xref: /linux/arch/x86/kvm/x86.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 
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 <linux/kernel_stat.h>
69 #include <asm/fpu/internal.h> /* Ugh! */
70 #include <asm/pvclock.h>
71 #include <asm/div64.h>
72 #include <asm/irq_remapping.h>
73 #include <asm/mshyperv.h>
74 #include <asm/hypervisor.h>
75 #include <asm/tlbflush.h>
76 #include <asm/intel_pt.h>
77 #include <asm/emulate_prefix.h>
78 #include <asm/sgx.h>
79 #include <clocksource/hyperv_timer.h>
80 
81 #define CREATE_TRACE_POINTS
82 #include "trace.h"
83 
84 #define MAX_IO_MSRS 256
85 #define KVM_MAX_MCE_BANKS 32
86 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
87 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
88 
89 #define emul_to_vcpu(ctxt) \
90 	((struct kvm_vcpu *)(ctxt)->vcpu)
91 
92 /* EFER defaults:
93  * - enable syscall per default because its emulated by KVM
94  * - enable LME and LMA per default on 64 bit KVM
95  */
96 #ifdef CONFIG_X86_64
97 static
98 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
99 #else
100 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
101 #endif
102 
103 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
104 
105 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
106                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
107 
108 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
109 static void process_nmi(struct kvm_vcpu *vcpu);
110 static void process_smi(struct kvm_vcpu *vcpu);
111 static void enter_smm(struct kvm_vcpu *vcpu);
112 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
113 static void store_regs(struct kvm_vcpu *vcpu);
114 static int sync_regs(struct kvm_vcpu *vcpu);
115 
116 struct kvm_x86_ops kvm_x86_ops __read_mostly;
117 EXPORT_SYMBOL_GPL(kvm_x86_ops);
118 
119 #define KVM_X86_OP(func)					     \
120 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
121 				*(((struct kvm_x86_ops *)0)->func));
122 #define KVM_X86_OP_NULL KVM_X86_OP
123 #include <asm/kvm-x86-ops.h>
124 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
126 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
127 
128 static bool __read_mostly ignore_msrs = 0;
129 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
130 
131 bool __read_mostly report_ignored_msrs = true;
132 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
133 EXPORT_SYMBOL_GPL(report_ignored_msrs);
134 
135 unsigned int min_timer_period_us = 200;
136 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
137 
138 static bool __read_mostly kvmclock_periodic_sync = true;
139 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
140 
141 bool __read_mostly kvm_has_tsc_control;
142 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
143 u32  __read_mostly kvm_max_guest_tsc_khz;
144 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
145 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
146 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
147 u64  __read_mostly kvm_max_tsc_scaling_ratio;
148 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
149 u64 __read_mostly kvm_default_tsc_scaling_ratio;
150 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
151 bool __read_mostly kvm_has_bus_lock_exit;
152 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
153 
154 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
155 static u32 __read_mostly tsc_tolerance_ppm = 250;
156 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
157 
158 /*
159  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
160  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
161  * advancement entirely.  Any other value is used as-is and disables adaptive
162  * tuning, i.e. allows privileged userspace to set an exact advancement time.
163  */
164 static int __read_mostly lapic_timer_advance_ns = -1;
165 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
166 
167 static bool __read_mostly vector_hashing = true;
168 module_param(vector_hashing, bool, S_IRUGO);
169 
170 bool __read_mostly enable_vmware_backdoor = false;
171 module_param(enable_vmware_backdoor, bool, S_IRUGO);
172 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
173 
174 static bool __read_mostly force_emulation_prefix = false;
175 module_param(force_emulation_prefix, bool, S_IRUGO);
176 
177 int __read_mostly pi_inject_timer = -1;
178 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
179 
180 /*
181  * Restoring the host value for MSRs that are only consumed when running in
182  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
183  * returns to userspace, i.e. the kernel can run with the guest's value.
184  */
185 #define KVM_MAX_NR_USER_RETURN_MSRS 16
186 
187 struct kvm_user_return_msrs {
188 	struct user_return_notifier urn;
189 	bool registered;
190 	struct kvm_user_return_msr_values {
191 		u64 host;
192 		u64 curr;
193 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
194 };
195 
196 u32 __read_mostly kvm_nr_uret_msrs;
197 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
198 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
199 static struct kvm_user_return_msrs __percpu *user_return_msrs;
200 
201 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
202 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
203 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
204 				| XFEATURE_MASK_PKRU)
205 
206 u64 __read_mostly host_efer;
207 EXPORT_SYMBOL_GPL(host_efer);
208 
209 bool __read_mostly allow_smaller_maxphyaddr = 0;
210 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
211 
212 u64 __read_mostly host_xss;
213 EXPORT_SYMBOL_GPL(host_xss);
214 u64 __read_mostly supported_xss;
215 EXPORT_SYMBOL_GPL(supported_xss);
216 
217 struct kvm_stats_debugfs_item debugfs_entries[] = {
218 	VCPU_STAT("pf_fixed", pf_fixed),
219 	VCPU_STAT("pf_guest", pf_guest),
220 	VCPU_STAT("tlb_flush", tlb_flush),
221 	VCPU_STAT("invlpg", invlpg),
222 	VCPU_STAT("exits", exits),
223 	VCPU_STAT("io_exits", io_exits),
224 	VCPU_STAT("mmio_exits", mmio_exits),
225 	VCPU_STAT("signal_exits", signal_exits),
226 	VCPU_STAT("irq_window", irq_window_exits),
227 	VCPU_STAT("nmi_window", nmi_window_exits),
228 	VCPU_STAT("halt_exits", halt_exits),
229 	VCPU_STAT("halt_successful_poll", halt_successful_poll),
230 	VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
231 	VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
232 	VCPU_STAT("halt_wakeup", halt_wakeup),
233 	VCPU_STAT("hypercalls", hypercalls),
234 	VCPU_STAT("request_irq", request_irq_exits),
235 	VCPU_STAT("irq_exits", irq_exits),
236 	VCPU_STAT("host_state_reload", host_state_reload),
237 	VCPU_STAT("fpu_reload", fpu_reload),
238 	VCPU_STAT("insn_emulation", insn_emulation),
239 	VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
240 	VCPU_STAT("irq_injections", irq_injections),
241 	VCPU_STAT("nmi_injections", nmi_injections),
242 	VCPU_STAT("req_event", req_event),
243 	VCPU_STAT("l1d_flush", l1d_flush),
244 	VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
245 	VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
246 	VCPU_STAT("nested_run", nested_run),
247 	VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
248 	VCPU_STAT("directed_yield_successful", directed_yield_successful),
249 	VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
250 	VM_STAT("mmu_pte_write", mmu_pte_write),
251 	VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
252 	VM_STAT("mmu_flooded", mmu_flooded),
253 	VM_STAT("mmu_recycled", mmu_recycled),
254 	VM_STAT("mmu_cache_miss", mmu_cache_miss),
255 	VM_STAT("mmu_unsync", mmu_unsync),
256 	VM_STAT("remote_tlb_flush", remote_tlb_flush),
257 	VM_STAT("largepages", lpages, .mode = 0444),
258 	VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
259 	VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
260 	{ NULL }
261 };
262 
263 u64 __read_mostly host_xcr0;
264 u64 __read_mostly supported_xcr0;
265 EXPORT_SYMBOL_GPL(supported_xcr0);
266 
267 static struct kmem_cache *x86_fpu_cache;
268 
269 static struct kmem_cache *x86_emulator_cache;
270 
271 /*
272  * When called, it means the previous get/set msr reached an invalid msr.
273  * Return true if we want to ignore/silent this failed msr access.
274  */
275 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
276 {
277 	const char *op = write ? "wrmsr" : "rdmsr";
278 
279 	if (ignore_msrs) {
280 		if (report_ignored_msrs)
281 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
282 				      op, msr, data);
283 		/* Mask the error */
284 		return true;
285 	} else {
286 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
287 				      op, msr, data);
288 		return false;
289 	}
290 }
291 
292 static struct kmem_cache *kvm_alloc_emulator_cache(void)
293 {
294 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
295 	unsigned int size = sizeof(struct x86_emulate_ctxt);
296 
297 	return kmem_cache_create_usercopy("x86_emulator", size,
298 					  __alignof__(struct x86_emulate_ctxt),
299 					  SLAB_ACCOUNT, useroffset,
300 					  size - useroffset, NULL);
301 }
302 
303 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
304 
305 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
306 {
307 	int i;
308 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
309 		vcpu->arch.apf.gfns[i] = ~0;
310 }
311 
312 static void kvm_on_user_return(struct user_return_notifier *urn)
313 {
314 	unsigned slot;
315 	struct kvm_user_return_msrs *msrs
316 		= container_of(urn, struct kvm_user_return_msrs, urn);
317 	struct kvm_user_return_msr_values *values;
318 	unsigned long flags;
319 
320 	/*
321 	 * Disabling irqs at this point since the following code could be
322 	 * interrupted and executed through kvm_arch_hardware_disable()
323 	 */
324 	local_irq_save(flags);
325 	if (msrs->registered) {
326 		msrs->registered = false;
327 		user_return_notifier_unregister(urn);
328 	}
329 	local_irq_restore(flags);
330 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
331 		values = &msrs->values[slot];
332 		if (values->host != values->curr) {
333 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
334 			values->curr = values->host;
335 		}
336 	}
337 }
338 
339 static int kvm_probe_user_return_msr(u32 msr)
340 {
341 	u64 val;
342 	int ret;
343 
344 	preempt_disable();
345 	ret = rdmsrl_safe(msr, &val);
346 	if (ret)
347 		goto out;
348 	ret = wrmsrl_safe(msr, val);
349 out:
350 	preempt_enable();
351 	return ret;
352 }
353 
354 int kvm_add_user_return_msr(u32 msr)
355 {
356 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
357 
358 	if (kvm_probe_user_return_msr(msr))
359 		return -1;
360 
361 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
362 	return kvm_nr_uret_msrs++;
363 }
364 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
365 
366 int kvm_find_user_return_msr(u32 msr)
367 {
368 	int i;
369 
370 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
371 		if (kvm_uret_msrs_list[i] == msr)
372 			return i;
373 	}
374 	return -1;
375 }
376 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
377 
378 static void kvm_user_return_msr_cpu_online(void)
379 {
380 	unsigned int cpu = smp_processor_id();
381 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
382 	u64 value;
383 	int i;
384 
385 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
386 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
387 		msrs->values[i].host = value;
388 		msrs->values[i].curr = value;
389 	}
390 }
391 
392 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
393 {
394 	unsigned int cpu = smp_processor_id();
395 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
396 	int err;
397 
398 	value = (value & mask) | (msrs->values[slot].host & ~mask);
399 	if (value == msrs->values[slot].curr)
400 		return 0;
401 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
402 	if (err)
403 		return 1;
404 
405 	msrs->values[slot].curr = value;
406 	if (!msrs->registered) {
407 		msrs->urn.on_user_return = kvm_on_user_return;
408 		user_return_notifier_register(&msrs->urn);
409 		msrs->registered = true;
410 	}
411 	return 0;
412 }
413 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
414 
415 static void drop_user_return_notifiers(void)
416 {
417 	unsigned int cpu = smp_processor_id();
418 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
419 
420 	if (msrs->registered)
421 		kvm_on_user_return(&msrs->urn);
422 }
423 
424 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
425 {
426 	return vcpu->arch.apic_base;
427 }
428 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
429 
430 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
431 {
432 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
433 }
434 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
435 
436 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
437 {
438 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
439 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
440 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
441 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
442 
443 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
444 		return 1;
445 	if (!msr_info->host_initiated) {
446 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
447 			return 1;
448 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
449 			return 1;
450 	}
451 
452 	kvm_lapic_set_base(vcpu, msr_info->data);
453 	kvm_recalculate_apic_map(vcpu->kvm);
454 	return 0;
455 }
456 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
457 
458 asmlinkage __visible noinstr void kvm_spurious_fault(void)
459 {
460 	/* Fault while not rebooting.  We want the trace. */
461 	BUG_ON(!kvm_rebooting);
462 }
463 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
464 
465 #define EXCPT_BENIGN		0
466 #define EXCPT_CONTRIBUTORY	1
467 #define EXCPT_PF		2
468 
469 static int exception_class(int vector)
470 {
471 	switch (vector) {
472 	case PF_VECTOR:
473 		return EXCPT_PF;
474 	case DE_VECTOR:
475 	case TS_VECTOR:
476 	case NP_VECTOR:
477 	case SS_VECTOR:
478 	case GP_VECTOR:
479 		return EXCPT_CONTRIBUTORY;
480 	default:
481 		break;
482 	}
483 	return EXCPT_BENIGN;
484 }
485 
486 #define EXCPT_FAULT		0
487 #define EXCPT_TRAP		1
488 #define EXCPT_ABORT		2
489 #define EXCPT_INTERRUPT		3
490 
491 static int exception_type(int vector)
492 {
493 	unsigned int mask;
494 
495 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
496 		return EXCPT_INTERRUPT;
497 
498 	mask = 1 << vector;
499 
500 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
501 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
502 		return EXCPT_TRAP;
503 
504 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
505 		return EXCPT_ABORT;
506 
507 	/* Reserved exceptions will result in fault */
508 	return EXCPT_FAULT;
509 }
510 
511 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
512 {
513 	unsigned nr = vcpu->arch.exception.nr;
514 	bool has_payload = vcpu->arch.exception.has_payload;
515 	unsigned long payload = vcpu->arch.exception.payload;
516 
517 	if (!has_payload)
518 		return;
519 
520 	switch (nr) {
521 	case DB_VECTOR:
522 		/*
523 		 * "Certain debug exceptions may clear bit 0-3.  The
524 		 * remaining contents of the DR6 register are never
525 		 * cleared by the processor".
526 		 */
527 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
528 		/*
529 		 * In order to reflect the #DB exception payload in guest
530 		 * dr6, three components need to be considered: active low
531 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
532 		 * DR6_BS and DR6_BT)
533 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
534 		 * In the target guest dr6:
535 		 * FIXED_1 bits should always be set.
536 		 * Active low bits should be cleared if 1-setting in payload.
537 		 * Active high bits should be set if 1-setting in payload.
538 		 *
539 		 * Note, the payload is compatible with the pending debug
540 		 * exceptions/exit qualification under VMX, that active_low bits
541 		 * are active high in payload.
542 		 * So they need to be flipped for DR6.
543 		 */
544 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
545 		vcpu->arch.dr6 |= payload;
546 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
547 
548 		/*
549 		 * The #DB payload is defined as compatible with the 'pending
550 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
551 		 * defined in the 'pending debug exceptions' field (enabled
552 		 * breakpoint), it is reserved and must be zero in DR6.
553 		 */
554 		vcpu->arch.dr6 &= ~BIT(12);
555 		break;
556 	case PF_VECTOR:
557 		vcpu->arch.cr2 = payload;
558 		break;
559 	}
560 
561 	vcpu->arch.exception.has_payload = false;
562 	vcpu->arch.exception.payload = 0;
563 }
564 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
565 
566 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
567 		unsigned nr, bool has_error, u32 error_code,
568 	        bool has_payload, unsigned long payload, bool reinject)
569 {
570 	u32 prev_nr;
571 	int class1, class2;
572 
573 	kvm_make_request(KVM_REQ_EVENT, vcpu);
574 
575 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
576 	queue:
577 		if (reinject) {
578 			/*
579 			 * On vmentry, vcpu->arch.exception.pending is only
580 			 * true if an event injection was blocked by
581 			 * nested_run_pending.  In that case, however,
582 			 * vcpu_enter_guest requests an immediate exit,
583 			 * and the guest shouldn't proceed far enough to
584 			 * need reinjection.
585 			 */
586 			WARN_ON_ONCE(vcpu->arch.exception.pending);
587 			vcpu->arch.exception.injected = true;
588 			if (WARN_ON_ONCE(has_payload)) {
589 				/*
590 				 * A reinjected event has already
591 				 * delivered its payload.
592 				 */
593 				has_payload = false;
594 				payload = 0;
595 			}
596 		} else {
597 			vcpu->arch.exception.pending = true;
598 			vcpu->arch.exception.injected = false;
599 		}
600 		vcpu->arch.exception.has_error_code = has_error;
601 		vcpu->arch.exception.nr = nr;
602 		vcpu->arch.exception.error_code = error_code;
603 		vcpu->arch.exception.has_payload = has_payload;
604 		vcpu->arch.exception.payload = payload;
605 		if (!is_guest_mode(vcpu))
606 			kvm_deliver_exception_payload(vcpu);
607 		return;
608 	}
609 
610 	/* to check exception */
611 	prev_nr = vcpu->arch.exception.nr;
612 	if (prev_nr == DF_VECTOR) {
613 		/* triple fault -> shutdown */
614 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
615 		return;
616 	}
617 	class1 = exception_class(prev_nr);
618 	class2 = exception_class(nr);
619 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
620 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
621 		/*
622 		 * Generate double fault per SDM Table 5-5.  Set
623 		 * exception.pending = true so that the double fault
624 		 * can trigger a nested vmexit.
625 		 */
626 		vcpu->arch.exception.pending = true;
627 		vcpu->arch.exception.injected = false;
628 		vcpu->arch.exception.has_error_code = true;
629 		vcpu->arch.exception.nr = DF_VECTOR;
630 		vcpu->arch.exception.error_code = 0;
631 		vcpu->arch.exception.has_payload = false;
632 		vcpu->arch.exception.payload = 0;
633 	} else
634 		/* replace previous exception with a new one in a hope
635 		   that instruction re-execution will regenerate lost
636 		   exception */
637 		goto queue;
638 }
639 
640 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
641 {
642 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
643 }
644 EXPORT_SYMBOL_GPL(kvm_queue_exception);
645 
646 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
647 {
648 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
649 }
650 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
651 
652 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
653 			   unsigned long payload)
654 {
655 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
656 }
657 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
658 
659 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
660 				    u32 error_code, unsigned long payload)
661 {
662 	kvm_multiple_exception(vcpu, nr, true, error_code,
663 			       true, payload, false);
664 }
665 
666 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
667 {
668 	if (err)
669 		kvm_inject_gp(vcpu, 0);
670 	else
671 		return kvm_skip_emulated_instruction(vcpu);
672 
673 	return 1;
674 }
675 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
676 
677 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
678 {
679 	++vcpu->stat.pf_guest;
680 	vcpu->arch.exception.nested_apf =
681 		is_guest_mode(vcpu) && fault->async_page_fault;
682 	if (vcpu->arch.exception.nested_apf) {
683 		vcpu->arch.apf.nested_apf_token = fault->address;
684 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
685 	} else {
686 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
687 					fault->address);
688 	}
689 }
690 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
691 
692 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
693 				    struct x86_exception *fault)
694 {
695 	struct kvm_mmu *fault_mmu;
696 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
697 
698 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
699 					       vcpu->arch.walk_mmu;
700 
701 	/*
702 	 * Invalidate the TLB entry for the faulting address, if it exists,
703 	 * else the access will fault indefinitely (and to emulate hardware).
704 	 */
705 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
706 	    !(fault->error_code & PFERR_RSVD_MASK))
707 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
708 				       fault_mmu->root_hpa);
709 
710 	fault_mmu->inject_page_fault(vcpu, fault);
711 	return fault->nested_page_fault;
712 }
713 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
714 
715 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
716 {
717 	atomic_inc(&vcpu->arch.nmi_queued);
718 	kvm_make_request(KVM_REQ_NMI, vcpu);
719 }
720 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
721 
722 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
723 {
724 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
725 }
726 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
727 
728 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
729 {
730 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
731 }
732 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
733 
734 /*
735  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
736  * a #GP and return false.
737  */
738 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
739 {
740 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
741 		return true;
742 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
743 	return false;
744 }
745 EXPORT_SYMBOL_GPL(kvm_require_cpl);
746 
747 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
748 {
749 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
750 		return true;
751 
752 	kvm_queue_exception(vcpu, UD_VECTOR);
753 	return false;
754 }
755 EXPORT_SYMBOL_GPL(kvm_require_dr);
756 
757 /*
758  * This function will be used to read from the physical memory of the currently
759  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
760  * can read from guest physical or from the guest's guest physical memory.
761  */
762 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
763 			    gfn_t ngfn, void *data, int offset, int len,
764 			    u32 access)
765 {
766 	struct x86_exception exception;
767 	gfn_t real_gfn;
768 	gpa_t ngpa;
769 
770 	ngpa     = gfn_to_gpa(ngfn);
771 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
772 	if (real_gfn == UNMAPPED_GVA)
773 		return -EFAULT;
774 
775 	real_gfn = gpa_to_gfn(real_gfn);
776 
777 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
778 }
779 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
780 
781 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
782 			       void *data, int offset, int len, u32 access)
783 {
784 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
785 				       data, offset, len, access);
786 }
787 
788 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
789 {
790 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
791 }
792 
793 /*
794  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
795  */
796 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
797 {
798 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
799 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
800 	int i;
801 	int ret;
802 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
803 
804 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
805 				      offset * sizeof(u64), sizeof(pdpte),
806 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
807 	if (ret < 0) {
808 		ret = 0;
809 		goto out;
810 	}
811 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
812 		if ((pdpte[i] & PT_PRESENT_MASK) &&
813 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
814 			ret = 0;
815 			goto out;
816 		}
817 	}
818 	ret = 1;
819 
820 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
821 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
822 
823 out:
824 
825 	return ret;
826 }
827 EXPORT_SYMBOL_GPL(load_pdptrs);
828 
829 bool pdptrs_changed(struct kvm_vcpu *vcpu)
830 {
831 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
832 	int offset;
833 	gfn_t gfn;
834 	int r;
835 
836 	if (!is_pae_paging(vcpu))
837 		return false;
838 
839 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
840 		return true;
841 
842 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
843 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
844 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
845 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
846 	if (r < 0)
847 		return true;
848 
849 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
850 }
851 EXPORT_SYMBOL_GPL(pdptrs_changed);
852 
853 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
854 {
855 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
856 
857 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
858 		kvm_clear_async_pf_completion_queue(vcpu);
859 		kvm_async_pf_hash_reset(vcpu);
860 	}
861 
862 	if ((cr0 ^ old_cr0) & update_bits)
863 		kvm_mmu_reset_context(vcpu);
864 
865 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
866 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
867 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
868 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
869 }
870 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
871 
872 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
873 {
874 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
875 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
876 
877 	cr0 |= X86_CR0_ET;
878 
879 #ifdef CONFIG_X86_64
880 	if (cr0 & 0xffffffff00000000UL)
881 		return 1;
882 #endif
883 
884 	cr0 &= ~CR0_RESERVED_BITS;
885 
886 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
887 		return 1;
888 
889 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
890 		return 1;
891 
892 #ifdef CONFIG_X86_64
893 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
894 	    (cr0 & X86_CR0_PG)) {
895 		int cs_db, cs_l;
896 
897 		if (!is_pae(vcpu))
898 			return 1;
899 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
900 		if (cs_l)
901 			return 1;
902 	}
903 #endif
904 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
905 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
906 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
907 		return 1;
908 
909 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
910 		return 1;
911 
912 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
913 
914 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
915 
916 	return 0;
917 }
918 EXPORT_SYMBOL_GPL(kvm_set_cr0);
919 
920 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
921 {
922 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
923 }
924 EXPORT_SYMBOL_GPL(kvm_lmsw);
925 
926 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
927 {
928 	if (vcpu->arch.guest_state_protected)
929 		return;
930 
931 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
932 
933 		if (vcpu->arch.xcr0 != host_xcr0)
934 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
935 
936 		if (vcpu->arch.xsaves_enabled &&
937 		    vcpu->arch.ia32_xss != host_xss)
938 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
939 	}
940 
941 	if (static_cpu_has(X86_FEATURE_PKU) &&
942 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
943 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
944 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
945 		__write_pkru(vcpu->arch.pkru);
946 }
947 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
948 
949 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
950 {
951 	if (vcpu->arch.guest_state_protected)
952 		return;
953 
954 	if (static_cpu_has(X86_FEATURE_PKU) &&
955 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
956 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
957 		vcpu->arch.pkru = rdpkru();
958 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
959 			__write_pkru(vcpu->arch.host_pkru);
960 	}
961 
962 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
963 
964 		if (vcpu->arch.xcr0 != host_xcr0)
965 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
966 
967 		if (vcpu->arch.xsaves_enabled &&
968 		    vcpu->arch.ia32_xss != host_xss)
969 			wrmsrl(MSR_IA32_XSS, host_xss);
970 	}
971 
972 }
973 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
974 
975 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
976 {
977 	u64 xcr0 = xcr;
978 	u64 old_xcr0 = vcpu->arch.xcr0;
979 	u64 valid_bits;
980 
981 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
982 	if (index != XCR_XFEATURE_ENABLED_MASK)
983 		return 1;
984 	if (!(xcr0 & XFEATURE_MASK_FP))
985 		return 1;
986 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
987 		return 1;
988 
989 	/*
990 	 * Do not allow the guest to set bits that we do not support
991 	 * saving.  However, xcr0 bit 0 is always set, even if the
992 	 * emulated CPU does not support XSAVE (see fx_init).
993 	 */
994 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
995 	if (xcr0 & ~valid_bits)
996 		return 1;
997 
998 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
999 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1000 		return 1;
1001 
1002 	if (xcr0 & XFEATURE_MASK_AVX512) {
1003 		if (!(xcr0 & XFEATURE_MASK_YMM))
1004 			return 1;
1005 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1006 			return 1;
1007 	}
1008 	vcpu->arch.xcr0 = xcr0;
1009 
1010 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1011 		kvm_update_cpuid_runtime(vcpu);
1012 	return 0;
1013 }
1014 
1015 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1016 {
1017 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1018 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1019 		kvm_inject_gp(vcpu, 0);
1020 		return 1;
1021 	}
1022 
1023 	return kvm_skip_emulated_instruction(vcpu);
1024 }
1025 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1026 
1027 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1028 {
1029 	if (cr4 & cr4_reserved_bits)
1030 		return false;
1031 
1032 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1033 		return false;
1034 
1035 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1036 }
1037 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1038 
1039 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1040 {
1041 	unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1042 				      X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1043 
1044 	if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1045 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1046 		kvm_mmu_reset_context(vcpu);
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1049 
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1051 {
1052 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1054 				   X86_CR4_SMEP;
1055 
1056 	if (!kvm_is_valid_cr4(vcpu, cr4))
1057 		return 1;
1058 
1059 	if (is_long_mode(vcpu)) {
1060 		if (!(cr4 & X86_CR4_PAE))
1061 			return 1;
1062 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1063 			return 1;
1064 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1066 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 				   kvm_read_cr3(vcpu)))
1068 		return 1;
1069 
1070 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1072 			return 1;
1073 
1074 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1076 			return 1;
1077 	}
1078 
1079 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1080 
1081 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1082 
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1086 
1087 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1088 {
1089 	bool skip_tlb_flush = false;
1090 #ifdef CONFIG_X86_64
1091 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1092 
1093 	if (pcid_enabled) {
1094 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1095 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1096 	}
1097 #endif
1098 
1099 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1100 		if (!skip_tlb_flush) {
1101 			kvm_mmu_sync_roots(vcpu);
1102 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1103 		}
1104 		return 0;
1105 	}
1106 
1107 	/*
1108 	 * Do not condition the GPA check on long mode, this helper is used to
1109 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1110 	 * the current vCPU mode is accurate.
1111 	 */
1112 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1113 		return 1;
1114 
1115 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1116 		return 1;
1117 
1118 	kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1119 	vcpu->arch.cr3 = cr3;
1120 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1121 
1122 	return 0;
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1125 
1126 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1127 {
1128 	if (cr8 & CR8_RESERVED_BITS)
1129 		return 1;
1130 	if (lapic_in_kernel(vcpu))
1131 		kvm_lapic_set_tpr(vcpu, cr8);
1132 	else
1133 		vcpu->arch.cr8 = cr8;
1134 	return 0;
1135 }
1136 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1137 
1138 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1139 {
1140 	if (lapic_in_kernel(vcpu))
1141 		return kvm_lapic_get_cr8(vcpu);
1142 	else
1143 		return vcpu->arch.cr8;
1144 }
1145 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1146 
1147 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1148 {
1149 	int i;
1150 
1151 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1152 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1153 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1154 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1155 	}
1156 }
1157 
1158 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1159 {
1160 	unsigned long dr7;
1161 
1162 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1163 		dr7 = vcpu->arch.guest_debug_dr7;
1164 	else
1165 		dr7 = vcpu->arch.dr7;
1166 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1167 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1168 	if (dr7 & DR7_BP_EN_MASK)
1169 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1170 }
1171 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1172 
1173 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1174 {
1175 	u64 fixed = DR6_FIXED_1;
1176 
1177 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1178 		fixed |= DR6_RTM;
1179 
1180 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1181 		fixed |= DR6_BUS_LOCK;
1182 	return fixed;
1183 }
1184 
1185 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1186 {
1187 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1188 
1189 	switch (dr) {
1190 	case 0 ... 3:
1191 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1192 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1193 			vcpu->arch.eff_db[dr] = val;
1194 		break;
1195 	case 4:
1196 	case 6:
1197 		if (!kvm_dr6_valid(val))
1198 			return 1; /* #GP */
1199 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1200 		break;
1201 	case 5:
1202 	default: /* 7 */
1203 		if (!kvm_dr7_valid(val))
1204 			return 1; /* #GP */
1205 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1206 		kvm_update_dr7(vcpu);
1207 		break;
1208 	}
1209 
1210 	return 0;
1211 }
1212 EXPORT_SYMBOL_GPL(kvm_set_dr);
1213 
1214 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1215 {
1216 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1217 
1218 	switch (dr) {
1219 	case 0 ... 3:
1220 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1221 		break;
1222 	case 4:
1223 	case 6:
1224 		*val = vcpu->arch.dr6;
1225 		break;
1226 	case 5:
1227 	default: /* 7 */
1228 		*val = vcpu->arch.dr7;
1229 		break;
1230 	}
1231 }
1232 EXPORT_SYMBOL_GPL(kvm_get_dr);
1233 
1234 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1235 {
1236 	u32 ecx = kvm_rcx_read(vcpu);
1237 	u64 data;
1238 
1239 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1240 		kvm_inject_gp(vcpu, 0);
1241 		return 1;
1242 	}
1243 
1244 	kvm_rax_write(vcpu, (u32)data);
1245 	kvm_rdx_write(vcpu, data >> 32);
1246 	return kvm_skip_emulated_instruction(vcpu);
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1249 
1250 /*
1251  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1252  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1253  *
1254  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1255  * extract the supported MSRs from the related const lists.
1256  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1257  * capabilities of the host cpu. This capabilities test skips MSRs that are
1258  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1259  * may depend on host virtualization features rather than host cpu features.
1260  */
1261 
1262 static const u32 msrs_to_save_all[] = {
1263 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1264 	MSR_STAR,
1265 #ifdef CONFIG_X86_64
1266 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1267 #endif
1268 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1269 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1270 	MSR_IA32_SPEC_CTRL,
1271 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1272 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1273 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1274 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1275 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1276 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1277 	MSR_IA32_UMWAIT_CONTROL,
1278 
1279 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1280 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1281 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1282 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1283 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1284 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1285 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1286 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1287 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1288 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1289 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1290 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1291 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1292 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1293 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1294 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1295 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1296 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1297 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1298 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1299 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1300 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1301 };
1302 
1303 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1304 static unsigned num_msrs_to_save;
1305 
1306 static const u32 emulated_msrs_all[] = {
1307 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1308 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1309 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1310 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1311 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1312 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1313 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1314 	HV_X64_MSR_RESET,
1315 	HV_X64_MSR_VP_INDEX,
1316 	HV_X64_MSR_VP_RUNTIME,
1317 	HV_X64_MSR_SCONTROL,
1318 	HV_X64_MSR_STIMER0_CONFIG,
1319 	HV_X64_MSR_VP_ASSIST_PAGE,
1320 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1321 	HV_X64_MSR_TSC_EMULATION_STATUS,
1322 	HV_X64_MSR_SYNDBG_OPTIONS,
1323 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1324 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1325 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1326 
1327 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1328 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1329 
1330 	MSR_IA32_TSC_ADJUST,
1331 	MSR_IA32_TSC_DEADLINE,
1332 	MSR_IA32_ARCH_CAPABILITIES,
1333 	MSR_IA32_PERF_CAPABILITIES,
1334 	MSR_IA32_MISC_ENABLE,
1335 	MSR_IA32_MCG_STATUS,
1336 	MSR_IA32_MCG_CTL,
1337 	MSR_IA32_MCG_EXT_CTL,
1338 	MSR_IA32_SMBASE,
1339 	MSR_SMI_COUNT,
1340 	MSR_PLATFORM_INFO,
1341 	MSR_MISC_FEATURES_ENABLES,
1342 	MSR_AMD64_VIRT_SPEC_CTRL,
1343 	MSR_IA32_POWER_CTL,
1344 	MSR_IA32_UCODE_REV,
1345 
1346 	/*
1347 	 * The following list leaves out MSRs whose values are determined
1348 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1349 	 * We always support the "true" VMX control MSRs, even if the host
1350 	 * processor does not, so I am putting these registers here rather
1351 	 * than in msrs_to_save_all.
1352 	 */
1353 	MSR_IA32_VMX_BASIC,
1354 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1355 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1356 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1357 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1358 	MSR_IA32_VMX_MISC,
1359 	MSR_IA32_VMX_CR0_FIXED0,
1360 	MSR_IA32_VMX_CR4_FIXED0,
1361 	MSR_IA32_VMX_VMCS_ENUM,
1362 	MSR_IA32_VMX_PROCBASED_CTLS2,
1363 	MSR_IA32_VMX_EPT_VPID_CAP,
1364 	MSR_IA32_VMX_VMFUNC,
1365 
1366 	MSR_K7_HWCR,
1367 	MSR_KVM_POLL_CONTROL,
1368 };
1369 
1370 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1371 static unsigned num_emulated_msrs;
1372 
1373 /*
1374  * List of msr numbers which are used to expose MSR-based features that
1375  * can be used by a hypervisor to validate requested CPU features.
1376  */
1377 static const u32 msr_based_features_all[] = {
1378 	MSR_IA32_VMX_BASIC,
1379 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1380 	MSR_IA32_VMX_PINBASED_CTLS,
1381 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1382 	MSR_IA32_VMX_PROCBASED_CTLS,
1383 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1384 	MSR_IA32_VMX_EXIT_CTLS,
1385 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1386 	MSR_IA32_VMX_ENTRY_CTLS,
1387 	MSR_IA32_VMX_MISC,
1388 	MSR_IA32_VMX_CR0_FIXED0,
1389 	MSR_IA32_VMX_CR0_FIXED1,
1390 	MSR_IA32_VMX_CR4_FIXED0,
1391 	MSR_IA32_VMX_CR4_FIXED1,
1392 	MSR_IA32_VMX_VMCS_ENUM,
1393 	MSR_IA32_VMX_PROCBASED_CTLS2,
1394 	MSR_IA32_VMX_EPT_VPID_CAP,
1395 	MSR_IA32_VMX_VMFUNC,
1396 
1397 	MSR_F10H_DECFG,
1398 	MSR_IA32_UCODE_REV,
1399 	MSR_IA32_ARCH_CAPABILITIES,
1400 	MSR_IA32_PERF_CAPABILITIES,
1401 };
1402 
1403 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1404 static unsigned int num_msr_based_features;
1405 
1406 static u64 kvm_get_arch_capabilities(void)
1407 {
1408 	u64 data = 0;
1409 
1410 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1411 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1412 
1413 	/*
1414 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1415 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1416 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1417 	 * L1 guests, so it need not worry about its own (L2) guests.
1418 	 */
1419 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1420 
1421 	/*
1422 	 * If we're doing cache flushes (either "always" or "cond")
1423 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1424 	 * If an outer hypervisor is doing the cache flush for us
1425 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1426 	 * capability to the guest too, and if EPT is disabled we're not
1427 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1428 	 * require a nested hypervisor to do a flush of its own.
1429 	 */
1430 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1431 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1432 
1433 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1434 		data |= ARCH_CAP_RDCL_NO;
1435 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1436 		data |= ARCH_CAP_SSB_NO;
1437 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1438 		data |= ARCH_CAP_MDS_NO;
1439 
1440 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1441 		/*
1442 		 * If RTM=0 because the kernel has disabled TSX, the host might
1443 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1444 		 * and therefore knows that there cannot be TAA) but keep
1445 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1446 		 * and we want to allow migrating those guests to tsx=off hosts.
1447 		 */
1448 		data &= ~ARCH_CAP_TAA_NO;
1449 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1450 		data |= ARCH_CAP_TAA_NO;
1451 	} else {
1452 		/*
1453 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1454 		 * host so the guest can choose between disabling TSX or
1455 		 * using VERW to clear CPU buffers.
1456 		 */
1457 	}
1458 
1459 	return data;
1460 }
1461 
1462 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1463 {
1464 	switch (msr->index) {
1465 	case MSR_IA32_ARCH_CAPABILITIES:
1466 		msr->data = kvm_get_arch_capabilities();
1467 		break;
1468 	case MSR_IA32_UCODE_REV:
1469 		rdmsrl_safe(msr->index, &msr->data);
1470 		break;
1471 	default:
1472 		return static_call(kvm_x86_get_msr_feature)(msr);
1473 	}
1474 	return 0;
1475 }
1476 
1477 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1478 {
1479 	struct kvm_msr_entry msr;
1480 	int r;
1481 
1482 	msr.index = index;
1483 	r = kvm_get_msr_feature(&msr);
1484 
1485 	if (r == KVM_MSR_RET_INVALID) {
1486 		/* Unconditionally clear the output for simplicity */
1487 		*data = 0;
1488 		if (kvm_msr_ignored_check(index, 0, false))
1489 			r = 0;
1490 	}
1491 
1492 	if (r)
1493 		return r;
1494 
1495 	*data = msr.data;
1496 
1497 	return 0;
1498 }
1499 
1500 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1501 {
1502 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1503 		return false;
1504 
1505 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1506 		return false;
1507 
1508 	if (efer & (EFER_LME | EFER_LMA) &&
1509 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1510 		return false;
1511 
1512 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1513 		return false;
1514 
1515 	return true;
1516 
1517 }
1518 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1519 {
1520 	if (efer & efer_reserved_bits)
1521 		return false;
1522 
1523 	return __kvm_valid_efer(vcpu, efer);
1524 }
1525 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1526 
1527 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1528 {
1529 	u64 old_efer = vcpu->arch.efer;
1530 	u64 efer = msr_info->data;
1531 	int r;
1532 
1533 	if (efer & efer_reserved_bits)
1534 		return 1;
1535 
1536 	if (!msr_info->host_initiated) {
1537 		if (!__kvm_valid_efer(vcpu, efer))
1538 			return 1;
1539 
1540 		if (is_paging(vcpu) &&
1541 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1542 			return 1;
1543 	}
1544 
1545 	efer &= ~EFER_LMA;
1546 	efer |= vcpu->arch.efer & EFER_LMA;
1547 
1548 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1549 	if (r) {
1550 		WARN_ON(r > 0);
1551 		return r;
1552 	}
1553 
1554 	/* Update reserved bits */
1555 	if ((efer ^ old_efer) & EFER_NX)
1556 		kvm_mmu_reset_context(vcpu);
1557 
1558 	return 0;
1559 }
1560 
1561 void kvm_enable_efer_bits(u64 mask)
1562 {
1563        efer_reserved_bits &= ~mask;
1564 }
1565 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1566 
1567 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1568 {
1569 	struct kvm_x86_msr_filter *msr_filter;
1570 	struct msr_bitmap_range *ranges;
1571 	struct kvm *kvm = vcpu->kvm;
1572 	bool allowed;
1573 	int idx;
1574 	u32 i;
1575 
1576 	/* x2APIC MSRs do not support filtering. */
1577 	if (index >= 0x800 && index <= 0x8ff)
1578 		return true;
1579 
1580 	idx = srcu_read_lock(&kvm->srcu);
1581 
1582 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1583 	if (!msr_filter) {
1584 		allowed = true;
1585 		goto out;
1586 	}
1587 
1588 	allowed = msr_filter->default_allow;
1589 	ranges = msr_filter->ranges;
1590 
1591 	for (i = 0; i < msr_filter->count; i++) {
1592 		u32 start = ranges[i].base;
1593 		u32 end = start + ranges[i].nmsrs;
1594 		u32 flags = ranges[i].flags;
1595 		unsigned long *bitmap = ranges[i].bitmap;
1596 
1597 		if ((index >= start) && (index < end) && (flags & type)) {
1598 			allowed = !!test_bit(index - start, bitmap);
1599 			break;
1600 		}
1601 	}
1602 
1603 out:
1604 	srcu_read_unlock(&kvm->srcu, idx);
1605 
1606 	return allowed;
1607 }
1608 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1609 
1610 /*
1611  * Write @data into the MSR specified by @index.  Select MSR specific fault
1612  * checks are bypassed if @host_initiated is %true.
1613  * Returns 0 on success, non-0 otherwise.
1614  * Assumes vcpu_load() was already called.
1615  */
1616 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1617 			 bool host_initiated)
1618 {
1619 	struct msr_data msr;
1620 
1621 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1622 		return KVM_MSR_RET_FILTERED;
1623 
1624 	switch (index) {
1625 	case MSR_FS_BASE:
1626 	case MSR_GS_BASE:
1627 	case MSR_KERNEL_GS_BASE:
1628 	case MSR_CSTAR:
1629 	case MSR_LSTAR:
1630 		if (is_noncanonical_address(data, vcpu))
1631 			return 1;
1632 		break;
1633 	case MSR_IA32_SYSENTER_EIP:
1634 	case MSR_IA32_SYSENTER_ESP:
1635 		/*
1636 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1637 		 * non-canonical address is written on Intel but not on
1638 		 * AMD (which ignores the top 32-bits, because it does
1639 		 * not implement 64-bit SYSENTER).
1640 		 *
1641 		 * 64-bit code should hence be able to write a non-canonical
1642 		 * value on AMD.  Making the address canonical ensures that
1643 		 * vmentry does not fail on Intel after writing a non-canonical
1644 		 * value, and that something deterministic happens if the guest
1645 		 * invokes 64-bit SYSENTER.
1646 		 */
1647 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1648 		break;
1649 	case MSR_TSC_AUX:
1650 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1651 			return 1;
1652 
1653 		if (!host_initiated &&
1654 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1655 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1656 			return 1;
1657 
1658 		/*
1659 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1660 		 * incomplete and conflicting architectural behavior.  Current
1661 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1662 		 * reserved and always read as zeros.  Enforce Intel's reserved
1663 		 * bits check if and only if the guest CPU is Intel, and clear
1664 		 * the bits in all other cases.  This ensures cross-vendor
1665 		 * migration will provide consistent behavior for the guest.
1666 		 */
1667 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1668 			return 1;
1669 
1670 		data = (u32)data;
1671 		break;
1672 	}
1673 
1674 	msr.data = data;
1675 	msr.index = index;
1676 	msr.host_initiated = host_initiated;
1677 
1678 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1679 }
1680 
1681 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1682 				     u32 index, u64 data, bool host_initiated)
1683 {
1684 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1685 
1686 	if (ret == KVM_MSR_RET_INVALID)
1687 		if (kvm_msr_ignored_check(index, data, true))
1688 			ret = 0;
1689 
1690 	return ret;
1691 }
1692 
1693 /*
1694  * Read the MSR specified by @index into @data.  Select MSR specific fault
1695  * checks are bypassed if @host_initiated is %true.
1696  * Returns 0 on success, non-0 otherwise.
1697  * Assumes vcpu_load() was already called.
1698  */
1699 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1700 		  bool host_initiated)
1701 {
1702 	struct msr_data msr;
1703 	int ret;
1704 
1705 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1706 		return KVM_MSR_RET_FILTERED;
1707 
1708 	switch (index) {
1709 	case MSR_TSC_AUX:
1710 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1711 			return 1;
1712 
1713 		if (!host_initiated &&
1714 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1715 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1716 			return 1;
1717 		break;
1718 	}
1719 
1720 	msr.index = index;
1721 	msr.host_initiated = host_initiated;
1722 
1723 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1724 	if (!ret)
1725 		*data = msr.data;
1726 	return ret;
1727 }
1728 
1729 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1730 				     u32 index, u64 *data, bool host_initiated)
1731 {
1732 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1733 
1734 	if (ret == KVM_MSR_RET_INVALID) {
1735 		/* Unconditionally clear *data for simplicity */
1736 		*data = 0;
1737 		if (kvm_msr_ignored_check(index, 0, false))
1738 			ret = 0;
1739 	}
1740 
1741 	return ret;
1742 }
1743 
1744 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1745 {
1746 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1747 }
1748 EXPORT_SYMBOL_GPL(kvm_get_msr);
1749 
1750 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1751 {
1752 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1753 }
1754 EXPORT_SYMBOL_GPL(kvm_set_msr);
1755 
1756 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1757 {
1758 	int err = vcpu->run->msr.error;
1759 	if (!err) {
1760 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1761 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1762 	}
1763 
1764 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1765 }
1766 
1767 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1768 {
1769 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1770 }
1771 
1772 static u64 kvm_msr_reason(int r)
1773 {
1774 	switch (r) {
1775 	case KVM_MSR_RET_INVALID:
1776 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1777 	case KVM_MSR_RET_FILTERED:
1778 		return KVM_MSR_EXIT_REASON_FILTER;
1779 	default:
1780 		return KVM_MSR_EXIT_REASON_INVAL;
1781 	}
1782 }
1783 
1784 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1785 			      u32 exit_reason, u64 data,
1786 			      int (*completion)(struct kvm_vcpu *vcpu),
1787 			      int r)
1788 {
1789 	u64 msr_reason = kvm_msr_reason(r);
1790 
1791 	/* Check if the user wanted to know about this MSR fault */
1792 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1793 		return 0;
1794 
1795 	vcpu->run->exit_reason = exit_reason;
1796 	vcpu->run->msr.error = 0;
1797 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1798 	vcpu->run->msr.reason = msr_reason;
1799 	vcpu->run->msr.index = index;
1800 	vcpu->run->msr.data = data;
1801 	vcpu->arch.complete_userspace_io = completion;
1802 
1803 	return 1;
1804 }
1805 
1806 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1807 {
1808 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1809 				   complete_emulated_rdmsr, r);
1810 }
1811 
1812 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1813 {
1814 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1815 				   complete_emulated_wrmsr, r);
1816 }
1817 
1818 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1819 {
1820 	u32 ecx = kvm_rcx_read(vcpu);
1821 	u64 data;
1822 	int r;
1823 
1824 	r = kvm_get_msr(vcpu, ecx, &data);
1825 
1826 	/* MSR read failed? See if we should ask user space */
1827 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1828 		/* Bounce to user space */
1829 		return 0;
1830 	}
1831 
1832 	if (!r) {
1833 		trace_kvm_msr_read(ecx, data);
1834 
1835 		kvm_rax_write(vcpu, data & -1u);
1836 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1837 	} else {
1838 		trace_kvm_msr_read_ex(ecx);
1839 	}
1840 
1841 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1842 }
1843 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1844 
1845 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1846 {
1847 	u32 ecx = kvm_rcx_read(vcpu);
1848 	u64 data = kvm_read_edx_eax(vcpu);
1849 	int r;
1850 
1851 	r = kvm_set_msr(vcpu, ecx, data);
1852 
1853 	/* MSR write failed? See if we should ask user space */
1854 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1855 		/* Bounce to user space */
1856 		return 0;
1857 
1858 	/* Signal all other negative errors to userspace */
1859 	if (r < 0)
1860 		return r;
1861 
1862 	if (!r)
1863 		trace_kvm_msr_write(ecx, data);
1864 	else
1865 		trace_kvm_msr_write_ex(ecx, data);
1866 
1867 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1868 }
1869 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1870 
1871 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1872 {
1873 	return kvm_skip_emulated_instruction(vcpu);
1874 }
1875 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1876 
1877 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1878 {
1879 	/* Treat an INVD instruction as a NOP and just skip it. */
1880 	return kvm_emulate_as_nop(vcpu);
1881 }
1882 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1883 
1884 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1885 {
1886 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1887 	return kvm_emulate_as_nop(vcpu);
1888 }
1889 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1890 
1891 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1892 {
1893 	kvm_queue_exception(vcpu, UD_VECTOR);
1894 	return 1;
1895 }
1896 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1897 
1898 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1899 {
1900 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1901 	return kvm_emulate_as_nop(vcpu);
1902 }
1903 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1904 
1905 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1906 {
1907 	xfer_to_guest_mode_prepare();
1908 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1909 		xfer_to_guest_mode_work_pending();
1910 }
1911 
1912 /*
1913  * The fast path for frequent and performance sensitive wrmsr emulation,
1914  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1915  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1916  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1917  * other cases which must be called after interrupts are enabled on the host.
1918  */
1919 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1920 {
1921 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1922 		return 1;
1923 
1924 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1925 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1926 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1927 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1928 
1929 		data &= ~(1 << 12);
1930 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1931 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1932 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1933 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1934 		return 0;
1935 	}
1936 
1937 	return 1;
1938 }
1939 
1940 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1941 {
1942 	if (!kvm_can_use_hv_timer(vcpu))
1943 		return 1;
1944 
1945 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1946 	return 0;
1947 }
1948 
1949 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1950 {
1951 	u32 msr = kvm_rcx_read(vcpu);
1952 	u64 data;
1953 	fastpath_t ret = EXIT_FASTPATH_NONE;
1954 
1955 	switch (msr) {
1956 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1957 		data = kvm_read_edx_eax(vcpu);
1958 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1959 			kvm_skip_emulated_instruction(vcpu);
1960 			ret = EXIT_FASTPATH_EXIT_HANDLED;
1961 		}
1962 		break;
1963 	case MSR_IA32_TSC_DEADLINE:
1964 		data = kvm_read_edx_eax(vcpu);
1965 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1966 			kvm_skip_emulated_instruction(vcpu);
1967 			ret = EXIT_FASTPATH_REENTER_GUEST;
1968 		}
1969 		break;
1970 	default:
1971 		break;
1972 	}
1973 
1974 	if (ret != EXIT_FASTPATH_NONE)
1975 		trace_kvm_msr_write(msr, data);
1976 
1977 	return ret;
1978 }
1979 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1980 
1981 /*
1982  * Adapt set_msr() to msr_io()'s calling convention
1983  */
1984 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1985 {
1986 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
1987 }
1988 
1989 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1990 {
1991 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1992 }
1993 
1994 #ifdef CONFIG_X86_64
1995 struct pvclock_clock {
1996 	int vclock_mode;
1997 	u64 cycle_last;
1998 	u64 mask;
1999 	u32 mult;
2000 	u32 shift;
2001 	u64 base_cycles;
2002 	u64 offset;
2003 };
2004 
2005 struct pvclock_gtod_data {
2006 	seqcount_t	seq;
2007 
2008 	struct pvclock_clock clock; /* extract of a clocksource struct */
2009 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2010 
2011 	ktime_t		offs_boot;
2012 	u64		wall_time_sec;
2013 };
2014 
2015 static struct pvclock_gtod_data pvclock_gtod_data;
2016 
2017 static void update_pvclock_gtod(struct timekeeper *tk)
2018 {
2019 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2020 
2021 	write_seqcount_begin(&vdata->seq);
2022 
2023 	/* copy pvclock gtod data */
2024 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2025 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2026 	vdata->clock.mask		= tk->tkr_mono.mask;
2027 	vdata->clock.mult		= tk->tkr_mono.mult;
2028 	vdata->clock.shift		= tk->tkr_mono.shift;
2029 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2030 	vdata->clock.offset		= tk->tkr_mono.base;
2031 
2032 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2033 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2034 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2035 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2036 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2037 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2038 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2039 
2040 	vdata->wall_time_sec            = tk->xtime_sec;
2041 
2042 	vdata->offs_boot		= tk->offs_boot;
2043 
2044 	write_seqcount_end(&vdata->seq);
2045 }
2046 
2047 static s64 get_kvmclock_base_ns(void)
2048 {
2049 	/* Count up from boot time, but with the frequency of the raw clock.  */
2050 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2051 }
2052 #else
2053 static s64 get_kvmclock_base_ns(void)
2054 {
2055 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2056 	return ktime_get_boottime_ns();
2057 }
2058 #endif
2059 
2060 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2061 {
2062 	int version;
2063 	int r;
2064 	struct pvclock_wall_clock wc;
2065 	u32 wc_sec_hi;
2066 	u64 wall_nsec;
2067 
2068 	if (!wall_clock)
2069 		return;
2070 
2071 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2072 	if (r)
2073 		return;
2074 
2075 	if (version & 1)
2076 		++version;  /* first time write, random junk */
2077 
2078 	++version;
2079 
2080 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2081 		return;
2082 
2083 	/*
2084 	 * The guest calculates current wall clock time by adding
2085 	 * system time (updated by kvm_guest_time_update below) to the
2086 	 * wall clock specified here.  We do the reverse here.
2087 	 */
2088 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2089 
2090 	wc.nsec = do_div(wall_nsec, 1000000000);
2091 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2092 	wc.version = version;
2093 
2094 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2095 
2096 	if (sec_hi_ofs) {
2097 		wc_sec_hi = wall_nsec >> 32;
2098 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2099 				&wc_sec_hi, sizeof(wc_sec_hi));
2100 	}
2101 
2102 	version++;
2103 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2104 }
2105 
2106 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2107 				  bool old_msr, bool host_initiated)
2108 {
2109 	struct kvm_arch *ka = &vcpu->kvm->arch;
2110 
2111 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2112 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2113 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2114 
2115 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2116 	}
2117 
2118 	vcpu->arch.time = system_time;
2119 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2120 
2121 	/* we verify if the enable bit is set... */
2122 	vcpu->arch.pv_time_enabled = false;
2123 	if (!(system_time & 1))
2124 		return;
2125 
2126 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2127 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2128 				       sizeof(struct pvclock_vcpu_time_info)))
2129 		vcpu->arch.pv_time_enabled = true;
2130 
2131 	return;
2132 }
2133 
2134 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2135 {
2136 	do_shl32_div32(dividend, divisor);
2137 	return dividend;
2138 }
2139 
2140 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2141 			       s8 *pshift, u32 *pmultiplier)
2142 {
2143 	uint64_t scaled64;
2144 	int32_t  shift = 0;
2145 	uint64_t tps64;
2146 	uint32_t tps32;
2147 
2148 	tps64 = base_hz;
2149 	scaled64 = scaled_hz;
2150 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2151 		tps64 >>= 1;
2152 		shift--;
2153 	}
2154 
2155 	tps32 = (uint32_t)tps64;
2156 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2157 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2158 			scaled64 >>= 1;
2159 		else
2160 			tps32 <<= 1;
2161 		shift++;
2162 	}
2163 
2164 	*pshift = shift;
2165 	*pmultiplier = div_frac(scaled64, tps32);
2166 }
2167 
2168 #ifdef CONFIG_X86_64
2169 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2170 #endif
2171 
2172 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2173 static unsigned long max_tsc_khz;
2174 
2175 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2176 {
2177 	u64 v = (u64)khz * (1000000 + ppm);
2178 	do_div(v, 1000000);
2179 	return v;
2180 }
2181 
2182 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2183 {
2184 	u64 ratio;
2185 
2186 	/* Guest TSC same frequency as host TSC? */
2187 	if (!scale) {
2188 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2189 		return 0;
2190 	}
2191 
2192 	/* TSC scaling supported? */
2193 	if (!kvm_has_tsc_control) {
2194 		if (user_tsc_khz > tsc_khz) {
2195 			vcpu->arch.tsc_catchup = 1;
2196 			vcpu->arch.tsc_always_catchup = 1;
2197 			return 0;
2198 		} else {
2199 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2200 			return -1;
2201 		}
2202 	}
2203 
2204 	/* TSC scaling required  - calculate ratio */
2205 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2206 				user_tsc_khz, tsc_khz);
2207 
2208 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2209 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2210 			            user_tsc_khz);
2211 		return -1;
2212 	}
2213 
2214 	vcpu->arch.tsc_scaling_ratio = ratio;
2215 	return 0;
2216 }
2217 
2218 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2219 {
2220 	u32 thresh_lo, thresh_hi;
2221 	int use_scaling = 0;
2222 
2223 	/* tsc_khz can be zero if TSC calibration fails */
2224 	if (user_tsc_khz == 0) {
2225 		/* set tsc_scaling_ratio to a safe value */
2226 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2227 		return -1;
2228 	}
2229 
2230 	/* Compute a scale to convert nanoseconds in TSC cycles */
2231 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2232 			   &vcpu->arch.virtual_tsc_shift,
2233 			   &vcpu->arch.virtual_tsc_mult);
2234 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2235 
2236 	/*
2237 	 * Compute the variation in TSC rate which is acceptable
2238 	 * within the range of tolerance and decide if the
2239 	 * rate being applied is within that bounds of the hardware
2240 	 * rate.  If so, no scaling or compensation need be done.
2241 	 */
2242 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2243 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2244 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2245 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2246 		use_scaling = 1;
2247 	}
2248 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2249 }
2250 
2251 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2252 {
2253 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2254 				      vcpu->arch.virtual_tsc_mult,
2255 				      vcpu->arch.virtual_tsc_shift);
2256 	tsc += vcpu->arch.this_tsc_write;
2257 	return tsc;
2258 }
2259 
2260 static inline int gtod_is_based_on_tsc(int mode)
2261 {
2262 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2263 }
2264 
2265 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2266 {
2267 #ifdef CONFIG_X86_64
2268 	bool vcpus_matched;
2269 	struct kvm_arch *ka = &vcpu->kvm->arch;
2270 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2271 
2272 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2273 			 atomic_read(&vcpu->kvm->online_vcpus));
2274 
2275 	/*
2276 	 * Once the masterclock is enabled, always perform request in
2277 	 * order to update it.
2278 	 *
2279 	 * In order to enable masterclock, the host clocksource must be TSC
2280 	 * and the vcpus need to have matched TSCs.  When that happens,
2281 	 * perform request to enable masterclock.
2282 	 */
2283 	if (ka->use_master_clock ||
2284 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2285 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2286 
2287 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2288 			    atomic_read(&vcpu->kvm->online_vcpus),
2289 		            ka->use_master_clock, gtod->clock.vclock_mode);
2290 #endif
2291 }
2292 
2293 /*
2294  * Multiply tsc by a fixed point number represented by ratio.
2295  *
2296  * The most significant 64-N bits (mult) of ratio represent the
2297  * integral part of the fixed point number; the remaining N bits
2298  * (frac) represent the fractional part, ie. ratio represents a fixed
2299  * point number (mult + frac * 2^(-N)).
2300  *
2301  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2302  */
2303 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2304 {
2305 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2306 }
2307 
2308 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2309 {
2310 	u64 _tsc = tsc;
2311 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
2312 
2313 	if (ratio != kvm_default_tsc_scaling_ratio)
2314 		_tsc = __scale_tsc(ratio, tsc);
2315 
2316 	return _tsc;
2317 }
2318 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2319 
2320 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2321 {
2322 	u64 tsc;
2323 
2324 	tsc = kvm_scale_tsc(vcpu, rdtsc());
2325 
2326 	return target_tsc - tsc;
2327 }
2328 
2329 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2330 {
2331 	return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2332 }
2333 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2334 
2335 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2336 {
2337 	vcpu->arch.l1_tsc_offset = offset;
2338 	vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2339 }
2340 
2341 static inline bool kvm_check_tsc_unstable(void)
2342 {
2343 #ifdef CONFIG_X86_64
2344 	/*
2345 	 * TSC is marked unstable when we're running on Hyper-V,
2346 	 * 'TSC page' clocksource is good.
2347 	 */
2348 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2349 		return false;
2350 #endif
2351 	return check_tsc_unstable();
2352 }
2353 
2354 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2355 {
2356 	struct kvm *kvm = vcpu->kvm;
2357 	u64 offset, ns, elapsed;
2358 	unsigned long flags;
2359 	bool matched;
2360 	bool already_matched;
2361 	bool synchronizing = false;
2362 
2363 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2364 	offset = kvm_compute_tsc_offset(vcpu, data);
2365 	ns = get_kvmclock_base_ns();
2366 	elapsed = ns - kvm->arch.last_tsc_nsec;
2367 
2368 	if (vcpu->arch.virtual_tsc_khz) {
2369 		if (data == 0) {
2370 			/*
2371 			 * detection of vcpu initialization -- need to sync
2372 			 * with other vCPUs. This particularly helps to keep
2373 			 * kvm_clock stable after CPU hotplug
2374 			 */
2375 			synchronizing = true;
2376 		} else {
2377 			u64 tsc_exp = kvm->arch.last_tsc_write +
2378 						nsec_to_cycles(vcpu, elapsed);
2379 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2380 			/*
2381 			 * Special case: TSC write with a small delta (1 second)
2382 			 * of virtual cycle time against real time is
2383 			 * interpreted as an attempt to synchronize the CPU.
2384 			 */
2385 			synchronizing = data < tsc_exp + tsc_hz &&
2386 					data + tsc_hz > tsc_exp;
2387 		}
2388 	}
2389 
2390 	/*
2391 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2392 	 * TSC, we add elapsed time in this computation.  We could let the
2393 	 * compensation code attempt to catch up if we fall behind, but
2394 	 * it's better to try to match offsets from the beginning.
2395          */
2396 	if (synchronizing &&
2397 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2398 		if (!kvm_check_tsc_unstable()) {
2399 			offset = kvm->arch.cur_tsc_offset;
2400 		} else {
2401 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2402 			data += delta;
2403 			offset = kvm_compute_tsc_offset(vcpu, data);
2404 		}
2405 		matched = true;
2406 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2407 	} else {
2408 		/*
2409 		 * We split periods of matched TSC writes into generations.
2410 		 * For each generation, we track the original measured
2411 		 * nanosecond time, offset, and write, so if TSCs are in
2412 		 * sync, we can match exact offset, and if not, we can match
2413 		 * exact software computation in compute_guest_tsc()
2414 		 *
2415 		 * These values are tracked in kvm->arch.cur_xxx variables.
2416 		 */
2417 		kvm->arch.cur_tsc_generation++;
2418 		kvm->arch.cur_tsc_nsec = ns;
2419 		kvm->arch.cur_tsc_write = data;
2420 		kvm->arch.cur_tsc_offset = offset;
2421 		matched = false;
2422 	}
2423 
2424 	/*
2425 	 * We also track th most recent recorded KHZ, write and time to
2426 	 * allow the matching interval to be extended at each write.
2427 	 */
2428 	kvm->arch.last_tsc_nsec = ns;
2429 	kvm->arch.last_tsc_write = data;
2430 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2431 
2432 	vcpu->arch.last_guest_tsc = data;
2433 
2434 	/* Keep track of which generation this VCPU has synchronized to */
2435 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2436 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2437 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2438 
2439 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2440 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2441 
2442 	spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2443 	if (!matched) {
2444 		kvm->arch.nr_vcpus_matched_tsc = 0;
2445 	} else if (!already_matched) {
2446 		kvm->arch.nr_vcpus_matched_tsc++;
2447 	}
2448 
2449 	kvm_track_tsc_matching(vcpu);
2450 	spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2451 }
2452 
2453 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2454 					   s64 adjustment)
2455 {
2456 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2457 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2458 }
2459 
2460 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2461 {
2462 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2463 		WARN_ON(adjustment < 0);
2464 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2465 	adjust_tsc_offset_guest(vcpu, adjustment);
2466 }
2467 
2468 #ifdef CONFIG_X86_64
2469 
2470 static u64 read_tsc(void)
2471 {
2472 	u64 ret = (u64)rdtsc_ordered();
2473 	u64 last = pvclock_gtod_data.clock.cycle_last;
2474 
2475 	if (likely(ret >= last))
2476 		return ret;
2477 
2478 	/*
2479 	 * GCC likes to generate cmov here, but this branch is extremely
2480 	 * predictable (it's just a function of time and the likely is
2481 	 * very likely) and there's a data dependence, so force GCC
2482 	 * to generate a branch instead.  I don't barrier() because
2483 	 * we don't actually need a barrier, and if this function
2484 	 * ever gets inlined it will generate worse code.
2485 	 */
2486 	asm volatile ("");
2487 	return last;
2488 }
2489 
2490 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2491 			  int *mode)
2492 {
2493 	long v;
2494 	u64 tsc_pg_val;
2495 
2496 	switch (clock->vclock_mode) {
2497 	case VDSO_CLOCKMODE_HVCLOCK:
2498 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2499 						  tsc_timestamp);
2500 		if (tsc_pg_val != U64_MAX) {
2501 			/* TSC page valid */
2502 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2503 			v = (tsc_pg_val - clock->cycle_last) &
2504 				clock->mask;
2505 		} else {
2506 			/* TSC page invalid */
2507 			*mode = VDSO_CLOCKMODE_NONE;
2508 		}
2509 		break;
2510 	case VDSO_CLOCKMODE_TSC:
2511 		*mode = VDSO_CLOCKMODE_TSC;
2512 		*tsc_timestamp = read_tsc();
2513 		v = (*tsc_timestamp - clock->cycle_last) &
2514 			clock->mask;
2515 		break;
2516 	default:
2517 		*mode = VDSO_CLOCKMODE_NONE;
2518 	}
2519 
2520 	if (*mode == VDSO_CLOCKMODE_NONE)
2521 		*tsc_timestamp = v = 0;
2522 
2523 	return v * clock->mult;
2524 }
2525 
2526 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2527 {
2528 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2529 	unsigned long seq;
2530 	int mode;
2531 	u64 ns;
2532 
2533 	do {
2534 		seq = read_seqcount_begin(&gtod->seq);
2535 		ns = gtod->raw_clock.base_cycles;
2536 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2537 		ns >>= gtod->raw_clock.shift;
2538 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2539 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2540 	*t = ns;
2541 
2542 	return mode;
2543 }
2544 
2545 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2546 {
2547 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2548 	unsigned long seq;
2549 	int mode;
2550 	u64 ns;
2551 
2552 	do {
2553 		seq = read_seqcount_begin(&gtod->seq);
2554 		ts->tv_sec = gtod->wall_time_sec;
2555 		ns = gtod->clock.base_cycles;
2556 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2557 		ns >>= gtod->clock.shift;
2558 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2559 
2560 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2561 	ts->tv_nsec = ns;
2562 
2563 	return mode;
2564 }
2565 
2566 /* returns true if host is using TSC based clocksource */
2567 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2568 {
2569 	/* checked again under seqlock below */
2570 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2571 		return false;
2572 
2573 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2574 						      tsc_timestamp));
2575 }
2576 
2577 /* returns true if host is using TSC based clocksource */
2578 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2579 					   u64 *tsc_timestamp)
2580 {
2581 	/* checked again under seqlock below */
2582 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2583 		return false;
2584 
2585 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2586 }
2587 #endif
2588 
2589 /*
2590  *
2591  * Assuming a stable TSC across physical CPUS, and a stable TSC
2592  * across virtual CPUs, the following condition is possible.
2593  * Each numbered line represents an event visible to both
2594  * CPUs at the next numbered event.
2595  *
2596  * "timespecX" represents host monotonic time. "tscX" represents
2597  * RDTSC value.
2598  *
2599  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2600  *
2601  * 1.  read timespec0,tsc0
2602  * 2.					| timespec1 = timespec0 + N
2603  * 					| tsc1 = tsc0 + M
2604  * 3. transition to guest		| transition to guest
2605  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2606  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2607  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2608  *
2609  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2610  *
2611  * 	- ret0 < ret1
2612  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2613  *		...
2614  *	- 0 < N - M => M < N
2615  *
2616  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2617  * always the case (the difference between two distinct xtime instances
2618  * might be smaller then the difference between corresponding TSC reads,
2619  * when updating guest vcpus pvclock areas).
2620  *
2621  * To avoid that problem, do not allow visibility of distinct
2622  * system_timestamp/tsc_timestamp values simultaneously: use a master
2623  * copy of host monotonic time values. Update that master copy
2624  * in lockstep.
2625  *
2626  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2627  *
2628  */
2629 
2630 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2631 {
2632 #ifdef CONFIG_X86_64
2633 	struct kvm_arch *ka = &kvm->arch;
2634 	int vclock_mode;
2635 	bool host_tsc_clocksource, vcpus_matched;
2636 
2637 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2638 			atomic_read(&kvm->online_vcpus));
2639 
2640 	/*
2641 	 * If the host uses TSC clock, then passthrough TSC as stable
2642 	 * to the guest.
2643 	 */
2644 	host_tsc_clocksource = kvm_get_time_and_clockread(
2645 					&ka->master_kernel_ns,
2646 					&ka->master_cycle_now);
2647 
2648 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2649 				&& !ka->backwards_tsc_observed
2650 				&& !ka->boot_vcpu_runs_old_kvmclock;
2651 
2652 	if (ka->use_master_clock)
2653 		atomic_set(&kvm_guest_has_master_clock, 1);
2654 
2655 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2656 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2657 					vcpus_matched);
2658 #endif
2659 }
2660 
2661 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2662 {
2663 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2664 }
2665 
2666 static void kvm_gen_update_masterclock(struct kvm *kvm)
2667 {
2668 #ifdef CONFIG_X86_64
2669 	int i;
2670 	struct kvm_vcpu *vcpu;
2671 	struct kvm_arch *ka = &kvm->arch;
2672 	unsigned long flags;
2673 
2674 	kvm_hv_invalidate_tsc_page(kvm);
2675 
2676 	kvm_make_mclock_inprogress_request(kvm);
2677 
2678 	/* no guest entries from this point */
2679 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2680 	pvclock_update_vm_gtod_copy(kvm);
2681 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2682 
2683 	kvm_for_each_vcpu(i, vcpu, kvm)
2684 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2685 
2686 	/* guest entries allowed */
2687 	kvm_for_each_vcpu(i, vcpu, kvm)
2688 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2689 #endif
2690 }
2691 
2692 u64 get_kvmclock_ns(struct kvm *kvm)
2693 {
2694 	struct kvm_arch *ka = &kvm->arch;
2695 	struct pvclock_vcpu_time_info hv_clock;
2696 	unsigned long flags;
2697 	u64 ret;
2698 
2699 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2700 	if (!ka->use_master_clock) {
2701 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2702 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2703 	}
2704 
2705 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2706 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2707 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2708 
2709 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2710 	get_cpu();
2711 
2712 	if (__this_cpu_read(cpu_tsc_khz)) {
2713 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2714 				   &hv_clock.tsc_shift,
2715 				   &hv_clock.tsc_to_system_mul);
2716 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2717 	} else
2718 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2719 
2720 	put_cpu();
2721 
2722 	return ret;
2723 }
2724 
2725 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2726 				   struct gfn_to_hva_cache *cache,
2727 				   unsigned int offset)
2728 {
2729 	struct kvm_vcpu_arch *vcpu = &v->arch;
2730 	struct pvclock_vcpu_time_info guest_hv_clock;
2731 
2732 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2733 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2734 		return;
2735 
2736 	/* This VCPU is paused, but it's legal for a guest to read another
2737 	 * VCPU's kvmclock, so we really have to follow the specification where
2738 	 * it says that version is odd if data is being modified, and even after
2739 	 * it is consistent.
2740 	 *
2741 	 * Version field updates must be kept separate.  This is because
2742 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2743 	 * writes within a string instruction are weakly ordered.  So there
2744 	 * are three writes overall.
2745 	 *
2746 	 * As a small optimization, only write the version field in the first
2747 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2748 	 * version field is the first in the struct.
2749 	 */
2750 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2751 
2752 	if (guest_hv_clock.version & 1)
2753 		++guest_hv_clock.version;  /* first time write, random junk */
2754 
2755 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2756 	kvm_write_guest_offset_cached(v->kvm, cache,
2757 				      &vcpu->hv_clock, offset,
2758 				      sizeof(vcpu->hv_clock.version));
2759 
2760 	smp_wmb();
2761 
2762 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2763 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2764 
2765 	if (vcpu->pvclock_set_guest_stopped_request) {
2766 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2767 		vcpu->pvclock_set_guest_stopped_request = false;
2768 	}
2769 
2770 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2771 
2772 	kvm_write_guest_offset_cached(v->kvm, cache,
2773 				      &vcpu->hv_clock, offset,
2774 				      sizeof(vcpu->hv_clock));
2775 
2776 	smp_wmb();
2777 
2778 	vcpu->hv_clock.version++;
2779 	kvm_write_guest_offset_cached(v->kvm, cache,
2780 				     &vcpu->hv_clock, offset,
2781 				     sizeof(vcpu->hv_clock.version));
2782 }
2783 
2784 static int kvm_guest_time_update(struct kvm_vcpu *v)
2785 {
2786 	unsigned long flags, tgt_tsc_khz;
2787 	struct kvm_vcpu_arch *vcpu = &v->arch;
2788 	struct kvm_arch *ka = &v->kvm->arch;
2789 	s64 kernel_ns;
2790 	u64 tsc_timestamp, host_tsc;
2791 	u8 pvclock_flags;
2792 	bool use_master_clock;
2793 
2794 	kernel_ns = 0;
2795 	host_tsc = 0;
2796 
2797 	/*
2798 	 * If the host uses TSC clock, then passthrough TSC as stable
2799 	 * to the guest.
2800 	 */
2801 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2802 	use_master_clock = ka->use_master_clock;
2803 	if (use_master_clock) {
2804 		host_tsc = ka->master_cycle_now;
2805 		kernel_ns = ka->master_kernel_ns;
2806 	}
2807 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2808 
2809 	/* Keep irq disabled to prevent changes to the clock */
2810 	local_irq_save(flags);
2811 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2812 	if (unlikely(tgt_tsc_khz == 0)) {
2813 		local_irq_restore(flags);
2814 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2815 		return 1;
2816 	}
2817 	if (!use_master_clock) {
2818 		host_tsc = rdtsc();
2819 		kernel_ns = get_kvmclock_base_ns();
2820 	}
2821 
2822 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2823 
2824 	/*
2825 	 * We may have to catch up the TSC to match elapsed wall clock
2826 	 * time for two reasons, even if kvmclock is used.
2827 	 *   1) CPU could have been running below the maximum TSC rate
2828 	 *   2) Broken TSC compensation resets the base at each VCPU
2829 	 *      entry to avoid unknown leaps of TSC even when running
2830 	 *      again on the same CPU.  This may cause apparent elapsed
2831 	 *      time to disappear, and the guest to stand still or run
2832 	 *	very slowly.
2833 	 */
2834 	if (vcpu->tsc_catchup) {
2835 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2836 		if (tsc > tsc_timestamp) {
2837 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2838 			tsc_timestamp = tsc;
2839 		}
2840 	}
2841 
2842 	local_irq_restore(flags);
2843 
2844 	/* With all the info we got, fill in the values */
2845 
2846 	if (kvm_has_tsc_control)
2847 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2848 
2849 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2850 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2851 				   &vcpu->hv_clock.tsc_shift,
2852 				   &vcpu->hv_clock.tsc_to_system_mul);
2853 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2854 	}
2855 
2856 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2857 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2858 	vcpu->last_guest_tsc = tsc_timestamp;
2859 
2860 	/* If the host uses TSC clocksource, then it is stable */
2861 	pvclock_flags = 0;
2862 	if (use_master_clock)
2863 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2864 
2865 	vcpu->hv_clock.flags = pvclock_flags;
2866 
2867 	if (vcpu->pv_time_enabled)
2868 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2869 	if (vcpu->xen.vcpu_info_set)
2870 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2871 				       offsetof(struct compat_vcpu_info, time));
2872 	if (vcpu->xen.vcpu_time_info_set)
2873 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2874 	if (v == kvm_get_vcpu(v->kvm, 0))
2875 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2876 	return 0;
2877 }
2878 
2879 /*
2880  * kvmclock updates which are isolated to a given vcpu, such as
2881  * vcpu->cpu migration, should not allow system_timestamp from
2882  * the rest of the vcpus to remain static. Otherwise ntp frequency
2883  * correction applies to one vcpu's system_timestamp but not
2884  * the others.
2885  *
2886  * So in those cases, request a kvmclock update for all vcpus.
2887  * We need to rate-limit these requests though, as they can
2888  * considerably slow guests that have a large number of vcpus.
2889  * The time for a remote vcpu to update its kvmclock is bound
2890  * by the delay we use to rate-limit the updates.
2891  */
2892 
2893 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2894 
2895 static void kvmclock_update_fn(struct work_struct *work)
2896 {
2897 	int i;
2898 	struct delayed_work *dwork = to_delayed_work(work);
2899 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2900 					   kvmclock_update_work);
2901 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2902 	struct kvm_vcpu *vcpu;
2903 
2904 	kvm_for_each_vcpu(i, vcpu, kvm) {
2905 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2906 		kvm_vcpu_kick(vcpu);
2907 	}
2908 }
2909 
2910 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2911 {
2912 	struct kvm *kvm = v->kvm;
2913 
2914 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2915 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2916 					KVMCLOCK_UPDATE_DELAY);
2917 }
2918 
2919 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2920 
2921 static void kvmclock_sync_fn(struct work_struct *work)
2922 {
2923 	struct delayed_work *dwork = to_delayed_work(work);
2924 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2925 					   kvmclock_sync_work);
2926 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2927 
2928 	if (!kvmclock_periodic_sync)
2929 		return;
2930 
2931 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2932 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2933 					KVMCLOCK_SYNC_PERIOD);
2934 }
2935 
2936 /*
2937  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2938  */
2939 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2940 {
2941 	/* McStatusWrEn enabled? */
2942 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2943 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2944 
2945 	return false;
2946 }
2947 
2948 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2949 {
2950 	u64 mcg_cap = vcpu->arch.mcg_cap;
2951 	unsigned bank_num = mcg_cap & 0xff;
2952 	u32 msr = msr_info->index;
2953 	u64 data = msr_info->data;
2954 
2955 	switch (msr) {
2956 	case MSR_IA32_MCG_STATUS:
2957 		vcpu->arch.mcg_status = data;
2958 		break;
2959 	case MSR_IA32_MCG_CTL:
2960 		if (!(mcg_cap & MCG_CTL_P) &&
2961 		    (data || !msr_info->host_initiated))
2962 			return 1;
2963 		if (data != 0 && data != ~(u64)0)
2964 			return 1;
2965 		vcpu->arch.mcg_ctl = data;
2966 		break;
2967 	default:
2968 		if (msr >= MSR_IA32_MC0_CTL &&
2969 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2970 			u32 offset = array_index_nospec(
2971 				msr - MSR_IA32_MC0_CTL,
2972 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2973 
2974 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2975 			 * some Linux kernels though clear bit 10 in bank 4 to
2976 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2977 			 * this to avoid an uncatched #GP in the guest
2978 			 */
2979 			if ((offset & 0x3) == 0 &&
2980 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2981 				return -1;
2982 
2983 			/* MCi_STATUS */
2984 			if (!msr_info->host_initiated &&
2985 			    (offset & 0x3) == 1 && data != 0) {
2986 				if (!can_set_mci_status(vcpu))
2987 					return -1;
2988 			}
2989 
2990 			vcpu->arch.mce_banks[offset] = data;
2991 			break;
2992 		}
2993 		return 1;
2994 	}
2995 	return 0;
2996 }
2997 
2998 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2999 {
3000 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3001 
3002 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3003 }
3004 
3005 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3006 {
3007 	gpa_t gpa = data & ~0x3f;
3008 
3009 	/* Bits 4:5 are reserved, Should be zero */
3010 	if (data & 0x30)
3011 		return 1;
3012 
3013 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3014 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3015 		return 1;
3016 
3017 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3018 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3019 		return 1;
3020 
3021 	if (!lapic_in_kernel(vcpu))
3022 		return data ? 1 : 0;
3023 
3024 	vcpu->arch.apf.msr_en_val = data;
3025 
3026 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3027 		kvm_clear_async_pf_completion_queue(vcpu);
3028 		kvm_async_pf_hash_reset(vcpu);
3029 		return 0;
3030 	}
3031 
3032 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3033 					sizeof(u64)))
3034 		return 1;
3035 
3036 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3037 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3038 
3039 	kvm_async_pf_wakeup_all(vcpu);
3040 
3041 	return 0;
3042 }
3043 
3044 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3045 {
3046 	/* Bits 8-63 are reserved */
3047 	if (data >> 8)
3048 		return 1;
3049 
3050 	if (!lapic_in_kernel(vcpu))
3051 		return 1;
3052 
3053 	vcpu->arch.apf.msr_int_val = data;
3054 
3055 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3056 
3057 	return 0;
3058 }
3059 
3060 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3061 {
3062 	vcpu->arch.pv_time_enabled = false;
3063 	vcpu->arch.time = 0;
3064 }
3065 
3066 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3067 {
3068 	++vcpu->stat.tlb_flush;
3069 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3070 }
3071 
3072 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3073 {
3074 	++vcpu->stat.tlb_flush;
3075 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3076 }
3077 
3078 static void record_steal_time(struct kvm_vcpu *vcpu)
3079 {
3080 	struct kvm_host_map map;
3081 	struct kvm_steal_time *st;
3082 
3083 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3084 		kvm_xen_runstate_set_running(vcpu);
3085 		return;
3086 	}
3087 
3088 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3089 		return;
3090 
3091 	/* -EAGAIN is returned in atomic context so we can just return. */
3092 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3093 			&map, &vcpu->arch.st.cache, false))
3094 		return;
3095 
3096 	st = map.hva +
3097 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3098 
3099 	/*
3100 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3101 	 * expensive IPIs.
3102 	 */
3103 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3104 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3105 				       st->preempted & KVM_VCPU_FLUSH_TLB);
3106 		if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
3107 			kvm_vcpu_flush_tlb_guest(vcpu);
3108 	}
3109 
3110 	vcpu->arch.st.preempted = 0;
3111 
3112 	if (st->version & 1)
3113 		st->version += 1;  /* first time write, random junk */
3114 
3115 	st->version += 1;
3116 
3117 	smp_wmb();
3118 
3119 	st->steal += current->sched_info.run_delay -
3120 		vcpu->arch.st.last_steal;
3121 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3122 
3123 	smp_wmb();
3124 
3125 	st->version += 1;
3126 
3127 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3128 }
3129 
3130 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3131 {
3132 	bool pr = false;
3133 	u32 msr = msr_info->index;
3134 	u64 data = msr_info->data;
3135 
3136 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3137 		return kvm_xen_write_hypercall_page(vcpu, data);
3138 
3139 	switch (msr) {
3140 	case MSR_AMD64_NB_CFG:
3141 	case MSR_IA32_UCODE_WRITE:
3142 	case MSR_VM_HSAVE_PA:
3143 	case MSR_AMD64_PATCH_LOADER:
3144 	case MSR_AMD64_BU_CFG2:
3145 	case MSR_AMD64_DC_CFG:
3146 	case MSR_F15H_EX_CFG:
3147 		break;
3148 
3149 	case MSR_IA32_UCODE_REV:
3150 		if (msr_info->host_initiated)
3151 			vcpu->arch.microcode_version = data;
3152 		break;
3153 	case MSR_IA32_ARCH_CAPABILITIES:
3154 		if (!msr_info->host_initiated)
3155 			return 1;
3156 		vcpu->arch.arch_capabilities = data;
3157 		break;
3158 	case MSR_IA32_PERF_CAPABILITIES: {
3159 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3160 
3161 		if (!msr_info->host_initiated)
3162 			return 1;
3163 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3164 			return 1;
3165 		if (data & ~msr_ent.data)
3166 			return 1;
3167 
3168 		vcpu->arch.perf_capabilities = data;
3169 
3170 		return 0;
3171 		}
3172 	case MSR_EFER:
3173 		return set_efer(vcpu, msr_info);
3174 	case MSR_K7_HWCR:
3175 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3176 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3177 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3178 
3179 		/* Handle McStatusWrEn */
3180 		if (data == BIT_ULL(18)) {
3181 			vcpu->arch.msr_hwcr = data;
3182 		} else if (data != 0) {
3183 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3184 				    data);
3185 			return 1;
3186 		}
3187 		break;
3188 	case MSR_FAM10H_MMIO_CONF_BASE:
3189 		if (data != 0) {
3190 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3191 				    "0x%llx\n", data);
3192 			return 1;
3193 		}
3194 		break;
3195 	case 0x200 ... 0x2ff:
3196 		return kvm_mtrr_set_msr(vcpu, msr, data);
3197 	case MSR_IA32_APICBASE:
3198 		return kvm_set_apic_base(vcpu, msr_info);
3199 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3200 		return kvm_x2apic_msr_write(vcpu, msr, data);
3201 	case MSR_IA32_TSC_DEADLINE:
3202 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3203 		break;
3204 	case MSR_IA32_TSC_ADJUST:
3205 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3206 			if (!msr_info->host_initiated) {
3207 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3208 				adjust_tsc_offset_guest(vcpu, adj);
3209 			}
3210 			vcpu->arch.ia32_tsc_adjust_msr = data;
3211 		}
3212 		break;
3213 	case MSR_IA32_MISC_ENABLE:
3214 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3215 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3216 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3217 				return 1;
3218 			vcpu->arch.ia32_misc_enable_msr = data;
3219 			kvm_update_cpuid_runtime(vcpu);
3220 		} else {
3221 			vcpu->arch.ia32_misc_enable_msr = data;
3222 		}
3223 		break;
3224 	case MSR_IA32_SMBASE:
3225 		if (!msr_info->host_initiated)
3226 			return 1;
3227 		vcpu->arch.smbase = data;
3228 		break;
3229 	case MSR_IA32_POWER_CTL:
3230 		vcpu->arch.msr_ia32_power_ctl = data;
3231 		break;
3232 	case MSR_IA32_TSC:
3233 		if (msr_info->host_initiated) {
3234 			kvm_synchronize_tsc(vcpu, data);
3235 		} else {
3236 			u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3237 			adjust_tsc_offset_guest(vcpu, adj);
3238 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3239 		}
3240 		break;
3241 	case MSR_IA32_XSS:
3242 		if (!msr_info->host_initiated &&
3243 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3244 			return 1;
3245 		/*
3246 		 * KVM supports exposing PT to the guest, but does not support
3247 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3248 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3249 		 */
3250 		if (data & ~supported_xss)
3251 			return 1;
3252 		vcpu->arch.ia32_xss = data;
3253 		break;
3254 	case MSR_SMI_COUNT:
3255 		if (!msr_info->host_initiated)
3256 			return 1;
3257 		vcpu->arch.smi_count = data;
3258 		break;
3259 	case MSR_KVM_WALL_CLOCK_NEW:
3260 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3261 			return 1;
3262 
3263 		vcpu->kvm->arch.wall_clock = data;
3264 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3265 		break;
3266 	case MSR_KVM_WALL_CLOCK:
3267 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3268 			return 1;
3269 
3270 		vcpu->kvm->arch.wall_clock = data;
3271 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3272 		break;
3273 	case MSR_KVM_SYSTEM_TIME_NEW:
3274 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3275 			return 1;
3276 
3277 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3278 		break;
3279 	case MSR_KVM_SYSTEM_TIME:
3280 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3281 			return 1;
3282 
3283 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3284 		break;
3285 	case MSR_KVM_ASYNC_PF_EN:
3286 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3287 			return 1;
3288 
3289 		if (kvm_pv_enable_async_pf(vcpu, data))
3290 			return 1;
3291 		break;
3292 	case MSR_KVM_ASYNC_PF_INT:
3293 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3294 			return 1;
3295 
3296 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3297 			return 1;
3298 		break;
3299 	case MSR_KVM_ASYNC_PF_ACK:
3300 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3301 			return 1;
3302 		if (data & 0x1) {
3303 			vcpu->arch.apf.pageready_pending = false;
3304 			kvm_check_async_pf_completion(vcpu);
3305 		}
3306 		break;
3307 	case MSR_KVM_STEAL_TIME:
3308 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3309 			return 1;
3310 
3311 		if (unlikely(!sched_info_on()))
3312 			return 1;
3313 
3314 		if (data & KVM_STEAL_RESERVED_MASK)
3315 			return 1;
3316 
3317 		vcpu->arch.st.msr_val = data;
3318 
3319 		if (!(data & KVM_MSR_ENABLED))
3320 			break;
3321 
3322 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3323 
3324 		break;
3325 	case MSR_KVM_PV_EOI_EN:
3326 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3327 			return 1;
3328 
3329 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3330 			return 1;
3331 		break;
3332 
3333 	case MSR_KVM_POLL_CONTROL:
3334 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3335 			return 1;
3336 
3337 		/* only enable bit supported */
3338 		if (data & (-1ULL << 1))
3339 			return 1;
3340 
3341 		vcpu->arch.msr_kvm_poll_control = data;
3342 		break;
3343 
3344 	case MSR_IA32_MCG_CTL:
3345 	case MSR_IA32_MCG_STATUS:
3346 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3347 		return set_msr_mce(vcpu, msr_info);
3348 
3349 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3350 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3351 		pr = true;
3352 		fallthrough;
3353 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3354 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3355 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3356 			return kvm_pmu_set_msr(vcpu, msr_info);
3357 
3358 		if (pr || data != 0)
3359 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3360 				    "0x%x data 0x%llx\n", msr, data);
3361 		break;
3362 	case MSR_K7_CLK_CTL:
3363 		/*
3364 		 * Ignore all writes to this no longer documented MSR.
3365 		 * Writes are only relevant for old K7 processors,
3366 		 * all pre-dating SVM, but a recommended workaround from
3367 		 * AMD for these chips. It is possible to specify the
3368 		 * affected processor models on the command line, hence
3369 		 * the need to ignore the workaround.
3370 		 */
3371 		break;
3372 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3373 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3374 	case HV_X64_MSR_SYNDBG_OPTIONS:
3375 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3376 	case HV_X64_MSR_CRASH_CTL:
3377 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3378 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3379 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3380 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3381 		return kvm_hv_set_msr_common(vcpu, msr, data,
3382 					     msr_info->host_initiated);
3383 	case MSR_IA32_BBL_CR_CTL3:
3384 		/* Drop writes to this legacy MSR -- see rdmsr
3385 		 * counterpart for further detail.
3386 		 */
3387 		if (report_ignored_msrs)
3388 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3389 				msr, data);
3390 		break;
3391 	case MSR_AMD64_OSVW_ID_LENGTH:
3392 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3393 			return 1;
3394 		vcpu->arch.osvw.length = data;
3395 		break;
3396 	case MSR_AMD64_OSVW_STATUS:
3397 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3398 			return 1;
3399 		vcpu->arch.osvw.status = data;
3400 		break;
3401 	case MSR_PLATFORM_INFO:
3402 		if (!msr_info->host_initiated ||
3403 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3404 		     cpuid_fault_enabled(vcpu)))
3405 			return 1;
3406 		vcpu->arch.msr_platform_info = data;
3407 		break;
3408 	case MSR_MISC_FEATURES_ENABLES:
3409 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3410 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3411 		     !supports_cpuid_fault(vcpu)))
3412 			return 1;
3413 		vcpu->arch.msr_misc_features_enables = data;
3414 		break;
3415 	default:
3416 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3417 			return kvm_pmu_set_msr(vcpu, msr_info);
3418 		return KVM_MSR_RET_INVALID;
3419 	}
3420 	return 0;
3421 }
3422 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3423 
3424 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3425 {
3426 	u64 data;
3427 	u64 mcg_cap = vcpu->arch.mcg_cap;
3428 	unsigned bank_num = mcg_cap & 0xff;
3429 
3430 	switch (msr) {
3431 	case MSR_IA32_P5_MC_ADDR:
3432 	case MSR_IA32_P5_MC_TYPE:
3433 		data = 0;
3434 		break;
3435 	case MSR_IA32_MCG_CAP:
3436 		data = vcpu->arch.mcg_cap;
3437 		break;
3438 	case MSR_IA32_MCG_CTL:
3439 		if (!(mcg_cap & MCG_CTL_P) && !host)
3440 			return 1;
3441 		data = vcpu->arch.mcg_ctl;
3442 		break;
3443 	case MSR_IA32_MCG_STATUS:
3444 		data = vcpu->arch.mcg_status;
3445 		break;
3446 	default:
3447 		if (msr >= MSR_IA32_MC0_CTL &&
3448 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3449 			u32 offset = array_index_nospec(
3450 				msr - MSR_IA32_MC0_CTL,
3451 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3452 
3453 			data = vcpu->arch.mce_banks[offset];
3454 			break;
3455 		}
3456 		return 1;
3457 	}
3458 	*pdata = data;
3459 	return 0;
3460 }
3461 
3462 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3463 {
3464 	switch (msr_info->index) {
3465 	case MSR_IA32_PLATFORM_ID:
3466 	case MSR_IA32_EBL_CR_POWERON:
3467 	case MSR_IA32_LASTBRANCHFROMIP:
3468 	case MSR_IA32_LASTBRANCHTOIP:
3469 	case MSR_IA32_LASTINTFROMIP:
3470 	case MSR_IA32_LASTINTTOIP:
3471 	case MSR_AMD64_SYSCFG:
3472 	case MSR_K8_TSEG_ADDR:
3473 	case MSR_K8_TSEG_MASK:
3474 	case MSR_VM_HSAVE_PA:
3475 	case MSR_K8_INT_PENDING_MSG:
3476 	case MSR_AMD64_NB_CFG:
3477 	case MSR_FAM10H_MMIO_CONF_BASE:
3478 	case MSR_AMD64_BU_CFG2:
3479 	case MSR_IA32_PERF_CTL:
3480 	case MSR_AMD64_DC_CFG:
3481 	case MSR_F15H_EX_CFG:
3482 	/*
3483 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3484 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3485 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3486 	 * so for existing CPU-specific MSRs.
3487 	 */
3488 	case MSR_RAPL_POWER_UNIT:
3489 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3490 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3491 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3492 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3493 		msr_info->data = 0;
3494 		break;
3495 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3496 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3497 			return kvm_pmu_get_msr(vcpu, msr_info);
3498 		if (!msr_info->host_initiated)
3499 			return 1;
3500 		msr_info->data = 0;
3501 		break;
3502 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3503 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3504 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3505 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3506 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3507 			return kvm_pmu_get_msr(vcpu, msr_info);
3508 		msr_info->data = 0;
3509 		break;
3510 	case MSR_IA32_UCODE_REV:
3511 		msr_info->data = vcpu->arch.microcode_version;
3512 		break;
3513 	case MSR_IA32_ARCH_CAPABILITIES:
3514 		if (!msr_info->host_initiated &&
3515 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3516 			return 1;
3517 		msr_info->data = vcpu->arch.arch_capabilities;
3518 		break;
3519 	case MSR_IA32_PERF_CAPABILITIES:
3520 		if (!msr_info->host_initiated &&
3521 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3522 			return 1;
3523 		msr_info->data = vcpu->arch.perf_capabilities;
3524 		break;
3525 	case MSR_IA32_POWER_CTL:
3526 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3527 		break;
3528 	case MSR_IA32_TSC: {
3529 		/*
3530 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3531 		 * even when not intercepted. AMD manual doesn't explicitly
3532 		 * state this but appears to behave the same.
3533 		 *
3534 		 * On userspace reads and writes, however, we unconditionally
3535 		 * return L1's TSC value to ensure backwards-compatible
3536 		 * behavior for migration.
3537 		 */
3538 		u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3539 							    vcpu->arch.tsc_offset;
3540 
3541 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3542 		break;
3543 	}
3544 	case MSR_MTRRcap:
3545 	case 0x200 ... 0x2ff:
3546 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3547 	case 0xcd: /* fsb frequency */
3548 		msr_info->data = 3;
3549 		break;
3550 		/*
3551 		 * MSR_EBC_FREQUENCY_ID
3552 		 * Conservative value valid for even the basic CPU models.
3553 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3554 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3555 		 * and 266MHz for model 3, or 4. Set Core Clock
3556 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3557 		 * 31:24) even though these are only valid for CPU
3558 		 * models > 2, however guests may end up dividing or
3559 		 * multiplying by zero otherwise.
3560 		 */
3561 	case MSR_EBC_FREQUENCY_ID:
3562 		msr_info->data = 1 << 24;
3563 		break;
3564 	case MSR_IA32_APICBASE:
3565 		msr_info->data = kvm_get_apic_base(vcpu);
3566 		break;
3567 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3568 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3569 	case MSR_IA32_TSC_DEADLINE:
3570 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3571 		break;
3572 	case MSR_IA32_TSC_ADJUST:
3573 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3574 		break;
3575 	case MSR_IA32_MISC_ENABLE:
3576 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3577 		break;
3578 	case MSR_IA32_SMBASE:
3579 		if (!msr_info->host_initiated)
3580 			return 1;
3581 		msr_info->data = vcpu->arch.smbase;
3582 		break;
3583 	case MSR_SMI_COUNT:
3584 		msr_info->data = vcpu->arch.smi_count;
3585 		break;
3586 	case MSR_IA32_PERF_STATUS:
3587 		/* TSC increment by tick */
3588 		msr_info->data = 1000ULL;
3589 		/* CPU multiplier */
3590 		msr_info->data |= (((uint64_t)4ULL) << 40);
3591 		break;
3592 	case MSR_EFER:
3593 		msr_info->data = vcpu->arch.efer;
3594 		break;
3595 	case MSR_KVM_WALL_CLOCK:
3596 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3597 			return 1;
3598 
3599 		msr_info->data = vcpu->kvm->arch.wall_clock;
3600 		break;
3601 	case MSR_KVM_WALL_CLOCK_NEW:
3602 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3603 			return 1;
3604 
3605 		msr_info->data = vcpu->kvm->arch.wall_clock;
3606 		break;
3607 	case MSR_KVM_SYSTEM_TIME:
3608 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3609 			return 1;
3610 
3611 		msr_info->data = vcpu->arch.time;
3612 		break;
3613 	case MSR_KVM_SYSTEM_TIME_NEW:
3614 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3615 			return 1;
3616 
3617 		msr_info->data = vcpu->arch.time;
3618 		break;
3619 	case MSR_KVM_ASYNC_PF_EN:
3620 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3621 			return 1;
3622 
3623 		msr_info->data = vcpu->arch.apf.msr_en_val;
3624 		break;
3625 	case MSR_KVM_ASYNC_PF_INT:
3626 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3627 			return 1;
3628 
3629 		msr_info->data = vcpu->arch.apf.msr_int_val;
3630 		break;
3631 	case MSR_KVM_ASYNC_PF_ACK:
3632 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3633 			return 1;
3634 
3635 		msr_info->data = 0;
3636 		break;
3637 	case MSR_KVM_STEAL_TIME:
3638 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3639 			return 1;
3640 
3641 		msr_info->data = vcpu->arch.st.msr_val;
3642 		break;
3643 	case MSR_KVM_PV_EOI_EN:
3644 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3645 			return 1;
3646 
3647 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3648 		break;
3649 	case MSR_KVM_POLL_CONTROL:
3650 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3651 			return 1;
3652 
3653 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3654 		break;
3655 	case MSR_IA32_P5_MC_ADDR:
3656 	case MSR_IA32_P5_MC_TYPE:
3657 	case MSR_IA32_MCG_CAP:
3658 	case MSR_IA32_MCG_CTL:
3659 	case MSR_IA32_MCG_STATUS:
3660 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3661 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3662 				   msr_info->host_initiated);
3663 	case MSR_IA32_XSS:
3664 		if (!msr_info->host_initiated &&
3665 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3666 			return 1;
3667 		msr_info->data = vcpu->arch.ia32_xss;
3668 		break;
3669 	case MSR_K7_CLK_CTL:
3670 		/*
3671 		 * Provide expected ramp-up count for K7. All other
3672 		 * are set to zero, indicating minimum divisors for
3673 		 * every field.
3674 		 *
3675 		 * This prevents guest kernels on AMD host with CPU
3676 		 * type 6, model 8 and higher from exploding due to
3677 		 * the rdmsr failing.
3678 		 */
3679 		msr_info->data = 0x20000000;
3680 		break;
3681 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3682 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3683 	case HV_X64_MSR_SYNDBG_OPTIONS:
3684 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3685 	case HV_X64_MSR_CRASH_CTL:
3686 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3687 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3688 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3689 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3690 		return kvm_hv_get_msr_common(vcpu,
3691 					     msr_info->index, &msr_info->data,
3692 					     msr_info->host_initiated);
3693 	case MSR_IA32_BBL_CR_CTL3:
3694 		/* This legacy MSR exists but isn't fully documented in current
3695 		 * silicon.  It is however accessed by winxp in very narrow
3696 		 * scenarios where it sets bit #19, itself documented as
3697 		 * a "reserved" bit.  Best effort attempt to source coherent
3698 		 * read data here should the balance of the register be
3699 		 * interpreted by the guest:
3700 		 *
3701 		 * L2 cache control register 3: 64GB range, 256KB size,
3702 		 * enabled, latency 0x1, configured
3703 		 */
3704 		msr_info->data = 0xbe702111;
3705 		break;
3706 	case MSR_AMD64_OSVW_ID_LENGTH:
3707 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3708 			return 1;
3709 		msr_info->data = vcpu->arch.osvw.length;
3710 		break;
3711 	case MSR_AMD64_OSVW_STATUS:
3712 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3713 			return 1;
3714 		msr_info->data = vcpu->arch.osvw.status;
3715 		break;
3716 	case MSR_PLATFORM_INFO:
3717 		if (!msr_info->host_initiated &&
3718 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3719 			return 1;
3720 		msr_info->data = vcpu->arch.msr_platform_info;
3721 		break;
3722 	case MSR_MISC_FEATURES_ENABLES:
3723 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3724 		break;
3725 	case MSR_K7_HWCR:
3726 		msr_info->data = vcpu->arch.msr_hwcr;
3727 		break;
3728 	default:
3729 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3730 			return kvm_pmu_get_msr(vcpu, msr_info);
3731 		return KVM_MSR_RET_INVALID;
3732 	}
3733 	return 0;
3734 }
3735 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3736 
3737 /*
3738  * Read or write a bunch of msrs. All parameters are kernel addresses.
3739  *
3740  * @return number of msrs set successfully.
3741  */
3742 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3743 		    struct kvm_msr_entry *entries,
3744 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3745 				  unsigned index, u64 *data))
3746 {
3747 	int i;
3748 
3749 	for (i = 0; i < msrs->nmsrs; ++i)
3750 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3751 			break;
3752 
3753 	return i;
3754 }
3755 
3756 /*
3757  * Read or write a bunch of msrs. Parameters are user addresses.
3758  *
3759  * @return number of msrs set successfully.
3760  */
3761 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3762 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3763 				unsigned index, u64 *data),
3764 		  int writeback)
3765 {
3766 	struct kvm_msrs msrs;
3767 	struct kvm_msr_entry *entries;
3768 	int r, n;
3769 	unsigned size;
3770 
3771 	r = -EFAULT;
3772 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3773 		goto out;
3774 
3775 	r = -E2BIG;
3776 	if (msrs.nmsrs >= MAX_IO_MSRS)
3777 		goto out;
3778 
3779 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3780 	entries = memdup_user(user_msrs->entries, size);
3781 	if (IS_ERR(entries)) {
3782 		r = PTR_ERR(entries);
3783 		goto out;
3784 	}
3785 
3786 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3787 	if (r < 0)
3788 		goto out_free;
3789 
3790 	r = -EFAULT;
3791 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3792 		goto out_free;
3793 
3794 	r = n;
3795 
3796 out_free:
3797 	kfree(entries);
3798 out:
3799 	return r;
3800 }
3801 
3802 static inline bool kvm_can_mwait_in_guest(void)
3803 {
3804 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3805 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3806 		boot_cpu_has(X86_FEATURE_ARAT);
3807 }
3808 
3809 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3810 					    struct kvm_cpuid2 __user *cpuid_arg)
3811 {
3812 	struct kvm_cpuid2 cpuid;
3813 	int r;
3814 
3815 	r = -EFAULT;
3816 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3817 		return r;
3818 
3819 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3820 	if (r)
3821 		return r;
3822 
3823 	r = -EFAULT;
3824 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3825 		return r;
3826 
3827 	return 0;
3828 }
3829 
3830 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3831 {
3832 	int r = 0;
3833 
3834 	switch (ext) {
3835 	case KVM_CAP_IRQCHIP:
3836 	case KVM_CAP_HLT:
3837 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3838 	case KVM_CAP_SET_TSS_ADDR:
3839 	case KVM_CAP_EXT_CPUID:
3840 	case KVM_CAP_EXT_EMUL_CPUID:
3841 	case KVM_CAP_CLOCKSOURCE:
3842 	case KVM_CAP_PIT:
3843 	case KVM_CAP_NOP_IO_DELAY:
3844 	case KVM_CAP_MP_STATE:
3845 	case KVM_CAP_SYNC_MMU:
3846 	case KVM_CAP_USER_NMI:
3847 	case KVM_CAP_REINJECT_CONTROL:
3848 	case KVM_CAP_IRQ_INJECT_STATUS:
3849 	case KVM_CAP_IOEVENTFD:
3850 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3851 	case KVM_CAP_PIT2:
3852 	case KVM_CAP_PIT_STATE2:
3853 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3854 	case KVM_CAP_VCPU_EVENTS:
3855 	case KVM_CAP_HYPERV:
3856 	case KVM_CAP_HYPERV_VAPIC:
3857 	case KVM_CAP_HYPERV_SPIN:
3858 	case KVM_CAP_HYPERV_SYNIC:
3859 	case KVM_CAP_HYPERV_SYNIC2:
3860 	case KVM_CAP_HYPERV_VP_INDEX:
3861 	case KVM_CAP_HYPERV_EVENTFD:
3862 	case KVM_CAP_HYPERV_TLBFLUSH:
3863 	case KVM_CAP_HYPERV_SEND_IPI:
3864 	case KVM_CAP_HYPERV_CPUID:
3865 	case KVM_CAP_SYS_HYPERV_CPUID:
3866 	case KVM_CAP_PCI_SEGMENT:
3867 	case KVM_CAP_DEBUGREGS:
3868 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3869 	case KVM_CAP_XSAVE:
3870 	case KVM_CAP_ASYNC_PF:
3871 	case KVM_CAP_ASYNC_PF_INT:
3872 	case KVM_CAP_GET_TSC_KHZ:
3873 	case KVM_CAP_KVMCLOCK_CTRL:
3874 	case KVM_CAP_READONLY_MEM:
3875 	case KVM_CAP_HYPERV_TIME:
3876 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3877 	case KVM_CAP_TSC_DEADLINE_TIMER:
3878 	case KVM_CAP_DISABLE_QUIRKS:
3879 	case KVM_CAP_SET_BOOT_CPU_ID:
3880  	case KVM_CAP_SPLIT_IRQCHIP:
3881 	case KVM_CAP_IMMEDIATE_EXIT:
3882 	case KVM_CAP_PMU_EVENT_FILTER:
3883 	case KVM_CAP_GET_MSR_FEATURES:
3884 	case KVM_CAP_MSR_PLATFORM_INFO:
3885 	case KVM_CAP_EXCEPTION_PAYLOAD:
3886 	case KVM_CAP_SET_GUEST_DEBUG:
3887 	case KVM_CAP_LAST_CPU:
3888 	case KVM_CAP_X86_USER_SPACE_MSR:
3889 	case KVM_CAP_X86_MSR_FILTER:
3890 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3891 #ifdef CONFIG_X86_SGX_KVM
3892 	case KVM_CAP_SGX_ATTRIBUTE:
3893 #endif
3894 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3895 		r = 1;
3896 		break;
3897 	case KVM_CAP_SET_GUEST_DEBUG2:
3898 		return KVM_GUESTDBG_VALID_MASK;
3899 #ifdef CONFIG_KVM_XEN
3900 	case KVM_CAP_XEN_HVM:
3901 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3902 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3903 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
3904 		if (sched_info_on())
3905 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3906 		break;
3907 #endif
3908 	case KVM_CAP_SYNC_REGS:
3909 		r = KVM_SYNC_X86_VALID_FIELDS;
3910 		break;
3911 	case KVM_CAP_ADJUST_CLOCK:
3912 		r = KVM_CLOCK_TSC_STABLE;
3913 		break;
3914 	case KVM_CAP_X86_DISABLE_EXITS:
3915 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3916 		      KVM_X86_DISABLE_EXITS_CSTATE;
3917 		if(kvm_can_mwait_in_guest())
3918 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3919 		break;
3920 	case KVM_CAP_X86_SMM:
3921 		/* SMBASE is usually relocated above 1M on modern chipsets,
3922 		 * and SMM handlers might indeed rely on 4G segment limits,
3923 		 * so do not report SMM to be available if real mode is
3924 		 * emulated via vm86 mode.  Still, do not go to great lengths
3925 		 * to avoid userspace's usage of the feature, because it is a
3926 		 * fringe case that is not enabled except via specific settings
3927 		 * of the module parameters.
3928 		 */
3929 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3930 		break;
3931 	case KVM_CAP_VAPIC:
3932 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3933 		break;
3934 	case KVM_CAP_NR_VCPUS:
3935 		r = KVM_SOFT_MAX_VCPUS;
3936 		break;
3937 	case KVM_CAP_MAX_VCPUS:
3938 		r = KVM_MAX_VCPUS;
3939 		break;
3940 	case KVM_CAP_MAX_VCPU_ID:
3941 		r = KVM_MAX_VCPU_ID;
3942 		break;
3943 	case KVM_CAP_PV_MMU:	/* obsolete */
3944 		r = 0;
3945 		break;
3946 	case KVM_CAP_MCE:
3947 		r = KVM_MAX_MCE_BANKS;
3948 		break;
3949 	case KVM_CAP_XCRS:
3950 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3951 		break;
3952 	case KVM_CAP_TSC_CONTROL:
3953 		r = kvm_has_tsc_control;
3954 		break;
3955 	case KVM_CAP_X2APIC_API:
3956 		r = KVM_X2APIC_API_VALID_FLAGS;
3957 		break;
3958 	case KVM_CAP_NESTED_STATE:
3959 		r = kvm_x86_ops.nested_ops->get_state ?
3960 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3961 		break;
3962 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3963 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3964 		break;
3965 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3966 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3967 		break;
3968 	case KVM_CAP_SMALLER_MAXPHYADDR:
3969 		r = (int) allow_smaller_maxphyaddr;
3970 		break;
3971 	case KVM_CAP_STEAL_TIME:
3972 		r = sched_info_on();
3973 		break;
3974 	case KVM_CAP_X86_BUS_LOCK_EXIT:
3975 		if (kvm_has_bus_lock_exit)
3976 			r = KVM_BUS_LOCK_DETECTION_OFF |
3977 			    KVM_BUS_LOCK_DETECTION_EXIT;
3978 		else
3979 			r = 0;
3980 		break;
3981 	default:
3982 		break;
3983 	}
3984 	return r;
3985 
3986 }
3987 
3988 long kvm_arch_dev_ioctl(struct file *filp,
3989 			unsigned int ioctl, unsigned long arg)
3990 {
3991 	void __user *argp = (void __user *)arg;
3992 	long r;
3993 
3994 	switch (ioctl) {
3995 	case KVM_GET_MSR_INDEX_LIST: {
3996 		struct kvm_msr_list __user *user_msr_list = argp;
3997 		struct kvm_msr_list msr_list;
3998 		unsigned n;
3999 
4000 		r = -EFAULT;
4001 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4002 			goto out;
4003 		n = msr_list.nmsrs;
4004 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4005 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4006 			goto out;
4007 		r = -E2BIG;
4008 		if (n < msr_list.nmsrs)
4009 			goto out;
4010 		r = -EFAULT;
4011 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4012 				 num_msrs_to_save * sizeof(u32)))
4013 			goto out;
4014 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4015 				 &emulated_msrs,
4016 				 num_emulated_msrs * sizeof(u32)))
4017 			goto out;
4018 		r = 0;
4019 		break;
4020 	}
4021 	case KVM_GET_SUPPORTED_CPUID:
4022 	case KVM_GET_EMULATED_CPUID: {
4023 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4024 		struct kvm_cpuid2 cpuid;
4025 
4026 		r = -EFAULT;
4027 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4028 			goto out;
4029 
4030 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4031 					    ioctl);
4032 		if (r)
4033 			goto out;
4034 
4035 		r = -EFAULT;
4036 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4037 			goto out;
4038 		r = 0;
4039 		break;
4040 	}
4041 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4042 		r = -EFAULT;
4043 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4044 				 sizeof(kvm_mce_cap_supported)))
4045 			goto out;
4046 		r = 0;
4047 		break;
4048 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4049 		struct kvm_msr_list __user *user_msr_list = argp;
4050 		struct kvm_msr_list msr_list;
4051 		unsigned int n;
4052 
4053 		r = -EFAULT;
4054 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4055 			goto out;
4056 		n = msr_list.nmsrs;
4057 		msr_list.nmsrs = num_msr_based_features;
4058 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4059 			goto out;
4060 		r = -E2BIG;
4061 		if (n < msr_list.nmsrs)
4062 			goto out;
4063 		r = -EFAULT;
4064 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4065 				 num_msr_based_features * sizeof(u32)))
4066 			goto out;
4067 		r = 0;
4068 		break;
4069 	}
4070 	case KVM_GET_MSRS:
4071 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4072 		break;
4073 	case KVM_GET_SUPPORTED_HV_CPUID:
4074 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4075 		break;
4076 	default:
4077 		r = -EINVAL;
4078 		break;
4079 	}
4080 out:
4081 	return r;
4082 }
4083 
4084 static void wbinvd_ipi(void *garbage)
4085 {
4086 	wbinvd();
4087 }
4088 
4089 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4090 {
4091 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4092 }
4093 
4094 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4095 {
4096 	/* Address WBINVD may be executed by guest */
4097 	if (need_emulate_wbinvd(vcpu)) {
4098 		if (static_call(kvm_x86_has_wbinvd_exit)())
4099 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4100 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4101 			smp_call_function_single(vcpu->cpu,
4102 					wbinvd_ipi, NULL, 1);
4103 	}
4104 
4105 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4106 
4107 	/* Save host pkru register if supported */
4108 	vcpu->arch.host_pkru = read_pkru();
4109 
4110 	/* Apply any externally detected TSC adjustments (due to suspend) */
4111 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4112 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4113 		vcpu->arch.tsc_offset_adjustment = 0;
4114 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4115 	}
4116 
4117 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4118 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4119 				rdtsc() - vcpu->arch.last_host_tsc;
4120 		if (tsc_delta < 0)
4121 			mark_tsc_unstable("KVM discovered backwards TSC");
4122 
4123 		if (kvm_check_tsc_unstable()) {
4124 			u64 offset = kvm_compute_tsc_offset(vcpu,
4125 						vcpu->arch.last_guest_tsc);
4126 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4127 			vcpu->arch.tsc_catchup = 1;
4128 		}
4129 
4130 		if (kvm_lapic_hv_timer_in_use(vcpu))
4131 			kvm_lapic_restart_hv_timer(vcpu);
4132 
4133 		/*
4134 		 * On a host with synchronized TSC, there is no need to update
4135 		 * kvmclock on vcpu->cpu migration
4136 		 */
4137 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4138 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4139 		if (vcpu->cpu != cpu)
4140 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4141 		vcpu->cpu = cpu;
4142 	}
4143 
4144 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4145 }
4146 
4147 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4148 {
4149 	struct kvm_host_map map;
4150 	struct kvm_steal_time *st;
4151 
4152 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4153 		return;
4154 
4155 	if (vcpu->arch.st.preempted)
4156 		return;
4157 
4158 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4159 			&vcpu->arch.st.cache, true))
4160 		return;
4161 
4162 	st = map.hva +
4163 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4164 
4165 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4166 
4167 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4168 }
4169 
4170 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4171 {
4172 	int idx;
4173 
4174 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4175 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4176 
4177 	/*
4178 	 * Take the srcu lock as memslots will be accessed to check the gfn
4179 	 * cache generation against the memslots generation.
4180 	 */
4181 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4182 	if (kvm_xen_msr_enabled(vcpu->kvm))
4183 		kvm_xen_runstate_set_preempted(vcpu);
4184 	else
4185 		kvm_steal_time_set_preempted(vcpu);
4186 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4187 
4188 	static_call(kvm_x86_vcpu_put)(vcpu);
4189 	vcpu->arch.last_host_tsc = rdtsc();
4190 	/*
4191 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4192 	 * on every vmexit, but if not, we might have a stale dr6 from the
4193 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4194 	 */
4195 	set_debugreg(0, 6);
4196 }
4197 
4198 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4199 				    struct kvm_lapic_state *s)
4200 {
4201 	if (vcpu->arch.apicv_active)
4202 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4203 
4204 	return kvm_apic_get_state(vcpu, s);
4205 }
4206 
4207 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4208 				    struct kvm_lapic_state *s)
4209 {
4210 	int r;
4211 
4212 	r = kvm_apic_set_state(vcpu, s);
4213 	if (r)
4214 		return r;
4215 	update_cr8_intercept(vcpu);
4216 
4217 	return 0;
4218 }
4219 
4220 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4221 {
4222 	/*
4223 	 * We can accept userspace's request for interrupt injection
4224 	 * as long as we have a place to store the interrupt number.
4225 	 * The actual injection will happen when the CPU is able to
4226 	 * deliver the interrupt.
4227 	 */
4228 	if (kvm_cpu_has_extint(vcpu))
4229 		return false;
4230 
4231 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4232 	return (!lapic_in_kernel(vcpu) ||
4233 		kvm_apic_accept_pic_intr(vcpu));
4234 }
4235 
4236 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4237 {
4238 	return kvm_arch_interrupt_allowed(vcpu) &&
4239 		kvm_cpu_accept_dm_intr(vcpu);
4240 }
4241 
4242 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4243 				    struct kvm_interrupt *irq)
4244 {
4245 	if (irq->irq >= KVM_NR_INTERRUPTS)
4246 		return -EINVAL;
4247 
4248 	if (!irqchip_in_kernel(vcpu->kvm)) {
4249 		kvm_queue_interrupt(vcpu, irq->irq, false);
4250 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4251 		return 0;
4252 	}
4253 
4254 	/*
4255 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4256 	 * fail for in-kernel 8259.
4257 	 */
4258 	if (pic_in_kernel(vcpu->kvm))
4259 		return -ENXIO;
4260 
4261 	if (vcpu->arch.pending_external_vector != -1)
4262 		return -EEXIST;
4263 
4264 	vcpu->arch.pending_external_vector = irq->irq;
4265 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4266 	return 0;
4267 }
4268 
4269 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4270 {
4271 	kvm_inject_nmi(vcpu);
4272 
4273 	return 0;
4274 }
4275 
4276 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4277 {
4278 	kvm_make_request(KVM_REQ_SMI, vcpu);
4279 
4280 	return 0;
4281 }
4282 
4283 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4284 					   struct kvm_tpr_access_ctl *tac)
4285 {
4286 	if (tac->flags)
4287 		return -EINVAL;
4288 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4289 	return 0;
4290 }
4291 
4292 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4293 					u64 mcg_cap)
4294 {
4295 	int r;
4296 	unsigned bank_num = mcg_cap & 0xff, bank;
4297 
4298 	r = -EINVAL;
4299 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4300 		goto out;
4301 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4302 		goto out;
4303 	r = 0;
4304 	vcpu->arch.mcg_cap = mcg_cap;
4305 	/* Init IA32_MCG_CTL to all 1s */
4306 	if (mcg_cap & MCG_CTL_P)
4307 		vcpu->arch.mcg_ctl = ~(u64)0;
4308 	/* Init IA32_MCi_CTL to all 1s */
4309 	for (bank = 0; bank < bank_num; bank++)
4310 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4311 
4312 	static_call(kvm_x86_setup_mce)(vcpu);
4313 out:
4314 	return r;
4315 }
4316 
4317 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4318 				      struct kvm_x86_mce *mce)
4319 {
4320 	u64 mcg_cap = vcpu->arch.mcg_cap;
4321 	unsigned bank_num = mcg_cap & 0xff;
4322 	u64 *banks = vcpu->arch.mce_banks;
4323 
4324 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4325 		return -EINVAL;
4326 	/*
4327 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4328 	 * reporting is disabled
4329 	 */
4330 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4331 	    vcpu->arch.mcg_ctl != ~(u64)0)
4332 		return 0;
4333 	banks += 4 * mce->bank;
4334 	/*
4335 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4336 	 * reporting is disabled for the bank
4337 	 */
4338 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4339 		return 0;
4340 	if (mce->status & MCI_STATUS_UC) {
4341 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4342 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4343 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4344 			return 0;
4345 		}
4346 		if (banks[1] & MCI_STATUS_VAL)
4347 			mce->status |= MCI_STATUS_OVER;
4348 		banks[2] = mce->addr;
4349 		banks[3] = mce->misc;
4350 		vcpu->arch.mcg_status = mce->mcg_status;
4351 		banks[1] = mce->status;
4352 		kvm_queue_exception(vcpu, MC_VECTOR);
4353 	} else if (!(banks[1] & MCI_STATUS_VAL)
4354 		   || !(banks[1] & MCI_STATUS_UC)) {
4355 		if (banks[1] & MCI_STATUS_VAL)
4356 			mce->status |= MCI_STATUS_OVER;
4357 		banks[2] = mce->addr;
4358 		banks[3] = mce->misc;
4359 		banks[1] = mce->status;
4360 	} else
4361 		banks[1] |= MCI_STATUS_OVER;
4362 	return 0;
4363 }
4364 
4365 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4366 					       struct kvm_vcpu_events *events)
4367 {
4368 	process_nmi(vcpu);
4369 
4370 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4371 		process_smi(vcpu);
4372 
4373 	/*
4374 	 * In guest mode, payload delivery should be deferred,
4375 	 * so that the L1 hypervisor can intercept #PF before
4376 	 * CR2 is modified (or intercept #DB before DR6 is
4377 	 * modified under nVMX). Unless the per-VM capability,
4378 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4379 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4380 	 * opportunistically defer the exception payload, deliver it if the
4381 	 * capability hasn't been requested before processing a
4382 	 * KVM_GET_VCPU_EVENTS.
4383 	 */
4384 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4385 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4386 		kvm_deliver_exception_payload(vcpu);
4387 
4388 	/*
4389 	 * The API doesn't provide the instruction length for software
4390 	 * exceptions, so don't report them. As long as the guest RIP
4391 	 * isn't advanced, we should expect to encounter the exception
4392 	 * again.
4393 	 */
4394 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4395 		events->exception.injected = 0;
4396 		events->exception.pending = 0;
4397 	} else {
4398 		events->exception.injected = vcpu->arch.exception.injected;
4399 		events->exception.pending = vcpu->arch.exception.pending;
4400 		/*
4401 		 * For ABI compatibility, deliberately conflate
4402 		 * pending and injected exceptions when
4403 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4404 		 */
4405 		if (!vcpu->kvm->arch.exception_payload_enabled)
4406 			events->exception.injected |=
4407 				vcpu->arch.exception.pending;
4408 	}
4409 	events->exception.nr = vcpu->arch.exception.nr;
4410 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4411 	events->exception.error_code = vcpu->arch.exception.error_code;
4412 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4413 	events->exception_payload = vcpu->arch.exception.payload;
4414 
4415 	events->interrupt.injected =
4416 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4417 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4418 	events->interrupt.soft = 0;
4419 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4420 
4421 	events->nmi.injected = vcpu->arch.nmi_injected;
4422 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4423 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4424 	events->nmi.pad = 0;
4425 
4426 	events->sipi_vector = 0; /* never valid when reporting to user space */
4427 
4428 	events->smi.smm = is_smm(vcpu);
4429 	events->smi.pending = vcpu->arch.smi_pending;
4430 	events->smi.smm_inside_nmi =
4431 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4432 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4433 
4434 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4435 			 | KVM_VCPUEVENT_VALID_SHADOW
4436 			 | KVM_VCPUEVENT_VALID_SMM);
4437 	if (vcpu->kvm->arch.exception_payload_enabled)
4438 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4439 
4440 	memset(&events->reserved, 0, sizeof(events->reserved));
4441 }
4442 
4443 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4444 
4445 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4446 					      struct kvm_vcpu_events *events)
4447 {
4448 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4449 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4450 			      | KVM_VCPUEVENT_VALID_SHADOW
4451 			      | KVM_VCPUEVENT_VALID_SMM
4452 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4453 		return -EINVAL;
4454 
4455 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4456 		if (!vcpu->kvm->arch.exception_payload_enabled)
4457 			return -EINVAL;
4458 		if (events->exception.pending)
4459 			events->exception.injected = 0;
4460 		else
4461 			events->exception_has_payload = 0;
4462 	} else {
4463 		events->exception.pending = 0;
4464 		events->exception_has_payload = 0;
4465 	}
4466 
4467 	if ((events->exception.injected || events->exception.pending) &&
4468 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4469 		return -EINVAL;
4470 
4471 	/* INITs are latched while in SMM */
4472 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4473 	    (events->smi.smm || events->smi.pending) &&
4474 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4475 		return -EINVAL;
4476 
4477 	process_nmi(vcpu);
4478 	vcpu->arch.exception.injected = events->exception.injected;
4479 	vcpu->arch.exception.pending = events->exception.pending;
4480 	vcpu->arch.exception.nr = events->exception.nr;
4481 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4482 	vcpu->arch.exception.error_code = events->exception.error_code;
4483 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4484 	vcpu->arch.exception.payload = events->exception_payload;
4485 
4486 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4487 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4488 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4489 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4490 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4491 						events->interrupt.shadow);
4492 
4493 	vcpu->arch.nmi_injected = events->nmi.injected;
4494 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4495 		vcpu->arch.nmi_pending = events->nmi.pending;
4496 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4497 
4498 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4499 	    lapic_in_kernel(vcpu))
4500 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4501 
4502 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4503 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4504 			if (events->smi.smm)
4505 				vcpu->arch.hflags |= HF_SMM_MASK;
4506 			else
4507 				vcpu->arch.hflags &= ~HF_SMM_MASK;
4508 			kvm_smm_changed(vcpu);
4509 		}
4510 
4511 		vcpu->arch.smi_pending = events->smi.pending;
4512 
4513 		if (events->smi.smm) {
4514 			if (events->smi.smm_inside_nmi)
4515 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4516 			else
4517 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4518 		}
4519 
4520 		if (lapic_in_kernel(vcpu)) {
4521 			if (events->smi.latched_init)
4522 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4523 			else
4524 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4525 		}
4526 	}
4527 
4528 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4529 
4530 	return 0;
4531 }
4532 
4533 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4534 					     struct kvm_debugregs *dbgregs)
4535 {
4536 	unsigned long val;
4537 
4538 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4539 	kvm_get_dr(vcpu, 6, &val);
4540 	dbgregs->dr6 = val;
4541 	dbgregs->dr7 = vcpu->arch.dr7;
4542 	dbgregs->flags = 0;
4543 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4544 }
4545 
4546 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4547 					    struct kvm_debugregs *dbgregs)
4548 {
4549 	if (dbgregs->flags)
4550 		return -EINVAL;
4551 
4552 	if (!kvm_dr6_valid(dbgregs->dr6))
4553 		return -EINVAL;
4554 	if (!kvm_dr7_valid(dbgregs->dr7))
4555 		return -EINVAL;
4556 
4557 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4558 	kvm_update_dr0123(vcpu);
4559 	vcpu->arch.dr6 = dbgregs->dr6;
4560 	vcpu->arch.dr7 = dbgregs->dr7;
4561 	kvm_update_dr7(vcpu);
4562 
4563 	return 0;
4564 }
4565 
4566 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4567 
4568 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4569 {
4570 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4571 	u64 xstate_bv = xsave->header.xfeatures;
4572 	u64 valid;
4573 
4574 	/*
4575 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4576 	 * leaves 0 and 1 in the loop below.
4577 	 */
4578 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4579 
4580 	/* Set XSTATE_BV */
4581 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4582 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4583 
4584 	/*
4585 	 * Copy each region from the possibly compacted offset to the
4586 	 * non-compacted offset.
4587 	 */
4588 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4589 	while (valid) {
4590 		u64 xfeature_mask = valid & -valid;
4591 		int xfeature_nr = fls64(xfeature_mask) - 1;
4592 		void *src = get_xsave_addr(xsave, xfeature_nr);
4593 
4594 		if (src) {
4595 			u32 size, offset, ecx, edx;
4596 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4597 				    &size, &offset, &ecx, &edx);
4598 			if (xfeature_nr == XFEATURE_PKRU)
4599 				memcpy(dest + offset, &vcpu->arch.pkru,
4600 				       sizeof(vcpu->arch.pkru));
4601 			else
4602 				memcpy(dest + offset, src, size);
4603 
4604 		}
4605 
4606 		valid -= xfeature_mask;
4607 	}
4608 }
4609 
4610 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4611 {
4612 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4613 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4614 	u64 valid;
4615 
4616 	/*
4617 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4618 	 * leaves 0 and 1 in the loop below.
4619 	 */
4620 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4621 
4622 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4623 	xsave->header.xfeatures = xstate_bv;
4624 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4625 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4626 
4627 	/*
4628 	 * Copy each region from the non-compacted offset to the
4629 	 * possibly compacted offset.
4630 	 */
4631 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4632 	while (valid) {
4633 		u64 xfeature_mask = valid & -valid;
4634 		int xfeature_nr = fls64(xfeature_mask) - 1;
4635 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4636 
4637 		if (dest) {
4638 			u32 size, offset, ecx, edx;
4639 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4640 				    &size, &offset, &ecx, &edx);
4641 			if (xfeature_nr == XFEATURE_PKRU)
4642 				memcpy(&vcpu->arch.pkru, src + offset,
4643 				       sizeof(vcpu->arch.pkru));
4644 			else
4645 				memcpy(dest, src + offset, size);
4646 		}
4647 
4648 		valid -= xfeature_mask;
4649 	}
4650 }
4651 
4652 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4653 					 struct kvm_xsave *guest_xsave)
4654 {
4655 	if (!vcpu->arch.guest_fpu)
4656 		return;
4657 
4658 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4659 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4660 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4661 	} else {
4662 		memcpy(guest_xsave->region,
4663 			&vcpu->arch.guest_fpu->state.fxsave,
4664 			sizeof(struct fxregs_state));
4665 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4666 			XFEATURE_MASK_FPSSE;
4667 	}
4668 }
4669 
4670 #define XSAVE_MXCSR_OFFSET 24
4671 
4672 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4673 					struct kvm_xsave *guest_xsave)
4674 {
4675 	u64 xstate_bv;
4676 	u32 mxcsr;
4677 
4678 	if (!vcpu->arch.guest_fpu)
4679 		return 0;
4680 
4681 	xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4682 	mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4683 
4684 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4685 		/*
4686 		 * Here we allow setting states that are not present in
4687 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4688 		 * with old userspace.
4689 		 */
4690 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4691 			return -EINVAL;
4692 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4693 	} else {
4694 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4695 			mxcsr & ~mxcsr_feature_mask)
4696 			return -EINVAL;
4697 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4698 			guest_xsave->region, sizeof(struct fxregs_state));
4699 	}
4700 	return 0;
4701 }
4702 
4703 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4704 					struct kvm_xcrs *guest_xcrs)
4705 {
4706 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4707 		guest_xcrs->nr_xcrs = 0;
4708 		return;
4709 	}
4710 
4711 	guest_xcrs->nr_xcrs = 1;
4712 	guest_xcrs->flags = 0;
4713 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4714 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4715 }
4716 
4717 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4718 				       struct kvm_xcrs *guest_xcrs)
4719 {
4720 	int i, r = 0;
4721 
4722 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4723 		return -EINVAL;
4724 
4725 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4726 		return -EINVAL;
4727 
4728 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4729 		/* Only support XCR0 currently */
4730 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4731 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4732 				guest_xcrs->xcrs[i].value);
4733 			break;
4734 		}
4735 	if (r)
4736 		r = -EINVAL;
4737 	return r;
4738 }
4739 
4740 /*
4741  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4742  * stopped by the hypervisor.  This function will be called from the host only.
4743  * EINVAL is returned when the host attempts to set the flag for a guest that
4744  * does not support pv clocks.
4745  */
4746 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4747 {
4748 	if (!vcpu->arch.pv_time_enabled)
4749 		return -EINVAL;
4750 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4751 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4752 	return 0;
4753 }
4754 
4755 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4756 				     struct kvm_enable_cap *cap)
4757 {
4758 	int r;
4759 	uint16_t vmcs_version;
4760 	void __user *user_ptr;
4761 
4762 	if (cap->flags)
4763 		return -EINVAL;
4764 
4765 	switch (cap->cap) {
4766 	case KVM_CAP_HYPERV_SYNIC2:
4767 		if (cap->args[0])
4768 			return -EINVAL;
4769 		fallthrough;
4770 
4771 	case KVM_CAP_HYPERV_SYNIC:
4772 		if (!irqchip_in_kernel(vcpu->kvm))
4773 			return -EINVAL;
4774 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4775 					     KVM_CAP_HYPERV_SYNIC2);
4776 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4777 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4778 			return -ENOTTY;
4779 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4780 		if (!r) {
4781 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4782 			if (copy_to_user(user_ptr, &vmcs_version,
4783 					 sizeof(vmcs_version)))
4784 				r = -EFAULT;
4785 		}
4786 		return r;
4787 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4788 		if (!kvm_x86_ops.enable_direct_tlbflush)
4789 			return -ENOTTY;
4790 
4791 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4792 
4793 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4794 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4795 		if (vcpu->arch.pv_cpuid.enforce)
4796 			kvm_update_pv_runtime(vcpu);
4797 
4798 		return 0;
4799 	default:
4800 		return -EINVAL;
4801 	}
4802 }
4803 
4804 long kvm_arch_vcpu_ioctl(struct file *filp,
4805 			 unsigned int ioctl, unsigned long arg)
4806 {
4807 	struct kvm_vcpu *vcpu = filp->private_data;
4808 	void __user *argp = (void __user *)arg;
4809 	int r;
4810 	union {
4811 		struct kvm_lapic_state *lapic;
4812 		struct kvm_xsave *xsave;
4813 		struct kvm_xcrs *xcrs;
4814 		void *buffer;
4815 	} u;
4816 
4817 	vcpu_load(vcpu);
4818 
4819 	u.buffer = NULL;
4820 	switch (ioctl) {
4821 	case KVM_GET_LAPIC: {
4822 		r = -EINVAL;
4823 		if (!lapic_in_kernel(vcpu))
4824 			goto out;
4825 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4826 				GFP_KERNEL_ACCOUNT);
4827 
4828 		r = -ENOMEM;
4829 		if (!u.lapic)
4830 			goto out;
4831 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4832 		if (r)
4833 			goto out;
4834 		r = -EFAULT;
4835 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4836 			goto out;
4837 		r = 0;
4838 		break;
4839 	}
4840 	case KVM_SET_LAPIC: {
4841 		r = -EINVAL;
4842 		if (!lapic_in_kernel(vcpu))
4843 			goto out;
4844 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4845 		if (IS_ERR(u.lapic)) {
4846 			r = PTR_ERR(u.lapic);
4847 			goto out_nofree;
4848 		}
4849 
4850 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4851 		break;
4852 	}
4853 	case KVM_INTERRUPT: {
4854 		struct kvm_interrupt irq;
4855 
4856 		r = -EFAULT;
4857 		if (copy_from_user(&irq, argp, sizeof(irq)))
4858 			goto out;
4859 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4860 		break;
4861 	}
4862 	case KVM_NMI: {
4863 		r = kvm_vcpu_ioctl_nmi(vcpu);
4864 		break;
4865 	}
4866 	case KVM_SMI: {
4867 		r = kvm_vcpu_ioctl_smi(vcpu);
4868 		break;
4869 	}
4870 	case KVM_SET_CPUID: {
4871 		struct kvm_cpuid __user *cpuid_arg = argp;
4872 		struct kvm_cpuid cpuid;
4873 
4874 		r = -EFAULT;
4875 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4876 			goto out;
4877 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4878 		break;
4879 	}
4880 	case KVM_SET_CPUID2: {
4881 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4882 		struct kvm_cpuid2 cpuid;
4883 
4884 		r = -EFAULT;
4885 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4886 			goto out;
4887 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4888 					      cpuid_arg->entries);
4889 		break;
4890 	}
4891 	case KVM_GET_CPUID2: {
4892 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4893 		struct kvm_cpuid2 cpuid;
4894 
4895 		r = -EFAULT;
4896 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4897 			goto out;
4898 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4899 					      cpuid_arg->entries);
4900 		if (r)
4901 			goto out;
4902 		r = -EFAULT;
4903 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4904 			goto out;
4905 		r = 0;
4906 		break;
4907 	}
4908 	case KVM_GET_MSRS: {
4909 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4910 		r = msr_io(vcpu, argp, do_get_msr, 1);
4911 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4912 		break;
4913 	}
4914 	case KVM_SET_MSRS: {
4915 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4916 		r = msr_io(vcpu, argp, do_set_msr, 0);
4917 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4918 		break;
4919 	}
4920 	case KVM_TPR_ACCESS_REPORTING: {
4921 		struct kvm_tpr_access_ctl tac;
4922 
4923 		r = -EFAULT;
4924 		if (copy_from_user(&tac, argp, sizeof(tac)))
4925 			goto out;
4926 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4927 		if (r)
4928 			goto out;
4929 		r = -EFAULT;
4930 		if (copy_to_user(argp, &tac, sizeof(tac)))
4931 			goto out;
4932 		r = 0;
4933 		break;
4934 	};
4935 	case KVM_SET_VAPIC_ADDR: {
4936 		struct kvm_vapic_addr va;
4937 		int idx;
4938 
4939 		r = -EINVAL;
4940 		if (!lapic_in_kernel(vcpu))
4941 			goto out;
4942 		r = -EFAULT;
4943 		if (copy_from_user(&va, argp, sizeof(va)))
4944 			goto out;
4945 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4946 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4947 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4948 		break;
4949 	}
4950 	case KVM_X86_SETUP_MCE: {
4951 		u64 mcg_cap;
4952 
4953 		r = -EFAULT;
4954 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4955 			goto out;
4956 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4957 		break;
4958 	}
4959 	case KVM_X86_SET_MCE: {
4960 		struct kvm_x86_mce mce;
4961 
4962 		r = -EFAULT;
4963 		if (copy_from_user(&mce, argp, sizeof(mce)))
4964 			goto out;
4965 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4966 		break;
4967 	}
4968 	case KVM_GET_VCPU_EVENTS: {
4969 		struct kvm_vcpu_events events;
4970 
4971 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4972 
4973 		r = -EFAULT;
4974 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4975 			break;
4976 		r = 0;
4977 		break;
4978 	}
4979 	case KVM_SET_VCPU_EVENTS: {
4980 		struct kvm_vcpu_events events;
4981 
4982 		r = -EFAULT;
4983 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4984 			break;
4985 
4986 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4987 		break;
4988 	}
4989 	case KVM_GET_DEBUGREGS: {
4990 		struct kvm_debugregs dbgregs;
4991 
4992 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4993 
4994 		r = -EFAULT;
4995 		if (copy_to_user(argp, &dbgregs,
4996 				 sizeof(struct kvm_debugregs)))
4997 			break;
4998 		r = 0;
4999 		break;
5000 	}
5001 	case KVM_SET_DEBUGREGS: {
5002 		struct kvm_debugregs dbgregs;
5003 
5004 		r = -EFAULT;
5005 		if (copy_from_user(&dbgregs, argp,
5006 				   sizeof(struct kvm_debugregs)))
5007 			break;
5008 
5009 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5010 		break;
5011 	}
5012 	case KVM_GET_XSAVE: {
5013 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5014 		r = -ENOMEM;
5015 		if (!u.xsave)
5016 			break;
5017 
5018 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5019 
5020 		r = -EFAULT;
5021 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5022 			break;
5023 		r = 0;
5024 		break;
5025 	}
5026 	case KVM_SET_XSAVE: {
5027 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5028 		if (IS_ERR(u.xsave)) {
5029 			r = PTR_ERR(u.xsave);
5030 			goto out_nofree;
5031 		}
5032 
5033 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5034 		break;
5035 	}
5036 	case KVM_GET_XCRS: {
5037 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5038 		r = -ENOMEM;
5039 		if (!u.xcrs)
5040 			break;
5041 
5042 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5043 
5044 		r = -EFAULT;
5045 		if (copy_to_user(argp, u.xcrs,
5046 				 sizeof(struct kvm_xcrs)))
5047 			break;
5048 		r = 0;
5049 		break;
5050 	}
5051 	case KVM_SET_XCRS: {
5052 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5053 		if (IS_ERR(u.xcrs)) {
5054 			r = PTR_ERR(u.xcrs);
5055 			goto out_nofree;
5056 		}
5057 
5058 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5059 		break;
5060 	}
5061 	case KVM_SET_TSC_KHZ: {
5062 		u32 user_tsc_khz;
5063 
5064 		r = -EINVAL;
5065 		user_tsc_khz = (u32)arg;
5066 
5067 		if (kvm_has_tsc_control &&
5068 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5069 			goto out;
5070 
5071 		if (user_tsc_khz == 0)
5072 			user_tsc_khz = tsc_khz;
5073 
5074 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5075 			r = 0;
5076 
5077 		goto out;
5078 	}
5079 	case KVM_GET_TSC_KHZ: {
5080 		r = vcpu->arch.virtual_tsc_khz;
5081 		goto out;
5082 	}
5083 	case KVM_KVMCLOCK_CTRL: {
5084 		r = kvm_set_guest_paused(vcpu);
5085 		goto out;
5086 	}
5087 	case KVM_ENABLE_CAP: {
5088 		struct kvm_enable_cap cap;
5089 
5090 		r = -EFAULT;
5091 		if (copy_from_user(&cap, argp, sizeof(cap)))
5092 			goto out;
5093 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5094 		break;
5095 	}
5096 	case KVM_GET_NESTED_STATE: {
5097 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5098 		u32 user_data_size;
5099 
5100 		r = -EINVAL;
5101 		if (!kvm_x86_ops.nested_ops->get_state)
5102 			break;
5103 
5104 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5105 		r = -EFAULT;
5106 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5107 			break;
5108 
5109 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5110 						     user_data_size);
5111 		if (r < 0)
5112 			break;
5113 
5114 		if (r > user_data_size) {
5115 			if (put_user(r, &user_kvm_nested_state->size))
5116 				r = -EFAULT;
5117 			else
5118 				r = -E2BIG;
5119 			break;
5120 		}
5121 
5122 		r = 0;
5123 		break;
5124 	}
5125 	case KVM_SET_NESTED_STATE: {
5126 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5127 		struct kvm_nested_state kvm_state;
5128 		int idx;
5129 
5130 		r = -EINVAL;
5131 		if (!kvm_x86_ops.nested_ops->set_state)
5132 			break;
5133 
5134 		r = -EFAULT;
5135 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5136 			break;
5137 
5138 		r = -EINVAL;
5139 		if (kvm_state.size < sizeof(kvm_state))
5140 			break;
5141 
5142 		if (kvm_state.flags &
5143 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5144 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5145 		      | KVM_STATE_NESTED_GIF_SET))
5146 			break;
5147 
5148 		/* nested_run_pending implies guest_mode.  */
5149 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5150 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5151 			break;
5152 
5153 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5154 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5155 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5156 		break;
5157 	}
5158 	case KVM_GET_SUPPORTED_HV_CPUID:
5159 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5160 		break;
5161 #ifdef CONFIG_KVM_XEN
5162 	case KVM_XEN_VCPU_GET_ATTR: {
5163 		struct kvm_xen_vcpu_attr xva;
5164 
5165 		r = -EFAULT;
5166 		if (copy_from_user(&xva, argp, sizeof(xva)))
5167 			goto out;
5168 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5169 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5170 			r = -EFAULT;
5171 		break;
5172 	}
5173 	case KVM_XEN_VCPU_SET_ATTR: {
5174 		struct kvm_xen_vcpu_attr xva;
5175 
5176 		r = -EFAULT;
5177 		if (copy_from_user(&xva, argp, sizeof(xva)))
5178 			goto out;
5179 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5180 		break;
5181 	}
5182 #endif
5183 	default:
5184 		r = -EINVAL;
5185 	}
5186 out:
5187 	kfree(u.buffer);
5188 out_nofree:
5189 	vcpu_put(vcpu);
5190 	return r;
5191 }
5192 
5193 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5194 {
5195 	return VM_FAULT_SIGBUS;
5196 }
5197 
5198 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5199 {
5200 	int ret;
5201 
5202 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5203 		return -EINVAL;
5204 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5205 	return ret;
5206 }
5207 
5208 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5209 					      u64 ident_addr)
5210 {
5211 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5212 }
5213 
5214 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5215 					 unsigned long kvm_nr_mmu_pages)
5216 {
5217 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5218 		return -EINVAL;
5219 
5220 	mutex_lock(&kvm->slots_lock);
5221 
5222 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5223 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5224 
5225 	mutex_unlock(&kvm->slots_lock);
5226 	return 0;
5227 }
5228 
5229 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5230 {
5231 	return kvm->arch.n_max_mmu_pages;
5232 }
5233 
5234 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5235 {
5236 	struct kvm_pic *pic = kvm->arch.vpic;
5237 	int r;
5238 
5239 	r = 0;
5240 	switch (chip->chip_id) {
5241 	case KVM_IRQCHIP_PIC_MASTER:
5242 		memcpy(&chip->chip.pic, &pic->pics[0],
5243 			sizeof(struct kvm_pic_state));
5244 		break;
5245 	case KVM_IRQCHIP_PIC_SLAVE:
5246 		memcpy(&chip->chip.pic, &pic->pics[1],
5247 			sizeof(struct kvm_pic_state));
5248 		break;
5249 	case KVM_IRQCHIP_IOAPIC:
5250 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5251 		break;
5252 	default:
5253 		r = -EINVAL;
5254 		break;
5255 	}
5256 	return r;
5257 }
5258 
5259 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5260 {
5261 	struct kvm_pic *pic = kvm->arch.vpic;
5262 	int r;
5263 
5264 	r = 0;
5265 	switch (chip->chip_id) {
5266 	case KVM_IRQCHIP_PIC_MASTER:
5267 		spin_lock(&pic->lock);
5268 		memcpy(&pic->pics[0], &chip->chip.pic,
5269 			sizeof(struct kvm_pic_state));
5270 		spin_unlock(&pic->lock);
5271 		break;
5272 	case KVM_IRQCHIP_PIC_SLAVE:
5273 		spin_lock(&pic->lock);
5274 		memcpy(&pic->pics[1], &chip->chip.pic,
5275 			sizeof(struct kvm_pic_state));
5276 		spin_unlock(&pic->lock);
5277 		break;
5278 	case KVM_IRQCHIP_IOAPIC:
5279 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5280 		break;
5281 	default:
5282 		r = -EINVAL;
5283 		break;
5284 	}
5285 	kvm_pic_update_irq(pic);
5286 	return r;
5287 }
5288 
5289 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5290 {
5291 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5292 
5293 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5294 
5295 	mutex_lock(&kps->lock);
5296 	memcpy(ps, &kps->channels, sizeof(*ps));
5297 	mutex_unlock(&kps->lock);
5298 	return 0;
5299 }
5300 
5301 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5302 {
5303 	int i;
5304 	struct kvm_pit *pit = kvm->arch.vpit;
5305 
5306 	mutex_lock(&pit->pit_state.lock);
5307 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5308 	for (i = 0; i < 3; i++)
5309 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5310 	mutex_unlock(&pit->pit_state.lock);
5311 	return 0;
5312 }
5313 
5314 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5315 {
5316 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5317 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5318 		sizeof(ps->channels));
5319 	ps->flags = kvm->arch.vpit->pit_state.flags;
5320 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5321 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5322 	return 0;
5323 }
5324 
5325 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5326 {
5327 	int start = 0;
5328 	int i;
5329 	u32 prev_legacy, cur_legacy;
5330 	struct kvm_pit *pit = kvm->arch.vpit;
5331 
5332 	mutex_lock(&pit->pit_state.lock);
5333 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5334 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5335 	if (!prev_legacy && cur_legacy)
5336 		start = 1;
5337 	memcpy(&pit->pit_state.channels, &ps->channels,
5338 	       sizeof(pit->pit_state.channels));
5339 	pit->pit_state.flags = ps->flags;
5340 	for (i = 0; i < 3; i++)
5341 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5342 				   start && i == 0);
5343 	mutex_unlock(&pit->pit_state.lock);
5344 	return 0;
5345 }
5346 
5347 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5348 				 struct kvm_reinject_control *control)
5349 {
5350 	struct kvm_pit *pit = kvm->arch.vpit;
5351 
5352 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5353 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5354 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5355 	 */
5356 	mutex_lock(&pit->pit_state.lock);
5357 	kvm_pit_set_reinject(pit, control->pit_reinject);
5358 	mutex_unlock(&pit->pit_state.lock);
5359 
5360 	return 0;
5361 }
5362 
5363 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5364 {
5365 
5366 	/*
5367 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5368 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5369 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5370 	 * VM-Exit.
5371 	 */
5372 	struct kvm_vcpu *vcpu;
5373 	int i;
5374 
5375 	kvm_for_each_vcpu(i, vcpu, kvm)
5376 		kvm_vcpu_kick(vcpu);
5377 }
5378 
5379 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5380 			bool line_status)
5381 {
5382 	if (!irqchip_in_kernel(kvm))
5383 		return -ENXIO;
5384 
5385 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5386 					irq_event->irq, irq_event->level,
5387 					line_status);
5388 	return 0;
5389 }
5390 
5391 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5392 			    struct kvm_enable_cap *cap)
5393 {
5394 	int r;
5395 
5396 	if (cap->flags)
5397 		return -EINVAL;
5398 
5399 	switch (cap->cap) {
5400 	case KVM_CAP_DISABLE_QUIRKS:
5401 		kvm->arch.disabled_quirks = cap->args[0];
5402 		r = 0;
5403 		break;
5404 	case KVM_CAP_SPLIT_IRQCHIP: {
5405 		mutex_lock(&kvm->lock);
5406 		r = -EINVAL;
5407 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5408 			goto split_irqchip_unlock;
5409 		r = -EEXIST;
5410 		if (irqchip_in_kernel(kvm))
5411 			goto split_irqchip_unlock;
5412 		if (kvm->created_vcpus)
5413 			goto split_irqchip_unlock;
5414 		r = kvm_setup_empty_irq_routing(kvm);
5415 		if (r)
5416 			goto split_irqchip_unlock;
5417 		/* Pairs with irqchip_in_kernel. */
5418 		smp_wmb();
5419 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5420 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5421 		r = 0;
5422 split_irqchip_unlock:
5423 		mutex_unlock(&kvm->lock);
5424 		break;
5425 	}
5426 	case KVM_CAP_X2APIC_API:
5427 		r = -EINVAL;
5428 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5429 			break;
5430 
5431 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5432 			kvm->arch.x2apic_format = true;
5433 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5434 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5435 
5436 		r = 0;
5437 		break;
5438 	case KVM_CAP_X86_DISABLE_EXITS:
5439 		r = -EINVAL;
5440 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5441 			break;
5442 
5443 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5444 			kvm_can_mwait_in_guest())
5445 			kvm->arch.mwait_in_guest = true;
5446 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5447 			kvm->arch.hlt_in_guest = true;
5448 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5449 			kvm->arch.pause_in_guest = true;
5450 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5451 			kvm->arch.cstate_in_guest = true;
5452 		r = 0;
5453 		break;
5454 	case KVM_CAP_MSR_PLATFORM_INFO:
5455 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5456 		r = 0;
5457 		break;
5458 	case KVM_CAP_EXCEPTION_PAYLOAD:
5459 		kvm->arch.exception_payload_enabled = cap->args[0];
5460 		r = 0;
5461 		break;
5462 	case KVM_CAP_X86_USER_SPACE_MSR:
5463 		kvm->arch.user_space_msr_mask = cap->args[0];
5464 		r = 0;
5465 		break;
5466 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5467 		r = -EINVAL;
5468 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5469 			break;
5470 
5471 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5472 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5473 			break;
5474 
5475 		if (kvm_has_bus_lock_exit &&
5476 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5477 			kvm->arch.bus_lock_detection_enabled = true;
5478 		r = 0;
5479 		break;
5480 #ifdef CONFIG_X86_SGX_KVM
5481 	case KVM_CAP_SGX_ATTRIBUTE: {
5482 		unsigned long allowed_attributes = 0;
5483 
5484 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5485 		if (r)
5486 			break;
5487 
5488 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5489 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5490 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5491 			kvm->arch.sgx_provisioning_allowed = true;
5492 		else
5493 			r = -EINVAL;
5494 		break;
5495 	}
5496 #endif
5497 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5498 		r = -EINVAL;
5499 		if (kvm_x86_ops.vm_copy_enc_context_from)
5500 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5501 		return r;
5502 	default:
5503 		r = -EINVAL;
5504 		break;
5505 	}
5506 	return r;
5507 }
5508 
5509 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5510 {
5511 	struct kvm_x86_msr_filter *msr_filter;
5512 
5513 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5514 	if (!msr_filter)
5515 		return NULL;
5516 
5517 	msr_filter->default_allow = default_allow;
5518 	return msr_filter;
5519 }
5520 
5521 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5522 {
5523 	u32 i;
5524 
5525 	if (!msr_filter)
5526 		return;
5527 
5528 	for (i = 0; i < msr_filter->count; i++)
5529 		kfree(msr_filter->ranges[i].bitmap);
5530 
5531 	kfree(msr_filter);
5532 }
5533 
5534 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5535 			      struct kvm_msr_filter_range *user_range)
5536 {
5537 	unsigned long *bitmap = NULL;
5538 	size_t bitmap_size;
5539 
5540 	if (!user_range->nmsrs)
5541 		return 0;
5542 
5543 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5544 		return -EINVAL;
5545 
5546 	if (!user_range->flags)
5547 		return -EINVAL;
5548 
5549 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5550 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5551 		return -EINVAL;
5552 
5553 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5554 	if (IS_ERR(bitmap))
5555 		return PTR_ERR(bitmap);
5556 
5557 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5558 		.flags = user_range->flags,
5559 		.base = user_range->base,
5560 		.nmsrs = user_range->nmsrs,
5561 		.bitmap = bitmap,
5562 	};
5563 
5564 	msr_filter->count++;
5565 	return 0;
5566 }
5567 
5568 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5569 {
5570 	struct kvm_msr_filter __user *user_msr_filter = argp;
5571 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5572 	struct kvm_msr_filter filter;
5573 	bool default_allow;
5574 	bool empty = true;
5575 	int r = 0;
5576 	u32 i;
5577 
5578 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5579 		return -EFAULT;
5580 
5581 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5582 		empty &= !filter.ranges[i].nmsrs;
5583 
5584 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5585 	if (empty && !default_allow)
5586 		return -EINVAL;
5587 
5588 	new_filter = kvm_alloc_msr_filter(default_allow);
5589 	if (!new_filter)
5590 		return -ENOMEM;
5591 
5592 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5593 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5594 		if (r) {
5595 			kvm_free_msr_filter(new_filter);
5596 			return r;
5597 		}
5598 	}
5599 
5600 	mutex_lock(&kvm->lock);
5601 
5602 	/* The per-VM filter is protected by kvm->lock... */
5603 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5604 
5605 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5606 	synchronize_srcu(&kvm->srcu);
5607 
5608 	kvm_free_msr_filter(old_filter);
5609 
5610 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5611 	mutex_unlock(&kvm->lock);
5612 
5613 	return 0;
5614 }
5615 
5616 long kvm_arch_vm_ioctl(struct file *filp,
5617 		       unsigned int ioctl, unsigned long arg)
5618 {
5619 	struct kvm *kvm = filp->private_data;
5620 	void __user *argp = (void __user *)arg;
5621 	int r = -ENOTTY;
5622 	/*
5623 	 * This union makes it completely explicit to gcc-3.x
5624 	 * that these two variables' stack usage should be
5625 	 * combined, not added together.
5626 	 */
5627 	union {
5628 		struct kvm_pit_state ps;
5629 		struct kvm_pit_state2 ps2;
5630 		struct kvm_pit_config pit_config;
5631 	} u;
5632 
5633 	switch (ioctl) {
5634 	case KVM_SET_TSS_ADDR:
5635 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5636 		break;
5637 	case KVM_SET_IDENTITY_MAP_ADDR: {
5638 		u64 ident_addr;
5639 
5640 		mutex_lock(&kvm->lock);
5641 		r = -EINVAL;
5642 		if (kvm->created_vcpus)
5643 			goto set_identity_unlock;
5644 		r = -EFAULT;
5645 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5646 			goto set_identity_unlock;
5647 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5648 set_identity_unlock:
5649 		mutex_unlock(&kvm->lock);
5650 		break;
5651 	}
5652 	case KVM_SET_NR_MMU_PAGES:
5653 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5654 		break;
5655 	case KVM_GET_NR_MMU_PAGES:
5656 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5657 		break;
5658 	case KVM_CREATE_IRQCHIP: {
5659 		mutex_lock(&kvm->lock);
5660 
5661 		r = -EEXIST;
5662 		if (irqchip_in_kernel(kvm))
5663 			goto create_irqchip_unlock;
5664 
5665 		r = -EINVAL;
5666 		if (kvm->created_vcpus)
5667 			goto create_irqchip_unlock;
5668 
5669 		r = kvm_pic_init(kvm);
5670 		if (r)
5671 			goto create_irqchip_unlock;
5672 
5673 		r = kvm_ioapic_init(kvm);
5674 		if (r) {
5675 			kvm_pic_destroy(kvm);
5676 			goto create_irqchip_unlock;
5677 		}
5678 
5679 		r = kvm_setup_default_irq_routing(kvm);
5680 		if (r) {
5681 			kvm_ioapic_destroy(kvm);
5682 			kvm_pic_destroy(kvm);
5683 			goto create_irqchip_unlock;
5684 		}
5685 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5686 		smp_wmb();
5687 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5688 	create_irqchip_unlock:
5689 		mutex_unlock(&kvm->lock);
5690 		break;
5691 	}
5692 	case KVM_CREATE_PIT:
5693 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5694 		goto create_pit;
5695 	case KVM_CREATE_PIT2:
5696 		r = -EFAULT;
5697 		if (copy_from_user(&u.pit_config, argp,
5698 				   sizeof(struct kvm_pit_config)))
5699 			goto out;
5700 	create_pit:
5701 		mutex_lock(&kvm->lock);
5702 		r = -EEXIST;
5703 		if (kvm->arch.vpit)
5704 			goto create_pit_unlock;
5705 		r = -ENOMEM;
5706 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5707 		if (kvm->arch.vpit)
5708 			r = 0;
5709 	create_pit_unlock:
5710 		mutex_unlock(&kvm->lock);
5711 		break;
5712 	case KVM_GET_IRQCHIP: {
5713 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5714 		struct kvm_irqchip *chip;
5715 
5716 		chip = memdup_user(argp, sizeof(*chip));
5717 		if (IS_ERR(chip)) {
5718 			r = PTR_ERR(chip);
5719 			goto out;
5720 		}
5721 
5722 		r = -ENXIO;
5723 		if (!irqchip_kernel(kvm))
5724 			goto get_irqchip_out;
5725 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5726 		if (r)
5727 			goto get_irqchip_out;
5728 		r = -EFAULT;
5729 		if (copy_to_user(argp, chip, sizeof(*chip)))
5730 			goto get_irqchip_out;
5731 		r = 0;
5732 	get_irqchip_out:
5733 		kfree(chip);
5734 		break;
5735 	}
5736 	case KVM_SET_IRQCHIP: {
5737 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5738 		struct kvm_irqchip *chip;
5739 
5740 		chip = memdup_user(argp, sizeof(*chip));
5741 		if (IS_ERR(chip)) {
5742 			r = PTR_ERR(chip);
5743 			goto out;
5744 		}
5745 
5746 		r = -ENXIO;
5747 		if (!irqchip_kernel(kvm))
5748 			goto set_irqchip_out;
5749 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5750 	set_irqchip_out:
5751 		kfree(chip);
5752 		break;
5753 	}
5754 	case KVM_GET_PIT: {
5755 		r = -EFAULT;
5756 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5757 			goto out;
5758 		r = -ENXIO;
5759 		if (!kvm->arch.vpit)
5760 			goto out;
5761 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5762 		if (r)
5763 			goto out;
5764 		r = -EFAULT;
5765 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5766 			goto out;
5767 		r = 0;
5768 		break;
5769 	}
5770 	case KVM_SET_PIT: {
5771 		r = -EFAULT;
5772 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5773 			goto out;
5774 		mutex_lock(&kvm->lock);
5775 		r = -ENXIO;
5776 		if (!kvm->arch.vpit)
5777 			goto set_pit_out;
5778 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5779 set_pit_out:
5780 		mutex_unlock(&kvm->lock);
5781 		break;
5782 	}
5783 	case KVM_GET_PIT2: {
5784 		r = -ENXIO;
5785 		if (!kvm->arch.vpit)
5786 			goto out;
5787 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5788 		if (r)
5789 			goto out;
5790 		r = -EFAULT;
5791 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5792 			goto out;
5793 		r = 0;
5794 		break;
5795 	}
5796 	case KVM_SET_PIT2: {
5797 		r = -EFAULT;
5798 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5799 			goto out;
5800 		mutex_lock(&kvm->lock);
5801 		r = -ENXIO;
5802 		if (!kvm->arch.vpit)
5803 			goto set_pit2_out;
5804 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5805 set_pit2_out:
5806 		mutex_unlock(&kvm->lock);
5807 		break;
5808 	}
5809 	case KVM_REINJECT_CONTROL: {
5810 		struct kvm_reinject_control control;
5811 		r =  -EFAULT;
5812 		if (copy_from_user(&control, argp, sizeof(control)))
5813 			goto out;
5814 		r = -ENXIO;
5815 		if (!kvm->arch.vpit)
5816 			goto out;
5817 		r = kvm_vm_ioctl_reinject(kvm, &control);
5818 		break;
5819 	}
5820 	case KVM_SET_BOOT_CPU_ID:
5821 		r = 0;
5822 		mutex_lock(&kvm->lock);
5823 		if (kvm->created_vcpus)
5824 			r = -EBUSY;
5825 		else
5826 			kvm->arch.bsp_vcpu_id = arg;
5827 		mutex_unlock(&kvm->lock);
5828 		break;
5829 #ifdef CONFIG_KVM_XEN
5830 	case KVM_XEN_HVM_CONFIG: {
5831 		struct kvm_xen_hvm_config xhc;
5832 		r = -EFAULT;
5833 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5834 			goto out;
5835 		r = kvm_xen_hvm_config(kvm, &xhc);
5836 		break;
5837 	}
5838 	case KVM_XEN_HVM_GET_ATTR: {
5839 		struct kvm_xen_hvm_attr xha;
5840 
5841 		r = -EFAULT;
5842 		if (copy_from_user(&xha, argp, sizeof(xha)))
5843 			goto out;
5844 		r = kvm_xen_hvm_get_attr(kvm, &xha);
5845 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5846 			r = -EFAULT;
5847 		break;
5848 	}
5849 	case KVM_XEN_HVM_SET_ATTR: {
5850 		struct kvm_xen_hvm_attr xha;
5851 
5852 		r = -EFAULT;
5853 		if (copy_from_user(&xha, argp, sizeof(xha)))
5854 			goto out;
5855 		r = kvm_xen_hvm_set_attr(kvm, &xha);
5856 		break;
5857 	}
5858 #endif
5859 	case KVM_SET_CLOCK: {
5860 		struct kvm_arch *ka = &kvm->arch;
5861 		struct kvm_clock_data user_ns;
5862 		u64 now_ns;
5863 
5864 		r = -EFAULT;
5865 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5866 			goto out;
5867 
5868 		r = -EINVAL;
5869 		if (user_ns.flags)
5870 			goto out;
5871 
5872 		r = 0;
5873 		/*
5874 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5875 		 * kvm_gen_update_masterclock() can be cut down to locked
5876 		 * pvclock_update_vm_gtod_copy().
5877 		 */
5878 		kvm_gen_update_masterclock(kvm);
5879 
5880 		/*
5881 		 * This pairs with kvm_guest_time_update(): when masterclock is
5882 		 * in use, we use master_kernel_ns + kvmclock_offset to set
5883 		 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5884 		 * is slightly ahead) here we risk going negative on unsigned
5885 		 * 'system_time' when 'user_ns.clock' is very small.
5886 		 */
5887 		spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5888 		if (kvm->arch.use_master_clock)
5889 			now_ns = ka->master_kernel_ns;
5890 		else
5891 			now_ns = get_kvmclock_base_ns();
5892 		ka->kvmclock_offset = user_ns.clock - now_ns;
5893 		spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5894 
5895 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5896 		break;
5897 	}
5898 	case KVM_GET_CLOCK: {
5899 		struct kvm_clock_data user_ns;
5900 		u64 now_ns;
5901 
5902 		now_ns = get_kvmclock_ns(kvm);
5903 		user_ns.clock = now_ns;
5904 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5905 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5906 
5907 		r = -EFAULT;
5908 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5909 			goto out;
5910 		r = 0;
5911 		break;
5912 	}
5913 	case KVM_MEMORY_ENCRYPT_OP: {
5914 		r = -ENOTTY;
5915 		if (kvm_x86_ops.mem_enc_op)
5916 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5917 		break;
5918 	}
5919 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5920 		struct kvm_enc_region region;
5921 
5922 		r = -EFAULT;
5923 		if (copy_from_user(&region, argp, sizeof(region)))
5924 			goto out;
5925 
5926 		r = -ENOTTY;
5927 		if (kvm_x86_ops.mem_enc_reg_region)
5928 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
5929 		break;
5930 	}
5931 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5932 		struct kvm_enc_region region;
5933 
5934 		r = -EFAULT;
5935 		if (copy_from_user(&region, argp, sizeof(region)))
5936 			goto out;
5937 
5938 		r = -ENOTTY;
5939 		if (kvm_x86_ops.mem_enc_unreg_region)
5940 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
5941 		break;
5942 	}
5943 	case KVM_HYPERV_EVENTFD: {
5944 		struct kvm_hyperv_eventfd hvevfd;
5945 
5946 		r = -EFAULT;
5947 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5948 			goto out;
5949 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5950 		break;
5951 	}
5952 	case KVM_SET_PMU_EVENT_FILTER:
5953 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5954 		break;
5955 	case KVM_X86_SET_MSR_FILTER:
5956 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5957 		break;
5958 	default:
5959 		r = -ENOTTY;
5960 	}
5961 out:
5962 	return r;
5963 }
5964 
5965 static void kvm_init_msr_list(void)
5966 {
5967 	struct x86_pmu_capability x86_pmu;
5968 	u32 dummy[2];
5969 	unsigned i;
5970 
5971 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5972 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5973 
5974 	perf_get_x86_pmu_capability(&x86_pmu);
5975 
5976 	num_msrs_to_save = 0;
5977 	num_emulated_msrs = 0;
5978 	num_msr_based_features = 0;
5979 
5980 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5981 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5982 			continue;
5983 
5984 		/*
5985 		 * Even MSRs that are valid in the host may not be exposed
5986 		 * to the guests in some cases.
5987 		 */
5988 		switch (msrs_to_save_all[i]) {
5989 		case MSR_IA32_BNDCFGS:
5990 			if (!kvm_mpx_supported())
5991 				continue;
5992 			break;
5993 		case MSR_TSC_AUX:
5994 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
5995 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
5996 				continue;
5997 			break;
5998 		case MSR_IA32_UMWAIT_CONTROL:
5999 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6000 				continue;
6001 			break;
6002 		case MSR_IA32_RTIT_CTL:
6003 		case MSR_IA32_RTIT_STATUS:
6004 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6005 				continue;
6006 			break;
6007 		case MSR_IA32_RTIT_CR3_MATCH:
6008 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6009 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6010 				continue;
6011 			break;
6012 		case MSR_IA32_RTIT_OUTPUT_BASE:
6013 		case MSR_IA32_RTIT_OUTPUT_MASK:
6014 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6015 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6016 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6017 				continue;
6018 			break;
6019 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6020 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6021 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6022 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6023 				continue;
6024 			break;
6025 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6026 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6027 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6028 				continue;
6029 			break;
6030 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6031 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6032 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6033 				continue;
6034 			break;
6035 		default:
6036 			break;
6037 		}
6038 
6039 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6040 	}
6041 
6042 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6043 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6044 			continue;
6045 
6046 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6047 	}
6048 
6049 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6050 		struct kvm_msr_entry msr;
6051 
6052 		msr.index = msr_based_features_all[i];
6053 		if (kvm_get_msr_feature(&msr))
6054 			continue;
6055 
6056 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6057 	}
6058 }
6059 
6060 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6061 			   const void *v)
6062 {
6063 	int handled = 0;
6064 	int n;
6065 
6066 	do {
6067 		n = min(len, 8);
6068 		if (!(lapic_in_kernel(vcpu) &&
6069 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6070 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6071 			break;
6072 		handled += n;
6073 		addr += n;
6074 		len -= n;
6075 		v += n;
6076 	} while (len);
6077 
6078 	return handled;
6079 }
6080 
6081 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6082 {
6083 	int handled = 0;
6084 	int n;
6085 
6086 	do {
6087 		n = min(len, 8);
6088 		if (!(lapic_in_kernel(vcpu) &&
6089 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6090 					 addr, n, v))
6091 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6092 			break;
6093 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6094 		handled += n;
6095 		addr += n;
6096 		len -= n;
6097 		v += n;
6098 	} while (len);
6099 
6100 	return handled;
6101 }
6102 
6103 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6104 			struct kvm_segment *var, int seg)
6105 {
6106 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6107 }
6108 
6109 void kvm_get_segment(struct kvm_vcpu *vcpu,
6110 		     struct kvm_segment *var, int seg)
6111 {
6112 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6113 }
6114 
6115 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6116 			   struct x86_exception *exception)
6117 {
6118 	gpa_t t_gpa;
6119 
6120 	BUG_ON(!mmu_is_nested(vcpu));
6121 
6122 	/* NPT walks are always user-walks */
6123 	access |= PFERR_USER_MASK;
6124 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6125 
6126 	return t_gpa;
6127 }
6128 
6129 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6130 			      struct x86_exception *exception)
6131 {
6132 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6133 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6134 }
6135 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6136 
6137  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6138 				struct x86_exception *exception)
6139 {
6140 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6141 	access |= PFERR_FETCH_MASK;
6142 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6143 }
6144 
6145 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6146 			       struct x86_exception *exception)
6147 {
6148 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6149 	access |= PFERR_WRITE_MASK;
6150 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6151 }
6152 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6153 
6154 /* uses this to access any guest's mapped memory without checking CPL */
6155 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6156 				struct x86_exception *exception)
6157 {
6158 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6159 }
6160 
6161 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6162 				      struct kvm_vcpu *vcpu, u32 access,
6163 				      struct x86_exception *exception)
6164 {
6165 	void *data = val;
6166 	int r = X86EMUL_CONTINUE;
6167 
6168 	while (bytes) {
6169 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6170 							    exception);
6171 		unsigned offset = addr & (PAGE_SIZE-1);
6172 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6173 		int ret;
6174 
6175 		if (gpa == UNMAPPED_GVA)
6176 			return X86EMUL_PROPAGATE_FAULT;
6177 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6178 					       offset, toread);
6179 		if (ret < 0) {
6180 			r = X86EMUL_IO_NEEDED;
6181 			goto out;
6182 		}
6183 
6184 		bytes -= toread;
6185 		data += toread;
6186 		addr += toread;
6187 	}
6188 out:
6189 	return r;
6190 }
6191 
6192 /* used for instruction fetching */
6193 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6194 				gva_t addr, void *val, unsigned int bytes,
6195 				struct x86_exception *exception)
6196 {
6197 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6198 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6199 	unsigned offset;
6200 	int ret;
6201 
6202 	/* Inline kvm_read_guest_virt_helper for speed.  */
6203 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6204 						    exception);
6205 	if (unlikely(gpa == UNMAPPED_GVA))
6206 		return X86EMUL_PROPAGATE_FAULT;
6207 
6208 	offset = addr & (PAGE_SIZE-1);
6209 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6210 		bytes = (unsigned)PAGE_SIZE - offset;
6211 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6212 				       offset, bytes);
6213 	if (unlikely(ret < 0))
6214 		return X86EMUL_IO_NEEDED;
6215 
6216 	return X86EMUL_CONTINUE;
6217 }
6218 
6219 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6220 			       gva_t addr, void *val, unsigned int bytes,
6221 			       struct x86_exception *exception)
6222 {
6223 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6224 
6225 	/*
6226 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6227 	 * is returned, but our callers are not ready for that and they blindly
6228 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6229 	 * uninitialized kernel stack memory into cr2 and error code.
6230 	 */
6231 	memset(exception, 0, sizeof(*exception));
6232 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6233 					  exception);
6234 }
6235 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6236 
6237 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6238 			     gva_t addr, void *val, unsigned int bytes,
6239 			     struct x86_exception *exception, bool system)
6240 {
6241 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6242 	u32 access = 0;
6243 
6244 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6245 		access |= PFERR_USER_MASK;
6246 
6247 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6248 }
6249 
6250 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6251 		unsigned long addr, void *val, unsigned int bytes)
6252 {
6253 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6254 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6255 
6256 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6257 }
6258 
6259 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6260 				      struct kvm_vcpu *vcpu, u32 access,
6261 				      struct x86_exception *exception)
6262 {
6263 	void *data = val;
6264 	int r = X86EMUL_CONTINUE;
6265 
6266 	while (bytes) {
6267 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6268 							     access,
6269 							     exception);
6270 		unsigned offset = addr & (PAGE_SIZE-1);
6271 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6272 		int ret;
6273 
6274 		if (gpa == UNMAPPED_GVA)
6275 			return X86EMUL_PROPAGATE_FAULT;
6276 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6277 		if (ret < 0) {
6278 			r = X86EMUL_IO_NEEDED;
6279 			goto out;
6280 		}
6281 
6282 		bytes -= towrite;
6283 		data += towrite;
6284 		addr += towrite;
6285 	}
6286 out:
6287 	return r;
6288 }
6289 
6290 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6291 			      unsigned int bytes, struct x86_exception *exception,
6292 			      bool system)
6293 {
6294 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6295 	u32 access = PFERR_WRITE_MASK;
6296 
6297 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6298 		access |= PFERR_USER_MASK;
6299 
6300 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6301 					   access, exception);
6302 }
6303 
6304 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6305 				unsigned int bytes, struct x86_exception *exception)
6306 {
6307 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6308 	vcpu->arch.l1tf_flush_l1d = true;
6309 
6310 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6311 					   PFERR_WRITE_MASK, exception);
6312 }
6313 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6314 
6315 int handle_ud(struct kvm_vcpu *vcpu)
6316 {
6317 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6318 	int emul_type = EMULTYPE_TRAP_UD;
6319 	char sig[5]; /* ud2; .ascii "kvm" */
6320 	struct x86_exception e;
6321 
6322 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6323 		return 1;
6324 
6325 	if (force_emulation_prefix &&
6326 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6327 				sig, sizeof(sig), &e) == 0 &&
6328 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6329 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6330 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6331 	}
6332 
6333 	return kvm_emulate_instruction(vcpu, emul_type);
6334 }
6335 EXPORT_SYMBOL_GPL(handle_ud);
6336 
6337 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6338 			    gpa_t gpa, bool write)
6339 {
6340 	/* For APIC access vmexit */
6341 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6342 		return 1;
6343 
6344 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6345 		trace_vcpu_match_mmio(gva, gpa, write, true);
6346 		return 1;
6347 	}
6348 
6349 	return 0;
6350 }
6351 
6352 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6353 				gpa_t *gpa, struct x86_exception *exception,
6354 				bool write)
6355 {
6356 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6357 		| (write ? PFERR_WRITE_MASK : 0);
6358 
6359 	/*
6360 	 * currently PKRU is only applied to ept enabled guest so
6361 	 * there is no pkey in EPT page table for L1 guest or EPT
6362 	 * shadow page table for L2 guest.
6363 	 */
6364 	if (vcpu_match_mmio_gva(vcpu, gva)
6365 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6366 				 vcpu->arch.mmio_access, 0, access)) {
6367 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6368 					(gva & (PAGE_SIZE - 1));
6369 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6370 		return 1;
6371 	}
6372 
6373 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6374 
6375 	if (*gpa == UNMAPPED_GVA)
6376 		return -1;
6377 
6378 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6379 }
6380 
6381 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6382 			const void *val, int bytes)
6383 {
6384 	int ret;
6385 
6386 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6387 	if (ret < 0)
6388 		return 0;
6389 	kvm_page_track_write(vcpu, gpa, val, bytes);
6390 	return 1;
6391 }
6392 
6393 struct read_write_emulator_ops {
6394 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6395 				  int bytes);
6396 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6397 				  void *val, int bytes);
6398 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6399 			       int bytes, void *val);
6400 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6401 				    void *val, int bytes);
6402 	bool write;
6403 };
6404 
6405 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6406 {
6407 	if (vcpu->mmio_read_completed) {
6408 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6409 			       vcpu->mmio_fragments[0].gpa, val);
6410 		vcpu->mmio_read_completed = 0;
6411 		return 1;
6412 	}
6413 
6414 	return 0;
6415 }
6416 
6417 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6418 			void *val, int bytes)
6419 {
6420 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6421 }
6422 
6423 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6424 			 void *val, int bytes)
6425 {
6426 	return emulator_write_phys(vcpu, gpa, val, bytes);
6427 }
6428 
6429 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6430 {
6431 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6432 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6433 }
6434 
6435 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6436 			  void *val, int bytes)
6437 {
6438 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6439 	return X86EMUL_IO_NEEDED;
6440 }
6441 
6442 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6443 			   void *val, int bytes)
6444 {
6445 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6446 
6447 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6448 	return X86EMUL_CONTINUE;
6449 }
6450 
6451 static const struct read_write_emulator_ops read_emultor = {
6452 	.read_write_prepare = read_prepare,
6453 	.read_write_emulate = read_emulate,
6454 	.read_write_mmio = vcpu_mmio_read,
6455 	.read_write_exit_mmio = read_exit_mmio,
6456 };
6457 
6458 static const struct read_write_emulator_ops write_emultor = {
6459 	.read_write_emulate = write_emulate,
6460 	.read_write_mmio = write_mmio,
6461 	.read_write_exit_mmio = write_exit_mmio,
6462 	.write = true,
6463 };
6464 
6465 static int emulator_read_write_onepage(unsigned long addr, void *val,
6466 				       unsigned int bytes,
6467 				       struct x86_exception *exception,
6468 				       struct kvm_vcpu *vcpu,
6469 				       const struct read_write_emulator_ops *ops)
6470 {
6471 	gpa_t gpa;
6472 	int handled, ret;
6473 	bool write = ops->write;
6474 	struct kvm_mmio_fragment *frag;
6475 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6476 
6477 	/*
6478 	 * If the exit was due to a NPF we may already have a GPA.
6479 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6480 	 * Note, this cannot be used on string operations since string
6481 	 * operation using rep will only have the initial GPA from the NPF
6482 	 * occurred.
6483 	 */
6484 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6485 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6486 		gpa = ctxt->gpa_val;
6487 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6488 	} else {
6489 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6490 		if (ret < 0)
6491 			return X86EMUL_PROPAGATE_FAULT;
6492 	}
6493 
6494 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6495 		return X86EMUL_CONTINUE;
6496 
6497 	/*
6498 	 * Is this MMIO handled locally?
6499 	 */
6500 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6501 	if (handled == bytes)
6502 		return X86EMUL_CONTINUE;
6503 
6504 	gpa += handled;
6505 	bytes -= handled;
6506 	val += handled;
6507 
6508 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6509 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6510 	frag->gpa = gpa;
6511 	frag->data = val;
6512 	frag->len = bytes;
6513 	return X86EMUL_CONTINUE;
6514 }
6515 
6516 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6517 			unsigned long addr,
6518 			void *val, unsigned int bytes,
6519 			struct x86_exception *exception,
6520 			const struct read_write_emulator_ops *ops)
6521 {
6522 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6523 	gpa_t gpa;
6524 	int rc;
6525 
6526 	if (ops->read_write_prepare &&
6527 		  ops->read_write_prepare(vcpu, val, bytes))
6528 		return X86EMUL_CONTINUE;
6529 
6530 	vcpu->mmio_nr_fragments = 0;
6531 
6532 	/* Crossing a page boundary? */
6533 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6534 		int now;
6535 
6536 		now = -addr & ~PAGE_MASK;
6537 		rc = emulator_read_write_onepage(addr, val, now, exception,
6538 						 vcpu, ops);
6539 
6540 		if (rc != X86EMUL_CONTINUE)
6541 			return rc;
6542 		addr += now;
6543 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6544 			addr = (u32)addr;
6545 		val += now;
6546 		bytes -= now;
6547 	}
6548 
6549 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6550 					 vcpu, ops);
6551 	if (rc != X86EMUL_CONTINUE)
6552 		return rc;
6553 
6554 	if (!vcpu->mmio_nr_fragments)
6555 		return rc;
6556 
6557 	gpa = vcpu->mmio_fragments[0].gpa;
6558 
6559 	vcpu->mmio_needed = 1;
6560 	vcpu->mmio_cur_fragment = 0;
6561 
6562 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6563 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6564 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6565 	vcpu->run->mmio.phys_addr = gpa;
6566 
6567 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6568 }
6569 
6570 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6571 				  unsigned long addr,
6572 				  void *val,
6573 				  unsigned int bytes,
6574 				  struct x86_exception *exception)
6575 {
6576 	return emulator_read_write(ctxt, addr, val, bytes,
6577 				   exception, &read_emultor);
6578 }
6579 
6580 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6581 			    unsigned long addr,
6582 			    const void *val,
6583 			    unsigned int bytes,
6584 			    struct x86_exception *exception)
6585 {
6586 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6587 				   exception, &write_emultor);
6588 }
6589 
6590 #define CMPXCHG_TYPE(t, ptr, old, new) \
6591 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6592 
6593 #ifdef CONFIG_X86_64
6594 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6595 #else
6596 #  define CMPXCHG64(ptr, old, new) \
6597 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6598 #endif
6599 
6600 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6601 				     unsigned long addr,
6602 				     const void *old,
6603 				     const void *new,
6604 				     unsigned int bytes,
6605 				     struct x86_exception *exception)
6606 {
6607 	struct kvm_host_map map;
6608 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6609 	u64 page_line_mask;
6610 	gpa_t gpa;
6611 	char *kaddr;
6612 	bool exchanged;
6613 
6614 	/* guests cmpxchg8b have to be emulated atomically */
6615 	if (bytes > 8 || (bytes & (bytes - 1)))
6616 		goto emul_write;
6617 
6618 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6619 
6620 	if (gpa == UNMAPPED_GVA ||
6621 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6622 		goto emul_write;
6623 
6624 	/*
6625 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6626 	 * enabled in the host and the access splits a cache line.
6627 	 */
6628 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6629 		page_line_mask = ~(cache_line_size() - 1);
6630 	else
6631 		page_line_mask = PAGE_MASK;
6632 
6633 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6634 		goto emul_write;
6635 
6636 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6637 		goto emul_write;
6638 
6639 	kaddr = map.hva + offset_in_page(gpa);
6640 
6641 	switch (bytes) {
6642 	case 1:
6643 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6644 		break;
6645 	case 2:
6646 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6647 		break;
6648 	case 4:
6649 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6650 		break;
6651 	case 8:
6652 		exchanged = CMPXCHG64(kaddr, old, new);
6653 		break;
6654 	default:
6655 		BUG();
6656 	}
6657 
6658 	kvm_vcpu_unmap(vcpu, &map, true);
6659 
6660 	if (!exchanged)
6661 		return X86EMUL_CMPXCHG_FAILED;
6662 
6663 	kvm_page_track_write(vcpu, gpa, new, bytes);
6664 
6665 	return X86EMUL_CONTINUE;
6666 
6667 emul_write:
6668 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6669 
6670 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6671 }
6672 
6673 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6674 {
6675 	int r = 0, i;
6676 
6677 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6678 		if (vcpu->arch.pio.in)
6679 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6680 					    vcpu->arch.pio.size, pd);
6681 		else
6682 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6683 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6684 					     pd);
6685 		if (r)
6686 			break;
6687 		pd += vcpu->arch.pio.size;
6688 	}
6689 	return r;
6690 }
6691 
6692 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6693 			       unsigned short port, void *val,
6694 			       unsigned int count, bool in)
6695 {
6696 	vcpu->arch.pio.port = port;
6697 	vcpu->arch.pio.in = in;
6698 	vcpu->arch.pio.count  = count;
6699 	vcpu->arch.pio.size = size;
6700 
6701 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6702 		vcpu->arch.pio.count = 0;
6703 		return 1;
6704 	}
6705 
6706 	vcpu->run->exit_reason = KVM_EXIT_IO;
6707 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6708 	vcpu->run->io.size = size;
6709 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6710 	vcpu->run->io.count = count;
6711 	vcpu->run->io.port = port;
6712 
6713 	return 0;
6714 }
6715 
6716 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6717 			   unsigned short port, void *val, unsigned int count)
6718 {
6719 	int ret;
6720 
6721 	if (vcpu->arch.pio.count)
6722 		goto data_avail;
6723 
6724 	memset(vcpu->arch.pio_data, 0, size * count);
6725 
6726 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6727 	if (ret) {
6728 data_avail:
6729 		memcpy(val, vcpu->arch.pio_data, size * count);
6730 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6731 		vcpu->arch.pio.count = 0;
6732 		return 1;
6733 	}
6734 
6735 	return 0;
6736 }
6737 
6738 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6739 				    int size, unsigned short port, void *val,
6740 				    unsigned int count)
6741 {
6742 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6743 
6744 }
6745 
6746 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6747 			    unsigned short port, const void *val,
6748 			    unsigned int count)
6749 {
6750 	memcpy(vcpu->arch.pio_data, val, size * count);
6751 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6752 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6753 }
6754 
6755 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6756 				     int size, unsigned short port,
6757 				     const void *val, unsigned int count)
6758 {
6759 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6760 }
6761 
6762 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6763 {
6764 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6765 }
6766 
6767 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6768 {
6769 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6770 }
6771 
6772 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6773 {
6774 	if (!need_emulate_wbinvd(vcpu))
6775 		return X86EMUL_CONTINUE;
6776 
6777 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
6778 		int cpu = get_cpu();
6779 
6780 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6781 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6782 				wbinvd_ipi, NULL, 1);
6783 		put_cpu();
6784 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6785 	} else
6786 		wbinvd();
6787 	return X86EMUL_CONTINUE;
6788 }
6789 
6790 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6791 {
6792 	kvm_emulate_wbinvd_noskip(vcpu);
6793 	return kvm_skip_emulated_instruction(vcpu);
6794 }
6795 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6796 
6797 
6798 
6799 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6800 {
6801 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6802 }
6803 
6804 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6805 			    unsigned long *dest)
6806 {
6807 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6808 }
6809 
6810 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6811 			   unsigned long value)
6812 {
6813 
6814 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6815 }
6816 
6817 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6818 {
6819 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6820 }
6821 
6822 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6823 {
6824 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6825 	unsigned long value;
6826 
6827 	switch (cr) {
6828 	case 0:
6829 		value = kvm_read_cr0(vcpu);
6830 		break;
6831 	case 2:
6832 		value = vcpu->arch.cr2;
6833 		break;
6834 	case 3:
6835 		value = kvm_read_cr3(vcpu);
6836 		break;
6837 	case 4:
6838 		value = kvm_read_cr4(vcpu);
6839 		break;
6840 	case 8:
6841 		value = kvm_get_cr8(vcpu);
6842 		break;
6843 	default:
6844 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6845 		return 0;
6846 	}
6847 
6848 	return value;
6849 }
6850 
6851 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6852 {
6853 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6854 	int res = 0;
6855 
6856 	switch (cr) {
6857 	case 0:
6858 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6859 		break;
6860 	case 2:
6861 		vcpu->arch.cr2 = val;
6862 		break;
6863 	case 3:
6864 		res = kvm_set_cr3(vcpu, val);
6865 		break;
6866 	case 4:
6867 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6868 		break;
6869 	case 8:
6870 		res = kvm_set_cr8(vcpu, val);
6871 		break;
6872 	default:
6873 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6874 		res = -1;
6875 	}
6876 
6877 	return res;
6878 }
6879 
6880 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6881 {
6882 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6883 }
6884 
6885 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6886 {
6887 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6888 }
6889 
6890 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6891 {
6892 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6893 }
6894 
6895 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6896 {
6897 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6898 }
6899 
6900 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6901 {
6902 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6903 }
6904 
6905 static unsigned long emulator_get_cached_segment_base(
6906 	struct x86_emulate_ctxt *ctxt, int seg)
6907 {
6908 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6909 }
6910 
6911 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6912 				 struct desc_struct *desc, u32 *base3,
6913 				 int seg)
6914 {
6915 	struct kvm_segment var;
6916 
6917 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6918 	*selector = var.selector;
6919 
6920 	if (var.unusable) {
6921 		memset(desc, 0, sizeof(*desc));
6922 		if (base3)
6923 			*base3 = 0;
6924 		return false;
6925 	}
6926 
6927 	if (var.g)
6928 		var.limit >>= 12;
6929 	set_desc_limit(desc, var.limit);
6930 	set_desc_base(desc, (unsigned long)var.base);
6931 #ifdef CONFIG_X86_64
6932 	if (base3)
6933 		*base3 = var.base >> 32;
6934 #endif
6935 	desc->type = var.type;
6936 	desc->s = var.s;
6937 	desc->dpl = var.dpl;
6938 	desc->p = var.present;
6939 	desc->avl = var.avl;
6940 	desc->l = var.l;
6941 	desc->d = var.db;
6942 	desc->g = var.g;
6943 
6944 	return true;
6945 }
6946 
6947 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6948 				 struct desc_struct *desc, u32 base3,
6949 				 int seg)
6950 {
6951 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6952 	struct kvm_segment var;
6953 
6954 	var.selector = selector;
6955 	var.base = get_desc_base(desc);
6956 #ifdef CONFIG_X86_64
6957 	var.base |= ((u64)base3) << 32;
6958 #endif
6959 	var.limit = get_desc_limit(desc);
6960 	if (desc->g)
6961 		var.limit = (var.limit << 12) | 0xfff;
6962 	var.type = desc->type;
6963 	var.dpl = desc->dpl;
6964 	var.db = desc->d;
6965 	var.s = desc->s;
6966 	var.l = desc->l;
6967 	var.g = desc->g;
6968 	var.avl = desc->avl;
6969 	var.present = desc->p;
6970 	var.unusable = !var.present;
6971 	var.padding = 0;
6972 
6973 	kvm_set_segment(vcpu, &var, seg);
6974 	return;
6975 }
6976 
6977 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6978 			    u32 msr_index, u64 *pdata)
6979 {
6980 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6981 	int r;
6982 
6983 	r = kvm_get_msr(vcpu, msr_index, pdata);
6984 
6985 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6986 		/* Bounce to user space */
6987 		return X86EMUL_IO_NEEDED;
6988 	}
6989 
6990 	return r;
6991 }
6992 
6993 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6994 			    u32 msr_index, u64 data)
6995 {
6996 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6997 	int r;
6998 
6999 	r = kvm_set_msr(vcpu, msr_index, data);
7000 
7001 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7002 		/* Bounce to user space */
7003 		return X86EMUL_IO_NEEDED;
7004 	}
7005 
7006 	return r;
7007 }
7008 
7009 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7010 {
7011 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7012 
7013 	return vcpu->arch.smbase;
7014 }
7015 
7016 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7017 {
7018 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7019 
7020 	vcpu->arch.smbase = smbase;
7021 }
7022 
7023 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7024 			      u32 pmc)
7025 {
7026 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7027 }
7028 
7029 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7030 			     u32 pmc, u64 *pdata)
7031 {
7032 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7033 }
7034 
7035 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7036 {
7037 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7038 }
7039 
7040 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7041 			      struct x86_instruction_info *info,
7042 			      enum x86_intercept_stage stage)
7043 {
7044 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7045 					    &ctxt->exception);
7046 }
7047 
7048 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7049 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7050 			      bool exact_only)
7051 {
7052 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7053 }
7054 
7055 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7056 {
7057 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7058 }
7059 
7060 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7061 {
7062 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7063 }
7064 
7065 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7066 {
7067 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7068 }
7069 
7070 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7071 {
7072 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7073 }
7074 
7075 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7076 {
7077 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7078 }
7079 
7080 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7081 {
7082 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7083 }
7084 
7085 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7086 {
7087 	return emul_to_vcpu(ctxt)->arch.hflags;
7088 }
7089 
7090 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7091 {
7092 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
7093 }
7094 
7095 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7096 				  const char *smstate)
7097 {
7098 	return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7099 }
7100 
7101 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7102 {
7103 	kvm_smm_changed(emul_to_vcpu(ctxt));
7104 }
7105 
7106 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7107 {
7108 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7109 }
7110 
7111 static const struct x86_emulate_ops emulate_ops = {
7112 	.read_gpr            = emulator_read_gpr,
7113 	.write_gpr           = emulator_write_gpr,
7114 	.read_std            = emulator_read_std,
7115 	.write_std           = emulator_write_std,
7116 	.read_phys           = kvm_read_guest_phys_system,
7117 	.fetch               = kvm_fetch_guest_virt,
7118 	.read_emulated       = emulator_read_emulated,
7119 	.write_emulated      = emulator_write_emulated,
7120 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7121 	.invlpg              = emulator_invlpg,
7122 	.pio_in_emulated     = emulator_pio_in_emulated,
7123 	.pio_out_emulated    = emulator_pio_out_emulated,
7124 	.get_segment         = emulator_get_segment,
7125 	.set_segment         = emulator_set_segment,
7126 	.get_cached_segment_base = emulator_get_cached_segment_base,
7127 	.get_gdt             = emulator_get_gdt,
7128 	.get_idt	     = emulator_get_idt,
7129 	.set_gdt             = emulator_set_gdt,
7130 	.set_idt	     = emulator_set_idt,
7131 	.get_cr              = emulator_get_cr,
7132 	.set_cr              = emulator_set_cr,
7133 	.cpl                 = emulator_get_cpl,
7134 	.get_dr              = emulator_get_dr,
7135 	.set_dr              = emulator_set_dr,
7136 	.get_smbase          = emulator_get_smbase,
7137 	.set_smbase          = emulator_set_smbase,
7138 	.set_msr             = emulator_set_msr,
7139 	.get_msr             = emulator_get_msr,
7140 	.check_pmc	     = emulator_check_pmc,
7141 	.read_pmc            = emulator_read_pmc,
7142 	.halt                = emulator_halt,
7143 	.wbinvd              = emulator_wbinvd,
7144 	.fix_hypercall       = emulator_fix_hypercall,
7145 	.intercept           = emulator_intercept,
7146 	.get_cpuid           = emulator_get_cpuid,
7147 	.guest_has_long_mode = emulator_guest_has_long_mode,
7148 	.guest_has_movbe     = emulator_guest_has_movbe,
7149 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7150 	.set_nmi_mask        = emulator_set_nmi_mask,
7151 	.get_hflags          = emulator_get_hflags,
7152 	.set_hflags          = emulator_set_hflags,
7153 	.pre_leave_smm       = emulator_pre_leave_smm,
7154 	.post_leave_smm      = emulator_post_leave_smm,
7155 	.set_xcr             = emulator_set_xcr,
7156 };
7157 
7158 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7159 {
7160 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7161 	/*
7162 	 * an sti; sti; sequence only disable interrupts for the first
7163 	 * instruction. So, if the last instruction, be it emulated or
7164 	 * not, left the system with the INT_STI flag enabled, it
7165 	 * means that the last instruction is an sti. We should not
7166 	 * leave the flag on in this case. The same goes for mov ss
7167 	 */
7168 	if (int_shadow & mask)
7169 		mask = 0;
7170 	if (unlikely(int_shadow || mask)) {
7171 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7172 		if (!mask)
7173 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7174 	}
7175 }
7176 
7177 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7178 {
7179 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7180 	if (ctxt->exception.vector == PF_VECTOR)
7181 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7182 
7183 	if (ctxt->exception.error_code_valid)
7184 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7185 				      ctxt->exception.error_code);
7186 	else
7187 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7188 	return false;
7189 }
7190 
7191 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7192 {
7193 	struct x86_emulate_ctxt *ctxt;
7194 
7195 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7196 	if (!ctxt) {
7197 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7198 		return NULL;
7199 	}
7200 
7201 	ctxt->vcpu = vcpu;
7202 	ctxt->ops = &emulate_ops;
7203 	vcpu->arch.emulate_ctxt = ctxt;
7204 
7205 	return ctxt;
7206 }
7207 
7208 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7209 {
7210 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7211 	int cs_db, cs_l;
7212 
7213 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7214 
7215 	ctxt->gpa_available = false;
7216 	ctxt->eflags = kvm_get_rflags(vcpu);
7217 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7218 
7219 	ctxt->eip = kvm_rip_read(vcpu);
7220 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7221 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7222 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7223 		     cs_db				? X86EMUL_MODE_PROT32 :
7224 							  X86EMUL_MODE_PROT16;
7225 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7226 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7227 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7228 
7229 	init_decode_cache(ctxt);
7230 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7231 }
7232 
7233 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7234 {
7235 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7236 	int ret;
7237 
7238 	init_emulate_ctxt(vcpu);
7239 
7240 	ctxt->op_bytes = 2;
7241 	ctxt->ad_bytes = 2;
7242 	ctxt->_eip = ctxt->eip + inc_eip;
7243 	ret = emulate_int_real(ctxt, irq);
7244 
7245 	if (ret != X86EMUL_CONTINUE) {
7246 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7247 	} else {
7248 		ctxt->eip = ctxt->_eip;
7249 		kvm_rip_write(vcpu, ctxt->eip);
7250 		kvm_set_rflags(vcpu, ctxt->eflags);
7251 	}
7252 }
7253 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7254 
7255 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7256 {
7257 	++vcpu->stat.insn_emulation_fail;
7258 	trace_kvm_emulate_insn_failed(vcpu);
7259 
7260 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7261 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7262 		return 1;
7263 	}
7264 
7265 	if (emulation_type & EMULTYPE_SKIP) {
7266 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7267 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7268 		vcpu->run->internal.ndata = 0;
7269 		return 0;
7270 	}
7271 
7272 	kvm_queue_exception(vcpu, UD_VECTOR);
7273 
7274 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7275 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7276 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7277 		vcpu->run->internal.ndata = 0;
7278 		return 0;
7279 	}
7280 
7281 	return 1;
7282 }
7283 
7284 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7285 				  bool write_fault_to_shadow_pgtable,
7286 				  int emulation_type)
7287 {
7288 	gpa_t gpa = cr2_or_gpa;
7289 	kvm_pfn_t pfn;
7290 
7291 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7292 		return false;
7293 
7294 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7295 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7296 		return false;
7297 
7298 	if (!vcpu->arch.mmu->direct_map) {
7299 		/*
7300 		 * Write permission should be allowed since only
7301 		 * write access need to be emulated.
7302 		 */
7303 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7304 
7305 		/*
7306 		 * If the mapping is invalid in guest, let cpu retry
7307 		 * it to generate fault.
7308 		 */
7309 		if (gpa == UNMAPPED_GVA)
7310 			return true;
7311 	}
7312 
7313 	/*
7314 	 * Do not retry the unhandleable instruction if it faults on the
7315 	 * readonly host memory, otherwise it will goto a infinite loop:
7316 	 * retry instruction -> write #PF -> emulation fail -> retry
7317 	 * instruction -> ...
7318 	 */
7319 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7320 
7321 	/*
7322 	 * If the instruction failed on the error pfn, it can not be fixed,
7323 	 * report the error to userspace.
7324 	 */
7325 	if (is_error_noslot_pfn(pfn))
7326 		return false;
7327 
7328 	kvm_release_pfn_clean(pfn);
7329 
7330 	/* The instructions are well-emulated on direct mmu. */
7331 	if (vcpu->arch.mmu->direct_map) {
7332 		unsigned int indirect_shadow_pages;
7333 
7334 		write_lock(&vcpu->kvm->mmu_lock);
7335 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7336 		write_unlock(&vcpu->kvm->mmu_lock);
7337 
7338 		if (indirect_shadow_pages)
7339 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7340 
7341 		return true;
7342 	}
7343 
7344 	/*
7345 	 * if emulation was due to access to shadowed page table
7346 	 * and it failed try to unshadow page and re-enter the
7347 	 * guest to let CPU execute the instruction.
7348 	 */
7349 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7350 
7351 	/*
7352 	 * If the access faults on its page table, it can not
7353 	 * be fixed by unprotecting shadow page and it should
7354 	 * be reported to userspace.
7355 	 */
7356 	return !write_fault_to_shadow_pgtable;
7357 }
7358 
7359 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7360 			      gpa_t cr2_or_gpa,  int emulation_type)
7361 {
7362 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7363 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7364 
7365 	last_retry_eip = vcpu->arch.last_retry_eip;
7366 	last_retry_addr = vcpu->arch.last_retry_addr;
7367 
7368 	/*
7369 	 * If the emulation is caused by #PF and it is non-page_table
7370 	 * writing instruction, it means the VM-EXIT is caused by shadow
7371 	 * page protected, we can zap the shadow page and retry this
7372 	 * instruction directly.
7373 	 *
7374 	 * Note: if the guest uses a non-page-table modifying instruction
7375 	 * on the PDE that points to the instruction, then we will unmap
7376 	 * the instruction and go to an infinite loop. So, we cache the
7377 	 * last retried eip and the last fault address, if we meet the eip
7378 	 * and the address again, we can break out of the potential infinite
7379 	 * loop.
7380 	 */
7381 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7382 
7383 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7384 		return false;
7385 
7386 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7387 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7388 		return false;
7389 
7390 	if (x86_page_table_writing_insn(ctxt))
7391 		return false;
7392 
7393 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7394 		return false;
7395 
7396 	vcpu->arch.last_retry_eip = ctxt->eip;
7397 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7398 
7399 	if (!vcpu->arch.mmu->direct_map)
7400 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7401 
7402 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7403 
7404 	return true;
7405 }
7406 
7407 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7408 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7409 
7410 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7411 {
7412 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7413 		/* This is a good place to trace that we are exiting SMM.  */
7414 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7415 
7416 		/* Process a latched INIT or SMI, if any.  */
7417 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7418 	}
7419 
7420 	kvm_mmu_reset_context(vcpu);
7421 }
7422 
7423 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7424 				unsigned long *db)
7425 {
7426 	u32 dr6 = 0;
7427 	int i;
7428 	u32 enable, rwlen;
7429 
7430 	enable = dr7;
7431 	rwlen = dr7 >> 16;
7432 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7433 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7434 			dr6 |= (1 << i);
7435 	return dr6;
7436 }
7437 
7438 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7439 {
7440 	struct kvm_run *kvm_run = vcpu->run;
7441 
7442 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7443 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7444 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7445 		kvm_run->debug.arch.exception = DB_VECTOR;
7446 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7447 		return 0;
7448 	}
7449 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7450 	return 1;
7451 }
7452 
7453 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7454 {
7455 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7456 	int r;
7457 
7458 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7459 	if (unlikely(!r))
7460 		return 0;
7461 
7462 	/*
7463 	 * rflags is the old, "raw" value of the flags.  The new value has
7464 	 * not been saved yet.
7465 	 *
7466 	 * This is correct even for TF set by the guest, because "the
7467 	 * processor will not generate this exception after the instruction
7468 	 * that sets the TF flag".
7469 	 */
7470 	if (unlikely(rflags & X86_EFLAGS_TF))
7471 		r = kvm_vcpu_do_singlestep(vcpu);
7472 	return r;
7473 }
7474 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7475 
7476 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7477 {
7478 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7479 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7480 		struct kvm_run *kvm_run = vcpu->run;
7481 		unsigned long eip = kvm_get_linear_rip(vcpu);
7482 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7483 					   vcpu->arch.guest_debug_dr7,
7484 					   vcpu->arch.eff_db);
7485 
7486 		if (dr6 != 0) {
7487 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7488 			kvm_run->debug.arch.pc = eip;
7489 			kvm_run->debug.arch.exception = DB_VECTOR;
7490 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7491 			*r = 0;
7492 			return true;
7493 		}
7494 	}
7495 
7496 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7497 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7498 		unsigned long eip = kvm_get_linear_rip(vcpu);
7499 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7500 					   vcpu->arch.dr7,
7501 					   vcpu->arch.db);
7502 
7503 		if (dr6 != 0) {
7504 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7505 			*r = 1;
7506 			return true;
7507 		}
7508 	}
7509 
7510 	return false;
7511 }
7512 
7513 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7514 {
7515 	switch (ctxt->opcode_len) {
7516 	case 1:
7517 		switch (ctxt->b) {
7518 		case 0xe4:	/* IN */
7519 		case 0xe5:
7520 		case 0xec:
7521 		case 0xed:
7522 		case 0xe6:	/* OUT */
7523 		case 0xe7:
7524 		case 0xee:
7525 		case 0xef:
7526 		case 0x6c:	/* INS */
7527 		case 0x6d:
7528 		case 0x6e:	/* OUTS */
7529 		case 0x6f:
7530 			return true;
7531 		}
7532 		break;
7533 	case 2:
7534 		switch (ctxt->b) {
7535 		case 0x33:	/* RDPMC */
7536 			return true;
7537 		}
7538 		break;
7539 	}
7540 
7541 	return false;
7542 }
7543 
7544 /*
7545  * Decode to be emulated instruction. Return EMULATION_OK if success.
7546  */
7547 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7548 				    void *insn, int insn_len)
7549 {
7550 	int r = EMULATION_OK;
7551 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7552 
7553 	init_emulate_ctxt(vcpu);
7554 
7555 	/*
7556 	 * We will reenter on the same instruction since we do not set
7557 	 * complete_userspace_io. This does not handle watchpoints yet,
7558 	 * those would be handled in the emulate_ops.
7559 	 */
7560 	if (!(emulation_type & EMULTYPE_SKIP) &&
7561 	    kvm_vcpu_check_breakpoint(vcpu, &r))
7562 		return r;
7563 
7564 	ctxt->interruptibility = 0;
7565 	ctxt->have_exception = false;
7566 	ctxt->exception.vector = -1;
7567 	ctxt->perm_ok = false;
7568 
7569 	ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7570 
7571 	r = x86_decode_insn(ctxt, insn, insn_len);
7572 
7573 	trace_kvm_emulate_insn_start(vcpu);
7574 	++vcpu->stat.insn_emulation;
7575 
7576 	return r;
7577 }
7578 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7579 
7580 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7581 			    int emulation_type, void *insn, int insn_len)
7582 {
7583 	int r;
7584 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7585 	bool writeback = true;
7586 	bool write_fault_to_spt;
7587 
7588 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7589 		return 1;
7590 
7591 	vcpu->arch.l1tf_flush_l1d = true;
7592 
7593 	/*
7594 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7595 	 * never reused.
7596 	 */
7597 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7598 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7599 
7600 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7601 		kvm_clear_exception_queue(vcpu);
7602 
7603 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
7604 						    insn, insn_len);
7605 		if (r != EMULATION_OK)  {
7606 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7607 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7608 				kvm_queue_exception(vcpu, UD_VECTOR);
7609 				return 1;
7610 			}
7611 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7612 						  write_fault_to_spt,
7613 						  emulation_type))
7614 				return 1;
7615 			if (ctxt->have_exception) {
7616 				/*
7617 				 * #UD should result in just EMULATION_FAILED, and trap-like
7618 				 * exception should not be encountered during decode.
7619 				 */
7620 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7621 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7622 				inject_emulated_exception(vcpu);
7623 				return 1;
7624 			}
7625 			return handle_emulation_failure(vcpu, emulation_type);
7626 		}
7627 	}
7628 
7629 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7630 	    !is_vmware_backdoor_opcode(ctxt)) {
7631 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7632 		return 1;
7633 	}
7634 
7635 	/*
7636 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7637 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7638 	 * updating interruptibility state and injecting single-step #DBs.
7639 	 */
7640 	if (emulation_type & EMULTYPE_SKIP) {
7641 		kvm_rip_write(vcpu, ctxt->_eip);
7642 		if (ctxt->eflags & X86_EFLAGS_RF)
7643 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7644 		return 1;
7645 	}
7646 
7647 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7648 		return 1;
7649 
7650 	/* this is needed for vmware backdoor interface to work since it
7651 	   changes registers values  during IO operation */
7652 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7653 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7654 		emulator_invalidate_register_cache(ctxt);
7655 	}
7656 
7657 restart:
7658 	if (emulation_type & EMULTYPE_PF) {
7659 		/* Save the faulting GPA (cr2) in the address field */
7660 		ctxt->exception.address = cr2_or_gpa;
7661 
7662 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7663 		if (vcpu->arch.mmu->direct_map) {
7664 			ctxt->gpa_available = true;
7665 			ctxt->gpa_val = cr2_or_gpa;
7666 		}
7667 	} else {
7668 		/* Sanitize the address out of an abundance of paranoia. */
7669 		ctxt->exception.address = 0;
7670 	}
7671 
7672 	r = x86_emulate_insn(ctxt);
7673 
7674 	if (r == EMULATION_INTERCEPTED)
7675 		return 1;
7676 
7677 	if (r == EMULATION_FAILED) {
7678 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7679 					emulation_type))
7680 			return 1;
7681 
7682 		return handle_emulation_failure(vcpu, emulation_type);
7683 	}
7684 
7685 	if (ctxt->have_exception) {
7686 		r = 1;
7687 		if (inject_emulated_exception(vcpu))
7688 			return r;
7689 	} else if (vcpu->arch.pio.count) {
7690 		if (!vcpu->arch.pio.in) {
7691 			/* FIXME: return into emulator if single-stepping.  */
7692 			vcpu->arch.pio.count = 0;
7693 		} else {
7694 			writeback = false;
7695 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7696 		}
7697 		r = 0;
7698 	} else if (vcpu->mmio_needed) {
7699 		++vcpu->stat.mmio_exits;
7700 
7701 		if (!vcpu->mmio_is_write)
7702 			writeback = false;
7703 		r = 0;
7704 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7705 	} else if (r == EMULATION_RESTART)
7706 		goto restart;
7707 	else
7708 		r = 1;
7709 
7710 	if (writeback) {
7711 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7712 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7713 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7714 		if (!ctxt->have_exception ||
7715 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7716 			kvm_rip_write(vcpu, ctxt->eip);
7717 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7718 				r = kvm_vcpu_do_singlestep(vcpu);
7719 			if (kvm_x86_ops.update_emulated_instruction)
7720 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
7721 			__kvm_set_rflags(vcpu, ctxt->eflags);
7722 		}
7723 
7724 		/*
7725 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7726 		 * do nothing, and it will be requested again as soon as
7727 		 * the shadow expires.  But we still need to check here,
7728 		 * because POPF has no interrupt shadow.
7729 		 */
7730 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7731 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7732 	} else
7733 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7734 
7735 	return r;
7736 }
7737 
7738 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7739 {
7740 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7741 }
7742 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7743 
7744 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7745 					void *insn, int insn_len)
7746 {
7747 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7748 }
7749 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7750 
7751 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7752 {
7753 	vcpu->arch.pio.count = 0;
7754 	return 1;
7755 }
7756 
7757 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7758 {
7759 	vcpu->arch.pio.count = 0;
7760 
7761 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7762 		return 1;
7763 
7764 	return kvm_skip_emulated_instruction(vcpu);
7765 }
7766 
7767 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7768 			    unsigned short port)
7769 {
7770 	unsigned long val = kvm_rax_read(vcpu);
7771 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7772 
7773 	if (ret)
7774 		return ret;
7775 
7776 	/*
7777 	 * Workaround userspace that relies on old KVM behavior of %rip being
7778 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7779 	 */
7780 	if (port == 0x7e &&
7781 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7782 		vcpu->arch.complete_userspace_io =
7783 			complete_fast_pio_out_port_0x7e;
7784 		kvm_skip_emulated_instruction(vcpu);
7785 	} else {
7786 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7787 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7788 	}
7789 	return 0;
7790 }
7791 
7792 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7793 {
7794 	unsigned long val;
7795 
7796 	/* We should only ever be called with arch.pio.count equal to 1 */
7797 	BUG_ON(vcpu->arch.pio.count != 1);
7798 
7799 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7800 		vcpu->arch.pio.count = 0;
7801 		return 1;
7802 	}
7803 
7804 	/* For size less than 4 we merge, else we zero extend */
7805 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7806 
7807 	/*
7808 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7809 	 * the copy and tracing
7810 	 */
7811 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7812 	kvm_rax_write(vcpu, val);
7813 
7814 	return kvm_skip_emulated_instruction(vcpu);
7815 }
7816 
7817 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7818 			   unsigned short port)
7819 {
7820 	unsigned long val;
7821 	int ret;
7822 
7823 	/* For size less than 4 we merge, else we zero extend */
7824 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7825 
7826 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7827 	if (ret) {
7828 		kvm_rax_write(vcpu, val);
7829 		return ret;
7830 	}
7831 
7832 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7833 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7834 
7835 	return 0;
7836 }
7837 
7838 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7839 {
7840 	int ret;
7841 
7842 	if (in)
7843 		ret = kvm_fast_pio_in(vcpu, size, port);
7844 	else
7845 		ret = kvm_fast_pio_out(vcpu, size, port);
7846 	return ret && kvm_skip_emulated_instruction(vcpu);
7847 }
7848 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7849 
7850 static int kvmclock_cpu_down_prep(unsigned int cpu)
7851 {
7852 	__this_cpu_write(cpu_tsc_khz, 0);
7853 	return 0;
7854 }
7855 
7856 static void tsc_khz_changed(void *data)
7857 {
7858 	struct cpufreq_freqs *freq = data;
7859 	unsigned long khz = 0;
7860 
7861 	if (data)
7862 		khz = freq->new;
7863 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7864 		khz = cpufreq_quick_get(raw_smp_processor_id());
7865 	if (!khz)
7866 		khz = tsc_khz;
7867 	__this_cpu_write(cpu_tsc_khz, khz);
7868 }
7869 
7870 #ifdef CONFIG_X86_64
7871 static void kvm_hyperv_tsc_notifier(void)
7872 {
7873 	struct kvm *kvm;
7874 	struct kvm_vcpu *vcpu;
7875 	int cpu;
7876 	unsigned long flags;
7877 
7878 	mutex_lock(&kvm_lock);
7879 	list_for_each_entry(kvm, &vm_list, vm_list)
7880 		kvm_make_mclock_inprogress_request(kvm);
7881 
7882 	hyperv_stop_tsc_emulation();
7883 
7884 	/* TSC frequency always matches when on Hyper-V */
7885 	for_each_present_cpu(cpu)
7886 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7887 	kvm_max_guest_tsc_khz = tsc_khz;
7888 
7889 	list_for_each_entry(kvm, &vm_list, vm_list) {
7890 		struct kvm_arch *ka = &kvm->arch;
7891 
7892 		spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7893 		pvclock_update_vm_gtod_copy(kvm);
7894 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7895 
7896 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7897 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7898 
7899 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7900 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7901 	}
7902 	mutex_unlock(&kvm_lock);
7903 }
7904 #endif
7905 
7906 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7907 {
7908 	struct kvm *kvm;
7909 	struct kvm_vcpu *vcpu;
7910 	int i, send_ipi = 0;
7911 
7912 	/*
7913 	 * We allow guests to temporarily run on slowing clocks,
7914 	 * provided we notify them after, or to run on accelerating
7915 	 * clocks, provided we notify them before.  Thus time never
7916 	 * goes backwards.
7917 	 *
7918 	 * However, we have a problem.  We can't atomically update
7919 	 * the frequency of a given CPU from this function; it is
7920 	 * merely a notifier, which can be called from any CPU.
7921 	 * Changing the TSC frequency at arbitrary points in time
7922 	 * requires a recomputation of local variables related to
7923 	 * the TSC for each VCPU.  We must flag these local variables
7924 	 * to be updated and be sure the update takes place with the
7925 	 * new frequency before any guests proceed.
7926 	 *
7927 	 * Unfortunately, the combination of hotplug CPU and frequency
7928 	 * change creates an intractable locking scenario; the order
7929 	 * of when these callouts happen is undefined with respect to
7930 	 * CPU hotplug, and they can race with each other.  As such,
7931 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7932 	 * undefined; you can actually have a CPU frequency change take
7933 	 * place in between the computation of X and the setting of the
7934 	 * variable.  To protect against this problem, all updates of
7935 	 * the per_cpu tsc_khz variable are done in an interrupt
7936 	 * protected IPI, and all callers wishing to update the value
7937 	 * must wait for a synchronous IPI to complete (which is trivial
7938 	 * if the caller is on the CPU already).  This establishes the
7939 	 * necessary total order on variable updates.
7940 	 *
7941 	 * Note that because a guest time update may take place
7942 	 * anytime after the setting of the VCPU's request bit, the
7943 	 * correct TSC value must be set before the request.  However,
7944 	 * to ensure the update actually makes it to any guest which
7945 	 * starts running in hardware virtualization between the set
7946 	 * and the acquisition of the spinlock, we must also ping the
7947 	 * CPU after setting the request bit.
7948 	 *
7949 	 */
7950 
7951 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7952 
7953 	mutex_lock(&kvm_lock);
7954 	list_for_each_entry(kvm, &vm_list, vm_list) {
7955 		kvm_for_each_vcpu(i, vcpu, kvm) {
7956 			if (vcpu->cpu != cpu)
7957 				continue;
7958 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7959 			if (vcpu->cpu != raw_smp_processor_id())
7960 				send_ipi = 1;
7961 		}
7962 	}
7963 	mutex_unlock(&kvm_lock);
7964 
7965 	if (freq->old < freq->new && send_ipi) {
7966 		/*
7967 		 * We upscale the frequency.  Must make the guest
7968 		 * doesn't see old kvmclock values while running with
7969 		 * the new frequency, otherwise we risk the guest sees
7970 		 * time go backwards.
7971 		 *
7972 		 * In case we update the frequency for another cpu
7973 		 * (which might be in guest context) send an interrupt
7974 		 * to kick the cpu out of guest context.  Next time
7975 		 * guest context is entered kvmclock will be updated,
7976 		 * so the guest will not see stale values.
7977 		 */
7978 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7979 	}
7980 }
7981 
7982 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7983 				     void *data)
7984 {
7985 	struct cpufreq_freqs *freq = data;
7986 	int cpu;
7987 
7988 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7989 		return 0;
7990 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7991 		return 0;
7992 
7993 	for_each_cpu(cpu, freq->policy->cpus)
7994 		__kvmclock_cpufreq_notifier(freq, cpu);
7995 
7996 	return 0;
7997 }
7998 
7999 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8000 	.notifier_call  = kvmclock_cpufreq_notifier
8001 };
8002 
8003 static int kvmclock_cpu_online(unsigned int cpu)
8004 {
8005 	tsc_khz_changed(NULL);
8006 	return 0;
8007 }
8008 
8009 static void kvm_timer_init(void)
8010 {
8011 	max_tsc_khz = tsc_khz;
8012 
8013 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8014 #ifdef CONFIG_CPU_FREQ
8015 		struct cpufreq_policy *policy;
8016 		int cpu;
8017 
8018 		cpu = get_cpu();
8019 		policy = cpufreq_cpu_get(cpu);
8020 		if (policy) {
8021 			if (policy->cpuinfo.max_freq)
8022 				max_tsc_khz = policy->cpuinfo.max_freq;
8023 			cpufreq_cpu_put(policy);
8024 		}
8025 		put_cpu();
8026 #endif
8027 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8028 					  CPUFREQ_TRANSITION_NOTIFIER);
8029 	}
8030 
8031 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8032 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8033 }
8034 
8035 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8036 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8037 
8038 int kvm_is_in_guest(void)
8039 {
8040 	return __this_cpu_read(current_vcpu) != NULL;
8041 }
8042 
8043 static int kvm_is_user_mode(void)
8044 {
8045 	int user_mode = 3;
8046 
8047 	if (__this_cpu_read(current_vcpu))
8048 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8049 
8050 	return user_mode != 0;
8051 }
8052 
8053 static unsigned long kvm_get_guest_ip(void)
8054 {
8055 	unsigned long ip = 0;
8056 
8057 	if (__this_cpu_read(current_vcpu))
8058 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8059 
8060 	return ip;
8061 }
8062 
8063 static void kvm_handle_intel_pt_intr(void)
8064 {
8065 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8066 
8067 	kvm_make_request(KVM_REQ_PMI, vcpu);
8068 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8069 			(unsigned long *)&vcpu->arch.pmu.global_status);
8070 }
8071 
8072 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8073 	.is_in_guest		= kvm_is_in_guest,
8074 	.is_user_mode		= kvm_is_user_mode,
8075 	.get_guest_ip		= kvm_get_guest_ip,
8076 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8077 };
8078 
8079 #ifdef CONFIG_X86_64
8080 static void pvclock_gtod_update_fn(struct work_struct *work)
8081 {
8082 	struct kvm *kvm;
8083 
8084 	struct kvm_vcpu *vcpu;
8085 	int i;
8086 
8087 	mutex_lock(&kvm_lock);
8088 	list_for_each_entry(kvm, &vm_list, vm_list)
8089 		kvm_for_each_vcpu(i, vcpu, kvm)
8090 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8091 	atomic_set(&kvm_guest_has_master_clock, 0);
8092 	mutex_unlock(&kvm_lock);
8093 }
8094 
8095 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8096 
8097 /*
8098  * Indirection to move queue_work() out of the tk_core.seq write held
8099  * region to prevent possible deadlocks against time accessors which
8100  * are invoked with work related locks held.
8101  */
8102 static void pvclock_irq_work_fn(struct irq_work *w)
8103 {
8104 	queue_work(system_long_wq, &pvclock_gtod_work);
8105 }
8106 
8107 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8108 
8109 /*
8110  * Notification about pvclock gtod data update.
8111  */
8112 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8113 			       void *priv)
8114 {
8115 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8116 	struct timekeeper *tk = priv;
8117 
8118 	update_pvclock_gtod(tk);
8119 
8120 	/*
8121 	 * Disable master clock if host does not trust, or does not use,
8122 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8123 	 * this is invoked with tk_core.seq write held.
8124 	 */
8125 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8126 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8127 		irq_work_queue(&pvclock_irq_work);
8128 	return 0;
8129 }
8130 
8131 static struct notifier_block pvclock_gtod_notifier = {
8132 	.notifier_call = pvclock_gtod_notify,
8133 };
8134 #endif
8135 
8136 int kvm_arch_init(void *opaque)
8137 {
8138 	struct kvm_x86_init_ops *ops = opaque;
8139 	int r;
8140 
8141 	if (kvm_x86_ops.hardware_enable) {
8142 		printk(KERN_ERR "kvm: already loaded the other module\n");
8143 		r = -EEXIST;
8144 		goto out;
8145 	}
8146 
8147 	if (!ops->cpu_has_kvm_support()) {
8148 		pr_err_ratelimited("kvm: no hardware support\n");
8149 		r = -EOPNOTSUPP;
8150 		goto out;
8151 	}
8152 	if (ops->disabled_by_bios()) {
8153 		pr_err_ratelimited("kvm: disabled by bios\n");
8154 		r = -EOPNOTSUPP;
8155 		goto out;
8156 	}
8157 
8158 	/*
8159 	 * KVM explicitly assumes that the guest has an FPU and
8160 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8161 	 * vCPU's FPU state as a fxregs_state struct.
8162 	 */
8163 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8164 		printk(KERN_ERR "kvm: inadequate fpu\n");
8165 		r = -EOPNOTSUPP;
8166 		goto out;
8167 	}
8168 
8169 	r = -ENOMEM;
8170 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8171 					  __alignof__(struct fpu), SLAB_ACCOUNT,
8172 					  NULL);
8173 	if (!x86_fpu_cache) {
8174 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8175 		goto out;
8176 	}
8177 
8178 	x86_emulator_cache = kvm_alloc_emulator_cache();
8179 	if (!x86_emulator_cache) {
8180 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8181 		goto out_free_x86_fpu_cache;
8182 	}
8183 
8184 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8185 	if (!user_return_msrs) {
8186 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8187 		goto out_free_x86_emulator_cache;
8188 	}
8189 	kvm_nr_uret_msrs = 0;
8190 
8191 	r = kvm_mmu_module_init();
8192 	if (r)
8193 		goto out_free_percpu;
8194 
8195 	kvm_timer_init();
8196 
8197 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8198 
8199 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8200 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8201 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8202 	}
8203 
8204 	if (pi_inject_timer == -1)
8205 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8206 #ifdef CONFIG_X86_64
8207 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8208 
8209 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8210 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8211 #endif
8212 
8213 	return 0;
8214 
8215 out_free_percpu:
8216 	free_percpu(user_return_msrs);
8217 out_free_x86_emulator_cache:
8218 	kmem_cache_destroy(x86_emulator_cache);
8219 out_free_x86_fpu_cache:
8220 	kmem_cache_destroy(x86_fpu_cache);
8221 out:
8222 	return r;
8223 }
8224 
8225 void kvm_arch_exit(void)
8226 {
8227 #ifdef CONFIG_X86_64
8228 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8229 		clear_hv_tscchange_cb();
8230 #endif
8231 	kvm_lapic_exit();
8232 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8233 
8234 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8235 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8236 					    CPUFREQ_TRANSITION_NOTIFIER);
8237 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8238 #ifdef CONFIG_X86_64
8239 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8240 	irq_work_sync(&pvclock_irq_work);
8241 	cancel_work_sync(&pvclock_gtod_work);
8242 #endif
8243 	kvm_x86_ops.hardware_enable = NULL;
8244 	kvm_mmu_module_exit();
8245 	free_percpu(user_return_msrs);
8246 	kmem_cache_destroy(x86_fpu_cache);
8247 #ifdef CONFIG_KVM_XEN
8248 	static_key_deferred_flush(&kvm_xen_enabled);
8249 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8250 #endif
8251 }
8252 
8253 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8254 {
8255 	++vcpu->stat.halt_exits;
8256 	if (lapic_in_kernel(vcpu)) {
8257 		vcpu->arch.mp_state = state;
8258 		return 1;
8259 	} else {
8260 		vcpu->run->exit_reason = reason;
8261 		return 0;
8262 	}
8263 }
8264 
8265 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8266 {
8267 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8268 }
8269 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8270 
8271 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8272 {
8273 	int ret = kvm_skip_emulated_instruction(vcpu);
8274 	/*
8275 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8276 	 * KVM_EXIT_DEBUG here.
8277 	 */
8278 	return kvm_vcpu_halt(vcpu) && ret;
8279 }
8280 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8281 
8282 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8283 {
8284 	int ret = kvm_skip_emulated_instruction(vcpu);
8285 
8286 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8287 }
8288 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8289 
8290 #ifdef CONFIG_X86_64
8291 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8292 			        unsigned long clock_type)
8293 {
8294 	struct kvm_clock_pairing clock_pairing;
8295 	struct timespec64 ts;
8296 	u64 cycle;
8297 	int ret;
8298 
8299 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8300 		return -KVM_EOPNOTSUPP;
8301 
8302 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8303 		return -KVM_EOPNOTSUPP;
8304 
8305 	clock_pairing.sec = ts.tv_sec;
8306 	clock_pairing.nsec = ts.tv_nsec;
8307 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8308 	clock_pairing.flags = 0;
8309 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8310 
8311 	ret = 0;
8312 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8313 			    sizeof(struct kvm_clock_pairing)))
8314 		ret = -KVM_EFAULT;
8315 
8316 	return ret;
8317 }
8318 #endif
8319 
8320 /*
8321  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8322  *
8323  * @apicid - apicid of vcpu to be kicked.
8324  */
8325 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8326 {
8327 	struct kvm_lapic_irq lapic_irq;
8328 
8329 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8330 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8331 	lapic_irq.level = 0;
8332 	lapic_irq.dest_id = apicid;
8333 	lapic_irq.msi_redir_hint = false;
8334 
8335 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8336 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8337 }
8338 
8339 bool kvm_apicv_activated(struct kvm *kvm)
8340 {
8341 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8342 }
8343 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8344 
8345 void kvm_apicv_init(struct kvm *kvm, bool enable)
8346 {
8347 	if (enable)
8348 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8349 			  &kvm->arch.apicv_inhibit_reasons);
8350 	else
8351 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8352 			&kvm->arch.apicv_inhibit_reasons);
8353 }
8354 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8355 
8356 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8357 {
8358 	struct kvm_vcpu *target = NULL;
8359 	struct kvm_apic_map *map;
8360 
8361 	vcpu->stat.directed_yield_attempted++;
8362 
8363 	rcu_read_lock();
8364 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8365 
8366 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8367 		target = map->phys_map[dest_id]->vcpu;
8368 
8369 	rcu_read_unlock();
8370 
8371 	if (!target || !READ_ONCE(target->ready))
8372 		goto no_yield;
8373 
8374 	/* Ignore requests to yield to self */
8375 	if (vcpu == target)
8376 		goto no_yield;
8377 
8378 	if (kvm_vcpu_yield_to(target) <= 0)
8379 		goto no_yield;
8380 
8381 	vcpu->stat.directed_yield_successful++;
8382 
8383 no_yield:
8384 	return;
8385 }
8386 
8387 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8388 {
8389 	unsigned long nr, a0, a1, a2, a3, ret;
8390 	int op_64_bit;
8391 
8392 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8393 		return kvm_xen_hypercall(vcpu);
8394 
8395 	if (kvm_hv_hypercall_enabled(vcpu))
8396 		return kvm_hv_hypercall(vcpu);
8397 
8398 	nr = kvm_rax_read(vcpu);
8399 	a0 = kvm_rbx_read(vcpu);
8400 	a1 = kvm_rcx_read(vcpu);
8401 	a2 = kvm_rdx_read(vcpu);
8402 	a3 = kvm_rsi_read(vcpu);
8403 
8404 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8405 
8406 	op_64_bit = is_64_bit_mode(vcpu);
8407 	if (!op_64_bit) {
8408 		nr &= 0xFFFFFFFF;
8409 		a0 &= 0xFFFFFFFF;
8410 		a1 &= 0xFFFFFFFF;
8411 		a2 &= 0xFFFFFFFF;
8412 		a3 &= 0xFFFFFFFF;
8413 	}
8414 
8415 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8416 		ret = -KVM_EPERM;
8417 		goto out;
8418 	}
8419 
8420 	ret = -KVM_ENOSYS;
8421 
8422 	switch (nr) {
8423 	case KVM_HC_VAPIC_POLL_IRQ:
8424 		ret = 0;
8425 		break;
8426 	case KVM_HC_KICK_CPU:
8427 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8428 			break;
8429 
8430 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8431 		kvm_sched_yield(vcpu, a1);
8432 		ret = 0;
8433 		break;
8434 #ifdef CONFIG_X86_64
8435 	case KVM_HC_CLOCK_PAIRING:
8436 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8437 		break;
8438 #endif
8439 	case KVM_HC_SEND_IPI:
8440 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8441 			break;
8442 
8443 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8444 		break;
8445 	case KVM_HC_SCHED_YIELD:
8446 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8447 			break;
8448 
8449 		kvm_sched_yield(vcpu, a0);
8450 		ret = 0;
8451 		break;
8452 	default:
8453 		ret = -KVM_ENOSYS;
8454 		break;
8455 	}
8456 out:
8457 	if (!op_64_bit)
8458 		ret = (u32)ret;
8459 	kvm_rax_write(vcpu, ret);
8460 
8461 	++vcpu->stat.hypercalls;
8462 	return kvm_skip_emulated_instruction(vcpu);
8463 }
8464 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8465 
8466 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8467 {
8468 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8469 	char instruction[3];
8470 	unsigned long rip = kvm_rip_read(vcpu);
8471 
8472 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8473 
8474 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8475 		&ctxt->exception);
8476 }
8477 
8478 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8479 {
8480 	return vcpu->run->request_interrupt_window &&
8481 		likely(!pic_in_kernel(vcpu->kvm));
8482 }
8483 
8484 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8485 {
8486 	struct kvm_run *kvm_run = vcpu->run;
8487 
8488 	/*
8489 	 * if_flag is obsolete and useless, so do not bother
8490 	 * setting it for SEV-ES guests.  Userspace can just
8491 	 * use kvm_run->ready_for_interrupt_injection.
8492 	 */
8493 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8494 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8495 
8496 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8497 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8498 	kvm_run->ready_for_interrupt_injection =
8499 		pic_in_kernel(vcpu->kvm) ||
8500 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8501 
8502 	if (is_smm(vcpu))
8503 		kvm_run->flags |= KVM_RUN_X86_SMM;
8504 }
8505 
8506 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8507 {
8508 	int max_irr, tpr;
8509 
8510 	if (!kvm_x86_ops.update_cr8_intercept)
8511 		return;
8512 
8513 	if (!lapic_in_kernel(vcpu))
8514 		return;
8515 
8516 	if (vcpu->arch.apicv_active)
8517 		return;
8518 
8519 	if (!vcpu->arch.apic->vapic_addr)
8520 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8521 	else
8522 		max_irr = -1;
8523 
8524 	if (max_irr != -1)
8525 		max_irr >>= 4;
8526 
8527 	tpr = kvm_lapic_get_cr8(vcpu);
8528 
8529 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8530 }
8531 
8532 
8533 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8534 {
8535 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8536 		return -EIO;
8537 
8538 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8539 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
8540 		return 1;
8541 	}
8542 
8543 	return kvm_x86_ops.nested_ops->check_events(vcpu);
8544 }
8545 
8546 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8547 {
8548 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8549 		vcpu->arch.exception.error_code = false;
8550 	static_call(kvm_x86_queue_exception)(vcpu);
8551 }
8552 
8553 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8554 {
8555 	int r;
8556 	bool can_inject = true;
8557 
8558 	/* try to reinject previous events if any */
8559 
8560 	if (vcpu->arch.exception.injected) {
8561 		kvm_inject_exception(vcpu);
8562 		can_inject = false;
8563 	}
8564 	/*
8565 	 * Do not inject an NMI or interrupt if there is a pending
8566 	 * exception.  Exceptions and interrupts are recognized at
8567 	 * instruction boundaries, i.e. the start of an instruction.
8568 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8569 	 * NMIs and interrupts, i.e. traps are recognized before an
8570 	 * NMI/interrupt that's pending on the same instruction.
8571 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8572 	 * priority, but are only generated (pended) during instruction
8573 	 * execution, i.e. a pending fault-like exception means the
8574 	 * fault occurred on the *previous* instruction and must be
8575 	 * serviced prior to recognizing any new events in order to
8576 	 * fully complete the previous instruction.
8577 	 */
8578 	else if (!vcpu->arch.exception.pending) {
8579 		if (vcpu->arch.nmi_injected) {
8580 			static_call(kvm_x86_set_nmi)(vcpu);
8581 			can_inject = false;
8582 		} else if (vcpu->arch.interrupt.injected) {
8583 			static_call(kvm_x86_set_irq)(vcpu);
8584 			can_inject = false;
8585 		}
8586 	}
8587 
8588 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8589 		     vcpu->arch.exception.pending);
8590 
8591 	/*
8592 	 * Call check_nested_events() even if we reinjected a previous event
8593 	 * in order for caller to determine if it should require immediate-exit
8594 	 * from L2 to L1 due to pending L1 events which require exit
8595 	 * from L2 to L1.
8596 	 */
8597 	if (is_guest_mode(vcpu)) {
8598 		r = kvm_check_nested_events(vcpu);
8599 		if (r < 0)
8600 			goto busy;
8601 	}
8602 
8603 	/* try to inject new event if pending */
8604 	if (vcpu->arch.exception.pending) {
8605 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8606 					vcpu->arch.exception.has_error_code,
8607 					vcpu->arch.exception.error_code);
8608 
8609 		vcpu->arch.exception.pending = false;
8610 		vcpu->arch.exception.injected = true;
8611 
8612 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8613 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8614 					     X86_EFLAGS_RF);
8615 
8616 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8617 			kvm_deliver_exception_payload(vcpu);
8618 			if (vcpu->arch.dr7 & DR7_GD) {
8619 				vcpu->arch.dr7 &= ~DR7_GD;
8620 				kvm_update_dr7(vcpu);
8621 			}
8622 		}
8623 
8624 		kvm_inject_exception(vcpu);
8625 		can_inject = false;
8626 	}
8627 
8628 	/*
8629 	 * Finally, inject interrupt events.  If an event cannot be injected
8630 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8631 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8632 	 * and can architecturally be injected, but we cannot do it right now:
8633 	 * an interrupt could have arrived just now and we have to inject it
8634 	 * as a vmexit, or there could already an event in the queue, which is
8635 	 * indicated by can_inject.  In that case we request an immediate exit
8636 	 * in order to make progress and get back here for another iteration.
8637 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8638 	 */
8639 	if (vcpu->arch.smi_pending) {
8640 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8641 		if (r < 0)
8642 			goto busy;
8643 		if (r) {
8644 			vcpu->arch.smi_pending = false;
8645 			++vcpu->arch.smi_count;
8646 			enter_smm(vcpu);
8647 			can_inject = false;
8648 		} else
8649 			static_call(kvm_x86_enable_smi_window)(vcpu);
8650 	}
8651 
8652 	if (vcpu->arch.nmi_pending) {
8653 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8654 		if (r < 0)
8655 			goto busy;
8656 		if (r) {
8657 			--vcpu->arch.nmi_pending;
8658 			vcpu->arch.nmi_injected = true;
8659 			static_call(kvm_x86_set_nmi)(vcpu);
8660 			can_inject = false;
8661 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8662 		}
8663 		if (vcpu->arch.nmi_pending)
8664 			static_call(kvm_x86_enable_nmi_window)(vcpu);
8665 	}
8666 
8667 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8668 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8669 		if (r < 0)
8670 			goto busy;
8671 		if (r) {
8672 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8673 			static_call(kvm_x86_set_irq)(vcpu);
8674 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8675 		}
8676 		if (kvm_cpu_has_injectable_intr(vcpu))
8677 			static_call(kvm_x86_enable_irq_window)(vcpu);
8678 	}
8679 
8680 	if (is_guest_mode(vcpu) &&
8681 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8682 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8683 		*req_immediate_exit = true;
8684 
8685 	WARN_ON(vcpu->arch.exception.pending);
8686 	return;
8687 
8688 busy:
8689 	*req_immediate_exit = true;
8690 	return;
8691 }
8692 
8693 static void process_nmi(struct kvm_vcpu *vcpu)
8694 {
8695 	unsigned limit = 2;
8696 
8697 	/*
8698 	 * x86 is limited to one NMI running, and one NMI pending after it.
8699 	 * If an NMI is already in progress, limit further NMIs to just one.
8700 	 * Otherwise, allow two (and we'll inject the first one immediately).
8701 	 */
8702 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8703 		limit = 1;
8704 
8705 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8706 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8707 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8708 }
8709 
8710 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8711 {
8712 	u32 flags = 0;
8713 	flags |= seg->g       << 23;
8714 	flags |= seg->db      << 22;
8715 	flags |= seg->l       << 21;
8716 	flags |= seg->avl     << 20;
8717 	flags |= seg->present << 15;
8718 	flags |= seg->dpl     << 13;
8719 	flags |= seg->s       << 12;
8720 	flags |= seg->type    << 8;
8721 	return flags;
8722 }
8723 
8724 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8725 {
8726 	struct kvm_segment seg;
8727 	int offset;
8728 
8729 	kvm_get_segment(vcpu, &seg, n);
8730 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8731 
8732 	if (n < 3)
8733 		offset = 0x7f84 + n * 12;
8734 	else
8735 		offset = 0x7f2c + (n - 3) * 12;
8736 
8737 	put_smstate(u32, buf, offset + 8, seg.base);
8738 	put_smstate(u32, buf, offset + 4, seg.limit);
8739 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8740 }
8741 
8742 #ifdef CONFIG_X86_64
8743 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8744 {
8745 	struct kvm_segment seg;
8746 	int offset;
8747 	u16 flags;
8748 
8749 	kvm_get_segment(vcpu, &seg, n);
8750 	offset = 0x7e00 + n * 16;
8751 
8752 	flags = enter_smm_get_segment_flags(&seg) >> 8;
8753 	put_smstate(u16, buf, offset, seg.selector);
8754 	put_smstate(u16, buf, offset + 2, flags);
8755 	put_smstate(u32, buf, offset + 4, seg.limit);
8756 	put_smstate(u64, buf, offset + 8, seg.base);
8757 }
8758 #endif
8759 
8760 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8761 {
8762 	struct desc_ptr dt;
8763 	struct kvm_segment seg;
8764 	unsigned long val;
8765 	int i;
8766 
8767 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8768 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8769 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8770 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8771 
8772 	for (i = 0; i < 8; i++)
8773 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8774 
8775 	kvm_get_dr(vcpu, 6, &val);
8776 	put_smstate(u32, buf, 0x7fcc, (u32)val);
8777 	kvm_get_dr(vcpu, 7, &val);
8778 	put_smstate(u32, buf, 0x7fc8, (u32)val);
8779 
8780 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8781 	put_smstate(u32, buf, 0x7fc4, seg.selector);
8782 	put_smstate(u32, buf, 0x7f64, seg.base);
8783 	put_smstate(u32, buf, 0x7f60, seg.limit);
8784 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8785 
8786 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8787 	put_smstate(u32, buf, 0x7fc0, seg.selector);
8788 	put_smstate(u32, buf, 0x7f80, seg.base);
8789 	put_smstate(u32, buf, 0x7f7c, seg.limit);
8790 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8791 
8792 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
8793 	put_smstate(u32, buf, 0x7f74, dt.address);
8794 	put_smstate(u32, buf, 0x7f70, dt.size);
8795 
8796 	static_call(kvm_x86_get_idt)(vcpu, &dt);
8797 	put_smstate(u32, buf, 0x7f58, dt.address);
8798 	put_smstate(u32, buf, 0x7f54, dt.size);
8799 
8800 	for (i = 0; i < 6; i++)
8801 		enter_smm_save_seg_32(vcpu, buf, i);
8802 
8803 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8804 
8805 	/* revision id */
8806 	put_smstate(u32, buf, 0x7efc, 0x00020000);
8807 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8808 }
8809 
8810 #ifdef CONFIG_X86_64
8811 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8812 {
8813 	struct desc_ptr dt;
8814 	struct kvm_segment seg;
8815 	unsigned long val;
8816 	int i;
8817 
8818 	for (i = 0; i < 16; i++)
8819 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8820 
8821 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8822 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8823 
8824 	kvm_get_dr(vcpu, 6, &val);
8825 	put_smstate(u64, buf, 0x7f68, val);
8826 	kvm_get_dr(vcpu, 7, &val);
8827 	put_smstate(u64, buf, 0x7f60, val);
8828 
8829 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8830 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8831 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8832 
8833 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8834 
8835 	/* revision id */
8836 	put_smstate(u32, buf, 0x7efc, 0x00020064);
8837 
8838 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8839 
8840 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8841 	put_smstate(u16, buf, 0x7e90, seg.selector);
8842 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8843 	put_smstate(u32, buf, 0x7e94, seg.limit);
8844 	put_smstate(u64, buf, 0x7e98, seg.base);
8845 
8846 	static_call(kvm_x86_get_idt)(vcpu, &dt);
8847 	put_smstate(u32, buf, 0x7e84, dt.size);
8848 	put_smstate(u64, buf, 0x7e88, dt.address);
8849 
8850 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8851 	put_smstate(u16, buf, 0x7e70, seg.selector);
8852 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8853 	put_smstate(u32, buf, 0x7e74, seg.limit);
8854 	put_smstate(u64, buf, 0x7e78, seg.base);
8855 
8856 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
8857 	put_smstate(u32, buf, 0x7e64, dt.size);
8858 	put_smstate(u64, buf, 0x7e68, dt.address);
8859 
8860 	for (i = 0; i < 6; i++)
8861 		enter_smm_save_seg_64(vcpu, buf, i);
8862 }
8863 #endif
8864 
8865 static void enter_smm(struct kvm_vcpu *vcpu)
8866 {
8867 	struct kvm_segment cs, ds;
8868 	struct desc_ptr dt;
8869 	char buf[512];
8870 	u32 cr0;
8871 
8872 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8873 	memset(buf, 0, 512);
8874 #ifdef CONFIG_X86_64
8875 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8876 		enter_smm_save_state_64(vcpu, buf);
8877 	else
8878 #endif
8879 		enter_smm_save_state_32(vcpu, buf);
8880 
8881 	/*
8882 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8883 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
8884 	 * the SMM state-save area.
8885 	 */
8886 	static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8887 
8888 	vcpu->arch.hflags |= HF_SMM_MASK;
8889 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8890 
8891 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8892 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8893 	else
8894 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8895 
8896 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8897 	kvm_rip_write(vcpu, 0x8000);
8898 
8899 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8900 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
8901 	vcpu->arch.cr0 = cr0;
8902 
8903 	static_call(kvm_x86_set_cr4)(vcpu, 0);
8904 
8905 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
8906 	dt.address = dt.size = 0;
8907 	static_call(kvm_x86_set_idt)(vcpu, &dt);
8908 
8909 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8910 
8911 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8912 	cs.base = vcpu->arch.smbase;
8913 
8914 	ds.selector = 0;
8915 	ds.base = 0;
8916 
8917 	cs.limit    = ds.limit = 0xffffffff;
8918 	cs.type     = ds.type = 0x3;
8919 	cs.dpl      = ds.dpl = 0;
8920 	cs.db       = ds.db = 0;
8921 	cs.s        = ds.s = 1;
8922 	cs.l        = ds.l = 0;
8923 	cs.g        = ds.g = 1;
8924 	cs.avl      = ds.avl = 0;
8925 	cs.present  = ds.present = 1;
8926 	cs.unusable = ds.unusable = 0;
8927 	cs.padding  = ds.padding = 0;
8928 
8929 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8930 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8931 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8932 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8933 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8934 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8935 
8936 #ifdef CONFIG_X86_64
8937 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8938 		static_call(kvm_x86_set_efer)(vcpu, 0);
8939 #endif
8940 
8941 	kvm_update_cpuid_runtime(vcpu);
8942 	kvm_mmu_reset_context(vcpu);
8943 }
8944 
8945 static void process_smi(struct kvm_vcpu *vcpu)
8946 {
8947 	vcpu->arch.smi_pending = true;
8948 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8949 }
8950 
8951 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8952 				       unsigned long *vcpu_bitmap)
8953 {
8954 	cpumask_var_t cpus;
8955 
8956 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8957 
8958 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8959 				    NULL, vcpu_bitmap, cpus);
8960 
8961 	free_cpumask_var(cpus);
8962 }
8963 
8964 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8965 {
8966 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8967 }
8968 
8969 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8970 {
8971 	if (!lapic_in_kernel(vcpu))
8972 		return;
8973 
8974 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8975 	kvm_apic_update_apicv(vcpu);
8976 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8977 }
8978 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8979 
8980 /*
8981  * NOTE: Do not hold any lock prior to calling this.
8982  *
8983  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8984  * locked, because it calls __x86_set_memory_region() which does
8985  * synchronize_srcu(&kvm->srcu).
8986  */
8987 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8988 {
8989 	struct kvm_vcpu *except;
8990 	unsigned long old, new, expected;
8991 
8992 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8993 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8994 		return;
8995 
8996 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8997 	do {
8998 		expected = new = old;
8999 		if (activate)
9000 			__clear_bit(bit, &new);
9001 		else
9002 			__set_bit(bit, &new);
9003 		if (new == old)
9004 			break;
9005 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9006 	} while (old != expected);
9007 
9008 	if (!!old == !!new)
9009 		return;
9010 
9011 	trace_kvm_apicv_update_request(activate, bit);
9012 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9013 		static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9014 
9015 	/*
9016 	 * Sending request to update APICV for all other vcpus,
9017 	 * while update the calling vcpu immediately instead of
9018 	 * waiting for another #VMEXIT to handle the request.
9019 	 */
9020 	except = kvm_get_running_vcpu();
9021 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9022 					 except);
9023 	if (except)
9024 		kvm_vcpu_update_apicv(except);
9025 }
9026 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9027 
9028 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9029 {
9030 	if (!kvm_apic_present(vcpu))
9031 		return;
9032 
9033 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9034 
9035 	if (irqchip_split(vcpu->kvm))
9036 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9037 	else {
9038 		if (vcpu->arch.apicv_active)
9039 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9040 		if (ioapic_in_kernel(vcpu->kvm))
9041 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9042 	}
9043 
9044 	if (is_guest_mode(vcpu))
9045 		vcpu->arch.load_eoi_exitmap_pending = true;
9046 	else
9047 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9048 }
9049 
9050 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9051 {
9052 	u64 eoi_exit_bitmap[4];
9053 
9054 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9055 		return;
9056 
9057 	if (to_hv_vcpu(vcpu))
9058 		bitmap_or((ulong *)eoi_exit_bitmap,
9059 			  vcpu->arch.ioapic_handled_vectors,
9060 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9061 
9062 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9063 }
9064 
9065 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9066 					    unsigned long start, unsigned long end)
9067 {
9068 	unsigned long apic_address;
9069 
9070 	/*
9071 	 * The physical address of apic access page is stored in the VMCS.
9072 	 * Update it when it becomes invalid.
9073 	 */
9074 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9075 	if (start <= apic_address && apic_address < end)
9076 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9077 }
9078 
9079 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9080 {
9081 	if (!lapic_in_kernel(vcpu))
9082 		return;
9083 
9084 	if (!kvm_x86_ops.set_apic_access_page_addr)
9085 		return;
9086 
9087 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9088 }
9089 
9090 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9091 {
9092 	smp_send_reschedule(vcpu->cpu);
9093 }
9094 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9095 
9096 /*
9097  * Returns 1 to let vcpu_run() continue the guest execution loop without
9098  * exiting to the userspace.  Otherwise, the value will be returned to the
9099  * userspace.
9100  */
9101 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9102 {
9103 	int r;
9104 	bool req_int_win =
9105 		dm_request_for_irq_injection(vcpu) &&
9106 		kvm_cpu_accept_dm_intr(vcpu);
9107 	fastpath_t exit_fastpath;
9108 
9109 	bool req_immediate_exit = false;
9110 
9111 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9112 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9113 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9114 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9115 		trace_kvm_dirty_ring_exit(vcpu);
9116 		r = 0;
9117 		goto out;
9118 	}
9119 
9120 	if (kvm_request_pending(vcpu)) {
9121 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9122 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9123 				r = 0;
9124 				goto out;
9125 			}
9126 		}
9127 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9128 			kvm_mmu_unload(vcpu);
9129 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9130 			__kvm_migrate_timers(vcpu);
9131 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9132 			kvm_gen_update_masterclock(vcpu->kvm);
9133 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9134 			kvm_gen_kvmclock_update(vcpu);
9135 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9136 			r = kvm_guest_time_update(vcpu);
9137 			if (unlikely(r))
9138 				goto out;
9139 		}
9140 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9141 			kvm_mmu_sync_roots(vcpu);
9142 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9143 			kvm_mmu_load_pgd(vcpu);
9144 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9145 			kvm_vcpu_flush_tlb_all(vcpu);
9146 
9147 			/* Flushing all ASIDs flushes the current ASID... */
9148 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9149 		}
9150 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9151 			kvm_vcpu_flush_tlb_current(vcpu);
9152 		if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9153 			kvm_vcpu_flush_tlb_guest(vcpu);
9154 
9155 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9156 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9157 			r = 0;
9158 			goto out;
9159 		}
9160 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9161 			if (is_guest_mode(vcpu)) {
9162 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9163 			} else {
9164 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9165 				vcpu->mmio_needed = 0;
9166 				r = 0;
9167 				goto out;
9168 			}
9169 		}
9170 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9171 			/* Page is swapped out. Do synthetic halt */
9172 			vcpu->arch.apf.halted = true;
9173 			r = 1;
9174 			goto out;
9175 		}
9176 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9177 			record_steal_time(vcpu);
9178 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9179 			process_smi(vcpu);
9180 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9181 			process_nmi(vcpu);
9182 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9183 			kvm_pmu_handle_event(vcpu);
9184 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9185 			kvm_pmu_deliver_pmi(vcpu);
9186 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9187 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9188 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9189 				     vcpu->arch.ioapic_handled_vectors)) {
9190 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9191 				vcpu->run->eoi.vector =
9192 						vcpu->arch.pending_ioapic_eoi;
9193 				r = 0;
9194 				goto out;
9195 			}
9196 		}
9197 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9198 			vcpu_scan_ioapic(vcpu);
9199 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9200 			vcpu_load_eoi_exitmap(vcpu);
9201 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9202 			kvm_vcpu_reload_apic_access_page(vcpu);
9203 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9204 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9205 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9206 			r = 0;
9207 			goto out;
9208 		}
9209 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9210 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9211 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9212 			r = 0;
9213 			goto out;
9214 		}
9215 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9216 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9217 
9218 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9219 			vcpu->run->hyperv = hv_vcpu->exit;
9220 			r = 0;
9221 			goto out;
9222 		}
9223 
9224 		/*
9225 		 * KVM_REQ_HV_STIMER has to be processed after
9226 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9227 		 * depend on the guest clock being up-to-date
9228 		 */
9229 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9230 			kvm_hv_process_stimers(vcpu);
9231 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9232 			kvm_vcpu_update_apicv(vcpu);
9233 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9234 			kvm_check_async_pf_completion(vcpu);
9235 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9236 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9237 
9238 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9239 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9240 	}
9241 
9242 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9243 	    kvm_xen_has_interrupt(vcpu)) {
9244 		++vcpu->stat.req_event;
9245 		kvm_apic_accept_events(vcpu);
9246 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9247 			r = 1;
9248 			goto out;
9249 		}
9250 
9251 		inject_pending_event(vcpu, &req_immediate_exit);
9252 		if (req_int_win)
9253 			static_call(kvm_x86_enable_irq_window)(vcpu);
9254 
9255 		if (kvm_lapic_enabled(vcpu)) {
9256 			update_cr8_intercept(vcpu);
9257 			kvm_lapic_sync_to_vapic(vcpu);
9258 		}
9259 	}
9260 
9261 	r = kvm_mmu_reload(vcpu);
9262 	if (unlikely(r)) {
9263 		goto cancel_injection;
9264 	}
9265 
9266 	preempt_disable();
9267 
9268 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9269 
9270 	/*
9271 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9272 	 * IPI are then delayed after guest entry, which ensures that they
9273 	 * result in virtual interrupt delivery.
9274 	 */
9275 	local_irq_disable();
9276 	vcpu->mode = IN_GUEST_MODE;
9277 
9278 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9279 
9280 	/*
9281 	 * 1) We should set ->mode before checking ->requests.  Please see
9282 	 * the comment in kvm_vcpu_exiting_guest_mode().
9283 	 *
9284 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9285 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9286 	 * (see vmx_deliver_posted_interrupt).
9287 	 *
9288 	 * 3) This also orders the write to mode from any reads to the page
9289 	 * tables done while the VCPU is running.  Please see the comment
9290 	 * in kvm_flush_remote_tlbs.
9291 	 */
9292 	smp_mb__after_srcu_read_unlock();
9293 
9294 	/*
9295 	 * This handles the case where a posted interrupt was
9296 	 * notified with kvm_vcpu_kick.
9297 	 */
9298 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9299 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9300 
9301 	if (kvm_vcpu_exit_request(vcpu)) {
9302 		vcpu->mode = OUTSIDE_GUEST_MODE;
9303 		smp_wmb();
9304 		local_irq_enable();
9305 		preempt_enable();
9306 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9307 		r = 1;
9308 		goto cancel_injection;
9309 	}
9310 
9311 	if (req_immediate_exit) {
9312 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9313 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9314 	}
9315 
9316 	fpregs_assert_state_consistent();
9317 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9318 		switch_fpu_return();
9319 
9320 	if (unlikely(vcpu->arch.switch_db_regs)) {
9321 		set_debugreg(0, 7);
9322 		set_debugreg(vcpu->arch.eff_db[0], 0);
9323 		set_debugreg(vcpu->arch.eff_db[1], 1);
9324 		set_debugreg(vcpu->arch.eff_db[2], 2);
9325 		set_debugreg(vcpu->arch.eff_db[3], 3);
9326 		set_debugreg(vcpu->arch.dr6, 6);
9327 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9328 	}
9329 
9330 	for (;;) {
9331 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9332 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9333 			break;
9334 
9335                 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9336 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9337 			break;
9338 		}
9339 
9340 		if (vcpu->arch.apicv_active)
9341 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9342         }
9343 
9344 	/*
9345 	 * Do this here before restoring debug registers on the host.  And
9346 	 * since we do this before handling the vmexit, a DR access vmexit
9347 	 * can (a) read the correct value of the debug registers, (b) set
9348 	 * KVM_DEBUGREG_WONT_EXIT again.
9349 	 */
9350 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9351 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9352 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9353 		kvm_update_dr0123(vcpu);
9354 		kvm_update_dr7(vcpu);
9355 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9356 	}
9357 
9358 	/*
9359 	 * If the guest has used debug registers, at least dr7
9360 	 * will be disabled while returning to the host.
9361 	 * If we don't have active breakpoints in the host, we don't
9362 	 * care about the messed up debug address registers. But if
9363 	 * we have some of them active, restore the old state.
9364 	 */
9365 	if (hw_breakpoint_active())
9366 		hw_breakpoint_restore();
9367 
9368 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9369 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9370 
9371 	vcpu->mode = OUTSIDE_GUEST_MODE;
9372 	smp_wmb();
9373 
9374 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9375 
9376 	/*
9377 	 * Consume any pending interrupts, including the possible source of
9378 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9379 	 * An instruction is required after local_irq_enable() to fully unblock
9380 	 * interrupts on processors that implement an interrupt shadow, the
9381 	 * stat.exits increment will do nicely.
9382 	 */
9383 	kvm_before_interrupt(vcpu);
9384 	local_irq_enable();
9385 	++vcpu->stat.exits;
9386 	local_irq_disable();
9387 	kvm_after_interrupt(vcpu);
9388 
9389 	/*
9390 	 * Wait until after servicing IRQs to account guest time so that any
9391 	 * ticks that occurred while running the guest are properly accounted
9392 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9393 	 * of accounting via context tracking, but the loss of accuracy is
9394 	 * acceptable for all known use cases.
9395 	 */
9396 	vtime_account_guest_exit();
9397 
9398 	if (lapic_in_kernel(vcpu)) {
9399 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9400 		if (delta != S64_MIN) {
9401 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9402 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9403 		}
9404 	}
9405 
9406 	local_irq_enable();
9407 	preempt_enable();
9408 
9409 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9410 
9411 	/*
9412 	 * Profile KVM exit RIPs:
9413 	 */
9414 	if (unlikely(prof_on == KVM_PROFILING)) {
9415 		unsigned long rip = kvm_rip_read(vcpu);
9416 		profile_hit(KVM_PROFILING, (void *)rip);
9417 	}
9418 
9419 	if (unlikely(vcpu->arch.tsc_always_catchup))
9420 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9421 
9422 	if (vcpu->arch.apic_attention)
9423 		kvm_lapic_sync_from_vapic(vcpu);
9424 
9425 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9426 	return r;
9427 
9428 cancel_injection:
9429 	if (req_immediate_exit)
9430 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9431 	static_call(kvm_x86_cancel_injection)(vcpu);
9432 	if (unlikely(vcpu->arch.apic_attention))
9433 		kvm_lapic_sync_from_vapic(vcpu);
9434 out:
9435 	return r;
9436 }
9437 
9438 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9439 {
9440 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9441 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9442 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9443 		kvm_vcpu_block(vcpu);
9444 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9445 
9446 		if (kvm_x86_ops.post_block)
9447 			static_call(kvm_x86_post_block)(vcpu);
9448 
9449 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9450 			return 1;
9451 	}
9452 
9453 	kvm_apic_accept_events(vcpu);
9454 	switch(vcpu->arch.mp_state) {
9455 	case KVM_MP_STATE_HALTED:
9456 	case KVM_MP_STATE_AP_RESET_HOLD:
9457 		vcpu->arch.pv.pv_unhalted = false;
9458 		vcpu->arch.mp_state =
9459 			KVM_MP_STATE_RUNNABLE;
9460 		fallthrough;
9461 	case KVM_MP_STATE_RUNNABLE:
9462 		vcpu->arch.apf.halted = false;
9463 		break;
9464 	case KVM_MP_STATE_INIT_RECEIVED:
9465 		break;
9466 	default:
9467 		return -EINTR;
9468 	}
9469 	return 1;
9470 }
9471 
9472 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9473 {
9474 	if (is_guest_mode(vcpu))
9475 		kvm_check_nested_events(vcpu);
9476 
9477 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9478 		!vcpu->arch.apf.halted);
9479 }
9480 
9481 static int vcpu_run(struct kvm_vcpu *vcpu)
9482 {
9483 	int r;
9484 	struct kvm *kvm = vcpu->kvm;
9485 
9486 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9487 	vcpu->arch.l1tf_flush_l1d = true;
9488 
9489 	for (;;) {
9490 		if (kvm_vcpu_running(vcpu)) {
9491 			r = vcpu_enter_guest(vcpu);
9492 		} else {
9493 			r = vcpu_block(kvm, vcpu);
9494 		}
9495 
9496 		if (r <= 0)
9497 			break;
9498 
9499 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9500 		if (kvm_cpu_has_pending_timer(vcpu))
9501 			kvm_inject_pending_timer_irqs(vcpu);
9502 
9503 		if (dm_request_for_irq_injection(vcpu) &&
9504 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9505 			r = 0;
9506 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9507 			++vcpu->stat.request_irq_exits;
9508 			break;
9509 		}
9510 
9511 		if (__xfer_to_guest_mode_work_pending()) {
9512 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9513 			r = xfer_to_guest_mode_handle_work(vcpu);
9514 			if (r)
9515 				return r;
9516 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9517 		}
9518 	}
9519 
9520 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9521 
9522 	return r;
9523 }
9524 
9525 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9526 {
9527 	int r;
9528 
9529 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9530 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9531 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9532 	return r;
9533 }
9534 
9535 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9536 {
9537 	BUG_ON(!vcpu->arch.pio.count);
9538 
9539 	return complete_emulated_io(vcpu);
9540 }
9541 
9542 /*
9543  * Implements the following, as a state machine:
9544  *
9545  * read:
9546  *   for each fragment
9547  *     for each mmio piece in the fragment
9548  *       write gpa, len
9549  *       exit
9550  *       copy data
9551  *   execute insn
9552  *
9553  * write:
9554  *   for each fragment
9555  *     for each mmio piece in the fragment
9556  *       write gpa, len
9557  *       copy data
9558  *       exit
9559  */
9560 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9561 {
9562 	struct kvm_run *run = vcpu->run;
9563 	struct kvm_mmio_fragment *frag;
9564 	unsigned len;
9565 
9566 	BUG_ON(!vcpu->mmio_needed);
9567 
9568 	/* Complete previous fragment */
9569 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9570 	len = min(8u, frag->len);
9571 	if (!vcpu->mmio_is_write)
9572 		memcpy(frag->data, run->mmio.data, len);
9573 
9574 	if (frag->len <= 8) {
9575 		/* Switch to the next fragment. */
9576 		frag++;
9577 		vcpu->mmio_cur_fragment++;
9578 	} else {
9579 		/* Go forward to the next mmio piece. */
9580 		frag->data += len;
9581 		frag->gpa += len;
9582 		frag->len -= len;
9583 	}
9584 
9585 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9586 		vcpu->mmio_needed = 0;
9587 
9588 		/* FIXME: return into emulator if single-stepping.  */
9589 		if (vcpu->mmio_is_write)
9590 			return 1;
9591 		vcpu->mmio_read_completed = 1;
9592 		return complete_emulated_io(vcpu);
9593 	}
9594 
9595 	run->exit_reason = KVM_EXIT_MMIO;
9596 	run->mmio.phys_addr = frag->gpa;
9597 	if (vcpu->mmio_is_write)
9598 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9599 	run->mmio.len = min(8u, frag->len);
9600 	run->mmio.is_write = vcpu->mmio_is_write;
9601 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9602 	return 0;
9603 }
9604 
9605 static void kvm_save_current_fpu(struct fpu *fpu)
9606 {
9607 	/*
9608 	 * If the target FPU state is not resident in the CPU registers, just
9609 	 * memcpy() from current, else save CPU state directly to the target.
9610 	 */
9611 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9612 		memcpy(&fpu->state, &current->thread.fpu.state,
9613 		       fpu_kernel_xstate_size);
9614 	else
9615 		copy_fpregs_to_fpstate(fpu);
9616 }
9617 
9618 /* Swap (qemu) user FPU context for the guest FPU context. */
9619 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9620 {
9621 	fpregs_lock();
9622 
9623 	kvm_save_current_fpu(vcpu->arch.user_fpu);
9624 
9625 	/*
9626 	 * Guests with protected state can't have it set by the hypervisor,
9627 	 * so skip trying to set it.
9628 	 */
9629 	if (vcpu->arch.guest_fpu)
9630 		/* PKRU is separately restored in kvm_x86_ops.run. */
9631 		__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9632 					~XFEATURE_MASK_PKRU);
9633 
9634 	fpregs_mark_activate();
9635 	fpregs_unlock();
9636 
9637 	trace_kvm_fpu(1);
9638 }
9639 
9640 /* When vcpu_run ends, restore user space FPU context. */
9641 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9642 {
9643 	fpregs_lock();
9644 
9645 	/*
9646 	 * Guests with protected state can't have it read by the hypervisor,
9647 	 * so skip trying to save it.
9648 	 */
9649 	if (vcpu->arch.guest_fpu)
9650 		kvm_save_current_fpu(vcpu->arch.guest_fpu);
9651 
9652 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9653 
9654 	fpregs_mark_activate();
9655 	fpregs_unlock();
9656 
9657 	++vcpu->stat.fpu_reload;
9658 	trace_kvm_fpu(0);
9659 }
9660 
9661 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9662 {
9663 	struct kvm_run *kvm_run = vcpu->run;
9664 	int r;
9665 
9666 	vcpu_load(vcpu);
9667 	kvm_sigset_activate(vcpu);
9668 	kvm_run->flags = 0;
9669 	kvm_load_guest_fpu(vcpu);
9670 
9671 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9672 		if (kvm_run->immediate_exit) {
9673 			r = -EINTR;
9674 			goto out;
9675 		}
9676 		kvm_vcpu_block(vcpu);
9677 		kvm_apic_accept_events(vcpu);
9678 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9679 		r = -EAGAIN;
9680 		if (signal_pending(current)) {
9681 			r = -EINTR;
9682 			kvm_run->exit_reason = KVM_EXIT_INTR;
9683 			++vcpu->stat.signal_exits;
9684 		}
9685 		goto out;
9686 	}
9687 
9688 	if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9689 		r = -EINVAL;
9690 		goto out;
9691 	}
9692 
9693 	if (kvm_run->kvm_dirty_regs) {
9694 		r = sync_regs(vcpu);
9695 		if (r != 0)
9696 			goto out;
9697 	}
9698 
9699 	/* re-sync apic's tpr */
9700 	if (!lapic_in_kernel(vcpu)) {
9701 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9702 			r = -EINVAL;
9703 			goto out;
9704 		}
9705 	}
9706 
9707 	if (unlikely(vcpu->arch.complete_userspace_io)) {
9708 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9709 		vcpu->arch.complete_userspace_io = NULL;
9710 		r = cui(vcpu);
9711 		if (r <= 0)
9712 			goto out;
9713 	} else
9714 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9715 
9716 	if (kvm_run->immediate_exit)
9717 		r = -EINTR;
9718 	else
9719 		r = vcpu_run(vcpu);
9720 
9721 out:
9722 	kvm_put_guest_fpu(vcpu);
9723 	if (kvm_run->kvm_valid_regs)
9724 		store_regs(vcpu);
9725 	post_kvm_run_save(vcpu);
9726 	kvm_sigset_deactivate(vcpu);
9727 
9728 	vcpu_put(vcpu);
9729 	return r;
9730 }
9731 
9732 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9733 {
9734 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9735 		/*
9736 		 * We are here if userspace calls get_regs() in the middle of
9737 		 * instruction emulation. Registers state needs to be copied
9738 		 * back from emulation context to vcpu. Userspace shouldn't do
9739 		 * that usually, but some bad designed PV devices (vmware
9740 		 * backdoor interface) need this to work
9741 		 */
9742 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9743 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9744 	}
9745 	regs->rax = kvm_rax_read(vcpu);
9746 	regs->rbx = kvm_rbx_read(vcpu);
9747 	regs->rcx = kvm_rcx_read(vcpu);
9748 	regs->rdx = kvm_rdx_read(vcpu);
9749 	regs->rsi = kvm_rsi_read(vcpu);
9750 	regs->rdi = kvm_rdi_read(vcpu);
9751 	regs->rsp = kvm_rsp_read(vcpu);
9752 	regs->rbp = kvm_rbp_read(vcpu);
9753 #ifdef CONFIG_X86_64
9754 	regs->r8 = kvm_r8_read(vcpu);
9755 	regs->r9 = kvm_r9_read(vcpu);
9756 	regs->r10 = kvm_r10_read(vcpu);
9757 	regs->r11 = kvm_r11_read(vcpu);
9758 	regs->r12 = kvm_r12_read(vcpu);
9759 	regs->r13 = kvm_r13_read(vcpu);
9760 	regs->r14 = kvm_r14_read(vcpu);
9761 	regs->r15 = kvm_r15_read(vcpu);
9762 #endif
9763 
9764 	regs->rip = kvm_rip_read(vcpu);
9765 	regs->rflags = kvm_get_rflags(vcpu);
9766 }
9767 
9768 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9769 {
9770 	vcpu_load(vcpu);
9771 	__get_regs(vcpu, regs);
9772 	vcpu_put(vcpu);
9773 	return 0;
9774 }
9775 
9776 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9777 {
9778 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9779 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9780 
9781 	kvm_rax_write(vcpu, regs->rax);
9782 	kvm_rbx_write(vcpu, regs->rbx);
9783 	kvm_rcx_write(vcpu, regs->rcx);
9784 	kvm_rdx_write(vcpu, regs->rdx);
9785 	kvm_rsi_write(vcpu, regs->rsi);
9786 	kvm_rdi_write(vcpu, regs->rdi);
9787 	kvm_rsp_write(vcpu, regs->rsp);
9788 	kvm_rbp_write(vcpu, regs->rbp);
9789 #ifdef CONFIG_X86_64
9790 	kvm_r8_write(vcpu, regs->r8);
9791 	kvm_r9_write(vcpu, regs->r9);
9792 	kvm_r10_write(vcpu, regs->r10);
9793 	kvm_r11_write(vcpu, regs->r11);
9794 	kvm_r12_write(vcpu, regs->r12);
9795 	kvm_r13_write(vcpu, regs->r13);
9796 	kvm_r14_write(vcpu, regs->r14);
9797 	kvm_r15_write(vcpu, regs->r15);
9798 #endif
9799 
9800 	kvm_rip_write(vcpu, regs->rip);
9801 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9802 
9803 	vcpu->arch.exception.pending = false;
9804 
9805 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9806 }
9807 
9808 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9809 {
9810 	vcpu_load(vcpu);
9811 	__set_regs(vcpu, regs);
9812 	vcpu_put(vcpu);
9813 	return 0;
9814 }
9815 
9816 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9817 {
9818 	struct kvm_segment cs;
9819 
9820 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9821 	*db = cs.db;
9822 	*l = cs.l;
9823 }
9824 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9825 
9826 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9827 {
9828 	struct desc_ptr dt;
9829 
9830 	if (vcpu->arch.guest_state_protected)
9831 		goto skip_protected_regs;
9832 
9833 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9834 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9835 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9836 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9837 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9838 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9839 
9840 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9841 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9842 
9843 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9844 	sregs->idt.limit = dt.size;
9845 	sregs->idt.base = dt.address;
9846 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9847 	sregs->gdt.limit = dt.size;
9848 	sregs->gdt.base = dt.address;
9849 
9850 	sregs->cr2 = vcpu->arch.cr2;
9851 	sregs->cr3 = kvm_read_cr3(vcpu);
9852 
9853 skip_protected_regs:
9854 	sregs->cr0 = kvm_read_cr0(vcpu);
9855 	sregs->cr4 = kvm_read_cr4(vcpu);
9856 	sregs->cr8 = kvm_get_cr8(vcpu);
9857 	sregs->efer = vcpu->arch.efer;
9858 	sregs->apic_base = kvm_get_apic_base(vcpu);
9859 
9860 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9861 
9862 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9863 		set_bit(vcpu->arch.interrupt.nr,
9864 			(unsigned long *)sregs->interrupt_bitmap);
9865 }
9866 
9867 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9868 				  struct kvm_sregs *sregs)
9869 {
9870 	vcpu_load(vcpu);
9871 	__get_sregs(vcpu, sregs);
9872 	vcpu_put(vcpu);
9873 	return 0;
9874 }
9875 
9876 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9877 				    struct kvm_mp_state *mp_state)
9878 {
9879 	vcpu_load(vcpu);
9880 	if (kvm_mpx_supported())
9881 		kvm_load_guest_fpu(vcpu);
9882 
9883 	kvm_apic_accept_events(vcpu);
9884 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9885 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9886 	    vcpu->arch.pv.pv_unhalted)
9887 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9888 	else
9889 		mp_state->mp_state = vcpu->arch.mp_state;
9890 
9891 	if (kvm_mpx_supported())
9892 		kvm_put_guest_fpu(vcpu);
9893 	vcpu_put(vcpu);
9894 	return 0;
9895 }
9896 
9897 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9898 				    struct kvm_mp_state *mp_state)
9899 {
9900 	int ret = -EINVAL;
9901 
9902 	vcpu_load(vcpu);
9903 
9904 	if (!lapic_in_kernel(vcpu) &&
9905 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9906 		goto out;
9907 
9908 	/*
9909 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9910 	 * INIT state; latched init should be reported using
9911 	 * KVM_SET_VCPU_EVENTS, so reject it here.
9912 	 */
9913 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9914 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9915 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9916 		goto out;
9917 
9918 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9919 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9920 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9921 	} else
9922 		vcpu->arch.mp_state = mp_state->mp_state;
9923 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9924 
9925 	ret = 0;
9926 out:
9927 	vcpu_put(vcpu);
9928 	return ret;
9929 }
9930 
9931 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9932 		    int reason, bool has_error_code, u32 error_code)
9933 {
9934 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9935 	int ret;
9936 
9937 	init_emulate_ctxt(vcpu);
9938 
9939 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9940 				   has_error_code, error_code);
9941 	if (ret) {
9942 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9943 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9944 		vcpu->run->internal.ndata = 0;
9945 		return 0;
9946 	}
9947 
9948 	kvm_rip_write(vcpu, ctxt->eip);
9949 	kvm_set_rflags(vcpu, ctxt->eflags);
9950 	return 1;
9951 }
9952 EXPORT_SYMBOL_GPL(kvm_task_switch);
9953 
9954 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9955 {
9956 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9957 		/*
9958 		 * When EFER.LME and CR0.PG are set, the processor is in
9959 		 * 64-bit mode (though maybe in a 32-bit code segment).
9960 		 * CR4.PAE and EFER.LMA must be set.
9961 		 */
9962 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9963 			return false;
9964 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9965 			return false;
9966 	} else {
9967 		/*
9968 		 * Not in 64-bit mode: EFER.LMA is clear and the code
9969 		 * segment cannot be 64-bit.
9970 		 */
9971 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9972 			return false;
9973 	}
9974 
9975 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
9976 }
9977 
9978 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9979 {
9980 	struct msr_data apic_base_msr;
9981 	int mmu_reset_needed = 0;
9982 	int pending_vec, max_bits, idx;
9983 	struct desc_ptr dt;
9984 	int ret = -EINVAL;
9985 
9986 	if (!kvm_is_valid_sregs(vcpu, sregs))
9987 		goto out;
9988 
9989 	apic_base_msr.data = sregs->apic_base;
9990 	apic_base_msr.host_initiated = true;
9991 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
9992 		goto out;
9993 
9994 	if (vcpu->arch.guest_state_protected)
9995 		goto skip_protected_regs;
9996 
9997 	dt.size = sregs->idt.limit;
9998 	dt.address = sregs->idt.base;
9999 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10000 	dt.size = sregs->gdt.limit;
10001 	dt.address = sregs->gdt.base;
10002 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10003 
10004 	vcpu->arch.cr2 = sregs->cr2;
10005 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10006 	vcpu->arch.cr3 = sregs->cr3;
10007 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10008 
10009 	kvm_set_cr8(vcpu, sregs->cr8);
10010 
10011 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10012 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10013 
10014 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10015 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10016 	vcpu->arch.cr0 = sregs->cr0;
10017 
10018 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10019 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10020 
10021 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10022 	if (is_pae_paging(vcpu)) {
10023 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10024 		mmu_reset_needed = 1;
10025 	}
10026 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10027 
10028 	if (mmu_reset_needed)
10029 		kvm_mmu_reset_context(vcpu);
10030 
10031 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10032 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10033 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10034 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10035 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10036 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10037 
10038 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10039 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10040 
10041 	update_cr8_intercept(vcpu);
10042 
10043 	/* Older userspace won't unhalt the vcpu on reset. */
10044 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10045 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10046 	    !is_protmode(vcpu))
10047 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10048 
10049 skip_protected_regs:
10050 	max_bits = KVM_NR_INTERRUPTS;
10051 	pending_vec = find_first_bit(
10052 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10053 	if (pending_vec < max_bits) {
10054 		kvm_queue_interrupt(vcpu, pending_vec, false);
10055 		pr_debug("Set back pending irq %d\n", pending_vec);
10056 	}
10057 
10058 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10059 
10060 	ret = 0;
10061 out:
10062 	return ret;
10063 }
10064 
10065 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10066 				  struct kvm_sregs *sregs)
10067 {
10068 	int ret;
10069 
10070 	vcpu_load(vcpu);
10071 	ret = __set_sregs(vcpu, sregs);
10072 	vcpu_put(vcpu);
10073 	return ret;
10074 }
10075 
10076 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10077 					struct kvm_guest_debug *dbg)
10078 {
10079 	unsigned long rflags;
10080 	int i, r;
10081 
10082 	if (vcpu->arch.guest_state_protected)
10083 		return -EINVAL;
10084 
10085 	vcpu_load(vcpu);
10086 
10087 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10088 		r = -EBUSY;
10089 		if (vcpu->arch.exception.pending)
10090 			goto out;
10091 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10092 			kvm_queue_exception(vcpu, DB_VECTOR);
10093 		else
10094 			kvm_queue_exception(vcpu, BP_VECTOR);
10095 	}
10096 
10097 	/*
10098 	 * Read rflags as long as potentially injected trace flags are still
10099 	 * filtered out.
10100 	 */
10101 	rflags = kvm_get_rflags(vcpu);
10102 
10103 	vcpu->guest_debug = dbg->control;
10104 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10105 		vcpu->guest_debug = 0;
10106 
10107 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10108 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10109 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10110 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10111 	} else {
10112 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10113 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10114 	}
10115 	kvm_update_dr7(vcpu);
10116 
10117 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10118 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
10119 			get_segment_base(vcpu, VCPU_SREG_CS);
10120 
10121 	/*
10122 	 * Trigger an rflags update that will inject or remove the trace
10123 	 * flags.
10124 	 */
10125 	kvm_set_rflags(vcpu, rflags);
10126 
10127 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10128 
10129 	r = 0;
10130 
10131 out:
10132 	vcpu_put(vcpu);
10133 	return r;
10134 }
10135 
10136 /*
10137  * Translate a guest virtual address to a guest physical address.
10138  */
10139 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10140 				    struct kvm_translation *tr)
10141 {
10142 	unsigned long vaddr = tr->linear_address;
10143 	gpa_t gpa;
10144 	int idx;
10145 
10146 	vcpu_load(vcpu);
10147 
10148 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10149 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10150 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10151 	tr->physical_address = gpa;
10152 	tr->valid = gpa != UNMAPPED_GVA;
10153 	tr->writeable = 1;
10154 	tr->usermode = 0;
10155 
10156 	vcpu_put(vcpu);
10157 	return 0;
10158 }
10159 
10160 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10161 {
10162 	struct fxregs_state *fxsave;
10163 
10164 	if (!vcpu->arch.guest_fpu)
10165 		return 0;
10166 
10167 	vcpu_load(vcpu);
10168 
10169 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10170 	memcpy(fpu->fpr, fxsave->st_space, 128);
10171 	fpu->fcw = fxsave->cwd;
10172 	fpu->fsw = fxsave->swd;
10173 	fpu->ftwx = fxsave->twd;
10174 	fpu->last_opcode = fxsave->fop;
10175 	fpu->last_ip = fxsave->rip;
10176 	fpu->last_dp = fxsave->rdp;
10177 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10178 
10179 	vcpu_put(vcpu);
10180 	return 0;
10181 }
10182 
10183 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10184 {
10185 	struct fxregs_state *fxsave;
10186 
10187 	if (!vcpu->arch.guest_fpu)
10188 		return 0;
10189 
10190 	vcpu_load(vcpu);
10191 
10192 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10193 
10194 	memcpy(fxsave->st_space, fpu->fpr, 128);
10195 	fxsave->cwd = fpu->fcw;
10196 	fxsave->swd = fpu->fsw;
10197 	fxsave->twd = fpu->ftwx;
10198 	fxsave->fop = fpu->last_opcode;
10199 	fxsave->rip = fpu->last_ip;
10200 	fxsave->rdp = fpu->last_dp;
10201 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10202 
10203 	vcpu_put(vcpu);
10204 	return 0;
10205 }
10206 
10207 static void store_regs(struct kvm_vcpu *vcpu)
10208 {
10209 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10210 
10211 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10212 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10213 
10214 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10215 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10216 
10217 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10218 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10219 				vcpu, &vcpu->run->s.regs.events);
10220 }
10221 
10222 static int sync_regs(struct kvm_vcpu *vcpu)
10223 {
10224 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10225 		return -EINVAL;
10226 
10227 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10228 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10229 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10230 	}
10231 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10232 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10233 			return -EINVAL;
10234 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10235 	}
10236 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10237 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10238 				vcpu, &vcpu->run->s.regs.events))
10239 			return -EINVAL;
10240 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10241 	}
10242 
10243 	return 0;
10244 }
10245 
10246 static void fx_init(struct kvm_vcpu *vcpu)
10247 {
10248 	if (!vcpu->arch.guest_fpu)
10249 		return;
10250 
10251 	fpstate_init(&vcpu->arch.guest_fpu->state);
10252 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10253 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10254 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
10255 
10256 	/*
10257 	 * Ensure guest xcr0 is valid for loading
10258 	 */
10259 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10260 
10261 	vcpu->arch.cr0 |= X86_CR0_ET;
10262 }
10263 
10264 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10265 {
10266 	if (vcpu->arch.guest_fpu) {
10267 		kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10268 		vcpu->arch.guest_fpu = NULL;
10269 	}
10270 }
10271 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10272 
10273 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10274 {
10275 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10276 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10277 			     "guest TSC will not be reliable\n");
10278 
10279 	return 0;
10280 }
10281 
10282 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10283 {
10284 	struct page *page;
10285 	int r;
10286 
10287 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10288 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10289 	else
10290 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10291 
10292 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10293 
10294 	r = kvm_mmu_create(vcpu);
10295 	if (r < 0)
10296 		return r;
10297 
10298 	if (irqchip_in_kernel(vcpu->kvm)) {
10299 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10300 		if (r < 0)
10301 			goto fail_mmu_destroy;
10302 		if (kvm_apicv_activated(vcpu->kvm))
10303 			vcpu->arch.apicv_active = true;
10304 	} else
10305 		static_branch_inc(&kvm_has_noapic_vcpu);
10306 
10307 	r = -ENOMEM;
10308 
10309 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10310 	if (!page)
10311 		goto fail_free_lapic;
10312 	vcpu->arch.pio_data = page_address(page);
10313 
10314 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10315 				       GFP_KERNEL_ACCOUNT);
10316 	if (!vcpu->arch.mce_banks)
10317 		goto fail_free_pio_data;
10318 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10319 
10320 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10321 				GFP_KERNEL_ACCOUNT))
10322 		goto fail_free_mce_banks;
10323 
10324 	if (!alloc_emulate_ctxt(vcpu))
10325 		goto free_wbinvd_dirty_mask;
10326 
10327 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10328 						GFP_KERNEL_ACCOUNT);
10329 	if (!vcpu->arch.user_fpu) {
10330 		pr_err("kvm: failed to allocate userspace's fpu\n");
10331 		goto free_emulate_ctxt;
10332 	}
10333 
10334 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10335 						 GFP_KERNEL_ACCOUNT);
10336 	if (!vcpu->arch.guest_fpu) {
10337 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10338 		goto free_user_fpu;
10339 	}
10340 	fx_init(vcpu);
10341 
10342 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10343 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10344 
10345 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10346 
10347 	kvm_async_pf_hash_reset(vcpu);
10348 	kvm_pmu_init(vcpu);
10349 
10350 	vcpu->arch.pending_external_vector = -1;
10351 	vcpu->arch.preempted_in_kernel = false;
10352 
10353 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10354 	if (r)
10355 		goto free_guest_fpu;
10356 
10357 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10358 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10359 	kvm_vcpu_mtrr_init(vcpu);
10360 	vcpu_load(vcpu);
10361 	kvm_vcpu_reset(vcpu, false);
10362 	kvm_init_mmu(vcpu, false);
10363 	vcpu_put(vcpu);
10364 	return 0;
10365 
10366 free_guest_fpu:
10367 	kvm_free_guest_fpu(vcpu);
10368 free_user_fpu:
10369 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10370 free_emulate_ctxt:
10371 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10372 free_wbinvd_dirty_mask:
10373 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10374 fail_free_mce_banks:
10375 	kfree(vcpu->arch.mce_banks);
10376 fail_free_pio_data:
10377 	free_page((unsigned long)vcpu->arch.pio_data);
10378 fail_free_lapic:
10379 	kvm_free_lapic(vcpu);
10380 fail_mmu_destroy:
10381 	kvm_mmu_destroy(vcpu);
10382 	return r;
10383 }
10384 
10385 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10386 {
10387 	struct kvm *kvm = vcpu->kvm;
10388 
10389 	if (mutex_lock_killable(&vcpu->mutex))
10390 		return;
10391 	vcpu_load(vcpu);
10392 	kvm_synchronize_tsc(vcpu, 0);
10393 	vcpu_put(vcpu);
10394 
10395 	/* poll control enabled by default */
10396 	vcpu->arch.msr_kvm_poll_control = 1;
10397 
10398 	mutex_unlock(&vcpu->mutex);
10399 
10400 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10401 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10402 						KVMCLOCK_SYNC_PERIOD);
10403 }
10404 
10405 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10406 {
10407 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10408 	int idx;
10409 
10410 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10411 
10412 	kvmclock_reset(vcpu);
10413 
10414 	static_call(kvm_x86_vcpu_free)(vcpu);
10415 
10416 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10417 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10418 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10419 	kvm_free_guest_fpu(vcpu);
10420 
10421 	kvm_hv_vcpu_uninit(vcpu);
10422 	kvm_pmu_destroy(vcpu);
10423 	kfree(vcpu->arch.mce_banks);
10424 	kvm_free_lapic(vcpu);
10425 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10426 	kvm_mmu_destroy(vcpu);
10427 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10428 	free_page((unsigned long)vcpu->arch.pio_data);
10429 	kvfree(vcpu->arch.cpuid_entries);
10430 	if (!lapic_in_kernel(vcpu))
10431 		static_branch_dec(&kvm_has_noapic_vcpu);
10432 }
10433 
10434 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10435 {
10436 	kvm_lapic_reset(vcpu, init_event);
10437 
10438 	vcpu->arch.hflags = 0;
10439 
10440 	vcpu->arch.smi_pending = 0;
10441 	vcpu->arch.smi_count = 0;
10442 	atomic_set(&vcpu->arch.nmi_queued, 0);
10443 	vcpu->arch.nmi_pending = 0;
10444 	vcpu->arch.nmi_injected = false;
10445 	kvm_clear_interrupt_queue(vcpu);
10446 	kvm_clear_exception_queue(vcpu);
10447 
10448 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10449 	kvm_update_dr0123(vcpu);
10450 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10451 	vcpu->arch.dr7 = DR7_FIXED_1;
10452 	kvm_update_dr7(vcpu);
10453 
10454 	vcpu->arch.cr2 = 0;
10455 
10456 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10457 	vcpu->arch.apf.msr_en_val = 0;
10458 	vcpu->arch.apf.msr_int_val = 0;
10459 	vcpu->arch.st.msr_val = 0;
10460 
10461 	kvmclock_reset(vcpu);
10462 
10463 	kvm_clear_async_pf_completion_queue(vcpu);
10464 	kvm_async_pf_hash_reset(vcpu);
10465 	vcpu->arch.apf.halted = false;
10466 
10467 	if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10468 		void *mpx_state_buffer;
10469 
10470 		/*
10471 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10472 		 * called with loaded FPU and does not let userspace fix the state.
10473 		 */
10474 		if (init_event)
10475 			kvm_put_guest_fpu(vcpu);
10476 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10477 					XFEATURE_BNDREGS);
10478 		if (mpx_state_buffer)
10479 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10480 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10481 					XFEATURE_BNDCSR);
10482 		if (mpx_state_buffer)
10483 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10484 		if (init_event)
10485 			kvm_load_guest_fpu(vcpu);
10486 	}
10487 
10488 	if (!init_event) {
10489 		kvm_pmu_reset(vcpu);
10490 		vcpu->arch.smbase = 0x30000;
10491 
10492 		vcpu->arch.msr_misc_features_enables = 0;
10493 
10494 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10495 	}
10496 
10497 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10498 	vcpu->arch.regs_avail = ~0;
10499 	vcpu->arch.regs_dirty = ~0;
10500 
10501 	vcpu->arch.ia32_xss = 0;
10502 
10503 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10504 }
10505 
10506 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10507 {
10508 	struct kvm_segment cs;
10509 
10510 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10511 	cs.selector = vector << 8;
10512 	cs.base = vector << 12;
10513 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10514 	kvm_rip_write(vcpu, 0);
10515 }
10516 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10517 
10518 int kvm_arch_hardware_enable(void)
10519 {
10520 	struct kvm *kvm;
10521 	struct kvm_vcpu *vcpu;
10522 	int i;
10523 	int ret;
10524 	u64 local_tsc;
10525 	u64 max_tsc = 0;
10526 	bool stable, backwards_tsc = false;
10527 
10528 	kvm_user_return_msr_cpu_online();
10529 	ret = static_call(kvm_x86_hardware_enable)();
10530 	if (ret != 0)
10531 		return ret;
10532 
10533 	local_tsc = rdtsc();
10534 	stable = !kvm_check_tsc_unstable();
10535 	list_for_each_entry(kvm, &vm_list, vm_list) {
10536 		kvm_for_each_vcpu(i, vcpu, kvm) {
10537 			if (!stable && vcpu->cpu == smp_processor_id())
10538 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10539 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10540 				backwards_tsc = true;
10541 				if (vcpu->arch.last_host_tsc > max_tsc)
10542 					max_tsc = vcpu->arch.last_host_tsc;
10543 			}
10544 		}
10545 	}
10546 
10547 	/*
10548 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10549 	 * platforms that reset TSC during suspend or hibernate actions, but
10550 	 * maintain synchronization.  We must compensate.  Fortunately, we can
10551 	 * detect that condition here, which happens early in CPU bringup,
10552 	 * before any KVM threads can be running.  Unfortunately, we can't
10553 	 * bring the TSCs fully up to date with real time, as we aren't yet far
10554 	 * enough into CPU bringup that we know how much real time has actually
10555 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10556 	 * variables that haven't been updated yet.
10557 	 *
10558 	 * So we simply find the maximum observed TSC above, then record the
10559 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
10560 	 * the adjustment will be applied.  Note that we accumulate
10561 	 * adjustments, in case multiple suspend cycles happen before some VCPU
10562 	 * gets a chance to run again.  In the event that no KVM threads get a
10563 	 * chance to run, we will miss the entire elapsed period, as we'll have
10564 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10565 	 * loose cycle time.  This isn't too big a deal, since the loss will be
10566 	 * uniform across all VCPUs (not to mention the scenario is extremely
10567 	 * unlikely). It is possible that a second hibernate recovery happens
10568 	 * much faster than a first, causing the observed TSC here to be
10569 	 * smaller; this would require additional padding adjustment, which is
10570 	 * why we set last_host_tsc to the local tsc observed here.
10571 	 *
10572 	 * N.B. - this code below runs only on platforms with reliable TSC,
10573 	 * as that is the only way backwards_tsc is set above.  Also note
10574 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10575 	 * have the same delta_cyc adjustment applied if backwards_tsc
10576 	 * is detected.  Note further, this adjustment is only done once,
10577 	 * as we reset last_host_tsc on all VCPUs to stop this from being
10578 	 * called multiple times (one for each physical CPU bringup).
10579 	 *
10580 	 * Platforms with unreliable TSCs don't have to deal with this, they
10581 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
10582 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
10583 	 * guarantee that they stay in perfect synchronization.
10584 	 */
10585 	if (backwards_tsc) {
10586 		u64 delta_cyc = max_tsc - local_tsc;
10587 		list_for_each_entry(kvm, &vm_list, vm_list) {
10588 			kvm->arch.backwards_tsc_observed = true;
10589 			kvm_for_each_vcpu(i, vcpu, kvm) {
10590 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
10591 				vcpu->arch.last_host_tsc = local_tsc;
10592 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10593 			}
10594 
10595 			/*
10596 			 * We have to disable TSC offset matching.. if you were
10597 			 * booting a VM while issuing an S4 host suspend....
10598 			 * you may have some problem.  Solving this issue is
10599 			 * left as an exercise to the reader.
10600 			 */
10601 			kvm->arch.last_tsc_nsec = 0;
10602 			kvm->arch.last_tsc_write = 0;
10603 		}
10604 
10605 	}
10606 	return 0;
10607 }
10608 
10609 void kvm_arch_hardware_disable(void)
10610 {
10611 	static_call(kvm_x86_hardware_disable)();
10612 	drop_user_return_notifiers();
10613 }
10614 
10615 int kvm_arch_hardware_setup(void *opaque)
10616 {
10617 	struct kvm_x86_init_ops *ops = opaque;
10618 	int r;
10619 
10620 	rdmsrl_safe(MSR_EFER, &host_efer);
10621 
10622 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10623 		rdmsrl(MSR_IA32_XSS, host_xss);
10624 
10625 	r = ops->hardware_setup();
10626 	if (r != 0)
10627 		return r;
10628 
10629 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10630 	kvm_ops_static_call_update();
10631 
10632 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10633 		supported_xss = 0;
10634 
10635 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10636 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10637 #undef __kvm_cpu_cap_has
10638 
10639 	if (kvm_has_tsc_control) {
10640 		/*
10641 		 * Make sure the user can only configure tsc_khz values that
10642 		 * fit into a signed integer.
10643 		 * A min value is not calculated because it will always
10644 		 * be 1 on all machines.
10645 		 */
10646 		u64 max = min(0x7fffffffULL,
10647 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10648 		kvm_max_guest_tsc_khz = max;
10649 
10650 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10651 	}
10652 
10653 	kvm_init_msr_list();
10654 	return 0;
10655 }
10656 
10657 void kvm_arch_hardware_unsetup(void)
10658 {
10659 	static_call(kvm_x86_hardware_unsetup)();
10660 }
10661 
10662 int kvm_arch_check_processor_compat(void *opaque)
10663 {
10664 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10665 	struct kvm_x86_init_ops *ops = opaque;
10666 
10667 	WARN_ON(!irqs_disabled());
10668 
10669 	if (__cr4_reserved_bits(cpu_has, c) !=
10670 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10671 		return -EIO;
10672 
10673 	return ops->check_processor_compatibility();
10674 }
10675 
10676 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10677 {
10678 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10679 }
10680 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10681 
10682 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10683 {
10684 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10685 }
10686 
10687 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10688 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10689 
10690 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10691 {
10692 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10693 
10694 	vcpu->arch.l1tf_flush_l1d = true;
10695 	if (pmu->version && unlikely(pmu->event_count)) {
10696 		pmu->need_cleanup = true;
10697 		kvm_make_request(KVM_REQ_PMU, vcpu);
10698 	}
10699 	static_call(kvm_x86_sched_in)(vcpu, cpu);
10700 }
10701 
10702 void kvm_arch_free_vm(struct kvm *kvm)
10703 {
10704 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
10705 	vfree(kvm);
10706 }
10707 
10708 
10709 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10710 {
10711 	if (type)
10712 		return -EINVAL;
10713 
10714 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10715 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10716 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10717 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10718 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10719 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10720 
10721 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10722 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10723 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10724 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10725 		&kvm->arch.irq_sources_bitmap);
10726 
10727 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10728 	mutex_init(&kvm->arch.apic_map_lock);
10729 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10730 
10731 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10732 	pvclock_update_vm_gtod_copy(kvm);
10733 
10734 	kvm->arch.guest_can_read_msr_platform_info = true;
10735 
10736 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10737 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10738 
10739 	kvm_hv_init_vm(kvm);
10740 	kvm_page_track_init(kvm);
10741 	kvm_mmu_init_vm(kvm);
10742 
10743 	return static_call(kvm_x86_vm_init)(kvm);
10744 }
10745 
10746 int kvm_arch_post_init_vm(struct kvm *kvm)
10747 {
10748 	return kvm_mmu_post_init_vm(kvm);
10749 }
10750 
10751 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10752 {
10753 	vcpu_load(vcpu);
10754 	kvm_mmu_unload(vcpu);
10755 	vcpu_put(vcpu);
10756 }
10757 
10758 static void kvm_free_vcpus(struct kvm *kvm)
10759 {
10760 	unsigned int i;
10761 	struct kvm_vcpu *vcpu;
10762 
10763 	/*
10764 	 * Unpin any mmu pages first.
10765 	 */
10766 	kvm_for_each_vcpu(i, vcpu, kvm) {
10767 		kvm_clear_async_pf_completion_queue(vcpu);
10768 		kvm_unload_vcpu_mmu(vcpu);
10769 	}
10770 	kvm_for_each_vcpu(i, vcpu, kvm)
10771 		kvm_vcpu_destroy(vcpu);
10772 
10773 	mutex_lock(&kvm->lock);
10774 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10775 		kvm->vcpus[i] = NULL;
10776 
10777 	atomic_set(&kvm->online_vcpus, 0);
10778 	mutex_unlock(&kvm->lock);
10779 }
10780 
10781 void kvm_arch_sync_events(struct kvm *kvm)
10782 {
10783 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10784 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10785 	kvm_free_pit(kvm);
10786 }
10787 
10788 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
10789 
10790 /**
10791  * __x86_set_memory_region: Setup KVM internal memory slot
10792  *
10793  * @kvm: the kvm pointer to the VM.
10794  * @id: the slot ID to setup.
10795  * @gpa: the GPA to install the slot (unused when @size == 0).
10796  * @size: the size of the slot. Set to zero to uninstall a slot.
10797  *
10798  * This function helps to setup a KVM internal memory slot.  Specify
10799  * @size > 0 to install a new slot, while @size == 0 to uninstall a
10800  * slot.  The return code can be one of the following:
10801  *
10802  *   HVA:           on success (uninstall will return a bogus HVA)
10803  *   -errno:        on error
10804  *
10805  * The caller should always use IS_ERR() to check the return value
10806  * before use.  Note, the KVM internal memory slots are guaranteed to
10807  * remain valid and unchanged until the VM is destroyed, i.e., the
10808  * GPA->HVA translation will not change.  However, the HVA is a user
10809  * address, i.e. its accessibility is not guaranteed, and must be
10810  * accessed via __copy_{to,from}_user().
10811  */
10812 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10813 				      u32 size)
10814 {
10815 	int i, r;
10816 	unsigned long hva, old_npages;
10817 	struct kvm_memslots *slots = kvm_memslots(kvm);
10818 	struct kvm_memory_slot *slot;
10819 
10820 	/* Called with kvm->slots_lock held.  */
10821 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10822 		return ERR_PTR_USR(-EINVAL);
10823 
10824 	slot = id_to_memslot(slots, id);
10825 	if (size) {
10826 		if (slot && slot->npages)
10827 			return ERR_PTR_USR(-EEXIST);
10828 
10829 		/*
10830 		 * MAP_SHARED to prevent internal slot pages from being moved
10831 		 * by fork()/COW.
10832 		 */
10833 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10834 			      MAP_SHARED | MAP_ANONYMOUS, 0);
10835 		if (IS_ERR((void *)hva))
10836 			return (void __user *)hva;
10837 	} else {
10838 		if (!slot || !slot->npages)
10839 			return NULL;
10840 
10841 		old_npages = slot->npages;
10842 		hva = slot->userspace_addr;
10843 	}
10844 
10845 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10846 		struct kvm_userspace_memory_region m;
10847 
10848 		m.slot = id | (i << 16);
10849 		m.flags = 0;
10850 		m.guest_phys_addr = gpa;
10851 		m.userspace_addr = hva;
10852 		m.memory_size = size;
10853 		r = __kvm_set_memory_region(kvm, &m);
10854 		if (r < 0)
10855 			return ERR_PTR_USR(r);
10856 	}
10857 
10858 	if (!size)
10859 		vm_munmap(hva, old_npages * PAGE_SIZE);
10860 
10861 	return (void __user *)hva;
10862 }
10863 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10864 
10865 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10866 {
10867 	kvm_mmu_pre_destroy_vm(kvm);
10868 }
10869 
10870 void kvm_arch_destroy_vm(struct kvm *kvm)
10871 {
10872 	if (current->mm == kvm->mm) {
10873 		/*
10874 		 * Free memory regions allocated on behalf of userspace,
10875 		 * unless the the memory map has changed due to process exit
10876 		 * or fd copying.
10877 		 */
10878 		mutex_lock(&kvm->slots_lock);
10879 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10880 					0, 0);
10881 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10882 					0, 0);
10883 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10884 		mutex_unlock(&kvm->slots_lock);
10885 	}
10886 	static_call_cond(kvm_x86_vm_destroy)(kvm);
10887 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10888 	kvm_pic_destroy(kvm);
10889 	kvm_ioapic_destroy(kvm);
10890 	kvm_free_vcpus(kvm);
10891 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10892 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10893 	kvm_mmu_uninit_vm(kvm);
10894 	kvm_page_track_cleanup(kvm);
10895 	kvm_xen_destroy_vm(kvm);
10896 	kvm_hv_destroy_vm(kvm);
10897 }
10898 
10899 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10900 {
10901 	int i;
10902 
10903 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10904 		kvfree(slot->arch.rmap[i]);
10905 		slot->arch.rmap[i] = NULL;
10906 
10907 		if (i == 0)
10908 			continue;
10909 
10910 		kvfree(slot->arch.lpage_info[i - 1]);
10911 		slot->arch.lpage_info[i - 1] = NULL;
10912 	}
10913 
10914 	kvm_page_track_free_memslot(slot);
10915 }
10916 
10917 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10918 				      unsigned long npages)
10919 {
10920 	int i;
10921 
10922 	/*
10923 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
10924 	 * old arrays will be freed by __kvm_set_memory_region() if installing
10925 	 * the new memslot is successful.
10926 	 */
10927 	memset(&slot->arch, 0, sizeof(slot->arch));
10928 
10929 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10930 		struct kvm_lpage_info *linfo;
10931 		unsigned long ugfn;
10932 		int lpages;
10933 		int level = i + 1;
10934 
10935 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
10936 				      slot->base_gfn, level) + 1;
10937 
10938 		slot->arch.rmap[i] =
10939 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10940 				 GFP_KERNEL_ACCOUNT);
10941 		if (!slot->arch.rmap[i])
10942 			goto out_free;
10943 		if (i == 0)
10944 			continue;
10945 
10946 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10947 		if (!linfo)
10948 			goto out_free;
10949 
10950 		slot->arch.lpage_info[i - 1] = linfo;
10951 
10952 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10953 			linfo[0].disallow_lpage = 1;
10954 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10955 			linfo[lpages - 1].disallow_lpage = 1;
10956 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
10957 		/*
10958 		 * If the gfn and userspace address are not aligned wrt each
10959 		 * other, disable large page support for this slot.
10960 		 */
10961 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10962 			unsigned long j;
10963 
10964 			for (j = 0; j < lpages; ++j)
10965 				linfo[j].disallow_lpage = 1;
10966 		}
10967 	}
10968 
10969 	if (kvm_page_track_create_memslot(slot, npages))
10970 		goto out_free;
10971 
10972 	return 0;
10973 
10974 out_free:
10975 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10976 		kvfree(slot->arch.rmap[i]);
10977 		slot->arch.rmap[i] = NULL;
10978 		if (i == 0)
10979 			continue;
10980 
10981 		kvfree(slot->arch.lpage_info[i - 1]);
10982 		slot->arch.lpage_info[i - 1] = NULL;
10983 	}
10984 	return -ENOMEM;
10985 }
10986 
10987 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10988 {
10989 	struct kvm_vcpu *vcpu;
10990 	int i;
10991 
10992 	/*
10993 	 * memslots->generation has been incremented.
10994 	 * mmio generation may have reached its maximum value.
10995 	 */
10996 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10997 
10998 	/* Force re-initialization of steal_time cache */
10999 	kvm_for_each_vcpu(i, vcpu, kvm)
11000 		kvm_vcpu_kick(vcpu);
11001 }
11002 
11003 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11004 				struct kvm_memory_slot *memslot,
11005 				const struct kvm_userspace_memory_region *mem,
11006 				enum kvm_mr_change change)
11007 {
11008 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11009 		return kvm_alloc_memslot_metadata(memslot,
11010 						  mem->memory_size >> PAGE_SHIFT);
11011 	return 0;
11012 }
11013 
11014 
11015 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11016 {
11017 	struct kvm_arch *ka = &kvm->arch;
11018 
11019 	if (!kvm_x86_ops.cpu_dirty_log_size)
11020 		return;
11021 
11022 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11023 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11024 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11025 
11026 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11027 }
11028 
11029 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11030 				     struct kvm_memory_slot *old,
11031 				     struct kvm_memory_slot *new,
11032 				     enum kvm_mr_change change)
11033 {
11034 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11035 
11036 	/*
11037 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11038 	 * applies to all operations.
11039 	 */
11040 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11041 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11042 
11043 	/*
11044 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11045 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11046 	 *
11047 	 * For a memslot with dirty logging disabled:
11048 	 * CREATE:      No dirty mappings will already exist.
11049 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11050 	 *		kvm_arch_flush_shadow_memslot()
11051 	 *
11052 	 * For a memslot with dirty logging enabled:
11053 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11054 	 *		and no dirty bits to clear.
11055 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11056 	 *		kvm_arch_flush_shadow_memslot().
11057 	 */
11058 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11059 		return;
11060 
11061 	/*
11062 	 * READONLY and non-flags changes were filtered out above, and the only
11063 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11064 	 * logging isn't being toggled on or off.
11065 	 */
11066 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11067 		return;
11068 
11069 	if (!log_dirty_pages) {
11070 		/*
11071 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11072 		 * large sptes have to be split.  If live migration succeeds,
11073 		 * the guest in the source machine will be destroyed and large
11074 		 * sptes will be created in the destination.  However, if the
11075 		 * guest continues to run in the source machine (for example if
11076 		 * live migration fails), small sptes will remain around and
11077 		 * cause bad performance.
11078 		 *
11079 		 * Scan sptes if dirty logging has been stopped, dropping those
11080 		 * which can be collapsed into a single large-page spte.  Later
11081 		 * page faults will create the large-page sptes.
11082 		 */
11083 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11084 	} else {
11085 		/* By default, write-protect everything to log writes. */
11086 		int level = PG_LEVEL_4K;
11087 
11088 		if (kvm_x86_ops.cpu_dirty_log_size) {
11089 			/*
11090 			 * Clear all dirty bits, unless pages are treated as
11091 			 * dirty from the get-go.
11092 			 */
11093 			if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
11094 				kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11095 
11096 			/*
11097 			 * Write-protect large pages on write so that dirty
11098 			 * logging happens at 4k granularity.  No need to
11099 			 * write-protect small SPTEs since write accesses are
11100 			 * logged by the CPU via dirty bits.
11101 			 */
11102 			level = PG_LEVEL_2M;
11103 		} else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
11104 			/*
11105 			 * If we're with initial-all-set, we don't need
11106 			 * to write protect any small page because
11107 			 * they're reported as dirty already.  However
11108 			 * we still need to write-protect huge pages
11109 			 * so that the page split can happen lazily on
11110 			 * the first write to the huge page.
11111 			 */
11112 			level = PG_LEVEL_2M;
11113 		}
11114 		kvm_mmu_slot_remove_write_access(kvm, new, level);
11115 	}
11116 }
11117 
11118 void kvm_arch_commit_memory_region(struct kvm *kvm,
11119 				const struct kvm_userspace_memory_region *mem,
11120 				struct kvm_memory_slot *old,
11121 				const struct kvm_memory_slot *new,
11122 				enum kvm_mr_change change)
11123 {
11124 	if (!kvm->arch.n_requested_mmu_pages)
11125 		kvm_mmu_change_mmu_pages(kvm,
11126 				kvm_mmu_calculate_default_mmu_pages(kvm));
11127 
11128 	/*
11129 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
11130 	 */
11131 	kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11132 
11133 	/* Free the arrays associated with the old memslot. */
11134 	if (change == KVM_MR_MOVE)
11135 		kvm_arch_free_memslot(kvm, old);
11136 }
11137 
11138 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11139 {
11140 	kvm_mmu_zap_all(kvm);
11141 }
11142 
11143 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11144 				   struct kvm_memory_slot *slot)
11145 {
11146 	kvm_page_track_flush_slot(kvm, slot);
11147 }
11148 
11149 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11150 {
11151 	return (is_guest_mode(vcpu) &&
11152 			kvm_x86_ops.guest_apic_has_interrupt &&
11153 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11154 }
11155 
11156 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11157 {
11158 	if (!list_empty_careful(&vcpu->async_pf.done))
11159 		return true;
11160 
11161 	if (kvm_apic_has_events(vcpu))
11162 		return true;
11163 
11164 	if (vcpu->arch.pv.pv_unhalted)
11165 		return true;
11166 
11167 	if (vcpu->arch.exception.pending)
11168 		return true;
11169 
11170 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11171 	    (vcpu->arch.nmi_pending &&
11172 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11173 		return true;
11174 
11175 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11176 	    (vcpu->arch.smi_pending &&
11177 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11178 		return true;
11179 
11180 	if (kvm_arch_interrupt_allowed(vcpu) &&
11181 	    (kvm_cpu_has_interrupt(vcpu) ||
11182 	    kvm_guest_apic_has_interrupt(vcpu)))
11183 		return true;
11184 
11185 	if (kvm_hv_has_stimer_pending(vcpu))
11186 		return true;
11187 
11188 	if (is_guest_mode(vcpu) &&
11189 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11190 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11191 		return true;
11192 
11193 	return false;
11194 }
11195 
11196 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11197 {
11198 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11199 }
11200 
11201 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11202 {
11203 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11204 		return true;
11205 
11206 	return false;
11207 }
11208 
11209 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11210 {
11211 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11212 		return true;
11213 
11214 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11215 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11216 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11217 		return true;
11218 
11219 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11220 }
11221 
11222 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11223 {
11224 	if (vcpu->arch.guest_state_protected)
11225 		return true;
11226 
11227 	return vcpu->arch.preempted_in_kernel;
11228 }
11229 
11230 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11231 {
11232 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11233 }
11234 
11235 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11236 {
11237 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11238 }
11239 
11240 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11241 {
11242 	/* Can't read the RIP when guest state is protected, just return 0 */
11243 	if (vcpu->arch.guest_state_protected)
11244 		return 0;
11245 
11246 	if (is_64_bit_mode(vcpu))
11247 		return kvm_rip_read(vcpu);
11248 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11249 		     kvm_rip_read(vcpu));
11250 }
11251 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11252 
11253 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11254 {
11255 	return kvm_get_linear_rip(vcpu) == linear_rip;
11256 }
11257 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11258 
11259 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11260 {
11261 	unsigned long rflags;
11262 
11263 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11264 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11265 		rflags &= ~X86_EFLAGS_TF;
11266 	return rflags;
11267 }
11268 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11269 
11270 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11271 {
11272 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11273 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11274 		rflags |= X86_EFLAGS_TF;
11275 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11276 }
11277 
11278 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11279 {
11280 	__kvm_set_rflags(vcpu, rflags);
11281 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11282 }
11283 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11284 
11285 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11286 {
11287 	int r;
11288 
11289 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11290 	      work->wakeup_all)
11291 		return;
11292 
11293 	r = kvm_mmu_reload(vcpu);
11294 	if (unlikely(r))
11295 		return;
11296 
11297 	if (!vcpu->arch.mmu->direct_map &&
11298 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11299 		return;
11300 
11301 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11302 }
11303 
11304 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11305 {
11306 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11307 
11308 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11309 }
11310 
11311 static inline u32 kvm_async_pf_next_probe(u32 key)
11312 {
11313 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11314 }
11315 
11316 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11317 {
11318 	u32 key = kvm_async_pf_hash_fn(gfn);
11319 
11320 	while (vcpu->arch.apf.gfns[key] != ~0)
11321 		key = kvm_async_pf_next_probe(key);
11322 
11323 	vcpu->arch.apf.gfns[key] = gfn;
11324 }
11325 
11326 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11327 {
11328 	int i;
11329 	u32 key = kvm_async_pf_hash_fn(gfn);
11330 
11331 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11332 		     (vcpu->arch.apf.gfns[key] != gfn &&
11333 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11334 		key = kvm_async_pf_next_probe(key);
11335 
11336 	return key;
11337 }
11338 
11339 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11340 {
11341 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11342 }
11343 
11344 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11345 {
11346 	u32 i, j, k;
11347 
11348 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11349 
11350 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11351 		return;
11352 
11353 	while (true) {
11354 		vcpu->arch.apf.gfns[i] = ~0;
11355 		do {
11356 			j = kvm_async_pf_next_probe(j);
11357 			if (vcpu->arch.apf.gfns[j] == ~0)
11358 				return;
11359 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11360 			/*
11361 			 * k lies cyclically in ]i,j]
11362 			 * |    i.k.j |
11363 			 * |....j i.k.| or  |.k..j i...|
11364 			 */
11365 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11366 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11367 		i = j;
11368 	}
11369 }
11370 
11371 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11372 {
11373 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11374 
11375 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11376 				      sizeof(reason));
11377 }
11378 
11379 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11380 {
11381 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11382 
11383 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11384 					     &token, offset, sizeof(token));
11385 }
11386 
11387 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11388 {
11389 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11390 	u32 val;
11391 
11392 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11393 					 &val, offset, sizeof(val)))
11394 		return false;
11395 
11396 	return !val;
11397 }
11398 
11399 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11400 {
11401 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11402 		return false;
11403 
11404 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11405 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11406 		return false;
11407 
11408 	return true;
11409 }
11410 
11411 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11412 {
11413 	if (unlikely(!lapic_in_kernel(vcpu) ||
11414 		     kvm_event_needs_reinjection(vcpu) ||
11415 		     vcpu->arch.exception.pending))
11416 		return false;
11417 
11418 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11419 		return false;
11420 
11421 	/*
11422 	 * If interrupts are off we cannot even use an artificial
11423 	 * halt state.
11424 	 */
11425 	return kvm_arch_interrupt_allowed(vcpu);
11426 }
11427 
11428 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11429 				     struct kvm_async_pf *work)
11430 {
11431 	struct x86_exception fault;
11432 
11433 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11434 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11435 
11436 	if (kvm_can_deliver_async_pf(vcpu) &&
11437 	    !apf_put_user_notpresent(vcpu)) {
11438 		fault.vector = PF_VECTOR;
11439 		fault.error_code_valid = true;
11440 		fault.error_code = 0;
11441 		fault.nested_page_fault = false;
11442 		fault.address = work->arch.token;
11443 		fault.async_page_fault = true;
11444 		kvm_inject_page_fault(vcpu, &fault);
11445 		return true;
11446 	} else {
11447 		/*
11448 		 * It is not possible to deliver a paravirtualized asynchronous
11449 		 * page fault, but putting the guest in an artificial halt state
11450 		 * can be beneficial nevertheless: if an interrupt arrives, we
11451 		 * can deliver it timely and perhaps the guest will schedule
11452 		 * another process.  When the instruction that triggered a page
11453 		 * fault is retried, hopefully the page will be ready in the host.
11454 		 */
11455 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11456 		return false;
11457 	}
11458 }
11459 
11460 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11461 				 struct kvm_async_pf *work)
11462 {
11463 	struct kvm_lapic_irq irq = {
11464 		.delivery_mode = APIC_DM_FIXED,
11465 		.vector = vcpu->arch.apf.vec
11466 	};
11467 
11468 	if (work->wakeup_all)
11469 		work->arch.token = ~0; /* broadcast wakeup */
11470 	else
11471 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11472 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11473 
11474 	if ((work->wakeup_all || work->notpresent_injected) &&
11475 	    kvm_pv_async_pf_enabled(vcpu) &&
11476 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11477 		vcpu->arch.apf.pageready_pending = true;
11478 		kvm_apic_set_irq(vcpu, &irq, NULL);
11479 	}
11480 
11481 	vcpu->arch.apf.halted = false;
11482 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11483 }
11484 
11485 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11486 {
11487 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11488 	if (!vcpu->arch.apf.pageready_pending)
11489 		kvm_vcpu_kick(vcpu);
11490 }
11491 
11492 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11493 {
11494 	if (!kvm_pv_async_pf_enabled(vcpu))
11495 		return true;
11496 	else
11497 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11498 }
11499 
11500 void kvm_arch_start_assignment(struct kvm *kvm)
11501 {
11502 	atomic_inc(&kvm->arch.assigned_device_count);
11503 }
11504 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11505 
11506 void kvm_arch_end_assignment(struct kvm *kvm)
11507 {
11508 	atomic_dec(&kvm->arch.assigned_device_count);
11509 }
11510 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11511 
11512 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11513 {
11514 	return atomic_read(&kvm->arch.assigned_device_count);
11515 }
11516 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11517 
11518 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11519 {
11520 	atomic_inc(&kvm->arch.noncoherent_dma_count);
11521 }
11522 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11523 
11524 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11525 {
11526 	atomic_dec(&kvm->arch.noncoherent_dma_count);
11527 }
11528 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11529 
11530 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11531 {
11532 	return atomic_read(&kvm->arch.noncoherent_dma_count);
11533 }
11534 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11535 
11536 bool kvm_arch_has_irq_bypass(void)
11537 {
11538 	return true;
11539 }
11540 
11541 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11542 				      struct irq_bypass_producer *prod)
11543 {
11544 	struct kvm_kernel_irqfd *irqfd =
11545 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11546 	int ret;
11547 
11548 	irqfd->producer = prod;
11549 	kvm_arch_start_assignment(irqfd->kvm);
11550 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11551 					 prod->irq, irqfd->gsi, 1);
11552 
11553 	if (ret)
11554 		kvm_arch_end_assignment(irqfd->kvm);
11555 
11556 	return ret;
11557 }
11558 
11559 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11560 				      struct irq_bypass_producer *prod)
11561 {
11562 	int ret;
11563 	struct kvm_kernel_irqfd *irqfd =
11564 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11565 
11566 	WARN_ON(irqfd->producer != prod);
11567 	irqfd->producer = NULL;
11568 
11569 	/*
11570 	 * When producer of consumer is unregistered, we change back to
11571 	 * remapped mode, so we can re-use the current implementation
11572 	 * when the irq is masked/disabled or the consumer side (KVM
11573 	 * int this case doesn't want to receive the interrupts.
11574 	*/
11575 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11576 	if (ret)
11577 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11578 		       " fails: %d\n", irqfd->consumer.token, ret);
11579 
11580 	kvm_arch_end_assignment(irqfd->kvm);
11581 }
11582 
11583 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11584 				   uint32_t guest_irq, bool set)
11585 {
11586 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11587 }
11588 
11589 bool kvm_vector_hashing_enabled(void)
11590 {
11591 	return vector_hashing;
11592 }
11593 
11594 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11595 {
11596 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11597 }
11598 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11599 
11600 
11601 int kvm_spec_ctrl_test_value(u64 value)
11602 {
11603 	/*
11604 	 * test that setting IA32_SPEC_CTRL to given value
11605 	 * is allowed by the host processor
11606 	 */
11607 
11608 	u64 saved_value;
11609 	unsigned long flags;
11610 	int ret = 0;
11611 
11612 	local_irq_save(flags);
11613 
11614 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11615 		ret = 1;
11616 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11617 		ret = 1;
11618 	else
11619 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11620 
11621 	local_irq_restore(flags);
11622 
11623 	return ret;
11624 }
11625 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11626 
11627 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11628 {
11629 	struct x86_exception fault;
11630 	u32 access = error_code &
11631 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11632 
11633 	if (!(error_code & PFERR_PRESENT_MASK) ||
11634 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11635 		/*
11636 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11637 		 * tables probably do not match the TLB.  Just proceed
11638 		 * with the error code that the processor gave.
11639 		 */
11640 		fault.vector = PF_VECTOR;
11641 		fault.error_code_valid = true;
11642 		fault.error_code = error_code;
11643 		fault.nested_page_fault = false;
11644 		fault.address = gva;
11645 	}
11646 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11647 }
11648 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11649 
11650 /*
11651  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11652  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11653  * indicates whether exit to userspace is needed.
11654  */
11655 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11656 			      struct x86_exception *e)
11657 {
11658 	if (r == X86EMUL_PROPAGATE_FAULT) {
11659 		kvm_inject_emulated_page_fault(vcpu, e);
11660 		return 1;
11661 	}
11662 
11663 	/*
11664 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11665 	 * while handling a VMX instruction KVM could've handled the request
11666 	 * correctly by exiting to userspace and performing I/O but there
11667 	 * doesn't seem to be a real use-case behind such requests, just return
11668 	 * KVM_EXIT_INTERNAL_ERROR for now.
11669 	 */
11670 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11671 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11672 	vcpu->run->internal.ndata = 0;
11673 
11674 	return 0;
11675 }
11676 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11677 
11678 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11679 {
11680 	bool pcid_enabled;
11681 	struct x86_exception e;
11682 	unsigned i;
11683 	unsigned long roots_to_free = 0;
11684 	struct {
11685 		u64 pcid;
11686 		u64 gla;
11687 	} operand;
11688 	int r;
11689 
11690 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11691 	if (r != X86EMUL_CONTINUE)
11692 		return kvm_handle_memory_failure(vcpu, r, &e);
11693 
11694 	if (operand.pcid >> 12 != 0) {
11695 		kvm_inject_gp(vcpu, 0);
11696 		return 1;
11697 	}
11698 
11699 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11700 
11701 	switch (type) {
11702 	case INVPCID_TYPE_INDIV_ADDR:
11703 		if ((!pcid_enabled && (operand.pcid != 0)) ||
11704 		    is_noncanonical_address(operand.gla, vcpu)) {
11705 			kvm_inject_gp(vcpu, 0);
11706 			return 1;
11707 		}
11708 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11709 		return kvm_skip_emulated_instruction(vcpu);
11710 
11711 	case INVPCID_TYPE_SINGLE_CTXT:
11712 		if (!pcid_enabled && (operand.pcid != 0)) {
11713 			kvm_inject_gp(vcpu, 0);
11714 			return 1;
11715 		}
11716 
11717 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11718 			kvm_mmu_sync_roots(vcpu);
11719 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11720 		}
11721 
11722 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11723 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11724 			    == operand.pcid)
11725 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11726 
11727 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11728 		/*
11729 		 * If neither the current cr3 nor any of the prev_roots use the
11730 		 * given PCID, then nothing needs to be done here because a
11731 		 * resync will happen anyway before switching to any other CR3.
11732 		 */
11733 
11734 		return kvm_skip_emulated_instruction(vcpu);
11735 
11736 	case INVPCID_TYPE_ALL_NON_GLOBAL:
11737 		/*
11738 		 * Currently, KVM doesn't mark global entries in the shadow
11739 		 * page tables, so a non-global flush just degenerates to a
11740 		 * global flush. If needed, we could optimize this later by
11741 		 * keeping track of global entries in shadow page tables.
11742 		 */
11743 
11744 		fallthrough;
11745 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
11746 		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11747 		return kvm_skip_emulated_instruction(vcpu);
11748 
11749 	default:
11750 		BUG(); /* We have already checked above that type <= 3 */
11751 	}
11752 }
11753 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11754 
11755 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11756 {
11757 	struct kvm_run *run = vcpu->run;
11758 	struct kvm_mmio_fragment *frag;
11759 	unsigned int len;
11760 
11761 	BUG_ON(!vcpu->mmio_needed);
11762 
11763 	/* Complete previous fragment */
11764 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11765 	len = min(8u, frag->len);
11766 	if (!vcpu->mmio_is_write)
11767 		memcpy(frag->data, run->mmio.data, len);
11768 
11769 	if (frag->len <= 8) {
11770 		/* Switch to the next fragment. */
11771 		frag++;
11772 		vcpu->mmio_cur_fragment++;
11773 	} else {
11774 		/* Go forward to the next mmio piece. */
11775 		frag->data += len;
11776 		frag->gpa += len;
11777 		frag->len -= len;
11778 	}
11779 
11780 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11781 		vcpu->mmio_needed = 0;
11782 
11783 		// VMG change, at this point, we're always done
11784 		// RIP has already been advanced
11785 		return 1;
11786 	}
11787 
11788 	// More MMIO is needed
11789 	run->mmio.phys_addr = frag->gpa;
11790 	run->mmio.len = min(8u, frag->len);
11791 	run->mmio.is_write = vcpu->mmio_is_write;
11792 	if (run->mmio.is_write)
11793 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11794 	run->exit_reason = KVM_EXIT_MMIO;
11795 
11796 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11797 
11798 	return 0;
11799 }
11800 
11801 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11802 			  void *data)
11803 {
11804 	int handled;
11805 	struct kvm_mmio_fragment *frag;
11806 
11807 	if (!data)
11808 		return -EINVAL;
11809 
11810 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11811 	if (handled == bytes)
11812 		return 1;
11813 
11814 	bytes -= handled;
11815 	gpa += handled;
11816 	data += handled;
11817 
11818 	/*TODO: Check if need to increment number of frags */
11819 	frag = vcpu->mmio_fragments;
11820 	vcpu->mmio_nr_fragments = 1;
11821 	frag->len = bytes;
11822 	frag->gpa = gpa;
11823 	frag->data = data;
11824 
11825 	vcpu->mmio_needed = 1;
11826 	vcpu->mmio_cur_fragment = 0;
11827 
11828 	vcpu->run->mmio.phys_addr = gpa;
11829 	vcpu->run->mmio.len = min(8u, frag->len);
11830 	vcpu->run->mmio.is_write = 1;
11831 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11832 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11833 
11834 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11835 
11836 	return 0;
11837 }
11838 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11839 
11840 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11841 			 void *data)
11842 {
11843 	int handled;
11844 	struct kvm_mmio_fragment *frag;
11845 
11846 	if (!data)
11847 		return -EINVAL;
11848 
11849 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11850 	if (handled == bytes)
11851 		return 1;
11852 
11853 	bytes -= handled;
11854 	gpa += handled;
11855 	data += handled;
11856 
11857 	/*TODO: Check if need to increment number of frags */
11858 	frag = vcpu->mmio_fragments;
11859 	vcpu->mmio_nr_fragments = 1;
11860 	frag->len = bytes;
11861 	frag->gpa = gpa;
11862 	frag->data = data;
11863 
11864 	vcpu->mmio_needed = 1;
11865 	vcpu->mmio_cur_fragment = 0;
11866 
11867 	vcpu->run->mmio.phys_addr = gpa;
11868 	vcpu->run->mmio.len = min(8u, frag->len);
11869 	vcpu->run->mmio.is_write = 0;
11870 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11871 
11872 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11873 
11874 	return 0;
11875 }
11876 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11877 
11878 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11879 {
11880 	memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11881 	       vcpu->arch.pio.count * vcpu->arch.pio.size);
11882 	vcpu->arch.pio.count = 0;
11883 
11884 	return 1;
11885 }
11886 
11887 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11888 			   unsigned int port, void *data,  unsigned int count)
11889 {
11890 	int ret;
11891 
11892 	ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11893 					data, count);
11894 	if (ret)
11895 		return ret;
11896 
11897 	vcpu->arch.pio.count = 0;
11898 
11899 	return 0;
11900 }
11901 
11902 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11903 			  unsigned int port, void *data, unsigned int count)
11904 {
11905 	int ret;
11906 
11907 	ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11908 				       data, count);
11909 	if (ret) {
11910 		vcpu->arch.pio.count = 0;
11911 	} else {
11912 		vcpu->arch.guest_ins_data = data;
11913 		vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11914 	}
11915 
11916 	return 0;
11917 }
11918 
11919 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11920 			 unsigned int port, void *data,  unsigned int count,
11921 			 int in)
11922 {
11923 	return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11924 		  : kvm_sev_es_outs(vcpu, size, port, data, count);
11925 }
11926 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11927 
11928 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11929 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11930 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11931 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11932 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11933 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11934 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11935 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11936 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11937 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11938 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11939 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11940 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11941 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11942 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11943 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11944 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11945 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11946 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11947 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11948 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11949 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11950 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11951 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11952 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11953 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11954 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
11955