xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 5f2fb52fac15a8a8e10ce020dd532504a8abfc4e)
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 "mmu.h"
22 #include "i8254.h"
23 #include "tss.h"
24 #include "kvm_cache_regs.h"
25 #include "x86.h"
26 #include "cpuid.h"
27 #include "pmu.h"
28 #include "hyperv.h"
29 
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
33 #include <linux/fs.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
56 
57 #include <trace/events/kvm.h>
58 
59 #include <asm/debugreg.h>
60 #include <asm/msr.h>
61 #include <asm/desc.h>
62 #include <asm/mce.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <asm/emulate_prefix.h>
72 #include <clocksource/hyperv_timer.h>
73 
74 #define CREATE_TRACE_POINTS
75 #include "trace.h"
76 
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 
82 #define emul_to_vcpu(ctxt) \
83 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
84 
85 /* EFER defaults:
86  * - enable syscall per default because its emulated by KVM
87  * - enable LME and LMA per default on 64 bit KVM
88  */
89 #ifdef CONFIG_X86_64
90 static
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 #else
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
94 #endif
95 
96 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
97 
98 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
99 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
100 
101 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
102                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
103 
104 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
105 static void process_nmi(struct kvm_vcpu *vcpu);
106 static void enter_smm(struct kvm_vcpu *vcpu);
107 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
108 static void store_regs(struct kvm_vcpu *vcpu);
109 static int sync_regs(struct kvm_vcpu *vcpu);
110 
111 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
112 EXPORT_SYMBOL_GPL(kvm_x86_ops);
113 
114 static bool __read_mostly ignore_msrs = 0;
115 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
116 
117 static bool __read_mostly report_ignored_msrs = true;
118 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
119 
120 unsigned int min_timer_period_us = 200;
121 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
122 
123 static bool __read_mostly kvmclock_periodic_sync = true;
124 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
125 
126 bool __read_mostly kvm_has_tsc_control;
127 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
128 u32  __read_mostly kvm_max_guest_tsc_khz;
129 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
130 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
131 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
132 u64  __read_mostly kvm_max_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
134 u64 __read_mostly kvm_default_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
136 
137 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
138 static u32 __read_mostly tsc_tolerance_ppm = 250;
139 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
140 
141 /*
142  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
143  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
144  * advancement entirely.  Any other value is used as-is and disables adaptive
145  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
146  */
147 static int __read_mostly lapic_timer_advance_ns = -1;
148 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
149 
150 static bool __read_mostly vector_hashing = true;
151 module_param(vector_hashing, bool, S_IRUGO);
152 
153 bool __read_mostly enable_vmware_backdoor = false;
154 module_param(enable_vmware_backdoor, bool, S_IRUGO);
155 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
156 
157 static bool __read_mostly force_emulation_prefix = false;
158 module_param(force_emulation_prefix, bool, S_IRUGO);
159 
160 int __read_mostly pi_inject_timer = -1;
161 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
162 
163 #define KVM_NR_SHARED_MSRS 16
164 
165 struct kvm_shared_msrs_global {
166 	int nr;
167 	u32 msrs[KVM_NR_SHARED_MSRS];
168 };
169 
170 struct kvm_shared_msrs {
171 	struct user_return_notifier urn;
172 	bool registered;
173 	struct kvm_shared_msr_values {
174 		u64 host;
175 		u64 curr;
176 	} values[KVM_NR_SHARED_MSRS];
177 };
178 
179 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
180 static struct kvm_shared_msrs __percpu *shared_msrs;
181 
182 static u64 __read_mostly host_xss;
183 
184 struct kvm_stats_debugfs_item debugfs_entries[] = {
185 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
186 	{ "pf_guest", VCPU_STAT(pf_guest) },
187 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
188 	{ "invlpg", VCPU_STAT(invlpg) },
189 	{ "exits", VCPU_STAT(exits) },
190 	{ "io_exits", VCPU_STAT(io_exits) },
191 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
192 	{ "signal_exits", VCPU_STAT(signal_exits) },
193 	{ "irq_window", VCPU_STAT(irq_window_exits) },
194 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
195 	{ "halt_exits", VCPU_STAT(halt_exits) },
196 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
197 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
198 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
199 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
200 	{ "hypercalls", VCPU_STAT(hypercalls) },
201 	{ "request_irq", VCPU_STAT(request_irq_exits) },
202 	{ "irq_exits", VCPU_STAT(irq_exits) },
203 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
204 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
205 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
206 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
207 	{ "irq_injections", VCPU_STAT(irq_injections) },
208 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
209 	{ "req_event", VCPU_STAT(req_event) },
210 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
211 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
212 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
213 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
214 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
215 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
216 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
217 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
218 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
219 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
220 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
221 	{ "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
222 	{ "max_mmu_page_hash_collisions",
223 		VM_STAT(max_mmu_page_hash_collisions) },
224 	{ NULL }
225 };
226 
227 u64 __read_mostly host_xcr0;
228 
229 struct kmem_cache *x86_fpu_cache;
230 EXPORT_SYMBOL_GPL(x86_fpu_cache);
231 
232 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
233 
234 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
235 {
236 	int i;
237 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
238 		vcpu->arch.apf.gfns[i] = ~0;
239 }
240 
241 static void kvm_on_user_return(struct user_return_notifier *urn)
242 {
243 	unsigned slot;
244 	struct kvm_shared_msrs *locals
245 		= container_of(urn, struct kvm_shared_msrs, urn);
246 	struct kvm_shared_msr_values *values;
247 	unsigned long flags;
248 
249 	/*
250 	 * Disabling irqs at this point since the following code could be
251 	 * interrupted and executed through kvm_arch_hardware_disable()
252 	 */
253 	local_irq_save(flags);
254 	if (locals->registered) {
255 		locals->registered = false;
256 		user_return_notifier_unregister(urn);
257 	}
258 	local_irq_restore(flags);
259 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
260 		values = &locals->values[slot];
261 		if (values->host != values->curr) {
262 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
263 			values->curr = values->host;
264 		}
265 	}
266 }
267 
268 void kvm_define_shared_msr(unsigned slot, u32 msr)
269 {
270 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
271 	shared_msrs_global.msrs[slot] = msr;
272 	if (slot >= shared_msrs_global.nr)
273 		shared_msrs_global.nr = slot + 1;
274 }
275 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
276 
277 static void kvm_shared_msr_cpu_online(void)
278 {
279 	unsigned int cpu = smp_processor_id();
280 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
281 	u64 value;
282 	int i;
283 
284 	for (i = 0; i < shared_msrs_global.nr; ++i) {
285 		rdmsrl_safe(shared_msrs_global.msrs[i], &value);
286 		smsr->values[i].host = value;
287 		smsr->values[i].curr = value;
288 	}
289 }
290 
291 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
292 {
293 	unsigned int cpu = smp_processor_id();
294 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
295 	int err;
296 
297 	value = (value & mask) | (smsr->values[slot].host & ~mask);
298 	if (value == smsr->values[slot].curr)
299 		return 0;
300 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
301 	if (err)
302 		return 1;
303 
304 	smsr->values[slot].curr = value;
305 	if (!smsr->registered) {
306 		smsr->urn.on_user_return = kvm_on_user_return;
307 		user_return_notifier_register(&smsr->urn);
308 		smsr->registered = true;
309 	}
310 	return 0;
311 }
312 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
313 
314 static void drop_user_return_notifiers(void)
315 {
316 	unsigned int cpu = smp_processor_id();
317 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
318 
319 	if (smsr->registered)
320 		kvm_on_user_return(&smsr->urn);
321 }
322 
323 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
324 {
325 	return vcpu->arch.apic_base;
326 }
327 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
328 
329 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
330 {
331 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
332 }
333 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
334 
335 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
336 {
337 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
338 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
339 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
340 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
341 
342 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
343 		return 1;
344 	if (!msr_info->host_initiated) {
345 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
346 			return 1;
347 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
348 			return 1;
349 	}
350 
351 	kvm_lapic_set_base(vcpu, msr_info->data);
352 	return 0;
353 }
354 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
355 
356 asmlinkage __visible void kvm_spurious_fault(void)
357 {
358 	/* Fault while not rebooting.  We want the trace. */
359 	BUG_ON(!kvm_rebooting);
360 }
361 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
362 
363 #define EXCPT_BENIGN		0
364 #define EXCPT_CONTRIBUTORY	1
365 #define EXCPT_PF		2
366 
367 static int exception_class(int vector)
368 {
369 	switch (vector) {
370 	case PF_VECTOR:
371 		return EXCPT_PF;
372 	case DE_VECTOR:
373 	case TS_VECTOR:
374 	case NP_VECTOR:
375 	case SS_VECTOR:
376 	case GP_VECTOR:
377 		return EXCPT_CONTRIBUTORY;
378 	default:
379 		break;
380 	}
381 	return EXCPT_BENIGN;
382 }
383 
384 #define EXCPT_FAULT		0
385 #define EXCPT_TRAP		1
386 #define EXCPT_ABORT		2
387 #define EXCPT_INTERRUPT		3
388 
389 static int exception_type(int vector)
390 {
391 	unsigned int mask;
392 
393 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
394 		return EXCPT_INTERRUPT;
395 
396 	mask = 1 << vector;
397 
398 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
399 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
400 		return EXCPT_TRAP;
401 
402 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
403 		return EXCPT_ABORT;
404 
405 	/* Reserved exceptions will result in fault */
406 	return EXCPT_FAULT;
407 }
408 
409 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
410 {
411 	unsigned nr = vcpu->arch.exception.nr;
412 	bool has_payload = vcpu->arch.exception.has_payload;
413 	unsigned long payload = vcpu->arch.exception.payload;
414 
415 	if (!has_payload)
416 		return;
417 
418 	switch (nr) {
419 	case DB_VECTOR:
420 		/*
421 		 * "Certain debug exceptions may clear bit 0-3.  The
422 		 * remaining contents of the DR6 register are never
423 		 * cleared by the processor".
424 		 */
425 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
426 		/*
427 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
428 		 */
429 		vcpu->arch.dr6 |= DR6_RTM;
430 		vcpu->arch.dr6 |= payload;
431 		/*
432 		 * Bit 16 should be set in the payload whenever the #DB
433 		 * exception should clear DR6.RTM. This makes the payload
434 		 * compatible with the pending debug exceptions under VMX.
435 		 * Though not currently documented in the SDM, this also
436 		 * makes the payload compatible with the exit qualification
437 		 * for #DB exceptions under VMX.
438 		 */
439 		vcpu->arch.dr6 ^= payload & DR6_RTM;
440 		break;
441 	case PF_VECTOR:
442 		vcpu->arch.cr2 = payload;
443 		break;
444 	}
445 
446 	vcpu->arch.exception.has_payload = false;
447 	vcpu->arch.exception.payload = 0;
448 }
449 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
450 
451 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
452 		unsigned nr, bool has_error, u32 error_code,
453 	        bool has_payload, unsigned long payload, bool reinject)
454 {
455 	u32 prev_nr;
456 	int class1, class2;
457 
458 	kvm_make_request(KVM_REQ_EVENT, vcpu);
459 
460 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
461 	queue:
462 		if (has_error && !is_protmode(vcpu))
463 			has_error = false;
464 		if (reinject) {
465 			/*
466 			 * On vmentry, vcpu->arch.exception.pending is only
467 			 * true if an event injection was blocked by
468 			 * nested_run_pending.  In that case, however,
469 			 * vcpu_enter_guest requests an immediate exit,
470 			 * and the guest shouldn't proceed far enough to
471 			 * need reinjection.
472 			 */
473 			WARN_ON_ONCE(vcpu->arch.exception.pending);
474 			vcpu->arch.exception.injected = true;
475 			if (WARN_ON_ONCE(has_payload)) {
476 				/*
477 				 * A reinjected event has already
478 				 * delivered its payload.
479 				 */
480 				has_payload = false;
481 				payload = 0;
482 			}
483 		} else {
484 			vcpu->arch.exception.pending = true;
485 			vcpu->arch.exception.injected = false;
486 		}
487 		vcpu->arch.exception.has_error_code = has_error;
488 		vcpu->arch.exception.nr = nr;
489 		vcpu->arch.exception.error_code = error_code;
490 		vcpu->arch.exception.has_payload = has_payload;
491 		vcpu->arch.exception.payload = payload;
492 		/*
493 		 * In guest mode, payload delivery should be deferred,
494 		 * so that the L1 hypervisor can intercept #PF before
495 		 * CR2 is modified (or intercept #DB before DR6 is
496 		 * modified under nVMX).  However, for ABI
497 		 * compatibility with KVM_GET_VCPU_EVENTS and
498 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
499 		 * delivery unless userspace has enabled this
500 		 * functionality via the per-VM capability,
501 		 * KVM_CAP_EXCEPTION_PAYLOAD.
502 		 */
503 		if (!vcpu->kvm->arch.exception_payload_enabled ||
504 		    !is_guest_mode(vcpu))
505 			kvm_deliver_exception_payload(vcpu);
506 		return;
507 	}
508 
509 	/* to check exception */
510 	prev_nr = vcpu->arch.exception.nr;
511 	if (prev_nr == DF_VECTOR) {
512 		/* triple fault -> shutdown */
513 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
514 		return;
515 	}
516 	class1 = exception_class(prev_nr);
517 	class2 = exception_class(nr);
518 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
519 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
520 		/*
521 		 * Generate double fault per SDM Table 5-5.  Set
522 		 * exception.pending = true so that the double fault
523 		 * can trigger a nested vmexit.
524 		 */
525 		vcpu->arch.exception.pending = true;
526 		vcpu->arch.exception.injected = false;
527 		vcpu->arch.exception.has_error_code = true;
528 		vcpu->arch.exception.nr = DF_VECTOR;
529 		vcpu->arch.exception.error_code = 0;
530 		vcpu->arch.exception.has_payload = false;
531 		vcpu->arch.exception.payload = 0;
532 	} else
533 		/* replace previous exception with a new one in a hope
534 		   that instruction re-execution will regenerate lost
535 		   exception */
536 		goto queue;
537 }
538 
539 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
540 {
541 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
542 }
543 EXPORT_SYMBOL_GPL(kvm_queue_exception);
544 
545 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
546 {
547 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
548 }
549 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
550 
551 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
552 				  unsigned long payload)
553 {
554 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
555 }
556 
557 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
558 				    u32 error_code, unsigned long payload)
559 {
560 	kvm_multiple_exception(vcpu, nr, true, error_code,
561 			       true, payload, false);
562 }
563 
564 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
565 {
566 	if (err)
567 		kvm_inject_gp(vcpu, 0);
568 	else
569 		return kvm_skip_emulated_instruction(vcpu);
570 
571 	return 1;
572 }
573 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
574 
575 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
576 {
577 	++vcpu->stat.pf_guest;
578 	vcpu->arch.exception.nested_apf =
579 		is_guest_mode(vcpu) && fault->async_page_fault;
580 	if (vcpu->arch.exception.nested_apf) {
581 		vcpu->arch.apf.nested_apf_token = fault->address;
582 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
583 	} else {
584 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
585 					fault->address);
586 	}
587 }
588 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
589 
590 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
591 {
592 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
593 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
594 	else
595 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
596 
597 	return fault->nested_page_fault;
598 }
599 
600 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
601 {
602 	atomic_inc(&vcpu->arch.nmi_queued);
603 	kvm_make_request(KVM_REQ_NMI, vcpu);
604 }
605 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
606 
607 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
608 {
609 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
610 }
611 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
612 
613 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
614 {
615 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
616 }
617 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
618 
619 /*
620  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
621  * a #GP and return false.
622  */
623 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
624 {
625 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
626 		return true;
627 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
628 	return false;
629 }
630 EXPORT_SYMBOL_GPL(kvm_require_cpl);
631 
632 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
633 {
634 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
635 		return true;
636 
637 	kvm_queue_exception(vcpu, UD_VECTOR);
638 	return false;
639 }
640 EXPORT_SYMBOL_GPL(kvm_require_dr);
641 
642 /*
643  * This function will be used to read from the physical memory of the currently
644  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
645  * can read from guest physical or from the guest's guest physical memory.
646  */
647 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
648 			    gfn_t ngfn, void *data, int offset, int len,
649 			    u32 access)
650 {
651 	struct x86_exception exception;
652 	gfn_t real_gfn;
653 	gpa_t ngpa;
654 
655 	ngpa     = gfn_to_gpa(ngfn);
656 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
657 	if (real_gfn == UNMAPPED_GVA)
658 		return -EFAULT;
659 
660 	real_gfn = gpa_to_gfn(real_gfn);
661 
662 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
663 }
664 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
665 
666 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
667 			       void *data, int offset, int len, u32 access)
668 {
669 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
670 				       data, offset, len, access);
671 }
672 
673 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
674 {
675 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
676 	       rsvd_bits(1, 2);
677 }
678 
679 /*
680  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
681  */
682 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
683 {
684 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
685 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
686 	int i;
687 	int ret;
688 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
689 
690 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
691 				      offset * sizeof(u64), sizeof(pdpte),
692 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
693 	if (ret < 0) {
694 		ret = 0;
695 		goto out;
696 	}
697 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
698 		if ((pdpte[i] & PT_PRESENT_MASK) &&
699 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
700 			ret = 0;
701 			goto out;
702 		}
703 	}
704 	ret = 1;
705 
706 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
707 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
708 
709 out:
710 
711 	return ret;
712 }
713 EXPORT_SYMBOL_GPL(load_pdptrs);
714 
715 bool pdptrs_changed(struct kvm_vcpu *vcpu)
716 {
717 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
718 	int offset;
719 	gfn_t gfn;
720 	int r;
721 
722 	if (!is_pae_paging(vcpu))
723 		return false;
724 
725 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
726 		return true;
727 
728 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
729 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
730 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
731 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
732 	if (r < 0)
733 		return true;
734 
735 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
736 }
737 EXPORT_SYMBOL_GPL(pdptrs_changed);
738 
739 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
740 {
741 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
742 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
743 
744 	cr0 |= X86_CR0_ET;
745 
746 #ifdef CONFIG_X86_64
747 	if (cr0 & 0xffffffff00000000UL)
748 		return 1;
749 #endif
750 
751 	cr0 &= ~CR0_RESERVED_BITS;
752 
753 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
754 		return 1;
755 
756 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
757 		return 1;
758 
759 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
760 #ifdef CONFIG_X86_64
761 		if ((vcpu->arch.efer & EFER_LME)) {
762 			int cs_db, cs_l;
763 
764 			if (!is_pae(vcpu))
765 				return 1;
766 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
767 			if (cs_l)
768 				return 1;
769 		} else
770 #endif
771 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
772 						 kvm_read_cr3(vcpu)))
773 			return 1;
774 	}
775 
776 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
777 		return 1;
778 
779 	kvm_x86_ops->set_cr0(vcpu, cr0);
780 
781 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
782 		kvm_clear_async_pf_completion_queue(vcpu);
783 		kvm_async_pf_hash_reset(vcpu);
784 	}
785 
786 	if ((cr0 ^ old_cr0) & update_bits)
787 		kvm_mmu_reset_context(vcpu);
788 
789 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
790 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
791 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
792 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
793 
794 	return 0;
795 }
796 EXPORT_SYMBOL_GPL(kvm_set_cr0);
797 
798 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
799 {
800 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
801 }
802 EXPORT_SYMBOL_GPL(kvm_lmsw);
803 
804 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
805 {
806 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
807 
808 		if (vcpu->arch.xcr0 != host_xcr0)
809 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
810 
811 		if (vcpu->arch.xsaves_enabled &&
812 		    vcpu->arch.ia32_xss != host_xss)
813 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
814 	}
815 }
816 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
817 
818 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
819 {
820 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
821 
822 		if (vcpu->arch.xcr0 != host_xcr0)
823 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
824 
825 		if (vcpu->arch.xsaves_enabled &&
826 		    vcpu->arch.ia32_xss != host_xss)
827 			wrmsrl(MSR_IA32_XSS, host_xss);
828 	}
829 
830 }
831 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
832 
833 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
834 {
835 	u64 xcr0 = xcr;
836 	u64 old_xcr0 = vcpu->arch.xcr0;
837 	u64 valid_bits;
838 
839 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
840 	if (index != XCR_XFEATURE_ENABLED_MASK)
841 		return 1;
842 	if (!(xcr0 & XFEATURE_MASK_FP))
843 		return 1;
844 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
845 		return 1;
846 
847 	/*
848 	 * Do not allow the guest to set bits that we do not support
849 	 * saving.  However, xcr0 bit 0 is always set, even if the
850 	 * emulated CPU does not support XSAVE (see fx_init).
851 	 */
852 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
853 	if (xcr0 & ~valid_bits)
854 		return 1;
855 
856 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
857 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
858 		return 1;
859 
860 	if (xcr0 & XFEATURE_MASK_AVX512) {
861 		if (!(xcr0 & XFEATURE_MASK_YMM))
862 			return 1;
863 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
864 			return 1;
865 	}
866 	vcpu->arch.xcr0 = xcr0;
867 
868 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
869 		kvm_update_cpuid(vcpu);
870 	return 0;
871 }
872 
873 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
874 {
875 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
876 	    __kvm_set_xcr(vcpu, index, xcr)) {
877 		kvm_inject_gp(vcpu, 0);
878 		return 1;
879 	}
880 	return 0;
881 }
882 EXPORT_SYMBOL_GPL(kvm_set_xcr);
883 
884 #define __cr4_reserved_bits(__cpu_has, __c)		\
885 ({							\
886 	u64 __reserved_bits = CR4_RESERVED_BITS;	\
887 							\
888 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))		\
889 		__reserved_bits |= X86_CR4_OSXSAVE;	\
890 	if (!__cpu_has(__c, X86_FEATURE_SMEP))		\
891 		__reserved_bits |= X86_CR4_SMEP;	\
892 	if (!__cpu_has(__c, X86_FEATURE_SMAP))		\
893 		__reserved_bits |= X86_CR4_SMAP;	\
894 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))	\
895 		__reserved_bits |= X86_CR4_FSGSBASE;	\
896 	if (!__cpu_has(__c, X86_FEATURE_PKU))		\
897 		__reserved_bits |= X86_CR4_PKE;		\
898 	if (!__cpu_has(__c, X86_FEATURE_LA57))		\
899 		__reserved_bits |= X86_CR4_LA57;	\
900 	__reserved_bits;				\
901 })
902 
903 static u64 kvm_host_cr4_reserved_bits(struct cpuinfo_x86 *c)
904 {
905 	u64 reserved_bits = __cr4_reserved_bits(cpu_has, c);
906 
907 	if (cpuid_ecx(0x7) & feature_bit(LA57))
908 		reserved_bits &= ~X86_CR4_LA57;
909 
910 	if (kvm_x86_ops->umip_emulated())
911 		reserved_bits &= ~X86_CR4_UMIP;
912 
913 	return reserved_bits;
914 }
915 
916 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
917 {
918 	if (cr4 & cr4_reserved_bits)
919 		return -EINVAL;
920 
921 	if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
922 		return -EINVAL;
923 
924 	return 0;
925 }
926 
927 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
928 {
929 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
930 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
931 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
932 
933 	if (kvm_valid_cr4(vcpu, cr4))
934 		return 1;
935 
936 	if (is_long_mode(vcpu)) {
937 		if (!(cr4 & X86_CR4_PAE))
938 			return 1;
939 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
940 		   && ((cr4 ^ old_cr4) & pdptr_bits)
941 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
942 				   kvm_read_cr3(vcpu)))
943 		return 1;
944 
945 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
946 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
947 			return 1;
948 
949 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
950 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
951 			return 1;
952 	}
953 
954 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
955 		return 1;
956 
957 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
958 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
959 		kvm_mmu_reset_context(vcpu);
960 
961 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
962 		kvm_update_cpuid(vcpu);
963 
964 	return 0;
965 }
966 EXPORT_SYMBOL_GPL(kvm_set_cr4);
967 
968 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
969 {
970 	bool skip_tlb_flush = false;
971 #ifdef CONFIG_X86_64
972 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
973 
974 	if (pcid_enabled) {
975 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
976 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
977 	}
978 #endif
979 
980 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
981 		if (!skip_tlb_flush) {
982 			kvm_mmu_sync_roots(vcpu);
983 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
984 		}
985 		return 0;
986 	}
987 
988 	if (is_long_mode(vcpu) &&
989 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
990 		return 1;
991 	else if (is_pae_paging(vcpu) &&
992 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
993 		return 1;
994 
995 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
996 	vcpu->arch.cr3 = cr3;
997 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
998 
999 	return 0;
1000 }
1001 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1002 
1003 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1004 {
1005 	if (cr8 & CR8_RESERVED_BITS)
1006 		return 1;
1007 	if (lapic_in_kernel(vcpu))
1008 		kvm_lapic_set_tpr(vcpu, cr8);
1009 	else
1010 		vcpu->arch.cr8 = cr8;
1011 	return 0;
1012 }
1013 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1014 
1015 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1016 {
1017 	if (lapic_in_kernel(vcpu))
1018 		return kvm_lapic_get_cr8(vcpu);
1019 	else
1020 		return vcpu->arch.cr8;
1021 }
1022 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1023 
1024 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1025 {
1026 	int i;
1027 
1028 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1029 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1030 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1031 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1032 	}
1033 }
1034 
1035 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1036 {
1037 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1038 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1039 }
1040 
1041 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1042 {
1043 	unsigned long dr7;
1044 
1045 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1046 		dr7 = vcpu->arch.guest_debug_dr7;
1047 	else
1048 		dr7 = vcpu->arch.dr7;
1049 	kvm_x86_ops->set_dr7(vcpu, dr7);
1050 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1051 	if (dr7 & DR7_BP_EN_MASK)
1052 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1053 }
1054 
1055 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1056 {
1057 	u64 fixed = DR6_FIXED_1;
1058 
1059 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1060 		fixed |= DR6_RTM;
1061 	return fixed;
1062 }
1063 
1064 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1065 {
1066 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1067 
1068 	switch (dr) {
1069 	case 0 ... 3:
1070 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1071 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1072 			vcpu->arch.eff_db[dr] = val;
1073 		break;
1074 	case 4:
1075 		/* fall through */
1076 	case 6:
1077 		if (val & 0xffffffff00000000ULL)
1078 			return -1; /* #GP */
1079 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1080 		kvm_update_dr6(vcpu);
1081 		break;
1082 	case 5:
1083 		/* fall through */
1084 	default: /* 7 */
1085 		if (!kvm_dr7_valid(val))
1086 			return -1; /* #GP */
1087 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1088 		kvm_update_dr7(vcpu);
1089 		break;
1090 	}
1091 
1092 	return 0;
1093 }
1094 
1095 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1096 {
1097 	if (__kvm_set_dr(vcpu, dr, val)) {
1098 		kvm_inject_gp(vcpu, 0);
1099 		return 1;
1100 	}
1101 	return 0;
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_set_dr);
1104 
1105 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1106 {
1107 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1108 
1109 	switch (dr) {
1110 	case 0 ... 3:
1111 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1112 		break;
1113 	case 4:
1114 		/* fall through */
1115 	case 6:
1116 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1117 			*val = vcpu->arch.dr6;
1118 		else
1119 			*val = kvm_x86_ops->get_dr6(vcpu);
1120 		break;
1121 	case 5:
1122 		/* fall through */
1123 	default: /* 7 */
1124 		*val = vcpu->arch.dr7;
1125 		break;
1126 	}
1127 	return 0;
1128 }
1129 EXPORT_SYMBOL_GPL(kvm_get_dr);
1130 
1131 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1132 {
1133 	u32 ecx = kvm_rcx_read(vcpu);
1134 	u64 data;
1135 	int err;
1136 
1137 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1138 	if (err)
1139 		return err;
1140 	kvm_rax_write(vcpu, (u32)data);
1141 	kvm_rdx_write(vcpu, data >> 32);
1142 	return err;
1143 }
1144 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1145 
1146 /*
1147  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1148  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1149  *
1150  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1151  * extract the supported MSRs from the related const lists.
1152  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1153  * capabilities of the host cpu. This capabilities test skips MSRs that are
1154  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1155  * may depend on host virtualization features rather than host cpu features.
1156  */
1157 
1158 static const u32 msrs_to_save_all[] = {
1159 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1160 	MSR_STAR,
1161 #ifdef CONFIG_X86_64
1162 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1163 #endif
1164 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1165 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1166 	MSR_IA32_SPEC_CTRL,
1167 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1168 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1169 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1170 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1171 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1172 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1173 	MSR_IA32_UMWAIT_CONTROL,
1174 
1175 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1176 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1177 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1178 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1179 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1180 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1181 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1182 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1183 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1184 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1185 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1186 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1187 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1188 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1189 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1190 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1191 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1192 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1193 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1194 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1195 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1196 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1197 };
1198 
1199 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1200 static unsigned num_msrs_to_save;
1201 
1202 static const u32 emulated_msrs_all[] = {
1203 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1204 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1205 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1206 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1207 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1208 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1209 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1210 	HV_X64_MSR_RESET,
1211 	HV_X64_MSR_VP_INDEX,
1212 	HV_X64_MSR_VP_RUNTIME,
1213 	HV_X64_MSR_SCONTROL,
1214 	HV_X64_MSR_STIMER0_CONFIG,
1215 	HV_X64_MSR_VP_ASSIST_PAGE,
1216 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1217 	HV_X64_MSR_TSC_EMULATION_STATUS,
1218 
1219 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1220 	MSR_KVM_PV_EOI_EN,
1221 
1222 	MSR_IA32_TSC_ADJUST,
1223 	MSR_IA32_TSCDEADLINE,
1224 	MSR_IA32_ARCH_CAPABILITIES,
1225 	MSR_IA32_MISC_ENABLE,
1226 	MSR_IA32_MCG_STATUS,
1227 	MSR_IA32_MCG_CTL,
1228 	MSR_IA32_MCG_EXT_CTL,
1229 	MSR_IA32_SMBASE,
1230 	MSR_SMI_COUNT,
1231 	MSR_PLATFORM_INFO,
1232 	MSR_MISC_FEATURES_ENABLES,
1233 	MSR_AMD64_VIRT_SPEC_CTRL,
1234 	MSR_IA32_POWER_CTL,
1235 	MSR_IA32_UCODE_REV,
1236 
1237 	/*
1238 	 * The following list leaves out MSRs whose values are determined
1239 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1240 	 * We always support the "true" VMX control MSRs, even if the host
1241 	 * processor does not, so I am putting these registers here rather
1242 	 * than in msrs_to_save_all.
1243 	 */
1244 	MSR_IA32_VMX_BASIC,
1245 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1246 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1247 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1248 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1249 	MSR_IA32_VMX_MISC,
1250 	MSR_IA32_VMX_CR0_FIXED0,
1251 	MSR_IA32_VMX_CR4_FIXED0,
1252 	MSR_IA32_VMX_VMCS_ENUM,
1253 	MSR_IA32_VMX_PROCBASED_CTLS2,
1254 	MSR_IA32_VMX_EPT_VPID_CAP,
1255 	MSR_IA32_VMX_VMFUNC,
1256 
1257 	MSR_K7_HWCR,
1258 	MSR_KVM_POLL_CONTROL,
1259 };
1260 
1261 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1262 static unsigned num_emulated_msrs;
1263 
1264 /*
1265  * List of msr numbers which are used to expose MSR-based features that
1266  * can be used by a hypervisor to validate requested CPU features.
1267  */
1268 static const u32 msr_based_features_all[] = {
1269 	MSR_IA32_VMX_BASIC,
1270 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1271 	MSR_IA32_VMX_PINBASED_CTLS,
1272 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1273 	MSR_IA32_VMX_PROCBASED_CTLS,
1274 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1275 	MSR_IA32_VMX_EXIT_CTLS,
1276 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1277 	MSR_IA32_VMX_ENTRY_CTLS,
1278 	MSR_IA32_VMX_MISC,
1279 	MSR_IA32_VMX_CR0_FIXED0,
1280 	MSR_IA32_VMX_CR0_FIXED1,
1281 	MSR_IA32_VMX_CR4_FIXED0,
1282 	MSR_IA32_VMX_CR4_FIXED1,
1283 	MSR_IA32_VMX_VMCS_ENUM,
1284 	MSR_IA32_VMX_PROCBASED_CTLS2,
1285 	MSR_IA32_VMX_EPT_VPID_CAP,
1286 	MSR_IA32_VMX_VMFUNC,
1287 
1288 	MSR_F10H_DECFG,
1289 	MSR_IA32_UCODE_REV,
1290 	MSR_IA32_ARCH_CAPABILITIES,
1291 };
1292 
1293 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1294 static unsigned int num_msr_based_features;
1295 
1296 static u64 kvm_get_arch_capabilities(void)
1297 {
1298 	u64 data = 0;
1299 
1300 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1301 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1302 
1303 	/*
1304 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1305 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1306 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1307 	 * L1 guests, so it need not worry about its own (L2) guests.
1308 	 */
1309 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1310 
1311 	/*
1312 	 * If we're doing cache flushes (either "always" or "cond")
1313 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1314 	 * If an outer hypervisor is doing the cache flush for us
1315 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1316 	 * capability to the guest too, and if EPT is disabled we're not
1317 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1318 	 * require a nested hypervisor to do a flush of its own.
1319 	 */
1320 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1321 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1322 
1323 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1324 		data |= ARCH_CAP_RDCL_NO;
1325 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1326 		data |= ARCH_CAP_SSB_NO;
1327 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1328 		data |= ARCH_CAP_MDS_NO;
1329 
1330 	/*
1331 	 * On TAA affected systems:
1332 	 *      - nothing to do if TSX is disabled on the host.
1333 	 *      - we emulate TSX_CTRL if present on the host.
1334 	 *	  This lets the guest use VERW to clear CPU buffers.
1335 	 */
1336 	if (!boot_cpu_has(X86_FEATURE_RTM))
1337 		data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1338 	else if (!boot_cpu_has_bug(X86_BUG_TAA))
1339 		data |= ARCH_CAP_TAA_NO;
1340 
1341 	return data;
1342 }
1343 
1344 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1345 {
1346 	switch (msr->index) {
1347 	case MSR_IA32_ARCH_CAPABILITIES:
1348 		msr->data = kvm_get_arch_capabilities();
1349 		break;
1350 	case MSR_IA32_UCODE_REV:
1351 		rdmsrl_safe(msr->index, &msr->data);
1352 		break;
1353 	default:
1354 		if (kvm_x86_ops->get_msr_feature(msr))
1355 			return 1;
1356 	}
1357 	return 0;
1358 }
1359 
1360 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1361 {
1362 	struct kvm_msr_entry msr;
1363 	int r;
1364 
1365 	msr.index = index;
1366 	r = kvm_get_msr_feature(&msr);
1367 	if (r)
1368 		return r;
1369 
1370 	*data = msr.data;
1371 
1372 	return 0;
1373 }
1374 
1375 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1376 {
1377 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1378 		return false;
1379 
1380 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1381 		return false;
1382 
1383 	if (efer & (EFER_LME | EFER_LMA) &&
1384 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1385 		return false;
1386 
1387 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1388 		return false;
1389 
1390 	return true;
1391 
1392 }
1393 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1394 {
1395 	if (efer & efer_reserved_bits)
1396 		return false;
1397 
1398 	return __kvm_valid_efer(vcpu, efer);
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1401 
1402 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1403 {
1404 	u64 old_efer = vcpu->arch.efer;
1405 	u64 efer = msr_info->data;
1406 
1407 	if (efer & efer_reserved_bits)
1408 		return 1;
1409 
1410 	if (!msr_info->host_initiated) {
1411 		if (!__kvm_valid_efer(vcpu, efer))
1412 			return 1;
1413 
1414 		if (is_paging(vcpu) &&
1415 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1416 			return 1;
1417 	}
1418 
1419 	efer &= ~EFER_LMA;
1420 	efer |= vcpu->arch.efer & EFER_LMA;
1421 
1422 	kvm_x86_ops->set_efer(vcpu, efer);
1423 
1424 	/* Update reserved bits */
1425 	if ((efer ^ old_efer) & EFER_NX)
1426 		kvm_mmu_reset_context(vcpu);
1427 
1428 	return 0;
1429 }
1430 
1431 void kvm_enable_efer_bits(u64 mask)
1432 {
1433        efer_reserved_bits &= ~mask;
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1436 
1437 /*
1438  * Write @data into the MSR specified by @index.  Select MSR specific fault
1439  * checks are bypassed if @host_initiated is %true.
1440  * Returns 0 on success, non-0 otherwise.
1441  * Assumes vcpu_load() was already called.
1442  */
1443 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1444 			 bool host_initiated)
1445 {
1446 	struct msr_data msr;
1447 
1448 	switch (index) {
1449 	case MSR_FS_BASE:
1450 	case MSR_GS_BASE:
1451 	case MSR_KERNEL_GS_BASE:
1452 	case MSR_CSTAR:
1453 	case MSR_LSTAR:
1454 		if (is_noncanonical_address(data, vcpu))
1455 			return 1;
1456 		break;
1457 	case MSR_IA32_SYSENTER_EIP:
1458 	case MSR_IA32_SYSENTER_ESP:
1459 		/*
1460 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1461 		 * non-canonical address is written on Intel but not on
1462 		 * AMD (which ignores the top 32-bits, because it does
1463 		 * not implement 64-bit SYSENTER).
1464 		 *
1465 		 * 64-bit code should hence be able to write a non-canonical
1466 		 * value on AMD.  Making the address canonical ensures that
1467 		 * vmentry does not fail on Intel after writing a non-canonical
1468 		 * value, and that something deterministic happens if the guest
1469 		 * invokes 64-bit SYSENTER.
1470 		 */
1471 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1472 	}
1473 
1474 	msr.data = data;
1475 	msr.index = index;
1476 	msr.host_initiated = host_initiated;
1477 
1478 	return kvm_x86_ops->set_msr(vcpu, &msr);
1479 }
1480 
1481 /*
1482  * Read the MSR specified by @index into @data.  Select MSR specific fault
1483  * checks are bypassed if @host_initiated is %true.
1484  * Returns 0 on success, non-0 otherwise.
1485  * Assumes vcpu_load() was already called.
1486  */
1487 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1488 		  bool host_initiated)
1489 {
1490 	struct msr_data msr;
1491 	int ret;
1492 
1493 	msr.index = index;
1494 	msr.host_initiated = host_initiated;
1495 
1496 	ret = kvm_x86_ops->get_msr(vcpu, &msr);
1497 	if (!ret)
1498 		*data = msr.data;
1499 	return ret;
1500 }
1501 
1502 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1503 {
1504 	return __kvm_get_msr(vcpu, index, data, false);
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_get_msr);
1507 
1508 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1509 {
1510 	return __kvm_set_msr(vcpu, index, data, false);
1511 }
1512 EXPORT_SYMBOL_GPL(kvm_set_msr);
1513 
1514 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1515 {
1516 	u32 ecx = kvm_rcx_read(vcpu);
1517 	u64 data;
1518 
1519 	if (kvm_get_msr(vcpu, ecx, &data)) {
1520 		trace_kvm_msr_read_ex(ecx);
1521 		kvm_inject_gp(vcpu, 0);
1522 		return 1;
1523 	}
1524 
1525 	trace_kvm_msr_read(ecx, data);
1526 
1527 	kvm_rax_write(vcpu, data & -1u);
1528 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1529 	return kvm_skip_emulated_instruction(vcpu);
1530 }
1531 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1532 
1533 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1534 {
1535 	u32 ecx = kvm_rcx_read(vcpu);
1536 	u64 data = kvm_read_edx_eax(vcpu);
1537 
1538 	if (kvm_set_msr(vcpu, ecx, data)) {
1539 		trace_kvm_msr_write_ex(ecx, data);
1540 		kvm_inject_gp(vcpu, 0);
1541 		return 1;
1542 	}
1543 
1544 	trace_kvm_msr_write(ecx, data);
1545 	return kvm_skip_emulated_instruction(vcpu);
1546 }
1547 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1548 
1549 /*
1550  * The fast path for frequent and performance sensitive wrmsr emulation,
1551  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1552  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1553  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1554  * other cases which must be called after interrupts are enabled on the host.
1555  */
1556 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1557 {
1558 	if (lapic_in_kernel(vcpu) && apic_x2apic_mode(vcpu->arch.apic) &&
1559 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1560 		((data & APIC_MODE_MASK) == APIC_DM_FIXED)) {
1561 
1562 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1563 		return kvm_lapic_reg_write(vcpu->arch.apic, APIC_ICR, (u32)data);
1564 	}
1565 
1566 	return 1;
1567 }
1568 
1569 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1570 {
1571 	u32 msr = kvm_rcx_read(vcpu);
1572 	u64 data = kvm_read_edx_eax(vcpu);
1573 	int ret = 0;
1574 
1575 	switch (msr) {
1576 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1577 		ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1578 		break;
1579 	default:
1580 		return EXIT_FASTPATH_NONE;
1581 	}
1582 
1583 	if (!ret) {
1584 		trace_kvm_msr_write(msr, data);
1585 		return EXIT_FASTPATH_SKIP_EMUL_INS;
1586 	}
1587 
1588 	return EXIT_FASTPATH_NONE;
1589 }
1590 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1591 
1592 /*
1593  * Adapt set_msr() to msr_io()'s calling convention
1594  */
1595 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1596 {
1597 	return __kvm_get_msr(vcpu, index, data, true);
1598 }
1599 
1600 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1601 {
1602 	return __kvm_set_msr(vcpu, index, *data, true);
1603 }
1604 
1605 #ifdef CONFIG_X86_64
1606 struct pvclock_clock {
1607 	int vclock_mode;
1608 	u64 cycle_last;
1609 	u64 mask;
1610 	u32 mult;
1611 	u32 shift;
1612 };
1613 
1614 struct pvclock_gtod_data {
1615 	seqcount_t	seq;
1616 
1617 	struct pvclock_clock clock; /* extract of a clocksource struct */
1618 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1619 
1620 	u64		boot_ns_raw;
1621 	u64		boot_ns;
1622 	u64		nsec_base;
1623 	u64		wall_time_sec;
1624 	u64		monotonic_raw_nsec;
1625 };
1626 
1627 static struct pvclock_gtod_data pvclock_gtod_data;
1628 
1629 static void update_pvclock_gtod(struct timekeeper *tk)
1630 {
1631 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1632 	u64 boot_ns, boot_ns_raw;
1633 
1634 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1635 	boot_ns_raw = ktime_to_ns(ktime_add(tk->tkr_raw.base, tk->offs_boot));
1636 
1637 	write_seqcount_begin(&vdata->seq);
1638 
1639 	/* copy pvclock gtod data */
1640 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1641 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1642 	vdata->clock.mask		= tk->tkr_mono.mask;
1643 	vdata->clock.mult		= tk->tkr_mono.mult;
1644 	vdata->clock.shift		= tk->tkr_mono.shift;
1645 
1646 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->archdata.vclock_mode;
1647 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1648 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1649 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1650 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1651 
1652 	vdata->boot_ns			= boot_ns;
1653 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1654 
1655 	vdata->wall_time_sec            = tk->xtime_sec;
1656 
1657 	vdata->boot_ns_raw		= boot_ns_raw;
1658 	vdata->monotonic_raw_nsec	= tk->tkr_raw.xtime_nsec;
1659 
1660 	write_seqcount_end(&vdata->seq);
1661 }
1662 #endif
1663 
1664 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1665 {
1666 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1667 	kvm_vcpu_kick(vcpu);
1668 }
1669 
1670 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1671 {
1672 	int version;
1673 	int r;
1674 	struct pvclock_wall_clock wc;
1675 	struct timespec64 boot;
1676 
1677 	if (!wall_clock)
1678 		return;
1679 
1680 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1681 	if (r)
1682 		return;
1683 
1684 	if (version & 1)
1685 		++version;  /* first time write, random junk */
1686 
1687 	++version;
1688 
1689 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1690 		return;
1691 
1692 	/*
1693 	 * The guest calculates current wall clock time by adding
1694 	 * system time (updated by kvm_guest_time_update below) to the
1695 	 * wall clock specified here.  guest system time equals host
1696 	 * system time for us, thus we must fill in host boot time here.
1697 	 */
1698 	getboottime64(&boot);
1699 
1700 	if (kvm->arch.kvmclock_offset) {
1701 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1702 		boot = timespec64_sub(boot, ts);
1703 	}
1704 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1705 	wc.nsec = boot.tv_nsec;
1706 	wc.version = version;
1707 
1708 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1709 
1710 	version++;
1711 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1712 }
1713 
1714 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1715 {
1716 	do_shl32_div32(dividend, divisor);
1717 	return dividend;
1718 }
1719 
1720 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1721 			       s8 *pshift, u32 *pmultiplier)
1722 {
1723 	uint64_t scaled64;
1724 	int32_t  shift = 0;
1725 	uint64_t tps64;
1726 	uint32_t tps32;
1727 
1728 	tps64 = base_hz;
1729 	scaled64 = scaled_hz;
1730 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1731 		tps64 >>= 1;
1732 		shift--;
1733 	}
1734 
1735 	tps32 = (uint32_t)tps64;
1736 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1737 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1738 			scaled64 >>= 1;
1739 		else
1740 			tps32 <<= 1;
1741 		shift++;
1742 	}
1743 
1744 	*pshift = shift;
1745 	*pmultiplier = div_frac(scaled64, tps32);
1746 }
1747 
1748 #ifdef CONFIG_X86_64
1749 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1750 #endif
1751 
1752 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1753 static unsigned long max_tsc_khz;
1754 
1755 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1756 {
1757 	u64 v = (u64)khz * (1000000 + ppm);
1758 	do_div(v, 1000000);
1759 	return v;
1760 }
1761 
1762 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1763 {
1764 	u64 ratio;
1765 
1766 	/* Guest TSC same frequency as host TSC? */
1767 	if (!scale) {
1768 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1769 		return 0;
1770 	}
1771 
1772 	/* TSC scaling supported? */
1773 	if (!kvm_has_tsc_control) {
1774 		if (user_tsc_khz > tsc_khz) {
1775 			vcpu->arch.tsc_catchup = 1;
1776 			vcpu->arch.tsc_always_catchup = 1;
1777 			return 0;
1778 		} else {
1779 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1780 			return -1;
1781 		}
1782 	}
1783 
1784 	/* TSC scaling required  - calculate ratio */
1785 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1786 				user_tsc_khz, tsc_khz);
1787 
1788 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1789 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1790 			            user_tsc_khz);
1791 		return -1;
1792 	}
1793 
1794 	vcpu->arch.tsc_scaling_ratio = ratio;
1795 	return 0;
1796 }
1797 
1798 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1799 {
1800 	u32 thresh_lo, thresh_hi;
1801 	int use_scaling = 0;
1802 
1803 	/* tsc_khz can be zero if TSC calibration fails */
1804 	if (user_tsc_khz == 0) {
1805 		/* set tsc_scaling_ratio to a safe value */
1806 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1807 		return -1;
1808 	}
1809 
1810 	/* Compute a scale to convert nanoseconds in TSC cycles */
1811 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1812 			   &vcpu->arch.virtual_tsc_shift,
1813 			   &vcpu->arch.virtual_tsc_mult);
1814 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1815 
1816 	/*
1817 	 * Compute the variation in TSC rate which is acceptable
1818 	 * within the range of tolerance and decide if the
1819 	 * rate being applied is within that bounds of the hardware
1820 	 * rate.  If so, no scaling or compensation need be done.
1821 	 */
1822 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1823 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1824 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1825 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1826 		use_scaling = 1;
1827 	}
1828 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1829 }
1830 
1831 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1832 {
1833 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1834 				      vcpu->arch.virtual_tsc_mult,
1835 				      vcpu->arch.virtual_tsc_shift);
1836 	tsc += vcpu->arch.this_tsc_write;
1837 	return tsc;
1838 }
1839 
1840 static inline int gtod_is_based_on_tsc(int mode)
1841 {
1842 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1843 }
1844 
1845 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1846 {
1847 #ifdef CONFIG_X86_64
1848 	bool vcpus_matched;
1849 	struct kvm_arch *ka = &vcpu->kvm->arch;
1850 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1851 
1852 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1853 			 atomic_read(&vcpu->kvm->online_vcpus));
1854 
1855 	/*
1856 	 * Once the masterclock is enabled, always perform request in
1857 	 * order to update it.
1858 	 *
1859 	 * In order to enable masterclock, the host clocksource must be TSC
1860 	 * and the vcpus need to have matched TSCs.  When that happens,
1861 	 * perform request to enable masterclock.
1862 	 */
1863 	if (ka->use_master_clock ||
1864 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1865 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1866 
1867 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1868 			    atomic_read(&vcpu->kvm->online_vcpus),
1869 		            ka->use_master_clock, gtod->clock.vclock_mode);
1870 #endif
1871 }
1872 
1873 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1874 {
1875 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1876 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1877 }
1878 
1879 /*
1880  * Multiply tsc by a fixed point number represented by ratio.
1881  *
1882  * The most significant 64-N bits (mult) of ratio represent the
1883  * integral part of the fixed point number; the remaining N bits
1884  * (frac) represent the fractional part, ie. ratio represents a fixed
1885  * point number (mult + frac * 2^(-N)).
1886  *
1887  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1888  */
1889 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1890 {
1891 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1892 }
1893 
1894 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1895 {
1896 	u64 _tsc = tsc;
1897 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1898 
1899 	if (ratio != kvm_default_tsc_scaling_ratio)
1900 		_tsc = __scale_tsc(ratio, tsc);
1901 
1902 	return _tsc;
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1905 
1906 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1907 {
1908 	u64 tsc;
1909 
1910 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1911 
1912 	return target_tsc - tsc;
1913 }
1914 
1915 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1916 {
1917 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1918 
1919 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1920 }
1921 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1922 
1923 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1924 {
1925 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1926 }
1927 
1928 static inline bool kvm_check_tsc_unstable(void)
1929 {
1930 #ifdef CONFIG_X86_64
1931 	/*
1932 	 * TSC is marked unstable when we're running on Hyper-V,
1933 	 * 'TSC page' clocksource is good.
1934 	 */
1935 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1936 		return false;
1937 #endif
1938 	return check_tsc_unstable();
1939 }
1940 
1941 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1942 {
1943 	struct kvm *kvm = vcpu->kvm;
1944 	u64 offset, ns, elapsed;
1945 	unsigned long flags;
1946 	bool matched;
1947 	bool already_matched;
1948 	u64 data = msr->data;
1949 	bool synchronizing = false;
1950 
1951 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1952 	offset = kvm_compute_tsc_offset(vcpu, data);
1953 	ns = ktime_get_boottime_ns();
1954 	elapsed = ns - kvm->arch.last_tsc_nsec;
1955 
1956 	if (vcpu->arch.virtual_tsc_khz) {
1957 		if (data == 0 && msr->host_initiated) {
1958 			/*
1959 			 * detection of vcpu initialization -- need to sync
1960 			 * with other vCPUs. This particularly helps to keep
1961 			 * kvm_clock stable after CPU hotplug
1962 			 */
1963 			synchronizing = true;
1964 		} else {
1965 			u64 tsc_exp = kvm->arch.last_tsc_write +
1966 						nsec_to_cycles(vcpu, elapsed);
1967 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1968 			/*
1969 			 * Special case: TSC write with a small delta (1 second)
1970 			 * of virtual cycle time against real time is
1971 			 * interpreted as an attempt to synchronize the CPU.
1972 			 */
1973 			synchronizing = data < tsc_exp + tsc_hz &&
1974 					data + tsc_hz > tsc_exp;
1975 		}
1976 	}
1977 
1978 	/*
1979 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1980 	 * TSC, we add elapsed time in this computation.  We could let the
1981 	 * compensation code attempt to catch up if we fall behind, but
1982 	 * it's better to try to match offsets from the beginning.
1983          */
1984 	if (synchronizing &&
1985 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1986 		if (!kvm_check_tsc_unstable()) {
1987 			offset = kvm->arch.cur_tsc_offset;
1988 		} else {
1989 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1990 			data += delta;
1991 			offset = kvm_compute_tsc_offset(vcpu, data);
1992 		}
1993 		matched = true;
1994 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1995 	} else {
1996 		/*
1997 		 * We split periods of matched TSC writes into generations.
1998 		 * For each generation, we track the original measured
1999 		 * nanosecond time, offset, and write, so if TSCs are in
2000 		 * sync, we can match exact offset, and if not, we can match
2001 		 * exact software computation in compute_guest_tsc()
2002 		 *
2003 		 * These values are tracked in kvm->arch.cur_xxx variables.
2004 		 */
2005 		kvm->arch.cur_tsc_generation++;
2006 		kvm->arch.cur_tsc_nsec = ns;
2007 		kvm->arch.cur_tsc_write = data;
2008 		kvm->arch.cur_tsc_offset = offset;
2009 		matched = false;
2010 	}
2011 
2012 	/*
2013 	 * We also track th most recent recorded KHZ, write and time to
2014 	 * allow the matching interval to be extended at each write.
2015 	 */
2016 	kvm->arch.last_tsc_nsec = ns;
2017 	kvm->arch.last_tsc_write = data;
2018 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2019 
2020 	vcpu->arch.last_guest_tsc = data;
2021 
2022 	/* Keep track of which generation this VCPU has synchronized to */
2023 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2024 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2025 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2026 
2027 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2028 		update_ia32_tsc_adjust_msr(vcpu, offset);
2029 
2030 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2031 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2032 
2033 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2034 	if (!matched) {
2035 		kvm->arch.nr_vcpus_matched_tsc = 0;
2036 	} else if (!already_matched) {
2037 		kvm->arch.nr_vcpus_matched_tsc++;
2038 	}
2039 
2040 	kvm_track_tsc_matching(vcpu);
2041 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2042 }
2043 
2044 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2045 
2046 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2047 					   s64 adjustment)
2048 {
2049 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
2050 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2051 }
2052 
2053 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2054 {
2055 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2056 		WARN_ON(adjustment < 0);
2057 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2058 	adjust_tsc_offset_guest(vcpu, adjustment);
2059 }
2060 
2061 #ifdef CONFIG_X86_64
2062 
2063 static u64 read_tsc(void)
2064 {
2065 	u64 ret = (u64)rdtsc_ordered();
2066 	u64 last = pvclock_gtod_data.clock.cycle_last;
2067 
2068 	if (likely(ret >= last))
2069 		return ret;
2070 
2071 	/*
2072 	 * GCC likes to generate cmov here, but this branch is extremely
2073 	 * predictable (it's just a function of time and the likely is
2074 	 * very likely) and there's a data dependence, so force GCC
2075 	 * to generate a branch instead.  I don't barrier() because
2076 	 * we don't actually need a barrier, and if this function
2077 	 * ever gets inlined it will generate worse code.
2078 	 */
2079 	asm volatile ("");
2080 	return last;
2081 }
2082 
2083 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2084 			  int *mode)
2085 {
2086 	long v;
2087 	u64 tsc_pg_val;
2088 
2089 	switch (clock->vclock_mode) {
2090 	case VCLOCK_HVCLOCK:
2091 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2092 						  tsc_timestamp);
2093 		if (tsc_pg_val != U64_MAX) {
2094 			/* TSC page valid */
2095 			*mode = VCLOCK_HVCLOCK;
2096 			v = (tsc_pg_val - clock->cycle_last) &
2097 				clock->mask;
2098 		} else {
2099 			/* TSC page invalid */
2100 			*mode = VCLOCK_NONE;
2101 		}
2102 		break;
2103 	case VCLOCK_TSC:
2104 		*mode = VCLOCK_TSC;
2105 		*tsc_timestamp = read_tsc();
2106 		v = (*tsc_timestamp - clock->cycle_last) &
2107 			clock->mask;
2108 		break;
2109 	default:
2110 		*mode = VCLOCK_NONE;
2111 	}
2112 
2113 	if (*mode == VCLOCK_NONE)
2114 		*tsc_timestamp = v = 0;
2115 
2116 	return v * clock->mult;
2117 }
2118 
2119 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2120 {
2121 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2122 	unsigned long seq;
2123 	int mode;
2124 	u64 ns;
2125 
2126 	do {
2127 		seq = read_seqcount_begin(&gtod->seq);
2128 		ns = gtod->monotonic_raw_nsec;
2129 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2130 		ns >>= gtod->clock.shift;
2131 		ns += gtod->boot_ns_raw;
2132 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2133 	*t = ns;
2134 
2135 	return mode;
2136 }
2137 
2138 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2139 {
2140 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2141 	unsigned long seq;
2142 	int mode;
2143 	u64 ns;
2144 
2145 	do {
2146 		seq = read_seqcount_begin(&gtod->seq);
2147 		ts->tv_sec = gtod->wall_time_sec;
2148 		ns = gtod->nsec_base;
2149 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2150 		ns >>= gtod->clock.shift;
2151 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2152 
2153 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2154 	ts->tv_nsec = ns;
2155 
2156 	return mode;
2157 }
2158 
2159 /* returns true if host is using TSC based clocksource */
2160 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2161 {
2162 	/* checked again under seqlock below */
2163 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2164 		return false;
2165 
2166 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2167 						      tsc_timestamp));
2168 }
2169 
2170 /* returns true if host is using TSC based clocksource */
2171 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2172 					   u64 *tsc_timestamp)
2173 {
2174 	/* checked again under seqlock below */
2175 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2176 		return false;
2177 
2178 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2179 }
2180 #endif
2181 
2182 /*
2183  *
2184  * Assuming a stable TSC across physical CPUS, and a stable TSC
2185  * across virtual CPUs, the following condition is possible.
2186  * Each numbered line represents an event visible to both
2187  * CPUs at the next numbered event.
2188  *
2189  * "timespecX" represents host monotonic time. "tscX" represents
2190  * RDTSC value.
2191  *
2192  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2193  *
2194  * 1.  read timespec0,tsc0
2195  * 2.					| timespec1 = timespec0 + N
2196  * 					| tsc1 = tsc0 + M
2197  * 3. transition to guest		| transition to guest
2198  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2199  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2200  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2201  *
2202  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2203  *
2204  * 	- ret0 < ret1
2205  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2206  *		...
2207  *	- 0 < N - M => M < N
2208  *
2209  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2210  * always the case (the difference between two distinct xtime instances
2211  * might be smaller then the difference between corresponding TSC reads,
2212  * when updating guest vcpus pvclock areas).
2213  *
2214  * To avoid that problem, do not allow visibility of distinct
2215  * system_timestamp/tsc_timestamp values simultaneously: use a master
2216  * copy of host monotonic time values. Update that master copy
2217  * in lockstep.
2218  *
2219  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2220  *
2221  */
2222 
2223 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2224 {
2225 #ifdef CONFIG_X86_64
2226 	struct kvm_arch *ka = &kvm->arch;
2227 	int vclock_mode;
2228 	bool host_tsc_clocksource, vcpus_matched;
2229 
2230 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2231 			atomic_read(&kvm->online_vcpus));
2232 
2233 	/*
2234 	 * If the host uses TSC clock, then passthrough TSC as stable
2235 	 * to the guest.
2236 	 */
2237 	host_tsc_clocksource = kvm_get_time_and_clockread(
2238 					&ka->master_kernel_ns,
2239 					&ka->master_cycle_now);
2240 
2241 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2242 				&& !ka->backwards_tsc_observed
2243 				&& !ka->boot_vcpu_runs_old_kvmclock;
2244 
2245 	if (ka->use_master_clock)
2246 		atomic_set(&kvm_guest_has_master_clock, 1);
2247 
2248 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2249 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2250 					vcpus_matched);
2251 #endif
2252 }
2253 
2254 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2255 {
2256 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2257 }
2258 
2259 static void kvm_gen_update_masterclock(struct kvm *kvm)
2260 {
2261 #ifdef CONFIG_X86_64
2262 	int i;
2263 	struct kvm_vcpu *vcpu;
2264 	struct kvm_arch *ka = &kvm->arch;
2265 
2266 	spin_lock(&ka->pvclock_gtod_sync_lock);
2267 	kvm_make_mclock_inprogress_request(kvm);
2268 	/* no guest entries from this point */
2269 	pvclock_update_vm_gtod_copy(kvm);
2270 
2271 	kvm_for_each_vcpu(i, vcpu, kvm)
2272 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2273 
2274 	/* guest entries allowed */
2275 	kvm_for_each_vcpu(i, vcpu, kvm)
2276 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2277 
2278 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2279 #endif
2280 }
2281 
2282 u64 get_kvmclock_ns(struct kvm *kvm)
2283 {
2284 	struct kvm_arch *ka = &kvm->arch;
2285 	struct pvclock_vcpu_time_info hv_clock;
2286 	u64 ret;
2287 
2288 	spin_lock(&ka->pvclock_gtod_sync_lock);
2289 	if (!ka->use_master_clock) {
2290 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2291 		return ktime_get_boottime_ns() + ka->kvmclock_offset;
2292 	}
2293 
2294 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2295 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2296 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2297 
2298 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2299 	get_cpu();
2300 
2301 	if (__this_cpu_read(cpu_tsc_khz)) {
2302 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2303 				   &hv_clock.tsc_shift,
2304 				   &hv_clock.tsc_to_system_mul);
2305 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2306 	} else
2307 		ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2308 
2309 	put_cpu();
2310 
2311 	return ret;
2312 }
2313 
2314 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2315 {
2316 	struct kvm_vcpu_arch *vcpu = &v->arch;
2317 	struct pvclock_vcpu_time_info guest_hv_clock;
2318 
2319 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2320 		&guest_hv_clock, sizeof(guest_hv_clock))))
2321 		return;
2322 
2323 	/* This VCPU is paused, but it's legal for a guest to read another
2324 	 * VCPU's kvmclock, so we really have to follow the specification where
2325 	 * it says that version is odd if data is being modified, and even after
2326 	 * it is consistent.
2327 	 *
2328 	 * Version field updates must be kept separate.  This is because
2329 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2330 	 * writes within a string instruction are weakly ordered.  So there
2331 	 * are three writes overall.
2332 	 *
2333 	 * As a small optimization, only write the version field in the first
2334 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2335 	 * version field is the first in the struct.
2336 	 */
2337 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2338 
2339 	if (guest_hv_clock.version & 1)
2340 		++guest_hv_clock.version;  /* first time write, random junk */
2341 
2342 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2343 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2344 				&vcpu->hv_clock,
2345 				sizeof(vcpu->hv_clock.version));
2346 
2347 	smp_wmb();
2348 
2349 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2350 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2351 
2352 	if (vcpu->pvclock_set_guest_stopped_request) {
2353 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2354 		vcpu->pvclock_set_guest_stopped_request = false;
2355 	}
2356 
2357 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2358 
2359 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2360 				&vcpu->hv_clock,
2361 				sizeof(vcpu->hv_clock));
2362 
2363 	smp_wmb();
2364 
2365 	vcpu->hv_clock.version++;
2366 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2367 				&vcpu->hv_clock,
2368 				sizeof(vcpu->hv_clock.version));
2369 }
2370 
2371 static int kvm_guest_time_update(struct kvm_vcpu *v)
2372 {
2373 	unsigned long flags, tgt_tsc_khz;
2374 	struct kvm_vcpu_arch *vcpu = &v->arch;
2375 	struct kvm_arch *ka = &v->kvm->arch;
2376 	s64 kernel_ns;
2377 	u64 tsc_timestamp, host_tsc;
2378 	u8 pvclock_flags;
2379 	bool use_master_clock;
2380 
2381 	kernel_ns = 0;
2382 	host_tsc = 0;
2383 
2384 	/*
2385 	 * If the host uses TSC clock, then passthrough TSC as stable
2386 	 * to the guest.
2387 	 */
2388 	spin_lock(&ka->pvclock_gtod_sync_lock);
2389 	use_master_clock = ka->use_master_clock;
2390 	if (use_master_clock) {
2391 		host_tsc = ka->master_cycle_now;
2392 		kernel_ns = ka->master_kernel_ns;
2393 	}
2394 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2395 
2396 	/* Keep irq disabled to prevent changes to the clock */
2397 	local_irq_save(flags);
2398 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2399 	if (unlikely(tgt_tsc_khz == 0)) {
2400 		local_irq_restore(flags);
2401 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2402 		return 1;
2403 	}
2404 	if (!use_master_clock) {
2405 		host_tsc = rdtsc();
2406 		kernel_ns = ktime_get_boottime_ns();
2407 	}
2408 
2409 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2410 
2411 	/*
2412 	 * We may have to catch up the TSC to match elapsed wall clock
2413 	 * time for two reasons, even if kvmclock is used.
2414 	 *   1) CPU could have been running below the maximum TSC rate
2415 	 *   2) Broken TSC compensation resets the base at each VCPU
2416 	 *      entry to avoid unknown leaps of TSC even when running
2417 	 *      again on the same CPU.  This may cause apparent elapsed
2418 	 *      time to disappear, and the guest to stand still or run
2419 	 *	very slowly.
2420 	 */
2421 	if (vcpu->tsc_catchup) {
2422 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2423 		if (tsc > tsc_timestamp) {
2424 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2425 			tsc_timestamp = tsc;
2426 		}
2427 	}
2428 
2429 	local_irq_restore(flags);
2430 
2431 	/* With all the info we got, fill in the values */
2432 
2433 	if (kvm_has_tsc_control)
2434 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2435 
2436 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2437 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2438 				   &vcpu->hv_clock.tsc_shift,
2439 				   &vcpu->hv_clock.tsc_to_system_mul);
2440 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2441 	}
2442 
2443 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2444 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2445 	vcpu->last_guest_tsc = tsc_timestamp;
2446 
2447 	/* If the host uses TSC clocksource, then it is stable */
2448 	pvclock_flags = 0;
2449 	if (use_master_clock)
2450 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2451 
2452 	vcpu->hv_clock.flags = pvclock_flags;
2453 
2454 	if (vcpu->pv_time_enabled)
2455 		kvm_setup_pvclock_page(v);
2456 	if (v == kvm_get_vcpu(v->kvm, 0))
2457 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2458 	return 0;
2459 }
2460 
2461 /*
2462  * kvmclock updates which are isolated to a given vcpu, such as
2463  * vcpu->cpu migration, should not allow system_timestamp from
2464  * the rest of the vcpus to remain static. Otherwise ntp frequency
2465  * correction applies to one vcpu's system_timestamp but not
2466  * the others.
2467  *
2468  * So in those cases, request a kvmclock update for all vcpus.
2469  * We need to rate-limit these requests though, as they can
2470  * considerably slow guests that have a large number of vcpus.
2471  * The time for a remote vcpu to update its kvmclock is bound
2472  * by the delay we use to rate-limit the updates.
2473  */
2474 
2475 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2476 
2477 static void kvmclock_update_fn(struct work_struct *work)
2478 {
2479 	int i;
2480 	struct delayed_work *dwork = to_delayed_work(work);
2481 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2482 					   kvmclock_update_work);
2483 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2484 	struct kvm_vcpu *vcpu;
2485 
2486 	kvm_for_each_vcpu(i, vcpu, kvm) {
2487 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2488 		kvm_vcpu_kick(vcpu);
2489 	}
2490 }
2491 
2492 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2493 {
2494 	struct kvm *kvm = v->kvm;
2495 
2496 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2497 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2498 					KVMCLOCK_UPDATE_DELAY);
2499 }
2500 
2501 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2502 
2503 static void kvmclock_sync_fn(struct work_struct *work)
2504 {
2505 	struct delayed_work *dwork = to_delayed_work(work);
2506 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2507 					   kvmclock_sync_work);
2508 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2509 
2510 	if (!kvmclock_periodic_sync)
2511 		return;
2512 
2513 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2514 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2515 					KVMCLOCK_SYNC_PERIOD);
2516 }
2517 
2518 /*
2519  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2520  */
2521 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2522 {
2523 	/* McStatusWrEn enabled? */
2524 	if (guest_cpuid_is_amd(vcpu))
2525 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2526 
2527 	return false;
2528 }
2529 
2530 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2531 {
2532 	u64 mcg_cap = vcpu->arch.mcg_cap;
2533 	unsigned bank_num = mcg_cap & 0xff;
2534 	u32 msr = msr_info->index;
2535 	u64 data = msr_info->data;
2536 
2537 	switch (msr) {
2538 	case MSR_IA32_MCG_STATUS:
2539 		vcpu->arch.mcg_status = data;
2540 		break;
2541 	case MSR_IA32_MCG_CTL:
2542 		if (!(mcg_cap & MCG_CTL_P) &&
2543 		    (data || !msr_info->host_initiated))
2544 			return 1;
2545 		if (data != 0 && data != ~(u64)0)
2546 			return 1;
2547 		vcpu->arch.mcg_ctl = data;
2548 		break;
2549 	default:
2550 		if (msr >= MSR_IA32_MC0_CTL &&
2551 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2552 			u32 offset = array_index_nospec(
2553 				msr - MSR_IA32_MC0_CTL,
2554 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2555 
2556 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2557 			 * some Linux kernels though clear bit 10 in bank 4 to
2558 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2559 			 * this to avoid an uncatched #GP in the guest
2560 			 */
2561 			if ((offset & 0x3) == 0 &&
2562 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2563 				return -1;
2564 
2565 			/* MCi_STATUS */
2566 			if (!msr_info->host_initiated &&
2567 			    (offset & 0x3) == 1 && data != 0) {
2568 				if (!can_set_mci_status(vcpu))
2569 					return -1;
2570 			}
2571 
2572 			vcpu->arch.mce_banks[offset] = data;
2573 			break;
2574 		}
2575 		return 1;
2576 	}
2577 	return 0;
2578 }
2579 
2580 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2581 {
2582 	struct kvm *kvm = vcpu->kvm;
2583 	int lm = is_long_mode(vcpu);
2584 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2585 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2586 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2587 		: kvm->arch.xen_hvm_config.blob_size_32;
2588 	u32 page_num = data & ~PAGE_MASK;
2589 	u64 page_addr = data & PAGE_MASK;
2590 	u8 *page;
2591 	int r;
2592 
2593 	r = -E2BIG;
2594 	if (page_num >= blob_size)
2595 		goto out;
2596 	r = -ENOMEM;
2597 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2598 	if (IS_ERR(page)) {
2599 		r = PTR_ERR(page);
2600 		goto out;
2601 	}
2602 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2603 		goto out_free;
2604 	r = 0;
2605 out_free:
2606 	kfree(page);
2607 out:
2608 	return r;
2609 }
2610 
2611 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2612 {
2613 	gpa_t gpa = data & ~0x3f;
2614 
2615 	/* Bits 3:5 are reserved, Should be zero */
2616 	if (data & 0x38)
2617 		return 1;
2618 
2619 	vcpu->arch.apf.msr_val = data;
2620 
2621 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2622 		kvm_clear_async_pf_completion_queue(vcpu);
2623 		kvm_async_pf_hash_reset(vcpu);
2624 		return 0;
2625 	}
2626 
2627 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2628 					sizeof(u32)))
2629 		return 1;
2630 
2631 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2632 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2633 	kvm_async_pf_wakeup_all(vcpu);
2634 	return 0;
2635 }
2636 
2637 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2638 {
2639 	vcpu->arch.pv_time_enabled = false;
2640 	vcpu->arch.time = 0;
2641 }
2642 
2643 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2644 {
2645 	++vcpu->stat.tlb_flush;
2646 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2647 }
2648 
2649 static void record_steal_time(struct kvm_vcpu *vcpu)
2650 {
2651 	struct kvm_host_map map;
2652 	struct kvm_steal_time *st;
2653 
2654 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2655 		return;
2656 
2657 	/* -EAGAIN is returned in atomic context so we can just return. */
2658 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2659 			&map, &vcpu->arch.st.cache, false))
2660 		return;
2661 
2662 	st = map.hva +
2663 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2664 
2665 	/*
2666 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2667 	 * expensive IPIs.
2668 	 */
2669 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2670 		st->preempted & KVM_VCPU_FLUSH_TLB);
2671 	if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2672 		kvm_vcpu_flush_tlb(vcpu, false);
2673 
2674 	vcpu->arch.st.preempted = 0;
2675 
2676 	if (st->version & 1)
2677 		st->version += 1;  /* first time write, random junk */
2678 
2679 	st->version += 1;
2680 
2681 	smp_wmb();
2682 
2683 	st->steal += current->sched_info.run_delay -
2684 		vcpu->arch.st.last_steal;
2685 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2686 
2687 	smp_wmb();
2688 
2689 	st->version += 1;
2690 
2691 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2692 }
2693 
2694 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2695 {
2696 	bool pr = false;
2697 	u32 msr = msr_info->index;
2698 	u64 data = msr_info->data;
2699 
2700 	switch (msr) {
2701 	case MSR_AMD64_NB_CFG:
2702 	case MSR_IA32_UCODE_WRITE:
2703 	case MSR_VM_HSAVE_PA:
2704 	case MSR_AMD64_PATCH_LOADER:
2705 	case MSR_AMD64_BU_CFG2:
2706 	case MSR_AMD64_DC_CFG:
2707 	case MSR_F15H_EX_CFG:
2708 		break;
2709 
2710 	case MSR_IA32_UCODE_REV:
2711 		if (msr_info->host_initiated)
2712 			vcpu->arch.microcode_version = data;
2713 		break;
2714 	case MSR_IA32_ARCH_CAPABILITIES:
2715 		if (!msr_info->host_initiated)
2716 			return 1;
2717 		vcpu->arch.arch_capabilities = data;
2718 		break;
2719 	case MSR_EFER:
2720 		return set_efer(vcpu, msr_info);
2721 	case MSR_K7_HWCR:
2722 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2723 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2724 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2725 
2726 		/* Handle McStatusWrEn */
2727 		if (data == BIT_ULL(18)) {
2728 			vcpu->arch.msr_hwcr = data;
2729 		} else if (data != 0) {
2730 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2731 				    data);
2732 			return 1;
2733 		}
2734 		break;
2735 	case MSR_FAM10H_MMIO_CONF_BASE:
2736 		if (data != 0) {
2737 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2738 				    "0x%llx\n", data);
2739 			return 1;
2740 		}
2741 		break;
2742 	case MSR_IA32_DEBUGCTLMSR:
2743 		if (!data) {
2744 			/* We support the non-activated case already */
2745 			break;
2746 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2747 			/* Values other than LBR and BTF are vendor-specific,
2748 			   thus reserved and should throw a #GP */
2749 			return 1;
2750 		}
2751 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2752 			    __func__, data);
2753 		break;
2754 	case 0x200 ... 0x2ff:
2755 		return kvm_mtrr_set_msr(vcpu, msr, data);
2756 	case MSR_IA32_APICBASE:
2757 		return kvm_set_apic_base(vcpu, msr_info);
2758 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2759 		return kvm_x2apic_msr_write(vcpu, msr, data);
2760 	case MSR_IA32_TSCDEADLINE:
2761 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2762 		break;
2763 	case MSR_IA32_TSC_ADJUST:
2764 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2765 			if (!msr_info->host_initiated) {
2766 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2767 				adjust_tsc_offset_guest(vcpu, adj);
2768 			}
2769 			vcpu->arch.ia32_tsc_adjust_msr = data;
2770 		}
2771 		break;
2772 	case MSR_IA32_MISC_ENABLE:
2773 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2774 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2775 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2776 				return 1;
2777 			vcpu->arch.ia32_misc_enable_msr = data;
2778 			kvm_update_cpuid(vcpu);
2779 		} else {
2780 			vcpu->arch.ia32_misc_enable_msr = data;
2781 		}
2782 		break;
2783 	case MSR_IA32_SMBASE:
2784 		if (!msr_info->host_initiated)
2785 			return 1;
2786 		vcpu->arch.smbase = data;
2787 		break;
2788 	case MSR_IA32_POWER_CTL:
2789 		vcpu->arch.msr_ia32_power_ctl = data;
2790 		break;
2791 	case MSR_IA32_TSC:
2792 		kvm_write_tsc(vcpu, msr_info);
2793 		break;
2794 	case MSR_IA32_XSS:
2795 		if (!msr_info->host_initiated &&
2796 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2797 			return 1;
2798 		/*
2799 		 * We do support PT if kvm_x86_ops->pt_supported(), but we do
2800 		 * not support IA32_XSS[bit 8]. Guests will have to use
2801 		 * RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT
2802 		 * MSRs.
2803 		 */
2804 		if (data != 0)
2805 			return 1;
2806 		vcpu->arch.ia32_xss = data;
2807 		break;
2808 	case MSR_SMI_COUNT:
2809 		if (!msr_info->host_initiated)
2810 			return 1;
2811 		vcpu->arch.smi_count = data;
2812 		break;
2813 	case MSR_KVM_WALL_CLOCK_NEW:
2814 	case MSR_KVM_WALL_CLOCK:
2815 		vcpu->kvm->arch.wall_clock = data;
2816 		kvm_write_wall_clock(vcpu->kvm, data);
2817 		break;
2818 	case MSR_KVM_SYSTEM_TIME_NEW:
2819 	case MSR_KVM_SYSTEM_TIME: {
2820 		struct kvm_arch *ka = &vcpu->kvm->arch;
2821 
2822 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2823 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2824 
2825 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2826 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2827 
2828 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2829 		}
2830 
2831 		vcpu->arch.time = data;
2832 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2833 
2834 		/* we verify if the enable bit is set... */
2835 		vcpu->arch.pv_time_enabled = false;
2836 		if (!(data & 1))
2837 			break;
2838 
2839 		if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2840 		     &vcpu->arch.pv_time, data & ~1ULL,
2841 		     sizeof(struct pvclock_vcpu_time_info)))
2842 			vcpu->arch.pv_time_enabled = true;
2843 
2844 		break;
2845 	}
2846 	case MSR_KVM_ASYNC_PF_EN:
2847 		if (kvm_pv_enable_async_pf(vcpu, data))
2848 			return 1;
2849 		break;
2850 	case MSR_KVM_STEAL_TIME:
2851 
2852 		if (unlikely(!sched_info_on()))
2853 			return 1;
2854 
2855 		if (data & KVM_STEAL_RESERVED_MASK)
2856 			return 1;
2857 
2858 		vcpu->arch.st.msr_val = data;
2859 
2860 		if (!(data & KVM_MSR_ENABLED))
2861 			break;
2862 
2863 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2864 
2865 		break;
2866 	case MSR_KVM_PV_EOI_EN:
2867 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2868 			return 1;
2869 		break;
2870 
2871 	case MSR_KVM_POLL_CONTROL:
2872 		/* only enable bit supported */
2873 		if (data & (-1ULL << 1))
2874 			return 1;
2875 
2876 		vcpu->arch.msr_kvm_poll_control = data;
2877 		break;
2878 
2879 	case MSR_IA32_MCG_CTL:
2880 	case MSR_IA32_MCG_STATUS:
2881 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2882 		return set_msr_mce(vcpu, msr_info);
2883 
2884 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2885 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2886 		pr = true; /* fall through */
2887 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2888 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2889 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2890 			return kvm_pmu_set_msr(vcpu, msr_info);
2891 
2892 		if (pr || data != 0)
2893 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2894 				    "0x%x data 0x%llx\n", msr, data);
2895 		break;
2896 	case MSR_K7_CLK_CTL:
2897 		/*
2898 		 * Ignore all writes to this no longer documented MSR.
2899 		 * Writes are only relevant for old K7 processors,
2900 		 * all pre-dating SVM, but a recommended workaround from
2901 		 * AMD for these chips. It is possible to specify the
2902 		 * affected processor models on the command line, hence
2903 		 * the need to ignore the workaround.
2904 		 */
2905 		break;
2906 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2907 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2908 	case HV_X64_MSR_CRASH_CTL:
2909 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2910 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2911 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2912 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2913 		return kvm_hv_set_msr_common(vcpu, msr, data,
2914 					     msr_info->host_initiated);
2915 	case MSR_IA32_BBL_CR_CTL3:
2916 		/* Drop writes to this legacy MSR -- see rdmsr
2917 		 * counterpart for further detail.
2918 		 */
2919 		if (report_ignored_msrs)
2920 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2921 				msr, data);
2922 		break;
2923 	case MSR_AMD64_OSVW_ID_LENGTH:
2924 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2925 			return 1;
2926 		vcpu->arch.osvw.length = data;
2927 		break;
2928 	case MSR_AMD64_OSVW_STATUS:
2929 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2930 			return 1;
2931 		vcpu->arch.osvw.status = data;
2932 		break;
2933 	case MSR_PLATFORM_INFO:
2934 		if (!msr_info->host_initiated ||
2935 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2936 		     cpuid_fault_enabled(vcpu)))
2937 			return 1;
2938 		vcpu->arch.msr_platform_info = data;
2939 		break;
2940 	case MSR_MISC_FEATURES_ENABLES:
2941 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2942 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2943 		     !supports_cpuid_fault(vcpu)))
2944 			return 1;
2945 		vcpu->arch.msr_misc_features_enables = data;
2946 		break;
2947 	default:
2948 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2949 			return xen_hvm_config(vcpu, data);
2950 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2951 			return kvm_pmu_set_msr(vcpu, msr_info);
2952 		if (!ignore_msrs) {
2953 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2954 				    msr, data);
2955 			return 1;
2956 		} else {
2957 			if (report_ignored_msrs)
2958 				vcpu_unimpl(vcpu,
2959 					"ignored wrmsr: 0x%x data 0x%llx\n",
2960 					msr, data);
2961 			break;
2962 		}
2963 	}
2964 	return 0;
2965 }
2966 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2967 
2968 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2969 {
2970 	u64 data;
2971 	u64 mcg_cap = vcpu->arch.mcg_cap;
2972 	unsigned bank_num = mcg_cap & 0xff;
2973 
2974 	switch (msr) {
2975 	case MSR_IA32_P5_MC_ADDR:
2976 	case MSR_IA32_P5_MC_TYPE:
2977 		data = 0;
2978 		break;
2979 	case MSR_IA32_MCG_CAP:
2980 		data = vcpu->arch.mcg_cap;
2981 		break;
2982 	case MSR_IA32_MCG_CTL:
2983 		if (!(mcg_cap & MCG_CTL_P) && !host)
2984 			return 1;
2985 		data = vcpu->arch.mcg_ctl;
2986 		break;
2987 	case MSR_IA32_MCG_STATUS:
2988 		data = vcpu->arch.mcg_status;
2989 		break;
2990 	default:
2991 		if (msr >= MSR_IA32_MC0_CTL &&
2992 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2993 			u32 offset = array_index_nospec(
2994 				msr - MSR_IA32_MC0_CTL,
2995 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2996 
2997 			data = vcpu->arch.mce_banks[offset];
2998 			break;
2999 		}
3000 		return 1;
3001 	}
3002 	*pdata = data;
3003 	return 0;
3004 }
3005 
3006 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3007 {
3008 	switch (msr_info->index) {
3009 	case MSR_IA32_PLATFORM_ID:
3010 	case MSR_IA32_EBL_CR_POWERON:
3011 	case MSR_IA32_DEBUGCTLMSR:
3012 	case MSR_IA32_LASTBRANCHFROMIP:
3013 	case MSR_IA32_LASTBRANCHTOIP:
3014 	case MSR_IA32_LASTINTFROMIP:
3015 	case MSR_IA32_LASTINTTOIP:
3016 	case MSR_K8_SYSCFG:
3017 	case MSR_K8_TSEG_ADDR:
3018 	case MSR_K8_TSEG_MASK:
3019 	case MSR_VM_HSAVE_PA:
3020 	case MSR_K8_INT_PENDING_MSG:
3021 	case MSR_AMD64_NB_CFG:
3022 	case MSR_FAM10H_MMIO_CONF_BASE:
3023 	case MSR_AMD64_BU_CFG2:
3024 	case MSR_IA32_PERF_CTL:
3025 	case MSR_AMD64_DC_CFG:
3026 	case MSR_F15H_EX_CFG:
3027 		msr_info->data = 0;
3028 		break;
3029 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3030 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3031 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3032 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3033 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3034 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3035 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3036 		msr_info->data = 0;
3037 		break;
3038 	case MSR_IA32_UCODE_REV:
3039 		msr_info->data = vcpu->arch.microcode_version;
3040 		break;
3041 	case MSR_IA32_ARCH_CAPABILITIES:
3042 		if (!msr_info->host_initiated &&
3043 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3044 			return 1;
3045 		msr_info->data = vcpu->arch.arch_capabilities;
3046 		break;
3047 	case MSR_IA32_POWER_CTL:
3048 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3049 		break;
3050 	case MSR_IA32_TSC:
3051 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3052 		break;
3053 	case MSR_MTRRcap:
3054 	case 0x200 ... 0x2ff:
3055 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3056 	case 0xcd: /* fsb frequency */
3057 		msr_info->data = 3;
3058 		break;
3059 		/*
3060 		 * MSR_EBC_FREQUENCY_ID
3061 		 * Conservative value valid for even the basic CPU models.
3062 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3063 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3064 		 * and 266MHz for model 3, or 4. Set Core Clock
3065 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3066 		 * 31:24) even though these are only valid for CPU
3067 		 * models > 2, however guests may end up dividing or
3068 		 * multiplying by zero otherwise.
3069 		 */
3070 	case MSR_EBC_FREQUENCY_ID:
3071 		msr_info->data = 1 << 24;
3072 		break;
3073 	case MSR_IA32_APICBASE:
3074 		msr_info->data = kvm_get_apic_base(vcpu);
3075 		break;
3076 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3077 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3078 		break;
3079 	case MSR_IA32_TSCDEADLINE:
3080 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3081 		break;
3082 	case MSR_IA32_TSC_ADJUST:
3083 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3084 		break;
3085 	case MSR_IA32_MISC_ENABLE:
3086 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3087 		break;
3088 	case MSR_IA32_SMBASE:
3089 		if (!msr_info->host_initiated)
3090 			return 1;
3091 		msr_info->data = vcpu->arch.smbase;
3092 		break;
3093 	case MSR_SMI_COUNT:
3094 		msr_info->data = vcpu->arch.smi_count;
3095 		break;
3096 	case MSR_IA32_PERF_STATUS:
3097 		/* TSC increment by tick */
3098 		msr_info->data = 1000ULL;
3099 		/* CPU multiplier */
3100 		msr_info->data |= (((uint64_t)4ULL) << 40);
3101 		break;
3102 	case MSR_EFER:
3103 		msr_info->data = vcpu->arch.efer;
3104 		break;
3105 	case MSR_KVM_WALL_CLOCK:
3106 	case MSR_KVM_WALL_CLOCK_NEW:
3107 		msr_info->data = vcpu->kvm->arch.wall_clock;
3108 		break;
3109 	case MSR_KVM_SYSTEM_TIME:
3110 	case MSR_KVM_SYSTEM_TIME_NEW:
3111 		msr_info->data = vcpu->arch.time;
3112 		break;
3113 	case MSR_KVM_ASYNC_PF_EN:
3114 		msr_info->data = vcpu->arch.apf.msr_val;
3115 		break;
3116 	case MSR_KVM_STEAL_TIME:
3117 		msr_info->data = vcpu->arch.st.msr_val;
3118 		break;
3119 	case MSR_KVM_PV_EOI_EN:
3120 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3121 		break;
3122 	case MSR_KVM_POLL_CONTROL:
3123 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3124 		break;
3125 	case MSR_IA32_P5_MC_ADDR:
3126 	case MSR_IA32_P5_MC_TYPE:
3127 	case MSR_IA32_MCG_CAP:
3128 	case MSR_IA32_MCG_CTL:
3129 	case MSR_IA32_MCG_STATUS:
3130 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3131 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3132 				   msr_info->host_initiated);
3133 	case MSR_IA32_XSS:
3134 		if (!msr_info->host_initiated &&
3135 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3136 			return 1;
3137 		msr_info->data = vcpu->arch.ia32_xss;
3138 		break;
3139 	case MSR_K7_CLK_CTL:
3140 		/*
3141 		 * Provide expected ramp-up count for K7. All other
3142 		 * are set to zero, indicating minimum divisors for
3143 		 * every field.
3144 		 *
3145 		 * This prevents guest kernels on AMD host with CPU
3146 		 * type 6, model 8 and higher from exploding due to
3147 		 * the rdmsr failing.
3148 		 */
3149 		msr_info->data = 0x20000000;
3150 		break;
3151 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3152 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3153 	case HV_X64_MSR_CRASH_CTL:
3154 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3155 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3156 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3157 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3158 		return kvm_hv_get_msr_common(vcpu,
3159 					     msr_info->index, &msr_info->data,
3160 					     msr_info->host_initiated);
3161 		break;
3162 	case MSR_IA32_BBL_CR_CTL3:
3163 		/* This legacy MSR exists but isn't fully documented in current
3164 		 * silicon.  It is however accessed by winxp in very narrow
3165 		 * scenarios where it sets bit #19, itself documented as
3166 		 * a "reserved" bit.  Best effort attempt to source coherent
3167 		 * read data here should the balance of the register be
3168 		 * interpreted by the guest:
3169 		 *
3170 		 * L2 cache control register 3: 64GB range, 256KB size,
3171 		 * enabled, latency 0x1, configured
3172 		 */
3173 		msr_info->data = 0xbe702111;
3174 		break;
3175 	case MSR_AMD64_OSVW_ID_LENGTH:
3176 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3177 			return 1;
3178 		msr_info->data = vcpu->arch.osvw.length;
3179 		break;
3180 	case MSR_AMD64_OSVW_STATUS:
3181 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3182 			return 1;
3183 		msr_info->data = vcpu->arch.osvw.status;
3184 		break;
3185 	case MSR_PLATFORM_INFO:
3186 		if (!msr_info->host_initiated &&
3187 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3188 			return 1;
3189 		msr_info->data = vcpu->arch.msr_platform_info;
3190 		break;
3191 	case MSR_MISC_FEATURES_ENABLES:
3192 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3193 		break;
3194 	case MSR_K7_HWCR:
3195 		msr_info->data = vcpu->arch.msr_hwcr;
3196 		break;
3197 	default:
3198 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3199 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3200 		if (!ignore_msrs) {
3201 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3202 					       msr_info->index);
3203 			return 1;
3204 		} else {
3205 			if (report_ignored_msrs)
3206 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3207 					msr_info->index);
3208 			msr_info->data = 0;
3209 		}
3210 		break;
3211 	}
3212 	return 0;
3213 }
3214 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3215 
3216 /*
3217  * Read or write a bunch of msrs. All parameters are kernel addresses.
3218  *
3219  * @return number of msrs set successfully.
3220  */
3221 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3222 		    struct kvm_msr_entry *entries,
3223 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3224 				  unsigned index, u64 *data))
3225 {
3226 	int i;
3227 
3228 	for (i = 0; i < msrs->nmsrs; ++i)
3229 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3230 			break;
3231 
3232 	return i;
3233 }
3234 
3235 /*
3236  * Read or write a bunch of msrs. Parameters are user addresses.
3237  *
3238  * @return number of msrs set successfully.
3239  */
3240 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3241 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3242 				unsigned index, u64 *data),
3243 		  int writeback)
3244 {
3245 	struct kvm_msrs msrs;
3246 	struct kvm_msr_entry *entries;
3247 	int r, n;
3248 	unsigned size;
3249 
3250 	r = -EFAULT;
3251 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3252 		goto out;
3253 
3254 	r = -E2BIG;
3255 	if (msrs.nmsrs >= MAX_IO_MSRS)
3256 		goto out;
3257 
3258 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3259 	entries = memdup_user(user_msrs->entries, size);
3260 	if (IS_ERR(entries)) {
3261 		r = PTR_ERR(entries);
3262 		goto out;
3263 	}
3264 
3265 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3266 	if (r < 0)
3267 		goto out_free;
3268 
3269 	r = -EFAULT;
3270 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3271 		goto out_free;
3272 
3273 	r = n;
3274 
3275 out_free:
3276 	kfree(entries);
3277 out:
3278 	return r;
3279 }
3280 
3281 static inline bool kvm_can_mwait_in_guest(void)
3282 {
3283 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3284 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3285 		boot_cpu_has(X86_FEATURE_ARAT);
3286 }
3287 
3288 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3289 {
3290 	int r = 0;
3291 
3292 	switch (ext) {
3293 	case KVM_CAP_IRQCHIP:
3294 	case KVM_CAP_HLT:
3295 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3296 	case KVM_CAP_SET_TSS_ADDR:
3297 	case KVM_CAP_EXT_CPUID:
3298 	case KVM_CAP_EXT_EMUL_CPUID:
3299 	case KVM_CAP_CLOCKSOURCE:
3300 	case KVM_CAP_PIT:
3301 	case KVM_CAP_NOP_IO_DELAY:
3302 	case KVM_CAP_MP_STATE:
3303 	case KVM_CAP_SYNC_MMU:
3304 	case KVM_CAP_USER_NMI:
3305 	case KVM_CAP_REINJECT_CONTROL:
3306 	case KVM_CAP_IRQ_INJECT_STATUS:
3307 	case KVM_CAP_IOEVENTFD:
3308 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3309 	case KVM_CAP_PIT2:
3310 	case KVM_CAP_PIT_STATE2:
3311 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3312 	case KVM_CAP_XEN_HVM:
3313 	case KVM_CAP_VCPU_EVENTS:
3314 	case KVM_CAP_HYPERV:
3315 	case KVM_CAP_HYPERV_VAPIC:
3316 	case KVM_CAP_HYPERV_SPIN:
3317 	case KVM_CAP_HYPERV_SYNIC:
3318 	case KVM_CAP_HYPERV_SYNIC2:
3319 	case KVM_CAP_HYPERV_VP_INDEX:
3320 	case KVM_CAP_HYPERV_EVENTFD:
3321 	case KVM_CAP_HYPERV_TLBFLUSH:
3322 	case KVM_CAP_HYPERV_SEND_IPI:
3323 	case KVM_CAP_HYPERV_CPUID:
3324 	case KVM_CAP_PCI_SEGMENT:
3325 	case KVM_CAP_DEBUGREGS:
3326 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3327 	case KVM_CAP_XSAVE:
3328 	case KVM_CAP_ASYNC_PF:
3329 	case KVM_CAP_GET_TSC_KHZ:
3330 	case KVM_CAP_KVMCLOCK_CTRL:
3331 	case KVM_CAP_READONLY_MEM:
3332 	case KVM_CAP_HYPERV_TIME:
3333 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3334 	case KVM_CAP_TSC_DEADLINE_TIMER:
3335 	case KVM_CAP_DISABLE_QUIRKS:
3336 	case KVM_CAP_SET_BOOT_CPU_ID:
3337  	case KVM_CAP_SPLIT_IRQCHIP:
3338 	case KVM_CAP_IMMEDIATE_EXIT:
3339 	case KVM_CAP_PMU_EVENT_FILTER:
3340 	case KVM_CAP_GET_MSR_FEATURES:
3341 	case KVM_CAP_MSR_PLATFORM_INFO:
3342 	case KVM_CAP_EXCEPTION_PAYLOAD:
3343 		r = 1;
3344 		break;
3345 	case KVM_CAP_SYNC_REGS:
3346 		r = KVM_SYNC_X86_VALID_FIELDS;
3347 		break;
3348 	case KVM_CAP_ADJUST_CLOCK:
3349 		r = KVM_CLOCK_TSC_STABLE;
3350 		break;
3351 	case KVM_CAP_X86_DISABLE_EXITS:
3352 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3353 		      KVM_X86_DISABLE_EXITS_CSTATE;
3354 		if(kvm_can_mwait_in_guest())
3355 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3356 		break;
3357 	case KVM_CAP_X86_SMM:
3358 		/* SMBASE is usually relocated above 1M on modern chipsets,
3359 		 * and SMM handlers might indeed rely on 4G segment limits,
3360 		 * so do not report SMM to be available if real mode is
3361 		 * emulated via vm86 mode.  Still, do not go to great lengths
3362 		 * to avoid userspace's usage of the feature, because it is a
3363 		 * fringe case that is not enabled except via specific settings
3364 		 * of the module parameters.
3365 		 */
3366 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3367 		break;
3368 	case KVM_CAP_VAPIC:
3369 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3370 		break;
3371 	case KVM_CAP_NR_VCPUS:
3372 		r = KVM_SOFT_MAX_VCPUS;
3373 		break;
3374 	case KVM_CAP_MAX_VCPUS:
3375 		r = KVM_MAX_VCPUS;
3376 		break;
3377 	case KVM_CAP_MAX_VCPU_ID:
3378 		r = KVM_MAX_VCPU_ID;
3379 		break;
3380 	case KVM_CAP_PV_MMU:	/* obsolete */
3381 		r = 0;
3382 		break;
3383 	case KVM_CAP_MCE:
3384 		r = KVM_MAX_MCE_BANKS;
3385 		break;
3386 	case KVM_CAP_XCRS:
3387 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3388 		break;
3389 	case KVM_CAP_TSC_CONTROL:
3390 		r = kvm_has_tsc_control;
3391 		break;
3392 	case KVM_CAP_X2APIC_API:
3393 		r = KVM_X2APIC_API_VALID_FLAGS;
3394 		break;
3395 	case KVM_CAP_NESTED_STATE:
3396 		r = kvm_x86_ops->get_nested_state ?
3397 			kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3398 		break;
3399 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3400 		r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3401 		break;
3402 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3403 		r = kvm_x86_ops->nested_enable_evmcs != NULL;
3404 		break;
3405 	default:
3406 		break;
3407 	}
3408 	return r;
3409 
3410 }
3411 
3412 long kvm_arch_dev_ioctl(struct file *filp,
3413 			unsigned int ioctl, unsigned long arg)
3414 {
3415 	void __user *argp = (void __user *)arg;
3416 	long r;
3417 
3418 	switch (ioctl) {
3419 	case KVM_GET_MSR_INDEX_LIST: {
3420 		struct kvm_msr_list __user *user_msr_list = argp;
3421 		struct kvm_msr_list msr_list;
3422 		unsigned n;
3423 
3424 		r = -EFAULT;
3425 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3426 			goto out;
3427 		n = msr_list.nmsrs;
3428 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3429 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3430 			goto out;
3431 		r = -E2BIG;
3432 		if (n < msr_list.nmsrs)
3433 			goto out;
3434 		r = -EFAULT;
3435 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3436 				 num_msrs_to_save * sizeof(u32)))
3437 			goto out;
3438 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3439 				 &emulated_msrs,
3440 				 num_emulated_msrs * sizeof(u32)))
3441 			goto out;
3442 		r = 0;
3443 		break;
3444 	}
3445 	case KVM_GET_SUPPORTED_CPUID:
3446 	case KVM_GET_EMULATED_CPUID: {
3447 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3448 		struct kvm_cpuid2 cpuid;
3449 
3450 		r = -EFAULT;
3451 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3452 			goto out;
3453 
3454 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3455 					    ioctl);
3456 		if (r)
3457 			goto out;
3458 
3459 		r = -EFAULT;
3460 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3461 			goto out;
3462 		r = 0;
3463 		break;
3464 	}
3465 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3466 		r = -EFAULT;
3467 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3468 				 sizeof(kvm_mce_cap_supported)))
3469 			goto out;
3470 		r = 0;
3471 		break;
3472 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3473 		struct kvm_msr_list __user *user_msr_list = argp;
3474 		struct kvm_msr_list msr_list;
3475 		unsigned int n;
3476 
3477 		r = -EFAULT;
3478 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3479 			goto out;
3480 		n = msr_list.nmsrs;
3481 		msr_list.nmsrs = num_msr_based_features;
3482 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3483 			goto out;
3484 		r = -E2BIG;
3485 		if (n < msr_list.nmsrs)
3486 			goto out;
3487 		r = -EFAULT;
3488 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3489 				 num_msr_based_features * sizeof(u32)))
3490 			goto out;
3491 		r = 0;
3492 		break;
3493 	}
3494 	case KVM_GET_MSRS:
3495 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3496 		break;
3497 	}
3498 	default:
3499 		r = -EINVAL;
3500 	}
3501 out:
3502 	return r;
3503 }
3504 
3505 static void wbinvd_ipi(void *garbage)
3506 {
3507 	wbinvd();
3508 }
3509 
3510 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3511 {
3512 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3513 }
3514 
3515 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3516 {
3517 	/* Address WBINVD may be executed by guest */
3518 	if (need_emulate_wbinvd(vcpu)) {
3519 		if (kvm_x86_ops->has_wbinvd_exit())
3520 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3521 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3522 			smp_call_function_single(vcpu->cpu,
3523 					wbinvd_ipi, NULL, 1);
3524 	}
3525 
3526 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3527 
3528 	/* Apply any externally detected TSC adjustments (due to suspend) */
3529 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3530 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3531 		vcpu->arch.tsc_offset_adjustment = 0;
3532 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3533 	}
3534 
3535 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3536 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3537 				rdtsc() - vcpu->arch.last_host_tsc;
3538 		if (tsc_delta < 0)
3539 			mark_tsc_unstable("KVM discovered backwards TSC");
3540 
3541 		if (kvm_check_tsc_unstable()) {
3542 			u64 offset = kvm_compute_tsc_offset(vcpu,
3543 						vcpu->arch.last_guest_tsc);
3544 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3545 			vcpu->arch.tsc_catchup = 1;
3546 		}
3547 
3548 		if (kvm_lapic_hv_timer_in_use(vcpu))
3549 			kvm_lapic_restart_hv_timer(vcpu);
3550 
3551 		/*
3552 		 * On a host with synchronized TSC, there is no need to update
3553 		 * kvmclock on vcpu->cpu migration
3554 		 */
3555 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3556 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3557 		if (vcpu->cpu != cpu)
3558 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3559 		vcpu->cpu = cpu;
3560 	}
3561 
3562 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3563 }
3564 
3565 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3566 {
3567 	struct kvm_host_map map;
3568 	struct kvm_steal_time *st;
3569 
3570 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3571 		return;
3572 
3573 	if (vcpu->arch.st.preempted)
3574 		return;
3575 
3576 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3577 			&vcpu->arch.st.cache, true))
3578 		return;
3579 
3580 	st = map.hva +
3581 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3582 
3583 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3584 
3585 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3586 }
3587 
3588 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3589 {
3590 	int idx;
3591 
3592 	if (vcpu->preempted)
3593 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3594 
3595 	/*
3596 	 * Disable page faults because we're in atomic context here.
3597 	 * kvm_write_guest_offset_cached() would call might_fault()
3598 	 * that relies on pagefault_disable() to tell if there's a
3599 	 * bug. NOTE: the write to guest memory may not go through if
3600 	 * during postcopy live migration or if there's heavy guest
3601 	 * paging.
3602 	 */
3603 	pagefault_disable();
3604 	/*
3605 	 * kvm_memslots() will be called by
3606 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3607 	 */
3608 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3609 	kvm_steal_time_set_preempted(vcpu);
3610 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3611 	pagefault_enable();
3612 	kvm_x86_ops->vcpu_put(vcpu);
3613 	vcpu->arch.last_host_tsc = rdtsc();
3614 	/*
3615 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3616 	 * on every vmexit, but if not, we might have a stale dr6 from the
3617 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3618 	 */
3619 	set_debugreg(0, 6);
3620 }
3621 
3622 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3623 				    struct kvm_lapic_state *s)
3624 {
3625 	if (vcpu->arch.apicv_active)
3626 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3627 
3628 	return kvm_apic_get_state(vcpu, s);
3629 }
3630 
3631 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3632 				    struct kvm_lapic_state *s)
3633 {
3634 	int r;
3635 
3636 	r = kvm_apic_set_state(vcpu, s);
3637 	if (r)
3638 		return r;
3639 	update_cr8_intercept(vcpu);
3640 
3641 	return 0;
3642 }
3643 
3644 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3645 {
3646 	return (!lapic_in_kernel(vcpu) ||
3647 		kvm_apic_accept_pic_intr(vcpu));
3648 }
3649 
3650 /*
3651  * if userspace requested an interrupt window, check that the
3652  * interrupt window is open.
3653  *
3654  * No need to exit to userspace if we already have an interrupt queued.
3655  */
3656 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3657 {
3658 	return kvm_arch_interrupt_allowed(vcpu) &&
3659 		!kvm_cpu_has_interrupt(vcpu) &&
3660 		!kvm_event_needs_reinjection(vcpu) &&
3661 		kvm_cpu_accept_dm_intr(vcpu);
3662 }
3663 
3664 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3665 				    struct kvm_interrupt *irq)
3666 {
3667 	if (irq->irq >= KVM_NR_INTERRUPTS)
3668 		return -EINVAL;
3669 
3670 	if (!irqchip_in_kernel(vcpu->kvm)) {
3671 		kvm_queue_interrupt(vcpu, irq->irq, false);
3672 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3673 		return 0;
3674 	}
3675 
3676 	/*
3677 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3678 	 * fail for in-kernel 8259.
3679 	 */
3680 	if (pic_in_kernel(vcpu->kvm))
3681 		return -ENXIO;
3682 
3683 	if (vcpu->arch.pending_external_vector != -1)
3684 		return -EEXIST;
3685 
3686 	vcpu->arch.pending_external_vector = irq->irq;
3687 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3688 	return 0;
3689 }
3690 
3691 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3692 {
3693 	kvm_inject_nmi(vcpu);
3694 
3695 	return 0;
3696 }
3697 
3698 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3699 {
3700 	kvm_make_request(KVM_REQ_SMI, vcpu);
3701 
3702 	return 0;
3703 }
3704 
3705 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3706 					   struct kvm_tpr_access_ctl *tac)
3707 {
3708 	if (tac->flags)
3709 		return -EINVAL;
3710 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3711 	return 0;
3712 }
3713 
3714 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3715 					u64 mcg_cap)
3716 {
3717 	int r;
3718 	unsigned bank_num = mcg_cap & 0xff, bank;
3719 
3720 	r = -EINVAL;
3721 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3722 		goto out;
3723 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3724 		goto out;
3725 	r = 0;
3726 	vcpu->arch.mcg_cap = mcg_cap;
3727 	/* Init IA32_MCG_CTL to all 1s */
3728 	if (mcg_cap & MCG_CTL_P)
3729 		vcpu->arch.mcg_ctl = ~(u64)0;
3730 	/* Init IA32_MCi_CTL to all 1s */
3731 	for (bank = 0; bank < bank_num; bank++)
3732 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3733 
3734 	kvm_x86_ops->setup_mce(vcpu);
3735 out:
3736 	return r;
3737 }
3738 
3739 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3740 				      struct kvm_x86_mce *mce)
3741 {
3742 	u64 mcg_cap = vcpu->arch.mcg_cap;
3743 	unsigned bank_num = mcg_cap & 0xff;
3744 	u64 *banks = vcpu->arch.mce_banks;
3745 
3746 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3747 		return -EINVAL;
3748 	/*
3749 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3750 	 * reporting is disabled
3751 	 */
3752 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3753 	    vcpu->arch.mcg_ctl != ~(u64)0)
3754 		return 0;
3755 	banks += 4 * mce->bank;
3756 	/*
3757 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3758 	 * reporting is disabled for the bank
3759 	 */
3760 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3761 		return 0;
3762 	if (mce->status & MCI_STATUS_UC) {
3763 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3764 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3765 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3766 			return 0;
3767 		}
3768 		if (banks[1] & MCI_STATUS_VAL)
3769 			mce->status |= MCI_STATUS_OVER;
3770 		banks[2] = mce->addr;
3771 		banks[3] = mce->misc;
3772 		vcpu->arch.mcg_status = mce->mcg_status;
3773 		banks[1] = mce->status;
3774 		kvm_queue_exception(vcpu, MC_VECTOR);
3775 	} else if (!(banks[1] & MCI_STATUS_VAL)
3776 		   || !(banks[1] & MCI_STATUS_UC)) {
3777 		if (banks[1] & MCI_STATUS_VAL)
3778 			mce->status |= MCI_STATUS_OVER;
3779 		banks[2] = mce->addr;
3780 		banks[3] = mce->misc;
3781 		banks[1] = mce->status;
3782 	} else
3783 		banks[1] |= MCI_STATUS_OVER;
3784 	return 0;
3785 }
3786 
3787 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3788 					       struct kvm_vcpu_events *events)
3789 {
3790 	process_nmi(vcpu);
3791 
3792 	/*
3793 	 * The API doesn't provide the instruction length for software
3794 	 * exceptions, so don't report them. As long as the guest RIP
3795 	 * isn't advanced, we should expect to encounter the exception
3796 	 * again.
3797 	 */
3798 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3799 		events->exception.injected = 0;
3800 		events->exception.pending = 0;
3801 	} else {
3802 		events->exception.injected = vcpu->arch.exception.injected;
3803 		events->exception.pending = vcpu->arch.exception.pending;
3804 		/*
3805 		 * For ABI compatibility, deliberately conflate
3806 		 * pending and injected exceptions when
3807 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3808 		 */
3809 		if (!vcpu->kvm->arch.exception_payload_enabled)
3810 			events->exception.injected |=
3811 				vcpu->arch.exception.pending;
3812 	}
3813 	events->exception.nr = vcpu->arch.exception.nr;
3814 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3815 	events->exception.error_code = vcpu->arch.exception.error_code;
3816 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3817 	events->exception_payload = vcpu->arch.exception.payload;
3818 
3819 	events->interrupt.injected =
3820 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3821 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3822 	events->interrupt.soft = 0;
3823 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3824 
3825 	events->nmi.injected = vcpu->arch.nmi_injected;
3826 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3827 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3828 	events->nmi.pad = 0;
3829 
3830 	events->sipi_vector = 0; /* never valid when reporting to user space */
3831 
3832 	events->smi.smm = is_smm(vcpu);
3833 	events->smi.pending = vcpu->arch.smi_pending;
3834 	events->smi.smm_inside_nmi =
3835 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3836 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3837 
3838 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3839 			 | KVM_VCPUEVENT_VALID_SHADOW
3840 			 | KVM_VCPUEVENT_VALID_SMM);
3841 	if (vcpu->kvm->arch.exception_payload_enabled)
3842 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3843 
3844 	memset(&events->reserved, 0, sizeof(events->reserved));
3845 }
3846 
3847 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3848 
3849 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3850 					      struct kvm_vcpu_events *events)
3851 {
3852 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3853 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3854 			      | KVM_VCPUEVENT_VALID_SHADOW
3855 			      | KVM_VCPUEVENT_VALID_SMM
3856 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3857 		return -EINVAL;
3858 
3859 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3860 		if (!vcpu->kvm->arch.exception_payload_enabled)
3861 			return -EINVAL;
3862 		if (events->exception.pending)
3863 			events->exception.injected = 0;
3864 		else
3865 			events->exception_has_payload = 0;
3866 	} else {
3867 		events->exception.pending = 0;
3868 		events->exception_has_payload = 0;
3869 	}
3870 
3871 	if ((events->exception.injected || events->exception.pending) &&
3872 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3873 		return -EINVAL;
3874 
3875 	/* INITs are latched while in SMM */
3876 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3877 	    (events->smi.smm || events->smi.pending) &&
3878 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3879 		return -EINVAL;
3880 
3881 	process_nmi(vcpu);
3882 	vcpu->arch.exception.injected = events->exception.injected;
3883 	vcpu->arch.exception.pending = events->exception.pending;
3884 	vcpu->arch.exception.nr = events->exception.nr;
3885 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3886 	vcpu->arch.exception.error_code = events->exception.error_code;
3887 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3888 	vcpu->arch.exception.payload = events->exception_payload;
3889 
3890 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3891 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3892 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3893 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3894 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3895 						  events->interrupt.shadow);
3896 
3897 	vcpu->arch.nmi_injected = events->nmi.injected;
3898 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3899 		vcpu->arch.nmi_pending = events->nmi.pending;
3900 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3901 
3902 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3903 	    lapic_in_kernel(vcpu))
3904 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3905 
3906 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3907 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3908 			if (events->smi.smm)
3909 				vcpu->arch.hflags |= HF_SMM_MASK;
3910 			else
3911 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3912 			kvm_smm_changed(vcpu);
3913 		}
3914 
3915 		vcpu->arch.smi_pending = events->smi.pending;
3916 
3917 		if (events->smi.smm) {
3918 			if (events->smi.smm_inside_nmi)
3919 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3920 			else
3921 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3922 		}
3923 
3924 		if (lapic_in_kernel(vcpu)) {
3925 			if (events->smi.latched_init)
3926 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3927 			else
3928 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3929 		}
3930 	}
3931 
3932 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3933 
3934 	return 0;
3935 }
3936 
3937 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3938 					     struct kvm_debugregs *dbgregs)
3939 {
3940 	unsigned long val;
3941 
3942 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3943 	kvm_get_dr(vcpu, 6, &val);
3944 	dbgregs->dr6 = val;
3945 	dbgregs->dr7 = vcpu->arch.dr7;
3946 	dbgregs->flags = 0;
3947 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3948 }
3949 
3950 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3951 					    struct kvm_debugregs *dbgregs)
3952 {
3953 	if (dbgregs->flags)
3954 		return -EINVAL;
3955 
3956 	if (dbgregs->dr6 & ~0xffffffffull)
3957 		return -EINVAL;
3958 	if (dbgregs->dr7 & ~0xffffffffull)
3959 		return -EINVAL;
3960 
3961 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3962 	kvm_update_dr0123(vcpu);
3963 	vcpu->arch.dr6 = dbgregs->dr6;
3964 	kvm_update_dr6(vcpu);
3965 	vcpu->arch.dr7 = dbgregs->dr7;
3966 	kvm_update_dr7(vcpu);
3967 
3968 	return 0;
3969 }
3970 
3971 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3972 
3973 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3974 {
3975 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3976 	u64 xstate_bv = xsave->header.xfeatures;
3977 	u64 valid;
3978 
3979 	/*
3980 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3981 	 * leaves 0 and 1 in the loop below.
3982 	 */
3983 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3984 
3985 	/* Set XSTATE_BV */
3986 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3987 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3988 
3989 	/*
3990 	 * Copy each region from the possibly compacted offset to the
3991 	 * non-compacted offset.
3992 	 */
3993 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3994 	while (valid) {
3995 		u64 xfeature_mask = valid & -valid;
3996 		int xfeature_nr = fls64(xfeature_mask) - 1;
3997 		void *src = get_xsave_addr(xsave, xfeature_nr);
3998 
3999 		if (src) {
4000 			u32 size, offset, ecx, edx;
4001 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4002 				    &size, &offset, &ecx, &edx);
4003 			if (xfeature_nr == XFEATURE_PKRU)
4004 				memcpy(dest + offset, &vcpu->arch.pkru,
4005 				       sizeof(vcpu->arch.pkru));
4006 			else
4007 				memcpy(dest + offset, src, size);
4008 
4009 		}
4010 
4011 		valid -= xfeature_mask;
4012 	}
4013 }
4014 
4015 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4016 {
4017 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4018 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4019 	u64 valid;
4020 
4021 	/*
4022 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4023 	 * leaves 0 and 1 in the loop below.
4024 	 */
4025 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4026 
4027 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4028 	xsave->header.xfeatures = xstate_bv;
4029 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4030 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4031 
4032 	/*
4033 	 * Copy each region from the non-compacted offset to the
4034 	 * possibly compacted offset.
4035 	 */
4036 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4037 	while (valid) {
4038 		u64 xfeature_mask = valid & -valid;
4039 		int xfeature_nr = fls64(xfeature_mask) - 1;
4040 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4041 
4042 		if (dest) {
4043 			u32 size, offset, ecx, edx;
4044 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4045 				    &size, &offset, &ecx, &edx);
4046 			if (xfeature_nr == XFEATURE_PKRU)
4047 				memcpy(&vcpu->arch.pkru, src + offset,
4048 				       sizeof(vcpu->arch.pkru));
4049 			else
4050 				memcpy(dest, src + offset, size);
4051 		}
4052 
4053 		valid -= xfeature_mask;
4054 	}
4055 }
4056 
4057 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4058 					 struct kvm_xsave *guest_xsave)
4059 {
4060 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4061 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4062 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4063 	} else {
4064 		memcpy(guest_xsave->region,
4065 			&vcpu->arch.guest_fpu->state.fxsave,
4066 			sizeof(struct fxregs_state));
4067 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4068 			XFEATURE_MASK_FPSSE;
4069 	}
4070 }
4071 
4072 #define XSAVE_MXCSR_OFFSET 24
4073 
4074 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4075 					struct kvm_xsave *guest_xsave)
4076 {
4077 	u64 xstate_bv =
4078 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4079 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4080 
4081 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4082 		/*
4083 		 * Here we allow setting states that are not present in
4084 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4085 		 * with old userspace.
4086 		 */
4087 		if (xstate_bv & ~kvm_supported_xcr0() ||
4088 			mxcsr & ~mxcsr_feature_mask)
4089 			return -EINVAL;
4090 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4091 	} else {
4092 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4093 			mxcsr & ~mxcsr_feature_mask)
4094 			return -EINVAL;
4095 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4096 			guest_xsave->region, sizeof(struct fxregs_state));
4097 	}
4098 	return 0;
4099 }
4100 
4101 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4102 					struct kvm_xcrs *guest_xcrs)
4103 {
4104 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4105 		guest_xcrs->nr_xcrs = 0;
4106 		return;
4107 	}
4108 
4109 	guest_xcrs->nr_xcrs = 1;
4110 	guest_xcrs->flags = 0;
4111 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4112 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4113 }
4114 
4115 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4116 				       struct kvm_xcrs *guest_xcrs)
4117 {
4118 	int i, r = 0;
4119 
4120 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4121 		return -EINVAL;
4122 
4123 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4124 		return -EINVAL;
4125 
4126 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4127 		/* Only support XCR0 currently */
4128 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4129 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4130 				guest_xcrs->xcrs[i].value);
4131 			break;
4132 		}
4133 	if (r)
4134 		r = -EINVAL;
4135 	return r;
4136 }
4137 
4138 /*
4139  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4140  * stopped by the hypervisor.  This function will be called from the host only.
4141  * EINVAL is returned when the host attempts to set the flag for a guest that
4142  * does not support pv clocks.
4143  */
4144 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4145 {
4146 	if (!vcpu->arch.pv_time_enabled)
4147 		return -EINVAL;
4148 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4149 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4150 	return 0;
4151 }
4152 
4153 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4154 				     struct kvm_enable_cap *cap)
4155 {
4156 	int r;
4157 	uint16_t vmcs_version;
4158 	void __user *user_ptr;
4159 
4160 	if (cap->flags)
4161 		return -EINVAL;
4162 
4163 	switch (cap->cap) {
4164 	case KVM_CAP_HYPERV_SYNIC2:
4165 		if (cap->args[0])
4166 			return -EINVAL;
4167 		/* fall through */
4168 
4169 	case KVM_CAP_HYPERV_SYNIC:
4170 		if (!irqchip_in_kernel(vcpu->kvm))
4171 			return -EINVAL;
4172 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4173 					     KVM_CAP_HYPERV_SYNIC2);
4174 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4175 		if (!kvm_x86_ops->nested_enable_evmcs)
4176 			return -ENOTTY;
4177 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4178 		if (!r) {
4179 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4180 			if (copy_to_user(user_ptr, &vmcs_version,
4181 					 sizeof(vmcs_version)))
4182 				r = -EFAULT;
4183 		}
4184 		return r;
4185 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4186 		if (!kvm_x86_ops->enable_direct_tlbflush)
4187 			return -ENOTTY;
4188 
4189 		return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4190 
4191 	default:
4192 		return -EINVAL;
4193 	}
4194 }
4195 
4196 long kvm_arch_vcpu_ioctl(struct file *filp,
4197 			 unsigned int ioctl, unsigned long arg)
4198 {
4199 	struct kvm_vcpu *vcpu = filp->private_data;
4200 	void __user *argp = (void __user *)arg;
4201 	int r;
4202 	union {
4203 		struct kvm_lapic_state *lapic;
4204 		struct kvm_xsave *xsave;
4205 		struct kvm_xcrs *xcrs;
4206 		void *buffer;
4207 	} u;
4208 
4209 	vcpu_load(vcpu);
4210 
4211 	u.buffer = NULL;
4212 	switch (ioctl) {
4213 	case KVM_GET_LAPIC: {
4214 		r = -EINVAL;
4215 		if (!lapic_in_kernel(vcpu))
4216 			goto out;
4217 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4218 				GFP_KERNEL_ACCOUNT);
4219 
4220 		r = -ENOMEM;
4221 		if (!u.lapic)
4222 			goto out;
4223 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4224 		if (r)
4225 			goto out;
4226 		r = -EFAULT;
4227 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4228 			goto out;
4229 		r = 0;
4230 		break;
4231 	}
4232 	case KVM_SET_LAPIC: {
4233 		r = -EINVAL;
4234 		if (!lapic_in_kernel(vcpu))
4235 			goto out;
4236 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4237 		if (IS_ERR(u.lapic)) {
4238 			r = PTR_ERR(u.lapic);
4239 			goto out_nofree;
4240 		}
4241 
4242 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4243 		break;
4244 	}
4245 	case KVM_INTERRUPT: {
4246 		struct kvm_interrupt irq;
4247 
4248 		r = -EFAULT;
4249 		if (copy_from_user(&irq, argp, sizeof(irq)))
4250 			goto out;
4251 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4252 		break;
4253 	}
4254 	case KVM_NMI: {
4255 		r = kvm_vcpu_ioctl_nmi(vcpu);
4256 		break;
4257 	}
4258 	case KVM_SMI: {
4259 		r = kvm_vcpu_ioctl_smi(vcpu);
4260 		break;
4261 	}
4262 	case KVM_SET_CPUID: {
4263 		struct kvm_cpuid __user *cpuid_arg = argp;
4264 		struct kvm_cpuid cpuid;
4265 
4266 		r = -EFAULT;
4267 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4268 			goto out;
4269 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4270 		break;
4271 	}
4272 	case KVM_SET_CPUID2: {
4273 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4274 		struct kvm_cpuid2 cpuid;
4275 
4276 		r = -EFAULT;
4277 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4278 			goto out;
4279 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4280 					      cpuid_arg->entries);
4281 		break;
4282 	}
4283 	case KVM_GET_CPUID2: {
4284 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4285 		struct kvm_cpuid2 cpuid;
4286 
4287 		r = -EFAULT;
4288 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4289 			goto out;
4290 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4291 					      cpuid_arg->entries);
4292 		if (r)
4293 			goto out;
4294 		r = -EFAULT;
4295 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4296 			goto out;
4297 		r = 0;
4298 		break;
4299 	}
4300 	case KVM_GET_MSRS: {
4301 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4302 		r = msr_io(vcpu, argp, do_get_msr, 1);
4303 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4304 		break;
4305 	}
4306 	case KVM_SET_MSRS: {
4307 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4308 		r = msr_io(vcpu, argp, do_set_msr, 0);
4309 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4310 		break;
4311 	}
4312 	case KVM_TPR_ACCESS_REPORTING: {
4313 		struct kvm_tpr_access_ctl tac;
4314 
4315 		r = -EFAULT;
4316 		if (copy_from_user(&tac, argp, sizeof(tac)))
4317 			goto out;
4318 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4319 		if (r)
4320 			goto out;
4321 		r = -EFAULT;
4322 		if (copy_to_user(argp, &tac, sizeof(tac)))
4323 			goto out;
4324 		r = 0;
4325 		break;
4326 	};
4327 	case KVM_SET_VAPIC_ADDR: {
4328 		struct kvm_vapic_addr va;
4329 		int idx;
4330 
4331 		r = -EINVAL;
4332 		if (!lapic_in_kernel(vcpu))
4333 			goto out;
4334 		r = -EFAULT;
4335 		if (copy_from_user(&va, argp, sizeof(va)))
4336 			goto out;
4337 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4338 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4339 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4340 		break;
4341 	}
4342 	case KVM_X86_SETUP_MCE: {
4343 		u64 mcg_cap;
4344 
4345 		r = -EFAULT;
4346 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4347 			goto out;
4348 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4349 		break;
4350 	}
4351 	case KVM_X86_SET_MCE: {
4352 		struct kvm_x86_mce mce;
4353 
4354 		r = -EFAULT;
4355 		if (copy_from_user(&mce, argp, sizeof(mce)))
4356 			goto out;
4357 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4358 		break;
4359 	}
4360 	case KVM_GET_VCPU_EVENTS: {
4361 		struct kvm_vcpu_events events;
4362 
4363 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4364 
4365 		r = -EFAULT;
4366 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4367 			break;
4368 		r = 0;
4369 		break;
4370 	}
4371 	case KVM_SET_VCPU_EVENTS: {
4372 		struct kvm_vcpu_events events;
4373 
4374 		r = -EFAULT;
4375 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4376 			break;
4377 
4378 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4379 		break;
4380 	}
4381 	case KVM_GET_DEBUGREGS: {
4382 		struct kvm_debugregs dbgregs;
4383 
4384 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4385 
4386 		r = -EFAULT;
4387 		if (copy_to_user(argp, &dbgregs,
4388 				 sizeof(struct kvm_debugregs)))
4389 			break;
4390 		r = 0;
4391 		break;
4392 	}
4393 	case KVM_SET_DEBUGREGS: {
4394 		struct kvm_debugregs dbgregs;
4395 
4396 		r = -EFAULT;
4397 		if (copy_from_user(&dbgregs, argp,
4398 				   sizeof(struct kvm_debugregs)))
4399 			break;
4400 
4401 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4402 		break;
4403 	}
4404 	case KVM_GET_XSAVE: {
4405 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4406 		r = -ENOMEM;
4407 		if (!u.xsave)
4408 			break;
4409 
4410 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4411 
4412 		r = -EFAULT;
4413 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4414 			break;
4415 		r = 0;
4416 		break;
4417 	}
4418 	case KVM_SET_XSAVE: {
4419 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4420 		if (IS_ERR(u.xsave)) {
4421 			r = PTR_ERR(u.xsave);
4422 			goto out_nofree;
4423 		}
4424 
4425 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4426 		break;
4427 	}
4428 	case KVM_GET_XCRS: {
4429 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4430 		r = -ENOMEM;
4431 		if (!u.xcrs)
4432 			break;
4433 
4434 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4435 
4436 		r = -EFAULT;
4437 		if (copy_to_user(argp, u.xcrs,
4438 				 sizeof(struct kvm_xcrs)))
4439 			break;
4440 		r = 0;
4441 		break;
4442 	}
4443 	case KVM_SET_XCRS: {
4444 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4445 		if (IS_ERR(u.xcrs)) {
4446 			r = PTR_ERR(u.xcrs);
4447 			goto out_nofree;
4448 		}
4449 
4450 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4451 		break;
4452 	}
4453 	case KVM_SET_TSC_KHZ: {
4454 		u32 user_tsc_khz;
4455 
4456 		r = -EINVAL;
4457 		user_tsc_khz = (u32)arg;
4458 
4459 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4460 			goto out;
4461 
4462 		if (user_tsc_khz == 0)
4463 			user_tsc_khz = tsc_khz;
4464 
4465 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4466 			r = 0;
4467 
4468 		goto out;
4469 	}
4470 	case KVM_GET_TSC_KHZ: {
4471 		r = vcpu->arch.virtual_tsc_khz;
4472 		goto out;
4473 	}
4474 	case KVM_KVMCLOCK_CTRL: {
4475 		r = kvm_set_guest_paused(vcpu);
4476 		goto out;
4477 	}
4478 	case KVM_ENABLE_CAP: {
4479 		struct kvm_enable_cap cap;
4480 
4481 		r = -EFAULT;
4482 		if (copy_from_user(&cap, argp, sizeof(cap)))
4483 			goto out;
4484 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4485 		break;
4486 	}
4487 	case KVM_GET_NESTED_STATE: {
4488 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4489 		u32 user_data_size;
4490 
4491 		r = -EINVAL;
4492 		if (!kvm_x86_ops->get_nested_state)
4493 			break;
4494 
4495 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4496 		r = -EFAULT;
4497 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4498 			break;
4499 
4500 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4501 						  user_data_size);
4502 		if (r < 0)
4503 			break;
4504 
4505 		if (r > user_data_size) {
4506 			if (put_user(r, &user_kvm_nested_state->size))
4507 				r = -EFAULT;
4508 			else
4509 				r = -E2BIG;
4510 			break;
4511 		}
4512 
4513 		r = 0;
4514 		break;
4515 	}
4516 	case KVM_SET_NESTED_STATE: {
4517 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4518 		struct kvm_nested_state kvm_state;
4519 		int idx;
4520 
4521 		r = -EINVAL;
4522 		if (!kvm_x86_ops->set_nested_state)
4523 			break;
4524 
4525 		r = -EFAULT;
4526 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4527 			break;
4528 
4529 		r = -EINVAL;
4530 		if (kvm_state.size < sizeof(kvm_state))
4531 			break;
4532 
4533 		if (kvm_state.flags &
4534 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4535 		      | KVM_STATE_NESTED_EVMCS))
4536 			break;
4537 
4538 		/* nested_run_pending implies guest_mode.  */
4539 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4540 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4541 			break;
4542 
4543 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4544 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4545 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4546 		break;
4547 	}
4548 	case KVM_GET_SUPPORTED_HV_CPUID: {
4549 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4550 		struct kvm_cpuid2 cpuid;
4551 
4552 		r = -EFAULT;
4553 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4554 			goto out;
4555 
4556 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4557 						cpuid_arg->entries);
4558 		if (r)
4559 			goto out;
4560 
4561 		r = -EFAULT;
4562 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4563 			goto out;
4564 		r = 0;
4565 		break;
4566 	}
4567 	default:
4568 		r = -EINVAL;
4569 	}
4570 out:
4571 	kfree(u.buffer);
4572 out_nofree:
4573 	vcpu_put(vcpu);
4574 	return r;
4575 }
4576 
4577 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4578 {
4579 	return VM_FAULT_SIGBUS;
4580 }
4581 
4582 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4583 {
4584 	int ret;
4585 
4586 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4587 		return -EINVAL;
4588 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4589 	return ret;
4590 }
4591 
4592 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4593 					      u64 ident_addr)
4594 {
4595 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4596 }
4597 
4598 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4599 					 unsigned long kvm_nr_mmu_pages)
4600 {
4601 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4602 		return -EINVAL;
4603 
4604 	mutex_lock(&kvm->slots_lock);
4605 
4606 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4607 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4608 
4609 	mutex_unlock(&kvm->slots_lock);
4610 	return 0;
4611 }
4612 
4613 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4614 {
4615 	return kvm->arch.n_max_mmu_pages;
4616 }
4617 
4618 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4619 {
4620 	struct kvm_pic *pic = kvm->arch.vpic;
4621 	int r;
4622 
4623 	r = 0;
4624 	switch (chip->chip_id) {
4625 	case KVM_IRQCHIP_PIC_MASTER:
4626 		memcpy(&chip->chip.pic, &pic->pics[0],
4627 			sizeof(struct kvm_pic_state));
4628 		break;
4629 	case KVM_IRQCHIP_PIC_SLAVE:
4630 		memcpy(&chip->chip.pic, &pic->pics[1],
4631 			sizeof(struct kvm_pic_state));
4632 		break;
4633 	case KVM_IRQCHIP_IOAPIC:
4634 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4635 		break;
4636 	default:
4637 		r = -EINVAL;
4638 		break;
4639 	}
4640 	return r;
4641 }
4642 
4643 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4644 {
4645 	struct kvm_pic *pic = kvm->arch.vpic;
4646 	int r;
4647 
4648 	r = 0;
4649 	switch (chip->chip_id) {
4650 	case KVM_IRQCHIP_PIC_MASTER:
4651 		spin_lock(&pic->lock);
4652 		memcpy(&pic->pics[0], &chip->chip.pic,
4653 			sizeof(struct kvm_pic_state));
4654 		spin_unlock(&pic->lock);
4655 		break;
4656 	case KVM_IRQCHIP_PIC_SLAVE:
4657 		spin_lock(&pic->lock);
4658 		memcpy(&pic->pics[1], &chip->chip.pic,
4659 			sizeof(struct kvm_pic_state));
4660 		spin_unlock(&pic->lock);
4661 		break;
4662 	case KVM_IRQCHIP_IOAPIC:
4663 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4664 		break;
4665 	default:
4666 		r = -EINVAL;
4667 		break;
4668 	}
4669 	kvm_pic_update_irq(pic);
4670 	return r;
4671 }
4672 
4673 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4674 {
4675 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4676 
4677 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4678 
4679 	mutex_lock(&kps->lock);
4680 	memcpy(ps, &kps->channels, sizeof(*ps));
4681 	mutex_unlock(&kps->lock);
4682 	return 0;
4683 }
4684 
4685 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4686 {
4687 	int i;
4688 	struct kvm_pit *pit = kvm->arch.vpit;
4689 
4690 	mutex_lock(&pit->pit_state.lock);
4691 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4692 	for (i = 0; i < 3; i++)
4693 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4694 	mutex_unlock(&pit->pit_state.lock);
4695 	return 0;
4696 }
4697 
4698 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4699 {
4700 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4701 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4702 		sizeof(ps->channels));
4703 	ps->flags = kvm->arch.vpit->pit_state.flags;
4704 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4705 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4706 	return 0;
4707 }
4708 
4709 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4710 {
4711 	int start = 0;
4712 	int i;
4713 	u32 prev_legacy, cur_legacy;
4714 	struct kvm_pit *pit = kvm->arch.vpit;
4715 
4716 	mutex_lock(&pit->pit_state.lock);
4717 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4718 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4719 	if (!prev_legacy && cur_legacy)
4720 		start = 1;
4721 	memcpy(&pit->pit_state.channels, &ps->channels,
4722 	       sizeof(pit->pit_state.channels));
4723 	pit->pit_state.flags = ps->flags;
4724 	for (i = 0; i < 3; i++)
4725 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4726 				   start && i == 0);
4727 	mutex_unlock(&pit->pit_state.lock);
4728 	return 0;
4729 }
4730 
4731 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4732 				 struct kvm_reinject_control *control)
4733 {
4734 	struct kvm_pit *pit = kvm->arch.vpit;
4735 
4736 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4737 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4738 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4739 	 */
4740 	mutex_lock(&pit->pit_state.lock);
4741 	kvm_pit_set_reinject(pit, control->pit_reinject);
4742 	mutex_unlock(&pit->pit_state.lock);
4743 
4744 	return 0;
4745 }
4746 
4747 /**
4748  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4749  * @kvm: kvm instance
4750  * @log: slot id and address to which we copy the log
4751  *
4752  * Steps 1-4 below provide general overview of dirty page logging. See
4753  * kvm_get_dirty_log_protect() function description for additional details.
4754  *
4755  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4756  * always flush the TLB (step 4) even if previous step failed  and the dirty
4757  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4758  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4759  * writes will be marked dirty for next log read.
4760  *
4761  *   1. Take a snapshot of the bit and clear it if needed.
4762  *   2. Write protect the corresponding page.
4763  *   3. Copy the snapshot to the userspace.
4764  *   4. Flush TLB's if needed.
4765  */
4766 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4767 {
4768 	bool flush = false;
4769 	int r;
4770 
4771 	mutex_lock(&kvm->slots_lock);
4772 
4773 	/*
4774 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4775 	 */
4776 	if (kvm_x86_ops->flush_log_dirty)
4777 		kvm_x86_ops->flush_log_dirty(kvm);
4778 
4779 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4780 
4781 	/*
4782 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4783 	 * kvm_mmu_slot_remove_write_access().
4784 	 */
4785 	lockdep_assert_held(&kvm->slots_lock);
4786 	if (flush)
4787 		kvm_flush_remote_tlbs(kvm);
4788 
4789 	mutex_unlock(&kvm->slots_lock);
4790 	return r;
4791 }
4792 
4793 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4794 {
4795 	bool flush = false;
4796 	int r;
4797 
4798 	mutex_lock(&kvm->slots_lock);
4799 
4800 	/*
4801 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4802 	 */
4803 	if (kvm_x86_ops->flush_log_dirty)
4804 		kvm_x86_ops->flush_log_dirty(kvm);
4805 
4806 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4807 
4808 	/*
4809 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4810 	 * kvm_mmu_slot_remove_write_access().
4811 	 */
4812 	lockdep_assert_held(&kvm->slots_lock);
4813 	if (flush)
4814 		kvm_flush_remote_tlbs(kvm);
4815 
4816 	mutex_unlock(&kvm->slots_lock);
4817 	return r;
4818 }
4819 
4820 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4821 			bool line_status)
4822 {
4823 	if (!irqchip_in_kernel(kvm))
4824 		return -ENXIO;
4825 
4826 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4827 					irq_event->irq, irq_event->level,
4828 					line_status);
4829 	return 0;
4830 }
4831 
4832 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4833 			    struct kvm_enable_cap *cap)
4834 {
4835 	int r;
4836 
4837 	if (cap->flags)
4838 		return -EINVAL;
4839 
4840 	switch (cap->cap) {
4841 	case KVM_CAP_DISABLE_QUIRKS:
4842 		kvm->arch.disabled_quirks = cap->args[0];
4843 		r = 0;
4844 		break;
4845 	case KVM_CAP_SPLIT_IRQCHIP: {
4846 		mutex_lock(&kvm->lock);
4847 		r = -EINVAL;
4848 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4849 			goto split_irqchip_unlock;
4850 		r = -EEXIST;
4851 		if (irqchip_in_kernel(kvm))
4852 			goto split_irqchip_unlock;
4853 		if (kvm->created_vcpus)
4854 			goto split_irqchip_unlock;
4855 		r = kvm_setup_empty_irq_routing(kvm);
4856 		if (r)
4857 			goto split_irqchip_unlock;
4858 		/* Pairs with irqchip_in_kernel. */
4859 		smp_wmb();
4860 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4861 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4862 		r = 0;
4863 split_irqchip_unlock:
4864 		mutex_unlock(&kvm->lock);
4865 		break;
4866 	}
4867 	case KVM_CAP_X2APIC_API:
4868 		r = -EINVAL;
4869 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4870 			break;
4871 
4872 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4873 			kvm->arch.x2apic_format = true;
4874 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4875 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4876 
4877 		r = 0;
4878 		break;
4879 	case KVM_CAP_X86_DISABLE_EXITS:
4880 		r = -EINVAL;
4881 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4882 			break;
4883 
4884 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4885 			kvm_can_mwait_in_guest())
4886 			kvm->arch.mwait_in_guest = true;
4887 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4888 			kvm->arch.hlt_in_guest = true;
4889 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4890 			kvm->arch.pause_in_guest = true;
4891 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4892 			kvm->arch.cstate_in_guest = true;
4893 		r = 0;
4894 		break;
4895 	case KVM_CAP_MSR_PLATFORM_INFO:
4896 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4897 		r = 0;
4898 		break;
4899 	case KVM_CAP_EXCEPTION_PAYLOAD:
4900 		kvm->arch.exception_payload_enabled = cap->args[0];
4901 		r = 0;
4902 		break;
4903 	default:
4904 		r = -EINVAL;
4905 		break;
4906 	}
4907 	return r;
4908 }
4909 
4910 long kvm_arch_vm_ioctl(struct file *filp,
4911 		       unsigned int ioctl, unsigned long arg)
4912 {
4913 	struct kvm *kvm = filp->private_data;
4914 	void __user *argp = (void __user *)arg;
4915 	int r = -ENOTTY;
4916 	/*
4917 	 * This union makes it completely explicit to gcc-3.x
4918 	 * that these two variables' stack usage should be
4919 	 * combined, not added together.
4920 	 */
4921 	union {
4922 		struct kvm_pit_state ps;
4923 		struct kvm_pit_state2 ps2;
4924 		struct kvm_pit_config pit_config;
4925 	} u;
4926 
4927 	switch (ioctl) {
4928 	case KVM_SET_TSS_ADDR:
4929 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4930 		break;
4931 	case KVM_SET_IDENTITY_MAP_ADDR: {
4932 		u64 ident_addr;
4933 
4934 		mutex_lock(&kvm->lock);
4935 		r = -EINVAL;
4936 		if (kvm->created_vcpus)
4937 			goto set_identity_unlock;
4938 		r = -EFAULT;
4939 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4940 			goto set_identity_unlock;
4941 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4942 set_identity_unlock:
4943 		mutex_unlock(&kvm->lock);
4944 		break;
4945 	}
4946 	case KVM_SET_NR_MMU_PAGES:
4947 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4948 		break;
4949 	case KVM_GET_NR_MMU_PAGES:
4950 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4951 		break;
4952 	case KVM_CREATE_IRQCHIP: {
4953 		mutex_lock(&kvm->lock);
4954 
4955 		r = -EEXIST;
4956 		if (irqchip_in_kernel(kvm))
4957 			goto create_irqchip_unlock;
4958 
4959 		r = -EINVAL;
4960 		if (kvm->created_vcpus)
4961 			goto create_irqchip_unlock;
4962 
4963 		r = kvm_pic_init(kvm);
4964 		if (r)
4965 			goto create_irqchip_unlock;
4966 
4967 		r = kvm_ioapic_init(kvm);
4968 		if (r) {
4969 			kvm_pic_destroy(kvm);
4970 			goto create_irqchip_unlock;
4971 		}
4972 
4973 		r = kvm_setup_default_irq_routing(kvm);
4974 		if (r) {
4975 			kvm_ioapic_destroy(kvm);
4976 			kvm_pic_destroy(kvm);
4977 			goto create_irqchip_unlock;
4978 		}
4979 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4980 		smp_wmb();
4981 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4982 	create_irqchip_unlock:
4983 		mutex_unlock(&kvm->lock);
4984 		break;
4985 	}
4986 	case KVM_CREATE_PIT:
4987 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4988 		goto create_pit;
4989 	case KVM_CREATE_PIT2:
4990 		r = -EFAULT;
4991 		if (copy_from_user(&u.pit_config, argp,
4992 				   sizeof(struct kvm_pit_config)))
4993 			goto out;
4994 	create_pit:
4995 		mutex_lock(&kvm->lock);
4996 		r = -EEXIST;
4997 		if (kvm->arch.vpit)
4998 			goto create_pit_unlock;
4999 		r = -ENOMEM;
5000 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5001 		if (kvm->arch.vpit)
5002 			r = 0;
5003 	create_pit_unlock:
5004 		mutex_unlock(&kvm->lock);
5005 		break;
5006 	case KVM_GET_IRQCHIP: {
5007 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5008 		struct kvm_irqchip *chip;
5009 
5010 		chip = memdup_user(argp, sizeof(*chip));
5011 		if (IS_ERR(chip)) {
5012 			r = PTR_ERR(chip);
5013 			goto out;
5014 		}
5015 
5016 		r = -ENXIO;
5017 		if (!irqchip_kernel(kvm))
5018 			goto get_irqchip_out;
5019 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5020 		if (r)
5021 			goto get_irqchip_out;
5022 		r = -EFAULT;
5023 		if (copy_to_user(argp, chip, sizeof(*chip)))
5024 			goto get_irqchip_out;
5025 		r = 0;
5026 	get_irqchip_out:
5027 		kfree(chip);
5028 		break;
5029 	}
5030 	case KVM_SET_IRQCHIP: {
5031 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5032 		struct kvm_irqchip *chip;
5033 
5034 		chip = memdup_user(argp, sizeof(*chip));
5035 		if (IS_ERR(chip)) {
5036 			r = PTR_ERR(chip);
5037 			goto out;
5038 		}
5039 
5040 		r = -ENXIO;
5041 		if (!irqchip_kernel(kvm))
5042 			goto set_irqchip_out;
5043 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5044 	set_irqchip_out:
5045 		kfree(chip);
5046 		break;
5047 	}
5048 	case KVM_GET_PIT: {
5049 		r = -EFAULT;
5050 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5051 			goto out;
5052 		r = -ENXIO;
5053 		if (!kvm->arch.vpit)
5054 			goto out;
5055 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5056 		if (r)
5057 			goto out;
5058 		r = -EFAULT;
5059 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5060 			goto out;
5061 		r = 0;
5062 		break;
5063 	}
5064 	case KVM_SET_PIT: {
5065 		r = -EFAULT;
5066 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5067 			goto out;
5068 		r = -ENXIO;
5069 		if (!kvm->arch.vpit)
5070 			goto out;
5071 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5072 		break;
5073 	}
5074 	case KVM_GET_PIT2: {
5075 		r = -ENXIO;
5076 		if (!kvm->arch.vpit)
5077 			goto out;
5078 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5079 		if (r)
5080 			goto out;
5081 		r = -EFAULT;
5082 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5083 			goto out;
5084 		r = 0;
5085 		break;
5086 	}
5087 	case KVM_SET_PIT2: {
5088 		r = -EFAULT;
5089 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5090 			goto out;
5091 		r = -ENXIO;
5092 		if (!kvm->arch.vpit)
5093 			goto out;
5094 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5095 		break;
5096 	}
5097 	case KVM_REINJECT_CONTROL: {
5098 		struct kvm_reinject_control control;
5099 		r =  -EFAULT;
5100 		if (copy_from_user(&control, argp, sizeof(control)))
5101 			goto out;
5102 		r = -ENXIO;
5103 		if (!kvm->arch.vpit)
5104 			goto out;
5105 		r = kvm_vm_ioctl_reinject(kvm, &control);
5106 		break;
5107 	}
5108 	case KVM_SET_BOOT_CPU_ID:
5109 		r = 0;
5110 		mutex_lock(&kvm->lock);
5111 		if (kvm->created_vcpus)
5112 			r = -EBUSY;
5113 		else
5114 			kvm->arch.bsp_vcpu_id = arg;
5115 		mutex_unlock(&kvm->lock);
5116 		break;
5117 	case KVM_XEN_HVM_CONFIG: {
5118 		struct kvm_xen_hvm_config xhc;
5119 		r = -EFAULT;
5120 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5121 			goto out;
5122 		r = -EINVAL;
5123 		if (xhc.flags)
5124 			goto out;
5125 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5126 		r = 0;
5127 		break;
5128 	}
5129 	case KVM_SET_CLOCK: {
5130 		struct kvm_clock_data user_ns;
5131 		u64 now_ns;
5132 
5133 		r = -EFAULT;
5134 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5135 			goto out;
5136 
5137 		r = -EINVAL;
5138 		if (user_ns.flags)
5139 			goto out;
5140 
5141 		r = 0;
5142 		/*
5143 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5144 		 * kvm_gen_update_masterclock() can be cut down to locked
5145 		 * pvclock_update_vm_gtod_copy().
5146 		 */
5147 		kvm_gen_update_masterclock(kvm);
5148 		now_ns = get_kvmclock_ns(kvm);
5149 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5150 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5151 		break;
5152 	}
5153 	case KVM_GET_CLOCK: {
5154 		struct kvm_clock_data user_ns;
5155 		u64 now_ns;
5156 
5157 		now_ns = get_kvmclock_ns(kvm);
5158 		user_ns.clock = now_ns;
5159 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5160 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5161 
5162 		r = -EFAULT;
5163 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5164 			goto out;
5165 		r = 0;
5166 		break;
5167 	}
5168 	case KVM_MEMORY_ENCRYPT_OP: {
5169 		r = -ENOTTY;
5170 		if (kvm_x86_ops->mem_enc_op)
5171 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
5172 		break;
5173 	}
5174 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5175 		struct kvm_enc_region region;
5176 
5177 		r = -EFAULT;
5178 		if (copy_from_user(&region, argp, sizeof(region)))
5179 			goto out;
5180 
5181 		r = -ENOTTY;
5182 		if (kvm_x86_ops->mem_enc_reg_region)
5183 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
5184 		break;
5185 	}
5186 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5187 		struct kvm_enc_region region;
5188 
5189 		r = -EFAULT;
5190 		if (copy_from_user(&region, argp, sizeof(region)))
5191 			goto out;
5192 
5193 		r = -ENOTTY;
5194 		if (kvm_x86_ops->mem_enc_unreg_region)
5195 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
5196 		break;
5197 	}
5198 	case KVM_HYPERV_EVENTFD: {
5199 		struct kvm_hyperv_eventfd hvevfd;
5200 
5201 		r = -EFAULT;
5202 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5203 			goto out;
5204 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5205 		break;
5206 	}
5207 	case KVM_SET_PMU_EVENT_FILTER:
5208 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5209 		break;
5210 	default:
5211 		r = -ENOTTY;
5212 	}
5213 out:
5214 	return r;
5215 }
5216 
5217 static void kvm_init_msr_list(void)
5218 {
5219 	struct x86_pmu_capability x86_pmu;
5220 	u32 dummy[2];
5221 	unsigned i;
5222 
5223 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5224 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5225 
5226 	perf_get_x86_pmu_capability(&x86_pmu);
5227 
5228 	num_msrs_to_save = 0;
5229 	num_emulated_msrs = 0;
5230 	num_msr_based_features = 0;
5231 
5232 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5233 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5234 			continue;
5235 
5236 		/*
5237 		 * Even MSRs that are valid in the host may not be exposed
5238 		 * to the guests in some cases.
5239 		 */
5240 		switch (msrs_to_save_all[i]) {
5241 		case MSR_IA32_BNDCFGS:
5242 			if (!kvm_mpx_supported())
5243 				continue;
5244 			break;
5245 		case MSR_TSC_AUX:
5246 			if (!kvm_x86_ops->rdtscp_supported())
5247 				continue;
5248 			break;
5249 		case MSR_IA32_RTIT_CTL:
5250 		case MSR_IA32_RTIT_STATUS:
5251 			if (!kvm_x86_ops->pt_supported())
5252 				continue;
5253 			break;
5254 		case MSR_IA32_RTIT_CR3_MATCH:
5255 			if (!kvm_x86_ops->pt_supported() ||
5256 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5257 				continue;
5258 			break;
5259 		case MSR_IA32_RTIT_OUTPUT_BASE:
5260 		case MSR_IA32_RTIT_OUTPUT_MASK:
5261 			if (!kvm_x86_ops->pt_supported() ||
5262 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5263 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5264 				continue;
5265 			break;
5266 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5267 			if (!kvm_x86_ops->pt_supported() ||
5268 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5269 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5270 				continue;
5271 			break;
5272 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5273 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5274 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5275 				continue;
5276 			break;
5277 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5278 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5279 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5280 				continue;
5281 		}
5282 		default:
5283 			break;
5284 		}
5285 
5286 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5287 	}
5288 
5289 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5290 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs_all[i]))
5291 			continue;
5292 
5293 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5294 	}
5295 
5296 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5297 		struct kvm_msr_entry msr;
5298 
5299 		msr.index = msr_based_features_all[i];
5300 		if (kvm_get_msr_feature(&msr))
5301 			continue;
5302 
5303 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5304 	}
5305 }
5306 
5307 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5308 			   const void *v)
5309 {
5310 	int handled = 0;
5311 	int n;
5312 
5313 	do {
5314 		n = min(len, 8);
5315 		if (!(lapic_in_kernel(vcpu) &&
5316 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5317 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5318 			break;
5319 		handled += n;
5320 		addr += n;
5321 		len -= n;
5322 		v += n;
5323 	} while (len);
5324 
5325 	return handled;
5326 }
5327 
5328 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5329 {
5330 	int handled = 0;
5331 	int n;
5332 
5333 	do {
5334 		n = min(len, 8);
5335 		if (!(lapic_in_kernel(vcpu) &&
5336 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5337 					 addr, n, v))
5338 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5339 			break;
5340 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5341 		handled += n;
5342 		addr += n;
5343 		len -= n;
5344 		v += n;
5345 	} while (len);
5346 
5347 	return handled;
5348 }
5349 
5350 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5351 			struct kvm_segment *var, int seg)
5352 {
5353 	kvm_x86_ops->set_segment(vcpu, var, seg);
5354 }
5355 
5356 void kvm_get_segment(struct kvm_vcpu *vcpu,
5357 		     struct kvm_segment *var, int seg)
5358 {
5359 	kvm_x86_ops->get_segment(vcpu, var, seg);
5360 }
5361 
5362 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5363 			   struct x86_exception *exception)
5364 {
5365 	gpa_t t_gpa;
5366 
5367 	BUG_ON(!mmu_is_nested(vcpu));
5368 
5369 	/* NPT walks are always user-walks */
5370 	access |= PFERR_USER_MASK;
5371 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5372 
5373 	return t_gpa;
5374 }
5375 
5376 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5377 			      struct x86_exception *exception)
5378 {
5379 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5380 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5381 }
5382 
5383  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5384 				struct x86_exception *exception)
5385 {
5386 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5387 	access |= PFERR_FETCH_MASK;
5388 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5389 }
5390 
5391 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5392 			       struct x86_exception *exception)
5393 {
5394 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5395 	access |= PFERR_WRITE_MASK;
5396 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5397 }
5398 
5399 /* uses this to access any guest's mapped memory without checking CPL */
5400 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5401 				struct x86_exception *exception)
5402 {
5403 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5404 }
5405 
5406 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5407 				      struct kvm_vcpu *vcpu, u32 access,
5408 				      struct x86_exception *exception)
5409 {
5410 	void *data = val;
5411 	int r = X86EMUL_CONTINUE;
5412 
5413 	while (bytes) {
5414 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5415 							    exception);
5416 		unsigned offset = addr & (PAGE_SIZE-1);
5417 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5418 		int ret;
5419 
5420 		if (gpa == UNMAPPED_GVA)
5421 			return X86EMUL_PROPAGATE_FAULT;
5422 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5423 					       offset, toread);
5424 		if (ret < 0) {
5425 			r = X86EMUL_IO_NEEDED;
5426 			goto out;
5427 		}
5428 
5429 		bytes -= toread;
5430 		data += toread;
5431 		addr += toread;
5432 	}
5433 out:
5434 	return r;
5435 }
5436 
5437 /* used for instruction fetching */
5438 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5439 				gva_t addr, void *val, unsigned int bytes,
5440 				struct x86_exception *exception)
5441 {
5442 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5443 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5444 	unsigned offset;
5445 	int ret;
5446 
5447 	/* Inline kvm_read_guest_virt_helper for speed.  */
5448 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5449 						    exception);
5450 	if (unlikely(gpa == UNMAPPED_GVA))
5451 		return X86EMUL_PROPAGATE_FAULT;
5452 
5453 	offset = addr & (PAGE_SIZE-1);
5454 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5455 		bytes = (unsigned)PAGE_SIZE - offset;
5456 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5457 				       offset, bytes);
5458 	if (unlikely(ret < 0))
5459 		return X86EMUL_IO_NEEDED;
5460 
5461 	return X86EMUL_CONTINUE;
5462 }
5463 
5464 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5465 			       gva_t addr, void *val, unsigned int bytes,
5466 			       struct x86_exception *exception)
5467 {
5468 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5469 
5470 	/*
5471 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5472 	 * is returned, but our callers are not ready for that and they blindly
5473 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5474 	 * uninitialized kernel stack memory into cr2 and error code.
5475 	 */
5476 	memset(exception, 0, sizeof(*exception));
5477 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5478 					  exception);
5479 }
5480 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5481 
5482 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5483 			     gva_t addr, void *val, unsigned int bytes,
5484 			     struct x86_exception *exception, bool system)
5485 {
5486 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5487 	u32 access = 0;
5488 
5489 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5490 		access |= PFERR_USER_MASK;
5491 
5492 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5493 }
5494 
5495 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5496 		unsigned long addr, void *val, unsigned int bytes)
5497 {
5498 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5499 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5500 
5501 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5502 }
5503 
5504 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5505 				      struct kvm_vcpu *vcpu, u32 access,
5506 				      struct x86_exception *exception)
5507 {
5508 	void *data = val;
5509 	int r = X86EMUL_CONTINUE;
5510 
5511 	while (bytes) {
5512 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5513 							     access,
5514 							     exception);
5515 		unsigned offset = addr & (PAGE_SIZE-1);
5516 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5517 		int ret;
5518 
5519 		if (gpa == UNMAPPED_GVA)
5520 			return X86EMUL_PROPAGATE_FAULT;
5521 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5522 		if (ret < 0) {
5523 			r = X86EMUL_IO_NEEDED;
5524 			goto out;
5525 		}
5526 
5527 		bytes -= towrite;
5528 		data += towrite;
5529 		addr += towrite;
5530 	}
5531 out:
5532 	return r;
5533 }
5534 
5535 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5536 			      unsigned int bytes, struct x86_exception *exception,
5537 			      bool system)
5538 {
5539 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5540 	u32 access = PFERR_WRITE_MASK;
5541 
5542 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5543 		access |= PFERR_USER_MASK;
5544 
5545 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5546 					   access, exception);
5547 }
5548 
5549 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5550 				unsigned int bytes, struct x86_exception *exception)
5551 {
5552 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5553 	vcpu->arch.l1tf_flush_l1d = true;
5554 
5555 	/*
5556 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5557 	 * is returned, but our callers are not ready for that and they blindly
5558 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5559 	 * uninitialized kernel stack memory into cr2 and error code.
5560 	 */
5561 	memset(exception, 0, sizeof(*exception));
5562 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5563 					   PFERR_WRITE_MASK, exception);
5564 }
5565 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5566 
5567 int handle_ud(struct kvm_vcpu *vcpu)
5568 {
5569 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5570 	int emul_type = EMULTYPE_TRAP_UD;
5571 	char sig[5]; /* ud2; .ascii "kvm" */
5572 	struct x86_exception e;
5573 
5574 	if (force_emulation_prefix &&
5575 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5576 				sig, sizeof(sig), &e) == 0 &&
5577 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5578 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5579 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5580 	}
5581 
5582 	return kvm_emulate_instruction(vcpu, emul_type);
5583 }
5584 EXPORT_SYMBOL_GPL(handle_ud);
5585 
5586 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5587 			    gpa_t gpa, bool write)
5588 {
5589 	/* For APIC access vmexit */
5590 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5591 		return 1;
5592 
5593 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5594 		trace_vcpu_match_mmio(gva, gpa, write, true);
5595 		return 1;
5596 	}
5597 
5598 	return 0;
5599 }
5600 
5601 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5602 				gpa_t *gpa, struct x86_exception *exception,
5603 				bool write)
5604 {
5605 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5606 		| (write ? PFERR_WRITE_MASK : 0);
5607 
5608 	/*
5609 	 * currently PKRU is only applied to ept enabled guest so
5610 	 * there is no pkey in EPT page table for L1 guest or EPT
5611 	 * shadow page table for L2 guest.
5612 	 */
5613 	if (vcpu_match_mmio_gva(vcpu, gva)
5614 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5615 				 vcpu->arch.mmio_access, 0, access)) {
5616 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5617 					(gva & (PAGE_SIZE - 1));
5618 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5619 		return 1;
5620 	}
5621 
5622 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5623 
5624 	if (*gpa == UNMAPPED_GVA)
5625 		return -1;
5626 
5627 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5628 }
5629 
5630 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5631 			const void *val, int bytes)
5632 {
5633 	int ret;
5634 
5635 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5636 	if (ret < 0)
5637 		return 0;
5638 	kvm_page_track_write(vcpu, gpa, val, bytes);
5639 	return 1;
5640 }
5641 
5642 struct read_write_emulator_ops {
5643 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5644 				  int bytes);
5645 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5646 				  void *val, int bytes);
5647 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5648 			       int bytes, void *val);
5649 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5650 				    void *val, int bytes);
5651 	bool write;
5652 };
5653 
5654 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5655 {
5656 	if (vcpu->mmio_read_completed) {
5657 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5658 			       vcpu->mmio_fragments[0].gpa, val);
5659 		vcpu->mmio_read_completed = 0;
5660 		return 1;
5661 	}
5662 
5663 	return 0;
5664 }
5665 
5666 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5667 			void *val, int bytes)
5668 {
5669 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5670 }
5671 
5672 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5673 			 void *val, int bytes)
5674 {
5675 	return emulator_write_phys(vcpu, gpa, val, bytes);
5676 }
5677 
5678 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5679 {
5680 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5681 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5682 }
5683 
5684 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5685 			  void *val, int bytes)
5686 {
5687 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5688 	return X86EMUL_IO_NEEDED;
5689 }
5690 
5691 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5692 			   void *val, int bytes)
5693 {
5694 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5695 
5696 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5697 	return X86EMUL_CONTINUE;
5698 }
5699 
5700 static const struct read_write_emulator_ops read_emultor = {
5701 	.read_write_prepare = read_prepare,
5702 	.read_write_emulate = read_emulate,
5703 	.read_write_mmio = vcpu_mmio_read,
5704 	.read_write_exit_mmio = read_exit_mmio,
5705 };
5706 
5707 static const struct read_write_emulator_ops write_emultor = {
5708 	.read_write_emulate = write_emulate,
5709 	.read_write_mmio = write_mmio,
5710 	.read_write_exit_mmio = write_exit_mmio,
5711 	.write = true,
5712 };
5713 
5714 static int emulator_read_write_onepage(unsigned long addr, void *val,
5715 				       unsigned int bytes,
5716 				       struct x86_exception *exception,
5717 				       struct kvm_vcpu *vcpu,
5718 				       const struct read_write_emulator_ops *ops)
5719 {
5720 	gpa_t gpa;
5721 	int handled, ret;
5722 	bool write = ops->write;
5723 	struct kvm_mmio_fragment *frag;
5724 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5725 
5726 	/*
5727 	 * If the exit was due to a NPF we may already have a GPA.
5728 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5729 	 * Note, this cannot be used on string operations since string
5730 	 * operation using rep will only have the initial GPA from the NPF
5731 	 * occurred.
5732 	 */
5733 	if (vcpu->arch.gpa_available &&
5734 	    emulator_can_use_gpa(ctxt) &&
5735 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5736 		gpa = vcpu->arch.gpa_val;
5737 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5738 	} else {
5739 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5740 		if (ret < 0)
5741 			return X86EMUL_PROPAGATE_FAULT;
5742 	}
5743 
5744 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5745 		return X86EMUL_CONTINUE;
5746 
5747 	/*
5748 	 * Is this MMIO handled locally?
5749 	 */
5750 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5751 	if (handled == bytes)
5752 		return X86EMUL_CONTINUE;
5753 
5754 	gpa += handled;
5755 	bytes -= handled;
5756 	val += handled;
5757 
5758 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5759 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5760 	frag->gpa = gpa;
5761 	frag->data = val;
5762 	frag->len = bytes;
5763 	return X86EMUL_CONTINUE;
5764 }
5765 
5766 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5767 			unsigned long addr,
5768 			void *val, unsigned int bytes,
5769 			struct x86_exception *exception,
5770 			const struct read_write_emulator_ops *ops)
5771 {
5772 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5773 	gpa_t gpa;
5774 	int rc;
5775 
5776 	if (ops->read_write_prepare &&
5777 		  ops->read_write_prepare(vcpu, val, bytes))
5778 		return X86EMUL_CONTINUE;
5779 
5780 	vcpu->mmio_nr_fragments = 0;
5781 
5782 	/* Crossing a page boundary? */
5783 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5784 		int now;
5785 
5786 		now = -addr & ~PAGE_MASK;
5787 		rc = emulator_read_write_onepage(addr, val, now, exception,
5788 						 vcpu, ops);
5789 
5790 		if (rc != X86EMUL_CONTINUE)
5791 			return rc;
5792 		addr += now;
5793 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5794 			addr = (u32)addr;
5795 		val += now;
5796 		bytes -= now;
5797 	}
5798 
5799 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5800 					 vcpu, ops);
5801 	if (rc != X86EMUL_CONTINUE)
5802 		return rc;
5803 
5804 	if (!vcpu->mmio_nr_fragments)
5805 		return rc;
5806 
5807 	gpa = vcpu->mmio_fragments[0].gpa;
5808 
5809 	vcpu->mmio_needed = 1;
5810 	vcpu->mmio_cur_fragment = 0;
5811 
5812 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5813 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5814 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5815 	vcpu->run->mmio.phys_addr = gpa;
5816 
5817 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5818 }
5819 
5820 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5821 				  unsigned long addr,
5822 				  void *val,
5823 				  unsigned int bytes,
5824 				  struct x86_exception *exception)
5825 {
5826 	return emulator_read_write(ctxt, addr, val, bytes,
5827 				   exception, &read_emultor);
5828 }
5829 
5830 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5831 			    unsigned long addr,
5832 			    const void *val,
5833 			    unsigned int bytes,
5834 			    struct x86_exception *exception)
5835 {
5836 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5837 				   exception, &write_emultor);
5838 }
5839 
5840 #define CMPXCHG_TYPE(t, ptr, old, new) \
5841 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5842 
5843 #ifdef CONFIG_X86_64
5844 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5845 #else
5846 #  define CMPXCHG64(ptr, old, new) \
5847 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5848 #endif
5849 
5850 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5851 				     unsigned long addr,
5852 				     const void *old,
5853 				     const void *new,
5854 				     unsigned int bytes,
5855 				     struct x86_exception *exception)
5856 {
5857 	struct kvm_host_map map;
5858 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5859 	gpa_t gpa;
5860 	char *kaddr;
5861 	bool exchanged;
5862 
5863 	/* guests cmpxchg8b have to be emulated atomically */
5864 	if (bytes > 8 || (bytes & (bytes - 1)))
5865 		goto emul_write;
5866 
5867 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5868 
5869 	if (gpa == UNMAPPED_GVA ||
5870 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5871 		goto emul_write;
5872 
5873 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5874 		goto emul_write;
5875 
5876 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5877 		goto emul_write;
5878 
5879 	kaddr = map.hva + offset_in_page(gpa);
5880 
5881 	switch (bytes) {
5882 	case 1:
5883 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5884 		break;
5885 	case 2:
5886 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5887 		break;
5888 	case 4:
5889 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5890 		break;
5891 	case 8:
5892 		exchanged = CMPXCHG64(kaddr, old, new);
5893 		break;
5894 	default:
5895 		BUG();
5896 	}
5897 
5898 	kvm_vcpu_unmap(vcpu, &map, true);
5899 
5900 	if (!exchanged)
5901 		return X86EMUL_CMPXCHG_FAILED;
5902 
5903 	kvm_page_track_write(vcpu, gpa, new, bytes);
5904 
5905 	return X86EMUL_CONTINUE;
5906 
5907 emul_write:
5908 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5909 
5910 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5911 }
5912 
5913 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5914 {
5915 	int r = 0, i;
5916 
5917 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5918 		if (vcpu->arch.pio.in)
5919 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5920 					    vcpu->arch.pio.size, pd);
5921 		else
5922 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5923 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5924 					     pd);
5925 		if (r)
5926 			break;
5927 		pd += vcpu->arch.pio.size;
5928 	}
5929 	return r;
5930 }
5931 
5932 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5933 			       unsigned short port, void *val,
5934 			       unsigned int count, bool in)
5935 {
5936 	vcpu->arch.pio.port = port;
5937 	vcpu->arch.pio.in = in;
5938 	vcpu->arch.pio.count  = count;
5939 	vcpu->arch.pio.size = size;
5940 
5941 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5942 		vcpu->arch.pio.count = 0;
5943 		return 1;
5944 	}
5945 
5946 	vcpu->run->exit_reason = KVM_EXIT_IO;
5947 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5948 	vcpu->run->io.size = size;
5949 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5950 	vcpu->run->io.count = count;
5951 	vcpu->run->io.port = port;
5952 
5953 	return 0;
5954 }
5955 
5956 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5957 				    int size, unsigned short port, void *val,
5958 				    unsigned int count)
5959 {
5960 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5961 	int ret;
5962 
5963 	if (vcpu->arch.pio.count)
5964 		goto data_avail;
5965 
5966 	memset(vcpu->arch.pio_data, 0, size * count);
5967 
5968 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5969 	if (ret) {
5970 data_avail:
5971 		memcpy(val, vcpu->arch.pio_data, size * count);
5972 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5973 		vcpu->arch.pio.count = 0;
5974 		return 1;
5975 	}
5976 
5977 	return 0;
5978 }
5979 
5980 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5981 				     int size, unsigned short port,
5982 				     const void *val, unsigned int count)
5983 {
5984 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5985 
5986 	memcpy(vcpu->arch.pio_data, val, size * count);
5987 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5988 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5989 }
5990 
5991 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5992 {
5993 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5994 }
5995 
5996 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5997 {
5998 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5999 }
6000 
6001 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6002 {
6003 	if (!need_emulate_wbinvd(vcpu))
6004 		return X86EMUL_CONTINUE;
6005 
6006 	if (kvm_x86_ops->has_wbinvd_exit()) {
6007 		int cpu = get_cpu();
6008 
6009 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6010 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6011 				wbinvd_ipi, NULL, 1);
6012 		put_cpu();
6013 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6014 	} else
6015 		wbinvd();
6016 	return X86EMUL_CONTINUE;
6017 }
6018 
6019 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6020 {
6021 	kvm_emulate_wbinvd_noskip(vcpu);
6022 	return kvm_skip_emulated_instruction(vcpu);
6023 }
6024 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6025 
6026 
6027 
6028 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6029 {
6030 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6031 }
6032 
6033 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6034 			   unsigned long *dest)
6035 {
6036 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6037 }
6038 
6039 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6040 			   unsigned long value)
6041 {
6042 
6043 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6044 }
6045 
6046 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6047 {
6048 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6049 }
6050 
6051 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6052 {
6053 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6054 	unsigned long value;
6055 
6056 	switch (cr) {
6057 	case 0:
6058 		value = kvm_read_cr0(vcpu);
6059 		break;
6060 	case 2:
6061 		value = vcpu->arch.cr2;
6062 		break;
6063 	case 3:
6064 		value = kvm_read_cr3(vcpu);
6065 		break;
6066 	case 4:
6067 		value = kvm_read_cr4(vcpu);
6068 		break;
6069 	case 8:
6070 		value = kvm_get_cr8(vcpu);
6071 		break;
6072 	default:
6073 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6074 		return 0;
6075 	}
6076 
6077 	return value;
6078 }
6079 
6080 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6081 {
6082 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6083 	int res = 0;
6084 
6085 	switch (cr) {
6086 	case 0:
6087 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6088 		break;
6089 	case 2:
6090 		vcpu->arch.cr2 = val;
6091 		break;
6092 	case 3:
6093 		res = kvm_set_cr3(vcpu, val);
6094 		break;
6095 	case 4:
6096 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6097 		break;
6098 	case 8:
6099 		res = kvm_set_cr8(vcpu, val);
6100 		break;
6101 	default:
6102 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6103 		res = -1;
6104 	}
6105 
6106 	return res;
6107 }
6108 
6109 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6110 {
6111 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6112 }
6113 
6114 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6115 {
6116 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6117 }
6118 
6119 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6120 {
6121 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6122 }
6123 
6124 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6125 {
6126 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6127 }
6128 
6129 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6130 {
6131 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6132 }
6133 
6134 static unsigned long emulator_get_cached_segment_base(
6135 	struct x86_emulate_ctxt *ctxt, int seg)
6136 {
6137 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6138 }
6139 
6140 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6141 				 struct desc_struct *desc, u32 *base3,
6142 				 int seg)
6143 {
6144 	struct kvm_segment var;
6145 
6146 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6147 	*selector = var.selector;
6148 
6149 	if (var.unusable) {
6150 		memset(desc, 0, sizeof(*desc));
6151 		if (base3)
6152 			*base3 = 0;
6153 		return false;
6154 	}
6155 
6156 	if (var.g)
6157 		var.limit >>= 12;
6158 	set_desc_limit(desc, var.limit);
6159 	set_desc_base(desc, (unsigned long)var.base);
6160 #ifdef CONFIG_X86_64
6161 	if (base3)
6162 		*base3 = var.base >> 32;
6163 #endif
6164 	desc->type = var.type;
6165 	desc->s = var.s;
6166 	desc->dpl = var.dpl;
6167 	desc->p = var.present;
6168 	desc->avl = var.avl;
6169 	desc->l = var.l;
6170 	desc->d = var.db;
6171 	desc->g = var.g;
6172 
6173 	return true;
6174 }
6175 
6176 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6177 				 struct desc_struct *desc, u32 base3,
6178 				 int seg)
6179 {
6180 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6181 	struct kvm_segment var;
6182 
6183 	var.selector = selector;
6184 	var.base = get_desc_base(desc);
6185 #ifdef CONFIG_X86_64
6186 	var.base |= ((u64)base3) << 32;
6187 #endif
6188 	var.limit = get_desc_limit(desc);
6189 	if (desc->g)
6190 		var.limit = (var.limit << 12) | 0xfff;
6191 	var.type = desc->type;
6192 	var.dpl = desc->dpl;
6193 	var.db = desc->d;
6194 	var.s = desc->s;
6195 	var.l = desc->l;
6196 	var.g = desc->g;
6197 	var.avl = desc->avl;
6198 	var.present = desc->p;
6199 	var.unusable = !var.present;
6200 	var.padding = 0;
6201 
6202 	kvm_set_segment(vcpu, &var, seg);
6203 	return;
6204 }
6205 
6206 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6207 			    u32 msr_index, u64 *pdata)
6208 {
6209 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6210 }
6211 
6212 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6213 			    u32 msr_index, u64 data)
6214 {
6215 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6216 }
6217 
6218 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6219 {
6220 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6221 
6222 	return vcpu->arch.smbase;
6223 }
6224 
6225 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6226 {
6227 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6228 
6229 	vcpu->arch.smbase = smbase;
6230 }
6231 
6232 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6233 			      u32 pmc)
6234 {
6235 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6236 }
6237 
6238 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6239 			     u32 pmc, u64 *pdata)
6240 {
6241 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6242 }
6243 
6244 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6245 {
6246 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6247 }
6248 
6249 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6250 			      struct x86_instruction_info *info,
6251 			      enum x86_intercept_stage stage)
6252 {
6253 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6254 }
6255 
6256 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6257 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6258 {
6259 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6260 }
6261 
6262 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6263 {
6264 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6265 }
6266 
6267 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6268 {
6269 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6270 }
6271 
6272 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6273 {
6274 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6275 }
6276 
6277 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6278 {
6279 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6280 }
6281 
6282 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6283 {
6284 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6285 }
6286 
6287 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6288 {
6289 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6290 }
6291 
6292 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6293 {
6294 	return emul_to_vcpu(ctxt)->arch.hflags;
6295 }
6296 
6297 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6298 {
6299 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6300 }
6301 
6302 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6303 				  const char *smstate)
6304 {
6305 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6306 }
6307 
6308 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6309 {
6310 	kvm_smm_changed(emul_to_vcpu(ctxt));
6311 }
6312 
6313 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6314 {
6315 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6316 }
6317 
6318 static const struct x86_emulate_ops emulate_ops = {
6319 	.read_gpr            = emulator_read_gpr,
6320 	.write_gpr           = emulator_write_gpr,
6321 	.read_std            = emulator_read_std,
6322 	.write_std           = emulator_write_std,
6323 	.read_phys           = kvm_read_guest_phys_system,
6324 	.fetch               = kvm_fetch_guest_virt,
6325 	.read_emulated       = emulator_read_emulated,
6326 	.write_emulated      = emulator_write_emulated,
6327 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6328 	.invlpg              = emulator_invlpg,
6329 	.pio_in_emulated     = emulator_pio_in_emulated,
6330 	.pio_out_emulated    = emulator_pio_out_emulated,
6331 	.get_segment         = emulator_get_segment,
6332 	.set_segment         = emulator_set_segment,
6333 	.get_cached_segment_base = emulator_get_cached_segment_base,
6334 	.get_gdt             = emulator_get_gdt,
6335 	.get_idt	     = emulator_get_idt,
6336 	.set_gdt             = emulator_set_gdt,
6337 	.set_idt	     = emulator_set_idt,
6338 	.get_cr              = emulator_get_cr,
6339 	.set_cr              = emulator_set_cr,
6340 	.cpl                 = emulator_get_cpl,
6341 	.get_dr              = emulator_get_dr,
6342 	.set_dr              = emulator_set_dr,
6343 	.get_smbase          = emulator_get_smbase,
6344 	.set_smbase          = emulator_set_smbase,
6345 	.set_msr             = emulator_set_msr,
6346 	.get_msr             = emulator_get_msr,
6347 	.check_pmc	     = emulator_check_pmc,
6348 	.read_pmc            = emulator_read_pmc,
6349 	.halt                = emulator_halt,
6350 	.wbinvd              = emulator_wbinvd,
6351 	.fix_hypercall       = emulator_fix_hypercall,
6352 	.intercept           = emulator_intercept,
6353 	.get_cpuid           = emulator_get_cpuid,
6354 	.guest_has_long_mode = emulator_guest_has_long_mode,
6355 	.guest_has_movbe     = emulator_guest_has_movbe,
6356 	.guest_has_fxsr      = emulator_guest_has_fxsr,
6357 	.set_nmi_mask        = emulator_set_nmi_mask,
6358 	.get_hflags          = emulator_get_hflags,
6359 	.set_hflags          = emulator_set_hflags,
6360 	.pre_leave_smm       = emulator_pre_leave_smm,
6361 	.post_leave_smm      = emulator_post_leave_smm,
6362 	.set_xcr             = emulator_set_xcr,
6363 };
6364 
6365 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6366 {
6367 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6368 	/*
6369 	 * an sti; sti; sequence only disable interrupts for the first
6370 	 * instruction. So, if the last instruction, be it emulated or
6371 	 * not, left the system with the INT_STI flag enabled, it
6372 	 * means that the last instruction is an sti. We should not
6373 	 * leave the flag on in this case. The same goes for mov ss
6374 	 */
6375 	if (int_shadow & mask)
6376 		mask = 0;
6377 	if (unlikely(int_shadow || mask)) {
6378 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6379 		if (!mask)
6380 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6381 	}
6382 }
6383 
6384 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6385 {
6386 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6387 	if (ctxt->exception.vector == PF_VECTOR)
6388 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6389 
6390 	if (ctxt->exception.error_code_valid)
6391 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6392 				      ctxt->exception.error_code);
6393 	else
6394 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6395 	return false;
6396 }
6397 
6398 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6399 {
6400 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6401 	int cs_db, cs_l;
6402 
6403 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6404 
6405 	ctxt->eflags = kvm_get_rflags(vcpu);
6406 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6407 
6408 	ctxt->eip = kvm_rip_read(vcpu);
6409 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6410 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6411 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6412 		     cs_db				? X86EMUL_MODE_PROT32 :
6413 							  X86EMUL_MODE_PROT16;
6414 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6415 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6416 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6417 
6418 	init_decode_cache(ctxt);
6419 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6420 }
6421 
6422 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6423 {
6424 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6425 	int ret;
6426 
6427 	init_emulate_ctxt(vcpu);
6428 
6429 	ctxt->op_bytes = 2;
6430 	ctxt->ad_bytes = 2;
6431 	ctxt->_eip = ctxt->eip + inc_eip;
6432 	ret = emulate_int_real(ctxt, irq);
6433 
6434 	if (ret != X86EMUL_CONTINUE) {
6435 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6436 	} else {
6437 		ctxt->eip = ctxt->_eip;
6438 		kvm_rip_write(vcpu, ctxt->eip);
6439 		kvm_set_rflags(vcpu, ctxt->eflags);
6440 	}
6441 }
6442 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6443 
6444 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6445 {
6446 	++vcpu->stat.insn_emulation_fail;
6447 	trace_kvm_emulate_insn_failed(vcpu);
6448 
6449 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6450 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6451 		return 1;
6452 	}
6453 
6454 	if (emulation_type & EMULTYPE_SKIP) {
6455 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6456 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6457 		vcpu->run->internal.ndata = 0;
6458 		return 0;
6459 	}
6460 
6461 	kvm_queue_exception(vcpu, UD_VECTOR);
6462 
6463 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6464 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6465 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6466 		vcpu->run->internal.ndata = 0;
6467 		return 0;
6468 	}
6469 
6470 	return 1;
6471 }
6472 
6473 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6474 				  bool write_fault_to_shadow_pgtable,
6475 				  int emulation_type)
6476 {
6477 	gpa_t gpa = cr2_or_gpa;
6478 	kvm_pfn_t pfn;
6479 
6480 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6481 		return false;
6482 
6483 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6484 		return false;
6485 
6486 	if (!vcpu->arch.mmu->direct_map) {
6487 		/*
6488 		 * Write permission should be allowed since only
6489 		 * write access need to be emulated.
6490 		 */
6491 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6492 
6493 		/*
6494 		 * If the mapping is invalid in guest, let cpu retry
6495 		 * it to generate fault.
6496 		 */
6497 		if (gpa == UNMAPPED_GVA)
6498 			return true;
6499 	}
6500 
6501 	/*
6502 	 * Do not retry the unhandleable instruction if it faults on the
6503 	 * readonly host memory, otherwise it will goto a infinite loop:
6504 	 * retry instruction -> write #PF -> emulation fail -> retry
6505 	 * instruction -> ...
6506 	 */
6507 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6508 
6509 	/*
6510 	 * If the instruction failed on the error pfn, it can not be fixed,
6511 	 * report the error to userspace.
6512 	 */
6513 	if (is_error_noslot_pfn(pfn))
6514 		return false;
6515 
6516 	kvm_release_pfn_clean(pfn);
6517 
6518 	/* The instructions are well-emulated on direct mmu. */
6519 	if (vcpu->arch.mmu->direct_map) {
6520 		unsigned int indirect_shadow_pages;
6521 
6522 		spin_lock(&vcpu->kvm->mmu_lock);
6523 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6524 		spin_unlock(&vcpu->kvm->mmu_lock);
6525 
6526 		if (indirect_shadow_pages)
6527 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6528 
6529 		return true;
6530 	}
6531 
6532 	/*
6533 	 * if emulation was due to access to shadowed page table
6534 	 * and it failed try to unshadow page and re-enter the
6535 	 * guest to let CPU execute the instruction.
6536 	 */
6537 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6538 
6539 	/*
6540 	 * If the access faults on its page table, it can not
6541 	 * be fixed by unprotecting shadow page and it should
6542 	 * be reported to userspace.
6543 	 */
6544 	return !write_fault_to_shadow_pgtable;
6545 }
6546 
6547 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6548 			      gpa_t cr2_or_gpa,  int emulation_type)
6549 {
6550 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6551 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6552 
6553 	last_retry_eip = vcpu->arch.last_retry_eip;
6554 	last_retry_addr = vcpu->arch.last_retry_addr;
6555 
6556 	/*
6557 	 * If the emulation is caused by #PF and it is non-page_table
6558 	 * writing instruction, it means the VM-EXIT is caused by shadow
6559 	 * page protected, we can zap the shadow page and retry this
6560 	 * instruction directly.
6561 	 *
6562 	 * Note: if the guest uses a non-page-table modifying instruction
6563 	 * on the PDE that points to the instruction, then we will unmap
6564 	 * the instruction and go to an infinite loop. So, we cache the
6565 	 * last retried eip and the last fault address, if we meet the eip
6566 	 * and the address again, we can break out of the potential infinite
6567 	 * loop.
6568 	 */
6569 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6570 
6571 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6572 		return false;
6573 
6574 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6575 		return false;
6576 
6577 	if (x86_page_table_writing_insn(ctxt))
6578 		return false;
6579 
6580 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6581 		return false;
6582 
6583 	vcpu->arch.last_retry_eip = ctxt->eip;
6584 	vcpu->arch.last_retry_addr = cr2_or_gpa;
6585 
6586 	if (!vcpu->arch.mmu->direct_map)
6587 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6588 
6589 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6590 
6591 	return true;
6592 }
6593 
6594 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6595 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6596 
6597 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6598 {
6599 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6600 		/* This is a good place to trace that we are exiting SMM.  */
6601 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6602 
6603 		/* Process a latched INIT or SMI, if any.  */
6604 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6605 	}
6606 
6607 	kvm_mmu_reset_context(vcpu);
6608 }
6609 
6610 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6611 				unsigned long *db)
6612 {
6613 	u32 dr6 = 0;
6614 	int i;
6615 	u32 enable, rwlen;
6616 
6617 	enable = dr7;
6618 	rwlen = dr7 >> 16;
6619 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6620 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6621 			dr6 |= (1 << i);
6622 	return dr6;
6623 }
6624 
6625 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6626 {
6627 	struct kvm_run *kvm_run = vcpu->run;
6628 
6629 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6630 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6631 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6632 		kvm_run->debug.arch.exception = DB_VECTOR;
6633 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6634 		return 0;
6635 	}
6636 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6637 	return 1;
6638 }
6639 
6640 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6641 {
6642 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6643 	int r;
6644 
6645 	r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6646 	if (unlikely(!r))
6647 		return 0;
6648 
6649 	/*
6650 	 * rflags is the old, "raw" value of the flags.  The new value has
6651 	 * not been saved yet.
6652 	 *
6653 	 * This is correct even for TF set by the guest, because "the
6654 	 * processor will not generate this exception after the instruction
6655 	 * that sets the TF flag".
6656 	 */
6657 	if (unlikely(rflags & X86_EFLAGS_TF))
6658 		r = kvm_vcpu_do_singlestep(vcpu);
6659 	return r;
6660 }
6661 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6662 
6663 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6664 {
6665 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6666 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6667 		struct kvm_run *kvm_run = vcpu->run;
6668 		unsigned long eip = kvm_get_linear_rip(vcpu);
6669 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6670 					   vcpu->arch.guest_debug_dr7,
6671 					   vcpu->arch.eff_db);
6672 
6673 		if (dr6 != 0) {
6674 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6675 			kvm_run->debug.arch.pc = eip;
6676 			kvm_run->debug.arch.exception = DB_VECTOR;
6677 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6678 			*r = 0;
6679 			return true;
6680 		}
6681 	}
6682 
6683 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6684 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6685 		unsigned long eip = kvm_get_linear_rip(vcpu);
6686 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6687 					   vcpu->arch.dr7,
6688 					   vcpu->arch.db);
6689 
6690 		if (dr6 != 0) {
6691 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6692 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6693 			kvm_queue_exception(vcpu, DB_VECTOR);
6694 			*r = 1;
6695 			return true;
6696 		}
6697 	}
6698 
6699 	return false;
6700 }
6701 
6702 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6703 {
6704 	switch (ctxt->opcode_len) {
6705 	case 1:
6706 		switch (ctxt->b) {
6707 		case 0xe4:	/* IN */
6708 		case 0xe5:
6709 		case 0xec:
6710 		case 0xed:
6711 		case 0xe6:	/* OUT */
6712 		case 0xe7:
6713 		case 0xee:
6714 		case 0xef:
6715 		case 0x6c:	/* INS */
6716 		case 0x6d:
6717 		case 0x6e:	/* OUTS */
6718 		case 0x6f:
6719 			return true;
6720 		}
6721 		break;
6722 	case 2:
6723 		switch (ctxt->b) {
6724 		case 0x33:	/* RDPMC */
6725 			return true;
6726 		}
6727 		break;
6728 	}
6729 
6730 	return false;
6731 }
6732 
6733 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6734 			    int emulation_type, void *insn, int insn_len)
6735 {
6736 	int r;
6737 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6738 	bool writeback = true;
6739 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6740 
6741 	vcpu->arch.l1tf_flush_l1d = true;
6742 
6743 	/*
6744 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6745 	 * never reused.
6746 	 */
6747 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6748 	kvm_clear_exception_queue(vcpu);
6749 
6750 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6751 		init_emulate_ctxt(vcpu);
6752 
6753 		/*
6754 		 * We will reenter on the same instruction since
6755 		 * we do not set complete_userspace_io.  This does not
6756 		 * handle watchpoints yet, those would be handled in
6757 		 * the emulate_ops.
6758 		 */
6759 		if (!(emulation_type & EMULTYPE_SKIP) &&
6760 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6761 			return r;
6762 
6763 		ctxt->interruptibility = 0;
6764 		ctxt->have_exception = false;
6765 		ctxt->exception.vector = -1;
6766 		ctxt->perm_ok = false;
6767 
6768 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6769 
6770 		r = x86_decode_insn(ctxt, insn, insn_len);
6771 
6772 		trace_kvm_emulate_insn_start(vcpu);
6773 		++vcpu->stat.insn_emulation;
6774 		if (r != EMULATION_OK)  {
6775 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6776 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6777 				kvm_queue_exception(vcpu, UD_VECTOR);
6778 				return 1;
6779 			}
6780 			if (reexecute_instruction(vcpu, cr2_or_gpa,
6781 						  write_fault_to_spt,
6782 						  emulation_type))
6783 				return 1;
6784 			if (ctxt->have_exception) {
6785 				/*
6786 				 * #UD should result in just EMULATION_FAILED, and trap-like
6787 				 * exception should not be encountered during decode.
6788 				 */
6789 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6790 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6791 				inject_emulated_exception(vcpu);
6792 				return 1;
6793 			}
6794 			return handle_emulation_failure(vcpu, emulation_type);
6795 		}
6796 	}
6797 
6798 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6799 	    !is_vmware_backdoor_opcode(ctxt)) {
6800 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6801 		return 1;
6802 	}
6803 
6804 	/*
6805 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6806 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6807 	 * updating interruptibility state and injecting single-step #DBs.
6808 	 */
6809 	if (emulation_type & EMULTYPE_SKIP) {
6810 		kvm_rip_write(vcpu, ctxt->_eip);
6811 		if (ctxt->eflags & X86_EFLAGS_RF)
6812 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6813 		return 1;
6814 	}
6815 
6816 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6817 		return 1;
6818 
6819 	/* this is needed for vmware backdoor interface to work since it
6820 	   changes registers values  during IO operation */
6821 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6822 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6823 		emulator_invalidate_register_cache(ctxt);
6824 	}
6825 
6826 restart:
6827 	/* Save the faulting GPA (cr2) in the address field */
6828 	ctxt->exception.address = cr2_or_gpa;
6829 
6830 	r = x86_emulate_insn(ctxt);
6831 
6832 	if (r == EMULATION_INTERCEPTED)
6833 		return 1;
6834 
6835 	if (r == EMULATION_FAILED) {
6836 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6837 					emulation_type))
6838 			return 1;
6839 
6840 		return handle_emulation_failure(vcpu, emulation_type);
6841 	}
6842 
6843 	if (ctxt->have_exception) {
6844 		r = 1;
6845 		if (inject_emulated_exception(vcpu))
6846 			return r;
6847 	} else if (vcpu->arch.pio.count) {
6848 		if (!vcpu->arch.pio.in) {
6849 			/* FIXME: return into emulator if single-stepping.  */
6850 			vcpu->arch.pio.count = 0;
6851 		} else {
6852 			writeback = false;
6853 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6854 		}
6855 		r = 0;
6856 	} else if (vcpu->mmio_needed) {
6857 		++vcpu->stat.mmio_exits;
6858 
6859 		if (!vcpu->mmio_is_write)
6860 			writeback = false;
6861 		r = 0;
6862 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6863 	} else if (r == EMULATION_RESTART)
6864 		goto restart;
6865 	else
6866 		r = 1;
6867 
6868 	if (writeback) {
6869 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6870 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6871 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6872 		if (!ctxt->have_exception ||
6873 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6874 			kvm_rip_write(vcpu, ctxt->eip);
6875 			if (r && ctxt->tf)
6876 				r = kvm_vcpu_do_singlestep(vcpu);
6877 			__kvm_set_rflags(vcpu, ctxt->eflags);
6878 		}
6879 
6880 		/*
6881 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6882 		 * do nothing, and it will be requested again as soon as
6883 		 * the shadow expires.  But we still need to check here,
6884 		 * because POPF has no interrupt shadow.
6885 		 */
6886 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6887 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6888 	} else
6889 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6890 
6891 	return r;
6892 }
6893 
6894 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6895 {
6896 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6897 }
6898 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6899 
6900 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6901 					void *insn, int insn_len)
6902 {
6903 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6904 }
6905 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6906 
6907 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6908 {
6909 	vcpu->arch.pio.count = 0;
6910 	return 1;
6911 }
6912 
6913 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6914 {
6915 	vcpu->arch.pio.count = 0;
6916 
6917 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6918 		return 1;
6919 
6920 	return kvm_skip_emulated_instruction(vcpu);
6921 }
6922 
6923 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6924 			    unsigned short port)
6925 {
6926 	unsigned long val = kvm_rax_read(vcpu);
6927 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6928 					    size, port, &val, 1);
6929 	if (ret)
6930 		return ret;
6931 
6932 	/*
6933 	 * Workaround userspace that relies on old KVM behavior of %rip being
6934 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6935 	 */
6936 	if (port == 0x7e &&
6937 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6938 		vcpu->arch.complete_userspace_io =
6939 			complete_fast_pio_out_port_0x7e;
6940 		kvm_skip_emulated_instruction(vcpu);
6941 	} else {
6942 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6943 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6944 	}
6945 	return 0;
6946 }
6947 
6948 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6949 {
6950 	unsigned long val;
6951 
6952 	/* We should only ever be called with arch.pio.count equal to 1 */
6953 	BUG_ON(vcpu->arch.pio.count != 1);
6954 
6955 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6956 		vcpu->arch.pio.count = 0;
6957 		return 1;
6958 	}
6959 
6960 	/* For size less than 4 we merge, else we zero extend */
6961 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6962 
6963 	/*
6964 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6965 	 * the copy and tracing
6966 	 */
6967 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6968 				 vcpu->arch.pio.port, &val, 1);
6969 	kvm_rax_write(vcpu, val);
6970 
6971 	return kvm_skip_emulated_instruction(vcpu);
6972 }
6973 
6974 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6975 			   unsigned short port)
6976 {
6977 	unsigned long val;
6978 	int ret;
6979 
6980 	/* For size less than 4 we merge, else we zero extend */
6981 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6982 
6983 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6984 				       &val, 1);
6985 	if (ret) {
6986 		kvm_rax_write(vcpu, val);
6987 		return ret;
6988 	}
6989 
6990 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6991 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6992 
6993 	return 0;
6994 }
6995 
6996 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6997 {
6998 	int ret;
6999 
7000 	if (in)
7001 		ret = kvm_fast_pio_in(vcpu, size, port);
7002 	else
7003 		ret = kvm_fast_pio_out(vcpu, size, port);
7004 	return ret && kvm_skip_emulated_instruction(vcpu);
7005 }
7006 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7007 
7008 static int kvmclock_cpu_down_prep(unsigned int cpu)
7009 {
7010 	__this_cpu_write(cpu_tsc_khz, 0);
7011 	return 0;
7012 }
7013 
7014 static void tsc_khz_changed(void *data)
7015 {
7016 	struct cpufreq_freqs *freq = data;
7017 	unsigned long khz = 0;
7018 
7019 	if (data)
7020 		khz = freq->new;
7021 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7022 		khz = cpufreq_quick_get(raw_smp_processor_id());
7023 	if (!khz)
7024 		khz = tsc_khz;
7025 	__this_cpu_write(cpu_tsc_khz, khz);
7026 }
7027 
7028 #ifdef CONFIG_X86_64
7029 static void kvm_hyperv_tsc_notifier(void)
7030 {
7031 	struct kvm *kvm;
7032 	struct kvm_vcpu *vcpu;
7033 	int cpu;
7034 
7035 	mutex_lock(&kvm_lock);
7036 	list_for_each_entry(kvm, &vm_list, vm_list)
7037 		kvm_make_mclock_inprogress_request(kvm);
7038 
7039 	hyperv_stop_tsc_emulation();
7040 
7041 	/* TSC frequency always matches when on Hyper-V */
7042 	for_each_present_cpu(cpu)
7043 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7044 	kvm_max_guest_tsc_khz = tsc_khz;
7045 
7046 	list_for_each_entry(kvm, &vm_list, vm_list) {
7047 		struct kvm_arch *ka = &kvm->arch;
7048 
7049 		spin_lock(&ka->pvclock_gtod_sync_lock);
7050 
7051 		pvclock_update_vm_gtod_copy(kvm);
7052 
7053 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7054 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7055 
7056 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7057 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7058 
7059 		spin_unlock(&ka->pvclock_gtod_sync_lock);
7060 	}
7061 	mutex_unlock(&kvm_lock);
7062 }
7063 #endif
7064 
7065 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7066 {
7067 	struct kvm *kvm;
7068 	struct kvm_vcpu *vcpu;
7069 	int i, send_ipi = 0;
7070 
7071 	/*
7072 	 * We allow guests to temporarily run on slowing clocks,
7073 	 * provided we notify them after, or to run on accelerating
7074 	 * clocks, provided we notify them before.  Thus time never
7075 	 * goes backwards.
7076 	 *
7077 	 * However, we have a problem.  We can't atomically update
7078 	 * the frequency of a given CPU from this function; it is
7079 	 * merely a notifier, which can be called from any CPU.
7080 	 * Changing the TSC frequency at arbitrary points in time
7081 	 * requires a recomputation of local variables related to
7082 	 * the TSC for each VCPU.  We must flag these local variables
7083 	 * to be updated and be sure the update takes place with the
7084 	 * new frequency before any guests proceed.
7085 	 *
7086 	 * Unfortunately, the combination of hotplug CPU and frequency
7087 	 * change creates an intractable locking scenario; the order
7088 	 * of when these callouts happen is undefined with respect to
7089 	 * CPU hotplug, and they can race with each other.  As such,
7090 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7091 	 * undefined; you can actually have a CPU frequency change take
7092 	 * place in between the computation of X and the setting of the
7093 	 * variable.  To protect against this problem, all updates of
7094 	 * the per_cpu tsc_khz variable are done in an interrupt
7095 	 * protected IPI, and all callers wishing to update the value
7096 	 * must wait for a synchronous IPI to complete (which is trivial
7097 	 * if the caller is on the CPU already).  This establishes the
7098 	 * necessary total order on variable updates.
7099 	 *
7100 	 * Note that because a guest time update may take place
7101 	 * anytime after the setting of the VCPU's request bit, the
7102 	 * correct TSC value must be set before the request.  However,
7103 	 * to ensure the update actually makes it to any guest which
7104 	 * starts running in hardware virtualization between the set
7105 	 * and the acquisition of the spinlock, we must also ping the
7106 	 * CPU after setting the request bit.
7107 	 *
7108 	 */
7109 
7110 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7111 
7112 	mutex_lock(&kvm_lock);
7113 	list_for_each_entry(kvm, &vm_list, vm_list) {
7114 		kvm_for_each_vcpu(i, vcpu, kvm) {
7115 			if (vcpu->cpu != cpu)
7116 				continue;
7117 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7118 			if (vcpu->cpu != raw_smp_processor_id())
7119 				send_ipi = 1;
7120 		}
7121 	}
7122 	mutex_unlock(&kvm_lock);
7123 
7124 	if (freq->old < freq->new && send_ipi) {
7125 		/*
7126 		 * We upscale the frequency.  Must make the guest
7127 		 * doesn't see old kvmclock values while running with
7128 		 * the new frequency, otherwise we risk the guest sees
7129 		 * time go backwards.
7130 		 *
7131 		 * In case we update the frequency for another cpu
7132 		 * (which might be in guest context) send an interrupt
7133 		 * to kick the cpu out of guest context.  Next time
7134 		 * guest context is entered kvmclock will be updated,
7135 		 * so the guest will not see stale values.
7136 		 */
7137 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7138 	}
7139 }
7140 
7141 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7142 				     void *data)
7143 {
7144 	struct cpufreq_freqs *freq = data;
7145 	int cpu;
7146 
7147 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7148 		return 0;
7149 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7150 		return 0;
7151 
7152 	for_each_cpu(cpu, freq->policy->cpus)
7153 		__kvmclock_cpufreq_notifier(freq, cpu);
7154 
7155 	return 0;
7156 }
7157 
7158 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7159 	.notifier_call  = kvmclock_cpufreq_notifier
7160 };
7161 
7162 static int kvmclock_cpu_online(unsigned int cpu)
7163 {
7164 	tsc_khz_changed(NULL);
7165 	return 0;
7166 }
7167 
7168 static void kvm_timer_init(void)
7169 {
7170 	max_tsc_khz = tsc_khz;
7171 
7172 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7173 #ifdef CONFIG_CPU_FREQ
7174 		struct cpufreq_policy policy;
7175 		int cpu;
7176 
7177 		memset(&policy, 0, sizeof(policy));
7178 		cpu = get_cpu();
7179 		cpufreq_get_policy(&policy, cpu);
7180 		if (policy.cpuinfo.max_freq)
7181 			max_tsc_khz = policy.cpuinfo.max_freq;
7182 		put_cpu();
7183 #endif
7184 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7185 					  CPUFREQ_TRANSITION_NOTIFIER);
7186 	}
7187 
7188 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7189 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7190 }
7191 
7192 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7193 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7194 
7195 int kvm_is_in_guest(void)
7196 {
7197 	return __this_cpu_read(current_vcpu) != NULL;
7198 }
7199 
7200 static int kvm_is_user_mode(void)
7201 {
7202 	int user_mode = 3;
7203 
7204 	if (__this_cpu_read(current_vcpu))
7205 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7206 
7207 	return user_mode != 0;
7208 }
7209 
7210 static unsigned long kvm_get_guest_ip(void)
7211 {
7212 	unsigned long ip = 0;
7213 
7214 	if (__this_cpu_read(current_vcpu))
7215 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7216 
7217 	return ip;
7218 }
7219 
7220 static void kvm_handle_intel_pt_intr(void)
7221 {
7222 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7223 
7224 	kvm_make_request(KVM_REQ_PMI, vcpu);
7225 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7226 			(unsigned long *)&vcpu->arch.pmu.global_status);
7227 }
7228 
7229 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7230 	.is_in_guest		= kvm_is_in_guest,
7231 	.is_user_mode		= kvm_is_user_mode,
7232 	.get_guest_ip		= kvm_get_guest_ip,
7233 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7234 };
7235 
7236 #ifdef CONFIG_X86_64
7237 static void pvclock_gtod_update_fn(struct work_struct *work)
7238 {
7239 	struct kvm *kvm;
7240 
7241 	struct kvm_vcpu *vcpu;
7242 	int i;
7243 
7244 	mutex_lock(&kvm_lock);
7245 	list_for_each_entry(kvm, &vm_list, vm_list)
7246 		kvm_for_each_vcpu(i, vcpu, kvm)
7247 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7248 	atomic_set(&kvm_guest_has_master_clock, 0);
7249 	mutex_unlock(&kvm_lock);
7250 }
7251 
7252 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7253 
7254 /*
7255  * Notification about pvclock gtod data update.
7256  */
7257 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7258 			       void *priv)
7259 {
7260 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7261 	struct timekeeper *tk = priv;
7262 
7263 	update_pvclock_gtod(tk);
7264 
7265 	/* disable master clock if host does not trust, or does not
7266 	 * use, TSC based clocksource.
7267 	 */
7268 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7269 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7270 		queue_work(system_long_wq, &pvclock_gtod_work);
7271 
7272 	return 0;
7273 }
7274 
7275 static struct notifier_block pvclock_gtod_notifier = {
7276 	.notifier_call = pvclock_gtod_notify,
7277 };
7278 #endif
7279 
7280 int kvm_arch_init(void *opaque)
7281 {
7282 	int r;
7283 	struct kvm_x86_ops *ops = opaque;
7284 
7285 	if (kvm_x86_ops) {
7286 		printk(KERN_ERR "kvm: already loaded the other module\n");
7287 		r = -EEXIST;
7288 		goto out;
7289 	}
7290 
7291 	if (!ops->cpu_has_kvm_support()) {
7292 		printk(KERN_ERR "kvm: no hardware support\n");
7293 		r = -EOPNOTSUPP;
7294 		goto out;
7295 	}
7296 	if (ops->disabled_by_bios()) {
7297 		printk(KERN_ERR "kvm: disabled by bios\n");
7298 		r = -EOPNOTSUPP;
7299 		goto out;
7300 	}
7301 
7302 	/*
7303 	 * KVM explicitly assumes that the guest has an FPU and
7304 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7305 	 * vCPU's FPU state as a fxregs_state struct.
7306 	 */
7307 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7308 		printk(KERN_ERR "kvm: inadequate fpu\n");
7309 		r = -EOPNOTSUPP;
7310 		goto out;
7311 	}
7312 
7313 	r = -ENOMEM;
7314 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7315 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7316 					  NULL);
7317 	if (!x86_fpu_cache) {
7318 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7319 		goto out;
7320 	}
7321 
7322 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7323 	if (!shared_msrs) {
7324 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7325 		goto out_free_x86_fpu_cache;
7326 	}
7327 
7328 	r = kvm_mmu_module_init();
7329 	if (r)
7330 		goto out_free_percpu;
7331 
7332 	kvm_x86_ops = ops;
7333 
7334 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7335 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7336 			PT_PRESENT_MASK, 0, sme_me_mask);
7337 	kvm_timer_init();
7338 
7339 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7340 
7341 	if (boot_cpu_has(X86_FEATURE_XSAVE))
7342 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7343 
7344 	kvm_lapic_init();
7345 	if (pi_inject_timer == -1)
7346 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7347 #ifdef CONFIG_X86_64
7348 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7349 
7350 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7351 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7352 #endif
7353 
7354 	return 0;
7355 
7356 out_free_percpu:
7357 	free_percpu(shared_msrs);
7358 out_free_x86_fpu_cache:
7359 	kmem_cache_destroy(x86_fpu_cache);
7360 out:
7361 	return r;
7362 }
7363 
7364 void kvm_arch_exit(void)
7365 {
7366 #ifdef CONFIG_X86_64
7367 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7368 		clear_hv_tscchange_cb();
7369 #endif
7370 	kvm_lapic_exit();
7371 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7372 
7373 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7374 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7375 					    CPUFREQ_TRANSITION_NOTIFIER);
7376 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7377 #ifdef CONFIG_X86_64
7378 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7379 #endif
7380 	kvm_x86_ops = NULL;
7381 	kvm_mmu_module_exit();
7382 	free_percpu(shared_msrs);
7383 	kmem_cache_destroy(x86_fpu_cache);
7384 }
7385 
7386 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7387 {
7388 	++vcpu->stat.halt_exits;
7389 	if (lapic_in_kernel(vcpu)) {
7390 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7391 		return 1;
7392 	} else {
7393 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7394 		return 0;
7395 	}
7396 }
7397 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7398 
7399 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7400 {
7401 	int ret = kvm_skip_emulated_instruction(vcpu);
7402 	/*
7403 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7404 	 * KVM_EXIT_DEBUG here.
7405 	 */
7406 	return kvm_vcpu_halt(vcpu) && ret;
7407 }
7408 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7409 
7410 #ifdef CONFIG_X86_64
7411 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7412 			        unsigned long clock_type)
7413 {
7414 	struct kvm_clock_pairing clock_pairing;
7415 	struct timespec64 ts;
7416 	u64 cycle;
7417 	int ret;
7418 
7419 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7420 		return -KVM_EOPNOTSUPP;
7421 
7422 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7423 		return -KVM_EOPNOTSUPP;
7424 
7425 	clock_pairing.sec = ts.tv_sec;
7426 	clock_pairing.nsec = ts.tv_nsec;
7427 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7428 	clock_pairing.flags = 0;
7429 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7430 
7431 	ret = 0;
7432 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7433 			    sizeof(struct kvm_clock_pairing)))
7434 		ret = -KVM_EFAULT;
7435 
7436 	return ret;
7437 }
7438 #endif
7439 
7440 /*
7441  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7442  *
7443  * @apicid - apicid of vcpu to be kicked.
7444  */
7445 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7446 {
7447 	struct kvm_lapic_irq lapic_irq;
7448 
7449 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
7450 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7451 	lapic_irq.level = 0;
7452 	lapic_irq.dest_id = apicid;
7453 	lapic_irq.msi_redir_hint = false;
7454 
7455 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7456 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7457 }
7458 
7459 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7460 {
7461 	if (!lapic_in_kernel(vcpu)) {
7462 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7463 		return;
7464 	}
7465 	if (!vcpu->arch.apicv_active)
7466 		return;
7467 
7468 	vcpu->arch.apicv_active = false;
7469 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7470 }
7471 
7472 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7473 {
7474 	struct kvm_vcpu *target = NULL;
7475 	struct kvm_apic_map *map;
7476 
7477 	rcu_read_lock();
7478 	map = rcu_dereference(kvm->arch.apic_map);
7479 
7480 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7481 		target = map->phys_map[dest_id]->vcpu;
7482 
7483 	rcu_read_unlock();
7484 
7485 	if (target && READ_ONCE(target->ready))
7486 		kvm_vcpu_yield_to(target);
7487 }
7488 
7489 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7490 {
7491 	unsigned long nr, a0, a1, a2, a3, ret;
7492 	int op_64_bit;
7493 
7494 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7495 		return kvm_hv_hypercall(vcpu);
7496 
7497 	nr = kvm_rax_read(vcpu);
7498 	a0 = kvm_rbx_read(vcpu);
7499 	a1 = kvm_rcx_read(vcpu);
7500 	a2 = kvm_rdx_read(vcpu);
7501 	a3 = kvm_rsi_read(vcpu);
7502 
7503 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7504 
7505 	op_64_bit = is_64_bit_mode(vcpu);
7506 	if (!op_64_bit) {
7507 		nr &= 0xFFFFFFFF;
7508 		a0 &= 0xFFFFFFFF;
7509 		a1 &= 0xFFFFFFFF;
7510 		a2 &= 0xFFFFFFFF;
7511 		a3 &= 0xFFFFFFFF;
7512 	}
7513 
7514 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7515 		ret = -KVM_EPERM;
7516 		goto out;
7517 	}
7518 
7519 	switch (nr) {
7520 	case KVM_HC_VAPIC_POLL_IRQ:
7521 		ret = 0;
7522 		break;
7523 	case KVM_HC_KICK_CPU:
7524 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7525 		kvm_sched_yield(vcpu->kvm, a1);
7526 		ret = 0;
7527 		break;
7528 #ifdef CONFIG_X86_64
7529 	case KVM_HC_CLOCK_PAIRING:
7530 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7531 		break;
7532 #endif
7533 	case KVM_HC_SEND_IPI:
7534 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7535 		break;
7536 	case KVM_HC_SCHED_YIELD:
7537 		kvm_sched_yield(vcpu->kvm, a0);
7538 		ret = 0;
7539 		break;
7540 	default:
7541 		ret = -KVM_ENOSYS;
7542 		break;
7543 	}
7544 out:
7545 	if (!op_64_bit)
7546 		ret = (u32)ret;
7547 	kvm_rax_write(vcpu, ret);
7548 
7549 	++vcpu->stat.hypercalls;
7550 	return kvm_skip_emulated_instruction(vcpu);
7551 }
7552 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7553 
7554 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7555 {
7556 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7557 	char instruction[3];
7558 	unsigned long rip = kvm_rip_read(vcpu);
7559 
7560 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7561 
7562 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7563 		&ctxt->exception);
7564 }
7565 
7566 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7567 {
7568 	return vcpu->run->request_interrupt_window &&
7569 		likely(!pic_in_kernel(vcpu->kvm));
7570 }
7571 
7572 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7573 {
7574 	struct kvm_run *kvm_run = vcpu->run;
7575 
7576 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7577 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7578 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7579 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7580 	kvm_run->ready_for_interrupt_injection =
7581 		pic_in_kernel(vcpu->kvm) ||
7582 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7583 }
7584 
7585 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7586 {
7587 	int max_irr, tpr;
7588 
7589 	if (!kvm_x86_ops->update_cr8_intercept)
7590 		return;
7591 
7592 	if (!lapic_in_kernel(vcpu))
7593 		return;
7594 
7595 	if (vcpu->arch.apicv_active)
7596 		return;
7597 
7598 	if (!vcpu->arch.apic->vapic_addr)
7599 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7600 	else
7601 		max_irr = -1;
7602 
7603 	if (max_irr != -1)
7604 		max_irr >>= 4;
7605 
7606 	tpr = kvm_lapic_get_cr8(vcpu);
7607 
7608 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7609 }
7610 
7611 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7612 {
7613 	int r;
7614 
7615 	/* try to reinject previous events if any */
7616 
7617 	if (vcpu->arch.exception.injected)
7618 		kvm_x86_ops->queue_exception(vcpu);
7619 	/*
7620 	 * Do not inject an NMI or interrupt if there is a pending
7621 	 * exception.  Exceptions and interrupts are recognized at
7622 	 * instruction boundaries, i.e. the start of an instruction.
7623 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7624 	 * NMIs and interrupts, i.e. traps are recognized before an
7625 	 * NMI/interrupt that's pending on the same instruction.
7626 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7627 	 * priority, but are only generated (pended) during instruction
7628 	 * execution, i.e. a pending fault-like exception means the
7629 	 * fault occurred on the *previous* instruction and must be
7630 	 * serviced prior to recognizing any new events in order to
7631 	 * fully complete the previous instruction.
7632 	 */
7633 	else if (!vcpu->arch.exception.pending) {
7634 		if (vcpu->arch.nmi_injected)
7635 			kvm_x86_ops->set_nmi(vcpu);
7636 		else if (vcpu->arch.interrupt.injected)
7637 			kvm_x86_ops->set_irq(vcpu);
7638 	}
7639 
7640 	/*
7641 	 * Call check_nested_events() even if we reinjected a previous event
7642 	 * in order for caller to determine if it should require immediate-exit
7643 	 * from L2 to L1 due to pending L1 events which require exit
7644 	 * from L2 to L1.
7645 	 */
7646 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7647 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7648 		if (r != 0)
7649 			return r;
7650 	}
7651 
7652 	/* try to inject new event if pending */
7653 	if (vcpu->arch.exception.pending) {
7654 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7655 					vcpu->arch.exception.has_error_code,
7656 					vcpu->arch.exception.error_code);
7657 
7658 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7659 		vcpu->arch.exception.pending = false;
7660 		vcpu->arch.exception.injected = true;
7661 
7662 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7663 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7664 					     X86_EFLAGS_RF);
7665 
7666 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7667 			/*
7668 			 * This code assumes that nSVM doesn't use
7669 			 * check_nested_events(). If it does, the
7670 			 * DR6/DR7 changes should happen before L1
7671 			 * gets a #VMEXIT for an intercepted #DB in
7672 			 * L2.  (Under VMX, on the other hand, the
7673 			 * DR6/DR7 changes should not happen in the
7674 			 * event of a VM-exit to L1 for an intercepted
7675 			 * #DB in L2.)
7676 			 */
7677 			kvm_deliver_exception_payload(vcpu);
7678 			if (vcpu->arch.dr7 & DR7_GD) {
7679 				vcpu->arch.dr7 &= ~DR7_GD;
7680 				kvm_update_dr7(vcpu);
7681 			}
7682 		}
7683 
7684 		kvm_x86_ops->queue_exception(vcpu);
7685 	}
7686 
7687 	/* Don't consider new event if we re-injected an event */
7688 	if (kvm_event_needs_reinjection(vcpu))
7689 		return 0;
7690 
7691 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7692 	    kvm_x86_ops->smi_allowed(vcpu)) {
7693 		vcpu->arch.smi_pending = false;
7694 		++vcpu->arch.smi_count;
7695 		enter_smm(vcpu);
7696 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7697 		--vcpu->arch.nmi_pending;
7698 		vcpu->arch.nmi_injected = true;
7699 		kvm_x86_ops->set_nmi(vcpu);
7700 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7701 		/*
7702 		 * Because interrupts can be injected asynchronously, we are
7703 		 * calling check_nested_events again here to avoid a race condition.
7704 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7705 		 * proposal and current concerns.  Perhaps we should be setting
7706 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7707 		 */
7708 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7709 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7710 			if (r != 0)
7711 				return r;
7712 		}
7713 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7714 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7715 					    false);
7716 			kvm_x86_ops->set_irq(vcpu);
7717 		}
7718 	}
7719 
7720 	return 0;
7721 }
7722 
7723 static void process_nmi(struct kvm_vcpu *vcpu)
7724 {
7725 	unsigned limit = 2;
7726 
7727 	/*
7728 	 * x86 is limited to one NMI running, and one NMI pending after it.
7729 	 * If an NMI is already in progress, limit further NMIs to just one.
7730 	 * Otherwise, allow two (and we'll inject the first one immediately).
7731 	 */
7732 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7733 		limit = 1;
7734 
7735 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7736 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7737 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7738 }
7739 
7740 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7741 {
7742 	u32 flags = 0;
7743 	flags |= seg->g       << 23;
7744 	flags |= seg->db      << 22;
7745 	flags |= seg->l       << 21;
7746 	flags |= seg->avl     << 20;
7747 	flags |= seg->present << 15;
7748 	flags |= seg->dpl     << 13;
7749 	flags |= seg->s       << 12;
7750 	flags |= seg->type    << 8;
7751 	return flags;
7752 }
7753 
7754 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7755 {
7756 	struct kvm_segment seg;
7757 	int offset;
7758 
7759 	kvm_get_segment(vcpu, &seg, n);
7760 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7761 
7762 	if (n < 3)
7763 		offset = 0x7f84 + n * 12;
7764 	else
7765 		offset = 0x7f2c + (n - 3) * 12;
7766 
7767 	put_smstate(u32, buf, offset + 8, seg.base);
7768 	put_smstate(u32, buf, offset + 4, seg.limit);
7769 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7770 }
7771 
7772 #ifdef CONFIG_X86_64
7773 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7774 {
7775 	struct kvm_segment seg;
7776 	int offset;
7777 	u16 flags;
7778 
7779 	kvm_get_segment(vcpu, &seg, n);
7780 	offset = 0x7e00 + n * 16;
7781 
7782 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7783 	put_smstate(u16, buf, offset, seg.selector);
7784 	put_smstate(u16, buf, offset + 2, flags);
7785 	put_smstate(u32, buf, offset + 4, seg.limit);
7786 	put_smstate(u64, buf, offset + 8, seg.base);
7787 }
7788 #endif
7789 
7790 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7791 {
7792 	struct desc_ptr dt;
7793 	struct kvm_segment seg;
7794 	unsigned long val;
7795 	int i;
7796 
7797 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7798 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7799 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7800 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7801 
7802 	for (i = 0; i < 8; i++)
7803 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7804 
7805 	kvm_get_dr(vcpu, 6, &val);
7806 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7807 	kvm_get_dr(vcpu, 7, &val);
7808 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7809 
7810 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7811 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7812 	put_smstate(u32, buf, 0x7f64, seg.base);
7813 	put_smstate(u32, buf, 0x7f60, seg.limit);
7814 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7815 
7816 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7817 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7818 	put_smstate(u32, buf, 0x7f80, seg.base);
7819 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7820 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7821 
7822 	kvm_x86_ops->get_gdt(vcpu, &dt);
7823 	put_smstate(u32, buf, 0x7f74, dt.address);
7824 	put_smstate(u32, buf, 0x7f70, dt.size);
7825 
7826 	kvm_x86_ops->get_idt(vcpu, &dt);
7827 	put_smstate(u32, buf, 0x7f58, dt.address);
7828 	put_smstate(u32, buf, 0x7f54, dt.size);
7829 
7830 	for (i = 0; i < 6; i++)
7831 		enter_smm_save_seg_32(vcpu, buf, i);
7832 
7833 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7834 
7835 	/* revision id */
7836 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7837 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7838 }
7839 
7840 #ifdef CONFIG_X86_64
7841 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7842 {
7843 	struct desc_ptr dt;
7844 	struct kvm_segment seg;
7845 	unsigned long val;
7846 	int i;
7847 
7848 	for (i = 0; i < 16; i++)
7849 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7850 
7851 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7852 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7853 
7854 	kvm_get_dr(vcpu, 6, &val);
7855 	put_smstate(u64, buf, 0x7f68, val);
7856 	kvm_get_dr(vcpu, 7, &val);
7857 	put_smstate(u64, buf, 0x7f60, val);
7858 
7859 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7860 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7861 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7862 
7863 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7864 
7865 	/* revision id */
7866 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7867 
7868 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7869 
7870 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7871 	put_smstate(u16, buf, 0x7e90, seg.selector);
7872 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7873 	put_smstate(u32, buf, 0x7e94, seg.limit);
7874 	put_smstate(u64, buf, 0x7e98, seg.base);
7875 
7876 	kvm_x86_ops->get_idt(vcpu, &dt);
7877 	put_smstate(u32, buf, 0x7e84, dt.size);
7878 	put_smstate(u64, buf, 0x7e88, dt.address);
7879 
7880 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7881 	put_smstate(u16, buf, 0x7e70, seg.selector);
7882 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7883 	put_smstate(u32, buf, 0x7e74, seg.limit);
7884 	put_smstate(u64, buf, 0x7e78, seg.base);
7885 
7886 	kvm_x86_ops->get_gdt(vcpu, &dt);
7887 	put_smstate(u32, buf, 0x7e64, dt.size);
7888 	put_smstate(u64, buf, 0x7e68, dt.address);
7889 
7890 	for (i = 0; i < 6; i++)
7891 		enter_smm_save_seg_64(vcpu, buf, i);
7892 }
7893 #endif
7894 
7895 static void enter_smm(struct kvm_vcpu *vcpu)
7896 {
7897 	struct kvm_segment cs, ds;
7898 	struct desc_ptr dt;
7899 	char buf[512];
7900 	u32 cr0;
7901 
7902 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7903 	memset(buf, 0, 512);
7904 #ifdef CONFIG_X86_64
7905 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7906 		enter_smm_save_state_64(vcpu, buf);
7907 	else
7908 #endif
7909 		enter_smm_save_state_32(vcpu, buf);
7910 
7911 	/*
7912 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7913 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7914 	 * the SMM state-save area.
7915 	 */
7916 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7917 
7918 	vcpu->arch.hflags |= HF_SMM_MASK;
7919 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7920 
7921 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7922 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7923 	else
7924 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7925 
7926 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7927 	kvm_rip_write(vcpu, 0x8000);
7928 
7929 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7930 	kvm_x86_ops->set_cr0(vcpu, cr0);
7931 	vcpu->arch.cr0 = cr0;
7932 
7933 	kvm_x86_ops->set_cr4(vcpu, 0);
7934 
7935 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7936 	dt.address = dt.size = 0;
7937 	kvm_x86_ops->set_idt(vcpu, &dt);
7938 
7939 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7940 
7941 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7942 	cs.base = vcpu->arch.smbase;
7943 
7944 	ds.selector = 0;
7945 	ds.base = 0;
7946 
7947 	cs.limit    = ds.limit = 0xffffffff;
7948 	cs.type     = ds.type = 0x3;
7949 	cs.dpl      = ds.dpl = 0;
7950 	cs.db       = ds.db = 0;
7951 	cs.s        = ds.s = 1;
7952 	cs.l        = ds.l = 0;
7953 	cs.g        = ds.g = 1;
7954 	cs.avl      = ds.avl = 0;
7955 	cs.present  = ds.present = 1;
7956 	cs.unusable = ds.unusable = 0;
7957 	cs.padding  = ds.padding = 0;
7958 
7959 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7960 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7961 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7962 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7963 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7964 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7965 
7966 #ifdef CONFIG_X86_64
7967 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7968 		kvm_x86_ops->set_efer(vcpu, 0);
7969 #endif
7970 
7971 	kvm_update_cpuid(vcpu);
7972 	kvm_mmu_reset_context(vcpu);
7973 }
7974 
7975 static void process_smi(struct kvm_vcpu *vcpu)
7976 {
7977 	vcpu->arch.smi_pending = true;
7978 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7979 }
7980 
7981 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
7982 				       unsigned long *vcpu_bitmap)
7983 {
7984 	cpumask_var_t cpus;
7985 
7986 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
7987 
7988 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
7989 				    vcpu_bitmap, cpus);
7990 
7991 	free_cpumask_var(cpus);
7992 }
7993 
7994 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7995 {
7996 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7997 }
7998 
7999 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8000 {
8001 	if (!kvm_apic_present(vcpu))
8002 		return;
8003 
8004 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8005 
8006 	if (irqchip_split(vcpu->kvm))
8007 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8008 	else {
8009 		if (vcpu->arch.apicv_active)
8010 			kvm_x86_ops->sync_pir_to_irr(vcpu);
8011 		if (ioapic_in_kernel(vcpu->kvm))
8012 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8013 	}
8014 
8015 	if (is_guest_mode(vcpu))
8016 		vcpu->arch.load_eoi_exitmap_pending = true;
8017 	else
8018 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8019 }
8020 
8021 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8022 {
8023 	u64 eoi_exit_bitmap[4];
8024 
8025 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8026 		return;
8027 
8028 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8029 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
8030 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8031 }
8032 
8033 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8034 		unsigned long start, unsigned long end,
8035 		bool blockable)
8036 {
8037 	unsigned long apic_address;
8038 
8039 	/*
8040 	 * The physical address of apic access page is stored in the VMCS.
8041 	 * Update it when it becomes invalid.
8042 	 */
8043 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8044 	if (start <= apic_address && apic_address < end)
8045 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8046 
8047 	return 0;
8048 }
8049 
8050 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8051 {
8052 	struct page *page = NULL;
8053 
8054 	if (!lapic_in_kernel(vcpu))
8055 		return;
8056 
8057 	if (!kvm_x86_ops->set_apic_access_page_addr)
8058 		return;
8059 
8060 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8061 	if (is_error_page(page))
8062 		return;
8063 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
8064 
8065 	/*
8066 	 * Do not pin apic access page in memory, the MMU notifier
8067 	 * will call us again if it is migrated or swapped out.
8068 	 */
8069 	put_page(page);
8070 }
8071 
8072 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8073 {
8074 	smp_send_reschedule(vcpu->cpu);
8075 }
8076 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8077 
8078 /*
8079  * Returns 1 to let vcpu_run() continue the guest execution loop without
8080  * exiting to the userspace.  Otherwise, the value will be returned to the
8081  * userspace.
8082  */
8083 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8084 {
8085 	int r;
8086 	bool req_int_win =
8087 		dm_request_for_irq_injection(vcpu) &&
8088 		kvm_cpu_accept_dm_intr(vcpu);
8089 	enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8090 
8091 	bool req_immediate_exit = false;
8092 
8093 	if (kvm_request_pending(vcpu)) {
8094 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8095 			if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
8096 				r = 0;
8097 				goto out;
8098 			}
8099 		}
8100 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8101 			kvm_mmu_unload(vcpu);
8102 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8103 			__kvm_migrate_timers(vcpu);
8104 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8105 			kvm_gen_update_masterclock(vcpu->kvm);
8106 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8107 			kvm_gen_kvmclock_update(vcpu);
8108 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8109 			r = kvm_guest_time_update(vcpu);
8110 			if (unlikely(r))
8111 				goto out;
8112 		}
8113 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8114 			kvm_mmu_sync_roots(vcpu);
8115 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
8116 			kvm_mmu_load_cr3(vcpu);
8117 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8118 			kvm_vcpu_flush_tlb(vcpu, true);
8119 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8120 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8121 			r = 0;
8122 			goto out;
8123 		}
8124 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8125 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8126 			vcpu->mmio_needed = 0;
8127 			r = 0;
8128 			goto out;
8129 		}
8130 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8131 			/* Page is swapped out. Do synthetic halt */
8132 			vcpu->arch.apf.halted = true;
8133 			r = 1;
8134 			goto out;
8135 		}
8136 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8137 			record_steal_time(vcpu);
8138 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8139 			process_smi(vcpu);
8140 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8141 			process_nmi(vcpu);
8142 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8143 			kvm_pmu_handle_event(vcpu);
8144 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8145 			kvm_pmu_deliver_pmi(vcpu);
8146 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8147 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8148 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8149 				     vcpu->arch.ioapic_handled_vectors)) {
8150 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8151 				vcpu->run->eoi.vector =
8152 						vcpu->arch.pending_ioapic_eoi;
8153 				r = 0;
8154 				goto out;
8155 			}
8156 		}
8157 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8158 			vcpu_scan_ioapic(vcpu);
8159 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8160 			vcpu_load_eoi_exitmap(vcpu);
8161 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8162 			kvm_vcpu_reload_apic_access_page(vcpu);
8163 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8164 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8165 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8166 			r = 0;
8167 			goto out;
8168 		}
8169 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8170 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8171 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8172 			r = 0;
8173 			goto out;
8174 		}
8175 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8176 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8177 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8178 			r = 0;
8179 			goto out;
8180 		}
8181 
8182 		/*
8183 		 * KVM_REQ_HV_STIMER has to be processed after
8184 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8185 		 * depend on the guest clock being up-to-date
8186 		 */
8187 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8188 			kvm_hv_process_stimers(vcpu);
8189 	}
8190 
8191 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8192 		++vcpu->stat.req_event;
8193 		kvm_apic_accept_events(vcpu);
8194 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8195 			r = 1;
8196 			goto out;
8197 		}
8198 
8199 		if (inject_pending_event(vcpu, req_int_win) != 0)
8200 			req_immediate_exit = true;
8201 		else {
8202 			/* Enable SMI/NMI/IRQ window open exits if needed.
8203 			 *
8204 			 * SMIs have three cases:
8205 			 * 1) They can be nested, and then there is nothing to
8206 			 *    do here because RSM will cause a vmexit anyway.
8207 			 * 2) There is an ISA-specific reason why SMI cannot be
8208 			 *    injected, and the moment when this changes can be
8209 			 *    intercepted.
8210 			 * 3) Or the SMI can be pending because
8211 			 *    inject_pending_event has completed the injection
8212 			 *    of an IRQ or NMI from the previous vmexit, and
8213 			 *    then we request an immediate exit to inject the
8214 			 *    SMI.
8215 			 */
8216 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
8217 				if (!kvm_x86_ops->enable_smi_window(vcpu))
8218 					req_immediate_exit = true;
8219 			if (vcpu->arch.nmi_pending)
8220 				kvm_x86_ops->enable_nmi_window(vcpu);
8221 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8222 				kvm_x86_ops->enable_irq_window(vcpu);
8223 			WARN_ON(vcpu->arch.exception.pending);
8224 		}
8225 
8226 		if (kvm_lapic_enabled(vcpu)) {
8227 			update_cr8_intercept(vcpu);
8228 			kvm_lapic_sync_to_vapic(vcpu);
8229 		}
8230 	}
8231 
8232 	r = kvm_mmu_reload(vcpu);
8233 	if (unlikely(r)) {
8234 		goto cancel_injection;
8235 	}
8236 
8237 	preempt_disable();
8238 
8239 	kvm_x86_ops->prepare_guest_switch(vcpu);
8240 
8241 	/*
8242 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8243 	 * IPI are then delayed after guest entry, which ensures that they
8244 	 * result in virtual interrupt delivery.
8245 	 */
8246 	local_irq_disable();
8247 	vcpu->mode = IN_GUEST_MODE;
8248 
8249 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8250 
8251 	/*
8252 	 * 1) We should set ->mode before checking ->requests.  Please see
8253 	 * the comment in kvm_vcpu_exiting_guest_mode().
8254 	 *
8255 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8256 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8257 	 * (see vmx_deliver_posted_interrupt).
8258 	 *
8259 	 * 3) This also orders the write to mode from any reads to the page
8260 	 * tables done while the VCPU is running.  Please see the comment
8261 	 * in kvm_flush_remote_tlbs.
8262 	 */
8263 	smp_mb__after_srcu_read_unlock();
8264 
8265 	/*
8266 	 * This handles the case where a posted interrupt was
8267 	 * notified with kvm_vcpu_kick.
8268 	 */
8269 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8270 		kvm_x86_ops->sync_pir_to_irr(vcpu);
8271 
8272 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8273 	    || need_resched() || signal_pending(current)) {
8274 		vcpu->mode = OUTSIDE_GUEST_MODE;
8275 		smp_wmb();
8276 		local_irq_enable();
8277 		preempt_enable();
8278 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8279 		r = 1;
8280 		goto cancel_injection;
8281 	}
8282 
8283 	if (req_immediate_exit) {
8284 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8285 		kvm_x86_ops->request_immediate_exit(vcpu);
8286 	}
8287 
8288 	trace_kvm_entry(vcpu->vcpu_id);
8289 	guest_enter_irqoff();
8290 
8291 	fpregs_assert_state_consistent();
8292 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8293 		switch_fpu_return();
8294 
8295 	if (unlikely(vcpu->arch.switch_db_regs)) {
8296 		set_debugreg(0, 7);
8297 		set_debugreg(vcpu->arch.eff_db[0], 0);
8298 		set_debugreg(vcpu->arch.eff_db[1], 1);
8299 		set_debugreg(vcpu->arch.eff_db[2], 2);
8300 		set_debugreg(vcpu->arch.eff_db[3], 3);
8301 		set_debugreg(vcpu->arch.dr6, 6);
8302 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8303 	}
8304 
8305 	kvm_x86_ops->run(vcpu);
8306 
8307 	/*
8308 	 * Do this here before restoring debug registers on the host.  And
8309 	 * since we do this before handling the vmexit, a DR access vmexit
8310 	 * can (a) read the correct value of the debug registers, (b) set
8311 	 * KVM_DEBUGREG_WONT_EXIT again.
8312 	 */
8313 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8314 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8315 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8316 		kvm_update_dr0123(vcpu);
8317 		kvm_update_dr6(vcpu);
8318 		kvm_update_dr7(vcpu);
8319 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8320 	}
8321 
8322 	/*
8323 	 * If the guest has used debug registers, at least dr7
8324 	 * will be disabled while returning to the host.
8325 	 * If we don't have active breakpoints in the host, we don't
8326 	 * care about the messed up debug address registers. But if
8327 	 * we have some of them active, restore the old state.
8328 	 */
8329 	if (hw_breakpoint_active())
8330 		hw_breakpoint_restore();
8331 
8332 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8333 
8334 	vcpu->mode = OUTSIDE_GUEST_MODE;
8335 	smp_wmb();
8336 
8337 	kvm_x86_ops->handle_exit_irqoff(vcpu, &exit_fastpath);
8338 
8339 	/*
8340 	 * Consume any pending interrupts, including the possible source of
8341 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8342 	 * An instruction is required after local_irq_enable() to fully unblock
8343 	 * interrupts on processors that implement an interrupt shadow, the
8344 	 * stat.exits increment will do nicely.
8345 	 */
8346 	kvm_before_interrupt(vcpu);
8347 	local_irq_enable();
8348 	++vcpu->stat.exits;
8349 	local_irq_disable();
8350 	kvm_after_interrupt(vcpu);
8351 
8352 	guest_exit_irqoff();
8353 	if (lapic_in_kernel(vcpu)) {
8354 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8355 		if (delta != S64_MIN) {
8356 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8357 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8358 		}
8359 	}
8360 
8361 	local_irq_enable();
8362 	preempt_enable();
8363 
8364 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8365 
8366 	/*
8367 	 * Profile KVM exit RIPs:
8368 	 */
8369 	if (unlikely(prof_on == KVM_PROFILING)) {
8370 		unsigned long rip = kvm_rip_read(vcpu);
8371 		profile_hit(KVM_PROFILING, (void *)rip);
8372 	}
8373 
8374 	if (unlikely(vcpu->arch.tsc_always_catchup))
8375 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8376 
8377 	if (vcpu->arch.apic_attention)
8378 		kvm_lapic_sync_from_vapic(vcpu);
8379 
8380 	vcpu->arch.gpa_available = false;
8381 	r = kvm_x86_ops->handle_exit(vcpu, exit_fastpath);
8382 	return r;
8383 
8384 cancel_injection:
8385 	kvm_x86_ops->cancel_injection(vcpu);
8386 	if (unlikely(vcpu->arch.apic_attention))
8387 		kvm_lapic_sync_from_vapic(vcpu);
8388 out:
8389 	return r;
8390 }
8391 
8392 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8393 {
8394 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8395 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8396 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8397 		kvm_vcpu_block(vcpu);
8398 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8399 
8400 		if (kvm_x86_ops->post_block)
8401 			kvm_x86_ops->post_block(vcpu);
8402 
8403 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8404 			return 1;
8405 	}
8406 
8407 	kvm_apic_accept_events(vcpu);
8408 	switch(vcpu->arch.mp_state) {
8409 	case KVM_MP_STATE_HALTED:
8410 		vcpu->arch.pv.pv_unhalted = false;
8411 		vcpu->arch.mp_state =
8412 			KVM_MP_STATE_RUNNABLE;
8413 		/* fall through */
8414 	case KVM_MP_STATE_RUNNABLE:
8415 		vcpu->arch.apf.halted = false;
8416 		break;
8417 	case KVM_MP_STATE_INIT_RECEIVED:
8418 		break;
8419 	default:
8420 		return -EINTR;
8421 		break;
8422 	}
8423 	return 1;
8424 }
8425 
8426 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8427 {
8428 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8429 		kvm_x86_ops->check_nested_events(vcpu, false);
8430 
8431 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8432 		!vcpu->arch.apf.halted);
8433 }
8434 
8435 static int vcpu_run(struct kvm_vcpu *vcpu)
8436 {
8437 	int r;
8438 	struct kvm *kvm = vcpu->kvm;
8439 
8440 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8441 	vcpu->arch.l1tf_flush_l1d = true;
8442 
8443 	for (;;) {
8444 		if (kvm_vcpu_running(vcpu)) {
8445 			r = vcpu_enter_guest(vcpu);
8446 		} else {
8447 			r = vcpu_block(kvm, vcpu);
8448 		}
8449 
8450 		if (r <= 0)
8451 			break;
8452 
8453 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8454 		if (kvm_cpu_has_pending_timer(vcpu))
8455 			kvm_inject_pending_timer_irqs(vcpu);
8456 
8457 		if (dm_request_for_irq_injection(vcpu) &&
8458 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8459 			r = 0;
8460 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8461 			++vcpu->stat.request_irq_exits;
8462 			break;
8463 		}
8464 
8465 		kvm_check_async_pf_completion(vcpu);
8466 
8467 		if (signal_pending(current)) {
8468 			r = -EINTR;
8469 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8470 			++vcpu->stat.signal_exits;
8471 			break;
8472 		}
8473 		if (need_resched()) {
8474 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8475 			cond_resched();
8476 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8477 		}
8478 	}
8479 
8480 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8481 
8482 	return r;
8483 }
8484 
8485 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8486 {
8487 	int r;
8488 
8489 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8490 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8491 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8492 	return r;
8493 }
8494 
8495 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8496 {
8497 	BUG_ON(!vcpu->arch.pio.count);
8498 
8499 	return complete_emulated_io(vcpu);
8500 }
8501 
8502 /*
8503  * Implements the following, as a state machine:
8504  *
8505  * read:
8506  *   for each fragment
8507  *     for each mmio piece in the fragment
8508  *       write gpa, len
8509  *       exit
8510  *       copy data
8511  *   execute insn
8512  *
8513  * write:
8514  *   for each fragment
8515  *     for each mmio piece in the fragment
8516  *       write gpa, len
8517  *       copy data
8518  *       exit
8519  */
8520 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8521 {
8522 	struct kvm_run *run = vcpu->run;
8523 	struct kvm_mmio_fragment *frag;
8524 	unsigned len;
8525 
8526 	BUG_ON(!vcpu->mmio_needed);
8527 
8528 	/* Complete previous fragment */
8529 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8530 	len = min(8u, frag->len);
8531 	if (!vcpu->mmio_is_write)
8532 		memcpy(frag->data, run->mmio.data, len);
8533 
8534 	if (frag->len <= 8) {
8535 		/* Switch to the next fragment. */
8536 		frag++;
8537 		vcpu->mmio_cur_fragment++;
8538 	} else {
8539 		/* Go forward to the next mmio piece. */
8540 		frag->data += len;
8541 		frag->gpa += len;
8542 		frag->len -= len;
8543 	}
8544 
8545 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8546 		vcpu->mmio_needed = 0;
8547 
8548 		/* FIXME: return into emulator if single-stepping.  */
8549 		if (vcpu->mmio_is_write)
8550 			return 1;
8551 		vcpu->mmio_read_completed = 1;
8552 		return complete_emulated_io(vcpu);
8553 	}
8554 
8555 	run->exit_reason = KVM_EXIT_MMIO;
8556 	run->mmio.phys_addr = frag->gpa;
8557 	if (vcpu->mmio_is_write)
8558 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8559 	run->mmio.len = min(8u, frag->len);
8560 	run->mmio.is_write = vcpu->mmio_is_write;
8561 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8562 	return 0;
8563 }
8564 
8565 static void kvm_save_current_fpu(struct fpu *fpu)
8566 {
8567 	/*
8568 	 * If the target FPU state is not resident in the CPU registers, just
8569 	 * memcpy() from current, else save CPU state directly to the target.
8570 	 */
8571 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8572 		memcpy(&fpu->state, &current->thread.fpu.state,
8573 		       fpu_kernel_xstate_size);
8574 	else
8575 		copy_fpregs_to_fpstate(fpu);
8576 }
8577 
8578 /* Swap (qemu) user FPU context for the guest FPU context. */
8579 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8580 {
8581 	fpregs_lock();
8582 
8583 	kvm_save_current_fpu(vcpu->arch.user_fpu);
8584 
8585 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8586 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8587 				~XFEATURE_MASK_PKRU);
8588 
8589 	fpregs_mark_activate();
8590 	fpregs_unlock();
8591 
8592 	trace_kvm_fpu(1);
8593 }
8594 
8595 /* When vcpu_run ends, restore user space FPU context. */
8596 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8597 {
8598 	fpregs_lock();
8599 
8600 	kvm_save_current_fpu(vcpu->arch.guest_fpu);
8601 
8602 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8603 
8604 	fpregs_mark_activate();
8605 	fpregs_unlock();
8606 
8607 	++vcpu->stat.fpu_reload;
8608 	trace_kvm_fpu(0);
8609 }
8610 
8611 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8612 {
8613 	int r;
8614 
8615 	vcpu_load(vcpu);
8616 	kvm_sigset_activate(vcpu);
8617 	kvm_load_guest_fpu(vcpu);
8618 
8619 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8620 		if (kvm_run->immediate_exit) {
8621 			r = -EINTR;
8622 			goto out;
8623 		}
8624 		kvm_vcpu_block(vcpu);
8625 		kvm_apic_accept_events(vcpu);
8626 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8627 		r = -EAGAIN;
8628 		if (signal_pending(current)) {
8629 			r = -EINTR;
8630 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8631 			++vcpu->stat.signal_exits;
8632 		}
8633 		goto out;
8634 	}
8635 
8636 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8637 		r = -EINVAL;
8638 		goto out;
8639 	}
8640 
8641 	if (vcpu->run->kvm_dirty_regs) {
8642 		r = sync_regs(vcpu);
8643 		if (r != 0)
8644 			goto out;
8645 	}
8646 
8647 	/* re-sync apic's tpr */
8648 	if (!lapic_in_kernel(vcpu)) {
8649 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8650 			r = -EINVAL;
8651 			goto out;
8652 		}
8653 	}
8654 
8655 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8656 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8657 		vcpu->arch.complete_userspace_io = NULL;
8658 		r = cui(vcpu);
8659 		if (r <= 0)
8660 			goto out;
8661 	} else
8662 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8663 
8664 	if (kvm_run->immediate_exit)
8665 		r = -EINTR;
8666 	else
8667 		r = vcpu_run(vcpu);
8668 
8669 out:
8670 	kvm_put_guest_fpu(vcpu);
8671 	if (vcpu->run->kvm_valid_regs)
8672 		store_regs(vcpu);
8673 	post_kvm_run_save(vcpu);
8674 	kvm_sigset_deactivate(vcpu);
8675 
8676 	vcpu_put(vcpu);
8677 	return r;
8678 }
8679 
8680 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8681 {
8682 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8683 		/*
8684 		 * We are here if userspace calls get_regs() in the middle of
8685 		 * instruction emulation. Registers state needs to be copied
8686 		 * back from emulation context to vcpu. Userspace shouldn't do
8687 		 * that usually, but some bad designed PV devices (vmware
8688 		 * backdoor interface) need this to work
8689 		 */
8690 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8691 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8692 	}
8693 	regs->rax = kvm_rax_read(vcpu);
8694 	regs->rbx = kvm_rbx_read(vcpu);
8695 	regs->rcx = kvm_rcx_read(vcpu);
8696 	regs->rdx = kvm_rdx_read(vcpu);
8697 	regs->rsi = kvm_rsi_read(vcpu);
8698 	regs->rdi = kvm_rdi_read(vcpu);
8699 	regs->rsp = kvm_rsp_read(vcpu);
8700 	regs->rbp = kvm_rbp_read(vcpu);
8701 #ifdef CONFIG_X86_64
8702 	regs->r8 = kvm_r8_read(vcpu);
8703 	regs->r9 = kvm_r9_read(vcpu);
8704 	regs->r10 = kvm_r10_read(vcpu);
8705 	regs->r11 = kvm_r11_read(vcpu);
8706 	regs->r12 = kvm_r12_read(vcpu);
8707 	regs->r13 = kvm_r13_read(vcpu);
8708 	regs->r14 = kvm_r14_read(vcpu);
8709 	regs->r15 = kvm_r15_read(vcpu);
8710 #endif
8711 
8712 	regs->rip = kvm_rip_read(vcpu);
8713 	regs->rflags = kvm_get_rflags(vcpu);
8714 }
8715 
8716 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8717 {
8718 	vcpu_load(vcpu);
8719 	__get_regs(vcpu, regs);
8720 	vcpu_put(vcpu);
8721 	return 0;
8722 }
8723 
8724 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8725 {
8726 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8727 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8728 
8729 	kvm_rax_write(vcpu, regs->rax);
8730 	kvm_rbx_write(vcpu, regs->rbx);
8731 	kvm_rcx_write(vcpu, regs->rcx);
8732 	kvm_rdx_write(vcpu, regs->rdx);
8733 	kvm_rsi_write(vcpu, regs->rsi);
8734 	kvm_rdi_write(vcpu, regs->rdi);
8735 	kvm_rsp_write(vcpu, regs->rsp);
8736 	kvm_rbp_write(vcpu, regs->rbp);
8737 #ifdef CONFIG_X86_64
8738 	kvm_r8_write(vcpu, regs->r8);
8739 	kvm_r9_write(vcpu, regs->r9);
8740 	kvm_r10_write(vcpu, regs->r10);
8741 	kvm_r11_write(vcpu, regs->r11);
8742 	kvm_r12_write(vcpu, regs->r12);
8743 	kvm_r13_write(vcpu, regs->r13);
8744 	kvm_r14_write(vcpu, regs->r14);
8745 	kvm_r15_write(vcpu, regs->r15);
8746 #endif
8747 
8748 	kvm_rip_write(vcpu, regs->rip);
8749 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8750 
8751 	vcpu->arch.exception.pending = false;
8752 
8753 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8754 }
8755 
8756 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8757 {
8758 	vcpu_load(vcpu);
8759 	__set_regs(vcpu, regs);
8760 	vcpu_put(vcpu);
8761 	return 0;
8762 }
8763 
8764 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8765 {
8766 	struct kvm_segment cs;
8767 
8768 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8769 	*db = cs.db;
8770 	*l = cs.l;
8771 }
8772 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8773 
8774 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8775 {
8776 	struct desc_ptr dt;
8777 
8778 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8779 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8780 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8781 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8782 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8783 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8784 
8785 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8786 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8787 
8788 	kvm_x86_ops->get_idt(vcpu, &dt);
8789 	sregs->idt.limit = dt.size;
8790 	sregs->idt.base = dt.address;
8791 	kvm_x86_ops->get_gdt(vcpu, &dt);
8792 	sregs->gdt.limit = dt.size;
8793 	sregs->gdt.base = dt.address;
8794 
8795 	sregs->cr0 = kvm_read_cr0(vcpu);
8796 	sregs->cr2 = vcpu->arch.cr2;
8797 	sregs->cr3 = kvm_read_cr3(vcpu);
8798 	sregs->cr4 = kvm_read_cr4(vcpu);
8799 	sregs->cr8 = kvm_get_cr8(vcpu);
8800 	sregs->efer = vcpu->arch.efer;
8801 	sregs->apic_base = kvm_get_apic_base(vcpu);
8802 
8803 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8804 
8805 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8806 		set_bit(vcpu->arch.interrupt.nr,
8807 			(unsigned long *)sregs->interrupt_bitmap);
8808 }
8809 
8810 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8811 				  struct kvm_sregs *sregs)
8812 {
8813 	vcpu_load(vcpu);
8814 	__get_sregs(vcpu, sregs);
8815 	vcpu_put(vcpu);
8816 	return 0;
8817 }
8818 
8819 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8820 				    struct kvm_mp_state *mp_state)
8821 {
8822 	vcpu_load(vcpu);
8823 	if (kvm_mpx_supported())
8824 		kvm_load_guest_fpu(vcpu);
8825 
8826 	kvm_apic_accept_events(vcpu);
8827 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8828 					vcpu->arch.pv.pv_unhalted)
8829 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8830 	else
8831 		mp_state->mp_state = vcpu->arch.mp_state;
8832 
8833 	if (kvm_mpx_supported())
8834 		kvm_put_guest_fpu(vcpu);
8835 	vcpu_put(vcpu);
8836 	return 0;
8837 }
8838 
8839 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8840 				    struct kvm_mp_state *mp_state)
8841 {
8842 	int ret = -EINVAL;
8843 
8844 	vcpu_load(vcpu);
8845 
8846 	if (!lapic_in_kernel(vcpu) &&
8847 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8848 		goto out;
8849 
8850 	/*
8851 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8852 	 * INIT state; latched init should be reported using
8853 	 * KVM_SET_VCPU_EVENTS, so reject it here.
8854 	 */
8855 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8856 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8857 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8858 		goto out;
8859 
8860 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8861 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8862 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8863 	} else
8864 		vcpu->arch.mp_state = mp_state->mp_state;
8865 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8866 
8867 	ret = 0;
8868 out:
8869 	vcpu_put(vcpu);
8870 	return ret;
8871 }
8872 
8873 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8874 		    int reason, bool has_error_code, u32 error_code)
8875 {
8876 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8877 	int ret;
8878 
8879 	init_emulate_ctxt(vcpu);
8880 
8881 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8882 				   has_error_code, error_code);
8883 	if (ret) {
8884 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8885 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8886 		vcpu->run->internal.ndata = 0;
8887 		return 0;
8888 	}
8889 
8890 	kvm_rip_write(vcpu, ctxt->eip);
8891 	kvm_set_rflags(vcpu, ctxt->eflags);
8892 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8893 	return 1;
8894 }
8895 EXPORT_SYMBOL_GPL(kvm_task_switch);
8896 
8897 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8898 {
8899 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8900 		/*
8901 		 * When EFER.LME and CR0.PG are set, the processor is in
8902 		 * 64-bit mode (though maybe in a 32-bit code segment).
8903 		 * CR4.PAE and EFER.LMA must be set.
8904 		 */
8905 		if (!(sregs->cr4 & X86_CR4_PAE)
8906 		    || !(sregs->efer & EFER_LMA))
8907 			return -EINVAL;
8908 	} else {
8909 		/*
8910 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8911 		 * segment cannot be 64-bit.
8912 		 */
8913 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8914 			return -EINVAL;
8915 	}
8916 
8917 	return kvm_valid_cr4(vcpu, sregs->cr4);
8918 }
8919 
8920 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8921 {
8922 	struct msr_data apic_base_msr;
8923 	int mmu_reset_needed = 0;
8924 	int cpuid_update_needed = 0;
8925 	int pending_vec, max_bits, idx;
8926 	struct desc_ptr dt;
8927 	int ret = -EINVAL;
8928 
8929 	if (kvm_valid_sregs(vcpu, sregs))
8930 		goto out;
8931 
8932 	apic_base_msr.data = sregs->apic_base;
8933 	apic_base_msr.host_initiated = true;
8934 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8935 		goto out;
8936 
8937 	dt.size = sregs->idt.limit;
8938 	dt.address = sregs->idt.base;
8939 	kvm_x86_ops->set_idt(vcpu, &dt);
8940 	dt.size = sregs->gdt.limit;
8941 	dt.address = sregs->gdt.base;
8942 	kvm_x86_ops->set_gdt(vcpu, &dt);
8943 
8944 	vcpu->arch.cr2 = sregs->cr2;
8945 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8946 	vcpu->arch.cr3 = sregs->cr3;
8947 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
8948 
8949 	kvm_set_cr8(vcpu, sregs->cr8);
8950 
8951 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8952 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8953 
8954 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8955 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8956 	vcpu->arch.cr0 = sregs->cr0;
8957 
8958 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8959 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8960 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8961 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8962 	if (cpuid_update_needed)
8963 		kvm_update_cpuid(vcpu);
8964 
8965 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8966 	if (is_pae_paging(vcpu)) {
8967 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8968 		mmu_reset_needed = 1;
8969 	}
8970 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8971 
8972 	if (mmu_reset_needed)
8973 		kvm_mmu_reset_context(vcpu);
8974 
8975 	max_bits = KVM_NR_INTERRUPTS;
8976 	pending_vec = find_first_bit(
8977 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8978 	if (pending_vec < max_bits) {
8979 		kvm_queue_interrupt(vcpu, pending_vec, false);
8980 		pr_debug("Set back pending irq %d\n", pending_vec);
8981 	}
8982 
8983 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8984 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8985 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8986 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8987 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8988 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8989 
8990 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8991 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8992 
8993 	update_cr8_intercept(vcpu);
8994 
8995 	/* Older userspace won't unhalt the vcpu on reset. */
8996 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8997 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8998 	    !is_protmode(vcpu))
8999 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9000 
9001 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9002 
9003 	ret = 0;
9004 out:
9005 	return ret;
9006 }
9007 
9008 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9009 				  struct kvm_sregs *sregs)
9010 {
9011 	int ret;
9012 
9013 	vcpu_load(vcpu);
9014 	ret = __set_sregs(vcpu, sregs);
9015 	vcpu_put(vcpu);
9016 	return ret;
9017 }
9018 
9019 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9020 					struct kvm_guest_debug *dbg)
9021 {
9022 	unsigned long rflags;
9023 	int i, r;
9024 
9025 	vcpu_load(vcpu);
9026 
9027 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9028 		r = -EBUSY;
9029 		if (vcpu->arch.exception.pending)
9030 			goto out;
9031 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9032 			kvm_queue_exception(vcpu, DB_VECTOR);
9033 		else
9034 			kvm_queue_exception(vcpu, BP_VECTOR);
9035 	}
9036 
9037 	/*
9038 	 * Read rflags as long as potentially injected trace flags are still
9039 	 * filtered out.
9040 	 */
9041 	rflags = kvm_get_rflags(vcpu);
9042 
9043 	vcpu->guest_debug = dbg->control;
9044 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9045 		vcpu->guest_debug = 0;
9046 
9047 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9048 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
9049 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9050 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9051 	} else {
9052 		for (i = 0; i < KVM_NR_DB_REGS; i++)
9053 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9054 	}
9055 	kvm_update_dr7(vcpu);
9056 
9057 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9058 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9059 			get_segment_base(vcpu, VCPU_SREG_CS);
9060 
9061 	/*
9062 	 * Trigger an rflags update that will inject or remove the trace
9063 	 * flags.
9064 	 */
9065 	kvm_set_rflags(vcpu, rflags);
9066 
9067 	kvm_x86_ops->update_bp_intercept(vcpu);
9068 
9069 	r = 0;
9070 
9071 out:
9072 	vcpu_put(vcpu);
9073 	return r;
9074 }
9075 
9076 /*
9077  * Translate a guest virtual address to a guest physical address.
9078  */
9079 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9080 				    struct kvm_translation *tr)
9081 {
9082 	unsigned long vaddr = tr->linear_address;
9083 	gpa_t gpa;
9084 	int idx;
9085 
9086 	vcpu_load(vcpu);
9087 
9088 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9089 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9090 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9091 	tr->physical_address = gpa;
9092 	tr->valid = gpa != UNMAPPED_GVA;
9093 	tr->writeable = 1;
9094 	tr->usermode = 0;
9095 
9096 	vcpu_put(vcpu);
9097 	return 0;
9098 }
9099 
9100 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9101 {
9102 	struct fxregs_state *fxsave;
9103 
9104 	vcpu_load(vcpu);
9105 
9106 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9107 	memcpy(fpu->fpr, fxsave->st_space, 128);
9108 	fpu->fcw = fxsave->cwd;
9109 	fpu->fsw = fxsave->swd;
9110 	fpu->ftwx = fxsave->twd;
9111 	fpu->last_opcode = fxsave->fop;
9112 	fpu->last_ip = fxsave->rip;
9113 	fpu->last_dp = fxsave->rdp;
9114 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9115 
9116 	vcpu_put(vcpu);
9117 	return 0;
9118 }
9119 
9120 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9121 {
9122 	struct fxregs_state *fxsave;
9123 
9124 	vcpu_load(vcpu);
9125 
9126 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9127 
9128 	memcpy(fxsave->st_space, fpu->fpr, 128);
9129 	fxsave->cwd = fpu->fcw;
9130 	fxsave->swd = fpu->fsw;
9131 	fxsave->twd = fpu->ftwx;
9132 	fxsave->fop = fpu->last_opcode;
9133 	fxsave->rip = fpu->last_ip;
9134 	fxsave->rdp = fpu->last_dp;
9135 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9136 
9137 	vcpu_put(vcpu);
9138 	return 0;
9139 }
9140 
9141 static void store_regs(struct kvm_vcpu *vcpu)
9142 {
9143 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9144 
9145 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9146 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9147 
9148 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9149 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9150 
9151 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9152 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9153 				vcpu, &vcpu->run->s.regs.events);
9154 }
9155 
9156 static int sync_regs(struct kvm_vcpu *vcpu)
9157 {
9158 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9159 		return -EINVAL;
9160 
9161 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9162 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9163 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9164 	}
9165 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9166 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9167 			return -EINVAL;
9168 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9169 	}
9170 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9171 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9172 				vcpu, &vcpu->run->s.regs.events))
9173 			return -EINVAL;
9174 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9175 	}
9176 
9177 	return 0;
9178 }
9179 
9180 static void fx_init(struct kvm_vcpu *vcpu)
9181 {
9182 	fpstate_init(&vcpu->arch.guest_fpu->state);
9183 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9184 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9185 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9186 
9187 	/*
9188 	 * Ensure guest xcr0 is valid for loading
9189 	 */
9190 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9191 
9192 	vcpu->arch.cr0 |= X86_CR0_ET;
9193 }
9194 
9195 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9196 {
9197 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9198 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9199 			     "guest TSC will not be reliable\n");
9200 
9201 	return 0;
9202 }
9203 
9204 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9205 {
9206 	struct page *page;
9207 	int r;
9208 
9209 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9210 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9211 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9212 	else
9213 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9214 
9215 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9216 
9217 	r = kvm_mmu_create(vcpu);
9218 	if (r < 0)
9219 		return r;
9220 
9221 	if (irqchip_in_kernel(vcpu->kvm)) {
9222 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu->kvm);
9223 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9224 		if (r < 0)
9225 			goto fail_mmu_destroy;
9226 	} else
9227 		static_key_slow_inc(&kvm_no_apic_vcpu);
9228 
9229 	r = -ENOMEM;
9230 
9231 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9232 	if (!page)
9233 		goto fail_free_lapic;
9234 	vcpu->arch.pio_data = page_address(page);
9235 
9236 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9237 				       GFP_KERNEL_ACCOUNT);
9238 	if (!vcpu->arch.mce_banks)
9239 		goto fail_free_pio_data;
9240 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9241 
9242 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9243 				GFP_KERNEL_ACCOUNT))
9244 		goto fail_free_mce_banks;
9245 
9246 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9247 						GFP_KERNEL_ACCOUNT);
9248 	if (!vcpu->arch.user_fpu) {
9249 		pr_err("kvm: failed to allocate userspace's fpu\n");
9250 		goto free_wbinvd_dirty_mask;
9251 	}
9252 
9253 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9254 						 GFP_KERNEL_ACCOUNT);
9255 	if (!vcpu->arch.guest_fpu) {
9256 		pr_err("kvm: failed to allocate vcpu's fpu\n");
9257 		goto free_user_fpu;
9258 	}
9259 	fx_init(vcpu);
9260 
9261 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9262 
9263 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9264 
9265 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9266 
9267 	kvm_async_pf_hash_reset(vcpu);
9268 	kvm_pmu_init(vcpu);
9269 
9270 	vcpu->arch.pending_external_vector = -1;
9271 	vcpu->arch.preempted_in_kernel = false;
9272 
9273 	kvm_hv_vcpu_init(vcpu);
9274 
9275 	r = kvm_x86_ops->vcpu_create(vcpu);
9276 	if (r)
9277 		goto free_guest_fpu;
9278 
9279 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9280 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9281 	kvm_vcpu_mtrr_init(vcpu);
9282 	vcpu_load(vcpu);
9283 	kvm_vcpu_reset(vcpu, false);
9284 	kvm_init_mmu(vcpu, false);
9285 	vcpu_put(vcpu);
9286 	return 0;
9287 
9288 free_guest_fpu:
9289 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9290 free_user_fpu:
9291 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9292 free_wbinvd_dirty_mask:
9293 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9294 fail_free_mce_banks:
9295 	kfree(vcpu->arch.mce_banks);
9296 fail_free_pio_data:
9297 	free_page((unsigned long)vcpu->arch.pio_data);
9298 fail_free_lapic:
9299 	kvm_free_lapic(vcpu);
9300 fail_mmu_destroy:
9301 	kvm_mmu_destroy(vcpu);
9302 	return r;
9303 }
9304 
9305 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9306 {
9307 	struct msr_data msr;
9308 	struct kvm *kvm = vcpu->kvm;
9309 
9310 	kvm_hv_vcpu_postcreate(vcpu);
9311 
9312 	if (mutex_lock_killable(&vcpu->mutex))
9313 		return;
9314 	vcpu_load(vcpu);
9315 	msr.data = 0x0;
9316 	msr.index = MSR_IA32_TSC;
9317 	msr.host_initiated = true;
9318 	kvm_write_tsc(vcpu, &msr);
9319 	vcpu_put(vcpu);
9320 
9321 	/* poll control enabled by default */
9322 	vcpu->arch.msr_kvm_poll_control = 1;
9323 
9324 	mutex_unlock(&vcpu->mutex);
9325 
9326 	if (!kvmclock_periodic_sync)
9327 		return;
9328 
9329 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9330 					KVMCLOCK_SYNC_PERIOD);
9331 }
9332 
9333 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9334 {
9335 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9336 	int idx;
9337 
9338 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
9339 
9340 	kvmclock_reset(vcpu);
9341 
9342 	kvm_x86_ops->vcpu_free(vcpu);
9343 
9344 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9345 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9346 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9347 
9348 	kvm_hv_vcpu_uninit(vcpu);
9349 	kvm_pmu_destroy(vcpu);
9350 	kfree(vcpu->arch.mce_banks);
9351 	kvm_free_lapic(vcpu);
9352 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9353 	kvm_mmu_destroy(vcpu);
9354 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9355 	free_page((unsigned long)vcpu->arch.pio_data);
9356 	if (!lapic_in_kernel(vcpu))
9357 		static_key_slow_dec(&kvm_no_apic_vcpu);
9358 }
9359 
9360 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9361 {
9362 	kvm_lapic_reset(vcpu, init_event);
9363 
9364 	vcpu->arch.hflags = 0;
9365 
9366 	vcpu->arch.smi_pending = 0;
9367 	vcpu->arch.smi_count = 0;
9368 	atomic_set(&vcpu->arch.nmi_queued, 0);
9369 	vcpu->arch.nmi_pending = 0;
9370 	vcpu->arch.nmi_injected = false;
9371 	kvm_clear_interrupt_queue(vcpu);
9372 	kvm_clear_exception_queue(vcpu);
9373 
9374 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9375 	kvm_update_dr0123(vcpu);
9376 	vcpu->arch.dr6 = DR6_INIT;
9377 	kvm_update_dr6(vcpu);
9378 	vcpu->arch.dr7 = DR7_FIXED_1;
9379 	kvm_update_dr7(vcpu);
9380 
9381 	vcpu->arch.cr2 = 0;
9382 
9383 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9384 	vcpu->arch.apf.msr_val = 0;
9385 	vcpu->arch.st.msr_val = 0;
9386 
9387 	kvmclock_reset(vcpu);
9388 
9389 	kvm_clear_async_pf_completion_queue(vcpu);
9390 	kvm_async_pf_hash_reset(vcpu);
9391 	vcpu->arch.apf.halted = false;
9392 
9393 	if (kvm_mpx_supported()) {
9394 		void *mpx_state_buffer;
9395 
9396 		/*
9397 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9398 		 * called with loaded FPU and does not let userspace fix the state.
9399 		 */
9400 		if (init_event)
9401 			kvm_put_guest_fpu(vcpu);
9402 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9403 					XFEATURE_BNDREGS);
9404 		if (mpx_state_buffer)
9405 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9406 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9407 					XFEATURE_BNDCSR);
9408 		if (mpx_state_buffer)
9409 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9410 		if (init_event)
9411 			kvm_load_guest_fpu(vcpu);
9412 	}
9413 
9414 	if (!init_event) {
9415 		kvm_pmu_reset(vcpu);
9416 		vcpu->arch.smbase = 0x30000;
9417 
9418 		vcpu->arch.msr_misc_features_enables = 0;
9419 
9420 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9421 	}
9422 
9423 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9424 	vcpu->arch.regs_avail = ~0;
9425 	vcpu->arch.regs_dirty = ~0;
9426 
9427 	vcpu->arch.ia32_xss = 0;
9428 
9429 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
9430 }
9431 
9432 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9433 {
9434 	struct kvm_segment cs;
9435 
9436 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9437 	cs.selector = vector << 8;
9438 	cs.base = vector << 12;
9439 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9440 	kvm_rip_write(vcpu, 0);
9441 }
9442 
9443 int kvm_arch_hardware_enable(void)
9444 {
9445 	struct kvm *kvm;
9446 	struct kvm_vcpu *vcpu;
9447 	int i;
9448 	int ret;
9449 	u64 local_tsc;
9450 	u64 max_tsc = 0;
9451 	bool stable, backwards_tsc = false;
9452 
9453 	kvm_shared_msr_cpu_online();
9454 	ret = kvm_x86_ops->hardware_enable();
9455 	if (ret != 0)
9456 		return ret;
9457 
9458 	local_tsc = rdtsc();
9459 	stable = !kvm_check_tsc_unstable();
9460 	list_for_each_entry(kvm, &vm_list, vm_list) {
9461 		kvm_for_each_vcpu(i, vcpu, kvm) {
9462 			if (!stable && vcpu->cpu == smp_processor_id())
9463 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9464 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9465 				backwards_tsc = true;
9466 				if (vcpu->arch.last_host_tsc > max_tsc)
9467 					max_tsc = vcpu->arch.last_host_tsc;
9468 			}
9469 		}
9470 	}
9471 
9472 	/*
9473 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9474 	 * platforms that reset TSC during suspend or hibernate actions, but
9475 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9476 	 * detect that condition here, which happens early in CPU bringup,
9477 	 * before any KVM threads can be running.  Unfortunately, we can't
9478 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9479 	 * enough into CPU bringup that we know how much real time has actually
9480 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9481 	 * variables that haven't been updated yet.
9482 	 *
9483 	 * So we simply find the maximum observed TSC above, then record the
9484 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9485 	 * the adjustment will be applied.  Note that we accumulate
9486 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9487 	 * gets a chance to run again.  In the event that no KVM threads get a
9488 	 * chance to run, we will miss the entire elapsed period, as we'll have
9489 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9490 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9491 	 * uniform across all VCPUs (not to mention the scenario is extremely
9492 	 * unlikely). It is possible that a second hibernate recovery happens
9493 	 * much faster than a first, causing the observed TSC here to be
9494 	 * smaller; this would require additional padding adjustment, which is
9495 	 * why we set last_host_tsc to the local tsc observed here.
9496 	 *
9497 	 * N.B. - this code below runs only on platforms with reliable TSC,
9498 	 * as that is the only way backwards_tsc is set above.  Also note
9499 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9500 	 * have the same delta_cyc adjustment applied if backwards_tsc
9501 	 * is detected.  Note further, this adjustment is only done once,
9502 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9503 	 * called multiple times (one for each physical CPU bringup).
9504 	 *
9505 	 * Platforms with unreliable TSCs don't have to deal with this, they
9506 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9507 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9508 	 * guarantee that they stay in perfect synchronization.
9509 	 */
9510 	if (backwards_tsc) {
9511 		u64 delta_cyc = max_tsc - local_tsc;
9512 		list_for_each_entry(kvm, &vm_list, vm_list) {
9513 			kvm->arch.backwards_tsc_observed = true;
9514 			kvm_for_each_vcpu(i, vcpu, kvm) {
9515 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9516 				vcpu->arch.last_host_tsc = local_tsc;
9517 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9518 			}
9519 
9520 			/*
9521 			 * We have to disable TSC offset matching.. if you were
9522 			 * booting a VM while issuing an S4 host suspend....
9523 			 * you may have some problem.  Solving this issue is
9524 			 * left as an exercise to the reader.
9525 			 */
9526 			kvm->arch.last_tsc_nsec = 0;
9527 			kvm->arch.last_tsc_write = 0;
9528 		}
9529 
9530 	}
9531 	return 0;
9532 }
9533 
9534 void kvm_arch_hardware_disable(void)
9535 {
9536 	kvm_x86_ops->hardware_disable();
9537 	drop_user_return_notifiers();
9538 }
9539 
9540 int kvm_arch_hardware_setup(void)
9541 {
9542 	int r;
9543 
9544 	r = kvm_x86_ops->hardware_setup();
9545 	if (r != 0)
9546 		return r;
9547 
9548 	cr4_reserved_bits = kvm_host_cr4_reserved_bits(&boot_cpu_data);
9549 
9550 	if (kvm_has_tsc_control) {
9551 		/*
9552 		 * Make sure the user can only configure tsc_khz values that
9553 		 * fit into a signed integer.
9554 		 * A min value is not calculated because it will always
9555 		 * be 1 on all machines.
9556 		 */
9557 		u64 max = min(0x7fffffffULL,
9558 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9559 		kvm_max_guest_tsc_khz = max;
9560 
9561 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9562 	}
9563 
9564 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9565 		rdmsrl(MSR_IA32_XSS, host_xss);
9566 
9567 	kvm_init_msr_list();
9568 	return 0;
9569 }
9570 
9571 void kvm_arch_hardware_unsetup(void)
9572 {
9573 	kvm_x86_ops->hardware_unsetup();
9574 }
9575 
9576 int kvm_arch_check_processor_compat(void)
9577 {
9578 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9579 
9580 	WARN_ON(!irqs_disabled());
9581 
9582 	if (kvm_host_cr4_reserved_bits(c) != cr4_reserved_bits)
9583 		return -EIO;
9584 
9585 	return kvm_x86_ops->check_processor_compatibility();
9586 }
9587 
9588 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9589 {
9590 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9591 }
9592 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9593 
9594 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9595 {
9596 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9597 }
9598 
9599 struct static_key kvm_no_apic_vcpu __read_mostly;
9600 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9601 
9602 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9603 {
9604 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9605 
9606 	vcpu->arch.l1tf_flush_l1d = true;
9607 	if (pmu->version && unlikely(pmu->event_count)) {
9608 		pmu->need_cleanup = true;
9609 		kvm_make_request(KVM_REQ_PMU, vcpu);
9610 	}
9611 	kvm_x86_ops->sched_in(vcpu, cpu);
9612 }
9613 
9614 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9615 {
9616 	if (type)
9617 		return -EINVAL;
9618 
9619 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9620 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9621 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9622 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9623 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9624 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9625 
9626 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9627 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9628 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9629 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9630 		&kvm->arch.irq_sources_bitmap);
9631 
9632 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9633 	mutex_init(&kvm->arch.apic_map_lock);
9634 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9635 
9636 	kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9637 	pvclock_update_vm_gtod_copy(kvm);
9638 
9639 	kvm->arch.guest_can_read_msr_platform_info = true;
9640 
9641 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9642 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9643 
9644 	kvm_hv_init_vm(kvm);
9645 	kvm_page_track_init(kvm);
9646 	kvm_mmu_init_vm(kvm);
9647 
9648 	return kvm_x86_ops->vm_init(kvm);
9649 }
9650 
9651 int kvm_arch_post_init_vm(struct kvm *kvm)
9652 {
9653 	return kvm_mmu_post_init_vm(kvm);
9654 }
9655 
9656 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9657 {
9658 	vcpu_load(vcpu);
9659 	kvm_mmu_unload(vcpu);
9660 	vcpu_put(vcpu);
9661 }
9662 
9663 static void kvm_free_vcpus(struct kvm *kvm)
9664 {
9665 	unsigned int i;
9666 	struct kvm_vcpu *vcpu;
9667 
9668 	/*
9669 	 * Unpin any mmu pages first.
9670 	 */
9671 	kvm_for_each_vcpu(i, vcpu, kvm) {
9672 		kvm_clear_async_pf_completion_queue(vcpu);
9673 		kvm_unload_vcpu_mmu(vcpu);
9674 	}
9675 	kvm_for_each_vcpu(i, vcpu, kvm)
9676 		kvm_vcpu_destroy(vcpu);
9677 
9678 	mutex_lock(&kvm->lock);
9679 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9680 		kvm->vcpus[i] = NULL;
9681 
9682 	atomic_set(&kvm->online_vcpus, 0);
9683 	mutex_unlock(&kvm->lock);
9684 }
9685 
9686 void kvm_arch_sync_events(struct kvm *kvm)
9687 {
9688 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9689 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9690 	kvm_free_pit(kvm);
9691 }
9692 
9693 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9694 {
9695 	int i, r;
9696 	unsigned long hva;
9697 	struct kvm_memslots *slots = kvm_memslots(kvm);
9698 	struct kvm_memory_slot *slot, old;
9699 
9700 	/* Called with kvm->slots_lock held.  */
9701 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9702 		return -EINVAL;
9703 
9704 	slot = id_to_memslot(slots, id);
9705 	if (size) {
9706 		if (slot->npages)
9707 			return -EEXIST;
9708 
9709 		/*
9710 		 * MAP_SHARED to prevent internal slot pages from being moved
9711 		 * by fork()/COW.
9712 		 */
9713 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9714 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9715 		if (IS_ERR((void *)hva))
9716 			return PTR_ERR((void *)hva);
9717 	} else {
9718 		if (!slot->npages)
9719 			return 0;
9720 
9721 		hva = 0;
9722 	}
9723 
9724 	old = *slot;
9725 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9726 		struct kvm_userspace_memory_region m;
9727 
9728 		m.slot = id | (i << 16);
9729 		m.flags = 0;
9730 		m.guest_phys_addr = gpa;
9731 		m.userspace_addr = hva;
9732 		m.memory_size = size;
9733 		r = __kvm_set_memory_region(kvm, &m);
9734 		if (r < 0)
9735 			return r;
9736 	}
9737 
9738 	if (!size)
9739 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9740 
9741 	return 0;
9742 }
9743 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9744 
9745 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9746 {
9747 	kvm_mmu_pre_destroy_vm(kvm);
9748 }
9749 
9750 void kvm_arch_destroy_vm(struct kvm *kvm)
9751 {
9752 	if (current->mm == kvm->mm) {
9753 		/*
9754 		 * Free memory regions allocated on behalf of userspace,
9755 		 * unless the the memory map has changed due to process exit
9756 		 * or fd copying.
9757 		 */
9758 		mutex_lock(&kvm->slots_lock);
9759 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9760 					0, 0);
9761 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9762 					0, 0);
9763 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9764 		mutex_unlock(&kvm->slots_lock);
9765 	}
9766 	if (kvm_x86_ops->vm_destroy)
9767 		kvm_x86_ops->vm_destroy(kvm);
9768 	kvm_pic_destroy(kvm);
9769 	kvm_ioapic_destroy(kvm);
9770 	kvm_free_vcpus(kvm);
9771 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9772 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9773 	kvm_mmu_uninit_vm(kvm);
9774 	kvm_page_track_cleanup(kvm);
9775 	kvm_hv_destroy_vm(kvm);
9776 }
9777 
9778 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9779 			   struct kvm_memory_slot *dont)
9780 {
9781 	int i;
9782 
9783 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9784 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9785 			kvfree(free->arch.rmap[i]);
9786 			free->arch.rmap[i] = NULL;
9787 		}
9788 		if (i == 0)
9789 			continue;
9790 
9791 		if (!dont || free->arch.lpage_info[i - 1] !=
9792 			     dont->arch.lpage_info[i - 1]) {
9793 			kvfree(free->arch.lpage_info[i - 1]);
9794 			free->arch.lpage_info[i - 1] = NULL;
9795 		}
9796 	}
9797 
9798 	kvm_page_track_free_memslot(free, dont);
9799 }
9800 
9801 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9802 			    unsigned long npages)
9803 {
9804 	int i;
9805 
9806 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9807 		struct kvm_lpage_info *linfo;
9808 		unsigned long ugfn;
9809 		int lpages;
9810 		int level = i + 1;
9811 
9812 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9813 				      slot->base_gfn, level) + 1;
9814 
9815 		slot->arch.rmap[i] =
9816 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9817 				 GFP_KERNEL_ACCOUNT);
9818 		if (!slot->arch.rmap[i])
9819 			goto out_free;
9820 		if (i == 0)
9821 			continue;
9822 
9823 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9824 		if (!linfo)
9825 			goto out_free;
9826 
9827 		slot->arch.lpage_info[i - 1] = linfo;
9828 
9829 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9830 			linfo[0].disallow_lpage = 1;
9831 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9832 			linfo[lpages - 1].disallow_lpage = 1;
9833 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9834 		/*
9835 		 * If the gfn and userspace address are not aligned wrt each
9836 		 * other, or if explicitly asked to, disable large page
9837 		 * support for this slot
9838 		 */
9839 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9840 		    !kvm_largepages_enabled()) {
9841 			unsigned long j;
9842 
9843 			for (j = 0; j < lpages; ++j)
9844 				linfo[j].disallow_lpage = 1;
9845 		}
9846 	}
9847 
9848 	if (kvm_page_track_create_memslot(slot, npages))
9849 		goto out_free;
9850 
9851 	return 0;
9852 
9853 out_free:
9854 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9855 		kvfree(slot->arch.rmap[i]);
9856 		slot->arch.rmap[i] = NULL;
9857 		if (i == 0)
9858 			continue;
9859 
9860 		kvfree(slot->arch.lpage_info[i - 1]);
9861 		slot->arch.lpage_info[i - 1] = NULL;
9862 	}
9863 	return -ENOMEM;
9864 }
9865 
9866 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9867 {
9868 	struct kvm_vcpu *vcpu;
9869 	int i;
9870 
9871 	/*
9872 	 * memslots->generation has been incremented.
9873 	 * mmio generation may have reached its maximum value.
9874 	 */
9875 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9876 
9877 	/* Force re-initialization of steal_time cache */
9878 	kvm_for_each_vcpu(i, vcpu, kvm)
9879 		kvm_vcpu_kick(vcpu);
9880 }
9881 
9882 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9883 				struct kvm_memory_slot *memslot,
9884 				const struct kvm_userspace_memory_region *mem,
9885 				enum kvm_mr_change change)
9886 {
9887 	return 0;
9888 }
9889 
9890 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9891 				     struct kvm_memory_slot *new)
9892 {
9893 	/* Still write protect RO slot */
9894 	if (new->flags & KVM_MEM_READONLY) {
9895 		kvm_mmu_slot_remove_write_access(kvm, new);
9896 		return;
9897 	}
9898 
9899 	/*
9900 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9901 	 *
9902 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9903 	 *
9904 	 *  - KVM_MR_CREATE with dirty logging is disabled
9905 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9906 	 *
9907 	 * The reason is, in case of PML, we need to set D-bit for any slots
9908 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9909 	 * logging in PML buffer (and potential PML buffer full VMEXIT). This
9910 	 * guarantees leaving PML enabled during guest's lifetime won't have
9911 	 * any additional overhead from PML when guest is running with dirty
9912 	 * logging disabled for memory slots.
9913 	 *
9914 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9915 	 * to dirty logging mode.
9916 	 *
9917 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9918 	 *
9919 	 * In case of write protect:
9920 	 *
9921 	 * Write protect all pages for dirty logging.
9922 	 *
9923 	 * All the sptes including the large sptes which point to this
9924 	 * slot are set to readonly. We can not create any new large
9925 	 * spte on this slot until the end of the logging.
9926 	 *
9927 	 * See the comments in fast_page_fault().
9928 	 */
9929 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9930 		if (kvm_x86_ops->slot_enable_log_dirty)
9931 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9932 		else
9933 			kvm_mmu_slot_remove_write_access(kvm, new);
9934 	} else {
9935 		if (kvm_x86_ops->slot_disable_log_dirty)
9936 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9937 	}
9938 }
9939 
9940 void kvm_arch_commit_memory_region(struct kvm *kvm,
9941 				const struct kvm_userspace_memory_region *mem,
9942 				const struct kvm_memory_slot *old,
9943 				const struct kvm_memory_slot *new,
9944 				enum kvm_mr_change change)
9945 {
9946 	if (!kvm->arch.n_requested_mmu_pages)
9947 		kvm_mmu_change_mmu_pages(kvm,
9948 				kvm_mmu_calculate_default_mmu_pages(kvm));
9949 
9950 	/*
9951 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9952 	 * sptes have to be split.  If live migration is successful, the guest
9953 	 * in the source machine will be destroyed and large sptes will be
9954 	 * created in the destination. However, if the guest continues to run
9955 	 * in the source machine (for example if live migration fails), small
9956 	 * sptes will remain around and cause bad performance.
9957 	 *
9958 	 * Scan sptes if dirty logging has been stopped, dropping those
9959 	 * which can be collapsed into a single large-page spte.  Later
9960 	 * page faults will create the large-page sptes.
9961 	 *
9962 	 * There is no need to do this in any of the following cases:
9963 	 * CREATE:	No dirty mappings will already exist.
9964 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
9965 	 *		kvm_arch_flush_shadow_memslot()
9966 	 */
9967 	if (change == KVM_MR_FLAGS_ONLY &&
9968 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9969 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9970 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9971 
9972 	/*
9973 	 * Set up write protection and/or dirty logging for the new slot.
9974 	 *
9975 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9976 	 * been zapped so no dirty logging staff is needed for old slot. For
9977 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9978 	 * new and it's also covered when dealing with the new slot.
9979 	 *
9980 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9981 	 */
9982 	if (change != KVM_MR_DELETE)
9983 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9984 }
9985 
9986 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9987 {
9988 	kvm_mmu_zap_all(kvm);
9989 }
9990 
9991 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9992 				   struct kvm_memory_slot *slot)
9993 {
9994 	kvm_page_track_flush_slot(kvm, slot);
9995 }
9996 
9997 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9998 {
9999 	return (is_guest_mode(vcpu) &&
10000 			kvm_x86_ops->guest_apic_has_interrupt &&
10001 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
10002 }
10003 
10004 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10005 {
10006 	if (!list_empty_careful(&vcpu->async_pf.done))
10007 		return true;
10008 
10009 	if (kvm_apic_has_events(vcpu))
10010 		return true;
10011 
10012 	if (vcpu->arch.pv.pv_unhalted)
10013 		return true;
10014 
10015 	if (vcpu->arch.exception.pending)
10016 		return true;
10017 
10018 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10019 	    (vcpu->arch.nmi_pending &&
10020 	     kvm_x86_ops->nmi_allowed(vcpu)))
10021 		return true;
10022 
10023 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10024 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
10025 		return true;
10026 
10027 	if (kvm_arch_interrupt_allowed(vcpu) &&
10028 	    (kvm_cpu_has_interrupt(vcpu) ||
10029 	    kvm_guest_apic_has_interrupt(vcpu)))
10030 		return true;
10031 
10032 	if (kvm_hv_has_stimer_pending(vcpu))
10033 		return true;
10034 
10035 	return false;
10036 }
10037 
10038 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10039 {
10040 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10041 }
10042 
10043 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10044 {
10045 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10046 		return true;
10047 
10048 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10049 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
10050 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
10051 		return true;
10052 
10053 	if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
10054 		return true;
10055 
10056 	return false;
10057 }
10058 
10059 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10060 {
10061 	return vcpu->arch.preempted_in_kernel;
10062 }
10063 
10064 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10065 {
10066 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10067 }
10068 
10069 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10070 {
10071 	return kvm_x86_ops->interrupt_allowed(vcpu);
10072 }
10073 
10074 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10075 {
10076 	if (is_64_bit_mode(vcpu))
10077 		return kvm_rip_read(vcpu);
10078 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10079 		     kvm_rip_read(vcpu));
10080 }
10081 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10082 
10083 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10084 {
10085 	return kvm_get_linear_rip(vcpu) == linear_rip;
10086 }
10087 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10088 
10089 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10090 {
10091 	unsigned long rflags;
10092 
10093 	rflags = kvm_x86_ops->get_rflags(vcpu);
10094 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10095 		rflags &= ~X86_EFLAGS_TF;
10096 	return rflags;
10097 }
10098 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10099 
10100 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10101 {
10102 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10103 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10104 		rflags |= X86_EFLAGS_TF;
10105 	kvm_x86_ops->set_rflags(vcpu, rflags);
10106 }
10107 
10108 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10109 {
10110 	__kvm_set_rflags(vcpu, rflags);
10111 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10112 }
10113 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10114 
10115 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10116 {
10117 	int r;
10118 
10119 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10120 	      work->wakeup_all)
10121 		return;
10122 
10123 	r = kvm_mmu_reload(vcpu);
10124 	if (unlikely(r))
10125 		return;
10126 
10127 	if (!vcpu->arch.mmu->direct_map &&
10128 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
10129 		return;
10130 
10131 	vcpu->arch.mmu->page_fault(vcpu, work->cr2_or_gpa, 0, true);
10132 }
10133 
10134 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10135 {
10136 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10137 }
10138 
10139 static inline u32 kvm_async_pf_next_probe(u32 key)
10140 {
10141 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10142 }
10143 
10144 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10145 {
10146 	u32 key = kvm_async_pf_hash_fn(gfn);
10147 
10148 	while (vcpu->arch.apf.gfns[key] != ~0)
10149 		key = kvm_async_pf_next_probe(key);
10150 
10151 	vcpu->arch.apf.gfns[key] = gfn;
10152 }
10153 
10154 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10155 {
10156 	int i;
10157 	u32 key = kvm_async_pf_hash_fn(gfn);
10158 
10159 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10160 		     (vcpu->arch.apf.gfns[key] != gfn &&
10161 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10162 		key = kvm_async_pf_next_probe(key);
10163 
10164 	return key;
10165 }
10166 
10167 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10168 {
10169 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10170 }
10171 
10172 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10173 {
10174 	u32 i, j, k;
10175 
10176 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10177 	while (true) {
10178 		vcpu->arch.apf.gfns[i] = ~0;
10179 		do {
10180 			j = kvm_async_pf_next_probe(j);
10181 			if (vcpu->arch.apf.gfns[j] == ~0)
10182 				return;
10183 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10184 			/*
10185 			 * k lies cyclically in ]i,j]
10186 			 * |    i.k.j |
10187 			 * |....j i.k.| or  |.k..j i...|
10188 			 */
10189 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10190 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10191 		i = j;
10192 	}
10193 }
10194 
10195 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10196 {
10197 
10198 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10199 				      sizeof(val));
10200 }
10201 
10202 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10203 {
10204 
10205 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10206 				      sizeof(u32));
10207 }
10208 
10209 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10210 {
10211 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10212 		return false;
10213 
10214 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10215 	    (vcpu->arch.apf.send_user_only &&
10216 	     kvm_x86_ops->get_cpl(vcpu) == 0))
10217 		return false;
10218 
10219 	return true;
10220 }
10221 
10222 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10223 {
10224 	if (unlikely(!lapic_in_kernel(vcpu) ||
10225 		     kvm_event_needs_reinjection(vcpu) ||
10226 		     vcpu->arch.exception.pending))
10227 		return false;
10228 
10229 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10230 		return false;
10231 
10232 	/*
10233 	 * If interrupts are off we cannot even use an artificial
10234 	 * halt state.
10235 	 */
10236 	return kvm_x86_ops->interrupt_allowed(vcpu);
10237 }
10238 
10239 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10240 				     struct kvm_async_pf *work)
10241 {
10242 	struct x86_exception fault;
10243 
10244 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10245 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10246 
10247 	if (kvm_can_deliver_async_pf(vcpu) &&
10248 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10249 		fault.vector = PF_VECTOR;
10250 		fault.error_code_valid = true;
10251 		fault.error_code = 0;
10252 		fault.nested_page_fault = false;
10253 		fault.address = work->arch.token;
10254 		fault.async_page_fault = true;
10255 		kvm_inject_page_fault(vcpu, &fault);
10256 	} else {
10257 		/*
10258 		 * It is not possible to deliver a paravirtualized asynchronous
10259 		 * page fault, but putting the guest in an artificial halt state
10260 		 * can be beneficial nevertheless: if an interrupt arrives, we
10261 		 * can deliver it timely and perhaps the guest will schedule
10262 		 * another process.  When the instruction that triggered a page
10263 		 * fault is retried, hopefully the page will be ready in the host.
10264 		 */
10265 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10266 	}
10267 }
10268 
10269 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10270 				 struct kvm_async_pf *work)
10271 {
10272 	struct x86_exception fault;
10273 	u32 val;
10274 
10275 	if (work->wakeup_all)
10276 		work->arch.token = ~0; /* broadcast wakeup */
10277 	else
10278 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10279 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10280 
10281 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10282 	    !apf_get_user(vcpu, &val)) {
10283 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10284 		    vcpu->arch.exception.pending &&
10285 		    vcpu->arch.exception.nr == PF_VECTOR &&
10286 		    !apf_put_user(vcpu, 0)) {
10287 			vcpu->arch.exception.injected = false;
10288 			vcpu->arch.exception.pending = false;
10289 			vcpu->arch.exception.nr = 0;
10290 			vcpu->arch.exception.has_error_code = false;
10291 			vcpu->arch.exception.error_code = 0;
10292 			vcpu->arch.exception.has_payload = false;
10293 			vcpu->arch.exception.payload = 0;
10294 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10295 			fault.vector = PF_VECTOR;
10296 			fault.error_code_valid = true;
10297 			fault.error_code = 0;
10298 			fault.nested_page_fault = false;
10299 			fault.address = work->arch.token;
10300 			fault.async_page_fault = true;
10301 			kvm_inject_page_fault(vcpu, &fault);
10302 		}
10303 	}
10304 	vcpu->arch.apf.halted = false;
10305 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10306 }
10307 
10308 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10309 {
10310 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10311 		return true;
10312 	else
10313 		return kvm_can_do_async_pf(vcpu);
10314 }
10315 
10316 void kvm_arch_start_assignment(struct kvm *kvm)
10317 {
10318 	atomic_inc(&kvm->arch.assigned_device_count);
10319 }
10320 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10321 
10322 void kvm_arch_end_assignment(struct kvm *kvm)
10323 {
10324 	atomic_dec(&kvm->arch.assigned_device_count);
10325 }
10326 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10327 
10328 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10329 {
10330 	return atomic_read(&kvm->arch.assigned_device_count);
10331 }
10332 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10333 
10334 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10335 {
10336 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10337 }
10338 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10339 
10340 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10341 {
10342 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10343 }
10344 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10345 
10346 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10347 {
10348 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10349 }
10350 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10351 
10352 bool kvm_arch_has_irq_bypass(void)
10353 {
10354 	return true;
10355 }
10356 
10357 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10358 				      struct irq_bypass_producer *prod)
10359 {
10360 	struct kvm_kernel_irqfd *irqfd =
10361 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10362 
10363 	irqfd->producer = prod;
10364 
10365 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10366 					   prod->irq, irqfd->gsi, 1);
10367 }
10368 
10369 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10370 				      struct irq_bypass_producer *prod)
10371 {
10372 	int ret;
10373 	struct kvm_kernel_irqfd *irqfd =
10374 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10375 
10376 	WARN_ON(irqfd->producer != prod);
10377 	irqfd->producer = NULL;
10378 
10379 	/*
10380 	 * When producer of consumer is unregistered, we change back to
10381 	 * remapped mode, so we can re-use the current implementation
10382 	 * when the irq is masked/disabled or the consumer side (KVM
10383 	 * int this case doesn't want to receive the interrupts.
10384 	*/
10385 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10386 	if (ret)
10387 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10388 		       " fails: %d\n", irqfd->consumer.token, ret);
10389 }
10390 
10391 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10392 				   uint32_t guest_irq, bool set)
10393 {
10394 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10395 }
10396 
10397 bool kvm_vector_hashing_enabled(void)
10398 {
10399 	return vector_hashing;
10400 }
10401 
10402 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10403 {
10404 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10405 }
10406 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10407 
10408 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10409 {
10410 	uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10411 
10412 	/* The STIBP bit doesn't fault even if it's not advertised */
10413 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10414 	    !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10415 		bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10416 	if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10417 	    !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10418 		bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10419 
10420 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10421 	    !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10422 		bits &= ~SPEC_CTRL_SSBD;
10423 	if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10424 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10425 		bits &= ~SPEC_CTRL_SSBD;
10426 
10427 	return bits;
10428 }
10429 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10430 
10431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10432 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10433 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10434 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10435 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10451