}
5523 	case KVM_SMI: {
5524 		r = kvm_vcpu_ioctl_smi(vcpu);
5525 		break;
5526 	}
5527 	case KVM_SET_CPUID: {
5528 		struct kvm_cpuid __user *cpuid_arg = argp;
5529 		struct kvm_cpuid cpuid;
5530 
5531 		r = -EFAULT;
5532 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5533 			goto out;
5534 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5535 		break;
5536 	}
5537 	case KVM_SET_CPUID2: {
5538 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5539 		struct kvm_cpuid2 cpuid;
5540 
5541 		r = -EFAULT;
5542 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5543 			goto out;
5544 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5545 					      cpuid_arg->entries);
5546 		break;
5547 	}
5548 	case KVM_GET_CPUID2: {
5549 		struct kvm_cpuid2 __user *cpuid_arg = argp;
5550 		struct kvm_cpuid2 cpuid;
5551 
5552 		r = -EFAULT;
5553 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5554 			goto out;
5555 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5556 					      cpuid_arg->entries);
5557 		if (r)
5558 			goto out;
5559 		r = -EFAULT;
5560 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5561 			goto out;
5562 		r = 0;
5563 		break;
5564 	}
5565 	case KVM_GET_MSRS: {
5566 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5567 		r = msr_io(vcpu, argp, do_get_msr, 1);
5568 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5569 		break;
5570 	}
5571 	case KVM_SET_MSRS: {
5572 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
5573 		r = msr_io(vcpu, argp, do_set_msr, 0);
5574 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5575 		break;
5576 	}
5577 	case KVM_TPR_ACCESS_REPORTING: {
5578 		struct kvm_tpr_access_ctl tac;
5579 
5580 		r = -EFAULT;
5581 		if (copy_from_user(&tac, argp, sizeof(tac)))
5582 			goto out;
5583 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5584 		if (r)
5585 			goto out;
5586 		r = -EFAULT;
5587 		if (copy_to_user(argp, &tac, sizeof(tac)))
5588 			goto out;
5589 		r = 0;
5590 		break;
5591 	};
5592 	case KVM_SET_VAPIC_ADDR: {
5593 		struct kvm_vapic_addr va;
5594 		int idx;
5595 
5596 		r = -EINVAL;
5597 		if (!lapic_in_kernel(vcpu))
5598 			goto out;
5599 		r = -EFAULT;
5600 		if (copy_from_user(&va, argp, sizeof(va)))
5601 			goto out;
5602 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5603 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5604 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5605 		break;
5606 	}
5607 	case KVM_X86_SETUP_MCE: {
5608 		u64 mcg_cap;
5609 
5610 		r = -EFAULT;
5611 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5612 			goto out;
5613 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5614 		break;
5615 	}
5616 	case KVM_X86_SET_MCE: {
5617 		struct kvm_x86_mce mce;
5618 
5619 		r = -EFAULT;
5620 		if (copy_from_user(&mce, argp, sizeof(mce)))
5621 			goto out;
5622 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5623 		break;
5624 	}
5625 	case KVM_GET_VCPU_EVENTS: {
5626 		struct kvm_vcpu_events events;
5627 
5628 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5629 
5630 		r = -EFAULT;
5631 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5632 			break;
5633 		r = 0;
5634 		break;
5635 	}
5636 	case KVM_SET_VCPU_EVENTS: {
5637 		struct kvm_vcpu_events events;
5638 
5639 		r = -EFAULT;
5640 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5641 			break;
5642 
5643 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5644 		break;
5645 	}
5646 	case KVM_GET_DEBUGREGS: {
5647 		struct kvm_debugregs dbgregs;
5648 
5649 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5650 
5651 		r = -EFAULT;
5652 		if (copy_to_user(argp, &dbgregs,
5653 				 sizeof(struct kvm_debugregs)))
5654 			break;
5655 		r = 0;
5656 		break;
5657 	}
5658 	case KVM_SET_DEBUGREGS: {
5659 		struct kvm_debugregs dbgregs;
5660 
5661 		r = -EFAULT;
5662 		if (copy_from_user(&dbgregs, argp,
5663 				   sizeof(struct kvm_debugregs)))
5664 			break;
5665 
5666 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5667 		break;
5668 	}
5669 	case KVM_GET_XSAVE: {
5670 		r = -EINVAL;
5671 		if (vcpu->arch.guest_fpu.uabi_size > sizeof(struct kvm_xsave))
5672 			break;
5673 
5674 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5675 		r = -ENOMEM;
5676 		if (!u.xsave)
5677 			break;
5678 
5679 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5680 
5681 		r = -EFAULT;
5682 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5683 			break;
5684 		r = 0;
5685 		break;
5686 	}
5687 	case KVM_SET_XSAVE: {
5688 		int size = vcpu->arch.guest_fpu.uabi_size;
5689 
5690 		u.xsave = memdup_user(argp, size);
5691 		if (IS_ERR(u.xsave)) {
5692 			r = PTR_ERR(u.xsave);
5693 			goto out_nofree;
5694 		}
5695 
5696 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5697 		break;
5698 	}
5699 
5700 	case KVM_GET_XSAVE2: {
5701 		int size = vcpu->arch.guest_fpu.uabi_size;
5702 
5703 		u.xsave = kzalloc(size, GFP_KERNEL_ACCOUNT);
5704 		r = -ENOMEM;
5705 		if (!u.xsave)
5706 			break;
5707 
5708 		kvm_vcpu_ioctl_x86_get_xsave2(vcpu, u.buffer, size);
5709 
5710 		r = -EFAULT;
5711 		if (copy_to_user(argp, u.xsave, size))
5712 			break;
5713 
5714 		r = 0;
5715 		break;
5716 	}
5717 
5718 	case KVM_GET_XCRS: {
5719 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5720 		r = -ENOMEM;
5721 		if (!u.xcrs)
5722 			break;
5723 
5724 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5725 
5726 		r = -EFAULT;
5727 		if (copy_to_user(argp, u.xcrs,
5728 				 sizeof(struct kvm_xcrs)))
5729 			break;
5730 		r = 0;
5731 		break;
5732 	}
5733 	case KVM_SET_XCRS: {
5734 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5735 		if (IS_ERR(u.xcrs)) {
5736 			r = PTR_ERR(u.xcrs);
5737 			goto out_nofree;
5738 		}
5739 
5740 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5741 		break;
5742 	}
5743 	case KVM_SET_TSC_KHZ: {
5744 		u32 user_tsc_khz;
5745 
5746 		r = -EINVAL;
5747 		user_tsc_khz = (u32)arg;
5748 
5749 		if (kvm_caps.has_tsc_control &&
5750 		    user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
5751 			goto out;
5752 
5753 		if (user_tsc_khz == 0)
5754 			user_tsc_khz = tsc_khz;
5755 
5756 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5757 			r = 0;
5758 
5759 		goto out;
5760 	}
5761 	case KVM_GET_TSC_KHZ: {
5762 		r = vcpu->arch.virtual_tsc_khz;
5763 		goto out;
5764 	}
5765 	case KVM_KVMCLOCK_CTRL: {
5766 		r = kvm_set_guest_paused(vcpu);
5767 		goto out;
5768 	}
5769 	case KVM_ENABLE_CAP: {
5770 		struct kvm_enable_cap cap;
5771 
5772 		r = -EFAULT;
5773 		if (copy_from_user(&cap, argp, sizeof(cap)))
5774 			goto out;
5775 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5776 		break;
5777 	}
5778 	case KVM_GET_NESTED_STATE: {
5779 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5780 		u32 user_data_size;
5781 
5782 		r = -EINVAL;
5783 		if (!kvm_x86_ops.nested_ops->get_state)
5784 			break;
5785 
5786 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5787 		r = -EFAULT;
5788 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5789 			break;
5790 
5791 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5792 						     user_data_size);
5793 		if (r < 0)
5794 			break;
5795 
5796 		if (r > user_data_size) {
5797 			if (put_user(r, &user_kvm_nested_state->size))
5798 				r = -EFAULT;
5799 			else
5800 				r = -E2BIG;
5801 			break;
5802 		}
5803 
5804 		r = 0;
5805 		break;
5806 	}
5807 	case KVM_SET_NESTED_STATE: {
5808 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5809 		struct kvm_nested_state kvm_state;
5810 		int idx;
5811 
5812 		r = -EINVAL;
5813 		if (!kvm_x86_ops.nested_ops->set_state)
5814 			break;
5815 
5816 		r = -EFAULT;
5817 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5818 			break;
5819 
5820 		r = -EINVAL;
5821 		if (kvm_state.size < sizeof(kvm_state))
5822 			break;
5823 
5824 		if (kvm_state.flags &
5825 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5826 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5827 		      | KVM_STATE_NESTED_GIF_SET))
5828 			break;
5829 
5830 		/* nested_run_pending implies guest_mode.  */
5831 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5832 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5833 			break;
5834 
5835 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5836 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5837 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5838 		break;
5839 	}
5840 	case KVM_GET_SUPPORTED_HV_CPUID:
5841 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5842 		break;
5843 #ifdef CONFIG_KVM_XEN
5844 	case KVM_XEN_VCPU_GET_ATTR: {
5845 		struct kvm_xen_vcpu_attr xva;
5846 
5847 		r = -EFAULT;
5848 		if (copy_from_user(&xva, argp, sizeof(xva)))
5849 			goto out;
5850 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5851 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5852 			r = -EFAULT;
5853 		break;
5854 	}
5855 	case KVM_XEN_VCPU_SET_ATTR: {
5856 		struct kvm_xen_vcpu_attr xva;
5857 
5858 		r = -EFAULT;
5859 		if (copy_from_user(&xva, argp, sizeof(xva)))
5860 			goto out;
5861 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5862 		break;
5863 	}
5864 #endif
5865 	case KVM_GET_SREGS2: {
5866 		u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5867 		r = -ENOMEM;
5868 		if (!u.sregs2)
5869 			goto out;
5870 		__get_sregs2(vcpu, u.sregs2);
5871 		r = -EFAULT;
5872 		if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5873 			goto out;
5874 		r = 0;
5875 		break;
5876 	}
5877 	case KVM_SET_SREGS2: {
5878 		u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5879 		if (IS_ERR(u.sregs2)) {
5880 			r = PTR_ERR(u.sregs2);
5881 			u.sregs2 = NULL;
5882 			goto out;
5883 		}
5884 		r = __set_sregs2(vcpu, u.sregs2);
5885 		break;
5886 	}
5887 	case KVM_HAS_DEVICE_ATTR:
5888 	case KVM_GET_DEVICE_ATTR:
5889 	case KVM_SET_DEVICE_ATTR:
5890 		r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5891 		break;
5892 	default:
5893 		r = -EINVAL;
5894 	}
5895 out:
5896 	kfree(u.buffer);
5897 out_nofree:
5898 	vcpu_put(vcpu);
5899 	return r;
5900 }
5901 
5902 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5903 {
5904 	return VM_FAULT_SIGBUS;
5905 }
5906 
5907 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5908 {
5909 	int ret;
5910 
5911 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5912 		return -EINVAL;
5913 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5914 	return ret;
5915 }
5916 
5917 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5918 					      u64 ident_addr)
5919 {
5920 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5921 }
5922 
5923 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5924 					 unsigned long kvm_nr_mmu_pages)
5925 {
5926 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5927 		return -EINVAL;
5928 
5929 	mutex_lock(&kvm->slots_lock);
5930 
5931 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5932 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5933 
5934 	mutex_unlock(&kvm->slots_lock);
5935 	return 0;
5936 }
5937 
5938 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5939 {
5940 	return kvm->arch.n_max_mmu_pages;
5941 }
5942 
5943 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5944 {
5945 	struct kvm_pic *pic = kvm->arch.vpic;
5946 	int r;
5947 
5948 	r = 0;
5949 	switch (chip->chip_id) {
5950 	case KVM_IRQCHIP_PIC_MASTER:
5951 		memcpy(&chip->chip.pic, &pic->pics[0],
5952 			sizeof(struct kvm_pic_state));
5953 		break;
5954 	case KVM_IRQCHIP_PIC_SLAVE:
5955 		memcpy(&chip->chip.pic, &pic->pics[1],
5956 			sizeof(struct kvm_pic_state));
5957 		break;
5958 	case KVM_IRQCHIP_IOAPIC:
5959 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5960 		break;
5961 	default:
5962 		r = -EINVAL;
5963 		break;
5964 	}
5965 	return r;
5966 }
5967 
5968 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5969 {
5970 	struct kvm_pic *pic = kvm->arch.vpic;
5971 	int r;
5972 
5973 	r = 0;
5974 	switch (chip->chip_id) {
5975 	case KVM_IRQCHIP_PIC_MASTER:
5976 		spin_lock(&pic->lock);
5977 		memcpy(&pic->pics[0], &chip->chip.pic,
5978 			sizeof(struct kvm_pic_state));
5979 		spin_unlock(&pic->lock);
5980 		break;
5981 	case KVM_IRQCHIP_PIC_SLAVE:
5982 		spin_lock(&pic->lock);
5983 		memcpy(&pic->pics[1], &chip->chip.pic,
5984 			sizeof(struct kvm_pic_state));
5985 		spin_unlock(&pic->lock);
5986 		break;
5987 	case KVM_IRQCHIP_IOAPIC:
5988 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5989 		break;
5990 	default:
5991 		r = -EINVAL;
5992 		break;
5993 	}
5994 	kvm_pic_update_irq(pic);
5995 	return r;
5996 }
5997 
5998 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5999 {
6000 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
6001 
6002 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
6003 
6004 	mutex_lock(&kps->lock);
6005 	memcpy(ps, &kps->channels, sizeof(*ps));
6006 	mutex_unlock(&kps->lock);
6007 	return 0;
6008 }
6009 
6010 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6011 {
6012 	int i;
6013 	struct kvm_pit *pit = kvm->arch.vpit;
6014 
6015 	mutex_lock(&pit->pit_state.lock);
6016 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
6017 	for (i = 0; i < 3; i++)
6018 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
6019 	mutex_unlock(&pit->pit_state.lock);
6020 	return 0;
6021 }
6022 
6023 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6024 {
6025 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
6026 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
6027 		sizeof(ps->channels));
6028 	ps->flags = kvm->arch.vpit->pit_state.flags;
6029 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
6030 	memset(&ps->reserved, 0, sizeof(ps->reserved));
6031 	return 0;
6032 }
6033 
6034 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6035 {
6036 	int start = 0;
6037 	int i;
6038 	u32 prev_legacy, cur_legacy;
6039 	struct kvm_pit *pit = kvm->arch.vpit;
6040 
6041 	mutex_lock(&pit->pit_state.lock);
6042 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
6043 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
6044 	if (!prev_legacy && cur_legacy)
6045 		start = 1;
6046 	memcpy(&pit->pit_state.channels, &ps->channels,
6047 	       sizeof(pit->pit_state.channels));
6048 	pit->pit_state.flags = ps->flags;
6049 	for (i = 0; i < 3; i++)
6050 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
6051 				   start && i == 0);
6052 	mutex_unlock(&pit->pit_state.lock);
6053 	return 0;
6054 }
6055 
6056 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
6057 				 struct kvm_reinject_control *control)
6058 {
6059 	struct kvm_pit *pit = kvm->arch.vpit;
6060 
6061 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
6062 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
6063 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
6064 	 */
6065 	mutex_lock(&pit->pit_state.lock);
6066 	kvm_pit_set_reinject(pit, control->pit_reinject);
6067 	mutex_unlock(&pit->pit_state.lock);
6068 
6069 	return 0;
6070 }
6071 
6072 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
6073 {
6074 
6075 	/*
6076 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
6077 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
6078 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
6079 	 * VM-Exit.
6080 	 */
6081 	struct kvm_vcpu *vcpu;
6082 	unsigned long i;
6083 
6084 	kvm_for_each_vcpu(i, vcpu, kvm)
6085 		kvm_vcpu_kick(vcpu);
6086 }
6087 
6088 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
6089 			bool line_status)
6090 {
6091 	if (!irqchip_in_kernel(kvm))
6092 		return -ENXIO;
6093 
6094 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
6095 					irq_event->irq, irq_event->level,
6096 					line_status);
6097 	return 0;
6098 }
6099 
6100 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
6101 			    struct kvm_enable_cap *cap)
6102 {
6103 	int r;
6104 
6105 	if (cap->flags)
6106 		return -EINVAL;
6107 
6108 	switch (cap->cap) {
6109 	case KVM_CAP_DISABLE_QUIRKS2:
6110 		r = -EINVAL;
6111 		if (cap->args[0] & ~KVM_X86_VALID_QUIRKS)
6112 			break;
6113 		fallthrough;
6114 	case KVM_CAP_DISABLE_QUIRKS:
6115 		kvm->arch.disabled_quirks = cap->args[0];
6116 		r = 0;
6117 		break;
6118 	case KVM_CAP_SPLIT_IRQCHIP: {
6119 		mutex_lock(&kvm->lock);
6120 		r = -EINVAL;
6121 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
6122 			goto split_irqchip_unlock;
6123 		r = -EEXIST;
6124 		if (irqchip_in_kernel(kvm))
6125 			goto split_irqchip_unlock;
6126 		if (kvm->created_vcpus)
6127 			goto split_irqchip_unlock;
6128 		r = kvm_setup_empty_irq_routing(kvm);
6129 		if (r)
6130 			goto split_irqchip_unlock;
6131 		/* Pairs with irqchip_in_kernel. */
6132 		smp_wmb();
6133 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
6134 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
6135 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6136 		r = 0;
6137 split_irqchip_unlock:
6138 		mutex_unlock(&kvm->lock);
6139 		break;
6140 	}
6141 	case KVM_CAP_X2APIC_API:
6142 		r = -EINVAL;
6143 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
6144 			break;
6145 
6146 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
6147 			kvm->arch.x2apic_format = true;
6148 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
6149 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
6150 
6151 		r = 0;
6152 		break;
6153 	case KVM_CAP_X86_DISABLE_EXITS:
6154 		r = -EINVAL;
6155 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
6156 			break;
6157 
6158 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
6159 			kvm_can_mwait_in_guest())
6160 			kvm->arch.mwait_in_guest = true;
6161 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
6162 			kvm->arch.hlt_in_guest = true;
6163 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
6164 			kvm->arch.pause_in_guest = true;
6165 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
6166 			kvm->arch.cstate_in_guest = true;
6167 		r = 0;
6168 		break;
6169 	case KVM_CAP_MSR_PLATFORM_INFO:
6170 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
6171 		r = 0;
6172 		break;
6173 	case KVM_CAP_EXCEPTION_PAYLOAD:
6174 		kvm->arch.exception_payload_enabled = cap->args[0];
6175 		r = 0;
6176 		break;
6177 	case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
6178 		kvm->arch.triple_fault_event = cap->args[0];
6179 		r = 0;
6180 		break;
6181 	case KVM_CAP_X86_USER_SPACE_MSR:
6182 		r = -EINVAL;
6183 		if (cap->args[0] & ~(KVM_MSR_EXIT_REASON_INVAL |
6184 				     KVM_MSR_EXIT_REASON_UNKNOWN |
6185 				     KVM_MSR_EXIT_REASON_FILTER))
6186 			break;
6187 		kvm->arch.user_space_msr_mask = cap->args[0];
6188 		r = 0;
6189 		break;
6190 	case KVM_CAP_X86_BUS_LOCK_EXIT:
6191 		r = -EINVAL;
6192 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
6193 			break;
6194 
6195 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
6196 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
6197 			break;
6198 
6199 		if (kvm_caps.has_bus_lock_exit &&
6200 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
6201 			kvm->arch.bus_lock_detection_enabled = true;
6202 		r = 0;
6203 		break;
6204 #ifdef CONFIG_X86_SGX_KVM
6205 	case KVM_CAP_SGX_ATTRIBUTE: {
6206 		unsigned long allowed_attributes = 0;
6207 
6208 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
6209 		if (r)
6210 			break;
6211 
6212 		/* KVM only supports the PROVISIONKEY privileged attribute. */
6213 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
6214 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
6215 			kvm->arch.sgx_provisioning_allowed = true;
6216 		else
6217 			r = -EINVAL;
6218 		break;
6219 	}
6220 #endif
6221 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
6222 		r = -EINVAL;
6223 		if (!kvm_x86_ops.vm_copy_enc_context_from)
6224 			break;
6225 
6226 		r = static_call(kvm_x86_vm_copy_enc_context_from)(kvm, cap->args[0]);
6227 		break;
6228 	case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
6229 		r = -EINVAL;
6230 		if (!kvm_x86_ops.vm_move_enc_context_from)
6231 			break;
6232 
6233 		r = static_call(kvm_x86_vm_move_enc_context_from)(kvm, cap->args[0]);
6234 		break;
6235 	case KVM_CAP_EXIT_HYPERCALL:
6236 		if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
6237 			r = -EINVAL;
6238 			break;
6239 		}
6240 		kvm->arch.hypercall_exit_enabled = cap->args[0];
6241 		r = 0;
6242 		break;
6243 	case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
6244 		r = -EINVAL;
6245 		if (cap->args[0] & ~1)
6246 			break;
6247 		kvm->arch.exit_on_emulation_error = cap->args[0];
6248 		r = 0;
6249 		break;
6250 	case KVM_CAP_PMU_CAPABILITY:
6251 		r = -EINVAL;
6252 		if (!enable_pmu || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
6253 			break;
6254 
6255 		mutex_lock(&kvm->lock);
6256 		if (!kvm->created_vcpus) {
6257 			kvm->arch.enable_pmu = !(cap->args[0] & KVM_PMU_CAP_DISABLE);
6258 			r = 0;
6259 		}
6260 		mutex_unlock(&kvm->lock);
6261 		break;
6262 	case KVM_CAP_MAX_VCPU_ID:
6263 		r = -EINVAL;
6264 		if (cap->args[0] > KVM_MAX_VCPU_IDS)
6265 			break;
6266 
6267 		mutex_lock(&kvm->lock);
6268 		if (kvm->arch.max_vcpu_ids == cap->args[0]) {
6269 			r = 0;
6270 		} else if (!kvm->arch.max_vcpu_ids) {
6271 			kvm->arch.max_vcpu_ids = cap->args[0];
6272 			r = 0;
6273 		}
6274 		mutex_unlock(&kvm->lock);
6275 		break;
6276 	case KVM_CAP_X86_NOTIFY_VMEXIT:
6277 		r = -EINVAL;
6278 		if ((u32)cap->args[0] & ~KVM_X86_NOTIFY_VMEXIT_VALID_BITS)
6279 			break;
6280 		if (!kvm_caps.has_notify_vmexit)
6281 			break;
6282 		if (!((u32)cap->args[0] & KVM_X86_NOTIFY_VMEXIT_ENABLED))
6283 			break;
6284 		mutex_lock(&kvm->lock);
6285 		if (!kvm->created_vcpus) {
6286 			kvm->arch.notify_window = cap->args[0] >> 32;
6287 			kvm->arch.notify_vmexit_flags = (u32)cap->args[0];
6288 			r = 0;
6289 		}
6290 		mutex_unlock(&kvm->lock);
6291 		break;
6292 	case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
6293 		r = -EINVAL;
6294 
6295 		/*
6296 		 * Since the risk of disabling NX hugepages is a guest crashing
6297 		 * the system, ensure the userspace process has permission to
6298 		 * reboot the system.
6299 		 *
6300 		 * Note that unlike the reboot() syscall, the process must have
6301 		 * this capability in the root namespace because exposing
6302 		 * /dev/kvm into a container does not limit the scope of the
6303 		 * iTLB multihit bug to that container. In other words,
6304 		 * this must use capable(), not ns_capable().
6305 		 */
6306 		if (!capable(CAP_SYS_BOOT)) {
6307 			r = -EPERM;
6308 			break;
6309 		}
6310 
6311 		if (cap->args[0])
6312 			break;
6313 
6314 		mutex_lock(&kvm->lock);
6315 		if (!kvm->created_vcpus) {
6316 			kvm->arch.disable_nx_huge_pages = true;
6317 			r = 0;
6318 		}
6319 		mutex_unlock(&kvm->lock);
6320 		break;
6321 	default:
6322 		r = -EINVAL;
6323 		break;
6324 	}
6325 	return r;
6326 }
6327 
6328 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
6329 {
6330 	struct kvm_x86_msr_filter *msr_filter;
6331 
6332 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
6333 	if (!msr_filter)
6334 		return NULL;
6335 
6336 	msr_filter->default_allow = default_allow;
6337 	return msr_filter;
6338 }
6339 
6340 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
6341 {
6342 	u32 i;
6343 
6344 	if (!msr_filter)
6345 		return;
6346 
6347 	for (i = 0; i < msr_filter->count; i++)
6348 		kfree(msr_filter->ranges[i].bitmap);
6349 
6350 	kfree(msr_filter);
6351 }
6352 
6353 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
6354 			      struct kvm_msr_filter_range *user_range)
6355 {
6356 	unsigned long *bitmap = NULL;
6357 	size_t bitmap_size;
6358 
6359 	if (!user_range->nmsrs)
6360 		return 0;
6361 
6362 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
6363 		return -EINVAL;
6364 
6365 	if (!user_range->flags)
6366 		return -EINVAL;
6367 
6368 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
6369 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
6370 		return -EINVAL;
6371 
6372 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
6373 	if (IS_ERR(bitmap))
6374 		return PTR_ERR(bitmap);
6375 
6376 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
6377 		.flags = user_range->flags,
6378 		.base = user_range->base,
6379 		.nmsrs = user_range->nmsrs,
6380 		.bitmap = bitmap,
6381 	};
6382 
6383 	msr_filter->count++;
6384 	return 0;
6385 }
6386 
6387 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
6388 {
6389 	struct kvm_msr_filter __user *user_msr_filter = argp;
6390 	struct kvm_x86_msr_filter *new_filter, *old_filter;
6391 	struct kvm_msr_filter filter;
6392 	bool default_allow;
6393 	bool empty = true;
6394 	int r = 0;
6395 	u32 i;
6396 
6397 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
6398 		return -EFAULT;
6399 
6400 	if (filter.flags & ~KVM_MSR_FILTER_DEFAULT_DENY)
6401 		return -EINVAL;
6402 
6403 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
6404 		empty &= !filter.ranges[i].nmsrs;
6405 
6406 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
6407 	if (empty && !default_allow)
6408 		return -EINVAL;
6409 
6410 	new_filter = kvm_alloc_msr_filter(default_allow);
6411 	if (!new_filter)
6412 		return -ENOMEM;
6413 
6414 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6415 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
6416 		if (r) {
6417 			kvm_free_msr_filter(new_filter);
6418 			return r;
6419 		}
6420 	}
6421 
6422 	mutex_lock(&kvm->lock);
6423 
6424 	/* The per-VM filter is protected by kvm->lock... */
6425 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
6426 
6427 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
6428 	synchronize_srcu(&kvm->srcu);
6429 
6430 	kvm_free_msr_filter(old_filter);
6431 
6432 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6433 	mutex_unlock(&kvm->lock);
6434 
6435 	return 0;
6436 }
6437 
6438 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6439 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6440 {
6441 	struct kvm_vcpu *vcpu;
6442 	unsigned long i;
6443 	int ret = 0;
6444 
6445 	mutex_lock(&kvm->lock);
6446 	kvm_for_each_vcpu(i, vcpu, kvm) {
6447 		if (!vcpu->arch.pv_time.active)
6448 			continue;
6449 
6450 		ret = kvm_set_guest_paused(vcpu);
6451 		if (ret) {
6452 			kvm_err("Failed to pause guest VCPU%d: %d\n",
6453 				vcpu->vcpu_id, ret);
6454 			break;
6455 		}
6456 	}
6457 	mutex_unlock(&kvm->lock);
6458 
6459 	return ret ? NOTIFY_BAD : NOTIFY_DONE;
6460 }
6461 
6462 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6463 {
6464 	switch (state) {
6465 	case PM_HIBERNATION_PREPARE:
6466 	case PM_SUSPEND_PREPARE:
6467 		return kvm_arch_suspend_notifier(kvm);
6468 	}
6469 
6470 	return NOTIFY_DONE;
6471 }
6472 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6473 
6474 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6475 {
6476 	struct kvm_clock_data data = { 0 };
6477 
6478 	get_kvmclock(kvm, &data);
6479 	if (copy_to_user(argp, &data, sizeof(data)))
6480 		return -EFAULT;
6481 
6482 	return 0;
6483 }
6484 
6485 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6486 {
6487 	struct kvm_arch *ka = &kvm->arch;
6488 	struct kvm_clock_data data;
6489 	u64 now_raw_ns;
6490 
6491 	if (copy_from_user(&data, argp, sizeof(data)))
6492 		return -EFAULT;
6493 
6494 	/*
6495 	 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6496 	 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6497 	 */
6498 	if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6499 		return -EINVAL;
6500 
6501 	kvm_hv_request_tsc_page_update(kvm);
6502 	kvm_start_pvclock_update(kvm);
6503 	pvclock_update_vm_gtod_copy(kvm);
6504 
6505 	/*
6506 	 * This pairs with kvm_guest_time_update(): when masterclock is
6507 	 * in use, we use master_kernel_ns + kvmclock_offset to set
6508 	 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6509 	 * is slightly ahead) here we risk going negative on unsigned
6510 	 * 'system_time' when 'data.clock' is very small.
6511 	 */
6512 	if (data.flags & KVM_CLOCK_REALTIME) {
6513 		u64 now_real_ns = ktime_get_real_ns();
6514 
6515 		/*
6516 		 * Avoid stepping the kvmclock backwards.
6517 		 */
6518 		if (now_real_ns > data.realtime)
6519 			data.clock += now_real_ns - data.realtime;
6520 	}
6521 
6522 	if (ka->use_master_clock)
6523 		now_raw_ns = ka->master_kernel_ns;
6524 	else
6525 		now_raw_ns = get_kvmclock_base_ns();
6526 	ka->kvmclock_offset = data.clock - now_raw_ns;
6527 	kvm_end_pvclock_update(kvm);
6528 	return 0;
6529 }
6530 
6531 long kvm_arch_vm_ioctl(struct file *filp,
6532 		       unsigned int ioctl, unsigned long arg)
6533 {
6534 	struct kvm *kvm = filp->private_data;
6535 	void __user *argp = (void __user *)arg;
6536 	int r = -ENOTTY;
6537 	/*
6538 	 * This union makes it completely explicit to gcc-3.x
6539 	 * that these two variables' stack usage should be
6540 	 * combined, not added together.
6541 	 */
6542 	union {
6543 		struct kvm_pit_state ps;
6544 		struct kvm_pit_state2 ps2;
6545 		struct kvm_pit_config pit_config;
6546 	} u;
6547 
6548 	switch (ioctl) {
6549 	case KVM_SET_TSS_ADDR:
6550 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6551 		break;
6552 	case KVM_SET_IDENTITY_MAP_ADDR: {
6553 		u64 ident_addr;
6554 
6555 		mutex_lock(&kvm->lock);
6556 		r = -EINVAL;
6557 		if (kvm->created_vcpus)
6558 			goto set_identity_unlock;
6559 		r = -EFAULT;
6560 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6561 			goto set_identity_unlock;
6562 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6563 set_identity_unlock:
6564 		mutex_unlock(&kvm->lock);
6565 		break;
6566 	}
6567 	case KVM_SET_NR_MMU_PAGES:
6568 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6569 		break;
6570 	case KVM_GET_NR_MMU_PAGES:
6571 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6572 		break;
6573 	case KVM_CREATE_IRQCHIP: {
6574 		mutex_lock(&kvm->lock);
6575 
6576 		r = -EEXIST;
6577 		if (irqchip_in_kernel(kvm))
6578 			goto create_irqchip_unlock;
6579 
6580 		r = -EINVAL;
6581 		if (kvm->created_vcpus)
6582 			goto create_irqchip_unlock;
6583 
6584 		r = kvm_pic_init(kvm);
6585 		if (r)
6586 			goto create_irqchip_unlock;
6587 
6588 		r = kvm_ioapic_init(kvm);
6589 		if (r) {
6590 			kvm_pic_destroy(kvm);
6591 			goto create_irqchip_unlock;
6592 		}
6593 
6594 		r = kvm_setup_default_irq_routing(kvm);
6595 		if (r) {
6596 			kvm_ioapic_destroy(kvm);
6597 			kvm_pic_destroy(kvm);
6598 			goto create_irqchip_unlock;
6599 		}
6600 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6601 		smp_wmb();
6602 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6603 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6604 	create_irqchip_unlock:
6605 		mutex_unlock(&kvm->lock);
6606 		break;
6607 	}
6608 	case KVM_CREATE_PIT:
6609 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6610 		goto create_pit;
6611 	case KVM_CREATE_PIT2:
6612 		r = -EFAULT;
6613 		if (copy_from_user(&u.pit_config, argp,
6614 				   sizeof(struct kvm_pit_config)))
6615 			goto out;
6616 	create_pit:
6617 		mutex_lock(&kvm->lock);
6618 		r = -EEXIST;
6619 		if (kvm->arch.vpit)
6620 			goto create_pit_unlock;
6621 		r = -ENOMEM;
6622 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6623 		if (kvm->arch.vpit)
6624 			r = 0;
6625 	create_pit_unlock:
6626 		mutex_unlock(&kvm->lock);
6627 		break;
6628 	case KVM_GET_IRQCHIP: {
6629 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6630 		struct kvm_irqchip *chip;
6631 
6632 		chip = memdup_user(argp, sizeof(*chip));
6633 		if (IS_ERR(chip)) {
6634 			r = PTR_ERR(chip);
6635 			goto out;
6636 		}
6637 
6638 		r = -ENXIO;
6639 		if (!irqchip_kernel(kvm))
6640 			goto get_irqchip_out;
6641 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6642 		if (r)
6643 			goto get_irqchip_out;
6644 		r = -EFAULT;
6645 		if (copy_to_user(argp, chip, sizeof(*chip)))
6646 			goto get_irqchip_out;
6647 		r = 0;
6648 	get_irqchip_out:
6649 		kfree(chip);
6650 		break;
6651 	}
6652 	case KVM_SET_IRQCHIP: {
6653 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6654 		struct kvm_irqchip *chip;
6655 
6656 		chip = memdup_user(argp, sizeof(*chip));
6657 		if (IS_ERR(chip)) {
6658 			r = PTR_ERR(chip);
6659 			goto out;
6660 		}
6661 
6662 		r = -ENXIO;
6663 		if (!irqchip_kernel(kvm))
6664 			goto set_irqchip_out;
6665 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6666 	set_irqchip_out:
6667 		kfree(chip);
6668 		break;
6669 	}
6670 	case KVM_GET_PIT: {
6671 		r = -EFAULT;
6672 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6673 			goto out;
6674 		r = -ENXIO;
6675 		if (!kvm->arch.vpit)
6676 			goto out;
6677 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6678 		if (r)
6679 			goto out;
6680 		r = -EFAULT;
6681 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6682 			goto out;
6683 		r = 0;
6684 		break;
6685 	}
6686 	case KVM_SET_PIT: {
6687 		r = -EFAULT;
6688 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6689 			goto out;
6690 		mutex_lock(&kvm->lock);
6691 		r = -ENXIO;
6692 		if (!kvm->arch.vpit)
6693 			goto set_pit_out;
6694 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6695 set_pit_out:
6696 		mutex_unlock(&kvm->lock);
6697 		break;
6698 	}
6699 	case KVM_GET_PIT2: {
6700 		r = -ENXIO;
6701 		if (!kvm->arch.vpit)
6702 			goto out;
6703 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6704 		if (r)
6705 			goto out;
6706 		r = -EFAULT;
6707 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6708 			goto out;
6709 		r = 0;
6710 		break;
6711 	}
6712 	case KVM_SET_PIT2: {
6713 		r = -EFAULT;
6714 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6715 			goto out;
6716 		mutex_lock(&kvm->lock);
6717 		r = -ENXIO;
6718 		if (!kvm->arch.vpit)
6719 			goto set_pit2_out;
6720 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6721 set_pit2_out:
6722 		mutex_unlock(&kvm->lock);
6723 		break;
6724 	}
6725 	case KVM_REINJECT_CONTROL: {
6726 		struct kvm_reinject_control control;
6727 		r =  -EFAULT;
6728 		if (copy_from_user(&control, argp, sizeof(control)))
6729 			goto out;
6730 		r = -ENXIO;
6731 		if (!kvm->arch.vpit)
6732 			goto out;
6733 		r = kvm_vm_ioctl_reinject(kvm, &control);
6734 		break;
6735 	}
6736 	case KVM_SET_BOOT_CPU_ID:
6737 		r = 0;
6738 		mutex_lock(&kvm->lock);
6739 		if (kvm->created_vcpus)
6740 			r = -EBUSY;
6741 		else
6742 			kvm->arch.bsp_vcpu_id = arg;
6743 		mutex_unlock(&kvm->lock);
6744 		break;
6745 #ifdef CONFIG_KVM_XEN
6746 	case KVM_XEN_HVM_CONFIG: {
6747 		struct kvm_xen_hvm_config xhc;
6748 		r = -EFAULT;
6749 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
6750 			goto out;
6751 		r = kvm_xen_hvm_config(kvm, &xhc);
6752 		break;
6753 	}
6754 	case KVM_XEN_HVM_GET_ATTR: {
6755 		struct kvm_xen_hvm_attr xha;
6756 
6757 		r = -EFAULT;
6758 		if (copy_from_user(&xha, argp, sizeof(xha)))
6759 			goto out;
6760 		r = kvm_xen_hvm_get_attr(kvm, &xha);
6761 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6762 			r = -EFAULT;
6763 		break;
6764 	}
6765 	case KVM_XEN_HVM_SET_ATTR: {
6766 		struct kvm_xen_hvm_attr xha;
6767 
6768 		r = -EFAULT;
6769 		if (copy_from_user(&xha, argp, sizeof(xha)))
6770 			goto out;
6771 		r = kvm_xen_hvm_set_attr(kvm, &xha);
6772 		break;
6773 	}
6774 	case KVM_XEN_HVM_EVTCHN_SEND: {
6775 		struct kvm_irq_routing_xen_evtchn uxe;
6776 
6777 		r = -EFAULT;
6778 		if (copy_from_user(&uxe, argp, sizeof(uxe)))
6779 			goto out;
6780 		r = kvm_xen_hvm_evtchn_send(kvm, &uxe);
6781 		break;
6782 	}
6783 #endif
6784 	case KVM_SET_CLOCK:
6785 		r = kvm_vm_ioctl_set_clock(kvm, argp);
6786 		break;
6787 	case KVM_GET_CLOCK:
6788 		r = kvm_vm_ioctl_get_clock(kvm, argp);
6789 		break;
6790 	case KVM_SET_TSC_KHZ: {
6791 		u32 user_tsc_khz;
6792 
6793 		r = -EINVAL;
6794 		user_tsc_khz = (u32)arg;
6795 
6796 		if (kvm_caps.has_tsc_control &&
6797 		    user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
6798 			goto out;
6799 
6800 		if (user_tsc_khz == 0)
6801 			user_tsc_khz = tsc_khz;
6802 
6803 		WRITE_ONCE(kvm->arch.default_tsc_khz, user_tsc_khz);
6804 		r = 0;
6805 
6806 		goto out;
6807 	}
6808 	case KVM_GET_TSC_KHZ: {
6809 		r = READ_ONCE(kvm->arch.default_tsc_khz);
6810 		goto out;
6811 	}
6812 	case KVM_MEMORY_ENCRYPT_OP: {
6813 		r = -ENOTTY;
6814 		if (!kvm_x86_ops.mem_enc_ioctl)
6815 			goto out;
6816 
6817 		r = static_call(kvm_x86_mem_enc_ioctl)(kvm, argp);
6818 		break;
6819 	}
6820 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
6821 		struct kvm_enc_region region;
6822 
6823 		r = -EFAULT;
6824 		if (copy_from_user(&region, argp, sizeof(region)))
6825 			goto out;
6826 
6827 		r = -ENOTTY;
6828 		if (!kvm_x86_ops.mem_enc_register_region)
6829 			goto out;
6830 
6831 		r = static_call(kvm_x86_mem_enc_register_region)(kvm, &region);
6832 		break;
6833 	}
6834 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6835 		struct kvm_enc_region region;
6836 
6837 		r = -EFAULT;
6838 		if (copy_from_user(&region, argp, sizeof(region)))
6839 			goto out;
6840 
6841 		r = -ENOTTY;
6842 		if (!kvm_x86_ops.mem_enc_unregister_region)
6843 			goto out;
6844 
6845 		r = static_call(kvm_x86_mem_enc_unregister_region)(kvm, &region);
6846 		break;
6847 	}
6848 	case KVM_HYPERV_EVENTFD: {
6849 		struct kvm_hyperv_eventfd hvevfd;
6850 
6851 		r = -EFAULT;
6852 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6853 			goto out;
6854 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6855 		break;
6856 	}
6857 	case KVM_SET_PMU_EVENT_FILTER:
6858 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6859 		break;
6860 	case KVM_X86_SET_MSR_FILTER:
6861 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6862 		break;
6863 	default:
6864 		r = -ENOTTY;
6865 	}
6866 out:
6867 	return r;
6868 }
6869 
6870 static void kvm_init_msr_list(void)
6871 {
6872 	u32 dummy[2];
6873 	unsigned i;
6874 
6875 	BUILD_BUG_ON_MSG(KVM_PMC_MAX_FIXED != 3,
6876 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
6877 
6878 	num_msrs_to_save = 0;
6879 	num_emulated_msrs = 0;
6880 	num_msr_based_features = 0;
6881 
6882 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6883 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6884 			continue;
6885 
6886 		/*
6887 		 * Even MSRs that are valid in the host may not be exposed
6888 		 * to the guests in some cases.
6889 		 */
6890 		switch (msrs_to_save_all[i]) {
6891 		case MSR_IA32_BNDCFGS:
6892 			if (!kvm_mpx_supported())
6893 				continue;
6894 			break;
6895 		case MSR_TSC_AUX:
6896 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6897 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6898 				continue;
6899 			break;
6900 		case MSR_IA32_UMWAIT_CONTROL:
6901 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6902 				continue;
6903 			break;
6904 		case MSR_IA32_RTIT_CTL:
6905 		case MSR_IA32_RTIT_STATUS:
6906 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6907 				continue;
6908 			break;
6909 		case MSR_IA32_RTIT_CR3_MATCH:
6910 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6911 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6912 				continue;
6913 			break;
6914 		case MSR_IA32_RTIT_OUTPUT_BASE:
6915 		case MSR_IA32_RTIT_OUTPUT_MASK:
6916 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6917 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6918 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6919 				continue;
6920 			break;
6921 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6922 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6923 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6924 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6925 				continue;
6926 			break;
6927 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6928 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6929 			    min(INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp))
6930 				continue;
6931 			break;
6932 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6933 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6934 			    min(INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp))
6935 				continue;
6936 			break;
6937 		case MSR_IA32_XFD:
6938 		case MSR_IA32_XFD_ERR:
6939 			if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
6940 				continue;
6941 			break;
6942 		default:
6943 			break;
6944 		}
6945 
6946 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6947 	}
6948 
6949 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6950 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6951 			continue;
6952 
6953 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6954 	}
6955 
6956 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6957 		struct kvm_msr_entry msr;
6958 
6959 		msr.index = msr_based_features_all[i];
6960 		if (kvm_get_msr_feature(&msr))
6961 			continue;
6962 
6963 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6964 	}
6965 }
6966 
6967 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6968 			   const void *v)
6969 {
6970 	int handled = 0;
6971 	int n;
6972 
6973 	do {
6974 		n = min(len, 8);
6975 		if (!(lapic_in_kernel(vcpu) &&
6976 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6977 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6978 			break;
6979 		handled += n;
6980 		addr += n;
6981 		len -= n;
6982 		v += n;
6983 	} while (len);
6984 
6985 	return handled;
6986 }
6987 
6988 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6989 {
6990 	int handled = 0;
6991 	int n;
6992 
6993 	do {
6994 		n = min(len, 8);
6995 		if (!(lapic_in_kernel(vcpu) &&
6996 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6997 					 addr, n, v))
6998 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6999 			break;
7000 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
7001 		handled += n;
7002 		addr += n;
7003 		len -= n;
7004 		v += n;
7005 	} while (len);
7006 
7007 	return handled;
7008 }
7009 
7010 static void kvm_set_segment(struct kvm_vcpu *vcpu,
7011 			struct kvm_segment *var, int seg)
7012 {
7013 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
7014 }
7015 
7016 void kvm_get_segment(struct kvm_vcpu *vcpu,
7017 		     struct kvm_segment *var, int seg)
7018 {
7019 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
7020 }
7021 
7022 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
7023 			   struct x86_exception *exception)
7024 {
7025 	struct kvm_mmu *mmu = vcpu->arch.mmu;
7026 	gpa_t t_gpa;
7027 
7028 	BUG_ON(!mmu_is_nested(vcpu));
7029 
7030 	/* NPT walks are always user-walks */
7031 	access |= PFERR_USER_MASK;
7032 	t_gpa  = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
7033 
7034 	return t_gpa;
7035 }
7036 
7037 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
7038 			      struct x86_exception *exception)
7039 {
7040 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7041 
7042 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7043 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7044 }
7045 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
7046 
7047  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
7048 				struct x86_exception *exception)
7049 {
7050 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7051 
7052 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7053 	access |= PFERR_FETCH_MASK;
7054 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7055 }
7056 
7057 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
7058 			       struct x86_exception *exception)
7059 {
7060 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7061 
7062 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7063 	access |= PFERR_WRITE_MASK;
7064 	return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7065 }
7066 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
7067 
7068 /* uses this to access any guest's mapped memory without checking CPL */
7069 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
7070 				struct x86_exception *exception)
7071 {
7072 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7073 
7074 	return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
7075 }
7076 
7077 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7078 				      struct kvm_vcpu *vcpu, u64 access,
7079 				      struct x86_exception *exception)
7080 {
7081 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7082 	void *data = val;
7083 	int r = X86EMUL_CONTINUE;
7084 
7085 	while (bytes) {
7086 		gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7087 		unsigned offset = addr & (PAGE_SIZE-1);
7088 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
7089 		int ret;
7090 
7091 		if (gpa == INVALID_GPA)
7092 			return X86EMUL_PROPAGATE_FAULT;
7093 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
7094 					       offset, toread);
7095 		if (ret < 0) {
7096 			r = X86EMUL_IO_NEEDED;
7097 			goto out;
7098 		}
7099 
7100 		bytes -= toread;
7101 		data += toread;
7102 		addr += toread;
7103 	}
7104 out:
7105 	return r;
7106 }
7107 
7108 /* used for instruction fetching */
7109 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
7110 				gva_t addr, void *val, unsigned int bytes,
7111 				struct x86_exception *exception)
7112 {
7113 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7114 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7115 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7116 	unsigned offset;
7117 	int ret;
7118 
7119 	/* Inline kvm_read_guest_virt_helper for speed.  */
7120 	gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
7121 				    exception);
7122 	if (unlikely(gpa == INVALID_GPA))
7123 		return X86EMUL_PROPAGATE_FAULT;
7124 
7125 	offset = addr & (PAGE_SIZE-1);
7126 	if (WARN_ON(offset + bytes > PAGE_SIZE))
7127 		bytes = (unsigned)PAGE_SIZE - offset;
7128 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
7129 				       offset, bytes);
7130 	if (unlikely(ret < 0))
7131 		return X86EMUL_IO_NEEDED;
7132 
7133 	return X86EMUL_CONTINUE;
7134 }
7135 
7136 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
7137 			       gva_t addr, void *val, unsigned int bytes,
7138 			       struct x86_exception *exception)
7139 {
7140 	u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7141 
7142 	/*
7143 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
7144 	 * is returned, but our callers are not ready for that and they blindly
7145 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
7146 	 * uninitialized kernel stack memory into cr2 and error code.
7147 	 */
7148 	memset(exception, 0, sizeof(*exception));
7149 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
7150 					  exception);
7151 }
7152 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
7153 
7154 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
7155 			     gva_t addr, void *val, unsigned int bytes,
7156 			     struct x86_exception *exception, bool system)
7157 {
7158 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7159 	u64 access = 0;
7160 
7161 	if (system)
7162 		access |= PFERR_IMPLICIT_ACCESS;
7163 	else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7164 		access |= PFERR_USER_MASK;
7165 
7166 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
7167 }
7168 
7169 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
7170 		unsigned long addr, void *val, unsigned int bytes)
7171 {
7172 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7173 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
7174 
7175 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
7176 }
7177 
7178 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7179 				      struct kvm_vcpu *vcpu, u64 access,
7180 				      struct x86_exception *exception)
7181 {
7182 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7183 	void *data = val;
7184 	int r = X86EMUL_CONTINUE;
7185 
7186 	while (bytes) {
7187 		gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7188 		unsigned offset = addr & (PAGE_SIZE-1);
7189 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
7190 		int ret;
7191 
7192 		if (gpa == INVALID_GPA)
7193 			return X86EMUL_PROPAGATE_FAULT;
7194 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
7195 		if (ret < 0) {
7196 			r = X86EMUL_IO_NEEDED;
7197 			goto out;
7198 		}
7199 
7200 		bytes -= towrite;
7201 		data += towrite;
7202 		addr += towrite;
7203 	}
7204 out:
7205 	return r;
7206 }
7207 
7208 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
7209 			      unsigned int bytes, struct x86_exception *exception,
7210 			      bool system)
7211 {
7212 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7213 	u64 access = PFERR_WRITE_MASK;
7214 
7215 	if (system)
7216 		access |= PFERR_IMPLICIT_ACCESS;
7217 	else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7218 		access |= PFERR_USER_MASK;
7219 
7220 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7221 					   access, exception);
7222 }
7223 
7224 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
7225 				unsigned int bytes, struct x86_exception *exception)
7226 {
7227 	/* kvm_write_guest_virt_system can pull in tons of pages. */
7228 	vcpu->arch.l1tf_flush_l1d = true;
7229 
7230 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7231 					   PFERR_WRITE_MASK, exception);
7232 }
7233 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
7234 
7235 static int kvm_can_emulate_insn(struct kvm_vcpu *vcpu, int emul_type,
7236 				void *insn, int insn_len)
7237 {
7238 	return static_call(kvm_x86_can_emulate_instruction)(vcpu, emul_type,
7239 							    insn, insn_len);
7240 }
7241 
7242 int handle_ud(struct kvm_vcpu *vcpu)
7243 {
7244 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
7245 	int emul_type = EMULTYPE_TRAP_UD;
7246 	char sig[5]; /* ud2; .ascii "kvm" */
7247 	struct x86_exception e;
7248 
7249 	if (unlikely(!kvm_can_emulate_insn(vcpu, emul_type, NULL, 0)))
7250 		return 1;
7251 
7252 	if (force_emulation_prefix &&
7253 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
7254 				sig, sizeof(sig), &e) == 0 &&
7255 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
7256 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
7257 		emul_type = EMULTYPE_TRAP_UD_FORCED;
7258 	}
7259 
7260 	return kvm_emulate_instruction(vcpu, emul_type);
7261 }
7262 EXPORT_SYMBOL_GPL(handle_ud);
7263 
7264 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7265 			    gpa_t gpa, bool write)
7266 {
7267 	/* For APIC access vmexit */
7268 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7269 		return 1;
7270 
7271 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
7272 		trace_vcpu_match_mmio(gva, gpa, write, true);
7273 		return 1;
7274 	}
7275 
7276 	return 0;
7277 }
7278 
7279 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7280 				gpa_t *gpa, struct x86_exception *exception,
7281 				bool write)
7282 {
7283 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7284 	u64 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
7285 		| (write ? PFERR_WRITE_MASK : 0);
7286 
7287 	/*
7288 	 * currently PKRU is only applied to ept enabled guest so
7289 	 * there is no pkey in EPT page table for L1 guest or EPT
7290 	 * shadow page table for L2 guest.
7291 	 */
7292 	if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
7293 	    !permission_fault(vcpu, vcpu->arch.walk_mmu,
7294 			      vcpu->arch.mmio_access, 0, access))) {
7295 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
7296 					(gva & (PAGE_SIZE - 1));
7297 		trace_vcpu_match_mmio(gva, *gpa, write, false);
7298 		return 1;
7299 	}
7300 
7301 	*gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7302 
7303 	if (*gpa == INVALID_GPA)
7304 		return -1;
7305 
7306 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
7307 }
7308 
7309 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
7310 			const void *val, int bytes)
7311 {
7312 	int ret;
7313 
7314 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
7315 	if (ret < 0)
7316 		return 0;
7317 	kvm_page_track_write(vcpu, gpa, val, bytes);
7318 	return 1;
7319 }
7320 
7321 struct read_write_emulator_ops {
7322 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
7323 				  int bytes);
7324 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
7325 				  void *val, int bytes);
7326 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7327 			       int bytes, void *val);
7328 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7329 				    void *val, int bytes);
7330 	bool write;
7331 };
7332 
7333 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
7334 {
7335 	if (vcpu->mmio_read_completed) {
7336 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
7337 			       vcpu->mmio_fragments[0].gpa, val);
7338 		vcpu->mmio_read_completed = 0;
7339 		return 1;
7340 	}
7341 
7342 	return 0;
7343 }
7344 
7345 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7346 			void *val, int bytes)
7347 {
7348 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
7349 }
7350 
7351 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7352 			 void *val, int bytes)
7353 {
7354 	return emulator_write_phys(vcpu, gpa, val, bytes);
7355 }
7356 
7357 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
7358 {
7359 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
7360 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
7361 }
7362 
7363 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7364 			  void *val, int bytes)
7365 {
7366 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
7367 	return X86EMUL_IO_NEEDED;
7368 }
7369 
7370 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7371 			   void *val, int bytes)
7372 {
7373 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
7374 
7375 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
7376 	return X86EMUL_CONTINUE;
7377 }
7378 
7379 static const struct read_write_emulator_ops read_emultor = {
7380 	.read_write_prepare = read_prepare,
7381 	.read_write_emulate = read_emulate,
7382 	.read_write_mmio = vcpu_mmio_read,
7383 	.read_write_exit_mmio = read_exit_mmio,
7384 };
7385 
7386 static const struct read_write_emulator_ops write_emultor = {
7387 	.read_write_emulate = write_emulate,
7388 	.read_write_mmio = write_mmio,
7389 	.read_write_exit_mmio = write_exit_mmio,
7390 	.write = true,
7391 };
7392 
7393 static int emulator_read_write_onepage(unsigned long addr, void *val,
7394 				       unsigned int bytes,
7395 				       struct x86_exception *exception,
7396 				       struct kvm_vcpu *vcpu,
7397 				       const struct read_write_emulator_ops *ops)
7398 {
7399 	gpa_t gpa;
7400 	int handled, ret;
7401 	bool write = ops->write;
7402 	struct kvm_mmio_fragment *frag;
7403 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7404 
7405 	/*
7406 	 * If the exit was due to a NPF we may already have a GPA.
7407 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
7408 	 * Note, this cannot be used on string operations since string
7409 	 * operation using rep will only have the initial GPA from the NPF
7410 	 * occurred.
7411 	 */
7412 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
7413 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
7414 		gpa = ctxt->gpa_val;
7415 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
7416 	} else {
7417 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
7418 		if (ret < 0)
7419 			return X86EMUL_PROPAGATE_FAULT;
7420 	}
7421 
7422 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
7423 		return X86EMUL_CONTINUE;
7424 
7425 	/*
7426 	 * Is this MMIO handled locally?
7427 	 */
7428 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
7429 	if (handled == bytes)
7430 		return X86EMUL_CONTINUE;
7431 
7432 	gpa += handled;
7433 	bytes -= handled;
7434 	val += handled;
7435 
7436 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
7437 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
7438 	frag->gpa = gpa;
7439 	frag->data = val;
7440 	frag->len = bytes;
7441 	return X86EMUL_CONTINUE;
7442 }
7443 
7444 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
7445 			unsigned long addr,
7446 			void *val, unsigned int bytes,
7447 			struct x86_exception *exception,
7448 			const struct read_write_emulator_ops *ops)
7449 {
7450 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7451 	gpa_t gpa;
7452 	int rc;
7453 
7454 	if (ops->read_write_prepare &&
7455 		  ops->read_write_prepare(vcpu, val, bytes))
7456 		return X86EMUL_CONTINUE;
7457 
7458 	vcpu->mmio_nr_fragments = 0;
7459 
7460 	/* Crossing a page boundary? */
7461 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
7462 		int now;
7463 
7464 		now = -addr & ~PAGE_MASK;
7465 		rc = emulator_read_write_onepage(addr, val, now, exception,
7466 						 vcpu, ops);
7467 
7468 		if (rc != X86EMUL_CONTINUE)
7469 			return rc;
7470 		addr += now;
7471 		if (ctxt->mode != X86EMUL_MODE_PROT64)
7472 			addr = (u32)addr;
7473 		val += now;
7474 		bytes -= now;
7475 	}
7476 
7477 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
7478 					 vcpu, ops);
7479 	if (rc != X86EMUL_CONTINUE)
7480 		return rc;
7481 
7482 	if (!vcpu->mmio_nr_fragments)
7483 		return rc;
7484 
7485 	gpa = vcpu->mmio_fragments[0].gpa;
7486 
7487 	vcpu->mmio_needed = 1;
7488 	vcpu->mmio_cur_fragment = 0;
7489 
7490 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7491 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7492 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
7493 	vcpu->run->mmio.phys_addr = gpa;
7494 
7495 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7496 }
7497 
7498 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7499 				  unsigned long addr,
7500 				  void *val,
7501 				  unsigned int bytes,
7502 				  struct x86_exception *exception)
7503 {
7504 	return emulator_read_write(ctxt, addr, val, bytes,
7505 				   exception, &read_emultor);
7506 }
7507 
7508 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7509 			    unsigned long addr,
7510 			    const void *val,
7511 			    unsigned int bytes,
7512 			    struct x86_exception *exception)
7513 {
7514 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
7515 				   exception, &write_emultor);
7516 }
7517 
7518 #define emulator_try_cmpxchg_user(t, ptr, old, new) \
7519 	(__try_cmpxchg_user((t __user *)(ptr), (t *)(old), *(t *)(new), efault ## t))
7520 
7521 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7522 				     unsigned long addr,
7523 				     const void *old,
7524 				     const void *new,
7525 				     unsigned int bytes,
7526 				     struct x86_exception *exception)
7527 {
7528 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7529 	u64 page_line_mask;
7530 	unsigned long hva;
7531 	gpa_t gpa;
7532 	int r;
7533 
7534 	/* guests cmpxchg8b have to be emulated atomically */
7535 	if (bytes > 8 || (bytes & (bytes - 1)))
7536 		goto emul_write;
7537 
7538 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7539 
7540 	if (gpa == INVALID_GPA ||
7541 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7542 		goto emul_write;
7543 
7544 	/*
7545 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
7546 	 * enabled in the host and the access splits a cache line.
7547 	 */
7548 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7549 		page_line_mask = ~(cache_line_size() - 1);
7550 	else
7551 		page_line_mask = PAGE_MASK;
7552 
7553 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7554 		goto emul_write;
7555 
7556 	hva = kvm_vcpu_gfn_to_hva(vcpu, gpa_to_gfn(gpa));
7557 	if (kvm_is_error_hva(hva))
7558 		goto emul_write;
7559 
7560 	hva += offset_in_page(gpa);
7561 
7562 	switch (bytes) {
7563 	case 1:
7564 		r = emulator_try_cmpxchg_user(u8, hva, old, new);
7565 		break;
7566 	case 2:
7567 		r = emulator_try_cmpxchg_user(u16, hva, old, new);
7568 		break;
7569 	case 4:
7570 		r = emulator_try_cmpxchg_user(u32, hva, old, new);
7571 		break;
7572 	case 8:
7573 		r = emulator_try_cmpxchg_user(u64, hva, old, new);
7574 		break;
7575 	default:
7576 		BUG();
7577 	}
7578 
7579 	if (r < 0)
7580 		return X86EMUL_UNHANDLEABLE;
7581 	if (r)
7582 		return X86EMUL_CMPXCHG_FAILED;
7583 
7584 	kvm_page_track_write(vcpu, gpa, new, bytes);
7585 
7586 	return X86EMUL_CONTINUE;
7587 
7588 emul_write:
7589 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7590 
7591 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7592 }
7593 
7594 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7595 			       unsigned short port, void *data,
7596 			       unsigned int count, bool in)
7597 {
7598 	unsigned i;
7599 	int r;
7600 
7601 	WARN_ON_ONCE(vcpu->arch.pio.count);
7602 	for (i = 0; i < count; i++) {
7603 		if (in)
7604 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, port, size, data);
7605 		else
7606 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, port, size, data);
7607 
7608 		if (r) {
7609 			if (i == 0)
7610 				goto userspace_io;
7611 
7612 			/*
7613 			 * Userspace must have unregistered the device while PIO
7614 			 * was running.  Drop writes / read as 0.
7615 			 */
7616 			if (in)
7617 				memset(data, 0, size * (count - i));
7618 			break;
7619 		}
7620 
7621 		data += size;
7622 	}
7623 	return 1;
7624 
7625 userspace_io:
7626 	vcpu->arch.pio.port = port;
7627 	vcpu->arch.pio.in = in;
7628 	vcpu->arch.pio.count = count;
7629 	vcpu->arch.pio.size = size;
7630 
7631 	if (in)
7632 		memset(vcpu->arch.pio_data, 0, size * count);
7633 	else
7634 		memcpy(vcpu->arch.pio_data, data, size * count);
7635 
7636 	vcpu->run->exit_reason = KVM_EXIT_IO;
7637 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7638 	vcpu->run->io.size = size;
7639 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7640 	vcpu->run->io.count = count;
7641 	vcpu->run->io.port = port;
7642 	return 0;
7643 }
7644 
7645 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7646       			   unsigned short port, void *val, unsigned int count)
7647 {
7648 	int r = emulator_pio_in_out(vcpu, size, port, val, count, true);
7649 	if (r)
7650 		trace_kvm_pio(KVM_PIO_IN, port, size, count, val);
7651 
7652 	return r;
7653 }
7654 
7655 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7656 {
7657 	int size = vcpu->arch.pio.size;
7658 	unsigned int count = vcpu->arch.pio.count;
7659 	memcpy(val, vcpu->arch.pio_data, size * count);
7660 	trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7661 	vcpu->arch.pio.count = 0;
7662 }
7663 
7664 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7665 				    int size, unsigned short port, void *val,
7666 				    unsigned int count)
7667 {
7668 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7669 	if (vcpu->arch.pio.count) {
7670 		/*
7671 		 * Complete a previous iteration that required userspace I/O.
7672 		 * Note, @count isn't guaranteed to match pio.count as userspace
7673 		 * can modify ECX before rerunning the vCPU.  Ignore any such
7674 		 * shenanigans as KVM doesn't support modifying the rep count,
7675 		 * and the emulator ensures @count doesn't overflow the buffer.
7676 		 */
7677 		complete_emulator_pio_in(vcpu, val);
7678 		return 1;
7679 	}
7680 
7681 	return emulator_pio_in(vcpu, size, port, val, count);
7682 }
7683 
7684 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7685 			    unsigned short port, const void *val,
7686 			    unsigned int count)
7687 {
7688 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, val);
7689 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
7690 }
7691 
7692 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7693 				     int size, unsigned short port,
7694 				     const void *val, unsigned int count)
7695 {
7696 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7697 }
7698 
7699 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7700 {
7701 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7702 }
7703 
7704 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7705 {
7706 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7707 }
7708 
7709 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7710 {
7711 	if (!need_emulate_wbinvd(vcpu))
7712 		return X86EMUL_CONTINUE;
7713 
7714 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
7715 		int cpu = get_cpu();
7716 
7717 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7718 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7719 				wbinvd_ipi, NULL, 1);
7720 		put_cpu();
7721 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7722 	} else
7723 		wbinvd();
7724 	return X86EMUL_CONTINUE;
7725 }
7726 
7727 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7728 {
7729 	kvm_emulate_wbinvd_noskip(vcpu);
7730 	return kvm_skip_emulated_instruction(vcpu);
7731 }
7732 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7733 
7734 
7735 
7736 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7737 {
7738 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7739 }
7740 
7741 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7742 			    unsigned long *dest)
7743 {
7744 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7745 }
7746 
7747 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7748 			   unsigned long value)
7749 {
7750 
7751 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7752 }
7753 
7754 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7755 {
7756 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7757 }
7758 
7759 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7760 {
7761 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7762 	unsigned long value;
7763 
7764 	switch (cr) {
7765 	case 0:
7766 		value = kvm_read_cr0(vcpu);
7767 		break;
7768 	case 2:
7769 		value = vcpu->arch.cr2;
7770 		break;
7771 	case 3:
7772 		value = kvm_read_cr3(vcpu);
7773 		break;
7774 	case 4:
7775 		value = kvm_read_cr4(vcpu);
7776 		break;
7777 	case 8:
7778 		value = kvm_get_cr8(vcpu);
7779 		break;
7780 	default:
7781 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7782 		return 0;
7783 	}
7784 
7785 	return value;
7786 }
7787 
7788 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7789 {
7790 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7791 	int res = 0;
7792 
7793 	switch (cr) {
7794 	case 0:
7795 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7796 		break;
7797 	case 2:
7798 		vcpu->arch.cr2 = val;
7799 		break;
7800 	case 3:
7801 		res = kvm_set_cr3(vcpu, val);
7802 		break;
7803 	case 4:
7804 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7805 		break;
7806 	case 8:
7807 		res = kvm_set_cr8(vcpu, val);
7808 		break;
7809 	default:
7810 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
7811 		res = -1;
7812 	}
7813 
7814 	return res;
7815 }
7816 
7817 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7818 {
7819 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7820 }
7821 
7822 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7823 {
7824 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7825 }
7826 
7827 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7828 {
7829 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7830 }
7831 
7832 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7833 {
7834 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7835 }
7836 
7837 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7838 {
7839 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7840 }
7841 
7842 static unsigned long emulator_get_cached_segment_base(
7843 	struct x86_emulate_ctxt *ctxt, int seg)
7844 {
7845 	return get_segment_base(emul_to_vcpu(ctxt), seg);
7846 }
7847 
7848 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7849 				 struct desc_struct *desc, u32 *base3,
7850 				 int seg)
7851 {
7852 	struct kvm_segment var;
7853 
7854 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7855 	*selector = var.selector;
7856 
7857 	if (var.unusable) {
7858 		memset(desc, 0, sizeof(*desc));
7859 		if (base3)
7860 			*base3 = 0;
7861 		return false;
7862 	}
7863 
7864 	if (var.g)
7865 		var.limit >>= 12;
7866 	set_desc_limit(desc, var.limit);
7867 	set_desc_base(desc, (unsigned long)var.base);
7868 #ifdef CONFIG_X86_64
7869 	if (base3)
7870 		*base3 = var.base >> 32;
7871 #endif
7872 	desc->type = var.type;
7873 	desc->s = var.s;
7874 	desc->dpl = var.dpl;
7875 	desc->p = var.present;
7876 	desc->avl = var.avl;
7877 	desc->l = var.l;
7878 	desc->d = var.db;
7879 	desc->g = var.g;
7880 
7881 	return true;
7882 }
7883 
7884 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7885 				 struct desc_struct *desc, u32 base3,
7886 				 int seg)
7887 {
7888 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7889 	struct kvm_segment var;
7890 
7891 	var.selector = selector;
7892 	var.base = get_desc_base(desc);
7893 #ifdef CONFIG_X86_64
7894 	var.base |= ((u64)base3) << 32;
7895 #endif
7896 	var.limit = get_desc_limit(desc);
7897 	if (desc->g)
7898 		var.limit = (var.limit << 12) | 0xfff;
7899 	var.type = desc->type;
7900 	var.dpl = desc->dpl;
7901 	var.db = desc->d;
7902 	var.s = desc->s;
7903 	var.l = desc->l;
7904 	var.g = desc->g;
7905 	var.avl = desc->avl;
7906 	var.present = desc->p;
7907 	var.unusable = !var.present;
7908 	var.padding = 0;
7909 
7910 	kvm_set_segment(vcpu, &var, seg);
7911 	return;
7912 }
7913 
7914 static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
7915 					u32 msr_index, u64 *pdata)
7916 {
7917 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7918 	int r;
7919 
7920 	r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
7921 
7922 	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
7923 				    complete_emulated_rdmsr, r)) {
7924 		/* Bounce to user space */
7925 		return X86EMUL_IO_NEEDED;
7926 	}
7927 
7928 	return r;
7929 }
7930 
7931 static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
7932 					u32 msr_index, u64 data)
7933 {
7934 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7935 	int r;
7936 
7937 	r = kvm_set_msr_with_filter(vcpu, msr_index, data);
7938 
7939 	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
7940 				    complete_emulated_msr_access, r)) {
7941 		/* Bounce to user space */
7942 		return X86EMUL_IO_NEEDED;
7943 	}
7944 
7945 	return r;
7946 }
7947 
7948 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7949 			    u32 msr_index, u64 *pdata)
7950 {
7951 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
7952 }
7953 
7954 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7955 			    u32 msr_index, u64 data)
7956 {
7957 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
7958 }
7959 
7960 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7961 {
7962 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7963 
7964 	return vcpu->arch.smbase;
7965 }
7966 
7967 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7968 {
7969 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7970 
7971 	vcpu->arch.smbase = smbase;
7972 }
7973 
7974 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7975 			      u32 pmc)
7976 {
7977 	if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7978 		return 0;
7979 	return -EINVAL;
7980 }
7981 
7982 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7983 			     u32 pmc, u64 *pdata)
7984 {
7985 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7986 }
7987 
7988 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7989 {
7990 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7991 }
7992 
7993 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7994 			      struct x86_instruction_info *info,
7995 			      enum x86_intercept_stage stage)
7996 {
7997 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7998 					    &ctxt->exception);
7999 }
8000 
8001 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
8002 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
8003 			      bool exact_only)
8004 {
8005 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
8006 }
8007 
8008 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
8009 {
8010 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
8011 }
8012 
8013 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
8014 {
8015 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
8016 }
8017 
8018 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
8019 {
8020 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
8021 }
8022 
8023 static bool emulator_guest_has_rdpid(struct x86_emulate_ctxt *ctxt)
8024 {
8025 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_RDPID);
8026 }
8027 
8028 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
8029 {
8030 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
8031 }
8032 
8033 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
8034 {
8035 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
8036 }
8037 
8038 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
8039 {
8040 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
8041 }
8042 
8043 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
8044 {
8045 	return emul_to_vcpu(ctxt)->arch.hflags;
8046 }
8047 
8048 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
8049 {
8050 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8051 
8052 	kvm_smm_changed(vcpu, false);
8053 }
8054 
8055 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
8056 				  const char *smstate)
8057 {
8058 	return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
8059 }
8060 
8061 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
8062 {
8063 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
8064 }
8065 
8066 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
8067 {
8068 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
8069 }
8070 
8071 static void emulator_vm_bugged(struct x86_emulate_ctxt *ctxt)
8072 {
8073 	struct kvm *kvm = emul_to_vcpu(ctxt)->kvm;
8074 
8075 	if (!kvm->vm_bugged)
8076 		kvm_vm_bugged(kvm);
8077 }
8078 
8079 static const struct x86_emulate_ops emulate_ops = {
8080 	.vm_bugged           = emulator_vm_bugged,
8081 	.read_gpr            = emulator_read_gpr,
8082 	.write_gpr           = emulator_write_gpr,
8083 	.read_std            = emulator_read_std,
8084 	.write_std           = emulator_write_std,
8085 	.read_phys           = kvm_read_guest_phys_system,
8086 	.fetch               = kvm_fetch_guest_virt,
8087 	.read_emulated       = emulator_read_emulated,
8088 	.write_emulated      = emulator_write_emulated,
8089 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
8090 	.invlpg              = emulator_invlpg,
8091 	.pio_in_emulated     = emulator_pio_in_emulated,
8092 	.pio_out_emulated    = emulator_pio_out_emulated,
8093 	.get_segment         = emulator_get_segment,
8094 	.set_segment         = emulator_set_segment,
8095 	.get_cached_segment_base = emulator_get_cached_segment_base,
8096 	.get_gdt             = emulator_get_gdt,
8097 	.get_idt	     = emulator_get_idt,
8098 	.set_gdt             = emulator_set_gdt,
8099 	.set_idt	     = emulator_set_idt,
8100 	.get_cr              = emulator_get_cr,
8101 	.set_cr              = emulator_set_cr,
8102 	.cpl                 = emulator_get_cpl,
8103 	.get_dr              = emulator_get_dr,
8104 	.set_dr              = emulator_set_dr,
8105 	.get_smbase          = emulator_get_smbase,
8106 	.set_smbase          = emulator_set_smbase,
8107 	.set_msr_with_filter = emulator_set_msr_with_filter,
8108 	.get_msr_with_filter = emulator_get_msr_with_filter,
8109 	.set_msr             = emulator_set_msr,
8110 	.get_msr             = emulator_get_msr,
8111 	.check_pmc	     = emulator_check_pmc,
8112 	.read_pmc            = emulator_read_pmc,
8113 	.halt                = emulator_halt,
8114 	.wbinvd              = emulator_wbinvd,
8115 	.fix_hypercall       = emulator_fix_hypercall,
8116 	.intercept           = emulator_intercept,
8117 	.get_cpuid           = emulator_get_cpuid,
8118 	.guest_has_long_mode = emulator_guest_has_long_mode,
8119 	.guest_has_movbe     = emulator_guest_has_movbe,
8120 	.guest_has_fxsr      = emulator_guest_has_fxsr,
8121 	.guest_has_rdpid     = emulator_guest_has_rdpid,
8122 	.set_nmi_mask        = emulator_set_nmi_mask,
8123 	.get_hflags          = emulator_get_hflags,
8124 	.exiting_smm         = emulator_exiting_smm,
8125 	.leave_smm           = emulator_leave_smm,
8126 	.triple_fault        = emulator_triple_fault,
8127 	.set_xcr             = emulator_set_xcr,
8128 };
8129 
8130 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
8131 {
8132 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8133 	/*
8134 	 * an sti; sti; sequence only disable interrupts for the first
8135 	 * instruction. So, if the last instruction, be it emulated or
8136 	 * not, left the system with the INT_STI flag enabled, it
8137 	 * means that the last instruction is an sti. We should not
8138 	 * leave the flag on in this case. The same goes for mov ss
8139 	 */
8140 	if (int_shadow & mask)
8141 		mask = 0;
8142 	if (unlikely(int_shadow || mask)) {
8143 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
8144 		if (!mask)
8145 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8146 	}
8147 }
8148 
8149 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
8150 {
8151 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8152 	if (ctxt->exception.vector == PF_VECTOR)
8153 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
8154 
8155 	if (ctxt->exception.error_code_valid)
8156 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
8157 				      ctxt->exception.error_code);
8158 	else
8159 		kvm_queue_exception(vcpu, ctxt->exception.vector);
8160 	return false;
8161 }
8162 
8163 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
8164 {
8165 	struct x86_emulate_ctxt *ctxt;
8166 
8167 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
8168 	if (!ctxt) {
8169 		pr_err("kvm: failed to allocate vcpu's emulator\n");
8170 		return NULL;
8171 	}
8172 
8173 	ctxt->vcpu = vcpu;
8174 	ctxt->ops = &emulate_ops;
8175 	vcpu->arch.emulate_ctxt = ctxt;
8176 
8177 	return ctxt;
8178 }
8179 
8180 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
8181 {
8182 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8183 	int cs_db, cs_l;
8184 
8185 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
8186 
8187 	ctxt->gpa_available = false;
8188 	ctxt->eflags = kvm_get_rflags(vcpu);
8189 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
8190 
8191 	ctxt->eip = kvm_rip_read(vcpu);
8192 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
8193 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
8194 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
8195 		     cs_db				? X86EMUL_MODE_PROT32 :
8196 							  X86EMUL_MODE_PROT16;
8197 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
8198 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
8199 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
8200 
8201 	ctxt->interruptibility = 0;
8202 	ctxt->have_exception = false;
8203 	ctxt->exception.vector = -1;
8204 	ctxt->perm_ok = false;
8205 
8206 	init_decode_cache(ctxt);
8207 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8208 }
8209 
8210 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
8211 {
8212 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8213 	int ret;
8214 
8215 	init_emulate_ctxt(vcpu);
8216 
8217 	ctxt->op_bytes = 2;
8218 	ctxt->ad_bytes = 2;
8219 	ctxt->_eip = ctxt->eip + inc_eip;
8220 	ret = emulate_int_real(ctxt, irq);
8221 
8222 	if (ret != X86EMUL_CONTINUE) {
8223 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
8224 	} else {
8225 		ctxt->eip = ctxt->_eip;
8226 		kvm_rip_write(vcpu, ctxt->eip);
8227 		kvm_set_rflags(vcpu, ctxt->eflags);
8228 	}
8229 }
8230 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
8231 
8232 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8233 					   u8 ndata, u8 *insn_bytes, u8 insn_size)
8234 {
8235 	struct kvm_run *run = vcpu->run;
8236 	u64 info[5];
8237 	u8 info_start;
8238 
8239 	/*
8240 	 * Zero the whole array used to retrieve the exit info, as casting to
8241 	 * u32 for select entries will leave some chunks uninitialized.
8242 	 */
8243 	memset(&info, 0, sizeof(info));
8244 
8245 	static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
8246 					   &info[2], (u32 *)&info[3],
8247 					   (u32 *)&info[4]);
8248 
8249 	run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8250 	run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
8251 
8252 	/*
8253 	 * There's currently space for 13 entries, but 5 are used for the exit
8254 	 * reason and info.  Restrict to 4 to reduce the maintenance burden
8255 	 * when expanding kvm_run.emulation_failure in the future.
8256 	 */
8257 	if (WARN_ON_ONCE(ndata > 4))
8258 		ndata = 4;
8259 
8260 	/* Always include the flags as a 'data' entry. */
8261 	info_start = 1;
8262 	run->emulation_failure.flags = 0;
8263 
8264 	if (insn_size) {
8265 		BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
8266 			      sizeof(run->emulation_failure.insn_bytes) != 16));
8267 		info_start += 2;
8268 		run->emulation_failure.flags |=
8269 			KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
8270 		run->emulation_failure.insn_size = insn_size;
8271 		memset(run->emulation_failure.insn_bytes, 0x90,
8272 		       sizeof(run->emulation_failure.insn_bytes));
8273 		memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
8274 	}
8275 
8276 	memcpy(&run->internal.data[info_start], info, sizeof(info));
8277 	memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
8278 	       ndata * sizeof(data[0]));
8279 
8280 	run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
8281 }
8282 
8283 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
8284 {
8285 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8286 
8287 	prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
8288 				       ctxt->fetch.end - ctxt->fetch.data);
8289 }
8290 
8291 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8292 					  u8 ndata)
8293 {
8294 	prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
8295 }
8296 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
8297 
8298 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
8299 {
8300 	__kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
8301 }
8302 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
8303 
8304 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
8305 {
8306 	struct kvm *kvm = vcpu->kvm;
8307 
8308 	++vcpu->stat.insn_emulation_fail;
8309 	trace_kvm_emulate_insn_failed(vcpu);
8310 
8311 	if (emulation_type & EMULTYPE_VMWARE_GP) {
8312 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8313 		return 1;
8314 	}
8315 
8316 	if (kvm->arch.exit_on_emulation_error ||
8317 	    (emulation_type & EMULTYPE_SKIP)) {
8318 		prepare_emulation_ctxt_failure_exit(vcpu);
8319 		return 0;
8320 	}
8321 
8322 	kvm_queue_exception(vcpu, UD_VECTOR);
8323 
8324 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
8325 		prepare_emulation_ctxt_failure_exit(vcpu);
8326 		return 0;
8327 	}
8328 
8329 	return 1;
8330 }
8331 
8332 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8333 				  bool write_fault_to_shadow_pgtable,
8334 				  int emulation_type)
8335 {
8336 	gpa_t gpa = cr2_or_gpa;
8337 	kvm_pfn_t pfn;
8338 
8339 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8340 		return false;
8341 
8342 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8343 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8344 		return false;
8345 
8346 	if (!vcpu->arch.mmu->root_role.direct) {
8347 		/*
8348 		 * Write permission should be allowed since only
8349 		 * write access need to be emulated.
8350 		 */
8351 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8352 
8353 		/*
8354 		 * If the mapping is invalid in guest, let cpu retry
8355 		 * it to generate fault.
8356 		 */
8357 		if (gpa == INVALID_GPA)
8358 			return true;
8359 	}
8360 
8361 	/*
8362 	 * Do not retry the unhandleable instruction if it faults on the
8363 	 * readonly host memory, otherwise it will goto a infinite loop:
8364 	 * retry instruction -> write #PF -> emulation fail -> retry
8365 	 * instruction -> ...
8366 	 */
8367 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
8368 
8369 	/*
8370 	 * If the instruction failed on the error pfn, it can not be fixed,
8371 	 * report the error to userspace.
8372 	 */
8373 	if (is_error_noslot_pfn(pfn))
8374 		return false;
8375 
8376 	kvm_release_pfn_clean(pfn);
8377 
8378 	/* The instructions are well-emulated on direct mmu. */
8379 	if (vcpu->arch.mmu->root_role.direct) {
8380 		unsigned int indirect_shadow_pages;
8381 
8382 		write_lock(&vcpu->kvm->mmu_lock);
8383 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
8384 		write_unlock(&vcpu->kvm->mmu_lock);
8385 
8386 		if (indirect_shadow_pages)
8387 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8388 
8389 		return true;
8390 	}
8391 
8392 	/*
8393 	 * if emulation was due to access to shadowed page table
8394 	 * and it failed try to unshadow page and re-enter the
8395 	 * guest to let CPU execute the instruction.
8396 	 */
8397 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8398 
8399 	/*
8400 	 * If the access faults on its page table, it can not
8401 	 * be fixed by unprotecting shadow page and it should
8402 	 * be reported to userspace.
8403 	 */
8404 	return !write_fault_to_shadow_pgtable;
8405 }
8406 
8407 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
8408 			      gpa_t cr2_or_gpa,  int emulation_type)
8409 {
8410 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8411 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
8412 
8413 	last_retry_eip = vcpu->arch.last_retry_eip;
8414 	last_retry_addr = vcpu->arch.last_retry_addr;
8415 
8416 	/*
8417 	 * If the emulation is caused by #PF and it is non-page_table
8418 	 * writing instruction, it means the VM-EXIT is caused by shadow
8419 	 * page protected, we can zap the shadow page and retry this
8420 	 * instruction directly.
8421 	 *
8422 	 * Note: if the guest uses a non-page-table modifying instruction
8423 	 * on the PDE that points to the instruction, then we will unmap
8424 	 * the instruction and go to an infinite loop. So, we cache the
8425 	 * last retried eip and the last fault address, if we meet the eip
8426 	 * and the address again, we can break out of the potential infinite
8427 	 * loop.
8428 	 */
8429 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
8430 
8431 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8432 		return false;
8433 
8434 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8435 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8436 		return false;
8437 
8438 	if (x86_page_table_writing_insn(ctxt))
8439 		return false;
8440 
8441 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
8442 		return false;
8443 
8444 	vcpu->arch.last_retry_eip = ctxt->eip;
8445 	vcpu->arch.last_retry_addr = cr2_or_gpa;
8446 
8447 	if (!vcpu->arch.mmu->root_role.direct)
8448 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8449 
8450 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8451 
8452 	return true;
8453 }
8454 
8455 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
8456 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
8457 
8458 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
8459 {
8460 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
8461 
8462 	if (entering_smm) {
8463 		vcpu->arch.hflags |= HF_SMM_MASK;
8464 	} else {
8465 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
8466 
8467 		/* Process a latched INIT or SMI, if any.  */
8468 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8469 
8470 		/*
8471 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
8472 		 * on SMM exit we still need to reload them from
8473 		 * guest memory
8474 		 */
8475 		vcpu->arch.pdptrs_from_userspace = false;
8476 	}
8477 
8478 	kvm_mmu_reset_context(vcpu);
8479 }
8480 
8481 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
8482 				unsigned long *db)
8483 {
8484 	u32 dr6 = 0;
8485 	int i;
8486 	u32 enable, rwlen;
8487 
8488 	enable = dr7;
8489 	rwlen = dr7 >> 16;
8490 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
8491 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
8492 			dr6 |= (1 << i);
8493 	return dr6;
8494 }
8495 
8496 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
8497 {
8498 	struct kvm_run *kvm_run = vcpu->run;
8499 
8500 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
8501 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
8502 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
8503 		kvm_run->debug.arch.exception = DB_VECTOR;
8504 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
8505 		return 0;
8506 	}
8507 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
8508 	return 1;
8509 }
8510 
8511 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
8512 {
8513 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8514 	int r;
8515 
8516 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8517 	if (unlikely(!r))
8518 		return 0;
8519 
8520 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8521 
8522 	/*
8523 	 * rflags is the old, "raw" value of the flags.  The new value has
8524 	 * not been saved yet.
8525 	 *
8526 	 * This is correct even for TF set by the guest, because "the
8527 	 * processor will not generate this exception after the instruction
8528 	 * that sets the TF flag".
8529 	 */
8530 	if (unlikely(rflags & X86_EFLAGS_TF))
8531 		r = kvm_vcpu_do_singlestep(vcpu);
8532 	return r;
8533 }
8534 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8535 
8536 static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu, int *r)
8537 {
8538 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8539 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8540 		struct kvm_run *kvm_run = vcpu->run;
8541 		unsigned long eip = kvm_get_linear_rip(vcpu);
8542 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8543 					   vcpu->arch.guest_debug_dr7,
8544 					   vcpu->arch.eff_db);
8545 
8546 		if (dr6 != 0) {
8547 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8548 			kvm_run->debug.arch.pc = eip;
8549 			kvm_run->debug.arch.exception = DB_VECTOR;
8550 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
8551 			*r = 0;
8552 			return true;
8553 		}
8554 	}
8555 
8556 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8557 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
8558 		unsigned long eip = kvm_get_linear_rip(vcpu);
8559 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8560 					   vcpu->arch.dr7,
8561 					   vcpu->arch.db);
8562 
8563 		if (dr6 != 0) {
8564 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8565 			*r = 1;
8566 			return true;
8567 		}
8568 	}
8569 
8570 	return false;
8571 }
8572 
8573 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8574 {
8575 	switch (ctxt->opcode_len) {
8576 	case 1:
8577 		switch (ctxt->b) {
8578 		case 0xe4:	/* IN */
8579 		case 0xe5:
8580 		case 0xec:
8581 		case 0xed:
8582 		case 0xe6:	/* OUT */
8583 		case 0xe7:
8584 		case 0xee:
8585 		case 0xef:
8586 		case 0x6c:	/* INS */
8587 		case 0x6d:
8588 		case 0x6e:	/* OUTS */
8589 		case 0x6f:
8590 			return true;
8591 		}
8592 		break;
8593 	case 2:
8594 		switch (ctxt->b) {
8595 		case 0x33:	/* RDPMC */
8596 			return true;
8597 		}
8598 		break;
8599 	}
8600 
8601 	return false;
8602 }
8603 
8604 /*
8605  * Decode an instruction for emulation.  The caller is responsible for handling
8606  * code breakpoints.  Note, manually detecting code breakpoints is unnecessary
8607  * (and wrong) when emulating on an intercepted fault-like exception[*], as
8608  * code breakpoints have higher priority and thus have already been done by
8609  * hardware.
8610  *
8611  * [*] Except #MC, which is higher priority, but KVM should never emulate in
8612  *     response to a machine check.
8613  */
8614 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8615 				    void *insn, int insn_len)
8616 {
8617 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8618 	int r;
8619 
8620 	init_emulate_ctxt(vcpu);
8621 
8622 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8623 
8624 	trace_kvm_emulate_insn_start(vcpu);
8625 	++vcpu->stat.insn_emulation;
8626 
8627 	return r;
8628 }
8629 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8630 
8631 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8632 			    int emulation_type, void *insn, int insn_len)
8633 {
8634 	int r;
8635 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8636 	bool writeback = true;
8637 	bool write_fault_to_spt;
8638 
8639 	if (unlikely(!kvm_can_emulate_insn(vcpu, emulation_type, insn, insn_len)))
8640 		return 1;
8641 
8642 	vcpu->arch.l1tf_flush_l1d = true;
8643 
8644 	/*
8645 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
8646 	 * never reused.
8647 	 */
8648 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8649 	vcpu->arch.write_fault_to_shadow_pgtable = false;
8650 
8651 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8652 		kvm_clear_exception_queue(vcpu);
8653 
8654 		/*
8655 		 * Return immediately if RIP hits a code breakpoint, such #DBs
8656 		 * are fault-like and are higher priority than any faults on
8657 		 * the code fetch itself.
8658 		 */
8659 		if (!(emulation_type & EMULTYPE_SKIP) &&
8660 		    kvm_vcpu_check_code_breakpoint(vcpu, &r))
8661 			return r;
8662 
8663 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
8664 						    insn, insn_len);
8665 		if (r != EMULATION_OK)  {
8666 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
8667 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8668 				kvm_queue_exception(vcpu, UD_VECTOR);
8669 				return 1;
8670 			}
8671 			if (reexecute_instruction(vcpu, cr2_or_gpa,
8672 						  write_fault_to_spt,
8673 						  emulation_type))
8674 				return 1;
8675 			if (ctxt->have_exception) {
8676 				/*
8677 				 * #UD should result in just EMULATION_FAILED, and trap-like
8678 				 * exception should not be encountered during decode.
8679 				 */
8680 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8681 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8682 				inject_emulated_exception(vcpu);
8683 				return 1;
8684 			}
8685 			return handle_emulation_failure(vcpu, emulation_type);
8686 		}
8687 	}
8688 
8689 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8690 	    !is_vmware_backdoor_opcode(ctxt)) {
8691 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8692 		return 1;
8693 	}
8694 
8695 	/*
8696 	 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8697 	 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8698 	 * The caller is responsible for updating interruptibility state and
8699 	 * injecting single-step #DBs.
8700 	 */
8701 	if (emulation_type & EMULTYPE_SKIP) {
8702 		if (ctxt->mode != X86EMUL_MODE_PROT64)
8703 			ctxt->eip = (u32)ctxt->_eip;
8704 		else
8705 			ctxt->eip = ctxt->_eip;
8706 
8707 		if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8708 			r = 1;
8709 			goto writeback;
8710 		}
8711 
8712 		kvm_rip_write(vcpu, ctxt->eip);
8713 		if (ctxt->eflags & X86_EFLAGS_RF)
8714 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8715 		return 1;
8716 	}
8717 
8718 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8719 		return 1;
8720 
8721 	/* this is needed for vmware backdoor interface to work since it
8722 	   changes registers values  during IO operation */
8723 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8724 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8725 		emulator_invalidate_register_cache(ctxt);
8726 	}
8727 
8728 restart:
8729 	if (emulation_type & EMULTYPE_PF) {
8730 		/* Save the faulting GPA (cr2) in the address field */
8731 		ctxt->exception.address = cr2_or_gpa;
8732 
8733 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
8734 		if (vcpu->arch.mmu->root_role.direct) {
8735 			ctxt->gpa_available = true;
8736 			ctxt->gpa_val = cr2_or_gpa;
8737 		}
8738 	} else {
8739 		/* Sanitize the address out of an abundance of paranoia. */
8740 		ctxt->exception.address = 0;
8741 	}
8742 
8743 	r = x86_emulate_insn(ctxt);
8744 
8745 	if (r == EMULATION_INTERCEPTED)
8746 		return 1;
8747 
8748 	if (r == EMULATION_FAILED) {
8749 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8750 					emulation_type))
8751 			return 1;
8752 
8753 		return handle_emulation_failure(vcpu, emulation_type);
8754 	}
8755 
8756 	if (ctxt->have_exception) {
8757 		r = 1;
8758 		if (inject_emulated_exception(vcpu))
8759 			return r;
8760 	} else if (vcpu->arch.pio.count) {
8761 		if (!vcpu->arch.pio.in) {
8762 			/* FIXME: return into emulator if single-stepping.  */
8763 			vcpu->arch.pio.count = 0;
8764 		} else {
8765 			writeback = false;
8766 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
8767 		}
8768 		r = 0;
8769 	} else if (vcpu->mmio_needed) {
8770 		++vcpu->stat.mmio_exits;
8771 
8772 		if (!vcpu->mmio_is_write)
8773 			writeback = false;
8774 		r = 0;
8775 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8776 	} else if (vcpu->arch.complete_userspace_io) {
8777 		writeback = false;
8778 		r = 0;
8779 	} else if (r == EMULATION_RESTART)
8780 		goto restart;
8781 	else
8782 		r = 1;
8783 
8784 writeback:
8785 	if (writeback) {
8786 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8787 		toggle_interruptibility(vcpu, ctxt->interruptibility);
8788 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8789 		if (!ctxt->have_exception ||
8790 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8791 			kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8792 			if (ctxt->is_branch)
8793 				kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
8794 			kvm_rip_write(vcpu, ctxt->eip);
8795 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8796 				r = kvm_vcpu_do_singlestep(vcpu);
8797 			static_call_cond(kvm_x86_update_emulated_instruction)(vcpu);
8798 			__kvm_set_rflags(vcpu, ctxt->eflags);
8799 		}
8800 
8801 		/*
8802 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8803 		 * do nothing, and it will be requested again as soon as
8804 		 * the shadow expires.  But we still need to check here,
8805 		 * because POPF has no interrupt shadow.
8806 		 */
8807 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8808 			kvm_make_request(KVM_REQ_EVENT, vcpu);
8809 	} else
8810 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8811 
8812 	return r;
8813 }
8814 
8815 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8816 {
8817 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8818 }
8819 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8820 
8821 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8822 					void *insn, int insn_len)
8823 {
8824 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8825 }
8826 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8827 
8828 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8829 {
8830 	vcpu->arch.pio.count = 0;
8831 	return 1;
8832 }
8833 
8834 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8835 {
8836 	vcpu->arch.pio.count = 0;
8837 
8838 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8839 		return 1;
8840 
8841 	return kvm_skip_emulated_instruction(vcpu);
8842 }
8843 
8844 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8845 			    unsigned short port)
8846 {
8847 	unsigned long val = kvm_rax_read(vcpu);
8848 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8849 
8850 	if (ret)
8851 		return ret;
8852 
8853 	/*
8854 	 * Workaround userspace that relies on old KVM behavior of %rip being
8855 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8856 	 */
8857 	if (port == 0x7e &&
8858 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8859 		vcpu->arch.complete_userspace_io =
8860 			complete_fast_pio_out_port_0x7e;
8861 		kvm_skip_emulated_instruction(vcpu);
8862 	} else {
8863 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8864 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8865 	}
8866 	return 0;
8867 }
8868 
8869 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8870 {
8871 	unsigned long val;
8872 
8873 	/* We should only ever be called with arch.pio.count equal to 1 */
8874 	BUG_ON(vcpu->arch.pio.count != 1);
8875 
8876 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8877 		vcpu->arch.pio.count = 0;
8878 		return 1;
8879 	}
8880 
8881 	/* For size less than 4 we merge, else we zero extend */
8882 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8883 
8884 	complete_emulator_pio_in(vcpu, &val);
8885 	kvm_rax_write(vcpu, val);
8886 
8887 	return kvm_skip_emulated_instruction(vcpu);
8888 }
8889 
8890 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8891 			   unsigned short port)
8892 {
8893 	unsigned long val;
8894 	int ret;
8895 
8896 	/* For size less than 4 we merge, else we zero extend */
8897 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8898 
8899 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
8900 	if (ret) {
8901 		kvm_rax_write(vcpu, val);
8902 		return ret;
8903 	}
8904 
8905 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8906 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8907 
8908 	return 0;
8909 }
8910 
8911 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8912 {
8913 	int ret;
8914 
8915 	if (in)
8916 		ret = kvm_fast_pio_in(vcpu, size, port);
8917 	else
8918 		ret = kvm_fast_pio_out(vcpu, size, port);
8919 	return ret && kvm_skip_emulated_instruction(vcpu);
8920 }
8921 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8922 
8923 static int kvmclock_cpu_down_prep(unsigned int cpu)
8924 {
8925 	__this_cpu_write(cpu_tsc_khz, 0);
8926 	return 0;
8927 }
8928 
8929 static void tsc_khz_changed(void *data)
8930 {
8931 	struct cpufreq_freqs *freq = data;
8932 	unsigned long khz = 0;
8933 
8934 	if (data)
8935 		khz = freq->new;
8936 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8937 		khz = cpufreq_quick_get(raw_smp_processor_id());
8938 	if (!khz)
8939 		khz = tsc_khz;
8940 	__this_cpu_write(cpu_tsc_khz, khz);
8941 }
8942 
8943 #ifdef CONFIG_X86_64
8944 static void kvm_hyperv_tsc_notifier(void)
8945 {
8946 	struct kvm *kvm;
8947 	int cpu;
8948 
8949 	mutex_lock(&kvm_lock);
8950 	list_for_each_entry(kvm, &vm_list, vm_list)
8951 		kvm_make_mclock_inprogress_request(kvm);
8952 
8953 	/* no guest entries from this point */
8954 	hyperv_stop_tsc_emulation();
8955 
8956 	/* TSC frequency always matches when on Hyper-V */
8957 	for_each_present_cpu(cpu)
8958 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8959 	kvm_caps.max_guest_tsc_khz = tsc_khz;
8960 
8961 	list_for_each_entry(kvm, &vm_list, vm_list) {
8962 		__kvm_start_pvclock_update(kvm);
8963 		pvclock_update_vm_gtod_copy(kvm);
8964 		kvm_end_pvclock_update(kvm);
8965 	}
8966 
8967 	mutex_unlock(&kvm_lock);
8968 }
8969 #endif
8970 
8971 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8972 {
8973 	struct kvm *kvm;
8974 	struct kvm_vcpu *vcpu;
8975 	int send_ipi = 0;
8976 	unsigned long i;
8977 
8978 	/*
8979 	 * We allow guests to temporarily run on slowing clocks,
8980 	 * provided we notify them after, or to run on accelerating
8981 	 * clocks, provided we notify them before.  Thus time never
8982 	 * goes backwards.
8983 	 *
8984 	 * However, we have a problem.  We can't atomically update
8985 	 * the frequency of a given CPU from this function; it is
8986 	 * merely a notifier, which can be called from any CPU.
8987 	 * Changing the TSC frequency at arbitrary points in time
8988 	 * requires a recomputation of local variables related to
8989 	 * the TSC for each VCPU.  We must flag these local variables
8990 	 * to be updated and be sure the update takes place with the
8991 	 * new frequency before any guests proceed.
8992 	 *
8993 	 * Unfortunately, the combination of hotplug CPU and frequency
8994 	 * change creates an intractable locking scenario; the order
8995 	 * of when these callouts happen is undefined with respect to
8996 	 * CPU hotplug, and they can race with each other.  As such,
8997 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8998 	 * undefined; you can actually have a CPU frequency change take
8999 	 * place in between the computation of X and the setting of the
9000 	 * variable.  To protect against this problem, all updates of
9001 	 * the per_cpu tsc_khz variable are done in an interrupt
9002 	 * protected IPI, and all callers wishing to update the value
9003 	 * must wait for a synchronous IPI to complete (which is trivial
9004 	 * if the caller is on the CPU already).  This establishes the
9005 	 * necessary total order on variable updates.
9006 	 *
9007 	 * Note that because a guest time update may take place
9008 	 * anytime after the setting of the VCPU's request bit, the
9009 	 * correct TSC value must be set before the request.  However,
9010 	 * to ensure the update actually makes it to any guest which
9011 	 * starts running in hardware virtualization between the set
9012 	 * and the acquisition of the spinlock, we must also ping the
9013 	 * CPU after setting the request bit.
9014 	 *
9015 	 */
9016 
9017 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9018 
9019 	mutex_lock(&kvm_lock);
9020 	list_for_each_entry(kvm, &vm_list, vm_list) {
9021 		kvm_for_each_vcpu(i, vcpu, kvm) {
9022 			if (vcpu->cpu != cpu)
9023 				continue;
9024 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9025 			if (vcpu->cpu != raw_smp_processor_id())
9026 				send_ipi = 1;
9027 		}
9028 	}
9029 	mutex_unlock(&kvm_lock);
9030 
9031 	if (freq->old < freq->new && send_ipi) {
9032 		/*
9033 		 * We upscale the frequency.  Must make the guest
9034 		 * doesn't see old kvmclock values while running with
9035 		 * the new frequency, otherwise we risk the guest sees
9036 		 * time go backwards.
9037 		 *
9038 		 * In case we update the frequency for another cpu
9039 		 * (which might be in guest context) send an interrupt
9040 		 * to kick the cpu out of guest context.  Next time
9041 		 * guest context is entered kvmclock will be updated,
9042 		 * so the guest will not see stale values.
9043 		 */
9044 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9045 	}
9046 }
9047 
9048 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
9049 				     void *data)
9050 {
9051 	struct cpufreq_freqs *freq = data;
9052 	int cpu;
9053 
9054 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
9055 		return 0;
9056 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
9057 		return 0;
9058 
9059 	for_each_cpu(cpu, freq->policy->cpus)
9060 		__kvmclock_cpufreq_notifier(freq, cpu);
9061 
9062 	return 0;
9063 }
9064 
9065 static struct notifier_block kvmclock_cpufreq_notifier_block = {
9066 	.notifier_call  = kvmclock_cpufreq_notifier
9067 };
9068 
9069 static int kvmclock_cpu_online(unsigned int cpu)
9070 {
9071 	tsc_khz_changed(NULL);
9072 	return 0;
9073 }
9074 
9075 static void kvm_timer_init(void)
9076 {
9077 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9078 		max_tsc_khz = tsc_khz;
9079 
9080 		if (IS_ENABLED(CONFIG_CPU_FREQ)) {
9081 			struct cpufreq_policy *policy;
9082 			int cpu;
9083 
9084 			cpu = get_cpu();
9085 			policy = cpufreq_cpu_get(cpu);
9086 			if (policy) {
9087 				if (policy->cpuinfo.max_freq)
9088 					max_tsc_khz = policy->cpuinfo.max_freq;
9089 				cpufreq_cpu_put(policy);
9090 			}
9091 			put_cpu();
9092 		}
9093 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
9094 					  CPUFREQ_TRANSITION_NOTIFIER);
9095 	}
9096 
9097 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
9098 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
9099 }
9100 
9101 #ifdef CONFIG_X86_64
9102 static void pvclock_gtod_update_fn(struct work_struct *work)
9103 {
9104 	struct kvm *kvm;
9105 	struct kvm_vcpu *vcpu;
9106 	unsigned long i;
9107 
9108 	mutex_lock(&kvm_lock);
9109 	list_for_each_entry(kvm, &vm_list, vm_list)
9110 		kvm_for_each_vcpu(i, vcpu, kvm)
9111 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9112 	atomic_set(&kvm_guest_has_master_clock, 0);
9113 	mutex_unlock(&kvm_lock);
9114 }
9115 
9116 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
9117 
9118 /*
9119  * Indirection to move queue_work() out of the tk_core.seq write held
9120  * region to prevent possible deadlocks against time accessors which
9121  * are invoked with work related locks held.
9122  */
9123 static void pvclock_irq_work_fn(struct irq_work *w)
9124 {
9125 	queue_work(system_long_wq, &pvclock_gtod_work);
9126 }
9127 
9128 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
9129 
9130 /*
9131  * Notification about pvclock gtod data update.
9132  */
9133 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
9134 			       void *priv)
9135 {
9136 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
9137 	struct timekeeper *tk = priv;
9138 
9139 	update_pvclock_gtod(tk);
9140 
9141 	/*
9142 	 * Disable master clock if host does not trust, or does not use,
9143 	 * TSC based clocksource. Delegate queue_work() to irq_work as
9144 	 * this is invoked with tk_core.seq write held.
9145 	 */
9146 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
9147 	    atomic_read(&kvm_guest_has_master_clock) != 0)
9148 		irq_work_queue(&pvclock_irq_work);
9149 	return 0;
9150 }
9151 
9152 static struct notifier_block pvclock_gtod_notifier = {
9153 	.notifier_call = pvclock_gtod_notify,
9154 };
9155 #endif
9156 
9157 int kvm_arch_init(void *opaque)
9158 {
9159 	struct kvm_x86_init_ops *ops = opaque;
9160 	u64 host_pat;
9161 	int r;
9162 
9163 	if (kvm_x86_ops.hardware_enable) {
9164 		pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
9165 		return -EEXIST;
9166 	}
9167 
9168 	if (!ops->cpu_has_kvm_support()) {
9169 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
9170 				   ops->runtime_ops->name);
9171 		return -EOPNOTSUPP;
9172 	}
9173 	if (ops->disabled_by_bios()) {
9174 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
9175 				   ops->runtime_ops->name);
9176 		return -EOPNOTSUPP;
9177 	}
9178 
9179 	/*
9180 	 * KVM explicitly assumes that the guest has an FPU and
9181 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
9182 	 * vCPU's FPU state as a fxregs_state struct.
9183 	 */
9184 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
9185 		printk(KERN_ERR "kvm: inadequate fpu\n");
9186 		return -EOPNOTSUPP;
9187 	}
9188 
9189 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9190 		pr_err("RT requires X86_FEATURE_CONSTANT_TSC\n");
9191 		return -EOPNOTSUPP;
9192 	}
9193 
9194 	/*
9195 	 * KVM assumes that PAT entry '0' encodes WB memtype and simply zeroes
9196 	 * the PAT bits in SPTEs.  Bail if PAT[0] is programmed to something
9197 	 * other than WB.  Note, EPT doesn't utilize the PAT, but don't bother
9198 	 * with an exception.  PAT[0] is set to WB on RESET and also by the
9199 	 * kernel, i.e. failure indicates a kernel bug or broken firmware.
9200 	 */
9201 	if (rdmsrl_safe(MSR_IA32_CR_PAT, &host_pat) ||
9202 	    (host_pat & GENMASK(2, 0)) != 6) {
9203 		pr_err("kvm: host PAT[0] is not WB\n");
9204 		return -EIO;
9205 	}
9206 
9207 	x86_emulator_cache = kvm_alloc_emulator_cache();
9208 	if (!x86_emulator_cache) {
9209 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
9210 		return -ENOMEM;
9211 	}
9212 
9213 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
9214 	if (!user_return_msrs) {
9215 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
9216 		r = -ENOMEM;
9217 		goto out_free_x86_emulator_cache;
9218 	}
9219 	kvm_nr_uret_msrs = 0;
9220 
9221 	r = kvm_mmu_vendor_module_init();
9222 	if (r)
9223 		goto out_free_percpu;
9224 
9225 	kvm_timer_init();
9226 
9227 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
9228 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
9229 		kvm_caps.supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
9230 	}
9231 
9232 	if (pi_inject_timer == -1)
9233 		pi_inject_timer = housekeeping_enabled(HK_TYPE_TIMER);
9234 #ifdef CONFIG_X86_64
9235 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
9236 
9237 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9238 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
9239 #endif
9240 
9241 	return 0;
9242 
9243 out_free_percpu:
9244 	free_percpu(user_return_msrs);
9245 out_free_x86_emulator_cache:
9246 	kmem_cache_destroy(x86_emulator_cache);
9247 	return r;
9248 }
9249 
9250 void kvm_arch_exit(void)
9251 {
9252 #ifdef CONFIG_X86_64
9253 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9254 		clear_hv_tscchange_cb();
9255 #endif
9256 	kvm_lapic_exit();
9257 
9258 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
9259 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
9260 					    CPUFREQ_TRANSITION_NOTIFIER);
9261 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
9262 #ifdef CONFIG_X86_64
9263 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
9264 	irq_work_sync(&pvclock_irq_work);
9265 	cancel_work_sync(&pvclock_gtod_work);
9266 #endif
9267 	kvm_x86_ops.hardware_enable = NULL;
9268 	kvm_mmu_vendor_module_exit();
9269 	free_percpu(user_return_msrs);
9270 	kmem_cache_destroy(x86_emulator_cache);
9271 #ifdef CONFIG_KVM_XEN
9272 	static_key_deferred_flush(&kvm_xen_enabled);
9273 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
9274 #endif
9275 }
9276 
9277 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
9278 {
9279 	/*
9280 	 * The vCPU has halted, e.g. executed HLT.  Update the run state if the
9281 	 * local APIC is in-kernel, the run loop will detect the non-runnable
9282 	 * state and halt the vCPU.  Exit to userspace if the local APIC is
9283 	 * managed by userspace, in which case userspace is responsible for
9284 	 * handling wake events.
9285 	 */
9286 	++vcpu->stat.halt_exits;
9287 	if (lapic_in_kernel(vcpu)) {
9288 		vcpu->arch.mp_state = state;
9289 		return 1;
9290 	} else {
9291 		vcpu->run->exit_reason = reason;
9292 		return 0;
9293 	}
9294 }
9295 
9296 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
9297 {
9298 	return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
9299 }
9300 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
9301 
9302 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
9303 {
9304 	int ret = kvm_skip_emulated_instruction(vcpu);
9305 	/*
9306 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
9307 	 * KVM_EXIT_DEBUG here.
9308 	 */
9309 	return kvm_emulate_halt_noskip(vcpu) && ret;
9310 }
9311 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
9312 
9313 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
9314 {
9315 	int ret = kvm_skip_emulated_instruction(vcpu);
9316 
9317 	return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
9318 					KVM_EXIT_AP_RESET_HOLD) && ret;
9319 }
9320 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
9321 
9322 #ifdef CONFIG_X86_64
9323 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
9324 			        unsigned long clock_type)
9325 {
9326 	struct kvm_clock_pairing clock_pairing;
9327 	struct timespec64 ts;
9328 	u64 cycle;
9329 	int ret;
9330 
9331 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
9332 		return -KVM_EOPNOTSUPP;
9333 
9334 	/*
9335 	 * When tsc is in permanent catchup mode guests won't be able to use
9336 	 * pvclock_read_retry loop to get consistent view of pvclock
9337 	 */
9338 	if (vcpu->arch.tsc_always_catchup)
9339 		return -KVM_EOPNOTSUPP;
9340 
9341 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
9342 		return -KVM_EOPNOTSUPP;
9343 
9344 	clock_pairing.sec = ts.tv_sec;
9345 	clock_pairing.nsec = ts.tv_nsec;
9346 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
9347 	clock_pairing.flags = 0;
9348 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
9349 
9350 	ret = 0;
9351 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
9352 			    sizeof(struct kvm_clock_pairing)))
9353 		ret = -KVM_EFAULT;
9354 
9355 	return ret;
9356 }
9357 #endif
9358 
9359 /*
9360  * kvm_pv_kick_cpu_op:  Kick a vcpu.
9361  *
9362  * @apicid - apicid of vcpu to be kicked.
9363  */
9364 static void kvm_pv_kick_cpu_op(struct kvm *kvm, int apicid)
9365 {
9366 	/*
9367 	 * All other fields are unused for APIC_DM_REMRD, but may be consumed by
9368 	 * common code, e.g. for tracing. Defer initialization to the compiler.
9369 	 */
9370 	struct kvm_lapic_irq lapic_irq = {
9371 		.delivery_mode = APIC_DM_REMRD,
9372 		.dest_mode = APIC_DEST_PHYSICAL,
9373 		.shorthand = APIC_DEST_NOSHORT,
9374 		.dest_id = apicid,
9375 	};
9376 
9377 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
9378 }
9379 
9380 bool kvm_apicv_activated(struct kvm *kvm)
9381 {
9382 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
9383 }
9384 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
9385 
9386 bool kvm_vcpu_apicv_activated(struct kvm_vcpu *vcpu)
9387 {
9388 	ulong vm_reasons = READ_ONCE(vcpu->kvm->arch.apicv_inhibit_reasons);
9389 	ulong vcpu_reasons = static_call(kvm_x86_vcpu_get_apicv_inhibit_reasons)(vcpu);
9390 
9391 	return (vm_reasons | vcpu_reasons) == 0;
9392 }
9393 EXPORT_SYMBOL_GPL(kvm_vcpu_apicv_activated);
9394 
9395 static void set_or_clear_apicv_inhibit(unsigned long *inhibits,
9396 				       enum kvm_apicv_inhibit reason, bool set)
9397 {
9398 	if (set)
9399 		__set_bit(reason, inhibits);
9400 	else
9401 		__clear_bit(reason, inhibits);
9402 
9403 	trace_kvm_apicv_inhibit_changed(reason, set, *inhibits);
9404 }
9405 
9406 static void kvm_apicv_init(struct kvm *kvm)
9407 {
9408 	unsigned long *inhibits = &kvm->arch.apicv_inhibit_reasons;
9409 
9410 	init_rwsem(&kvm->arch.apicv_update_lock);
9411 
9412 	set_or_clear_apicv_inhibit(inhibits, APICV_INHIBIT_REASON_ABSENT, true);
9413 
9414 	if (!enable_apicv)
9415 		set_or_clear_apicv_inhibit(inhibits,
9416 					   APICV_INHIBIT_REASON_DISABLE, true);
9417 }
9418 
9419 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
9420 {
9421 	struct kvm_vcpu *target = NULL;
9422 	struct kvm_apic_map *map;
9423 
9424 	vcpu->stat.directed_yield_attempted++;
9425 
9426 	if (single_task_running())
9427 		goto no_yield;
9428 
9429 	rcu_read_lock();
9430 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
9431 
9432 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
9433 		target = map->phys_map[dest_id]->vcpu;
9434 
9435 	rcu_read_unlock();
9436 
9437 	if (!target || !READ_ONCE(target->ready))
9438 		goto no_yield;
9439 
9440 	/* Ignore requests to yield to self */
9441 	if (vcpu == target)
9442 		goto no_yield;
9443 
9444 	if (kvm_vcpu_yield_to(target) <= 0)
9445 		goto no_yield;
9446 
9447 	vcpu->stat.directed_yield_successful++;
9448 
9449 no_yield:
9450 	return;
9451 }
9452 
9453 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
9454 {
9455 	u64 ret = vcpu->run->hypercall.ret;
9456 
9457 	if (!is_64_bit_mode(vcpu))
9458 		ret = (u32)ret;
9459 	kvm_rax_write(vcpu, ret);
9460 	++vcpu->stat.hypercalls;
9461 	return kvm_skip_emulated_instruction(vcpu);
9462 }
9463 
9464 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
9465 {
9466 	unsigned long nr, a0, a1, a2, a3, ret;
9467 	int op_64_bit;
9468 
9469 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
9470 		return kvm_xen_hypercall(vcpu);
9471 
9472 	if (kvm_hv_hypercall_enabled(vcpu))
9473 		return kvm_hv_hypercall(vcpu);
9474 
9475 	nr = kvm_rax_read(vcpu);
9476 	a0 = kvm_rbx_read(vcpu);
9477 	a1 = kvm_rcx_read(vcpu);
9478 	a2 = kvm_rdx_read(vcpu);
9479 	a3 = kvm_rsi_read(vcpu);
9480 
9481 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
9482 
9483 	op_64_bit = is_64_bit_hypercall(vcpu);
9484 	if (!op_64_bit) {
9485 		nr &= 0xFFFFFFFF;
9486 		a0 &= 0xFFFFFFFF;
9487 		a1 &= 0xFFFFFFFF;
9488 		a2 &= 0xFFFFFFFF;
9489 		a3 &= 0xFFFFFFFF;
9490 	}
9491 
9492 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
9493 		ret = -KVM_EPERM;
9494 		goto out;
9495 	}
9496 
9497 	ret = -KVM_ENOSYS;
9498 
9499 	switch (nr) {
9500 	case KVM_HC_VAPIC_POLL_IRQ:
9501 		ret = 0;
9502 		break;
9503 	case KVM_HC_KICK_CPU:
9504 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
9505 			break;
9506 
9507 		kvm_pv_kick_cpu_op(vcpu->kvm, a1);
9508 		kvm_sched_yield(vcpu, a1);
9509 		ret = 0;
9510 		break;
9511 #ifdef CONFIG_X86_64
9512 	case KVM_HC_CLOCK_PAIRING:
9513 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
9514 		break;
9515 #endif
9516 	case KVM_HC_SEND_IPI:
9517 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
9518 			break;
9519 
9520 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
9521 		break;
9522 	case KVM_HC_SCHED_YIELD:
9523 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9524 			break;
9525 
9526 		kvm_sched_yield(vcpu, a0);
9527 		ret = 0;
9528 		break;
9529 	case KVM_HC_MAP_GPA_RANGE: {
9530 		u64 gpa = a0, npages = a1, attrs = a2;
9531 
9532 		ret = -KVM_ENOSYS;
9533 		if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9534 			break;
9535 
9536 		if (!PAGE_ALIGNED(gpa) || !npages ||
9537 		    gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9538 			ret = -KVM_EINVAL;
9539 			break;
9540 		}
9541 
9542 		vcpu->run->exit_reason        = KVM_EXIT_HYPERCALL;
9543 		vcpu->run->hypercall.nr       = KVM_HC_MAP_GPA_RANGE;
9544 		vcpu->run->hypercall.args[0]  = gpa;
9545 		vcpu->run->hypercall.args[1]  = npages;
9546 		vcpu->run->hypercall.args[2]  = attrs;
9547 		vcpu->run->hypercall.longmode = op_64_bit;
9548 		vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9549 		return 0;
9550 	}
9551 	default:
9552 		ret = -KVM_ENOSYS;
9553 		break;
9554 	}
9555 out:
9556 	if (!op_64_bit)
9557 		ret = (u32)ret;
9558 	kvm_rax_write(vcpu, ret);
9559 
9560 	++vcpu->stat.hypercalls;
9561 	return kvm_skip_emulated_instruction(vcpu);
9562 }
9563 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9564 
9565 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9566 {
9567 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9568 	char instruction[3];
9569 	unsigned long rip = kvm_rip_read(vcpu);
9570 
9571 	/*
9572 	 * If the quirk is disabled, synthesize a #UD and let the guest pick up
9573 	 * the pieces.
9574 	 */
9575 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_FIX_HYPERCALL_INSN)) {
9576 		ctxt->exception.error_code_valid = false;
9577 		ctxt->exception.vector = UD_VECTOR;
9578 		ctxt->have_exception = true;
9579 		return X86EMUL_PROPAGATE_FAULT;
9580 	}
9581 
9582 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9583 
9584 	return emulator_write_emulated(ctxt, rip, instruction, 3,
9585 		&ctxt->exception);
9586 }
9587 
9588 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9589 {
9590 	return vcpu->run->request_interrupt_window &&
9591 		likely(!pic_in_kernel(vcpu->kvm));
9592 }
9593 
9594 /* Called within kvm->srcu read side.  */
9595 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9596 {
9597 	struct kvm_run *kvm_run = vcpu->run;
9598 
9599 	kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9600 	kvm_run->cr8 = kvm_get_cr8(vcpu);
9601 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
9602 
9603 	kvm_run->ready_for_interrupt_injection =
9604 		pic_in_kernel(vcpu->kvm) ||
9605 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
9606 
9607 	if (is_smm(vcpu))
9608 		kvm_run->flags |= KVM_RUN_X86_SMM;
9609 }
9610 
9611 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9612 {
9613 	int max_irr, tpr;
9614 
9615 	if (!kvm_x86_ops.update_cr8_intercept)
9616 		return;
9617 
9618 	if (!lapic_in_kernel(vcpu))
9619 		return;
9620 
9621 	if (vcpu->arch.apic->apicv_active)
9622 		return;
9623 
9624 	if (!vcpu->arch.apic->vapic_addr)
9625 		max_irr = kvm_lapic_find_highest_irr(vcpu);
9626 	else
9627 		max_irr = -1;
9628 
9629 	if (max_irr != -1)
9630 		max_irr >>= 4;
9631 
9632 	tpr = kvm_lapic_get_cr8(vcpu);
9633 
9634 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9635 }
9636 
9637 
9638 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9639 {
9640 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9641 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
9642 		return 1;
9643 	}
9644 
9645 	return kvm_x86_ops.nested_ops->check_events(vcpu);
9646 }
9647 
9648 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9649 {
9650 	trace_kvm_inj_exception(vcpu->arch.exception.nr,
9651 				vcpu->arch.exception.has_error_code,
9652 				vcpu->arch.exception.error_code,
9653 				vcpu->arch.exception.injected);
9654 
9655 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9656 		vcpu->arch.exception.error_code = false;
9657 	static_call(kvm_x86_queue_exception)(vcpu);
9658 }
9659 
9660 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9661 {
9662 	int r;
9663 	bool can_inject = true;
9664 
9665 	/* try to reinject previous events if any */
9666 
9667 	if (vcpu->arch.exception.injected) {
9668 		kvm_inject_exception(vcpu);
9669 		can_inject = false;
9670 	}
9671 	/*
9672 	 * Do not inject an NMI or interrupt if there is a pending
9673 	 * exception.  Exceptions and interrupts are recognized at
9674 	 * instruction boundaries, i.e. the start of an instruction.
9675 	 * Trap-like exceptions, e.g. #DB, have higher priority than
9676 	 * NMIs and interrupts, i.e. traps are recognized before an
9677 	 * NMI/interrupt that's pending on the same instruction.
9678 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9679 	 * priority, but are only generated (pended) during instruction
9680 	 * execution, i.e. a pending fault-like exception means the
9681 	 * fault occurred on the *previous* instruction and must be
9682 	 * serviced prior to recognizing any new events in order to
9683 	 * fully complete the previous instruction.
9684 	 */
9685 	else if (!vcpu->arch.exception.pending) {
9686 		if (vcpu->arch.nmi_injected) {
9687 			static_call(kvm_x86_inject_nmi)(vcpu);
9688 			can_inject = false;
9689 		} else if (vcpu->arch.interrupt.injected) {
9690 			static_call(kvm_x86_inject_irq)(vcpu, true);
9691 			can_inject = false;
9692 		}
9693 	}
9694 
9695 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
9696 		     vcpu->arch.exception.pending);
9697 
9698 	/*
9699 	 * Call check_nested_events() even if we reinjected a previous event
9700 	 * in order for caller to determine if it should require immediate-exit
9701 	 * from L2 to L1 due to pending L1 events which require exit
9702 	 * from L2 to L1.
9703 	 */
9704 	if (is_guest_mode(vcpu)) {
9705 		r = kvm_check_nested_events(vcpu);
9706 		if (r < 0)
9707 			goto out;
9708 	}
9709 
9710 	/* try to inject new event if pending */
9711 	if (vcpu->arch.exception.pending) {
9712 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9713 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9714 					     X86_EFLAGS_RF);
9715 
9716 		if (vcpu->arch.exception.nr == DB_VECTOR) {
9717 			kvm_deliver_exception_payload(vcpu);
9718 			if (vcpu->arch.dr7 & DR7_GD) {
9719 				vcpu->arch.dr7 &= ~DR7_GD;
9720 				kvm_update_dr7(vcpu);
9721 			}
9722 		}
9723 
9724 		kvm_inject_exception(vcpu);
9725 
9726 		vcpu->arch.exception.pending = false;
9727 		vcpu->arch.exception.injected = true;
9728 
9729 		can_inject = false;
9730 	}
9731 
9732 	/* Don't inject interrupts if the user asked to avoid doing so */
9733 	if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9734 		return 0;
9735 
9736 	/*
9737 	 * Finally, inject interrupt events.  If an event cannot be injected
9738 	 * due to architectural conditions (e.g. IF=0) a window-open exit
9739 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
9740 	 * and can architecturally be injected, but we cannot do it right now:
9741 	 * an interrupt could have arrived just now and we have to inject it
9742 	 * as a vmexit, or there could already an event in the queue, which is
9743 	 * indicated by can_inject.  In that case we request an immediate exit
9744 	 * in order to make progress and get back here for another iteration.
9745 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9746 	 */
9747 	if (vcpu->arch.smi_pending) {
9748 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9749 		if (r < 0)
9750 			goto out;
9751 		if (r) {
9752 			vcpu->arch.smi_pending = false;
9753 			++vcpu->arch.smi_count;
9754 			enter_smm(vcpu);
9755 			can_inject = false;
9756 		} else
9757 			static_call(kvm_x86_enable_smi_window)(vcpu);
9758 	}
9759 
9760 	if (vcpu->arch.nmi_pending) {
9761 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9762 		if (r < 0)
9763 			goto out;
9764 		if (r) {
9765 			--vcpu->arch.nmi_pending;
9766 			vcpu->arch.nmi_injected = true;
9767 			static_call(kvm_x86_inject_nmi)(vcpu);
9768 			can_inject = false;
9769 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9770 		}
9771 		if (vcpu->arch.nmi_pending)
9772 			static_call(kvm_x86_enable_nmi_window)(vcpu);
9773 	}
9774 
9775 	if (kvm_cpu_has_injectable_intr(vcpu)) {
9776 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9777 		if (r < 0)
9778 			goto out;
9779 		if (r) {
9780 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9781 			static_call(kvm_x86_inject_irq)(vcpu, false);
9782 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9783 		}
9784 		if (kvm_cpu_has_injectable_intr(vcpu))
9785 			static_call(kvm_x86_enable_irq_window)(vcpu);
9786 	}
9787 
9788 	if (is_guest_mode(vcpu) &&
9789 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
9790 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9791 		*req_immediate_exit = true;
9792 
9793 	WARN_ON(vcpu->arch.exception.pending);
9794 	return 0;
9795 
9796 out:
9797 	if (r == -EBUSY) {
9798 		*req_immediate_exit = true;
9799 		r = 0;
9800 	}
9801 	return r;
9802 }
9803 
9804 static void process_nmi(struct kvm_vcpu *vcpu)
9805 {
9806 	unsigned limit = 2;
9807 
9808 	/*
9809 	 * x86 is limited to one NMI running, and one NMI pending after it.
9810 	 * If an NMI is already in progress, limit further NMIs to just one.
9811 	 * Otherwise, allow two (and we'll inject the first one immediately).
9812 	 */
9813 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9814 		limit = 1;
9815 
9816 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9817 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9818 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9819 }
9820 
9821 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9822 {
9823 	u32 flags = 0;
9824 	flags |= seg->g       << 23;
9825 	flags |= seg->db      << 22;
9826 	flags |= seg->l       << 21;
9827 	flags |= seg->avl     << 20;
9828 	flags |= seg->present << 15;
9829 	flags |= seg->dpl     << 13;
9830 	flags |= seg->s       << 12;
9831 	flags |= seg->type    << 8;
9832 	return flags;
9833 }
9834 
9835 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9836 {
9837 	struct kvm_segment seg;
9838 	int offset;
9839 
9840 	kvm_get_segment(vcpu, &seg, n);
9841 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9842 
9843 	if (n < 3)
9844 		offset = 0x7f84 + n * 12;
9845 	else
9846 		offset = 0x7f2c + (n - 3) * 12;
9847 
9848 	put_smstate(u32, buf, offset + 8, seg.base);
9849 	put_smstate(u32, buf, offset + 4, seg.limit);
9850 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9851 }
9852 
9853 #ifdef CONFIG_X86_64
9854 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9855 {
9856 	struct kvm_segment seg;
9857 	int offset;
9858 	u16 flags;
9859 
9860 	kvm_get_segment(vcpu, &seg, n);
9861 	offset = 0x7e00 + n * 16;
9862 
9863 	flags = enter_smm_get_segment_flags(&seg) >> 8;
9864 	put_smstate(u16, buf, offset, seg.selector);
9865 	put_smstate(u16, buf, offset + 2, flags);
9866 	put_smstate(u32, buf, offset + 4, seg.limit);
9867 	put_smstate(u64, buf, offset + 8, seg.base);
9868 }
9869 #endif
9870 
9871 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9872 {
9873 	struct desc_ptr dt;
9874 	struct kvm_segment seg;
9875 	unsigned long val;
9876 	int i;
9877 
9878 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9879 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9880 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9881 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9882 
9883 	for (i = 0; i < 8; i++)
9884 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9885 
9886 	kvm_get_dr(vcpu, 6, &val);
9887 	put_smstate(u32, buf, 0x7fcc, (u32)val);
9888 	kvm_get_dr(vcpu, 7, &val);
9889 	put_smstate(u32, buf, 0x7fc8, (u32)val);
9890 
9891 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9892 	put_smstate(u32, buf, 0x7fc4, seg.selector);
9893 	put_smstate(u32, buf, 0x7f64, seg.base);
9894 	put_smstate(u32, buf, 0x7f60, seg.limit);
9895 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9896 
9897 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9898 	put_smstate(u32, buf, 0x7fc0, seg.selector);
9899 	put_smstate(u32, buf, 0x7f80, seg.base);
9900 	put_smstate(u32, buf, 0x7f7c, seg.limit);
9901 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9902 
9903 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9904 	put_smstate(u32, buf, 0x7f74, dt.address);
9905 	put_smstate(u32, buf, 0x7f70, dt.size);
9906 
9907 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9908 	put_smstate(u32, buf, 0x7f58, dt.address);
9909 	put_smstate(u32, buf, 0x7f54, dt.size);
9910 
9911 	for (i = 0; i < 6; i++)
9912 		enter_smm_save_seg_32(vcpu, buf, i);
9913 
9914 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9915 
9916 	/* revision id */
9917 	put_smstate(u32, buf, 0x7efc, 0x00020000);
9918 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9919 }
9920 
9921 #ifdef CONFIG_X86_64
9922 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9923 {
9924 	struct desc_ptr dt;
9925 	struct kvm_segment seg;
9926 	unsigned long val;
9927 	int i;
9928 
9929 	for (i = 0; i < 16; i++)
9930 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9931 
9932 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9933 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9934 
9935 	kvm_get_dr(vcpu, 6, &val);
9936 	put_smstate(u64, buf, 0x7f68, val);
9937 	kvm_get_dr(vcpu, 7, &val);
9938 	put_smstate(u64, buf, 0x7f60, val);
9939 
9940 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9941 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9942 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9943 
9944 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9945 
9946 	/* revision id */
9947 	put_smstate(u32, buf, 0x7efc, 0x00020064);
9948 
9949 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9950 
9951 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9952 	put_smstate(u16, buf, 0x7e90, seg.selector);
9953 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9954 	put_smstate(u32, buf, 0x7e94, seg.limit);
9955 	put_smstate(u64, buf, 0x7e98, seg.base);
9956 
9957 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9958 	put_smstate(u32, buf, 0x7e84, dt.size);
9959 	put_smstate(u64, buf, 0x7e88, dt.address);
9960 
9961 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9962 	put_smstate(u16, buf, 0x7e70, seg.selector);
9963 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9964 	put_smstate(u32, buf, 0x7e74, seg.limit);
9965 	put_smstate(u64, buf, 0x7e78, seg.base);
9966 
9967 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9968 	put_smstate(u32, buf, 0x7e64, dt.size);
9969 	put_smstate(u64, buf, 0x7e68, dt.address);
9970 
9971 	for (i = 0; i < 6; i++)
9972 		enter_smm_save_seg_64(vcpu, buf, i);
9973 }
9974 #endif
9975 
9976 static void enter_smm(struct kvm_vcpu *vcpu)
9977 {
9978 	struct kvm_segment cs, ds;
9979 	struct desc_ptr dt;
9980 	unsigned long cr0;
9981 	char buf[512];
9982 
9983 	memset(buf, 0, 512);
9984 #ifdef CONFIG_X86_64
9985 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9986 		enter_smm_save_state_64(vcpu, buf);
9987 	else
9988 #endif
9989 		enter_smm_save_state_32(vcpu, buf);
9990 
9991 	/*
9992 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9993 	 * state (e.g. leave guest mode) after we've saved the state into the
9994 	 * SMM state-save area.
9995 	 */
9996 	static_call(kvm_x86_enter_smm)(vcpu, buf);
9997 
9998 	kvm_smm_changed(vcpu, true);
9999 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
10000 
10001 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
10002 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
10003 	else
10004 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
10005 
10006 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
10007 	kvm_rip_write(vcpu, 0x8000);
10008 
10009 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
10010 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
10011 	vcpu->arch.cr0 = cr0;
10012 
10013 	static_call(kvm_x86_set_cr4)(vcpu, 0);
10014 
10015 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
10016 	dt.address = dt.size = 0;
10017 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10018 
10019 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
10020 
10021 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
10022 	cs.base = vcpu->arch.smbase;
10023 
10024 	ds.selector = 0;
10025 	ds.base = 0;
10026 
10027 	cs.limit    = ds.limit = 0xffffffff;
10028 	cs.type     = ds.type = 0x3;
10029 	cs.dpl      = ds.dpl = 0;
10030 	cs.db       = ds.db = 0;
10031 	cs.s        = ds.s = 1;
10032 	cs.l        = ds.l = 0;
10033 	cs.g        = ds.g = 1;
10034 	cs.avl      = ds.avl = 0;
10035 	cs.present  = ds.present = 1;
10036 	cs.unusable = ds.unusable = 0;
10037 	cs.padding  = ds.padding = 0;
10038 
10039 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10040 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
10041 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
10042 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
10043 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
10044 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
10045 
10046 #ifdef CONFIG_X86_64
10047 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
10048 		static_call(kvm_x86_set_efer)(vcpu, 0);
10049 #endif
10050 
10051 	kvm_update_cpuid_runtime(vcpu);
10052 	kvm_mmu_reset_context(vcpu);
10053 }
10054 
10055 static void process_smi(struct kvm_vcpu *vcpu)
10056 {
10057 	vcpu->arch.smi_pending = true;
10058 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10059 }
10060 
10061 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
10062 				       unsigned long *vcpu_bitmap)
10063 {
10064 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
10065 }
10066 
10067 void kvm_make_scan_ioapic_request(struct kvm *kvm)
10068 {
10069 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
10070 }
10071 
10072 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10073 {
10074 	struct kvm_lapic *apic = vcpu->arch.apic;
10075 	bool activate;
10076 
10077 	if (!lapic_in_kernel(vcpu))
10078 		return;
10079 
10080 	down_read(&vcpu->kvm->arch.apicv_update_lock);
10081 	preempt_disable();
10082 
10083 	/* Do not activate APICV when APIC is disabled */
10084 	activate = kvm_vcpu_apicv_activated(vcpu) &&
10085 		   (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED);
10086 
10087 	if (apic->apicv_active == activate)
10088 		goto out;
10089 
10090 	apic->apicv_active = activate;
10091 	kvm_apic_update_apicv(vcpu);
10092 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
10093 
10094 	/*
10095 	 * When APICv gets disabled, we may still have injected interrupts
10096 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
10097 	 * still active when the interrupt got accepted. Make sure
10098 	 * inject_pending_event() is called to check for that.
10099 	 */
10100 	if (!apic->apicv_active)
10101 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10102 
10103 out:
10104 	preempt_enable();
10105 	up_read(&vcpu->kvm->arch.apicv_update_lock);
10106 }
10107 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
10108 
10109 void __kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10110 				      enum kvm_apicv_inhibit reason, bool set)
10111 {
10112 	unsigned long old, new;
10113 
10114 	lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
10115 
10116 	if (!static_call(kvm_x86_check_apicv_inhibit_reasons)(reason))
10117 		return;
10118 
10119 	old = new = kvm->arch.apicv_inhibit_reasons;
10120 
10121 	set_or_clear_apicv_inhibit(&new, reason, set);
10122 
10123 	if (!!old != !!new) {
10124 		/*
10125 		 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
10126 		 * false positives in the sanity check WARN in svm_vcpu_run().
10127 		 * This task will wait for all vCPUs to ack the kick IRQ before
10128 		 * updating apicv_inhibit_reasons, and all other vCPUs will
10129 		 * block on acquiring apicv_update_lock so that vCPUs can't
10130 		 * redo svm_vcpu_run() without seeing the new inhibit state.
10131 		 *
10132 		 * Note, holding apicv_update_lock and taking it in the read
10133 		 * side (handling the request) also prevents other vCPUs from
10134 		 * servicing the request with a stale apicv_inhibit_reasons.
10135 		 */
10136 		kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
10137 		kvm->arch.apicv_inhibit_reasons = new;
10138 		if (new) {
10139 			unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
10140 			kvm_zap_gfn_range(kvm, gfn, gfn+1);
10141 		}
10142 	} else {
10143 		kvm->arch.apicv_inhibit_reasons = new;
10144 	}
10145 }
10146 
10147 void kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10148 				    enum kvm_apicv_inhibit reason, bool set)
10149 {
10150 	if (!enable_apicv)
10151 		return;
10152 
10153 	down_write(&kvm->arch.apicv_update_lock);
10154 	__kvm_set_or_clear_apicv_inhibit(kvm, reason, set);
10155 	up_write(&kvm->arch.apicv_update_lock);
10156 }
10157 EXPORT_SYMBOL_GPL(kvm_set_or_clear_apicv_inhibit);
10158 
10159 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
10160 {
10161 	if (!kvm_apic_present(vcpu))
10162 		return;
10163 
10164 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
10165 
10166 	if (irqchip_split(vcpu->kvm))
10167 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
10168 	else {
10169 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10170 		if (ioapic_in_kernel(vcpu->kvm))
10171 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
10172 	}
10173 
10174 	if (is_guest_mode(vcpu))
10175 		vcpu->arch.load_eoi_exitmap_pending = true;
10176 	else
10177 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
10178 }
10179 
10180 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
10181 {
10182 	u64 eoi_exit_bitmap[4];
10183 
10184 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
10185 		return;
10186 
10187 	if (to_hv_vcpu(vcpu)) {
10188 		bitmap_or((ulong *)eoi_exit_bitmap,
10189 			  vcpu->arch.ioapic_handled_vectors,
10190 			  to_hv_synic(vcpu)->vec_bitmap, 256);
10191 		static_call_cond(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
10192 		return;
10193 	}
10194 
10195 	static_call_cond(kvm_x86_load_eoi_exitmap)(
10196 		vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
10197 }
10198 
10199 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
10200 					    unsigned long start, unsigned long end)
10201 {
10202 	unsigned long apic_address;
10203 
10204 	/*
10205 	 * The physical address of apic access page is stored in the VMCS.
10206 	 * Update it when it becomes invalid.
10207 	 */
10208 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
10209 	if (start <= apic_address && apic_address < end)
10210 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
10211 }
10212 
10213 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
10214 {
10215 	static_call_cond(kvm_x86_guest_memory_reclaimed)(kvm);
10216 }
10217 
10218 static void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
10219 {
10220 	if (!lapic_in_kernel(vcpu))
10221 		return;
10222 
10223 	static_call_cond(kvm_x86_set_apic_access_page_addr)(vcpu);
10224 }
10225 
10226 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
10227 {
10228 	smp_send_reschedule(vcpu->cpu);
10229 }
10230 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
10231 
10232 /*
10233  * Called within kvm->srcu read side.
10234  * Returns 1 to let vcpu_run() continue the guest execution loop without
10235  * exiting to the userspace.  Otherwise, the value will be returned to the
10236  * userspace.
10237  */
10238 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
10239 {
10240 	int r;
10241 	bool req_int_win =
10242 		dm_request_for_irq_injection(vcpu) &&
10243 		kvm_cpu_accept_dm_intr(vcpu);
10244 	fastpath_t exit_fastpath;
10245 
10246 	bool req_immediate_exit = false;
10247 
10248 	/* Forbid vmenter if vcpu dirty ring is soft-full */
10249 	if (unlikely(vcpu->kvm->dirty_ring_size &&
10250 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
10251 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
10252 		trace_kvm_dirty_ring_exit(vcpu);
10253 		r = 0;
10254 		goto out;
10255 	}
10256 
10257 	if (kvm_request_pending(vcpu)) {
10258 		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
10259 			r = -EIO;
10260 			goto out;
10261 		}
10262 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
10263 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
10264 				r = 0;
10265 				goto out;
10266 			}
10267 		}
10268 		if (kvm_check_request(KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, vcpu))
10269 			kvm_mmu_free_obsolete_roots(vcpu);
10270 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
10271 			__kvm_migrate_timers(vcpu);
10272 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
10273 			kvm_update_masterclock(vcpu->kvm);
10274 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
10275 			kvm_gen_kvmclock_update(vcpu);
10276 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
10277 			r = kvm_guest_time_update(vcpu);
10278 			if (unlikely(r))
10279 				goto out;
10280 		}
10281 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
10282 			kvm_mmu_sync_roots(vcpu);
10283 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
10284 			kvm_mmu_load_pgd(vcpu);
10285 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
10286 			kvm_vcpu_flush_tlb_all(vcpu);
10287 
10288 			/* Flushing all ASIDs flushes the current ASID... */
10289 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
10290 		}
10291 		kvm_service_local_tlb_flush_requests(vcpu);
10292 
10293 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
10294 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
10295 			r = 0;
10296 			goto out;
10297 		}
10298 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10299 			if (is_guest_mode(vcpu)) {
10300 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
10301 			} else {
10302 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
10303 				vcpu->mmio_needed = 0;
10304 				r = 0;
10305 				goto out;
10306 			}
10307 		}
10308 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
10309 			/* Page is swapped out. Do synthetic halt */
10310 			vcpu->arch.apf.halted = true;
10311 			r = 1;
10312 			goto out;
10313 		}
10314 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
10315 			record_steal_time(vcpu);
10316 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
10317 			process_smi(vcpu);
10318 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
10319 			process_nmi(vcpu);
10320 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
10321 			kvm_pmu_handle_event(vcpu);
10322 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
10323 			kvm_pmu_deliver_pmi(vcpu);
10324 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
10325 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
10326 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
10327 				     vcpu->arch.ioapic_handled_vectors)) {
10328 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
10329 				vcpu->run->eoi.vector =
10330 						vcpu->arch.pending_ioapic_eoi;
10331 				r = 0;
10332 				goto out;
10333 			}
10334 		}
10335 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
10336 			vcpu_scan_ioapic(vcpu);
10337 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
10338 			vcpu_load_eoi_exitmap(vcpu);
10339 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
10340 			kvm_vcpu_reload_apic_access_page(vcpu);
10341 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
10342 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10343 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
10344 			vcpu->run->system_event.ndata = 0;
10345 			r = 0;
10346 			goto out;
10347 		}
10348 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
10349 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10350 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
10351 			vcpu->run->system_event.ndata = 0;
10352 			r = 0;
10353 			goto out;
10354 		}
10355 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
10356 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
10357 
10358 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
10359 			vcpu->run->hyperv = hv_vcpu->exit;
10360 			r = 0;
10361 			goto out;
10362 		}
10363 
10364 		/*
10365 		 * KVM_REQ_HV_STIMER has to be processed after
10366 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
10367 		 * depend on the guest clock being up-to-date
10368 		 */
10369 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
10370 			kvm_hv_process_stimers(vcpu);
10371 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
10372 			kvm_vcpu_update_apicv(vcpu);
10373 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
10374 			kvm_check_async_pf_completion(vcpu);
10375 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
10376 			static_call(kvm_x86_msr_filter_changed)(vcpu);
10377 
10378 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
10379 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
10380 	}
10381 
10382 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
10383 	    kvm_xen_has_interrupt(vcpu)) {
10384 		++vcpu->stat.req_event;
10385 		r = kvm_apic_accept_events(vcpu);
10386 		if (r < 0) {
10387 			r = 0;
10388 			goto out;
10389 		}
10390 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
10391 			r = 1;
10392 			goto out;
10393 		}
10394 
10395 		r = inject_pending_event(vcpu, &req_immediate_exit);
10396 		if (r < 0) {
10397 			r = 0;
10398 			goto out;
10399 		}
10400 		if (req_int_win)
10401 			static_call(kvm_x86_enable_irq_window)(vcpu);
10402 
10403 		if (kvm_lapic_enabled(vcpu)) {
10404 			update_cr8_intercept(vcpu);
10405 			kvm_lapic_sync_to_vapic(vcpu);
10406 		}
10407 	}
10408 
10409 	r = kvm_mmu_reload(vcpu);
10410 	if (unlikely(r)) {
10411 		goto cancel_injection;
10412 	}
10413 
10414 	preempt_disable();
10415 
10416 	static_call(kvm_x86_prepare_switch_to_guest)(vcpu);
10417 
10418 	/*
10419 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
10420 	 * IPI are then delayed after guest entry, which ensures that they
10421 	 * result in virtual interrupt delivery.
10422 	 */
10423 	local_irq_disable();
10424 
10425 	/* Store vcpu->apicv_active before vcpu->mode.  */
10426 	smp_store_release(&vcpu->mode, IN_GUEST_MODE);
10427 
10428 	kvm_vcpu_srcu_read_unlock(vcpu);
10429 
10430 	/*
10431 	 * 1) We should set ->mode before checking ->requests.  Please see
10432 	 * the comment in kvm_vcpu_exiting_guest_mode().
10433 	 *
10434 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
10435 	 * pairs with the memory barrier implicit in pi_test_and_set_on
10436 	 * (see vmx_deliver_posted_interrupt).
10437 	 *
10438 	 * 3) This also orders the write to mode from any reads to the page
10439 	 * tables done while the VCPU is running.  Please see the comment
10440 	 * in kvm_flush_remote_tlbs.
10441 	 */
10442 	smp_mb__after_srcu_read_unlock();
10443 
10444 	/*
10445 	 * Process pending posted interrupts to handle the case where the
10446 	 * notification IRQ arrived in the host, or was never sent (because the
10447 	 * target vCPU wasn't running).  Do this regardless of the vCPU's APICv
10448 	 * status, KVM doesn't update assigned devices when APICv is inhibited,
10449 	 * i.e. they can post interrupts even if APICv is temporarily disabled.
10450 	 */
10451 	if (kvm_lapic_enabled(vcpu))
10452 		static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10453 
10454 	if (kvm_vcpu_exit_request(vcpu)) {
10455 		vcpu->mode = OUTSIDE_GUEST_MODE;
10456 		smp_wmb();
10457 		local_irq_enable();
10458 		preempt_enable();
10459 		kvm_vcpu_srcu_read_lock(vcpu);
10460 		r = 1;
10461 		goto cancel_injection;
10462 	}
10463 
10464 	if (req_immediate_exit) {
10465 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10466 		static_call(kvm_x86_request_immediate_exit)(vcpu);
10467 	}
10468 
10469 	fpregs_assert_state_consistent();
10470 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
10471 		switch_fpu_return();
10472 
10473 	if (vcpu->arch.guest_fpu.xfd_err)
10474 		wrmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
10475 
10476 	if (unlikely(vcpu->arch.switch_db_regs)) {
10477 		set_debugreg(0, 7);
10478 		set_debugreg(vcpu->arch.eff_db[0], 0);
10479 		set_debugreg(vcpu->arch.eff_db[1], 1);
10480 		set_debugreg(vcpu->arch.eff_db[2], 2);
10481 		set_debugreg(vcpu->arch.eff_db[3], 3);
10482 	} else if (unlikely(hw_breakpoint_active())) {
10483 		set_debugreg(0, 7);
10484 	}
10485 
10486 	guest_timing_enter_irqoff();
10487 
10488 	for (;;) {
10489 		/*
10490 		 * Assert that vCPU vs. VM APICv state is consistent.  An APICv
10491 		 * update must kick and wait for all vCPUs before toggling the
10492 		 * per-VM state, and responsing vCPUs must wait for the update
10493 		 * to complete before servicing KVM_REQ_APICV_UPDATE.
10494 		 */
10495 		WARN_ON_ONCE((kvm_vcpu_apicv_activated(vcpu) != kvm_vcpu_apicv_active(vcpu)) &&
10496 			     (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED));
10497 
10498 		exit_fastpath = static_call(kvm_x86_vcpu_run)(vcpu);
10499 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
10500 			break;
10501 
10502 		if (kvm_lapic_enabled(vcpu))
10503 			static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10504 
10505 		if (unlikely(kvm_vcpu_exit_request(vcpu))) {
10506 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
10507 			break;
10508 		}
10509 	}
10510 
10511 	/*
10512 	 * Do this here before restoring debug registers on the host.  And
10513 	 * since we do this before handling the vmexit, a DR access vmexit
10514 	 * can (a) read the correct value of the debug registers, (b) set
10515 	 * KVM_DEBUGREG_WONT_EXIT again.
10516 	 */
10517 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
10518 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
10519 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
10520 		kvm_update_dr0123(vcpu);
10521 		kvm_update_dr7(vcpu);
10522 	}
10523 
10524 	/*
10525 	 * If the guest has used debug registers, at least dr7
10526 	 * will be disabled while returning to the host.
10527 	 * If we don't have active breakpoints in the host, we don't
10528 	 * care about the messed up debug address registers. But if
10529 	 * we have some of them active, restore the old state.
10530 	 */
10531 	if (hw_breakpoint_active())
10532 		hw_breakpoint_restore();
10533 
10534 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
10535 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
10536 
10537 	vcpu->mode = OUTSIDE_GUEST_MODE;
10538 	smp_wmb();
10539 
10540 	/*
10541 	 * Sync xfd before calling handle_exit_irqoff() which may
10542 	 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
10543 	 * in #NM irqoff handler).
10544 	 */
10545 	if (vcpu->arch.xfd_no_write_intercept)
10546 		fpu_sync_guest_vmexit_xfd_state();
10547 
10548 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
10549 
10550 	if (vcpu->arch.guest_fpu.xfd_err)
10551 		wrmsrl(MSR_IA32_XFD_ERR, 0);
10552 
10553 	/*
10554 	 * Consume any pending interrupts, including the possible source of
10555 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10556 	 * An instruction is required after local_irq_enable() to fully unblock
10557 	 * interrupts on processors that implement an interrupt shadow, the
10558 	 * stat.exits increment will do nicely.
10559 	 */
10560 	kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
10561 	local_irq_enable();
10562 	++vcpu->stat.exits;
10563 	local_irq_disable();
10564 	kvm_after_interrupt(vcpu);
10565 
10566 	/*
10567 	 * Wait until after servicing IRQs to account guest time so that any
10568 	 * ticks that occurred while running the guest are properly accounted
10569 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
10570 	 * of accounting via context tracking, but the loss of accuracy is
10571 	 * acceptable for all known use cases.
10572 	 */
10573 	guest_timing_exit_irqoff();
10574 
10575 	local_irq_enable();
10576 	preempt_enable();
10577 
10578 	kvm_vcpu_srcu_read_lock(vcpu);
10579 
10580 	/*
10581 	 * Profile KVM exit RIPs:
10582 	 */
10583 	if (unlikely(prof_on == KVM_PROFILING)) {
10584 		unsigned long rip = kvm_rip_read(vcpu);
10585 		profile_hit(KVM_PROFILING, (void *)rip);
10586 	}
10587 
10588 	if (unlikely(vcpu->arch.tsc_always_catchup))
10589 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10590 
10591 	if (vcpu->arch.apic_attention)
10592 		kvm_lapic_sync_from_vapic(vcpu);
10593 
10594 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10595 	return r;
10596 
10597 cancel_injection:
10598 	if (req_immediate_exit)
10599 		kvm_make_request(KVM_REQ_EVENT, vcpu);
10600 	static_call(kvm_x86_cancel_injection)(vcpu);
10601 	if (unlikely(vcpu->arch.apic_attention))
10602 		kvm_lapic_sync_from_vapic(vcpu);
10603 out:
10604 	return r;
10605 }
10606 
10607 /* Called within kvm->srcu read side.  */
10608 static inline int vcpu_block(struct kvm_vcpu *vcpu)
10609 {
10610 	bool hv_timer;
10611 
10612 	if (!kvm_arch_vcpu_runnable(vcpu)) {
10613 		/*
10614 		 * Switch to the software timer before halt-polling/blocking as
10615 		 * the guest's timer may be a break event for the vCPU, and the
10616 		 * hypervisor timer runs only when the CPU is in guest mode.
10617 		 * Switch before halt-polling so that KVM recognizes an expired
10618 		 * timer before blocking.
10619 		 */
10620 		hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
10621 		if (hv_timer)
10622 			kvm_lapic_switch_to_sw_timer(vcpu);
10623 
10624 		kvm_vcpu_srcu_read_unlock(vcpu);
10625 		if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10626 			kvm_vcpu_halt(vcpu);
10627 		else
10628 			kvm_vcpu_block(vcpu);
10629 		kvm_vcpu_srcu_read_lock(vcpu);
10630 
10631 		if (hv_timer)
10632 			kvm_lapic_switch_to_hv_timer(vcpu);
10633 
10634 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10635 			return 1;
10636 	}
10637 
10638 	if (kvm_apic_accept_events(vcpu) < 0)
10639 		return 0;
10640 	switch(vcpu->arch.mp_state) {
10641 	case KVM_MP_STATE_HALTED:
10642 	case KVM_MP_STATE_AP_RESET_HOLD:
10643 		vcpu->arch.pv.pv_unhalted = false;
10644 		vcpu->arch.mp_state =
10645 			KVM_MP_STATE_RUNNABLE;
10646 		fallthrough;
10647 	case KVM_MP_STATE_RUNNABLE:
10648 		vcpu->arch.apf.halted = false;
10649 		break;
10650 	case KVM_MP_STATE_INIT_RECEIVED:
10651 		break;
10652 	default:
10653 		return -EINTR;
10654 	}
10655 	return 1;
10656 }
10657 
10658 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10659 {
10660 	if (is_guest_mode(vcpu))
10661 		kvm_check_nested_events(vcpu);
10662 
10663 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10664 		!vcpu->arch.apf.halted);
10665 }
10666 
10667 /* Called within kvm->srcu read side.  */
10668 static int vcpu_run(struct kvm_vcpu *vcpu)
10669 {
10670 	int r;
10671 
10672 	vcpu->arch.l1tf_flush_l1d = true;
10673 
10674 	for (;;) {
10675 		/*
10676 		 * If another guest vCPU requests a PV TLB flush in the middle
10677 		 * of instruction emulation, the rest of the emulation could
10678 		 * use a stale page translation. Assume that any code after
10679 		 * this point can start executing an instruction.
10680 		 */
10681 		vcpu->arch.at_instruction_boundary = false;
10682 		if (kvm_vcpu_running(vcpu)) {
10683 			r = vcpu_enter_guest(vcpu);
10684 		} else {
10685 			r = vcpu_block(vcpu);
10686 		}
10687 
10688 		if (r <= 0)
10689 			break;
10690 
10691 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10692 		if (kvm_xen_has_pending_events(vcpu))
10693 			kvm_xen_inject_pending_events(vcpu);
10694 
10695 		if (kvm_cpu_has_pending_timer(vcpu))
10696 			kvm_inject_pending_timer_irqs(vcpu);
10697 
10698 		if (dm_request_for_irq_injection(vcpu) &&
10699 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10700 			r = 0;
10701 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10702 			++vcpu->stat.request_irq_exits;
10703 			break;
10704 		}
10705 
10706 		if (__xfer_to_guest_mode_work_pending()) {
10707 			kvm_vcpu_srcu_read_unlock(vcpu);
10708 			r = xfer_to_guest_mode_handle_work(vcpu);
10709 			kvm_vcpu_srcu_read_lock(vcpu);
10710 			if (r)
10711 				return r;
10712 		}
10713 	}
10714 
10715 	return r;
10716 }
10717 
10718 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10719 {
10720 	return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10721 }
10722 
10723 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10724 {
10725 	BUG_ON(!vcpu->arch.pio.count);
10726 
10727 	return complete_emulated_io(vcpu);
10728 }
10729 
10730 /*
10731  * Implements the following, as a state machine:
10732  *
10733  * read:
10734  *   for each fragment
10735  *     for each mmio piece in the fragment
10736  *       write gpa, len
10737  *       exit
10738  *       copy data
10739  *   execute insn
10740  *
10741  * write:
10742  *   for each fragment
10743  *     for each mmio piece in the fragment
10744  *       write gpa, len
10745  *       copy data
10746  *       exit
10747  */
10748 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10749 {
10750 	struct kvm_run *run = vcpu->run;
10751 	struct kvm_mmio_fragment *frag;
10752 	unsigned len;
10753 
10754 	BUG_ON(!vcpu->mmio_needed);
10755 
10756 	/* Complete previous fragment */
10757 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10758 	len = min(8u, frag->len);
10759 	if (!vcpu->mmio_is_write)
10760 		memcpy(frag->data, run->mmio.data, len);
10761 
10762 	if (frag->len <= 8) {
10763 		/* Switch to the next fragment. */
10764 		frag++;
10765 		vcpu->mmio_cur_fragment++;
10766 	} else {
10767 		/* Go forward to the next mmio piece. */
10768 		frag->data += len;
10769 		frag->gpa += len;
10770 		frag->len -= len;
10771 	}
10772 
10773 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10774 		vcpu->mmio_needed = 0;
10775 
10776 		/* FIXME: return into emulator if single-stepping.  */
10777 		if (vcpu->mmio_is_write)
10778 			return 1;
10779 		vcpu->mmio_read_completed = 1;
10780 		return complete_emulated_io(vcpu);
10781 	}
10782 
10783 	run->exit_reason = KVM_EXIT_MMIO;
10784 	run->mmio.phys_addr = frag->gpa;
10785 	if (vcpu->mmio_is_write)
10786 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10787 	run->mmio.len = min(8u, frag->len);
10788 	run->mmio.is_write = vcpu->mmio_is_write;
10789 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10790 	return 0;
10791 }
10792 
10793 /* Swap (qemu) user FPU context for the guest FPU context. */
10794 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10795 {
10796 	/* Exclude PKRU, it's restored separately immediately after VM-Exit. */
10797 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10798 	trace_kvm_fpu(1);
10799 }
10800 
10801 /* When vcpu_run ends, restore user space FPU context. */
10802 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10803 {
10804 	fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10805 	++vcpu->stat.fpu_reload;
10806 	trace_kvm_fpu(0);
10807 }
10808 
10809 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10810 {
10811 	struct kvm_run *kvm_run = vcpu->run;
10812 	int r;
10813 
10814 	vcpu_load(vcpu);
10815 	kvm_sigset_activate(vcpu);
10816 	kvm_run->flags = 0;
10817 	kvm_load_guest_fpu(vcpu);
10818 
10819 	kvm_vcpu_srcu_read_lock(vcpu);
10820 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10821 		if (kvm_run->immediate_exit) {
10822 			r = -EINTR;
10823 			goto out;
10824 		}
10825 		/*
10826 		 * It should be impossible for the hypervisor timer to be in
10827 		 * use before KVM has ever run the vCPU.
10828 		 */
10829 		WARN_ON_ONCE(kvm_lapic_hv_timer_in_use(vcpu));
10830 
10831 		kvm_vcpu_srcu_read_unlock(vcpu);
10832 		kvm_vcpu_block(vcpu);
10833 		kvm_vcpu_srcu_read_lock(vcpu);
10834 
10835 		if (kvm_apic_accept_events(vcpu) < 0) {
10836 			r = 0;
10837 			goto out;
10838 		}
10839 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10840 		r = -EAGAIN;
10841 		if (signal_pending(current)) {
10842 			r = -EINTR;
10843 			kvm_run->exit_reason = KVM_EXIT_INTR;
10844 			++vcpu->stat.signal_exits;
10845 		}
10846 		goto out;
10847 	}
10848 
10849 	if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10850 	    (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10851 		r = -EINVAL;
10852 		goto out;
10853 	}
10854 
10855 	if (kvm_run->kvm_dirty_regs) {
10856 		r = sync_regs(vcpu);
10857 		if (r != 0)
10858 			goto out;
10859 	}
10860 
10861 	/* re-sync apic's tpr */
10862 	if (!lapic_in_kernel(vcpu)) {
10863 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10864 			r = -EINVAL;
10865 			goto out;
10866 		}
10867 	}
10868 
10869 	if (unlikely(vcpu->arch.complete_userspace_io)) {
10870 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10871 		vcpu->arch.complete_userspace_io = NULL;
10872 		r = cui(vcpu);
10873 		if (r <= 0)
10874 			goto out;
10875 	} else {
10876 		WARN_ON_ONCE(vcpu->arch.pio.count);
10877 		WARN_ON_ONCE(vcpu->mmio_needed);
10878 	}
10879 
10880 	if (kvm_run->immediate_exit) {
10881 		r = -EINTR;
10882 		goto out;
10883 	}
10884 
10885 	r = static_call(kvm_x86_vcpu_pre_run)(vcpu);
10886 	if (r <= 0)
10887 		goto out;
10888 
10889 	r = vcpu_run(vcpu);
10890 
10891 out:
10892 	kvm_put_guest_fpu(vcpu);
10893 	if (kvm_run->kvm_valid_regs)
10894 		store_regs(vcpu);
10895 	post_kvm_run_save(vcpu);
10896 	kvm_vcpu_srcu_read_unlock(vcpu);
10897 
10898 	kvm_sigset_deactivate(vcpu);
10899 	vcpu_put(vcpu);
10900 	return r;
10901 }
10902 
10903 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10904 {
10905 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10906 		/*
10907 		 * We are here if userspace calls get_regs() in the middle of
10908 		 * instruction emulation. Registers state needs to be copied
10909 		 * back from emulation context to vcpu. Userspace shouldn't do
10910 		 * that usually, but some bad designed PV devices (vmware
10911 		 * backdoor interface) need this to work
10912 		 */
10913 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10914 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10915 	}
10916 	regs->rax = kvm_rax_read(vcpu);
10917 	regs->rbx = kvm_rbx_read(vcpu);
10918 	regs->rcx = kvm_rcx_read(vcpu);
10919 	regs->rdx = kvm_rdx_read(vcpu);
10920 	regs->rsi = kvm_rsi_read(vcpu);
10921 	regs->rdi = kvm_rdi_read(vcpu);
10922 	regs->rsp = kvm_rsp_read(vcpu);
10923 	regs->rbp = kvm_rbp_read(vcpu);
10924 #ifdef CONFIG_X86_64
10925 	regs->r8 = kvm_r8_read(vcpu);
10926 	regs->r9 = kvm_r9_read(vcpu);
10927 	regs->r10 = kvm_r10_read(vcpu);
10928 	regs->r11 = kvm_r11_read(vcpu);
10929 	regs->r12 = kvm_r12_read(vcpu);
10930 	regs->r13 = kvm_r13_read(vcpu);
10931 	regs->r14 = kvm_r14_read(vcpu);
10932 	regs->r15 = kvm_r15_read(vcpu);
10933 #endif
10934 
10935 	regs->rip = kvm_rip_read(vcpu);
10936 	regs->rflags = kvm_get_rflags(vcpu);
10937 }
10938 
10939 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10940 {
10941 	vcpu_load(vcpu);
10942 	__get_regs(vcpu, regs);
10943 	vcpu_put(vcpu);
10944 	return 0;
10945 }
10946 
10947 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10948 {
10949 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10950 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10951 
10952 	kvm_rax_write(vcpu, regs->rax);
10953 	kvm_rbx_write(vcpu, regs->rbx);
10954 	kvm_rcx_write(vcpu, regs->rcx);
10955 	kvm_rdx_write(vcpu, regs->rdx);
10956 	kvm_rsi_write(vcpu, regs->rsi);
10957 	kvm_rdi_write(vcpu, regs->rdi);
10958 	kvm_rsp_write(vcpu, regs->rsp);
10959 	kvm_rbp_write(vcpu, regs->rbp);
10960 #ifdef CONFIG_X86_64
10961 	kvm_r8_write(vcpu, regs->r8);
10962 	kvm_r9_write(vcpu, regs->r9);
10963 	kvm_r10_write(vcpu, regs->r10);
10964 	kvm_r11_write(vcpu, regs->r11);
10965 	kvm_r12_write(vcpu, regs->r12);
10966 	kvm_r13_write(vcpu, regs->r13);
10967 	kvm_r14_write(vcpu, regs->r14);
10968 	kvm_r15_write(vcpu, regs->r15);
10969 #endif
10970 
10971 	kvm_rip_write(vcpu, regs->rip);
10972 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10973 
10974 	vcpu->arch.exception.pending = false;
10975 
10976 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10977 }
10978 
10979 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10980 {
10981 	vcpu_load(vcpu);
10982 	__set_regs(vcpu, regs);
10983 	vcpu_put(vcpu);
10984 	return 0;
10985 }
10986 
10987 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10988 {
10989 	struct desc_ptr dt;
10990 
10991 	if (vcpu->arch.guest_state_protected)
10992 		goto skip_protected_regs;
10993 
10994 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10995 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10996 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10997 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10998 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10999 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11000 
11001 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11002 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11003 
11004 	static_call(kvm_x86_get_idt)(vcpu, &dt);
11005 	sregs->idt.limit = dt.size;
11006 	sregs->idt.base = dt.address;
11007 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
11008 	sregs->gdt.limit = dt.size;
11009 	sregs->gdt.base = dt.address;
11010 
11011 	sregs->cr2 = vcpu->arch.cr2;
11012 	sregs->cr3 = kvm_read_cr3(vcpu);
11013 
11014 skip_protected_regs:
11015 	sregs->cr0 = kvm_read_cr0(vcpu);
11016 	sregs->cr4 = kvm_read_cr4(vcpu);
11017 	sregs->cr8 = kvm_get_cr8(vcpu);
11018 	sregs->efer = vcpu->arch.efer;
11019 	sregs->apic_base = kvm_get_apic_base(vcpu);
11020 }
11021 
11022 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11023 {
11024 	__get_sregs_common(vcpu, sregs);
11025 
11026 	if (vcpu->arch.guest_state_protected)
11027 		return;
11028 
11029 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
11030 		set_bit(vcpu->arch.interrupt.nr,
11031 			(unsigned long *)sregs->interrupt_bitmap);
11032 }
11033 
11034 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11035 {
11036 	int i;
11037 
11038 	__get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
11039 
11040 	if (vcpu->arch.guest_state_protected)
11041 		return;
11042 
11043 	if (is_pae_paging(vcpu)) {
11044 		for (i = 0 ; i < 4 ; i++)
11045 			sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
11046 		sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
11047 	}
11048 }
11049 
11050 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
11051 				  struct kvm_sregs *sregs)
11052 {
11053 	vcpu_load(vcpu);
11054 	__get_sregs(vcpu, sregs);
11055 	vcpu_put(vcpu);
11056 	return 0;
11057 }
11058 
11059 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
11060 				    struct kvm_mp_state *mp_state)
11061 {
11062 	int r;
11063 
11064 	vcpu_load(vcpu);
11065 	if (kvm_mpx_supported())
11066 		kvm_load_guest_fpu(vcpu);
11067 
11068 	r = kvm_apic_accept_events(vcpu);
11069 	if (r < 0)
11070 		goto out;
11071 	r = 0;
11072 
11073 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
11074 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
11075 	    vcpu->arch.pv.pv_unhalted)
11076 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
11077 	else
11078 		mp_state->mp_state = vcpu->arch.mp_state;
11079 
11080 out:
11081 	if (kvm_mpx_supported())
11082 		kvm_put_guest_fpu(vcpu);
11083 	vcpu_put(vcpu);
11084 	return r;
11085 }
11086 
11087 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
11088 				    struct kvm_mp_state *mp_state)
11089 {
11090 	int ret = -EINVAL;
11091 
11092 	vcpu_load(vcpu);
11093 
11094 	if (!lapic_in_kernel(vcpu) &&
11095 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
11096 		goto out;
11097 
11098 	/*
11099 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
11100 	 * INIT state; latched init should be reported using
11101 	 * KVM_SET_VCPU_EVENTS, so reject it here.
11102 	 */
11103 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
11104 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
11105 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
11106 		goto out;
11107 
11108 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
11109 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
11110 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
11111 	} else
11112 		vcpu->arch.mp_state = mp_state->mp_state;
11113 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11114 
11115 	ret = 0;
11116 out:
11117 	vcpu_put(vcpu);
11118 	return ret;
11119 }
11120 
11121 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
11122 		    int reason, bool has_error_code, u32 error_code)
11123 {
11124 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
11125 	int ret;
11126 
11127 	init_emulate_ctxt(vcpu);
11128 
11129 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
11130 				   has_error_code, error_code);
11131 	if (ret) {
11132 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11133 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11134 		vcpu->run->internal.ndata = 0;
11135 		return 0;
11136 	}
11137 
11138 	kvm_rip_write(vcpu, ctxt->eip);
11139 	kvm_set_rflags(vcpu, ctxt->eflags);
11140 	return 1;
11141 }
11142 EXPORT_SYMBOL_GPL(kvm_task_switch);
11143 
11144 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11145 {
11146 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
11147 		/*
11148 		 * When EFER.LME and CR0.PG are set, the processor is in
11149 		 * 64-bit mode (though maybe in a 32-bit code segment).
11150 		 * CR4.PAE and EFER.LMA must be set.
11151 		 */
11152 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
11153 			return false;
11154 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
11155 			return false;
11156 	} else {
11157 		/*
11158 		 * Not in 64-bit mode: EFER.LMA is clear and the code
11159 		 * segment cannot be 64-bit.
11160 		 */
11161 		if (sregs->efer & EFER_LMA || sregs->cs.l)
11162 			return false;
11163 	}
11164 
11165 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
11166 }
11167 
11168 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
11169 		int *mmu_reset_needed, bool update_pdptrs)
11170 {
11171 	struct msr_data apic_base_msr;
11172 	int idx;
11173 	struct desc_ptr dt;
11174 
11175 	if (!kvm_is_valid_sregs(vcpu, sregs))
11176 		return -EINVAL;
11177 
11178 	apic_base_msr.data = sregs->apic_base;
11179 	apic_base_msr.host_initiated = true;
11180 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
11181 		return -EINVAL;
11182 
11183 	if (vcpu->arch.guest_state_protected)
11184 		return 0;
11185 
11186 	dt.size = sregs->idt.limit;
11187 	dt.address = sregs->idt.base;
11188 	static_call(kvm_x86_set_idt)(vcpu, &dt);
11189 	dt.size = sregs->gdt.limit;
11190 	dt.address = sregs->gdt.base;
11191 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
11192 
11193 	vcpu->arch.cr2 = sregs->cr2;
11194 	*mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
11195 	vcpu->arch.cr3 = sregs->cr3;
11196 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11197 	static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
11198 
11199 	kvm_set_cr8(vcpu, sregs->cr8);
11200 
11201 	*mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
11202 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
11203 
11204 	*mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
11205 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
11206 	vcpu->arch.cr0 = sregs->cr0;
11207 
11208 	*mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
11209 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
11210 
11211 	if (update_pdptrs) {
11212 		idx = srcu_read_lock(&vcpu->kvm->srcu);
11213 		if (is_pae_paging(vcpu)) {
11214 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
11215 			*mmu_reset_needed = 1;
11216 		}
11217 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
11218 	}
11219 
11220 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11221 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11222 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11223 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11224 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11225 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11226 
11227 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11228 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11229 
11230 	update_cr8_intercept(vcpu);
11231 
11232 	/* Older userspace won't unhalt the vcpu on reset. */
11233 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
11234 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
11235 	    !is_protmode(vcpu))
11236 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11237 
11238 	return 0;
11239 }
11240 
11241 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11242 {
11243 	int pending_vec, max_bits;
11244 	int mmu_reset_needed = 0;
11245 	int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
11246 
11247 	if (ret)
11248 		return ret;
11249 
11250 	if (mmu_reset_needed)
11251 		kvm_mmu_reset_context(vcpu);
11252 
11253 	max_bits = KVM_NR_INTERRUPTS;
11254 	pending_vec = find_first_bit(
11255 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
11256 
11257 	if (pending_vec < max_bits) {
11258 		kvm_queue_interrupt(vcpu, pending_vec, false);
11259 		pr_debug("Set back pending irq %d\n", pending_vec);
11260 		kvm_make_request(KVM_REQ_EVENT, vcpu);
11261 	}
11262 	return 0;
11263 }
11264 
11265 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11266 {
11267 	int mmu_reset_needed = 0;
11268 	bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
11269 	bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
11270 		!(sregs2->efer & EFER_LMA);
11271 	int i, ret;
11272 
11273 	if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
11274 		return -EINVAL;
11275 
11276 	if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
11277 		return -EINVAL;
11278 
11279 	ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
11280 				 &mmu_reset_needed, !valid_pdptrs);
11281 	if (ret)
11282 		return ret;
11283 
11284 	if (valid_pdptrs) {
11285 		for (i = 0; i < 4 ; i++)
11286 			kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
11287 
11288 		kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
11289 		mmu_reset_needed = 1;
11290 		vcpu->arch.pdptrs_from_userspace = true;
11291 	}
11292 	if (mmu_reset_needed)
11293 		kvm_mmu_reset_context(vcpu);
11294 	return 0;
11295 }
11296 
11297 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
11298 				  struct kvm_sregs *sregs)
11299 {
11300 	int ret;
11301 
11302 	vcpu_load(vcpu);
11303 	ret = __set_sregs(vcpu, sregs);
11304 	vcpu_put(vcpu);
11305 	return ret;
11306 }
11307 
11308 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
11309 {
11310 	bool set = false;
11311 	struct kvm_vcpu *vcpu;
11312 	unsigned long i;
11313 
11314 	if (!enable_apicv)
11315 		return;
11316 
11317 	down_write(&kvm->arch.apicv_update_lock);
11318 
11319 	kvm_for_each_vcpu(i, vcpu, kvm) {
11320 		if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
11321 			set = true;
11322 			break;
11323 		}
11324 	}
11325 	__kvm_set_or_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_BLOCKIRQ, set);
11326 	up_write(&kvm->arch.apicv_update_lock);
11327 }
11328 
11329 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
11330 					struct kvm_guest_debug *dbg)
11331 {
11332 	unsigned long rflags;
11333 	int i, r;
11334 
11335 	if (vcpu->arch.guest_state_protected)
11336 		return -EINVAL;
11337 
11338 	vcpu_load(vcpu);
11339 
11340 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
11341 		r = -EBUSY;
11342 		if (vcpu->arch.exception.pending)
11343 			goto out;
11344 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
11345 			kvm_queue_exception(vcpu, DB_VECTOR);
11346 		else
11347 			kvm_queue_exception(vcpu, BP_VECTOR);
11348 	}
11349 
11350 	/*
11351 	 * Read rflags as long as potentially injected trace flags are still
11352 	 * filtered out.
11353 	 */
11354 	rflags = kvm_get_rflags(vcpu);
11355 
11356 	vcpu->guest_debug = dbg->control;
11357 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
11358 		vcpu->guest_debug = 0;
11359 
11360 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
11361 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
11362 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
11363 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
11364 	} else {
11365 		for (i = 0; i < KVM_NR_DB_REGS; i++)
11366 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
11367 	}
11368 	kvm_update_dr7(vcpu);
11369 
11370 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11371 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
11372 
11373 	/*
11374 	 * Trigger an rflags update that will inject or remove the trace
11375 	 * flags.
11376 	 */
11377 	kvm_set_rflags(vcpu, rflags);
11378 
11379 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11380 
11381 	kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
11382 
11383 	r = 0;
11384 
11385 out:
11386 	vcpu_put(vcpu);
11387 	return r;
11388 }
11389 
11390 /*
11391  * Translate a guest virtual address to a guest physical address.
11392  */
11393 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
11394 				    struct kvm_translation *tr)
11395 {
11396 	unsigned long vaddr = tr->linear_address;
11397 	gpa_t gpa;
11398 	int idx;
11399 
11400 	vcpu_load(vcpu);
11401 
11402 	idx = srcu_read_lock(&vcpu->kvm->srcu);
11403 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
11404 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
11405 	tr->physical_address = gpa;
11406 	tr->valid = gpa != INVALID_GPA;
11407 	tr->writeable = 1;
11408 	tr->usermode = 0;
11409 
11410 	vcpu_put(vcpu);
11411 	return 0;
11412 }
11413 
11414 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11415 {
11416 	struct fxregs_state *fxsave;
11417 
11418 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11419 		return 0;
11420 
11421 	vcpu_load(vcpu);
11422 
11423 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11424 	memcpy(fpu->fpr, fxsave->st_space, 128);
11425 	fpu->fcw = fxsave->cwd;
11426 	fpu->fsw = fxsave->swd;
11427 	fpu->ftwx = fxsave->twd;
11428 	fpu->last_opcode = fxsave->fop;
11429 	fpu->last_ip = fxsave->rip;
11430 	fpu->last_dp = fxsave->rdp;
11431 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
11432 
11433 	vcpu_put(vcpu);
11434 	return 0;
11435 }
11436 
11437 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11438 {
11439 	struct fxregs_state *fxsave;
11440 
11441 	if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11442 		return 0;
11443 
11444 	vcpu_load(vcpu);
11445 
11446 	fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11447 
11448 	memcpy(fxsave->st_space, fpu->fpr, 128);
11449 	fxsave->cwd = fpu->fcw;
11450 	fxsave->swd = fpu->fsw;
11451 	fxsave->twd = fpu->ftwx;
11452 	fxsave->fop = fpu->last_opcode;
11453 	fxsave->rip = fpu->last_ip;
11454 	fxsave->rdp = fpu->last_dp;
11455 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
11456 
11457 	vcpu_put(vcpu);
11458 	return 0;
11459 }
11460 
11461 static void store_regs(struct kvm_vcpu *vcpu)
11462 {
11463 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
11464 
11465 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
11466 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
11467 
11468 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
11469 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
11470 
11471 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
11472 		kvm_vcpu_ioctl_x86_get_vcpu_events(
11473 				vcpu, &vcpu->run->s.regs.events);
11474 }
11475 
11476 static int sync_regs(struct kvm_vcpu *vcpu)
11477 {
11478 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
11479 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
11480 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
11481 	}
11482 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
11483 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
11484 			return -EINVAL;
11485 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
11486 	}
11487 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
11488 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
11489 				vcpu, &vcpu->run->s.regs.events))
11490 			return -EINVAL;
11491 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
11492 	}
11493 
11494 	return 0;
11495 }
11496 
11497 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
11498 {
11499 	if (kvm_check_tsc_unstable() && kvm->created_vcpus)
11500 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
11501 			     "guest TSC will not be reliable\n");
11502 
11503 	if (!kvm->arch.max_vcpu_ids)
11504 		kvm->arch.max_vcpu_ids = KVM_MAX_VCPU_IDS;
11505 
11506 	if (id >= kvm->arch.max_vcpu_ids)
11507 		return -EINVAL;
11508 
11509 	return static_call(kvm_x86_vcpu_precreate)(kvm);
11510 }
11511 
11512 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
11513 {
11514 	struct page *page;
11515 	int r;
11516 
11517 	vcpu->arch.last_vmentry_cpu = -1;
11518 	vcpu->arch.regs_avail = ~0;
11519 	vcpu->arch.regs_dirty = ~0;
11520 
11521 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
11522 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11523 	else
11524 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11525 
11526 	r = kvm_mmu_create(vcpu);
11527 	if (r < 0)
11528 		return r;
11529 
11530 	if (irqchip_in_kernel(vcpu->kvm)) {
11531 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11532 		if (r < 0)
11533 			goto fail_mmu_destroy;
11534 
11535 		/*
11536 		 * Defer evaluating inhibits until the vCPU is first run, as
11537 		 * this vCPU will not get notified of any changes until this
11538 		 * vCPU is visible to other vCPUs (marked online and added to
11539 		 * the set of vCPUs).  Opportunistically mark APICv active as
11540 		 * VMX in particularly is highly unlikely to have inhibits.
11541 		 * Ignore the current per-VM APICv state so that vCPU creation
11542 		 * is guaranteed to run with a deterministic value, the request
11543 		 * will ensure the vCPU gets the correct state before VM-Entry.
11544 		 */
11545 		if (enable_apicv) {
11546 			vcpu->arch.apic->apicv_active = true;
11547 			kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
11548 		}
11549 	} else
11550 		static_branch_inc(&kvm_has_noapic_vcpu);
11551 
11552 	r = -ENOMEM;
11553 
11554 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11555 	if (!page)
11556 		goto fail_free_lapic;
11557 	vcpu->arch.pio_data = page_address(page);
11558 
11559 	vcpu->arch.mce_banks = kcalloc(KVM_MAX_MCE_BANKS * 4, sizeof(u64),
11560 				       GFP_KERNEL_ACCOUNT);
11561 	vcpu->arch.mci_ctl2_banks = kcalloc(KVM_MAX_MCE_BANKS, sizeof(u64),
11562 					    GFP_KERNEL_ACCOUNT);
11563 	if (!vcpu->arch.mce_banks || !vcpu->arch.mci_ctl2_banks)
11564 		goto fail_free_pio_data;
11565 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11566 
11567 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11568 				GFP_KERNEL_ACCOUNT))
11569 		goto fail_free_mce_banks;
11570 
11571 	if (!alloc_emulate_ctxt(vcpu))
11572 		goto free_wbinvd_dirty_mask;
11573 
11574 	if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
11575 		pr_err("kvm: failed to allocate vcpu's fpu\n");
11576 		goto free_emulate_ctxt;
11577 	}
11578 
11579 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11580 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11581 
11582 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11583 
11584 	kvm_async_pf_hash_reset(vcpu);
11585 	kvm_pmu_init(vcpu);
11586 
11587 	vcpu->arch.pending_external_vector = -1;
11588 	vcpu->arch.preempted_in_kernel = false;
11589 
11590 #if IS_ENABLED(CONFIG_HYPERV)
11591 	vcpu->arch.hv_root_tdp = INVALID_PAGE;
11592 #endif
11593 
11594 	r = static_call(kvm_x86_vcpu_create)(vcpu);
11595 	if (r)
11596 		goto free_guest_fpu;
11597 
11598 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11599 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11600 	kvm_xen_init_vcpu(vcpu);
11601 	kvm_vcpu_mtrr_init(vcpu);
11602 	vcpu_load(vcpu);
11603 	kvm_set_tsc_khz(vcpu, vcpu->kvm->arch.default_tsc_khz);
11604 	kvm_vcpu_reset(vcpu, false);
11605 	kvm_init_mmu(vcpu);
11606 	vcpu_put(vcpu);
11607 	return 0;
11608 
11609 free_guest_fpu:
11610 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11611 free_emulate_ctxt:
11612 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11613 free_wbinvd_dirty_mask:
11614 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11615 fail_free_mce_banks:
11616 	kfree(vcpu->arch.mce_banks);
11617 	kfree(vcpu->arch.mci_ctl2_banks);
11618 fail_free_pio_data:
11619 	free_page((unsigned long)vcpu->arch.pio_data);
11620 fail_free_lapic:
11621 	kvm_free_lapic(vcpu);
11622 fail_mmu_destroy:
11623 	kvm_mmu_destroy(vcpu);
11624 	return r;
11625 }
11626 
11627 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11628 {
11629 	struct kvm *kvm = vcpu->kvm;
11630 
11631 	if (mutex_lock_killable(&vcpu->mutex))
11632 		return;
11633 	vcpu_load(vcpu);
11634 	kvm_synchronize_tsc(vcpu, 0);
11635 	vcpu_put(vcpu);
11636 
11637 	/* poll control enabled by default */
11638 	vcpu->arch.msr_kvm_poll_control = 1;
11639 
11640 	mutex_unlock(&vcpu->mutex);
11641 
11642 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11643 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11644 						KVMCLOCK_SYNC_PERIOD);
11645 }
11646 
11647 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11648 {
11649 	int idx;
11650 
11651 	kvmclock_reset(vcpu);
11652 
11653 	static_call(kvm_x86_vcpu_free)(vcpu);
11654 
11655 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11656 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11657 	fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11658 
11659 	kvm_xen_destroy_vcpu(vcpu);
11660 	kvm_hv_vcpu_uninit(vcpu);
11661 	kvm_pmu_destroy(vcpu);
11662 	kfree(vcpu->arch.mce_banks);
11663 	kfree(vcpu->arch.mci_ctl2_banks);
11664 	kvm_free_lapic(vcpu);
11665 	idx = srcu_read_lock(&vcpu->kvm->srcu);
11666 	kvm_mmu_destroy(vcpu);
11667 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
11668 	free_page((unsigned long)vcpu->arch.pio_data);
11669 	kvfree(vcpu->arch.cpuid_entries);
11670 	if (!lapic_in_kernel(vcpu))
11671 		static_branch_dec(&kvm_has_noapic_vcpu);
11672 }
11673 
11674 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11675 {
11676 	struct kvm_cpuid_entry2 *cpuid_0x1;
11677 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
11678 	unsigned long new_cr0;
11679 
11680 	/*
11681 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11682 	 * to handle side effects.  RESET emulation hits those flows and relies
11683 	 * on emulated/virtualized registers, including those that are loaded
11684 	 * into hardware, to be zeroed at vCPU creation.  Use CRs as a sentinel
11685 	 * to detect improper or missing initialization.
11686 	 */
11687 	WARN_ON_ONCE(!init_event &&
11688 		     (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11689 
11690 	kvm_lapic_reset(vcpu, init_event);
11691 
11692 	vcpu->arch.hflags = 0;
11693 
11694 	vcpu->arch.smi_pending = 0;
11695 	vcpu->arch.smi_count = 0;
11696 	atomic_set(&vcpu->arch.nmi_queued, 0);
11697 	vcpu->arch.nmi_pending = 0;
11698 	vcpu->arch.nmi_injected = false;
11699 	kvm_clear_interrupt_queue(vcpu);
11700 	kvm_clear_exception_queue(vcpu);
11701 
11702 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11703 	kvm_update_dr0123(vcpu);
11704 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11705 	vcpu->arch.dr7 = DR7_FIXED_1;
11706 	kvm_update_dr7(vcpu);
11707 
11708 	vcpu->arch.cr2 = 0;
11709 
11710 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11711 	vcpu->arch.apf.msr_en_val = 0;
11712 	vcpu->arch.apf.msr_int_val = 0;
11713 	vcpu->arch.st.msr_val = 0;
11714 
11715 	kvmclock_reset(vcpu);
11716 
11717 	kvm_clear_async_pf_completion_queue(vcpu);
11718 	kvm_async_pf_hash_reset(vcpu);
11719 	vcpu->arch.apf.halted = false;
11720 
11721 	if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11722 		struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11723 
11724 		/*
11725 		 * To avoid have the INIT path from kvm_apic_has_events() that be
11726 		 * called with loaded FPU and does not let userspace fix the state.
11727 		 */
11728 		if (init_event)
11729 			kvm_put_guest_fpu(vcpu);
11730 
11731 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11732 		fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11733 
11734 		if (init_event)
11735 			kvm_load_guest_fpu(vcpu);
11736 	}
11737 
11738 	if (!init_event) {
11739 		kvm_pmu_reset(vcpu);
11740 		vcpu->arch.smbase = 0x30000;
11741 
11742 		vcpu->arch.msr_misc_features_enables = 0;
11743 		vcpu->arch.ia32_misc_enable_msr = MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL |
11744 						  MSR_IA32_MISC_ENABLE_BTS_UNAVAIL;
11745 
11746 		__kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
11747 		__kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
11748 	}
11749 
11750 	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11751 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11752 	kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11753 
11754 	/*
11755 	 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11756 	 * if no CPUID match is found.  Note, it's impossible to get a match at
11757 	 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11758 	 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11759 	 * on RESET.  But, go through the motions in case that's ever remedied.
11760 	 */
11761 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1);
11762 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11763 
11764 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11765 
11766 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11767 	kvm_rip_write(vcpu, 0xfff0);
11768 
11769 	vcpu->arch.cr3 = 0;
11770 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11771 
11772 	/*
11773 	 * CR0.CD/NW are set on RESET, preserved on INIT.  Note, some versions
11774 	 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11775 	 * (or qualify) that with a footnote stating that CD/NW are preserved.
11776 	 */
11777 	new_cr0 = X86_CR0_ET;
11778 	if (init_event)
11779 		new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11780 	else
11781 		new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11782 
11783 	static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11784 	static_call(kvm_x86_set_cr4)(vcpu, 0);
11785 	static_call(kvm_x86_set_efer)(vcpu, 0);
11786 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
11787 
11788 	/*
11789 	 * On the standard CR0/CR4/EFER modification paths, there are several
11790 	 * complex conditions determining whether the MMU has to be reset and/or
11791 	 * which PCIDs have to be flushed.  However, CR0.WP and the paging-related
11792 	 * bits in CR4 and EFER are irrelevant if CR0.PG was '0'; and a reset+flush
11793 	 * is needed anyway if CR0.PG was '1' (which can only happen for INIT, as
11794 	 * CR0 will be '0' prior to RESET).  So we only need to check CR0.PG here.
11795 	 */
11796 	if (old_cr0 & X86_CR0_PG) {
11797 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11798 		kvm_mmu_reset_context(vcpu);
11799 	}
11800 
11801 	/*
11802 	 * Intel's SDM states that all TLB entries are flushed on INIT.  AMD's
11803 	 * APM states the TLBs are untouched by INIT, but it also states that
11804 	 * the TLBs are flushed on "External initialization of the processor."
11805 	 * Flush the guest TLB regardless of vendor, there is no meaningful
11806 	 * benefit in relying on the guest to flush the TLB immediately after
11807 	 * INIT.  A spurious TLB flush is benign and likely negligible from a
11808 	 * performance perspective.
11809 	 */
11810 	if (init_event)
11811 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11812 }
11813 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11814 
11815 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11816 {
11817 	struct kvm_segment cs;
11818 
11819 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11820 	cs.selector = vector << 8;
11821 	cs.base = vector << 12;
11822 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11823 	kvm_rip_write(vcpu, 0);
11824 }
11825 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11826 
11827 int kvm_arch_hardware_enable(void)
11828 {
11829 	struct kvm *kvm;
11830 	struct kvm_vcpu *vcpu;
11831 	unsigned long i;
11832 	int ret;
11833 	u64 local_tsc;
11834 	u64 max_tsc = 0;
11835 	bool stable, backwards_tsc = false;
11836 
11837 	kvm_user_return_msr_cpu_online();
11838 	ret = static_call(kvm_x86_hardware_enable)();
11839 	if (ret != 0)
11840 		return ret;
11841 
11842 	local_tsc = rdtsc();
11843 	stable = !kvm_check_tsc_unstable();
11844 	list_for_each_entry(kvm, &vm_list, vm_list) {
11845 		kvm_for_each_vcpu(i, vcpu, kvm) {
11846 			if (!stable && vcpu->cpu == smp_processor_id())
11847 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11848 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11849 				backwards_tsc = true;
11850 				if (vcpu->arch.last_host_tsc > max_tsc)
11851 					max_tsc = vcpu->arch.last_host_tsc;
11852 			}
11853 		}
11854 	}
11855 
11856 	/*
11857 	 * Sometimes, even reliable TSCs go backwards.  This happens on
11858 	 * platforms that reset TSC during suspend or hibernate actions, but
11859 	 * maintain synchronization.  We must compensate.  Fortunately, we can
11860 	 * detect that condition here, which happens early in CPU bringup,
11861 	 * before any KVM threads can be running.  Unfortunately, we can't
11862 	 * bring the TSCs fully up to date with real time, as we aren't yet far
11863 	 * enough into CPU bringup that we know how much real time has actually
11864 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11865 	 * variables that haven't been updated yet.
11866 	 *
11867 	 * So we simply find the maximum observed TSC above, then record the
11868 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
11869 	 * the adjustment will be applied.  Note that we accumulate
11870 	 * adjustments, in case multiple suspend cycles happen before some VCPU
11871 	 * gets a chance to run again.  In the event that no KVM threads get a
11872 	 * chance to run, we will miss the entire elapsed period, as we'll have
11873 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11874 	 * loose cycle time.  This isn't too big a deal, since the loss will be
11875 	 * uniform across all VCPUs (not to mention the scenario is extremely
11876 	 * unlikely). It is possible that a second hibernate recovery happens
11877 	 * much faster than a first, causing the observed TSC here to be
11878 	 * smaller; this would require additional padding adjustment, which is
11879 	 * why we set last_host_tsc to the local tsc observed here.
11880 	 *
11881 	 * N.B. - this code below runs only on platforms with reliable TSC,
11882 	 * as that is the only way backwards_tsc is set above.  Also note
11883 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11884 	 * have the same delta_cyc adjustment applied if backwards_tsc
11885 	 * is detected.  Note further, this adjustment is only done once,
11886 	 * as we reset last_host_tsc on all VCPUs to stop this from being
11887 	 * called multiple times (one for each physical CPU bringup).
11888 	 *
11889 	 * Platforms with unreliable TSCs don't have to deal with this, they
11890 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
11891 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
11892 	 * guarantee that they stay in perfect synchronization.
11893 	 */
11894 	if (backwards_tsc) {
11895 		u64 delta_cyc = max_tsc - local_tsc;
11896 		list_for_each_entry(kvm, &vm_list, vm_list) {
11897 			kvm->arch.backwards_tsc_observed = true;
11898 			kvm_for_each_vcpu(i, vcpu, kvm) {
11899 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
11900 				vcpu->arch.last_host_tsc = local_tsc;
11901 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11902 			}
11903 
11904 			/*
11905 			 * We have to disable TSC offset matching.. if you were
11906 			 * booting a VM while issuing an S4 host suspend....
11907 			 * you may have some problem.  Solving this issue is
11908 			 * left as an exercise to the reader.
11909 			 */
11910 			kvm->arch.last_tsc_nsec = 0;
11911 			kvm->arch.last_tsc_write = 0;
11912 		}
11913 
11914 	}
11915 	return 0;
11916 }
11917 
11918 void kvm_arch_hardware_disable(void)
11919 {
11920 	static_call(kvm_x86_hardware_disable)();
11921 	drop_user_return_notifiers();
11922 }
11923 
11924 static inline void kvm_ops_update(struct kvm_x86_init_ops *ops)
11925 {
11926 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11927 
11928 #define __KVM_X86_OP(func) \
11929 	static_call_update(kvm_x86_##func, kvm_x86_ops.func);
11930 #define KVM_X86_OP(func) \
11931 	WARN_ON(!kvm_x86_ops.func); __KVM_X86_OP(func)
11932 #define KVM_X86_OP_OPTIONAL __KVM_X86_OP
11933 #define KVM_X86_OP_OPTIONAL_RET0(func) \
11934 	static_call_update(kvm_x86_##func, (void *)kvm_x86_ops.func ? : \
11935 					   (void *)__static_call_return0);
11936 #include <asm/kvm-x86-ops.h>
11937 #undef __KVM_X86_OP
11938 
11939 	kvm_pmu_ops_update(ops->pmu_ops);
11940 }
11941 
11942 int kvm_arch_hardware_setup(void *opaque)
11943 {
11944 	struct kvm_x86_init_ops *ops = opaque;
11945 	int r;
11946 
11947 	rdmsrl_safe(MSR_EFER, &host_efer);
11948 
11949 	if (boot_cpu_has(X86_FEATURE_XSAVES))
11950 		rdmsrl(MSR_IA32_XSS, host_xss);
11951 
11952 	kvm_init_pmu_capability();
11953 
11954 	r = ops->hardware_setup();
11955 	if (r != 0)
11956 		return r;
11957 
11958 	kvm_ops_update(ops);
11959 
11960 	kvm_register_perf_callbacks(ops->handle_intel_pt_intr);
11961 
11962 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11963 		kvm_caps.supported_xss = 0;
11964 
11965 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11966 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11967 #undef __kvm_cpu_cap_has
11968 
11969 	if (kvm_caps.has_tsc_control) {
11970 		/*
11971 		 * Make sure the user can only configure tsc_khz values that
11972 		 * fit into a signed integer.
11973 		 * A min value is not calculated because it will always
11974 		 * be 1 on all machines.
11975 		 */
11976 		u64 max = min(0x7fffffffULL,
11977 			      __scale_tsc(kvm_caps.max_tsc_scaling_ratio, tsc_khz));
11978 		kvm_caps.max_guest_tsc_khz = max;
11979 	}
11980 	kvm_caps.default_tsc_scaling_ratio = 1ULL << kvm_caps.tsc_scaling_ratio_frac_bits;
11981 	kvm_init_msr_list();
11982 	return 0;
11983 }
11984 
11985 void kvm_arch_hardware_unsetup(void)
11986 {
11987 	kvm_unregister_perf_callbacks();
11988 
11989 	static_call(kvm_x86_hardware_unsetup)();
11990 }
11991 
11992 int kvm_arch_check_processor_compat(void *opaque)
11993 {
11994 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11995 	struct kvm_x86_init_ops *ops = opaque;
11996 
11997 	WARN_ON(!irqs_disabled());
11998 
11999 	if (__cr4_reserved_bits(cpu_has, c) !=
12000 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
12001 		return -EIO;
12002 
12003 	return ops->check_processor_compatibility();
12004 }
12005 
12006 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
12007 {
12008 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
12009 }
12010 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
12011 
12012 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
12013 {
12014 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
12015 }
12016 
12017 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
12018 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
12019 
12020 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
12021 {
12022 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
12023 
12024 	vcpu->arch.l1tf_flush_l1d = true;
12025 	if (pmu->version && unlikely(pmu->event_count)) {
12026 		pmu->need_cleanup = true;
12027 		kvm_make_request(KVM_REQ_PMU, vcpu);
12028 	}
12029 	static_call(kvm_x86_sched_in)(vcpu, cpu);
12030 }
12031 
12032 void kvm_arch_free_vm(struct kvm *kvm)
12033 {
12034 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
12035 	__kvm_arch_free_vm(kvm);
12036 }
12037 
12038 
12039 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
12040 {
12041 	int ret;
12042 	unsigned long flags;
12043 
12044 	if (type)
12045 		return -EINVAL;
12046 
12047 	ret = kvm_page_track_init(kvm);
12048 	if (ret)
12049 		goto out;
12050 
12051 	ret = kvm_mmu_init_vm(kvm);
12052 	if (ret)
12053 		goto out_page_track;
12054 
12055 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
12056 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
12057 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
12058 
12059 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
12060 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
12061 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
12062 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
12063 		&kvm->arch.irq_sources_bitmap);
12064 
12065 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
12066 	mutex_init(&kvm->arch.apic_map_lock);
12067 	seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
12068 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
12069 
12070 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
12071 	pvclock_update_vm_gtod_copy(kvm);
12072 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
12073 
12074 	kvm->arch.default_tsc_khz = max_tsc_khz ? : tsc_khz;
12075 	kvm->arch.guest_can_read_msr_platform_info = true;
12076 	kvm->arch.enable_pmu = enable_pmu;
12077 
12078 #if IS_ENABLED(CONFIG_HYPERV)
12079 	spin_lock_init(&kvm->arch.hv_root_tdp_lock);
12080 	kvm->arch.hv_root_tdp = INVALID_PAGE;
12081 #endif
12082 
12083 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
12084 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
12085 
12086 	kvm_apicv_init(kvm);
12087 	kvm_hv_init_vm(kvm);
12088 	kvm_xen_init_vm(kvm);
12089 
12090 	return static_call(kvm_x86_vm_init)(kvm);
12091 
12092 out_page_track:
12093 	kvm_page_track_cleanup(kvm);
12094 out:
12095 	return ret;
12096 }
12097 
12098 int kvm_arch_post_init_vm(struct kvm *kvm)
12099 {
12100 	return kvm_mmu_post_init_vm(kvm);
12101 }
12102 
12103 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
12104 {
12105 	vcpu_load(vcpu);
12106 	kvm_mmu_unload(vcpu);
12107 	vcpu_put(vcpu);
12108 }
12109 
12110 static void kvm_unload_vcpu_mmus(struct kvm *kvm)
12111 {
12112 	unsigned long i;
12113 	struct kvm_vcpu *vcpu;
12114 
12115 	kvm_for_each_vcpu(i, vcpu, kvm) {
12116 		kvm_clear_async_pf_completion_queue(vcpu);
12117 		kvm_unload_vcpu_mmu(vcpu);
12118 	}
12119 }
12120 
12121 void kvm_arch_sync_events(struct kvm *kvm)
12122 {
12123 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
12124 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
12125 	kvm_free_pit(kvm);
12126 }
12127 
12128 /**
12129  * __x86_set_memory_region: Setup KVM internal memory slot
12130  *
12131  * @kvm: the kvm pointer to the VM.
12132  * @id: the slot ID to setup.
12133  * @gpa: the GPA to install the slot (unused when @size == 0).
12134  * @size: the size of the slot. Set to zero to uninstall a slot.
12135  *
12136  * This function helps to setup a KVM internal memory slot.  Specify
12137  * @size > 0 to install a new slot, while @size == 0 to uninstall a
12138  * slot.  The return code can be one of the following:
12139  *
12140  *   HVA:           on success (uninstall will return a bogus HVA)
12141  *   -errno:        on error
12142  *
12143  * The caller should always use IS_ERR() to check the return value
12144  * before use.  Note, the KVM internal memory slots are guaranteed to
12145  * remain valid and unchanged until the VM is destroyed, i.e., the
12146  * GPA->HVA translation will not change.  However, the HVA is a user
12147  * address, i.e. its accessibility is not guaranteed, and must be
12148  * accessed via __copy_{to,from}_user().
12149  */
12150 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
12151 				      u32 size)
12152 {
12153 	int i, r;
12154 	unsigned long hva, old_npages;
12155 	struct kvm_memslots *slots = kvm_memslots(kvm);
12156 	struct kvm_memory_slot *slot;
12157 
12158 	/* Called with kvm->slots_lock held.  */
12159 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
12160 		return ERR_PTR_USR(-EINVAL);
12161 
12162 	slot = id_to_memslot(slots, id);
12163 	if (size) {
12164 		if (slot && slot->npages)
12165 			return ERR_PTR_USR(-EEXIST);
12166 
12167 		/*
12168 		 * MAP_SHARED to prevent internal slot pages from being moved
12169 		 * by fork()/COW.
12170 		 */
12171 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
12172 			      MAP_SHARED | MAP_ANONYMOUS, 0);
12173 		if (IS_ERR((void *)hva))
12174 			return (void __user *)hva;
12175 	} else {
12176 		if (!slot || !slot->npages)
12177 			return NULL;
12178 
12179 		old_npages = slot->npages;
12180 		hva = slot->userspace_addr;
12181 	}
12182 
12183 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
12184 		struct kvm_userspace_memory_region m;
12185 
12186 		m.slot = id | (i << 16);
12187 		m.flags = 0;
12188 		m.guest_phys_addr = gpa;
12189 		m.userspace_addr = hva;
12190 		m.memory_size = size;
12191 		r = __kvm_set_memory_region(kvm, &m);
12192 		if (r < 0)
12193 			return ERR_PTR_USR(r);
12194 	}
12195 
12196 	if (!size)
12197 		vm_munmap(hva, old_npages * PAGE_SIZE);
12198 
12199 	return (void __user *)hva;
12200 }
12201 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
12202 
12203 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
12204 {
12205 	kvm_mmu_pre_destroy_vm(kvm);
12206 }
12207 
12208 void kvm_arch_destroy_vm(struct kvm *kvm)
12209 {
12210 	if (current->mm == kvm->mm) {
12211 		/*
12212 		 * Free memory regions allocated on behalf of userspace,
12213 		 * unless the memory map has changed due to process exit
12214 		 * or fd copying.
12215 		 */
12216 		mutex_lock(&kvm->slots_lock);
12217 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
12218 					0, 0);
12219 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
12220 					0, 0);
12221 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
12222 		mutex_unlock(&kvm->slots_lock);
12223 	}
12224 	kvm_unload_vcpu_mmus(kvm);
12225 	static_call_cond(kvm_x86_vm_destroy)(kvm);
12226 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
12227 	kvm_pic_destroy(kvm);
12228 	kvm_ioapic_destroy(kvm);
12229 	kvm_destroy_vcpus(kvm);
12230 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
12231 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
12232 	kvm_mmu_uninit_vm(kvm);
12233 	kvm_page_track_cleanup(kvm);
12234 	kvm_xen_destroy_vm(kvm);
12235 	kvm_hv_destroy_vm(kvm);
12236 }
12237 
12238 static void memslot_rmap_free(struct kvm_memory_slot *slot)
12239 {
12240 	int i;
12241 
12242 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12243 		kvfree(slot->arch.rmap[i]);
12244 		slot->arch.rmap[i] = NULL;
12245 	}
12246 }
12247 
12248 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
12249 {
12250 	int i;
12251 
12252 	memslot_rmap_free(slot);
12253 
12254 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12255 		kvfree(slot->arch.lpage_info[i - 1]);
12256 		slot->arch.lpage_info[i - 1] = NULL;
12257 	}
12258 
12259 	kvm_page_track_free_memslot(slot);
12260 }
12261 
12262 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
12263 {
12264 	const int sz = sizeof(*slot->arch.rmap[0]);
12265 	int i;
12266 
12267 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12268 		int level = i + 1;
12269 		int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12270 
12271 		if (slot->arch.rmap[i])
12272 			continue;
12273 
12274 		slot->arch.rmap[i] = __vcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
12275 		if (!slot->arch.rmap[i]) {
12276 			memslot_rmap_free(slot);
12277 			return -ENOMEM;
12278 		}
12279 	}
12280 
12281 	return 0;
12282 }
12283 
12284 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
12285 				      struct kvm_memory_slot *slot)
12286 {
12287 	unsigned long npages = slot->npages;
12288 	int i, r;
12289 
12290 	/*
12291 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
12292 	 * old arrays will be freed by __kvm_set_memory_region() if installing
12293 	 * the new memslot is successful.
12294 	 */
12295 	memset(&slot->arch, 0, sizeof(slot->arch));
12296 
12297 	if (kvm_memslots_have_rmaps(kvm)) {
12298 		r = memslot_rmap_alloc(slot, npages);
12299 		if (r)
12300 			return r;
12301 	}
12302 
12303 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12304 		struct kvm_lpage_info *linfo;
12305 		unsigned long ugfn;
12306 		int lpages;
12307 		int level = i + 1;
12308 
12309 		lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12310 
12311 		linfo = __vcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
12312 		if (!linfo)
12313 			goto out_free;
12314 
12315 		slot->arch.lpage_info[i - 1] = linfo;
12316 
12317 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
12318 			linfo[0].disallow_lpage = 1;
12319 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
12320 			linfo[lpages - 1].disallow_lpage = 1;
12321 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
12322 		/*
12323 		 * If the gfn and userspace address are not aligned wrt each
12324 		 * other, disable large page support for this slot.
12325 		 */
12326 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
12327 			unsigned long j;
12328 
12329 			for (j = 0; j < lpages; ++j)
12330 				linfo[j].disallow_lpage = 1;
12331 		}
12332 	}
12333 
12334 	if (kvm_page_track_create_memslot(kvm, slot, npages))
12335 		goto out_free;
12336 
12337 	return 0;
12338 
12339 out_free:
12340 	memslot_rmap_free(slot);
12341 
12342 	for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12343 		kvfree(slot->arch.lpage_info[i - 1]);
12344 		slot->arch.lpage_info[i - 1] = NULL;
12345 	}
12346 	return -ENOMEM;
12347 }
12348 
12349 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
12350 {
12351 	struct kvm_vcpu *vcpu;
12352 	unsigned long i;
12353 
12354 	/*
12355 	 * memslots->generation has been incremented.
12356 	 * mmio generation may have reached its maximum value.
12357 	 */
12358 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
12359 
12360 	/* Force re-initialization of steal_time cache */
12361 	kvm_for_each_vcpu(i, vcpu, kvm)
12362 		kvm_vcpu_kick(vcpu);
12363 }
12364 
12365 int kvm_arch_prepare_memory_region(struct kvm *kvm,
12366 				   const struct kvm_memory_slot *old,
12367 				   struct kvm_memory_slot *new,
12368 				   enum kvm_mr_change change)
12369 {
12370 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) {
12371 		if ((new->base_gfn + new->npages - 1) > kvm_mmu_max_gfn())
12372 			return -EINVAL;
12373 
12374 		return kvm_alloc_memslot_metadata(kvm, new);
12375 	}
12376 
12377 	if (change == KVM_MR_FLAGS_ONLY)
12378 		memcpy(&new->arch, &old->arch, sizeof(old->arch));
12379 	else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
12380 		return -EIO;
12381 
12382 	return 0;
12383 }
12384 
12385 
12386 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
12387 {
12388 	struct kvm_arch *ka = &kvm->arch;
12389 
12390 	if (!kvm_x86_ops.cpu_dirty_log_size)
12391 		return;
12392 
12393 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
12394 	    (!enable && --ka->cpu_dirty_logging_count == 0))
12395 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
12396 
12397 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
12398 }
12399 
12400 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
12401 				     struct kvm_memory_slot *old,
12402 				     const struct kvm_memory_slot *new,
12403 				     enum kvm_mr_change change)
12404 {
12405 	u32 old_flags = old ? old->flags : 0;
12406 	u32 new_flags = new ? new->flags : 0;
12407 	bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
12408 
12409 	/*
12410 	 * Update CPU dirty logging if dirty logging is being toggled.  This
12411 	 * applies to all operations.
12412 	 */
12413 	if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
12414 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
12415 
12416 	/*
12417 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
12418 	 * made writable) or CREATE/MOVE/DELETE of a slot.
12419 	 *
12420 	 * For a memslot with dirty logging disabled:
12421 	 * CREATE:      No dirty mappings will already exist.
12422 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
12423 	 *		kvm_arch_flush_shadow_memslot()
12424 	 *
12425 	 * For a memslot with dirty logging enabled:
12426 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
12427 	 *		and no dirty bits to clear.
12428 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
12429 	 *		kvm_arch_flush_shadow_memslot().
12430 	 */
12431 	if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
12432 		return;
12433 
12434 	/*
12435 	 * READONLY and non-flags changes were filtered out above, and the only
12436 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
12437 	 * logging isn't being toggled on or off.
12438 	 */
12439 	if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
12440 		return;
12441 
12442 	if (!log_dirty_pages) {
12443 		/*
12444 		 * Dirty logging tracks sptes in 4k granularity, meaning that
12445 		 * large sptes have to be split.  If live migration succeeds,
12446 		 * the guest in the source machine will be destroyed and large
12447 		 * sptes will be created in the destination.  However, if the
12448 		 * guest continues to run in the source machine (for example if
12449 		 * live migration fails), small sptes will remain around and
12450 		 * cause bad performance.
12451 		 *
12452 		 * Scan sptes if dirty logging has been stopped, dropping those
12453 		 * which can be collapsed into a single large-page spte.  Later
12454 		 * page faults will create the large-page sptes.
12455 		 */
12456 		kvm_mmu_zap_collapsible_sptes(kvm, new);
12457 	} else {
12458 		/*
12459 		 * Initially-all-set does not require write protecting any page,
12460 		 * because they're all assumed to be dirty.
12461 		 */
12462 		if (kvm_dirty_log_manual_protect_and_init_set(kvm))
12463 			return;
12464 
12465 		if (READ_ONCE(eager_page_split))
12466 			kvm_mmu_slot_try_split_huge_pages(kvm, new, PG_LEVEL_4K);
12467 
12468 		if (kvm_x86_ops.cpu_dirty_log_size) {
12469 			kvm_mmu_slot_leaf_clear_dirty(kvm, new);
12470 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
12471 		} else {
12472 			kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
12473 		}
12474 	}
12475 }
12476 
12477 void kvm_arch_commit_memory_region(struct kvm *kvm,
12478 				struct kvm_memory_slot *old,
12479 				const struct kvm_memory_slot *new,
12480 				enum kvm_mr_change change)
12481 {
12482 	if (!kvm->arch.n_requested_mmu_pages &&
12483 	    (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
12484 		unsigned long nr_mmu_pages;
12485 
12486 		nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
12487 		nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
12488 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
12489 	}
12490 
12491 	kvm_mmu_slot_apply_flags(kvm, old, new, change);
12492 
12493 	/* Free the arrays associated with the old memslot. */
12494 	if (change == KVM_MR_MOVE)
12495 		kvm_arch_free_memslot(kvm, old);
12496 }
12497 
12498 void kvm_arch_flush_shadow_all(struct kvm *kvm)
12499 {
12500 	kvm_mmu_zap_all(kvm);
12501 }
12502 
12503 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
12504 				   struct kvm_memory_slot *slot)
12505 {
12506 	kvm_page_track_flush_slot(kvm, slot);
12507 }
12508 
12509 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
12510 {
12511 	return (is_guest_mode(vcpu) &&
12512 		static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
12513 }
12514 
12515 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
12516 {
12517 	if (!list_empty_careful(&vcpu->async_pf.done))
12518 		return true;
12519 
12520 	if (kvm_apic_has_events(vcpu))
12521 		return true;
12522 
12523 	if (vcpu->arch.pv.pv_unhalted)
12524 		return true;
12525 
12526 	if (vcpu->arch.exception.pending)
12527 		return true;
12528 
12529 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12530 	    (vcpu->arch.nmi_pending &&
12531 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
12532 		return true;
12533 
12534 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
12535 	    (vcpu->arch.smi_pending &&
12536 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
12537 		return true;
12538 
12539 	if (kvm_arch_interrupt_allowed(vcpu) &&
12540 	    (kvm_cpu_has_interrupt(vcpu) ||
12541 	    kvm_guest_apic_has_interrupt(vcpu)))
12542 		return true;
12543 
12544 	if (kvm_hv_has_stimer_pending(vcpu))
12545 		return true;
12546 
12547 	if (is_guest_mode(vcpu) &&
12548 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
12549 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
12550 		return true;
12551 
12552 	if (kvm_xen_has_pending_events(vcpu))
12553 		return true;
12554 
12555 	if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu))
12556 		return true;
12557 
12558 	return false;
12559 }
12560 
12561 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
12562 {
12563 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
12564 }
12565 
12566 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
12567 {
12568 	if (kvm_vcpu_apicv_active(vcpu) &&
12569 	    static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
12570 		return true;
12571 
12572 	return false;
12573 }
12574 
12575 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12576 {
12577 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12578 		return true;
12579 
12580 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12581 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
12582 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
12583 		return true;
12584 
12585 	return kvm_arch_dy_has_pending_interrupt(vcpu);
12586 }
12587 
12588 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12589 {
12590 	if (vcpu->arch.guest_state_protected)
12591 		return true;
12592 
12593 	return vcpu->arch.preempted_in_kernel;
12594 }
12595 
12596 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
12597 {
12598 	return kvm_rip_read(vcpu);
12599 }
12600 
12601 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12602 {
12603 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12604 }
12605 
12606 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12607 {
12608 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12609 }
12610 
12611 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12612 {
12613 	/* Can't read the RIP when guest state is protected, just return 0 */
12614 	if (vcpu->arch.guest_state_protected)
12615 		return 0;
12616 
12617 	if (is_64_bit_mode(vcpu))
12618 		return kvm_rip_read(vcpu);
12619 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12620 		     kvm_rip_read(vcpu));
12621 }
12622 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12623 
12624 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12625 {
12626 	return kvm_get_linear_rip(vcpu) == linear_rip;
12627 }
12628 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12629 
12630 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12631 {
12632 	unsigned long rflags;
12633 
12634 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
12635 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12636 		rflags &= ~X86_EFLAGS_TF;
12637 	return rflags;
12638 }
12639 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12640 
12641 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12642 {
12643 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12644 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12645 		rflags |= X86_EFLAGS_TF;
12646 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
12647 }
12648 
12649 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12650 {
12651 	__kvm_set_rflags(vcpu, rflags);
12652 	kvm_make_request(KVM_REQ_EVENT, vcpu);
12653 }
12654 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12655 
12656 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12657 {
12658 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12659 
12660 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12661 }
12662 
12663 static inline u32 kvm_async_pf_next_probe(u32 key)
12664 {
12665 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12666 }
12667 
12668 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12669 {
12670 	u32 key = kvm_async_pf_hash_fn(gfn);
12671 
12672 	while (vcpu->arch.apf.gfns[key] != ~0)
12673 		key = kvm_async_pf_next_probe(key);
12674 
12675 	vcpu->arch.apf.gfns[key] = gfn;
12676 }
12677 
12678 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12679 {
12680 	int i;
12681 	u32 key = kvm_async_pf_hash_fn(gfn);
12682 
12683 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
12684 		     (vcpu->arch.apf.gfns[key] != gfn &&
12685 		      vcpu->arch.apf.gfns[key] != ~0); i++)
12686 		key = kvm_async_pf_next_probe(key);
12687 
12688 	return key;
12689 }
12690 
12691 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12692 {
12693 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12694 }
12695 
12696 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12697 {
12698 	u32 i, j, k;
12699 
12700 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12701 
12702 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12703 		return;
12704 
12705 	while (true) {
12706 		vcpu->arch.apf.gfns[i] = ~0;
12707 		do {
12708 			j = kvm_async_pf_next_probe(j);
12709 			if (vcpu->arch.apf.gfns[j] == ~0)
12710 				return;
12711 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12712 			/*
12713 			 * k lies cyclically in ]i,j]
12714 			 * |    i.k.j |
12715 			 * |....j i.k.| or  |.k..j i...|
12716 			 */
12717 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12718 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12719 		i = j;
12720 	}
12721 }
12722 
12723 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12724 {
12725 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12726 
12727 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12728 				      sizeof(reason));
12729 }
12730 
12731 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12732 {
12733 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12734 
12735 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12736 					     &token, offset, sizeof(token));
12737 }
12738 
12739 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12740 {
12741 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12742 	u32 val;
12743 
12744 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12745 					 &val, offset, sizeof(val)))
12746 		return false;
12747 
12748 	return !val;
12749 }
12750 
12751 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12752 {
12753 
12754 	if (!kvm_pv_async_pf_enabled(vcpu))
12755 		return false;
12756 
12757 	if (vcpu->arch.apf.send_user_only &&
12758 	    static_call(kvm_x86_get_cpl)(vcpu) == 0)
12759 		return false;
12760 
12761 	if (is_guest_mode(vcpu)) {
12762 		/*
12763 		 * L1 needs to opt into the special #PF vmexits that are
12764 		 * used to deliver async page faults.
12765 		 */
12766 		return vcpu->arch.apf.delivery_as_pf_vmexit;
12767 	} else {
12768 		/*
12769 		 * Play it safe in case the guest temporarily disables paging.
12770 		 * The real mode IDT in particular is unlikely to have a #PF
12771 		 * exception setup.
12772 		 */
12773 		return is_paging(vcpu);
12774 	}
12775 }
12776 
12777 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12778 {
12779 	if (unlikely(!lapic_in_kernel(vcpu) ||
12780 		     kvm_event_needs_reinjection(vcpu) ||
12781 		     vcpu->arch.exception.pending))
12782 		return false;
12783 
12784 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12785 		return false;
12786 
12787 	/*
12788 	 * If interrupts are off we cannot even use an artificial
12789 	 * halt state.
12790 	 */
12791 	return kvm_arch_interrupt_allowed(vcpu);
12792 }
12793 
12794 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12795 				     struct kvm_async_pf *work)
12796 {
12797 	struct x86_exception fault;
12798 
12799 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12800 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12801 
12802 	if (kvm_can_deliver_async_pf(vcpu) &&
12803 	    !apf_put_user_notpresent(vcpu)) {
12804 		fault.vector = PF_VECTOR;
12805 		fault.error_code_valid = true;
12806 		fault.error_code = 0;
12807 		fault.nested_page_fault = false;
12808 		fault.address = work->arch.token;
12809 		fault.async_page_fault = true;
12810 		kvm_inject_page_fault(vcpu, &fault);
12811 		return true;
12812 	} else {
12813 		/*
12814 		 * It is not possible to deliver a paravirtualized asynchronous
12815 		 * page fault, but putting the guest in an artificial halt state
12816 		 * can be beneficial nevertheless: if an interrupt arrives, we
12817 		 * can deliver it timely and perhaps the guest will schedule
12818 		 * another process.  When the instruction that triggered a page
12819 		 * fault is retried, hopefully the page will be ready in the host.
12820 		 */
12821 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12822 		return false;
12823 	}
12824 }
12825 
12826 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12827 				 struct kvm_async_pf *work)
12828 {
12829 	struct kvm_lapic_irq irq = {
12830 		.delivery_mode = APIC_DM_FIXED,
12831 		.vector = vcpu->arch.apf.vec
12832 	};
12833 
12834 	if (work->wakeup_all)
12835 		work->arch.token = ~0; /* broadcast wakeup */
12836 	else
12837 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12838 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12839 
12840 	if ((work->wakeup_all || work->notpresent_injected) &&
12841 	    kvm_pv_async_pf_enabled(vcpu) &&
12842 	    !apf_put_user_ready(vcpu, work->arch.token)) {
12843 		vcpu->arch.apf.pageready_pending = true;
12844 		kvm_apic_set_irq(vcpu, &irq, NULL);
12845 	}
12846 
12847 	vcpu->arch.apf.halted = false;
12848 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12849 }
12850 
12851 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12852 {
12853 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
12854 	if (!vcpu->arch.apf.pageready_pending)
12855 		kvm_vcpu_kick(vcpu);
12856 }
12857 
12858 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12859 {
12860 	if (!kvm_pv_async_pf_enabled(vcpu))
12861 		return true;
12862 	else
12863 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12864 }
12865 
12866 void kvm_arch_start_assignment(struct kvm *kvm)
12867 {
12868 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12869 		static_call_cond(kvm_x86_pi_start_assignment)(kvm);
12870 }
12871 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12872 
12873 void kvm_arch_end_assignment(struct kvm *kvm)
12874 {
12875 	atomic_dec(&kvm->arch.assigned_device_count);
12876 }
12877 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12878 
12879 bool noinstr kvm_arch_has_assigned_device(struct kvm *kvm)
12880 {
12881 	return arch_atomic_read(&kvm->arch.assigned_device_count);
12882 }
12883 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12884 
12885 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12886 {
12887 	atomic_inc(&kvm->arch.noncoherent_dma_count);
12888 }
12889 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12890 
12891 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12892 {
12893 	atomic_dec(&kvm->arch.noncoherent_dma_count);
12894 }
12895 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12896 
12897 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12898 {
12899 	return atomic_read(&kvm->arch.noncoherent_dma_count);
12900 }
12901 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12902 
12903 bool kvm_arch_has_irq_bypass(void)
12904 {
12905 	return true;
12906 }
12907 
12908 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12909 				      struct irq_bypass_producer *prod)
12910 {
12911 	struct kvm_kernel_irqfd *irqfd =
12912 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12913 	int ret;
12914 
12915 	irqfd->producer = prod;
12916 	kvm_arch_start_assignment(irqfd->kvm);
12917 	ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm,
12918 					 prod->irq, irqfd->gsi, 1);
12919 
12920 	if (ret)
12921 		kvm_arch_end_assignment(irqfd->kvm);
12922 
12923 	return ret;
12924 }
12925 
12926 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12927 				      struct irq_bypass_producer *prod)
12928 {
12929 	int ret;
12930 	struct kvm_kernel_irqfd *irqfd =
12931 		container_of(cons, struct kvm_kernel_irqfd, consumer);
12932 
12933 	WARN_ON(irqfd->producer != prod);
12934 	irqfd->producer = NULL;
12935 
12936 	/*
12937 	 * When producer of consumer is unregistered, we change back to
12938 	 * remapped mode, so we can re-use the current implementation
12939 	 * when the irq is masked/disabled or the consumer side (KVM
12940 	 * int this case doesn't want to receive the interrupts.
12941 	*/
12942 	ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12943 	if (ret)
12944 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12945 		       " fails: %d\n", irqfd->consumer.token, ret);
12946 
12947 	kvm_arch_end_assignment(irqfd->kvm);
12948 }
12949 
12950 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12951 				   uint32_t guest_irq, bool set)
12952 {
12953 	return static_call(kvm_x86_pi_update_irte)(kvm, host_irq, guest_irq, set);
12954 }
12955 
12956 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12957 				  struct kvm_kernel_irq_routing_entry *new)
12958 {
12959 	if (new->type != KVM_IRQ_ROUTING_MSI)
12960 		return true;
12961 
12962 	return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12963 }
12964 
12965 bool kvm_vector_hashing_enabled(void)
12966 {
12967 	return vector_hashing;
12968 }
12969 
12970 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12971 {
12972 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12973 }
12974 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12975 
12976 
12977 int kvm_spec_ctrl_test_value(u64 value)
12978 {
12979 	/*
12980 	 * test that setting IA32_SPEC_CTRL to given value
12981 	 * is allowed by the host processor
12982 	 */
12983 
12984 	u64 saved_value;
12985 	unsigned long flags;
12986 	int ret = 0;
12987 
12988 	local_irq_save(flags);
12989 
12990 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12991 		ret = 1;
12992 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12993 		ret = 1;
12994 	else
12995 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12996 
12997 	local_irq_restore(flags);
12998 
12999 	return ret;
13000 }
13001 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
13002 
13003 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
13004 {
13005 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
13006 	struct x86_exception fault;
13007 	u64 access = error_code &
13008 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
13009 
13010 	if (!(error_code & PFERR_PRESENT_MASK) ||
13011 	    mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != INVALID_GPA) {
13012 		/*
13013 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
13014 		 * tables probably do not match the TLB.  Just proceed
13015 		 * with the error code that the processor gave.
13016 		 */
13017 		fault.vector = PF_VECTOR;
13018 		fault.error_code_valid = true;
13019 		fault.error_code = error_code;
13020 		fault.nested_page_fault = false;
13021 		fault.address = gva;
13022 	}
13023 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
13024 }
13025 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
13026 
13027 /*
13028  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
13029  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
13030  * indicates whether exit to userspace is needed.
13031  */
13032 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
13033 			      struct x86_exception *e)
13034 {
13035 	if (r == X86EMUL_PROPAGATE_FAULT) {
13036 		kvm_inject_emulated_page_fault(vcpu, e);
13037 		return 1;
13038 	}
13039 
13040 	/*
13041 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
13042 	 * while handling a VMX instruction KVM could've handled the request
13043 	 * correctly by exiting to userspace and performing I/O but there
13044 	 * doesn't seem to be a real use-case behind such requests, just return
13045 	 * KVM_EXIT_INTERNAL_ERROR for now.
13046 	 */
13047 	kvm_prepare_emulation_failure_exit(vcpu);
13048 
13049 	return 0;
13050 }
13051 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
13052 
13053 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
13054 {
13055 	bool pcid_enabled;
13056 	struct x86_exception e;
13057 	struct {
13058 		u64 pcid;
13059 		u64 gla;
13060 	} operand;
13061 	int r;
13062 
13063 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
13064 	if (r != X86EMUL_CONTINUE)
13065 		return kvm_handle_memory_failure(vcpu, r, &e);
13066 
13067 	if (operand.pcid >> 12 != 0) {
13068 		kvm_inject_gp(vcpu, 0);
13069 		return 1;
13070 	}
13071 
13072 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
13073 
13074 	switch (type) {
13075 	case INVPCID_TYPE_INDIV_ADDR:
13076 		if ((!pcid_enabled && (operand.pcid != 0)) ||
13077 		    is_noncanonical_address(operand.gla, vcpu)) {
13078 			kvm_inject_gp(vcpu, 0);
13079 			return 1;
13080 		}
13081 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
13082 		return kvm_skip_emulated_instruction(vcpu);
13083 
13084 	case INVPCID_TYPE_SINGLE_CTXT:
13085 		if (!pcid_enabled && (operand.pcid != 0)) {
13086 			kvm_inject_gp(vcpu, 0);
13087 			return 1;
13088 		}
13089 
13090 		kvm_invalidate_pcid(vcpu, operand.pcid);
13091 		return kvm_skip_emulated_instruction(vcpu);
13092 
13093 	case INVPCID_TYPE_ALL_NON_GLOBAL:
13094 		/*
13095 		 * Currently, KVM doesn't mark global entries in the shadow
13096 		 * page tables, so a non-global flush just degenerates to a
13097 		 * global flush. If needed, we could optimize this later by
13098 		 * keeping track of global entries in shadow page tables.
13099 		 */
13100 
13101 		fallthrough;
13102 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
13103 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
13104 		return kvm_skip_emulated_instruction(vcpu);
13105 
13106 	default:
13107 		kvm_inject_gp(vcpu, 0);
13108 		return 1;
13109 	}
13110 }
13111 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
13112 
13113 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
13114 {
13115 	struct kvm_run *run = vcpu->run;
13116 	struct kvm_mmio_fragment *frag;
13117 	unsigned int len;
13118 
13119 	BUG_ON(!vcpu->mmio_needed);
13120 
13121 	/* Complete previous fragment */
13122 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
13123 	len = min(8u, frag->len);
13124 	if (!vcpu->mmio_is_write)
13125 		memcpy(frag->data, run->mmio.data, len);
13126 
13127 	if (frag->len <= 8) {
13128 		/* Switch to the next fragment. */
13129 		frag++;
13130 		vcpu->mmio_cur_fragment++;
13131 	} else {
13132 		/* Go forward to the next mmio piece. */
13133 		frag->data += len;
13134 		frag->gpa += len;
13135 		frag->len -= len;
13136 	}
13137 
13138 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
13139 		vcpu->mmio_needed = 0;
13140 
13141 		// VMG change, at this point, we're always done
13142 		// RIP has already been advanced
13143 		return 1;
13144 	}
13145 
13146 	// More MMIO is needed
13147 	run->mmio.phys_addr = frag->gpa;
13148 	run->mmio.len = min(8u, frag->len);
13149 	run->mmio.is_write = vcpu->mmio_is_write;
13150 	if (run->mmio.is_write)
13151 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
13152 	run->exit_reason = KVM_EXIT_MMIO;
13153 
13154 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13155 
13156 	return 0;
13157 }
13158 
13159 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13160 			  void *data)
13161 {
13162 	int handled;
13163 	struct kvm_mmio_fragment *frag;
13164 
13165 	if (!data)
13166 		return -EINVAL;
13167 
13168 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13169 	if (handled == bytes)
13170 		return 1;
13171 
13172 	bytes -= handled;
13173 	gpa += handled;
13174 	data += handled;
13175 
13176 	/*TODO: Check if need to increment number of frags */
13177 	frag = vcpu->mmio_fragments;
13178 	vcpu->mmio_nr_fragments = 1;
13179 	frag->len = bytes;
13180 	frag->gpa = gpa;
13181 	frag->data = data;
13182 
13183 	vcpu->mmio_needed = 1;
13184 	vcpu->mmio_cur_fragment = 0;
13185 
13186 	vcpu->run->mmio.phys_addr = gpa;
13187 	vcpu->run->mmio.len = min(8u, frag->len);
13188 	vcpu->run->mmio.is_write = 1;
13189 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
13190 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
13191 
13192 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13193 
13194 	return 0;
13195 }
13196 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
13197 
13198 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13199 			 void *data)
13200 {
13201 	int handled;
13202 	struct kvm_mmio_fragment *frag;
13203 
13204 	if (!data)
13205 		return -EINVAL;
13206 
13207 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13208 	if (handled == bytes)
13209 		return 1;
13210 
13211 	bytes -= handled;
13212 	gpa += handled;
13213 	data += handled;
13214 
13215 	/*TODO: Check if need to increment number of frags */
13216 	frag = vcpu->mmio_fragments;
13217 	vcpu->mmio_nr_fragments = 1;
13218 	frag->len = bytes;
13219 	frag->gpa = gpa;
13220 	frag->data = data;
13221 
13222 	vcpu->mmio_needed = 1;
13223 	vcpu->mmio_cur_fragment = 0;
13224 
13225 	vcpu->run->mmio.phys_addr = gpa;
13226 	vcpu->run->mmio.len = min(8u, frag->len);
13227 	vcpu->run->mmio.is_write = 0;
13228 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
13229 
13230 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13231 
13232 	return 0;
13233 }
13234 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
13235 
13236 static void advance_sev_es_emulated_pio(struct kvm_vcpu *vcpu, unsigned count, int size)
13237 {
13238 	vcpu->arch.sev_pio_count -= count;
13239 	vcpu->arch.sev_pio_data += count * size;
13240 }
13241 
13242 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13243 			   unsigned int port);
13244 
13245 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
13246 {
13247 	int size = vcpu->arch.pio.size;
13248 	int port = vcpu->arch.pio.port;
13249 
13250 	vcpu->arch.pio.count = 0;
13251 	if (vcpu->arch.sev_pio_count)
13252 		return kvm_sev_es_outs(vcpu, size, port);
13253 	return 1;
13254 }
13255 
13256 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13257 			   unsigned int port)
13258 {
13259 	for (;;) {
13260 		unsigned int count =
13261 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13262 		int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
13263 
13264 		/* memcpy done already by emulator_pio_out.  */
13265 		advance_sev_es_emulated_pio(vcpu, count, size);
13266 		if (!ret)
13267 			break;
13268 
13269 		/* Emulation done by the kernel.  */
13270 		if (!vcpu->arch.sev_pio_count)
13271 			return 1;
13272 	}
13273 
13274 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
13275 	return 0;
13276 }
13277 
13278 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13279 			  unsigned int port);
13280 
13281 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
13282 {
13283 	unsigned count = vcpu->arch.pio.count;
13284 	int size = vcpu->arch.pio.size;
13285 	int port = vcpu->arch.pio.port;
13286 
13287 	complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
13288 	advance_sev_es_emulated_pio(vcpu, count, size);
13289 	if (vcpu->arch.sev_pio_count)
13290 		return kvm_sev_es_ins(vcpu, size, port);
13291 	return 1;
13292 }
13293 
13294 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13295 			  unsigned int port)
13296 {
13297 	for (;;) {
13298 		unsigned int count =
13299 			min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13300 		if (!emulator_pio_in(vcpu, size, port, vcpu->arch.sev_pio_data, count))
13301 			break;
13302 
13303 		/* Emulation done by the kernel.  */
13304 		advance_sev_es_emulated_pio(vcpu, count, size);
13305 		if (!vcpu->arch.sev_pio_count)
13306 			return 1;
13307 	}
13308 
13309 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
13310 	return 0;
13311 }
13312 
13313 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
13314 			 unsigned int port, void *data,  unsigned int count,
13315 			 int in)
13316 {
13317 	vcpu->arch.sev_pio_data = data;
13318 	vcpu->arch.sev_pio_count = count;
13319 	return in ? kvm_sev_es_ins(vcpu, size, port)
13320 		  : kvm_sev_es_outs(vcpu, size, port);
13321 }
13322 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
13323 
13324 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
13325 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
13326 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
13327 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
13328 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
13329 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
13330 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
13331 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
13332 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
13333 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
13334 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
13335 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
13336 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
13337 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
13338 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
13339 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
13340 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
13341 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
13342 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
13343 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
13344 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
13345 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
13346 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_kick_vcpu_slowpath);
13347 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_doorbell);
13348 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
13349 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
13350 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
13351 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
13352 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
13353 
13354 static int __init kvm_x86_init(void)
13355 {
13356 	kvm_mmu_x86_module_init();
13357 	return 0;
13358 }
13359 module_init(kvm_x86_init);
13360 
13361 static void __exit kvm_x86_exit(void)
13362 {
13363 	/*
13364 	 * If module_init() is implemented, module_exit() must also be
13365 	 * implemented to allow module unload.
13366 	 */
13367 }
13368 module_exit(kvm_x86_exit);
13369