xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 0a671dc5)
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 <clocksource/hyperv_timer.h>
72 
73 #define CREATE_TRACE_POINTS
74 #include "trace.h"
75 
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
80 
81 #define emul_to_vcpu(ctxt) \
82 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83 
84 /* EFER defaults:
85  * - enable syscall per default because its emulated by KVM
86  * - enable LME and LMA per default on 64 bit KVM
87  */
88 #ifdef CONFIG_X86_64
89 static
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
91 #else
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93 #endif
94 
95 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
96 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
97 
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
100 
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
107 
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
110 
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
113 
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
116 
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
119 
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
122 
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32  __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64  __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
133 
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
137 
138 /*
139  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
140  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
141  * advancement entirely.  Any other value is used as-is and disables adaptive
142  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
143  */
144 static int __read_mostly lapic_timer_advance_ns = -1;
145 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
146 
147 static bool __read_mostly vector_hashing = true;
148 module_param(vector_hashing, bool, S_IRUGO);
149 
150 bool __read_mostly enable_vmware_backdoor = false;
151 module_param(enable_vmware_backdoor, bool, S_IRUGO);
152 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
153 
154 static bool __read_mostly force_emulation_prefix = false;
155 module_param(force_emulation_prefix, bool, S_IRUGO);
156 
157 int __read_mostly pi_inject_timer = -1;
158 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
159 
160 #define KVM_NR_SHARED_MSRS 16
161 
162 struct kvm_shared_msrs_global {
163 	int nr;
164 	u32 msrs[KVM_NR_SHARED_MSRS];
165 };
166 
167 struct kvm_shared_msrs {
168 	struct user_return_notifier urn;
169 	bool registered;
170 	struct kvm_shared_msr_values {
171 		u64 host;
172 		u64 curr;
173 	} values[KVM_NR_SHARED_MSRS];
174 };
175 
176 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
177 static struct kvm_shared_msrs __percpu *shared_msrs;
178 
179 struct kvm_stats_debugfs_item debugfs_entries[] = {
180 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
181 	{ "pf_guest", VCPU_STAT(pf_guest) },
182 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
183 	{ "invlpg", VCPU_STAT(invlpg) },
184 	{ "exits", VCPU_STAT(exits) },
185 	{ "io_exits", VCPU_STAT(io_exits) },
186 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
187 	{ "signal_exits", VCPU_STAT(signal_exits) },
188 	{ "irq_window", VCPU_STAT(irq_window_exits) },
189 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
190 	{ "halt_exits", VCPU_STAT(halt_exits) },
191 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
192 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
193 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
194 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
195 	{ "hypercalls", VCPU_STAT(hypercalls) },
196 	{ "request_irq", VCPU_STAT(request_irq_exits) },
197 	{ "irq_exits", VCPU_STAT(irq_exits) },
198 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
199 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
200 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
201 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
202 	{ "irq_injections", VCPU_STAT(irq_injections) },
203 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
204 	{ "req_event", VCPU_STAT(req_event) },
205 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
206 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
207 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
208 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
209 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
210 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
211 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
212 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
213 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
214 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
215 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
216 	{ "max_mmu_page_hash_collisions",
217 		VM_STAT(max_mmu_page_hash_collisions) },
218 	{ NULL }
219 };
220 
221 u64 __read_mostly host_xcr0;
222 
223 struct kmem_cache *x86_fpu_cache;
224 EXPORT_SYMBOL_GPL(x86_fpu_cache);
225 
226 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
227 
228 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
229 {
230 	int i;
231 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
232 		vcpu->arch.apf.gfns[i] = ~0;
233 }
234 
235 static void kvm_on_user_return(struct user_return_notifier *urn)
236 {
237 	unsigned slot;
238 	struct kvm_shared_msrs *locals
239 		= container_of(urn, struct kvm_shared_msrs, urn);
240 	struct kvm_shared_msr_values *values;
241 	unsigned long flags;
242 
243 	/*
244 	 * Disabling irqs at this point since the following code could be
245 	 * interrupted and executed through kvm_arch_hardware_disable()
246 	 */
247 	local_irq_save(flags);
248 	if (locals->registered) {
249 		locals->registered = false;
250 		user_return_notifier_unregister(urn);
251 	}
252 	local_irq_restore(flags);
253 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
254 		values = &locals->values[slot];
255 		if (values->host != values->curr) {
256 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
257 			values->curr = values->host;
258 		}
259 	}
260 }
261 
262 static void shared_msr_update(unsigned slot, u32 msr)
263 {
264 	u64 value;
265 	unsigned int cpu = smp_processor_id();
266 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
267 
268 	/* only read, and nobody should modify it at this time,
269 	 * so don't need lock */
270 	if (slot >= shared_msrs_global.nr) {
271 		printk(KERN_ERR "kvm: invalid MSR slot!");
272 		return;
273 	}
274 	rdmsrl_safe(msr, &value);
275 	smsr->values[slot].host = value;
276 	smsr->values[slot].curr = value;
277 }
278 
279 void kvm_define_shared_msr(unsigned slot, u32 msr)
280 {
281 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
282 	shared_msrs_global.msrs[slot] = msr;
283 	if (slot >= shared_msrs_global.nr)
284 		shared_msrs_global.nr = slot + 1;
285 }
286 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
287 
288 static void kvm_shared_msr_cpu_online(void)
289 {
290 	unsigned i;
291 
292 	for (i = 0; i < shared_msrs_global.nr; ++i)
293 		shared_msr_update(i, shared_msrs_global.msrs[i]);
294 }
295 
296 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
297 {
298 	unsigned int cpu = smp_processor_id();
299 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
300 	int err;
301 
302 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
303 		return 0;
304 	smsr->values[slot].curr = value;
305 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
306 	if (err)
307 		return 1;
308 
309 	if (!smsr->registered) {
310 		smsr->urn.on_user_return = kvm_on_user_return;
311 		user_return_notifier_register(&smsr->urn);
312 		smsr->registered = true;
313 	}
314 	return 0;
315 }
316 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
317 
318 static void drop_user_return_notifiers(void)
319 {
320 	unsigned int cpu = smp_processor_id();
321 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
322 
323 	if (smsr->registered)
324 		kvm_on_user_return(&smsr->urn);
325 }
326 
327 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
328 {
329 	return vcpu->arch.apic_base;
330 }
331 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
332 
333 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
334 {
335 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
336 }
337 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
338 
339 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
340 {
341 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
342 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
343 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
344 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
345 
346 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
347 		return 1;
348 	if (!msr_info->host_initiated) {
349 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
350 			return 1;
351 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
352 			return 1;
353 	}
354 
355 	kvm_lapic_set_base(vcpu, msr_info->data);
356 	return 0;
357 }
358 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
359 
360 asmlinkage __visible void kvm_spurious_fault(void)
361 {
362 	/* Fault while not rebooting.  We want the trace. */
363 	if (!kvm_rebooting)
364 		BUG();
365 }
366 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
367 
368 #define EXCPT_BENIGN		0
369 #define EXCPT_CONTRIBUTORY	1
370 #define EXCPT_PF		2
371 
372 static int exception_class(int vector)
373 {
374 	switch (vector) {
375 	case PF_VECTOR:
376 		return EXCPT_PF;
377 	case DE_VECTOR:
378 	case TS_VECTOR:
379 	case NP_VECTOR:
380 	case SS_VECTOR:
381 	case GP_VECTOR:
382 		return EXCPT_CONTRIBUTORY;
383 	default:
384 		break;
385 	}
386 	return EXCPT_BENIGN;
387 }
388 
389 #define EXCPT_FAULT		0
390 #define EXCPT_TRAP		1
391 #define EXCPT_ABORT		2
392 #define EXCPT_INTERRUPT		3
393 
394 static int exception_type(int vector)
395 {
396 	unsigned int mask;
397 
398 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
399 		return EXCPT_INTERRUPT;
400 
401 	mask = 1 << vector;
402 
403 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
404 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
405 		return EXCPT_TRAP;
406 
407 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
408 		return EXCPT_ABORT;
409 
410 	/* Reserved exceptions will result in fault */
411 	return EXCPT_FAULT;
412 }
413 
414 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
415 {
416 	unsigned nr = vcpu->arch.exception.nr;
417 	bool has_payload = vcpu->arch.exception.has_payload;
418 	unsigned long payload = vcpu->arch.exception.payload;
419 
420 	if (!has_payload)
421 		return;
422 
423 	switch (nr) {
424 	case DB_VECTOR:
425 		/*
426 		 * "Certain debug exceptions may clear bit 0-3.  The
427 		 * remaining contents of the DR6 register are never
428 		 * cleared by the processor".
429 		 */
430 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
431 		/*
432 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
433 		 */
434 		vcpu->arch.dr6 |= DR6_RTM;
435 		vcpu->arch.dr6 |= payload;
436 		/*
437 		 * Bit 16 should be set in the payload whenever the #DB
438 		 * exception should clear DR6.RTM. This makes the payload
439 		 * compatible with the pending debug exceptions under VMX.
440 		 * Though not currently documented in the SDM, this also
441 		 * makes the payload compatible with the exit qualification
442 		 * for #DB exceptions under VMX.
443 		 */
444 		vcpu->arch.dr6 ^= payload & DR6_RTM;
445 		break;
446 	case PF_VECTOR:
447 		vcpu->arch.cr2 = payload;
448 		break;
449 	}
450 
451 	vcpu->arch.exception.has_payload = false;
452 	vcpu->arch.exception.payload = 0;
453 }
454 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
455 
456 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
457 		unsigned nr, bool has_error, u32 error_code,
458 	        bool has_payload, unsigned long payload, bool reinject)
459 {
460 	u32 prev_nr;
461 	int class1, class2;
462 
463 	kvm_make_request(KVM_REQ_EVENT, vcpu);
464 
465 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
466 	queue:
467 		if (has_error && !is_protmode(vcpu))
468 			has_error = false;
469 		if (reinject) {
470 			/*
471 			 * On vmentry, vcpu->arch.exception.pending is only
472 			 * true if an event injection was blocked by
473 			 * nested_run_pending.  In that case, however,
474 			 * vcpu_enter_guest requests an immediate exit,
475 			 * and the guest shouldn't proceed far enough to
476 			 * need reinjection.
477 			 */
478 			WARN_ON_ONCE(vcpu->arch.exception.pending);
479 			vcpu->arch.exception.injected = true;
480 			if (WARN_ON_ONCE(has_payload)) {
481 				/*
482 				 * A reinjected event has already
483 				 * delivered its payload.
484 				 */
485 				has_payload = false;
486 				payload = 0;
487 			}
488 		} else {
489 			vcpu->arch.exception.pending = true;
490 			vcpu->arch.exception.injected = false;
491 		}
492 		vcpu->arch.exception.has_error_code = has_error;
493 		vcpu->arch.exception.nr = nr;
494 		vcpu->arch.exception.error_code = error_code;
495 		vcpu->arch.exception.has_payload = has_payload;
496 		vcpu->arch.exception.payload = payload;
497 		/*
498 		 * In guest mode, payload delivery should be deferred,
499 		 * so that the L1 hypervisor can intercept #PF before
500 		 * CR2 is modified (or intercept #DB before DR6 is
501 		 * modified under nVMX).  However, for ABI
502 		 * compatibility with KVM_GET_VCPU_EVENTS and
503 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
504 		 * delivery unless userspace has enabled this
505 		 * functionality via the per-VM capability,
506 		 * KVM_CAP_EXCEPTION_PAYLOAD.
507 		 */
508 		if (!vcpu->kvm->arch.exception_payload_enabled ||
509 		    !is_guest_mode(vcpu))
510 			kvm_deliver_exception_payload(vcpu);
511 		return;
512 	}
513 
514 	/* to check exception */
515 	prev_nr = vcpu->arch.exception.nr;
516 	if (prev_nr == DF_VECTOR) {
517 		/* triple fault -> shutdown */
518 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
519 		return;
520 	}
521 	class1 = exception_class(prev_nr);
522 	class2 = exception_class(nr);
523 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
524 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
525 		/*
526 		 * Generate double fault per SDM Table 5-5.  Set
527 		 * exception.pending = true so that the double fault
528 		 * can trigger a nested vmexit.
529 		 */
530 		vcpu->arch.exception.pending = true;
531 		vcpu->arch.exception.injected = false;
532 		vcpu->arch.exception.has_error_code = true;
533 		vcpu->arch.exception.nr = DF_VECTOR;
534 		vcpu->arch.exception.error_code = 0;
535 		vcpu->arch.exception.has_payload = false;
536 		vcpu->arch.exception.payload = 0;
537 	} else
538 		/* replace previous exception with a new one in a hope
539 		   that instruction re-execution will regenerate lost
540 		   exception */
541 		goto queue;
542 }
543 
544 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
545 {
546 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
547 }
548 EXPORT_SYMBOL_GPL(kvm_queue_exception);
549 
550 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
551 {
552 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
553 }
554 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
555 
556 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
557 				  unsigned long payload)
558 {
559 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
560 }
561 
562 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
563 				    u32 error_code, unsigned long payload)
564 {
565 	kvm_multiple_exception(vcpu, nr, true, error_code,
566 			       true, payload, false);
567 }
568 
569 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
570 {
571 	if (err)
572 		kvm_inject_gp(vcpu, 0);
573 	else
574 		return kvm_skip_emulated_instruction(vcpu);
575 
576 	return 1;
577 }
578 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
579 
580 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
581 {
582 	++vcpu->stat.pf_guest;
583 	vcpu->arch.exception.nested_apf =
584 		is_guest_mode(vcpu) && fault->async_page_fault;
585 	if (vcpu->arch.exception.nested_apf) {
586 		vcpu->arch.apf.nested_apf_token = fault->address;
587 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
588 	} else {
589 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
590 					fault->address);
591 	}
592 }
593 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
594 
595 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
596 {
597 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
598 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
599 	else
600 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
601 
602 	return fault->nested_page_fault;
603 }
604 
605 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
606 {
607 	atomic_inc(&vcpu->arch.nmi_queued);
608 	kvm_make_request(KVM_REQ_NMI, vcpu);
609 }
610 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
611 
612 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
613 {
614 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
615 }
616 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
617 
618 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
619 {
620 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
621 }
622 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
623 
624 /*
625  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
626  * a #GP and return false.
627  */
628 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
629 {
630 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
631 		return true;
632 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
633 	return false;
634 }
635 EXPORT_SYMBOL_GPL(kvm_require_cpl);
636 
637 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
638 {
639 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
640 		return true;
641 
642 	kvm_queue_exception(vcpu, UD_VECTOR);
643 	return false;
644 }
645 EXPORT_SYMBOL_GPL(kvm_require_dr);
646 
647 /*
648  * This function will be used to read from the physical memory of the currently
649  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
650  * can read from guest physical or from the guest's guest physical memory.
651  */
652 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
653 			    gfn_t ngfn, void *data, int offset, int len,
654 			    u32 access)
655 {
656 	struct x86_exception exception;
657 	gfn_t real_gfn;
658 	gpa_t ngpa;
659 
660 	ngpa     = gfn_to_gpa(ngfn);
661 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
662 	if (real_gfn == UNMAPPED_GVA)
663 		return -EFAULT;
664 
665 	real_gfn = gpa_to_gfn(real_gfn);
666 
667 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
668 }
669 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
670 
671 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
672 			       void *data, int offset, int len, u32 access)
673 {
674 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
675 				       data, offset, len, access);
676 }
677 
678 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
679 {
680 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
681 	       rsvd_bits(1, 2);
682 }
683 
684 /*
685  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
686  */
687 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
688 {
689 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
690 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
691 	int i;
692 	int ret;
693 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
694 
695 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
696 				      offset * sizeof(u64), sizeof(pdpte),
697 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
698 	if (ret < 0) {
699 		ret = 0;
700 		goto out;
701 	}
702 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
703 		if ((pdpte[i] & PT_PRESENT_MASK) &&
704 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
705 			ret = 0;
706 			goto out;
707 		}
708 	}
709 	ret = 1;
710 
711 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
712 	__set_bit(VCPU_EXREG_PDPTR,
713 		  (unsigned long *)&vcpu->arch.regs_avail);
714 	__set_bit(VCPU_EXREG_PDPTR,
715 		  (unsigned long *)&vcpu->arch.regs_dirty);
716 out:
717 
718 	return ret;
719 }
720 EXPORT_SYMBOL_GPL(load_pdptrs);
721 
722 bool pdptrs_changed(struct kvm_vcpu *vcpu)
723 {
724 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
725 	bool changed = true;
726 	int offset;
727 	gfn_t gfn;
728 	int r;
729 
730 	if (!is_pae_paging(vcpu))
731 		return false;
732 
733 	if (!test_bit(VCPU_EXREG_PDPTR,
734 		      (unsigned long *)&vcpu->arch.regs_avail))
735 		return true;
736 
737 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
738 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
739 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
740 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
741 	if (r < 0)
742 		goto out;
743 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
744 out:
745 
746 	return changed;
747 }
748 EXPORT_SYMBOL_GPL(pdptrs_changed);
749 
750 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
751 {
752 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
753 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
754 
755 	cr0 |= X86_CR0_ET;
756 
757 #ifdef CONFIG_X86_64
758 	if (cr0 & 0xffffffff00000000UL)
759 		return 1;
760 #endif
761 
762 	cr0 &= ~CR0_RESERVED_BITS;
763 
764 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
765 		return 1;
766 
767 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
768 		return 1;
769 
770 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
771 #ifdef CONFIG_X86_64
772 		if ((vcpu->arch.efer & EFER_LME)) {
773 			int cs_db, cs_l;
774 
775 			if (!is_pae(vcpu))
776 				return 1;
777 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
778 			if (cs_l)
779 				return 1;
780 		} else
781 #endif
782 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
783 						 kvm_read_cr3(vcpu)))
784 			return 1;
785 	}
786 
787 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
788 		return 1;
789 
790 	kvm_x86_ops->set_cr0(vcpu, cr0);
791 
792 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
793 		kvm_clear_async_pf_completion_queue(vcpu);
794 		kvm_async_pf_hash_reset(vcpu);
795 	}
796 
797 	if ((cr0 ^ old_cr0) & update_bits)
798 		kvm_mmu_reset_context(vcpu);
799 
800 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
801 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
802 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
803 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
804 
805 	return 0;
806 }
807 EXPORT_SYMBOL_GPL(kvm_set_cr0);
808 
809 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
810 {
811 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
812 }
813 EXPORT_SYMBOL_GPL(kvm_lmsw);
814 
815 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
816 {
817 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
818 			!vcpu->guest_xcr0_loaded) {
819 		/* kvm_set_xcr() also depends on this */
820 		if (vcpu->arch.xcr0 != host_xcr0)
821 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
822 		vcpu->guest_xcr0_loaded = 1;
823 	}
824 }
825 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
826 
827 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
828 {
829 	if (vcpu->guest_xcr0_loaded) {
830 		if (vcpu->arch.xcr0 != host_xcr0)
831 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
832 		vcpu->guest_xcr0_loaded = 0;
833 	}
834 }
835 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
836 
837 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
838 {
839 	u64 xcr0 = xcr;
840 	u64 old_xcr0 = vcpu->arch.xcr0;
841 	u64 valid_bits;
842 
843 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
844 	if (index != XCR_XFEATURE_ENABLED_MASK)
845 		return 1;
846 	if (!(xcr0 & XFEATURE_MASK_FP))
847 		return 1;
848 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
849 		return 1;
850 
851 	/*
852 	 * Do not allow the guest to set bits that we do not support
853 	 * saving.  However, xcr0 bit 0 is always set, even if the
854 	 * emulated CPU does not support XSAVE (see fx_init).
855 	 */
856 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
857 	if (xcr0 & ~valid_bits)
858 		return 1;
859 
860 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
861 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
862 		return 1;
863 
864 	if (xcr0 & XFEATURE_MASK_AVX512) {
865 		if (!(xcr0 & XFEATURE_MASK_YMM))
866 			return 1;
867 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
868 			return 1;
869 	}
870 	vcpu->arch.xcr0 = xcr0;
871 
872 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
873 		kvm_update_cpuid(vcpu);
874 	return 0;
875 }
876 
877 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
878 {
879 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
880 	    __kvm_set_xcr(vcpu, index, xcr)) {
881 		kvm_inject_gp(vcpu, 0);
882 		return 1;
883 	}
884 	return 0;
885 }
886 EXPORT_SYMBOL_GPL(kvm_set_xcr);
887 
888 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
889 {
890 	if (cr4 & CR4_RESERVED_BITS)
891 		return -EINVAL;
892 
893 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
894 		return -EINVAL;
895 
896 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
897 		return -EINVAL;
898 
899 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
900 		return -EINVAL;
901 
902 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
903 		return -EINVAL;
904 
905 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
906 		return -EINVAL;
907 
908 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
909 		return -EINVAL;
910 
911 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
912 		return -EINVAL;
913 
914 	return 0;
915 }
916 
917 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
918 {
919 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
920 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
921 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
922 
923 	if (kvm_valid_cr4(vcpu, cr4))
924 		return 1;
925 
926 	if (is_long_mode(vcpu)) {
927 		if (!(cr4 & X86_CR4_PAE))
928 			return 1;
929 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
930 		   && ((cr4 ^ old_cr4) & pdptr_bits)
931 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
932 				   kvm_read_cr3(vcpu)))
933 		return 1;
934 
935 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
936 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
937 			return 1;
938 
939 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
940 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
941 			return 1;
942 	}
943 
944 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
945 		return 1;
946 
947 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
948 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
949 		kvm_mmu_reset_context(vcpu);
950 
951 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
952 		kvm_update_cpuid(vcpu);
953 
954 	return 0;
955 }
956 EXPORT_SYMBOL_GPL(kvm_set_cr4);
957 
958 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
959 {
960 	bool skip_tlb_flush = false;
961 #ifdef CONFIG_X86_64
962 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
963 
964 	if (pcid_enabled) {
965 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
966 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
967 	}
968 #endif
969 
970 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
971 		if (!skip_tlb_flush) {
972 			kvm_mmu_sync_roots(vcpu);
973 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
974 		}
975 		return 0;
976 	}
977 
978 	if (is_long_mode(vcpu) &&
979 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
980 		return 1;
981 	else if (is_pae_paging(vcpu) &&
982 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
983 		return 1;
984 
985 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
986 	vcpu->arch.cr3 = cr3;
987 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
988 
989 	return 0;
990 }
991 EXPORT_SYMBOL_GPL(kvm_set_cr3);
992 
993 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
994 {
995 	if (cr8 & CR8_RESERVED_BITS)
996 		return 1;
997 	if (lapic_in_kernel(vcpu))
998 		kvm_lapic_set_tpr(vcpu, cr8);
999 	else
1000 		vcpu->arch.cr8 = cr8;
1001 	return 0;
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1004 
1005 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1006 {
1007 	if (lapic_in_kernel(vcpu))
1008 		return kvm_lapic_get_cr8(vcpu);
1009 	else
1010 		return vcpu->arch.cr8;
1011 }
1012 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1013 
1014 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1015 {
1016 	int i;
1017 
1018 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1019 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1020 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1021 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1022 	}
1023 }
1024 
1025 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1026 {
1027 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1028 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1029 }
1030 
1031 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1032 {
1033 	unsigned long dr7;
1034 
1035 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1036 		dr7 = vcpu->arch.guest_debug_dr7;
1037 	else
1038 		dr7 = vcpu->arch.dr7;
1039 	kvm_x86_ops->set_dr7(vcpu, dr7);
1040 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1041 	if (dr7 & DR7_BP_EN_MASK)
1042 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1043 }
1044 
1045 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1046 {
1047 	u64 fixed = DR6_FIXED_1;
1048 
1049 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1050 		fixed |= DR6_RTM;
1051 	return fixed;
1052 }
1053 
1054 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1055 {
1056 	switch (dr) {
1057 	case 0 ... 3:
1058 		vcpu->arch.db[dr] = val;
1059 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1060 			vcpu->arch.eff_db[dr] = val;
1061 		break;
1062 	case 4:
1063 		/* fall through */
1064 	case 6:
1065 		if (val & 0xffffffff00000000ULL)
1066 			return -1; /* #GP */
1067 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1068 		kvm_update_dr6(vcpu);
1069 		break;
1070 	case 5:
1071 		/* fall through */
1072 	default: /* 7 */
1073 		if (val & 0xffffffff00000000ULL)
1074 			return -1; /* #GP */
1075 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1076 		kvm_update_dr7(vcpu);
1077 		break;
1078 	}
1079 
1080 	return 0;
1081 }
1082 
1083 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1084 {
1085 	if (__kvm_set_dr(vcpu, dr, val)) {
1086 		kvm_inject_gp(vcpu, 0);
1087 		return 1;
1088 	}
1089 	return 0;
1090 }
1091 EXPORT_SYMBOL_GPL(kvm_set_dr);
1092 
1093 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1094 {
1095 	switch (dr) {
1096 	case 0 ... 3:
1097 		*val = vcpu->arch.db[dr];
1098 		break;
1099 	case 4:
1100 		/* fall through */
1101 	case 6:
1102 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1103 			*val = vcpu->arch.dr6;
1104 		else
1105 			*val = kvm_x86_ops->get_dr6(vcpu);
1106 		break;
1107 	case 5:
1108 		/* fall through */
1109 	default: /* 7 */
1110 		*val = vcpu->arch.dr7;
1111 		break;
1112 	}
1113 	return 0;
1114 }
1115 EXPORT_SYMBOL_GPL(kvm_get_dr);
1116 
1117 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1118 {
1119 	u32 ecx = kvm_rcx_read(vcpu);
1120 	u64 data;
1121 	int err;
1122 
1123 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1124 	if (err)
1125 		return err;
1126 	kvm_rax_write(vcpu, (u32)data);
1127 	kvm_rdx_write(vcpu, data >> 32);
1128 	return err;
1129 }
1130 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1131 
1132 /*
1133  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1134  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1135  *
1136  * This list is modified at module load time to reflect the
1137  * capabilities of the host cpu. This capabilities test skips MSRs that are
1138  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1139  * may depend on host virtualization features rather than host cpu features.
1140  */
1141 
1142 static u32 msrs_to_save[] = {
1143 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1144 	MSR_STAR,
1145 #ifdef CONFIG_X86_64
1146 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1147 #endif
1148 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1149 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1150 	MSR_IA32_SPEC_CTRL,
1151 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1152 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1153 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1154 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1155 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1156 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1157 	MSR_IA32_UMWAIT_CONTROL,
1158 
1159 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1160 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1161 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1162 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1163 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1164 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1165 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1166 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1167 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1168 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1169 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1170 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1171 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1172 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1173 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1174 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1175 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1176 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1177 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1178 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1179 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1180 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1181 };
1182 
1183 static unsigned num_msrs_to_save;
1184 
1185 static u32 emulated_msrs[] = {
1186 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1187 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1188 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1189 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1190 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1191 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1192 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1193 	HV_X64_MSR_RESET,
1194 	HV_X64_MSR_VP_INDEX,
1195 	HV_X64_MSR_VP_RUNTIME,
1196 	HV_X64_MSR_SCONTROL,
1197 	HV_X64_MSR_STIMER0_CONFIG,
1198 	HV_X64_MSR_VP_ASSIST_PAGE,
1199 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1200 	HV_X64_MSR_TSC_EMULATION_STATUS,
1201 
1202 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1203 	MSR_KVM_PV_EOI_EN,
1204 
1205 	MSR_IA32_TSC_ADJUST,
1206 	MSR_IA32_TSCDEADLINE,
1207 	MSR_IA32_ARCH_CAPABILITIES,
1208 	MSR_IA32_MISC_ENABLE,
1209 	MSR_IA32_MCG_STATUS,
1210 	MSR_IA32_MCG_CTL,
1211 	MSR_IA32_MCG_EXT_CTL,
1212 	MSR_IA32_SMBASE,
1213 	MSR_SMI_COUNT,
1214 	MSR_PLATFORM_INFO,
1215 	MSR_MISC_FEATURES_ENABLES,
1216 	MSR_AMD64_VIRT_SPEC_CTRL,
1217 	MSR_IA32_POWER_CTL,
1218 
1219 	/*
1220 	 * The following list leaves out MSRs whose values are determined
1221 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1222 	 * We always support the "true" VMX control MSRs, even if the host
1223 	 * processor does not, so I am putting these registers here rather
1224 	 * than in msrs_to_save.
1225 	 */
1226 	MSR_IA32_VMX_BASIC,
1227 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1228 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1229 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1230 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1231 	MSR_IA32_VMX_MISC,
1232 	MSR_IA32_VMX_CR0_FIXED0,
1233 	MSR_IA32_VMX_CR4_FIXED0,
1234 	MSR_IA32_VMX_VMCS_ENUM,
1235 	MSR_IA32_VMX_PROCBASED_CTLS2,
1236 	MSR_IA32_VMX_EPT_VPID_CAP,
1237 	MSR_IA32_VMX_VMFUNC,
1238 
1239 	MSR_K7_HWCR,
1240 	MSR_KVM_POLL_CONTROL,
1241 };
1242 
1243 static unsigned num_emulated_msrs;
1244 
1245 /*
1246  * List of msr numbers which are used to expose MSR-based features that
1247  * can be used by a hypervisor to validate requested CPU features.
1248  */
1249 static u32 msr_based_features[] = {
1250 	MSR_IA32_VMX_BASIC,
1251 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1252 	MSR_IA32_VMX_PINBASED_CTLS,
1253 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1254 	MSR_IA32_VMX_PROCBASED_CTLS,
1255 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1256 	MSR_IA32_VMX_EXIT_CTLS,
1257 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1258 	MSR_IA32_VMX_ENTRY_CTLS,
1259 	MSR_IA32_VMX_MISC,
1260 	MSR_IA32_VMX_CR0_FIXED0,
1261 	MSR_IA32_VMX_CR0_FIXED1,
1262 	MSR_IA32_VMX_CR4_FIXED0,
1263 	MSR_IA32_VMX_CR4_FIXED1,
1264 	MSR_IA32_VMX_VMCS_ENUM,
1265 	MSR_IA32_VMX_PROCBASED_CTLS2,
1266 	MSR_IA32_VMX_EPT_VPID_CAP,
1267 	MSR_IA32_VMX_VMFUNC,
1268 
1269 	MSR_F10H_DECFG,
1270 	MSR_IA32_UCODE_REV,
1271 	MSR_IA32_ARCH_CAPABILITIES,
1272 };
1273 
1274 static unsigned int num_msr_based_features;
1275 
1276 static u64 kvm_get_arch_capabilities(void)
1277 {
1278 	u64 data = 0;
1279 
1280 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1281 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1282 
1283 	/*
1284 	 * If we're doing cache flushes (either "always" or "cond")
1285 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1286 	 * If an outer hypervisor is doing the cache flush for us
1287 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1288 	 * capability to the guest too, and if EPT is disabled we're not
1289 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1290 	 * require a nested hypervisor to do a flush of its own.
1291 	 */
1292 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1293 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1294 
1295 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1296 		data |= ARCH_CAP_RDCL_NO;
1297 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1298 		data |= ARCH_CAP_SSB_NO;
1299 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1300 		data |= ARCH_CAP_MDS_NO;
1301 
1302 	return data;
1303 }
1304 
1305 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1306 {
1307 	switch (msr->index) {
1308 	case MSR_IA32_ARCH_CAPABILITIES:
1309 		msr->data = kvm_get_arch_capabilities();
1310 		break;
1311 	case MSR_IA32_UCODE_REV:
1312 		rdmsrl_safe(msr->index, &msr->data);
1313 		break;
1314 	default:
1315 		if (kvm_x86_ops->get_msr_feature(msr))
1316 			return 1;
1317 	}
1318 	return 0;
1319 }
1320 
1321 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1322 {
1323 	struct kvm_msr_entry msr;
1324 	int r;
1325 
1326 	msr.index = index;
1327 	r = kvm_get_msr_feature(&msr);
1328 	if (r)
1329 		return r;
1330 
1331 	*data = msr.data;
1332 
1333 	return 0;
1334 }
1335 
1336 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1337 {
1338 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1339 		return false;
1340 
1341 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1342 		return false;
1343 
1344 	if (efer & (EFER_LME | EFER_LMA) &&
1345 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1346 		return false;
1347 
1348 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1349 		return false;
1350 
1351 	return true;
1352 
1353 }
1354 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1355 {
1356 	if (efer & efer_reserved_bits)
1357 		return false;
1358 
1359 	return __kvm_valid_efer(vcpu, efer);
1360 }
1361 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1362 
1363 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1364 {
1365 	u64 old_efer = vcpu->arch.efer;
1366 	u64 efer = msr_info->data;
1367 
1368 	if (efer & efer_reserved_bits)
1369 		return 1;
1370 
1371 	if (!msr_info->host_initiated) {
1372 		if (!__kvm_valid_efer(vcpu, efer))
1373 			return 1;
1374 
1375 		if (is_paging(vcpu) &&
1376 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1377 			return 1;
1378 	}
1379 
1380 	efer &= ~EFER_LMA;
1381 	efer |= vcpu->arch.efer & EFER_LMA;
1382 
1383 	kvm_x86_ops->set_efer(vcpu, efer);
1384 
1385 	/* Update reserved bits */
1386 	if ((efer ^ old_efer) & EFER_NX)
1387 		kvm_mmu_reset_context(vcpu);
1388 
1389 	return 0;
1390 }
1391 
1392 void kvm_enable_efer_bits(u64 mask)
1393 {
1394        efer_reserved_bits &= ~mask;
1395 }
1396 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1397 
1398 /*
1399  * Write @data into the MSR specified by @index.  Select MSR specific fault
1400  * checks are bypassed if @host_initiated is %true.
1401  * Returns 0 on success, non-0 otherwise.
1402  * Assumes vcpu_load() was already called.
1403  */
1404 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1405 			 bool host_initiated)
1406 {
1407 	struct msr_data msr;
1408 
1409 	switch (index) {
1410 	case MSR_FS_BASE:
1411 	case MSR_GS_BASE:
1412 	case MSR_KERNEL_GS_BASE:
1413 	case MSR_CSTAR:
1414 	case MSR_LSTAR:
1415 		if (is_noncanonical_address(data, vcpu))
1416 			return 1;
1417 		break;
1418 	case MSR_IA32_SYSENTER_EIP:
1419 	case MSR_IA32_SYSENTER_ESP:
1420 		/*
1421 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1422 		 * non-canonical address is written on Intel but not on
1423 		 * AMD (which ignores the top 32-bits, because it does
1424 		 * not implement 64-bit SYSENTER).
1425 		 *
1426 		 * 64-bit code should hence be able to write a non-canonical
1427 		 * value on AMD.  Making the address canonical ensures that
1428 		 * vmentry does not fail on Intel after writing a non-canonical
1429 		 * value, and that something deterministic happens if the guest
1430 		 * invokes 64-bit SYSENTER.
1431 		 */
1432 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1433 	}
1434 
1435 	msr.data = data;
1436 	msr.index = index;
1437 	msr.host_initiated = host_initiated;
1438 
1439 	return kvm_x86_ops->set_msr(vcpu, &msr);
1440 }
1441 
1442 /*
1443  * Read the MSR specified by @index into @data.  Select MSR specific fault
1444  * checks are bypassed if @host_initiated is %true.
1445  * Returns 0 on success, non-0 otherwise.
1446  * Assumes vcpu_load() was already called.
1447  */
1448 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1449 			 bool host_initiated)
1450 {
1451 	struct msr_data msr;
1452 	int ret;
1453 
1454 	msr.index = index;
1455 	msr.host_initiated = host_initiated;
1456 
1457 	ret = kvm_x86_ops->get_msr(vcpu, &msr);
1458 	if (!ret)
1459 		*data = msr.data;
1460 	return ret;
1461 }
1462 
1463 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1464 {
1465 	return __kvm_get_msr(vcpu, index, data, false);
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_get_msr);
1468 
1469 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1470 {
1471 	return __kvm_set_msr(vcpu, index, data, false);
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_set_msr);
1474 
1475 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1476 {
1477 	u32 ecx = kvm_rcx_read(vcpu);
1478 	u64 data;
1479 
1480 	if (kvm_get_msr(vcpu, ecx, &data)) {
1481 		trace_kvm_msr_read_ex(ecx);
1482 		kvm_inject_gp(vcpu, 0);
1483 		return 1;
1484 	}
1485 
1486 	trace_kvm_msr_read(ecx, data);
1487 
1488 	kvm_rax_write(vcpu, data & -1u);
1489 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1490 	return kvm_skip_emulated_instruction(vcpu);
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1493 
1494 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1495 {
1496 	u32 ecx = kvm_rcx_read(vcpu);
1497 	u64 data = kvm_read_edx_eax(vcpu);
1498 
1499 	if (kvm_set_msr(vcpu, ecx, data)) {
1500 		trace_kvm_msr_write_ex(ecx, data);
1501 		kvm_inject_gp(vcpu, 0);
1502 		return 1;
1503 	}
1504 
1505 	trace_kvm_msr_write(ecx, data);
1506 	return kvm_skip_emulated_instruction(vcpu);
1507 }
1508 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1509 
1510 /*
1511  * Adapt set_msr() to msr_io()'s calling convention
1512  */
1513 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1514 {
1515 	return __kvm_get_msr(vcpu, index, data, true);
1516 }
1517 
1518 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1519 {
1520 	return __kvm_set_msr(vcpu, index, *data, true);
1521 }
1522 
1523 #ifdef CONFIG_X86_64
1524 struct pvclock_gtod_data {
1525 	seqcount_t	seq;
1526 
1527 	struct { /* extract of a clocksource struct */
1528 		int vclock_mode;
1529 		u64	cycle_last;
1530 		u64	mask;
1531 		u32	mult;
1532 		u32	shift;
1533 	} clock;
1534 
1535 	u64		boot_ns;
1536 	u64		nsec_base;
1537 	u64		wall_time_sec;
1538 };
1539 
1540 static struct pvclock_gtod_data pvclock_gtod_data;
1541 
1542 static void update_pvclock_gtod(struct timekeeper *tk)
1543 {
1544 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1545 	u64 boot_ns;
1546 
1547 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1548 
1549 	write_seqcount_begin(&vdata->seq);
1550 
1551 	/* copy pvclock gtod data */
1552 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1553 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1554 	vdata->clock.mask		= tk->tkr_mono.mask;
1555 	vdata->clock.mult		= tk->tkr_mono.mult;
1556 	vdata->clock.shift		= tk->tkr_mono.shift;
1557 
1558 	vdata->boot_ns			= boot_ns;
1559 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1560 
1561 	vdata->wall_time_sec            = tk->xtime_sec;
1562 
1563 	write_seqcount_end(&vdata->seq);
1564 }
1565 #endif
1566 
1567 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1568 {
1569 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1570 	kvm_vcpu_kick(vcpu);
1571 }
1572 
1573 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1574 {
1575 	int version;
1576 	int r;
1577 	struct pvclock_wall_clock wc;
1578 	struct timespec64 boot;
1579 
1580 	if (!wall_clock)
1581 		return;
1582 
1583 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1584 	if (r)
1585 		return;
1586 
1587 	if (version & 1)
1588 		++version;  /* first time write, random junk */
1589 
1590 	++version;
1591 
1592 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1593 		return;
1594 
1595 	/*
1596 	 * The guest calculates current wall clock time by adding
1597 	 * system time (updated by kvm_guest_time_update below) to the
1598 	 * wall clock specified here.  guest system time equals host
1599 	 * system time for us, thus we must fill in host boot time here.
1600 	 */
1601 	getboottime64(&boot);
1602 
1603 	if (kvm->arch.kvmclock_offset) {
1604 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1605 		boot = timespec64_sub(boot, ts);
1606 	}
1607 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1608 	wc.nsec = boot.tv_nsec;
1609 	wc.version = version;
1610 
1611 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1612 
1613 	version++;
1614 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1615 }
1616 
1617 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1618 {
1619 	do_shl32_div32(dividend, divisor);
1620 	return dividend;
1621 }
1622 
1623 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1624 			       s8 *pshift, u32 *pmultiplier)
1625 {
1626 	uint64_t scaled64;
1627 	int32_t  shift = 0;
1628 	uint64_t tps64;
1629 	uint32_t tps32;
1630 
1631 	tps64 = base_hz;
1632 	scaled64 = scaled_hz;
1633 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1634 		tps64 >>= 1;
1635 		shift--;
1636 	}
1637 
1638 	tps32 = (uint32_t)tps64;
1639 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1640 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1641 			scaled64 >>= 1;
1642 		else
1643 			tps32 <<= 1;
1644 		shift++;
1645 	}
1646 
1647 	*pshift = shift;
1648 	*pmultiplier = div_frac(scaled64, tps32);
1649 }
1650 
1651 #ifdef CONFIG_X86_64
1652 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1653 #endif
1654 
1655 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1656 static unsigned long max_tsc_khz;
1657 
1658 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1659 {
1660 	u64 v = (u64)khz * (1000000 + ppm);
1661 	do_div(v, 1000000);
1662 	return v;
1663 }
1664 
1665 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1666 {
1667 	u64 ratio;
1668 
1669 	/* Guest TSC same frequency as host TSC? */
1670 	if (!scale) {
1671 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1672 		return 0;
1673 	}
1674 
1675 	/* TSC scaling supported? */
1676 	if (!kvm_has_tsc_control) {
1677 		if (user_tsc_khz > tsc_khz) {
1678 			vcpu->arch.tsc_catchup = 1;
1679 			vcpu->arch.tsc_always_catchup = 1;
1680 			return 0;
1681 		} else {
1682 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1683 			return -1;
1684 		}
1685 	}
1686 
1687 	/* TSC scaling required  - calculate ratio */
1688 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1689 				user_tsc_khz, tsc_khz);
1690 
1691 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1692 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1693 			            user_tsc_khz);
1694 		return -1;
1695 	}
1696 
1697 	vcpu->arch.tsc_scaling_ratio = ratio;
1698 	return 0;
1699 }
1700 
1701 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1702 {
1703 	u32 thresh_lo, thresh_hi;
1704 	int use_scaling = 0;
1705 
1706 	/* tsc_khz can be zero if TSC calibration fails */
1707 	if (user_tsc_khz == 0) {
1708 		/* set tsc_scaling_ratio to a safe value */
1709 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1710 		return -1;
1711 	}
1712 
1713 	/* Compute a scale to convert nanoseconds in TSC cycles */
1714 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1715 			   &vcpu->arch.virtual_tsc_shift,
1716 			   &vcpu->arch.virtual_tsc_mult);
1717 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1718 
1719 	/*
1720 	 * Compute the variation in TSC rate which is acceptable
1721 	 * within the range of tolerance and decide if the
1722 	 * rate being applied is within that bounds of the hardware
1723 	 * rate.  If so, no scaling or compensation need be done.
1724 	 */
1725 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1726 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1727 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1728 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1729 		use_scaling = 1;
1730 	}
1731 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1732 }
1733 
1734 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1735 {
1736 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1737 				      vcpu->arch.virtual_tsc_mult,
1738 				      vcpu->arch.virtual_tsc_shift);
1739 	tsc += vcpu->arch.this_tsc_write;
1740 	return tsc;
1741 }
1742 
1743 static inline int gtod_is_based_on_tsc(int mode)
1744 {
1745 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1746 }
1747 
1748 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1749 {
1750 #ifdef CONFIG_X86_64
1751 	bool vcpus_matched;
1752 	struct kvm_arch *ka = &vcpu->kvm->arch;
1753 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1754 
1755 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1756 			 atomic_read(&vcpu->kvm->online_vcpus));
1757 
1758 	/*
1759 	 * Once the masterclock is enabled, always perform request in
1760 	 * order to update it.
1761 	 *
1762 	 * In order to enable masterclock, the host clocksource must be TSC
1763 	 * and the vcpus need to have matched TSCs.  When that happens,
1764 	 * perform request to enable masterclock.
1765 	 */
1766 	if (ka->use_master_clock ||
1767 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1768 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1769 
1770 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1771 			    atomic_read(&vcpu->kvm->online_vcpus),
1772 		            ka->use_master_clock, gtod->clock.vclock_mode);
1773 #endif
1774 }
1775 
1776 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1777 {
1778 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1779 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1780 }
1781 
1782 /*
1783  * Multiply tsc by a fixed point number represented by ratio.
1784  *
1785  * The most significant 64-N bits (mult) of ratio represent the
1786  * integral part of the fixed point number; the remaining N bits
1787  * (frac) represent the fractional part, ie. ratio represents a fixed
1788  * point number (mult + frac * 2^(-N)).
1789  *
1790  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1791  */
1792 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1793 {
1794 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1795 }
1796 
1797 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1798 {
1799 	u64 _tsc = tsc;
1800 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1801 
1802 	if (ratio != kvm_default_tsc_scaling_ratio)
1803 		_tsc = __scale_tsc(ratio, tsc);
1804 
1805 	return _tsc;
1806 }
1807 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1808 
1809 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1810 {
1811 	u64 tsc;
1812 
1813 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1814 
1815 	return target_tsc - tsc;
1816 }
1817 
1818 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1819 {
1820 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1821 
1822 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1823 }
1824 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1825 
1826 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1827 {
1828 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1829 }
1830 
1831 static inline bool kvm_check_tsc_unstable(void)
1832 {
1833 #ifdef CONFIG_X86_64
1834 	/*
1835 	 * TSC is marked unstable when we're running on Hyper-V,
1836 	 * 'TSC page' clocksource is good.
1837 	 */
1838 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1839 		return false;
1840 #endif
1841 	return check_tsc_unstable();
1842 }
1843 
1844 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1845 {
1846 	struct kvm *kvm = vcpu->kvm;
1847 	u64 offset, ns, elapsed;
1848 	unsigned long flags;
1849 	bool matched;
1850 	bool already_matched;
1851 	u64 data = msr->data;
1852 	bool synchronizing = false;
1853 
1854 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1855 	offset = kvm_compute_tsc_offset(vcpu, data);
1856 	ns = ktime_get_boottime_ns();
1857 	elapsed = ns - kvm->arch.last_tsc_nsec;
1858 
1859 	if (vcpu->arch.virtual_tsc_khz) {
1860 		if (data == 0 && msr->host_initiated) {
1861 			/*
1862 			 * detection of vcpu initialization -- need to sync
1863 			 * with other vCPUs. This particularly helps to keep
1864 			 * kvm_clock stable after CPU hotplug
1865 			 */
1866 			synchronizing = true;
1867 		} else {
1868 			u64 tsc_exp = kvm->arch.last_tsc_write +
1869 						nsec_to_cycles(vcpu, elapsed);
1870 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1871 			/*
1872 			 * Special case: TSC write with a small delta (1 second)
1873 			 * of virtual cycle time against real time is
1874 			 * interpreted as an attempt to synchronize the CPU.
1875 			 */
1876 			synchronizing = data < tsc_exp + tsc_hz &&
1877 					data + tsc_hz > tsc_exp;
1878 		}
1879 	}
1880 
1881 	/*
1882 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1883 	 * TSC, we add elapsed time in this computation.  We could let the
1884 	 * compensation code attempt to catch up if we fall behind, but
1885 	 * it's better to try to match offsets from the beginning.
1886          */
1887 	if (synchronizing &&
1888 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1889 		if (!kvm_check_tsc_unstable()) {
1890 			offset = kvm->arch.cur_tsc_offset;
1891 		} else {
1892 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1893 			data += delta;
1894 			offset = kvm_compute_tsc_offset(vcpu, data);
1895 		}
1896 		matched = true;
1897 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1898 	} else {
1899 		/*
1900 		 * We split periods of matched TSC writes into generations.
1901 		 * For each generation, we track the original measured
1902 		 * nanosecond time, offset, and write, so if TSCs are in
1903 		 * sync, we can match exact offset, and if not, we can match
1904 		 * exact software computation in compute_guest_tsc()
1905 		 *
1906 		 * These values are tracked in kvm->arch.cur_xxx variables.
1907 		 */
1908 		kvm->arch.cur_tsc_generation++;
1909 		kvm->arch.cur_tsc_nsec = ns;
1910 		kvm->arch.cur_tsc_write = data;
1911 		kvm->arch.cur_tsc_offset = offset;
1912 		matched = false;
1913 	}
1914 
1915 	/*
1916 	 * We also track th most recent recorded KHZ, write and time to
1917 	 * allow the matching interval to be extended at each write.
1918 	 */
1919 	kvm->arch.last_tsc_nsec = ns;
1920 	kvm->arch.last_tsc_write = data;
1921 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1922 
1923 	vcpu->arch.last_guest_tsc = data;
1924 
1925 	/* Keep track of which generation this VCPU has synchronized to */
1926 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1927 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1928 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1929 
1930 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1931 		update_ia32_tsc_adjust_msr(vcpu, offset);
1932 
1933 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1934 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1935 
1936 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1937 	if (!matched) {
1938 		kvm->arch.nr_vcpus_matched_tsc = 0;
1939 	} else if (!already_matched) {
1940 		kvm->arch.nr_vcpus_matched_tsc++;
1941 	}
1942 
1943 	kvm_track_tsc_matching(vcpu);
1944 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1945 }
1946 
1947 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1948 
1949 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1950 					   s64 adjustment)
1951 {
1952 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1953 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1954 }
1955 
1956 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1957 {
1958 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1959 		WARN_ON(adjustment < 0);
1960 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1961 	adjust_tsc_offset_guest(vcpu, adjustment);
1962 }
1963 
1964 #ifdef CONFIG_X86_64
1965 
1966 static u64 read_tsc(void)
1967 {
1968 	u64 ret = (u64)rdtsc_ordered();
1969 	u64 last = pvclock_gtod_data.clock.cycle_last;
1970 
1971 	if (likely(ret >= last))
1972 		return ret;
1973 
1974 	/*
1975 	 * GCC likes to generate cmov here, but this branch is extremely
1976 	 * predictable (it's just a function of time and the likely is
1977 	 * very likely) and there's a data dependence, so force GCC
1978 	 * to generate a branch instead.  I don't barrier() because
1979 	 * we don't actually need a barrier, and if this function
1980 	 * ever gets inlined it will generate worse code.
1981 	 */
1982 	asm volatile ("");
1983 	return last;
1984 }
1985 
1986 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1987 {
1988 	long v;
1989 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1990 	u64 tsc_pg_val;
1991 
1992 	switch (gtod->clock.vclock_mode) {
1993 	case VCLOCK_HVCLOCK:
1994 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1995 						  tsc_timestamp);
1996 		if (tsc_pg_val != U64_MAX) {
1997 			/* TSC page valid */
1998 			*mode = VCLOCK_HVCLOCK;
1999 			v = (tsc_pg_val - gtod->clock.cycle_last) &
2000 				gtod->clock.mask;
2001 		} else {
2002 			/* TSC page invalid */
2003 			*mode = VCLOCK_NONE;
2004 		}
2005 		break;
2006 	case VCLOCK_TSC:
2007 		*mode = VCLOCK_TSC;
2008 		*tsc_timestamp = read_tsc();
2009 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
2010 			gtod->clock.mask;
2011 		break;
2012 	default:
2013 		*mode = VCLOCK_NONE;
2014 	}
2015 
2016 	if (*mode == VCLOCK_NONE)
2017 		*tsc_timestamp = v = 0;
2018 
2019 	return v * gtod->clock.mult;
2020 }
2021 
2022 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2023 {
2024 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2025 	unsigned long seq;
2026 	int mode;
2027 	u64 ns;
2028 
2029 	do {
2030 		seq = read_seqcount_begin(&gtod->seq);
2031 		ns = gtod->nsec_base;
2032 		ns += vgettsc(tsc_timestamp, &mode);
2033 		ns >>= gtod->clock.shift;
2034 		ns += gtod->boot_ns;
2035 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2036 	*t = ns;
2037 
2038 	return mode;
2039 }
2040 
2041 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2042 {
2043 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2044 	unsigned long seq;
2045 	int mode;
2046 	u64 ns;
2047 
2048 	do {
2049 		seq = read_seqcount_begin(&gtod->seq);
2050 		ts->tv_sec = gtod->wall_time_sec;
2051 		ns = gtod->nsec_base;
2052 		ns += vgettsc(tsc_timestamp, &mode);
2053 		ns >>= gtod->clock.shift;
2054 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2055 
2056 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2057 	ts->tv_nsec = ns;
2058 
2059 	return mode;
2060 }
2061 
2062 /* returns true if host is using TSC based clocksource */
2063 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2064 {
2065 	/* checked again under seqlock below */
2066 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2067 		return false;
2068 
2069 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2070 						      tsc_timestamp));
2071 }
2072 
2073 /* returns true if host is using TSC based clocksource */
2074 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2075 					   u64 *tsc_timestamp)
2076 {
2077 	/* checked again under seqlock below */
2078 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2079 		return false;
2080 
2081 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2082 }
2083 #endif
2084 
2085 /*
2086  *
2087  * Assuming a stable TSC across physical CPUS, and a stable TSC
2088  * across virtual CPUs, the following condition is possible.
2089  * Each numbered line represents an event visible to both
2090  * CPUs at the next numbered event.
2091  *
2092  * "timespecX" represents host monotonic time. "tscX" represents
2093  * RDTSC value.
2094  *
2095  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2096  *
2097  * 1.  read timespec0,tsc0
2098  * 2.					| timespec1 = timespec0 + N
2099  * 					| tsc1 = tsc0 + M
2100  * 3. transition to guest		| transition to guest
2101  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2102  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2103  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2104  *
2105  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2106  *
2107  * 	- ret0 < ret1
2108  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2109  *		...
2110  *	- 0 < N - M => M < N
2111  *
2112  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2113  * always the case (the difference between two distinct xtime instances
2114  * might be smaller then the difference between corresponding TSC reads,
2115  * when updating guest vcpus pvclock areas).
2116  *
2117  * To avoid that problem, do not allow visibility of distinct
2118  * system_timestamp/tsc_timestamp values simultaneously: use a master
2119  * copy of host monotonic time values. Update that master copy
2120  * in lockstep.
2121  *
2122  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2123  *
2124  */
2125 
2126 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2127 {
2128 #ifdef CONFIG_X86_64
2129 	struct kvm_arch *ka = &kvm->arch;
2130 	int vclock_mode;
2131 	bool host_tsc_clocksource, vcpus_matched;
2132 
2133 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2134 			atomic_read(&kvm->online_vcpus));
2135 
2136 	/*
2137 	 * If the host uses TSC clock, then passthrough TSC as stable
2138 	 * to the guest.
2139 	 */
2140 	host_tsc_clocksource = kvm_get_time_and_clockread(
2141 					&ka->master_kernel_ns,
2142 					&ka->master_cycle_now);
2143 
2144 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2145 				&& !ka->backwards_tsc_observed
2146 				&& !ka->boot_vcpu_runs_old_kvmclock;
2147 
2148 	if (ka->use_master_clock)
2149 		atomic_set(&kvm_guest_has_master_clock, 1);
2150 
2151 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2152 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2153 					vcpus_matched);
2154 #endif
2155 }
2156 
2157 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2158 {
2159 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2160 }
2161 
2162 static void kvm_gen_update_masterclock(struct kvm *kvm)
2163 {
2164 #ifdef CONFIG_X86_64
2165 	int i;
2166 	struct kvm_vcpu *vcpu;
2167 	struct kvm_arch *ka = &kvm->arch;
2168 
2169 	spin_lock(&ka->pvclock_gtod_sync_lock);
2170 	kvm_make_mclock_inprogress_request(kvm);
2171 	/* no guest entries from this point */
2172 	pvclock_update_vm_gtod_copy(kvm);
2173 
2174 	kvm_for_each_vcpu(i, vcpu, kvm)
2175 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2176 
2177 	/* guest entries allowed */
2178 	kvm_for_each_vcpu(i, vcpu, kvm)
2179 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2180 
2181 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2182 #endif
2183 }
2184 
2185 u64 get_kvmclock_ns(struct kvm *kvm)
2186 {
2187 	struct kvm_arch *ka = &kvm->arch;
2188 	struct pvclock_vcpu_time_info hv_clock;
2189 	u64 ret;
2190 
2191 	spin_lock(&ka->pvclock_gtod_sync_lock);
2192 	if (!ka->use_master_clock) {
2193 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2194 		return ktime_get_boottime_ns() + ka->kvmclock_offset;
2195 	}
2196 
2197 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2198 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2199 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2200 
2201 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2202 	get_cpu();
2203 
2204 	if (__this_cpu_read(cpu_tsc_khz)) {
2205 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2206 				   &hv_clock.tsc_shift,
2207 				   &hv_clock.tsc_to_system_mul);
2208 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2209 	} else
2210 		ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2211 
2212 	put_cpu();
2213 
2214 	return ret;
2215 }
2216 
2217 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2218 {
2219 	struct kvm_vcpu_arch *vcpu = &v->arch;
2220 	struct pvclock_vcpu_time_info guest_hv_clock;
2221 
2222 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2223 		&guest_hv_clock, sizeof(guest_hv_clock))))
2224 		return;
2225 
2226 	/* This VCPU is paused, but it's legal for a guest to read another
2227 	 * VCPU's kvmclock, so we really have to follow the specification where
2228 	 * it says that version is odd if data is being modified, and even after
2229 	 * it is consistent.
2230 	 *
2231 	 * Version field updates must be kept separate.  This is because
2232 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2233 	 * writes within a string instruction are weakly ordered.  So there
2234 	 * are three writes overall.
2235 	 *
2236 	 * As a small optimization, only write the version field in the first
2237 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2238 	 * version field is the first in the struct.
2239 	 */
2240 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2241 
2242 	if (guest_hv_clock.version & 1)
2243 		++guest_hv_clock.version;  /* first time write, random junk */
2244 
2245 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2246 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2247 				&vcpu->hv_clock,
2248 				sizeof(vcpu->hv_clock.version));
2249 
2250 	smp_wmb();
2251 
2252 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2253 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2254 
2255 	if (vcpu->pvclock_set_guest_stopped_request) {
2256 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2257 		vcpu->pvclock_set_guest_stopped_request = false;
2258 	}
2259 
2260 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2261 
2262 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2263 				&vcpu->hv_clock,
2264 				sizeof(vcpu->hv_clock));
2265 
2266 	smp_wmb();
2267 
2268 	vcpu->hv_clock.version++;
2269 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2270 				&vcpu->hv_clock,
2271 				sizeof(vcpu->hv_clock.version));
2272 }
2273 
2274 static int kvm_guest_time_update(struct kvm_vcpu *v)
2275 {
2276 	unsigned long flags, tgt_tsc_khz;
2277 	struct kvm_vcpu_arch *vcpu = &v->arch;
2278 	struct kvm_arch *ka = &v->kvm->arch;
2279 	s64 kernel_ns;
2280 	u64 tsc_timestamp, host_tsc;
2281 	u8 pvclock_flags;
2282 	bool use_master_clock;
2283 
2284 	kernel_ns = 0;
2285 	host_tsc = 0;
2286 
2287 	/*
2288 	 * If the host uses TSC clock, then passthrough TSC as stable
2289 	 * to the guest.
2290 	 */
2291 	spin_lock(&ka->pvclock_gtod_sync_lock);
2292 	use_master_clock = ka->use_master_clock;
2293 	if (use_master_clock) {
2294 		host_tsc = ka->master_cycle_now;
2295 		kernel_ns = ka->master_kernel_ns;
2296 	}
2297 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2298 
2299 	/* Keep irq disabled to prevent changes to the clock */
2300 	local_irq_save(flags);
2301 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2302 	if (unlikely(tgt_tsc_khz == 0)) {
2303 		local_irq_restore(flags);
2304 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2305 		return 1;
2306 	}
2307 	if (!use_master_clock) {
2308 		host_tsc = rdtsc();
2309 		kernel_ns = ktime_get_boottime_ns();
2310 	}
2311 
2312 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2313 
2314 	/*
2315 	 * We may have to catch up the TSC to match elapsed wall clock
2316 	 * time for two reasons, even if kvmclock is used.
2317 	 *   1) CPU could have been running below the maximum TSC rate
2318 	 *   2) Broken TSC compensation resets the base at each VCPU
2319 	 *      entry to avoid unknown leaps of TSC even when running
2320 	 *      again on the same CPU.  This may cause apparent elapsed
2321 	 *      time to disappear, and the guest to stand still or run
2322 	 *	very slowly.
2323 	 */
2324 	if (vcpu->tsc_catchup) {
2325 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2326 		if (tsc > tsc_timestamp) {
2327 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2328 			tsc_timestamp = tsc;
2329 		}
2330 	}
2331 
2332 	local_irq_restore(flags);
2333 
2334 	/* With all the info we got, fill in the values */
2335 
2336 	if (kvm_has_tsc_control)
2337 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2338 
2339 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2340 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2341 				   &vcpu->hv_clock.tsc_shift,
2342 				   &vcpu->hv_clock.tsc_to_system_mul);
2343 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2344 	}
2345 
2346 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2347 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2348 	vcpu->last_guest_tsc = tsc_timestamp;
2349 
2350 	/* If the host uses TSC clocksource, then it is stable */
2351 	pvclock_flags = 0;
2352 	if (use_master_clock)
2353 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2354 
2355 	vcpu->hv_clock.flags = pvclock_flags;
2356 
2357 	if (vcpu->pv_time_enabled)
2358 		kvm_setup_pvclock_page(v);
2359 	if (v == kvm_get_vcpu(v->kvm, 0))
2360 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2361 	return 0;
2362 }
2363 
2364 /*
2365  * kvmclock updates which are isolated to a given vcpu, such as
2366  * vcpu->cpu migration, should not allow system_timestamp from
2367  * the rest of the vcpus to remain static. Otherwise ntp frequency
2368  * correction applies to one vcpu's system_timestamp but not
2369  * the others.
2370  *
2371  * So in those cases, request a kvmclock update for all vcpus.
2372  * We need to rate-limit these requests though, as they can
2373  * considerably slow guests that have a large number of vcpus.
2374  * The time for a remote vcpu to update its kvmclock is bound
2375  * by the delay we use to rate-limit the updates.
2376  */
2377 
2378 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2379 
2380 static void kvmclock_update_fn(struct work_struct *work)
2381 {
2382 	int i;
2383 	struct delayed_work *dwork = to_delayed_work(work);
2384 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2385 					   kvmclock_update_work);
2386 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2387 	struct kvm_vcpu *vcpu;
2388 
2389 	kvm_for_each_vcpu(i, vcpu, kvm) {
2390 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2391 		kvm_vcpu_kick(vcpu);
2392 	}
2393 }
2394 
2395 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2396 {
2397 	struct kvm *kvm = v->kvm;
2398 
2399 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2400 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2401 					KVMCLOCK_UPDATE_DELAY);
2402 }
2403 
2404 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2405 
2406 static void kvmclock_sync_fn(struct work_struct *work)
2407 {
2408 	struct delayed_work *dwork = to_delayed_work(work);
2409 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2410 					   kvmclock_sync_work);
2411 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2412 
2413 	if (!kvmclock_periodic_sync)
2414 		return;
2415 
2416 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2417 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2418 					KVMCLOCK_SYNC_PERIOD);
2419 }
2420 
2421 /*
2422  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2423  */
2424 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2425 {
2426 	/* McStatusWrEn enabled? */
2427 	if (guest_cpuid_is_amd(vcpu))
2428 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2429 
2430 	return false;
2431 }
2432 
2433 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2434 {
2435 	u64 mcg_cap = vcpu->arch.mcg_cap;
2436 	unsigned bank_num = mcg_cap & 0xff;
2437 	u32 msr = msr_info->index;
2438 	u64 data = msr_info->data;
2439 
2440 	switch (msr) {
2441 	case MSR_IA32_MCG_STATUS:
2442 		vcpu->arch.mcg_status = data;
2443 		break;
2444 	case MSR_IA32_MCG_CTL:
2445 		if (!(mcg_cap & MCG_CTL_P) &&
2446 		    (data || !msr_info->host_initiated))
2447 			return 1;
2448 		if (data != 0 && data != ~(u64)0)
2449 			return 1;
2450 		vcpu->arch.mcg_ctl = data;
2451 		break;
2452 	default:
2453 		if (msr >= MSR_IA32_MC0_CTL &&
2454 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2455 			u32 offset = msr - MSR_IA32_MC0_CTL;
2456 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2457 			 * some Linux kernels though clear bit 10 in bank 4 to
2458 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2459 			 * this to avoid an uncatched #GP in the guest
2460 			 */
2461 			if ((offset & 0x3) == 0 &&
2462 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2463 				return -1;
2464 
2465 			/* MCi_STATUS */
2466 			if (!msr_info->host_initiated &&
2467 			    (offset & 0x3) == 1 && data != 0) {
2468 				if (!can_set_mci_status(vcpu))
2469 					return -1;
2470 			}
2471 
2472 			vcpu->arch.mce_banks[offset] = data;
2473 			break;
2474 		}
2475 		return 1;
2476 	}
2477 	return 0;
2478 }
2479 
2480 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2481 {
2482 	struct kvm *kvm = vcpu->kvm;
2483 	int lm = is_long_mode(vcpu);
2484 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2485 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2486 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2487 		: kvm->arch.xen_hvm_config.blob_size_32;
2488 	u32 page_num = data & ~PAGE_MASK;
2489 	u64 page_addr = data & PAGE_MASK;
2490 	u8 *page;
2491 	int r;
2492 
2493 	r = -E2BIG;
2494 	if (page_num >= blob_size)
2495 		goto out;
2496 	r = -ENOMEM;
2497 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2498 	if (IS_ERR(page)) {
2499 		r = PTR_ERR(page);
2500 		goto out;
2501 	}
2502 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2503 		goto out_free;
2504 	r = 0;
2505 out_free:
2506 	kfree(page);
2507 out:
2508 	return r;
2509 }
2510 
2511 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2512 {
2513 	gpa_t gpa = data & ~0x3f;
2514 
2515 	/* Bits 3:5 are reserved, Should be zero */
2516 	if (data & 0x38)
2517 		return 1;
2518 
2519 	vcpu->arch.apf.msr_val = data;
2520 
2521 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2522 		kvm_clear_async_pf_completion_queue(vcpu);
2523 		kvm_async_pf_hash_reset(vcpu);
2524 		return 0;
2525 	}
2526 
2527 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2528 					sizeof(u32)))
2529 		return 1;
2530 
2531 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2532 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2533 	kvm_async_pf_wakeup_all(vcpu);
2534 	return 0;
2535 }
2536 
2537 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2538 {
2539 	vcpu->arch.pv_time_enabled = false;
2540 }
2541 
2542 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2543 {
2544 	++vcpu->stat.tlb_flush;
2545 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2546 }
2547 
2548 static void record_steal_time(struct kvm_vcpu *vcpu)
2549 {
2550 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2551 		return;
2552 
2553 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2554 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2555 		return;
2556 
2557 	/*
2558 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2559 	 * expensive IPIs.
2560 	 */
2561 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2562 		vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2563 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2564 		kvm_vcpu_flush_tlb(vcpu, false);
2565 
2566 	if (vcpu->arch.st.steal.version & 1)
2567 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2568 
2569 	vcpu->arch.st.steal.version += 1;
2570 
2571 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2572 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2573 
2574 	smp_wmb();
2575 
2576 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2577 		vcpu->arch.st.last_steal;
2578 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2579 
2580 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2581 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2582 
2583 	smp_wmb();
2584 
2585 	vcpu->arch.st.steal.version += 1;
2586 
2587 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2588 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2589 }
2590 
2591 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2592 {
2593 	bool pr = false;
2594 	u32 msr = msr_info->index;
2595 	u64 data = msr_info->data;
2596 
2597 	switch (msr) {
2598 	case MSR_AMD64_NB_CFG:
2599 	case MSR_IA32_UCODE_WRITE:
2600 	case MSR_VM_HSAVE_PA:
2601 	case MSR_AMD64_PATCH_LOADER:
2602 	case MSR_AMD64_BU_CFG2:
2603 	case MSR_AMD64_DC_CFG:
2604 	case MSR_F15H_EX_CFG:
2605 		break;
2606 
2607 	case MSR_IA32_UCODE_REV:
2608 		if (msr_info->host_initiated)
2609 			vcpu->arch.microcode_version = data;
2610 		break;
2611 	case MSR_IA32_ARCH_CAPABILITIES:
2612 		if (!msr_info->host_initiated)
2613 			return 1;
2614 		vcpu->arch.arch_capabilities = data;
2615 		break;
2616 	case MSR_EFER:
2617 		return set_efer(vcpu, msr_info);
2618 	case MSR_K7_HWCR:
2619 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2620 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2621 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2622 
2623 		/* Handle McStatusWrEn */
2624 		if (data == BIT_ULL(18)) {
2625 			vcpu->arch.msr_hwcr = data;
2626 		} else if (data != 0) {
2627 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2628 				    data);
2629 			return 1;
2630 		}
2631 		break;
2632 	case MSR_FAM10H_MMIO_CONF_BASE:
2633 		if (data != 0) {
2634 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2635 				    "0x%llx\n", data);
2636 			return 1;
2637 		}
2638 		break;
2639 	case MSR_IA32_DEBUGCTLMSR:
2640 		if (!data) {
2641 			/* We support the non-activated case already */
2642 			break;
2643 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2644 			/* Values other than LBR and BTF are vendor-specific,
2645 			   thus reserved and should throw a #GP */
2646 			return 1;
2647 		}
2648 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2649 			    __func__, data);
2650 		break;
2651 	case 0x200 ... 0x2ff:
2652 		return kvm_mtrr_set_msr(vcpu, msr, data);
2653 	case MSR_IA32_APICBASE:
2654 		return kvm_set_apic_base(vcpu, msr_info);
2655 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2656 		return kvm_x2apic_msr_write(vcpu, msr, data);
2657 	case MSR_IA32_TSCDEADLINE:
2658 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2659 		break;
2660 	case MSR_IA32_TSC_ADJUST:
2661 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2662 			if (!msr_info->host_initiated) {
2663 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2664 				adjust_tsc_offset_guest(vcpu, adj);
2665 			}
2666 			vcpu->arch.ia32_tsc_adjust_msr = data;
2667 		}
2668 		break;
2669 	case MSR_IA32_MISC_ENABLE:
2670 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2671 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2672 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2673 				return 1;
2674 			vcpu->arch.ia32_misc_enable_msr = data;
2675 			kvm_update_cpuid(vcpu);
2676 		} else {
2677 			vcpu->arch.ia32_misc_enable_msr = data;
2678 		}
2679 		break;
2680 	case MSR_IA32_SMBASE:
2681 		if (!msr_info->host_initiated)
2682 			return 1;
2683 		vcpu->arch.smbase = data;
2684 		break;
2685 	case MSR_IA32_POWER_CTL:
2686 		vcpu->arch.msr_ia32_power_ctl = data;
2687 		break;
2688 	case MSR_IA32_TSC:
2689 		kvm_write_tsc(vcpu, msr_info);
2690 		break;
2691 	case MSR_SMI_COUNT:
2692 		if (!msr_info->host_initiated)
2693 			return 1;
2694 		vcpu->arch.smi_count = data;
2695 		break;
2696 	case MSR_KVM_WALL_CLOCK_NEW:
2697 	case MSR_KVM_WALL_CLOCK:
2698 		vcpu->kvm->arch.wall_clock = data;
2699 		kvm_write_wall_clock(vcpu->kvm, data);
2700 		break;
2701 	case MSR_KVM_SYSTEM_TIME_NEW:
2702 	case MSR_KVM_SYSTEM_TIME: {
2703 		struct kvm_arch *ka = &vcpu->kvm->arch;
2704 
2705 		kvmclock_reset(vcpu);
2706 
2707 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2708 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2709 
2710 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2711 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2712 
2713 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2714 		}
2715 
2716 		vcpu->arch.time = data;
2717 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2718 
2719 		/* we verify if the enable bit is set... */
2720 		if (!(data & 1))
2721 			break;
2722 
2723 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2724 		     &vcpu->arch.pv_time, data & ~1ULL,
2725 		     sizeof(struct pvclock_vcpu_time_info)))
2726 			vcpu->arch.pv_time_enabled = false;
2727 		else
2728 			vcpu->arch.pv_time_enabled = true;
2729 
2730 		break;
2731 	}
2732 	case MSR_KVM_ASYNC_PF_EN:
2733 		if (kvm_pv_enable_async_pf(vcpu, data))
2734 			return 1;
2735 		break;
2736 	case MSR_KVM_STEAL_TIME:
2737 
2738 		if (unlikely(!sched_info_on()))
2739 			return 1;
2740 
2741 		if (data & KVM_STEAL_RESERVED_MASK)
2742 			return 1;
2743 
2744 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2745 						data & KVM_STEAL_VALID_BITS,
2746 						sizeof(struct kvm_steal_time)))
2747 			return 1;
2748 
2749 		vcpu->arch.st.msr_val = data;
2750 
2751 		if (!(data & KVM_MSR_ENABLED))
2752 			break;
2753 
2754 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2755 
2756 		break;
2757 	case MSR_KVM_PV_EOI_EN:
2758 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2759 			return 1;
2760 		break;
2761 
2762 	case MSR_KVM_POLL_CONTROL:
2763 		/* only enable bit supported */
2764 		if (data & (-1ULL << 1))
2765 			return 1;
2766 
2767 		vcpu->arch.msr_kvm_poll_control = data;
2768 		break;
2769 
2770 	case MSR_IA32_MCG_CTL:
2771 	case MSR_IA32_MCG_STATUS:
2772 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2773 		return set_msr_mce(vcpu, msr_info);
2774 
2775 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2776 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2777 		pr = true; /* fall through */
2778 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2779 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2780 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2781 			return kvm_pmu_set_msr(vcpu, msr_info);
2782 
2783 		if (pr || data != 0)
2784 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2785 				    "0x%x data 0x%llx\n", msr, data);
2786 		break;
2787 	case MSR_K7_CLK_CTL:
2788 		/*
2789 		 * Ignore all writes to this no longer documented MSR.
2790 		 * Writes are only relevant for old K7 processors,
2791 		 * all pre-dating SVM, but a recommended workaround from
2792 		 * AMD for these chips. It is possible to specify the
2793 		 * affected processor models on the command line, hence
2794 		 * the need to ignore the workaround.
2795 		 */
2796 		break;
2797 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2798 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2799 	case HV_X64_MSR_CRASH_CTL:
2800 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2801 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2802 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2803 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2804 		return kvm_hv_set_msr_common(vcpu, msr, data,
2805 					     msr_info->host_initiated);
2806 	case MSR_IA32_BBL_CR_CTL3:
2807 		/* Drop writes to this legacy MSR -- see rdmsr
2808 		 * counterpart for further detail.
2809 		 */
2810 		if (report_ignored_msrs)
2811 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2812 				msr, data);
2813 		break;
2814 	case MSR_AMD64_OSVW_ID_LENGTH:
2815 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2816 			return 1;
2817 		vcpu->arch.osvw.length = data;
2818 		break;
2819 	case MSR_AMD64_OSVW_STATUS:
2820 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2821 			return 1;
2822 		vcpu->arch.osvw.status = data;
2823 		break;
2824 	case MSR_PLATFORM_INFO:
2825 		if (!msr_info->host_initiated ||
2826 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2827 		     cpuid_fault_enabled(vcpu)))
2828 			return 1;
2829 		vcpu->arch.msr_platform_info = data;
2830 		break;
2831 	case MSR_MISC_FEATURES_ENABLES:
2832 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2833 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2834 		     !supports_cpuid_fault(vcpu)))
2835 			return 1;
2836 		vcpu->arch.msr_misc_features_enables = data;
2837 		break;
2838 	default:
2839 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2840 			return xen_hvm_config(vcpu, data);
2841 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2842 			return kvm_pmu_set_msr(vcpu, msr_info);
2843 		if (!ignore_msrs) {
2844 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2845 				    msr, data);
2846 			return 1;
2847 		} else {
2848 			if (report_ignored_msrs)
2849 				vcpu_unimpl(vcpu,
2850 					"ignored wrmsr: 0x%x data 0x%llx\n",
2851 					msr, data);
2852 			break;
2853 		}
2854 	}
2855 	return 0;
2856 }
2857 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2858 
2859 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2860 {
2861 	u64 data;
2862 	u64 mcg_cap = vcpu->arch.mcg_cap;
2863 	unsigned bank_num = mcg_cap & 0xff;
2864 
2865 	switch (msr) {
2866 	case MSR_IA32_P5_MC_ADDR:
2867 	case MSR_IA32_P5_MC_TYPE:
2868 		data = 0;
2869 		break;
2870 	case MSR_IA32_MCG_CAP:
2871 		data = vcpu->arch.mcg_cap;
2872 		break;
2873 	case MSR_IA32_MCG_CTL:
2874 		if (!(mcg_cap & MCG_CTL_P) && !host)
2875 			return 1;
2876 		data = vcpu->arch.mcg_ctl;
2877 		break;
2878 	case MSR_IA32_MCG_STATUS:
2879 		data = vcpu->arch.mcg_status;
2880 		break;
2881 	default:
2882 		if (msr >= MSR_IA32_MC0_CTL &&
2883 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2884 			u32 offset = msr - MSR_IA32_MC0_CTL;
2885 			data = vcpu->arch.mce_banks[offset];
2886 			break;
2887 		}
2888 		return 1;
2889 	}
2890 	*pdata = data;
2891 	return 0;
2892 }
2893 
2894 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2895 {
2896 	switch (msr_info->index) {
2897 	case MSR_IA32_PLATFORM_ID:
2898 	case MSR_IA32_EBL_CR_POWERON:
2899 	case MSR_IA32_DEBUGCTLMSR:
2900 	case MSR_IA32_LASTBRANCHFROMIP:
2901 	case MSR_IA32_LASTBRANCHTOIP:
2902 	case MSR_IA32_LASTINTFROMIP:
2903 	case MSR_IA32_LASTINTTOIP:
2904 	case MSR_K8_SYSCFG:
2905 	case MSR_K8_TSEG_ADDR:
2906 	case MSR_K8_TSEG_MASK:
2907 	case MSR_VM_HSAVE_PA:
2908 	case MSR_K8_INT_PENDING_MSG:
2909 	case MSR_AMD64_NB_CFG:
2910 	case MSR_FAM10H_MMIO_CONF_BASE:
2911 	case MSR_AMD64_BU_CFG2:
2912 	case MSR_IA32_PERF_CTL:
2913 	case MSR_AMD64_DC_CFG:
2914 	case MSR_F15H_EX_CFG:
2915 		msr_info->data = 0;
2916 		break;
2917 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2918 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2919 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2920 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2921 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2922 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2923 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2924 		msr_info->data = 0;
2925 		break;
2926 	case MSR_IA32_UCODE_REV:
2927 		msr_info->data = vcpu->arch.microcode_version;
2928 		break;
2929 	case MSR_IA32_ARCH_CAPABILITIES:
2930 		if (!msr_info->host_initiated &&
2931 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2932 			return 1;
2933 		msr_info->data = vcpu->arch.arch_capabilities;
2934 		break;
2935 	case MSR_IA32_POWER_CTL:
2936 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2937 		break;
2938 	case MSR_IA32_TSC:
2939 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2940 		break;
2941 	case MSR_MTRRcap:
2942 	case 0x200 ... 0x2ff:
2943 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2944 	case 0xcd: /* fsb frequency */
2945 		msr_info->data = 3;
2946 		break;
2947 		/*
2948 		 * MSR_EBC_FREQUENCY_ID
2949 		 * Conservative value valid for even the basic CPU models.
2950 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2951 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2952 		 * and 266MHz for model 3, or 4. Set Core Clock
2953 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2954 		 * 31:24) even though these are only valid for CPU
2955 		 * models > 2, however guests may end up dividing or
2956 		 * multiplying by zero otherwise.
2957 		 */
2958 	case MSR_EBC_FREQUENCY_ID:
2959 		msr_info->data = 1 << 24;
2960 		break;
2961 	case MSR_IA32_APICBASE:
2962 		msr_info->data = kvm_get_apic_base(vcpu);
2963 		break;
2964 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2965 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2966 		break;
2967 	case MSR_IA32_TSCDEADLINE:
2968 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2969 		break;
2970 	case MSR_IA32_TSC_ADJUST:
2971 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2972 		break;
2973 	case MSR_IA32_MISC_ENABLE:
2974 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2975 		break;
2976 	case MSR_IA32_SMBASE:
2977 		if (!msr_info->host_initiated)
2978 			return 1;
2979 		msr_info->data = vcpu->arch.smbase;
2980 		break;
2981 	case MSR_SMI_COUNT:
2982 		msr_info->data = vcpu->arch.smi_count;
2983 		break;
2984 	case MSR_IA32_PERF_STATUS:
2985 		/* TSC increment by tick */
2986 		msr_info->data = 1000ULL;
2987 		/* CPU multiplier */
2988 		msr_info->data |= (((uint64_t)4ULL) << 40);
2989 		break;
2990 	case MSR_EFER:
2991 		msr_info->data = vcpu->arch.efer;
2992 		break;
2993 	case MSR_KVM_WALL_CLOCK:
2994 	case MSR_KVM_WALL_CLOCK_NEW:
2995 		msr_info->data = vcpu->kvm->arch.wall_clock;
2996 		break;
2997 	case MSR_KVM_SYSTEM_TIME:
2998 	case MSR_KVM_SYSTEM_TIME_NEW:
2999 		msr_info->data = vcpu->arch.time;
3000 		break;
3001 	case MSR_KVM_ASYNC_PF_EN:
3002 		msr_info->data = vcpu->arch.apf.msr_val;
3003 		break;
3004 	case MSR_KVM_STEAL_TIME:
3005 		msr_info->data = vcpu->arch.st.msr_val;
3006 		break;
3007 	case MSR_KVM_PV_EOI_EN:
3008 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3009 		break;
3010 	case MSR_KVM_POLL_CONTROL:
3011 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3012 		break;
3013 	case MSR_IA32_P5_MC_ADDR:
3014 	case MSR_IA32_P5_MC_TYPE:
3015 	case MSR_IA32_MCG_CAP:
3016 	case MSR_IA32_MCG_CTL:
3017 	case MSR_IA32_MCG_STATUS:
3018 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3019 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3020 				   msr_info->host_initiated);
3021 	case MSR_K7_CLK_CTL:
3022 		/*
3023 		 * Provide expected ramp-up count for K7. All other
3024 		 * are set to zero, indicating minimum divisors for
3025 		 * every field.
3026 		 *
3027 		 * This prevents guest kernels on AMD host with CPU
3028 		 * type 6, model 8 and higher from exploding due to
3029 		 * the rdmsr failing.
3030 		 */
3031 		msr_info->data = 0x20000000;
3032 		break;
3033 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3034 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3035 	case HV_X64_MSR_CRASH_CTL:
3036 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3037 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3038 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3039 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3040 		return kvm_hv_get_msr_common(vcpu,
3041 					     msr_info->index, &msr_info->data,
3042 					     msr_info->host_initiated);
3043 		break;
3044 	case MSR_IA32_BBL_CR_CTL3:
3045 		/* This legacy MSR exists but isn't fully documented in current
3046 		 * silicon.  It is however accessed by winxp in very narrow
3047 		 * scenarios where it sets bit #19, itself documented as
3048 		 * a "reserved" bit.  Best effort attempt to source coherent
3049 		 * read data here should the balance of the register be
3050 		 * interpreted by the guest:
3051 		 *
3052 		 * L2 cache control register 3: 64GB range, 256KB size,
3053 		 * enabled, latency 0x1, configured
3054 		 */
3055 		msr_info->data = 0xbe702111;
3056 		break;
3057 	case MSR_AMD64_OSVW_ID_LENGTH:
3058 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3059 			return 1;
3060 		msr_info->data = vcpu->arch.osvw.length;
3061 		break;
3062 	case MSR_AMD64_OSVW_STATUS:
3063 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3064 			return 1;
3065 		msr_info->data = vcpu->arch.osvw.status;
3066 		break;
3067 	case MSR_PLATFORM_INFO:
3068 		if (!msr_info->host_initiated &&
3069 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3070 			return 1;
3071 		msr_info->data = vcpu->arch.msr_platform_info;
3072 		break;
3073 	case MSR_MISC_FEATURES_ENABLES:
3074 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3075 		break;
3076 	case MSR_K7_HWCR:
3077 		msr_info->data = vcpu->arch.msr_hwcr;
3078 		break;
3079 	default:
3080 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3081 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3082 		if (!ignore_msrs) {
3083 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3084 					       msr_info->index);
3085 			return 1;
3086 		} else {
3087 			if (report_ignored_msrs)
3088 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3089 					msr_info->index);
3090 			msr_info->data = 0;
3091 		}
3092 		break;
3093 	}
3094 	return 0;
3095 }
3096 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3097 
3098 /*
3099  * Read or write a bunch of msrs. All parameters are kernel addresses.
3100  *
3101  * @return number of msrs set successfully.
3102  */
3103 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3104 		    struct kvm_msr_entry *entries,
3105 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3106 				  unsigned index, u64 *data))
3107 {
3108 	int i;
3109 
3110 	for (i = 0; i < msrs->nmsrs; ++i)
3111 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3112 			break;
3113 
3114 	return i;
3115 }
3116 
3117 /*
3118  * Read or write a bunch of msrs. Parameters are user addresses.
3119  *
3120  * @return number of msrs set successfully.
3121  */
3122 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3123 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3124 				unsigned index, u64 *data),
3125 		  int writeback)
3126 {
3127 	struct kvm_msrs msrs;
3128 	struct kvm_msr_entry *entries;
3129 	int r, n;
3130 	unsigned size;
3131 
3132 	r = -EFAULT;
3133 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3134 		goto out;
3135 
3136 	r = -E2BIG;
3137 	if (msrs.nmsrs >= MAX_IO_MSRS)
3138 		goto out;
3139 
3140 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3141 	entries = memdup_user(user_msrs->entries, size);
3142 	if (IS_ERR(entries)) {
3143 		r = PTR_ERR(entries);
3144 		goto out;
3145 	}
3146 
3147 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3148 	if (r < 0)
3149 		goto out_free;
3150 
3151 	r = -EFAULT;
3152 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3153 		goto out_free;
3154 
3155 	r = n;
3156 
3157 out_free:
3158 	kfree(entries);
3159 out:
3160 	return r;
3161 }
3162 
3163 static inline bool kvm_can_mwait_in_guest(void)
3164 {
3165 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3166 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3167 		boot_cpu_has(X86_FEATURE_ARAT);
3168 }
3169 
3170 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3171 {
3172 	int r = 0;
3173 
3174 	switch (ext) {
3175 	case KVM_CAP_IRQCHIP:
3176 	case KVM_CAP_HLT:
3177 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3178 	case KVM_CAP_SET_TSS_ADDR:
3179 	case KVM_CAP_EXT_CPUID:
3180 	case KVM_CAP_EXT_EMUL_CPUID:
3181 	case KVM_CAP_CLOCKSOURCE:
3182 	case KVM_CAP_PIT:
3183 	case KVM_CAP_NOP_IO_DELAY:
3184 	case KVM_CAP_MP_STATE:
3185 	case KVM_CAP_SYNC_MMU:
3186 	case KVM_CAP_USER_NMI:
3187 	case KVM_CAP_REINJECT_CONTROL:
3188 	case KVM_CAP_IRQ_INJECT_STATUS:
3189 	case KVM_CAP_IOEVENTFD:
3190 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3191 	case KVM_CAP_PIT2:
3192 	case KVM_CAP_PIT_STATE2:
3193 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3194 	case KVM_CAP_XEN_HVM:
3195 	case KVM_CAP_VCPU_EVENTS:
3196 	case KVM_CAP_HYPERV:
3197 	case KVM_CAP_HYPERV_VAPIC:
3198 	case KVM_CAP_HYPERV_SPIN:
3199 	case KVM_CAP_HYPERV_SYNIC:
3200 	case KVM_CAP_HYPERV_SYNIC2:
3201 	case KVM_CAP_HYPERV_VP_INDEX:
3202 	case KVM_CAP_HYPERV_EVENTFD:
3203 	case KVM_CAP_HYPERV_TLBFLUSH:
3204 	case KVM_CAP_HYPERV_SEND_IPI:
3205 	case KVM_CAP_HYPERV_CPUID:
3206 	case KVM_CAP_PCI_SEGMENT:
3207 	case KVM_CAP_DEBUGREGS:
3208 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3209 	case KVM_CAP_XSAVE:
3210 	case KVM_CAP_ASYNC_PF:
3211 	case KVM_CAP_GET_TSC_KHZ:
3212 	case KVM_CAP_KVMCLOCK_CTRL:
3213 	case KVM_CAP_READONLY_MEM:
3214 	case KVM_CAP_HYPERV_TIME:
3215 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3216 	case KVM_CAP_TSC_DEADLINE_TIMER:
3217 	case KVM_CAP_DISABLE_QUIRKS:
3218 	case KVM_CAP_SET_BOOT_CPU_ID:
3219  	case KVM_CAP_SPLIT_IRQCHIP:
3220 	case KVM_CAP_IMMEDIATE_EXIT:
3221 	case KVM_CAP_PMU_EVENT_FILTER:
3222 	case KVM_CAP_GET_MSR_FEATURES:
3223 	case KVM_CAP_MSR_PLATFORM_INFO:
3224 	case KVM_CAP_EXCEPTION_PAYLOAD:
3225 		r = 1;
3226 		break;
3227 	case KVM_CAP_SYNC_REGS:
3228 		r = KVM_SYNC_X86_VALID_FIELDS;
3229 		break;
3230 	case KVM_CAP_ADJUST_CLOCK:
3231 		r = KVM_CLOCK_TSC_STABLE;
3232 		break;
3233 	case KVM_CAP_X86_DISABLE_EXITS:
3234 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3235 		      KVM_X86_DISABLE_EXITS_CSTATE;
3236 		if(kvm_can_mwait_in_guest())
3237 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3238 		break;
3239 	case KVM_CAP_X86_SMM:
3240 		/* SMBASE is usually relocated above 1M on modern chipsets,
3241 		 * and SMM handlers might indeed rely on 4G segment limits,
3242 		 * so do not report SMM to be available if real mode is
3243 		 * emulated via vm86 mode.  Still, do not go to great lengths
3244 		 * to avoid userspace's usage of the feature, because it is a
3245 		 * fringe case that is not enabled except via specific settings
3246 		 * of the module parameters.
3247 		 */
3248 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3249 		break;
3250 	case KVM_CAP_VAPIC:
3251 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3252 		break;
3253 	case KVM_CAP_NR_VCPUS:
3254 		r = KVM_SOFT_MAX_VCPUS;
3255 		break;
3256 	case KVM_CAP_MAX_VCPUS:
3257 		r = KVM_MAX_VCPUS;
3258 		break;
3259 	case KVM_CAP_MAX_VCPU_ID:
3260 		r = KVM_MAX_VCPU_ID;
3261 		break;
3262 	case KVM_CAP_PV_MMU:	/* obsolete */
3263 		r = 0;
3264 		break;
3265 	case KVM_CAP_MCE:
3266 		r = KVM_MAX_MCE_BANKS;
3267 		break;
3268 	case KVM_CAP_XCRS:
3269 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3270 		break;
3271 	case KVM_CAP_TSC_CONTROL:
3272 		r = kvm_has_tsc_control;
3273 		break;
3274 	case KVM_CAP_X2APIC_API:
3275 		r = KVM_X2APIC_API_VALID_FLAGS;
3276 		break;
3277 	case KVM_CAP_NESTED_STATE:
3278 		r = kvm_x86_ops->get_nested_state ?
3279 			kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3280 		break;
3281 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3282 		r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3283 		break;
3284 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3285 		r = kvm_x86_ops->nested_enable_evmcs != NULL;
3286 		break;
3287 	default:
3288 		break;
3289 	}
3290 	return r;
3291 
3292 }
3293 
3294 long kvm_arch_dev_ioctl(struct file *filp,
3295 			unsigned int ioctl, unsigned long arg)
3296 {
3297 	void __user *argp = (void __user *)arg;
3298 	long r;
3299 
3300 	switch (ioctl) {
3301 	case KVM_GET_MSR_INDEX_LIST: {
3302 		struct kvm_msr_list __user *user_msr_list = argp;
3303 		struct kvm_msr_list msr_list;
3304 		unsigned n;
3305 
3306 		r = -EFAULT;
3307 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3308 			goto out;
3309 		n = msr_list.nmsrs;
3310 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3311 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3312 			goto out;
3313 		r = -E2BIG;
3314 		if (n < msr_list.nmsrs)
3315 			goto out;
3316 		r = -EFAULT;
3317 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3318 				 num_msrs_to_save * sizeof(u32)))
3319 			goto out;
3320 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3321 				 &emulated_msrs,
3322 				 num_emulated_msrs * sizeof(u32)))
3323 			goto out;
3324 		r = 0;
3325 		break;
3326 	}
3327 	case KVM_GET_SUPPORTED_CPUID:
3328 	case KVM_GET_EMULATED_CPUID: {
3329 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3330 		struct kvm_cpuid2 cpuid;
3331 
3332 		r = -EFAULT;
3333 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3334 			goto out;
3335 
3336 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3337 					    ioctl);
3338 		if (r)
3339 			goto out;
3340 
3341 		r = -EFAULT;
3342 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3343 			goto out;
3344 		r = 0;
3345 		break;
3346 	}
3347 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3348 		r = -EFAULT;
3349 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3350 				 sizeof(kvm_mce_cap_supported)))
3351 			goto out;
3352 		r = 0;
3353 		break;
3354 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3355 		struct kvm_msr_list __user *user_msr_list = argp;
3356 		struct kvm_msr_list msr_list;
3357 		unsigned int n;
3358 
3359 		r = -EFAULT;
3360 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3361 			goto out;
3362 		n = msr_list.nmsrs;
3363 		msr_list.nmsrs = num_msr_based_features;
3364 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3365 			goto out;
3366 		r = -E2BIG;
3367 		if (n < msr_list.nmsrs)
3368 			goto out;
3369 		r = -EFAULT;
3370 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3371 				 num_msr_based_features * sizeof(u32)))
3372 			goto out;
3373 		r = 0;
3374 		break;
3375 	}
3376 	case KVM_GET_MSRS:
3377 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3378 		break;
3379 	}
3380 	default:
3381 		r = -EINVAL;
3382 	}
3383 out:
3384 	return r;
3385 }
3386 
3387 static void wbinvd_ipi(void *garbage)
3388 {
3389 	wbinvd();
3390 }
3391 
3392 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3393 {
3394 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3395 }
3396 
3397 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3398 {
3399 	/* Address WBINVD may be executed by guest */
3400 	if (need_emulate_wbinvd(vcpu)) {
3401 		if (kvm_x86_ops->has_wbinvd_exit())
3402 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3403 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3404 			smp_call_function_single(vcpu->cpu,
3405 					wbinvd_ipi, NULL, 1);
3406 	}
3407 
3408 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3409 
3410 	fpregs_assert_state_consistent();
3411 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
3412 		switch_fpu_return();
3413 
3414 	/* Apply any externally detected TSC adjustments (due to suspend) */
3415 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3416 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3417 		vcpu->arch.tsc_offset_adjustment = 0;
3418 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3419 	}
3420 
3421 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3422 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3423 				rdtsc() - vcpu->arch.last_host_tsc;
3424 		if (tsc_delta < 0)
3425 			mark_tsc_unstable("KVM discovered backwards TSC");
3426 
3427 		if (kvm_check_tsc_unstable()) {
3428 			u64 offset = kvm_compute_tsc_offset(vcpu,
3429 						vcpu->arch.last_guest_tsc);
3430 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3431 			vcpu->arch.tsc_catchup = 1;
3432 		}
3433 
3434 		if (kvm_lapic_hv_timer_in_use(vcpu))
3435 			kvm_lapic_restart_hv_timer(vcpu);
3436 
3437 		/*
3438 		 * On a host with synchronized TSC, there is no need to update
3439 		 * kvmclock on vcpu->cpu migration
3440 		 */
3441 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3442 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3443 		if (vcpu->cpu != cpu)
3444 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3445 		vcpu->cpu = cpu;
3446 	}
3447 
3448 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3449 }
3450 
3451 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3452 {
3453 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3454 		return;
3455 
3456 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3457 
3458 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3459 			&vcpu->arch.st.steal.preempted,
3460 			offsetof(struct kvm_steal_time, preempted),
3461 			sizeof(vcpu->arch.st.steal.preempted));
3462 }
3463 
3464 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3465 {
3466 	int idx;
3467 
3468 	if (vcpu->preempted)
3469 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3470 
3471 	/*
3472 	 * Disable page faults because we're in atomic context here.
3473 	 * kvm_write_guest_offset_cached() would call might_fault()
3474 	 * that relies on pagefault_disable() to tell if there's a
3475 	 * bug. NOTE: the write to guest memory may not go through if
3476 	 * during postcopy live migration or if there's heavy guest
3477 	 * paging.
3478 	 */
3479 	pagefault_disable();
3480 	/*
3481 	 * kvm_memslots() will be called by
3482 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3483 	 */
3484 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3485 	kvm_steal_time_set_preempted(vcpu);
3486 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3487 	pagefault_enable();
3488 	kvm_x86_ops->vcpu_put(vcpu);
3489 	vcpu->arch.last_host_tsc = rdtsc();
3490 	/*
3491 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3492 	 * on every vmexit, but if not, we might have a stale dr6 from the
3493 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3494 	 */
3495 	set_debugreg(0, 6);
3496 }
3497 
3498 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3499 				    struct kvm_lapic_state *s)
3500 {
3501 	if (vcpu->arch.apicv_active)
3502 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3503 
3504 	return kvm_apic_get_state(vcpu, s);
3505 }
3506 
3507 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3508 				    struct kvm_lapic_state *s)
3509 {
3510 	int r;
3511 
3512 	r = kvm_apic_set_state(vcpu, s);
3513 	if (r)
3514 		return r;
3515 	update_cr8_intercept(vcpu);
3516 
3517 	return 0;
3518 }
3519 
3520 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3521 {
3522 	return (!lapic_in_kernel(vcpu) ||
3523 		kvm_apic_accept_pic_intr(vcpu));
3524 }
3525 
3526 /*
3527  * if userspace requested an interrupt window, check that the
3528  * interrupt window is open.
3529  *
3530  * No need to exit to userspace if we already have an interrupt queued.
3531  */
3532 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3533 {
3534 	return kvm_arch_interrupt_allowed(vcpu) &&
3535 		!kvm_cpu_has_interrupt(vcpu) &&
3536 		!kvm_event_needs_reinjection(vcpu) &&
3537 		kvm_cpu_accept_dm_intr(vcpu);
3538 }
3539 
3540 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3541 				    struct kvm_interrupt *irq)
3542 {
3543 	if (irq->irq >= KVM_NR_INTERRUPTS)
3544 		return -EINVAL;
3545 
3546 	if (!irqchip_in_kernel(vcpu->kvm)) {
3547 		kvm_queue_interrupt(vcpu, irq->irq, false);
3548 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3549 		return 0;
3550 	}
3551 
3552 	/*
3553 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3554 	 * fail for in-kernel 8259.
3555 	 */
3556 	if (pic_in_kernel(vcpu->kvm))
3557 		return -ENXIO;
3558 
3559 	if (vcpu->arch.pending_external_vector != -1)
3560 		return -EEXIST;
3561 
3562 	vcpu->arch.pending_external_vector = irq->irq;
3563 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3564 	return 0;
3565 }
3566 
3567 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3568 {
3569 	kvm_inject_nmi(vcpu);
3570 
3571 	return 0;
3572 }
3573 
3574 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3575 {
3576 	kvm_make_request(KVM_REQ_SMI, vcpu);
3577 
3578 	return 0;
3579 }
3580 
3581 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3582 					   struct kvm_tpr_access_ctl *tac)
3583 {
3584 	if (tac->flags)
3585 		return -EINVAL;
3586 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3587 	return 0;
3588 }
3589 
3590 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3591 					u64 mcg_cap)
3592 {
3593 	int r;
3594 	unsigned bank_num = mcg_cap & 0xff, bank;
3595 
3596 	r = -EINVAL;
3597 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3598 		goto out;
3599 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3600 		goto out;
3601 	r = 0;
3602 	vcpu->arch.mcg_cap = mcg_cap;
3603 	/* Init IA32_MCG_CTL to all 1s */
3604 	if (mcg_cap & MCG_CTL_P)
3605 		vcpu->arch.mcg_ctl = ~(u64)0;
3606 	/* Init IA32_MCi_CTL to all 1s */
3607 	for (bank = 0; bank < bank_num; bank++)
3608 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3609 
3610 	kvm_x86_ops->setup_mce(vcpu);
3611 out:
3612 	return r;
3613 }
3614 
3615 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3616 				      struct kvm_x86_mce *mce)
3617 {
3618 	u64 mcg_cap = vcpu->arch.mcg_cap;
3619 	unsigned bank_num = mcg_cap & 0xff;
3620 	u64 *banks = vcpu->arch.mce_banks;
3621 
3622 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3623 		return -EINVAL;
3624 	/*
3625 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3626 	 * reporting is disabled
3627 	 */
3628 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3629 	    vcpu->arch.mcg_ctl != ~(u64)0)
3630 		return 0;
3631 	banks += 4 * mce->bank;
3632 	/*
3633 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3634 	 * reporting is disabled for the bank
3635 	 */
3636 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3637 		return 0;
3638 	if (mce->status & MCI_STATUS_UC) {
3639 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3640 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3641 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3642 			return 0;
3643 		}
3644 		if (banks[1] & MCI_STATUS_VAL)
3645 			mce->status |= MCI_STATUS_OVER;
3646 		banks[2] = mce->addr;
3647 		banks[3] = mce->misc;
3648 		vcpu->arch.mcg_status = mce->mcg_status;
3649 		banks[1] = mce->status;
3650 		kvm_queue_exception(vcpu, MC_VECTOR);
3651 	} else if (!(banks[1] & MCI_STATUS_VAL)
3652 		   || !(banks[1] & MCI_STATUS_UC)) {
3653 		if (banks[1] & MCI_STATUS_VAL)
3654 			mce->status |= MCI_STATUS_OVER;
3655 		banks[2] = mce->addr;
3656 		banks[3] = mce->misc;
3657 		banks[1] = mce->status;
3658 	} else
3659 		banks[1] |= MCI_STATUS_OVER;
3660 	return 0;
3661 }
3662 
3663 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3664 					       struct kvm_vcpu_events *events)
3665 {
3666 	process_nmi(vcpu);
3667 
3668 	/*
3669 	 * The API doesn't provide the instruction length for software
3670 	 * exceptions, so don't report them. As long as the guest RIP
3671 	 * isn't advanced, we should expect to encounter the exception
3672 	 * again.
3673 	 */
3674 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3675 		events->exception.injected = 0;
3676 		events->exception.pending = 0;
3677 	} else {
3678 		events->exception.injected = vcpu->arch.exception.injected;
3679 		events->exception.pending = vcpu->arch.exception.pending;
3680 		/*
3681 		 * For ABI compatibility, deliberately conflate
3682 		 * pending and injected exceptions when
3683 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3684 		 */
3685 		if (!vcpu->kvm->arch.exception_payload_enabled)
3686 			events->exception.injected |=
3687 				vcpu->arch.exception.pending;
3688 	}
3689 	events->exception.nr = vcpu->arch.exception.nr;
3690 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3691 	events->exception.error_code = vcpu->arch.exception.error_code;
3692 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3693 	events->exception_payload = vcpu->arch.exception.payload;
3694 
3695 	events->interrupt.injected =
3696 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3697 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3698 	events->interrupt.soft = 0;
3699 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3700 
3701 	events->nmi.injected = vcpu->arch.nmi_injected;
3702 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3703 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3704 	events->nmi.pad = 0;
3705 
3706 	events->sipi_vector = 0; /* never valid when reporting to user space */
3707 
3708 	events->smi.smm = is_smm(vcpu);
3709 	events->smi.pending = vcpu->arch.smi_pending;
3710 	events->smi.smm_inside_nmi =
3711 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3712 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3713 
3714 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3715 			 | KVM_VCPUEVENT_VALID_SHADOW
3716 			 | KVM_VCPUEVENT_VALID_SMM);
3717 	if (vcpu->kvm->arch.exception_payload_enabled)
3718 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3719 
3720 	memset(&events->reserved, 0, sizeof(events->reserved));
3721 }
3722 
3723 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3724 
3725 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3726 					      struct kvm_vcpu_events *events)
3727 {
3728 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3729 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3730 			      | KVM_VCPUEVENT_VALID_SHADOW
3731 			      | KVM_VCPUEVENT_VALID_SMM
3732 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3733 		return -EINVAL;
3734 
3735 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3736 		if (!vcpu->kvm->arch.exception_payload_enabled)
3737 			return -EINVAL;
3738 		if (events->exception.pending)
3739 			events->exception.injected = 0;
3740 		else
3741 			events->exception_has_payload = 0;
3742 	} else {
3743 		events->exception.pending = 0;
3744 		events->exception_has_payload = 0;
3745 	}
3746 
3747 	if ((events->exception.injected || events->exception.pending) &&
3748 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3749 		return -EINVAL;
3750 
3751 	/* INITs are latched while in SMM */
3752 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3753 	    (events->smi.smm || events->smi.pending) &&
3754 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3755 		return -EINVAL;
3756 
3757 	process_nmi(vcpu);
3758 	vcpu->arch.exception.injected = events->exception.injected;
3759 	vcpu->arch.exception.pending = events->exception.pending;
3760 	vcpu->arch.exception.nr = events->exception.nr;
3761 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3762 	vcpu->arch.exception.error_code = events->exception.error_code;
3763 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3764 	vcpu->arch.exception.payload = events->exception_payload;
3765 
3766 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3767 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3768 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3769 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3770 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3771 						  events->interrupt.shadow);
3772 
3773 	vcpu->arch.nmi_injected = events->nmi.injected;
3774 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3775 		vcpu->arch.nmi_pending = events->nmi.pending;
3776 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3777 
3778 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3779 	    lapic_in_kernel(vcpu))
3780 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3781 
3782 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3783 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3784 			if (events->smi.smm)
3785 				vcpu->arch.hflags |= HF_SMM_MASK;
3786 			else
3787 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3788 			kvm_smm_changed(vcpu);
3789 		}
3790 
3791 		vcpu->arch.smi_pending = events->smi.pending;
3792 
3793 		if (events->smi.smm) {
3794 			if (events->smi.smm_inside_nmi)
3795 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3796 			else
3797 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3798 			if (lapic_in_kernel(vcpu)) {
3799 				if (events->smi.latched_init)
3800 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3801 				else
3802 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3803 			}
3804 		}
3805 	}
3806 
3807 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3808 
3809 	return 0;
3810 }
3811 
3812 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3813 					     struct kvm_debugregs *dbgregs)
3814 {
3815 	unsigned long val;
3816 
3817 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3818 	kvm_get_dr(vcpu, 6, &val);
3819 	dbgregs->dr6 = val;
3820 	dbgregs->dr7 = vcpu->arch.dr7;
3821 	dbgregs->flags = 0;
3822 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3823 }
3824 
3825 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3826 					    struct kvm_debugregs *dbgregs)
3827 {
3828 	if (dbgregs->flags)
3829 		return -EINVAL;
3830 
3831 	if (dbgregs->dr6 & ~0xffffffffull)
3832 		return -EINVAL;
3833 	if (dbgregs->dr7 & ~0xffffffffull)
3834 		return -EINVAL;
3835 
3836 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3837 	kvm_update_dr0123(vcpu);
3838 	vcpu->arch.dr6 = dbgregs->dr6;
3839 	kvm_update_dr6(vcpu);
3840 	vcpu->arch.dr7 = dbgregs->dr7;
3841 	kvm_update_dr7(vcpu);
3842 
3843 	return 0;
3844 }
3845 
3846 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3847 
3848 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3849 {
3850 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3851 	u64 xstate_bv = xsave->header.xfeatures;
3852 	u64 valid;
3853 
3854 	/*
3855 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3856 	 * leaves 0 and 1 in the loop below.
3857 	 */
3858 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3859 
3860 	/* Set XSTATE_BV */
3861 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3862 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3863 
3864 	/*
3865 	 * Copy each region from the possibly compacted offset to the
3866 	 * non-compacted offset.
3867 	 */
3868 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3869 	while (valid) {
3870 		u64 xfeature_mask = valid & -valid;
3871 		int xfeature_nr = fls64(xfeature_mask) - 1;
3872 		void *src = get_xsave_addr(xsave, xfeature_nr);
3873 
3874 		if (src) {
3875 			u32 size, offset, ecx, edx;
3876 			cpuid_count(XSTATE_CPUID, xfeature_nr,
3877 				    &size, &offset, &ecx, &edx);
3878 			if (xfeature_nr == XFEATURE_PKRU)
3879 				memcpy(dest + offset, &vcpu->arch.pkru,
3880 				       sizeof(vcpu->arch.pkru));
3881 			else
3882 				memcpy(dest + offset, src, size);
3883 
3884 		}
3885 
3886 		valid -= xfeature_mask;
3887 	}
3888 }
3889 
3890 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3891 {
3892 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3893 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3894 	u64 valid;
3895 
3896 	/*
3897 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3898 	 * leaves 0 and 1 in the loop below.
3899 	 */
3900 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3901 
3902 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3903 	xsave->header.xfeatures = xstate_bv;
3904 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3905 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3906 
3907 	/*
3908 	 * Copy each region from the non-compacted offset to the
3909 	 * possibly compacted offset.
3910 	 */
3911 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3912 	while (valid) {
3913 		u64 xfeature_mask = valid & -valid;
3914 		int xfeature_nr = fls64(xfeature_mask) - 1;
3915 		void *dest = get_xsave_addr(xsave, xfeature_nr);
3916 
3917 		if (dest) {
3918 			u32 size, offset, ecx, edx;
3919 			cpuid_count(XSTATE_CPUID, xfeature_nr,
3920 				    &size, &offset, &ecx, &edx);
3921 			if (xfeature_nr == XFEATURE_PKRU)
3922 				memcpy(&vcpu->arch.pkru, src + offset,
3923 				       sizeof(vcpu->arch.pkru));
3924 			else
3925 				memcpy(dest, src + offset, size);
3926 		}
3927 
3928 		valid -= xfeature_mask;
3929 	}
3930 }
3931 
3932 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3933 					 struct kvm_xsave *guest_xsave)
3934 {
3935 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3936 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3937 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3938 	} else {
3939 		memcpy(guest_xsave->region,
3940 			&vcpu->arch.guest_fpu->state.fxsave,
3941 			sizeof(struct fxregs_state));
3942 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3943 			XFEATURE_MASK_FPSSE;
3944 	}
3945 }
3946 
3947 #define XSAVE_MXCSR_OFFSET 24
3948 
3949 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3950 					struct kvm_xsave *guest_xsave)
3951 {
3952 	u64 xstate_bv =
3953 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3954 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3955 
3956 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3957 		/*
3958 		 * Here we allow setting states that are not present in
3959 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3960 		 * with old userspace.
3961 		 */
3962 		if (xstate_bv & ~kvm_supported_xcr0() ||
3963 			mxcsr & ~mxcsr_feature_mask)
3964 			return -EINVAL;
3965 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3966 	} else {
3967 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3968 			mxcsr & ~mxcsr_feature_mask)
3969 			return -EINVAL;
3970 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3971 			guest_xsave->region, sizeof(struct fxregs_state));
3972 	}
3973 	return 0;
3974 }
3975 
3976 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3977 					struct kvm_xcrs *guest_xcrs)
3978 {
3979 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3980 		guest_xcrs->nr_xcrs = 0;
3981 		return;
3982 	}
3983 
3984 	guest_xcrs->nr_xcrs = 1;
3985 	guest_xcrs->flags = 0;
3986 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3987 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3988 }
3989 
3990 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3991 				       struct kvm_xcrs *guest_xcrs)
3992 {
3993 	int i, r = 0;
3994 
3995 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3996 		return -EINVAL;
3997 
3998 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3999 		return -EINVAL;
4000 
4001 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4002 		/* Only support XCR0 currently */
4003 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4004 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4005 				guest_xcrs->xcrs[i].value);
4006 			break;
4007 		}
4008 	if (r)
4009 		r = -EINVAL;
4010 	return r;
4011 }
4012 
4013 /*
4014  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4015  * stopped by the hypervisor.  This function will be called from the host only.
4016  * EINVAL is returned when the host attempts to set the flag for a guest that
4017  * does not support pv clocks.
4018  */
4019 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4020 {
4021 	if (!vcpu->arch.pv_time_enabled)
4022 		return -EINVAL;
4023 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4024 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4025 	return 0;
4026 }
4027 
4028 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4029 				     struct kvm_enable_cap *cap)
4030 {
4031 	int r;
4032 	uint16_t vmcs_version;
4033 	void __user *user_ptr;
4034 
4035 	if (cap->flags)
4036 		return -EINVAL;
4037 
4038 	switch (cap->cap) {
4039 	case KVM_CAP_HYPERV_SYNIC2:
4040 		if (cap->args[0])
4041 			return -EINVAL;
4042 		/* fall through */
4043 
4044 	case KVM_CAP_HYPERV_SYNIC:
4045 		if (!irqchip_in_kernel(vcpu->kvm))
4046 			return -EINVAL;
4047 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4048 					     KVM_CAP_HYPERV_SYNIC2);
4049 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4050 		if (!kvm_x86_ops->nested_enable_evmcs)
4051 			return -ENOTTY;
4052 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4053 		if (!r) {
4054 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4055 			if (copy_to_user(user_ptr, &vmcs_version,
4056 					 sizeof(vmcs_version)))
4057 				r = -EFAULT;
4058 		}
4059 		return r;
4060 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4061 		if (!kvm_x86_ops->enable_direct_tlbflush)
4062 			return -ENOTTY;
4063 
4064 		return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4065 
4066 	default:
4067 		return -EINVAL;
4068 	}
4069 }
4070 
4071 long kvm_arch_vcpu_ioctl(struct file *filp,
4072 			 unsigned int ioctl, unsigned long arg)
4073 {
4074 	struct kvm_vcpu *vcpu = filp->private_data;
4075 	void __user *argp = (void __user *)arg;
4076 	int r;
4077 	union {
4078 		struct kvm_lapic_state *lapic;
4079 		struct kvm_xsave *xsave;
4080 		struct kvm_xcrs *xcrs;
4081 		void *buffer;
4082 	} u;
4083 
4084 	vcpu_load(vcpu);
4085 
4086 	u.buffer = NULL;
4087 	switch (ioctl) {
4088 	case KVM_GET_LAPIC: {
4089 		r = -EINVAL;
4090 		if (!lapic_in_kernel(vcpu))
4091 			goto out;
4092 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4093 				GFP_KERNEL_ACCOUNT);
4094 
4095 		r = -ENOMEM;
4096 		if (!u.lapic)
4097 			goto out;
4098 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4099 		if (r)
4100 			goto out;
4101 		r = -EFAULT;
4102 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4103 			goto out;
4104 		r = 0;
4105 		break;
4106 	}
4107 	case KVM_SET_LAPIC: {
4108 		r = -EINVAL;
4109 		if (!lapic_in_kernel(vcpu))
4110 			goto out;
4111 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4112 		if (IS_ERR(u.lapic)) {
4113 			r = PTR_ERR(u.lapic);
4114 			goto out_nofree;
4115 		}
4116 
4117 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4118 		break;
4119 	}
4120 	case KVM_INTERRUPT: {
4121 		struct kvm_interrupt irq;
4122 
4123 		r = -EFAULT;
4124 		if (copy_from_user(&irq, argp, sizeof(irq)))
4125 			goto out;
4126 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4127 		break;
4128 	}
4129 	case KVM_NMI: {
4130 		r = kvm_vcpu_ioctl_nmi(vcpu);
4131 		break;
4132 	}
4133 	case KVM_SMI: {
4134 		r = kvm_vcpu_ioctl_smi(vcpu);
4135 		break;
4136 	}
4137 	case KVM_SET_CPUID: {
4138 		struct kvm_cpuid __user *cpuid_arg = argp;
4139 		struct kvm_cpuid cpuid;
4140 
4141 		r = -EFAULT;
4142 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4143 			goto out;
4144 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4145 		break;
4146 	}
4147 	case KVM_SET_CPUID2: {
4148 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4149 		struct kvm_cpuid2 cpuid;
4150 
4151 		r = -EFAULT;
4152 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4153 			goto out;
4154 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4155 					      cpuid_arg->entries);
4156 		break;
4157 	}
4158 	case KVM_GET_CPUID2: {
4159 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4160 		struct kvm_cpuid2 cpuid;
4161 
4162 		r = -EFAULT;
4163 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4164 			goto out;
4165 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4166 					      cpuid_arg->entries);
4167 		if (r)
4168 			goto out;
4169 		r = -EFAULT;
4170 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4171 			goto out;
4172 		r = 0;
4173 		break;
4174 	}
4175 	case KVM_GET_MSRS: {
4176 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4177 		r = msr_io(vcpu, argp, do_get_msr, 1);
4178 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4179 		break;
4180 	}
4181 	case KVM_SET_MSRS: {
4182 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4183 		r = msr_io(vcpu, argp, do_set_msr, 0);
4184 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4185 		break;
4186 	}
4187 	case KVM_TPR_ACCESS_REPORTING: {
4188 		struct kvm_tpr_access_ctl tac;
4189 
4190 		r = -EFAULT;
4191 		if (copy_from_user(&tac, argp, sizeof(tac)))
4192 			goto out;
4193 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4194 		if (r)
4195 			goto out;
4196 		r = -EFAULT;
4197 		if (copy_to_user(argp, &tac, sizeof(tac)))
4198 			goto out;
4199 		r = 0;
4200 		break;
4201 	};
4202 	case KVM_SET_VAPIC_ADDR: {
4203 		struct kvm_vapic_addr va;
4204 		int idx;
4205 
4206 		r = -EINVAL;
4207 		if (!lapic_in_kernel(vcpu))
4208 			goto out;
4209 		r = -EFAULT;
4210 		if (copy_from_user(&va, argp, sizeof(va)))
4211 			goto out;
4212 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4213 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4214 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4215 		break;
4216 	}
4217 	case KVM_X86_SETUP_MCE: {
4218 		u64 mcg_cap;
4219 
4220 		r = -EFAULT;
4221 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4222 			goto out;
4223 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4224 		break;
4225 	}
4226 	case KVM_X86_SET_MCE: {
4227 		struct kvm_x86_mce mce;
4228 
4229 		r = -EFAULT;
4230 		if (copy_from_user(&mce, argp, sizeof(mce)))
4231 			goto out;
4232 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4233 		break;
4234 	}
4235 	case KVM_GET_VCPU_EVENTS: {
4236 		struct kvm_vcpu_events events;
4237 
4238 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4239 
4240 		r = -EFAULT;
4241 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4242 			break;
4243 		r = 0;
4244 		break;
4245 	}
4246 	case KVM_SET_VCPU_EVENTS: {
4247 		struct kvm_vcpu_events events;
4248 
4249 		r = -EFAULT;
4250 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4251 			break;
4252 
4253 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4254 		break;
4255 	}
4256 	case KVM_GET_DEBUGREGS: {
4257 		struct kvm_debugregs dbgregs;
4258 
4259 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4260 
4261 		r = -EFAULT;
4262 		if (copy_to_user(argp, &dbgregs,
4263 				 sizeof(struct kvm_debugregs)))
4264 			break;
4265 		r = 0;
4266 		break;
4267 	}
4268 	case KVM_SET_DEBUGREGS: {
4269 		struct kvm_debugregs dbgregs;
4270 
4271 		r = -EFAULT;
4272 		if (copy_from_user(&dbgregs, argp,
4273 				   sizeof(struct kvm_debugregs)))
4274 			break;
4275 
4276 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4277 		break;
4278 	}
4279 	case KVM_GET_XSAVE: {
4280 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4281 		r = -ENOMEM;
4282 		if (!u.xsave)
4283 			break;
4284 
4285 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4286 
4287 		r = -EFAULT;
4288 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4289 			break;
4290 		r = 0;
4291 		break;
4292 	}
4293 	case KVM_SET_XSAVE: {
4294 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4295 		if (IS_ERR(u.xsave)) {
4296 			r = PTR_ERR(u.xsave);
4297 			goto out_nofree;
4298 		}
4299 
4300 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4301 		break;
4302 	}
4303 	case KVM_GET_XCRS: {
4304 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4305 		r = -ENOMEM;
4306 		if (!u.xcrs)
4307 			break;
4308 
4309 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4310 
4311 		r = -EFAULT;
4312 		if (copy_to_user(argp, u.xcrs,
4313 				 sizeof(struct kvm_xcrs)))
4314 			break;
4315 		r = 0;
4316 		break;
4317 	}
4318 	case KVM_SET_XCRS: {
4319 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4320 		if (IS_ERR(u.xcrs)) {
4321 			r = PTR_ERR(u.xcrs);
4322 			goto out_nofree;
4323 		}
4324 
4325 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4326 		break;
4327 	}
4328 	case KVM_SET_TSC_KHZ: {
4329 		u32 user_tsc_khz;
4330 
4331 		r = -EINVAL;
4332 		user_tsc_khz = (u32)arg;
4333 
4334 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4335 			goto out;
4336 
4337 		if (user_tsc_khz == 0)
4338 			user_tsc_khz = tsc_khz;
4339 
4340 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4341 			r = 0;
4342 
4343 		goto out;
4344 	}
4345 	case KVM_GET_TSC_KHZ: {
4346 		r = vcpu->arch.virtual_tsc_khz;
4347 		goto out;
4348 	}
4349 	case KVM_KVMCLOCK_CTRL: {
4350 		r = kvm_set_guest_paused(vcpu);
4351 		goto out;
4352 	}
4353 	case KVM_ENABLE_CAP: {
4354 		struct kvm_enable_cap cap;
4355 
4356 		r = -EFAULT;
4357 		if (copy_from_user(&cap, argp, sizeof(cap)))
4358 			goto out;
4359 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4360 		break;
4361 	}
4362 	case KVM_GET_NESTED_STATE: {
4363 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4364 		u32 user_data_size;
4365 
4366 		r = -EINVAL;
4367 		if (!kvm_x86_ops->get_nested_state)
4368 			break;
4369 
4370 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4371 		r = -EFAULT;
4372 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4373 			break;
4374 
4375 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4376 						  user_data_size);
4377 		if (r < 0)
4378 			break;
4379 
4380 		if (r > user_data_size) {
4381 			if (put_user(r, &user_kvm_nested_state->size))
4382 				r = -EFAULT;
4383 			else
4384 				r = -E2BIG;
4385 			break;
4386 		}
4387 
4388 		r = 0;
4389 		break;
4390 	}
4391 	case KVM_SET_NESTED_STATE: {
4392 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4393 		struct kvm_nested_state kvm_state;
4394 
4395 		r = -EINVAL;
4396 		if (!kvm_x86_ops->set_nested_state)
4397 			break;
4398 
4399 		r = -EFAULT;
4400 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4401 			break;
4402 
4403 		r = -EINVAL;
4404 		if (kvm_state.size < sizeof(kvm_state))
4405 			break;
4406 
4407 		if (kvm_state.flags &
4408 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4409 		      | KVM_STATE_NESTED_EVMCS))
4410 			break;
4411 
4412 		/* nested_run_pending implies guest_mode.  */
4413 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4414 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4415 			break;
4416 
4417 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4418 		break;
4419 	}
4420 	case KVM_GET_SUPPORTED_HV_CPUID: {
4421 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4422 		struct kvm_cpuid2 cpuid;
4423 
4424 		r = -EFAULT;
4425 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4426 			goto out;
4427 
4428 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4429 						cpuid_arg->entries);
4430 		if (r)
4431 			goto out;
4432 
4433 		r = -EFAULT;
4434 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4435 			goto out;
4436 		r = 0;
4437 		break;
4438 	}
4439 	default:
4440 		r = -EINVAL;
4441 	}
4442 out:
4443 	kfree(u.buffer);
4444 out_nofree:
4445 	vcpu_put(vcpu);
4446 	return r;
4447 }
4448 
4449 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4450 {
4451 	return VM_FAULT_SIGBUS;
4452 }
4453 
4454 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4455 {
4456 	int ret;
4457 
4458 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4459 		return -EINVAL;
4460 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4461 	return ret;
4462 }
4463 
4464 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4465 					      u64 ident_addr)
4466 {
4467 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4468 }
4469 
4470 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4471 					 unsigned long kvm_nr_mmu_pages)
4472 {
4473 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4474 		return -EINVAL;
4475 
4476 	mutex_lock(&kvm->slots_lock);
4477 
4478 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4479 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4480 
4481 	mutex_unlock(&kvm->slots_lock);
4482 	return 0;
4483 }
4484 
4485 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4486 {
4487 	return kvm->arch.n_max_mmu_pages;
4488 }
4489 
4490 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4491 {
4492 	struct kvm_pic *pic = kvm->arch.vpic;
4493 	int r;
4494 
4495 	r = 0;
4496 	switch (chip->chip_id) {
4497 	case KVM_IRQCHIP_PIC_MASTER:
4498 		memcpy(&chip->chip.pic, &pic->pics[0],
4499 			sizeof(struct kvm_pic_state));
4500 		break;
4501 	case KVM_IRQCHIP_PIC_SLAVE:
4502 		memcpy(&chip->chip.pic, &pic->pics[1],
4503 			sizeof(struct kvm_pic_state));
4504 		break;
4505 	case KVM_IRQCHIP_IOAPIC:
4506 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4507 		break;
4508 	default:
4509 		r = -EINVAL;
4510 		break;
4511 	}
4512 	return r;
4513 }
4514 
4515 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4516 {
4517 	struct kvm_pic *pic = kvm->arch.vpic;
4518 	int r;
4519 
4520 	r = 0;
4521 	switch (chip->chip_id) {
4522 	case KVM_IRQCHIP_PIC_MASTER:
4523 		spin_lock(&pic->lock);
4524 		memcpy(&pic->pics[0], &chip->chip.pic,
4525 			sizeof(struct kvm_pic_state));
4526 		spin_unlock(&pic->lock);
4527 		break;
4528 	case KVM_IRQCHIP_PIC_SLAVE:
4529 		spin_lock(&pic->lock);
4530 		memcpy(&pic->pics[1], &chip->chip.pic,
4531 			sizeof(struct kvm_pic_state));
4532 		spin_unlock(&pic->lock);
4533 		break;
4534 	case KVM_IRQCHIP_IOAPIC:
4535 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4536 		break;
4537 	default:
4538 		r = -EINVAL;
4539 		break;
4540 	}
4541 	kvm_pic_update_irq(pic);
4542 	return r;
4543 }
4544 
4545 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4546 {
4547 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4548 
4549 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4550 
4551 	mutex_lock(&kps->lock);
4552 	memcpy(ps, &kps->channels, sizeof(*ps));
4553 	mutex_unlock(&kps->lock);
4554 	return 0;
4555 }
4556 
4557 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4558 {
4559 	int i;
4560 	struct kvm_pit *pit = kvm->arch.vpit;
4561 
4562 	mutex_lock(&pit->pit_state.lock);
4563 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4564 	for (i = 0; i < 3; i++)
4565 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4566 	mutex_unlock(&pit->pit_state.lock);
4567 	return 0;
4568 }
4569 
4570 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4571 {
4572 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4573 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4574 		sizeof(ps->channels));
4575 	ps->flags = kvm->arch.vpit->pit_state.flags;
4576 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4577 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4578 	return 0;
4579 }
4580 
4581 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4582 {
4583 	int start = 0;
4584 	int i;
4585 	u32 prev_legacy, cur_legacy;
4586 	struct kvm_pit *pit = kvm->arch.vpit;
4587 
4588 	mutex_lock(&pit->pit_state.lock);
4589 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4590 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4591 	if (!prev_legacy && cur_legacy)
4592 		start = 1;
4593 	memcpy(&pit->pit_state.channels, &ps->channels,
4594 	       sizeof(pit->pit_state.channels));
4595 	pit->pit_state.flags = ps->flags;
4596 	for (i = 0; i < 3; i++)
4597 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4598 				   start && i == 0);
4599 	mutex_unlock(&pit->pit_state.lock);
4600 	return 0;
4601 }
4602 
4603 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4604 				 struct kvm_reinject_control *control)
4605 {
4606 	struct kvm_pit *pit = kvm->arch.vpit;
4607 
4608 	if (!pit)
4609 		return -ENXIO;
4610 
4611 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4612 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4613 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4614 	 */
4615 	mutex_lock(&pit->pit_state.lock);
4616 	kvm_pit_set_reinject(pit, control->pit_reinject);
4617 	mutex_unlock(&pit->pit_state.lock);
4618 
4619 	return 0;
4620 }
4621 
4622 /**
4623  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4624  * @kvm: kvm instance
4625  * @log: slot id and address to which we copy the log
4626  *
4627  * Steps 1-4 below provide general overview of dirty page logging. See
4628  * kvm_get_dirty_log_protect() function description for additional details.
4629  *
4630  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4631  * always flush the TLB (step 4) even if previous step failed  and the dirty
4632  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4633  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4634  * writes will be marked dirty for next log read.
4635  *
4636  *   1. Take a snapshot of the bit and clear it if needed.
4637  *   2. Write protect the corresponding page.
4638  *   3. Copy the snapshot to the userspace.
4639  *   4. Flush TLB's if needed.
4640  */
4641 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4642 {
4643 	bool flush = false;
4644 	int r;
4645 
4646 	mutex_lock(&kvm->slots_lock);
4647 
4648 	/*
4649 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4650 	 */
4651 	if (kvm_x86_ops->flush_log_dirty)
4652 		kvm_x86_ops->flush_log_dirty(kvm);
4653 
4654 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4655 
4656 	/*
4657 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4658 	 * kvm_mmu_slot_remove_write_access().
4659 	 */
4660 	lockdep_assert_held(&kvm->slots_lock);
4661 	if (flush)
4662 		kvm_flush_remote_tlbs(kvm);
4663 
4664 	mutex_unlock(&kvm->slots_lock);
4665 	return r;
4666 }
4667 
4668 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4669 {
4670 	bool flush = false;
4671 	int r;
4672 
4673 	mutex_lock(&kvm->slots_lock);
4674 
4675 	/*
4676 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4677 	 */
4678 	if (kvm_x86_ops->flush_log_dirty)
4679 		kvm_x86_ops->flush_log_dirty(kvm);
4680 
4681 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4682 
4683 	/*
4684 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4685 	 * kvm_mmu_slot_remove_write_access().
4686 	 */
4687 	lockdep_assert_held(&kvm->slots_lock);
4688 	if (flush)
4689 		kvm_flush_remote_tlbs(kvm);
4690 
4691 	mutex_unlock(&kvm->slots_lock);
4692 	return r;
4693 }
4694 
4695 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4696 			bool line_status)
4697 {
4698 	if (!irqchip_in_kernel(kvm))
4699 		return -ENXIO;
4700 
4701 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4702 					irq_event->irq, irq_event->level,
4703 					line_status);
4704 	return 0;
4705 }
4706 
4707 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4708 			    struct kvm_enable_cap *cap)
4709 {
4710 	int r;
4711 
4712 	if (cap->flags)
4713 		return -EINVAL;
4714 
4715 	switch (cap->cap) {
4716 	case KVM_CAP_DISABLE_QUIRKS:
4717 		kvm->arch.disabled_quirks = cap->args[0];
4718 		r = 0;
4719 		break;
4720 	case KVM_CAP_SPLIT_IRQCHIP: {
4721 		mutex_lock(&kvm->lock);
4722 		r = -EINVAL;
4723 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4724 			goto split_irqchip_unlock;
4725 		r = -EEXIST;
4726 		if (irqchip_in_kernel(kvm))
4727 			goto split_irqchip_unlock;
4728 		if (kvm->created_vcpus)
4729 			goto split_irqchip_unlock;
4730 		r = kvm_setup_empty_irq_routing(kvm);
4731 		if (r)
4732 			goto split_irqchip_unlock;
4733 		/* Pairs with irqchip_in_kernel. */
4734 		smp_wmb();
4735 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4736 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4737 		r = 0;
4738 split_irqchip_unlock:
4739 		mutex_unlock(&kvm->lock);
4740 		break;
4741 	}
4742 	case KVM_CAP_X2APIC_API:
4743 		r = -EINVAL;
4744 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4745 			break;
4746 
4747 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4748 			kvm->arch.x2apic_format = true;
4749 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4750 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4751 
4752 		r = 0;
4753 		break;
4754 	case KVM_CAP_X86_DISABLE_EXITS:
4755 		r = -EINVAL;
4756 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4757 			break;
4758 
4759 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4760 			kvm_can_mwait_in_guest())
4761 			kvm->arch.mwait_in_guest = true;
4762 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4763 			kvm->arch.hlt_in_guest = true;
4764 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4765 			kvm->arch.pause_in_guest = true;
4766 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4767 			kvm->arch.cstate_in_guest = true;
4768 		r = 0;
4769 		break;
4770 	case KVM_CAP_MSR_PLATFORM_INFO:
4771 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4772 		r = 0;
4773 		break;
4774 	case KVM_CAP_EXCEPTION_PAYLOAD:
4775 		kvm->arch.exception_payload_enabled = cap->args[0];
4776 		r = 0;
4777 		break;
4778 	default:
4779 		r = -EINVAL;
4780 		break;
4781 	}
4782 	return r;
4783 }
4784 
4785 long kvm_arch_vm_ioctl(struct file *filp,
4786 		       unsigned int ioctl, unsigned long arg)
4787 {
4788 	struct kvm *kvm = filp->private_data;
4789 	void __user *argp = (void __user *)arg;
4790 	int r = -ENOTTY;
4791 	/*
4792 	 * This union makes it completely explicit to gcc-3.x
4793 	 * that these two variables' stack usage should be
4794 	 * combined, not added together.
4795 	 */
4796 	union {
4797 		struct kvm_pit_state ps;
4798 		struct kvm_pit_state2 ps2;
4799 		struct kvm_pit_config pit_config;
4800 	} u;
4801 
4802 	switch (ioctl) {
4803 	case KVM_SET_TSS_ADDR:
4804 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4805 		break;
4806 	case KVM_SET_IDENTITY_MAP_ADDR: {
4807 		u64 ident_addr;
4808 
4809 		mutex_lock(&kvm->lock);
4810 		r = -EINVAL;
4811 		if (kvm->created_vcpus)
4812 			goto set_identity_unlock;
4813 		r = -EFAULT;
4814 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4815 			goto set_identity_unlock;
4816 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4817 set_identity_unlock:
4818 		mutex_unlock(&kvm->lock);
4819 		break;
4820 	}
4821 	case KVM_SET_NR_MMU_PAGES:
4822 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4823 		break;
4824 	case KVM_GET_NR_MMU_PAGES:
4825 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4826 		break;
4827 	case KVM_CREATE_IRQCHIP: {
4828 		mutex_lock(&kvm->lock);
4829 
4830 		r = -EEXIST;
4831 		if (irqchip_in_kernel(kvm))
4832 			goto create_irqchip_unlock;
4833 
4834 		r = -EINVAL;
4835 		if (kvm->created_vcpus)
4836 			goto create_irqchip_unlock;
4837 
4838 		r = kvm_pic_init(kvm);
4839 		if (r)
4840 			goto create_irqchip_unlock;
4841 
4842 		r = kvm_ioapic_init(kvm);
4843 		if (r) {
4844 			kvm_pic_destroy(kvm);
4845 			goto create_irqchip_unlock;
4846 		}
4847 
4848 		r = kvm_setup_default_irq_routing(kvm);
4849 		if (r) {
4850 			kvm_ioapic_destroy(kvm);
4851 			kvm_pic_destroy(kvm);
4852 			goto create_irqchip_unlock;
4853 		}
4854 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4855 		smp_wmb();
4856 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4857 	create_irqchip_unlock:
4858 		mutex_unlock(&kvm->lock);
4859 		break;
4860 	}
4861 	case KVM_CREATE_PIT:
4862 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4863 		goto create_pit;
4864 	case KVM_CREATE_PIT2:
4865 		r = -EFAULT;
4866 		if (copy_from_user(&u.pit_config, argp,
4867 				   sizeof(struct kvm_pit_config)))
4868 			goto out;
4869 	create_pit:
4870 		mutex_lock(&kvm->lock);
4871 		r = -EEXIST;
4872 		if (kvm->arch.vpit)
4873 			goto create_pit_unlock;
4874 		r = -ENOMEM;
4875 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4876 		if (kvm->arch.vpit)
4877 			r = 0;
4878 	create_pit_unlock:
4879 		mutex_unlock(&kvm->lock);
4880 		break;
4881 	case KVM_GET_IRQCHIP: {
4882 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4883 		struct kvm_irqchip *chip;
4884 
4885 		chip = memdup_user(argp, sizeof(*chip));
4886 		if (IS_ERR(chip)) {
4887 			r = PTR_ERR(chip);
4888 			goto out;
4889 		}
4890 
4891 		r = -ENXIO;
4892 		if (!irqchip_kernel(kvm))
4893 			goto get_irqchip_out;
4894 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4895 		if (r)
4896 			goto get_irqchip_out;
4897 		r = -EFAULT;
4898 		if (copy_to_user(argp, chip, sizeof(*chip)))
4899 			goto get_irqchip_out;
4900 		r = 0;
4901 	get_irqchip_out:
4902 		kfree(chip);
4903 		break;
4904 	}
4905 	case KVM_SET_IRQCHIP: {
4906 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4907 		struct kvm_irqchip *chip;
4908 
4909 		chip = memdup_user(argp, sizeof(*chip));
4910 		if (IS_ERR(chip)) {
4911 			r = PTR_ERR(chip);
4912 			goto out;
4913 		}
4914 
4915 		r = -ENXIO;
4916 		if (!irqchip_kernel(kvm))
4917 			goto set_irqchip_out;
4918 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4919 		if (r)
4920 			goto set_irqchip_out;
4921 		r = 0;
4922 	set_irqchip_out:
4923 		kfree(chip);
4924 		break;
4925 	}
4926 	case KVM_GET_PIT: {
4927 		r = -EFAULT;
4928 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4929 			goto out;
4930 		r = -ENXIO;
4931 		if (!kvm->arch.vpit)
4932 			goto out;
4933 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4934 		if (r)
4935 			goto out;
4936 		r = -EFAULT;
4937 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4938 			goto out;
4939 		r = 0;
4940 		break;
4941 	}
4942 	case KVM_SET_PIT: {
4943 		r = -EFAULT;
4944 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4945 			goto out;
4946 		r = -ENXIO;
4947 		if (!kvm->arch.vpit)
4948 			goto out;
4949 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4950 		break;
4951 	}
4952 	case KVM_GET_PIT2: {
4953 		r = -ENXIO;
4954 		if (!kvm->arch.vpit)
4955 			goto out;
4956 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4957 		if (r)
4958 			goto out;
4959 		r = -EFAULT;
4960 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4961 			goto out;
4962 		r = 0;
4963 		break;
4964 	}
4965 	case KVM_SET_PIT2: {
4966 		r = -EFAULT;
4967 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4968 			goto out;
4969 		r = -ENXIO;
4970 		if (!kvm->arch.vpit)
4971 			goto out;
4972 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4973 		break;
4974 	}
4975 	case KVM_REINJECT_CONTROL: {
4976 		struct kvm_reinject_control control;
4977 		r =  -EFAULT;
4978 		if (copy_from_user(&control, argp, sizeof(control)))
4979 			goto out;
4980 		r = kvm_vm_ioctl_reinject(kvm, &control);
4981 		break;
4982 	}
4983 	case KVM_SET_BOOT_CPU_ID:
4984 		r = 0;
4985 		mutex_lock(&kvm->lock);
4986 		if (kvm->created_vcpus)
4987 			r = -EBUSY;
4988 		else
4989 			kvm->arch.bsp_vcpu_id = arg;
4990 		mutex_unlock(&kvm->lock);
4991 		break;
4992 	case KVM_XEN_HVM_CONFIG: {
4993 		struct kvm_xen_hvm_config xhc;
4994 		r = -EFAULT;
4995 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4996 			goto out;
4997 		r = -EINVAL;
4998 		if (xhc.flags)
4999 			goto out;
5000 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5001 		r = 0;
5002 		break;
5003 	}
5004 	case KVM_SET_CLOCK: {
5005 		struct kvm_clock_data user_ns;
5006 		u64 now_ns;
5007 
5008 		r = -EFAULT;
5009 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5010 			goto out;
5011 
5012 		r = -EINVAL;
5013 		if (user_ns.flags)
5014 			goto out;
5015 
5016 		r = 0;
5017 		/*
5018 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5019 		 * kvm_gen_update_masterclock() can be cut down to locked
5020 		 * pvclock_update_vm_gtod_copy().
5021 		 */
5022 		kvm_gen_update_masterclock(kvm);
5023 		now_ns = get_kvmclock_ns(kvm);
5024 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5025 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5026 		break;
5027 	}
5028 	case KVM_GET_CLOCK: {
5029 		struct kvm_clock_data user_ns;
5030 		u64 now_ns;
5031 
5032 		now_ns = get_kvmclock_ns(kvm);
5033 		user_ns.clock = now_ns;
5034 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5035 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5036 
5037 		r = -EFAULT;
5038 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5039 			goto out;
5040 		r = 0;
5041 		break;
5042 	}
5043 	case KVM_MEMORY_ENCRYPT_OP: {
5044 		r = -ENOTTY;
5045 		if (kvm_x86_ops->mem_enc_op)
5046 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
5047 		break;
5048 	}
5049 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5050 		struct kvm_enc_region region;
5051 
5052 		r = -EFAULT;
5053 		if (copy_from_user(&region, argp, sizeof(region)))
5054 			goto out;
5055 
5056 		r = -ENOTTY;
5057 		if (kvm_x86_ops->mem_enc_reg_region)
5058 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
5059 		break;
5060 	}
5061 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5062 		struct kvm_enc_region region;
5063 
5064 		r = -EFAULT;
5065 		if (copy_from_user(&region, argp, sizeof(region)))
5066 			goto out;
5067 
5068 		r = -ENOTTY;
5069 		if (kvm_x86_ops->mem_enc_unreg_region)
5070 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
5071 		break;
5072 	}
5073 	case KVM_HYPERV_EVENTFD: {
5074 		struct kvm_hyperv_eventfd hvevfd;
5075 
5076 		r = -EFAULT;
5077 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5078 			goto out;
5079 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5080 		break;
5081 	}
5082 	case KVM_SET_PMU_EVENT_FILTER:
5083 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5084 		break;
5085 	default:
5086 		r = -ENOTTY;
5087 	}
5088 out:
5089 	return r;
5090 }
5091 
5092 static void kvm_init_msr_list(void)
5093 {
5094 	struct x86_pmu_capability x86_pmu;
5095 	u32 dummy[2];
5096 	unsigned i, j;
5097 
5098 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5099 			 "Please update the fixed PMCs in msrs_to_save[]");
5100 
5101 	perf_get_x86_pmu_capability(&x86_pmu);
5102 
5103 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5104 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5105 			continue;
5106 
5107 		/*
5108 		 * Even MSRs that are valid in the host may not be exposed
5109 		 * to the guests in some cases.
5110 		 */
5111 		switch (msrs_to_save[i]) {
5112 		case MSR_IA32_BNDCFGS:
5113 			if (!kvm_mpx_supported())
5114 				continue;
5115 			break;
5116 		case MSR_TSC_AUX:
5117 			if (!kvm_x86_ops->rdtscp_supported())
5118 				continue;
5119 			break;
5120 		case MSR_IA32_RTIT_CTL:
5121 		case MSR_IA32_RTIT_STATUS:
5122 			if (!kvm_x86_ops->pt_supported())
5123 				continue;
5124 			break;
5125 		case MSR_IA32_RTIT_CR3_MATCH:
5126 			if (!kvm_x86_ops->pt_supported() ||
5127 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5128 				continue;
5129 			break;
5130 		case MSR_IA32_RTIT_OUTPUT_BASE:
5131 		case MSR_IA32_RTIT_OUTPUT_MASK:
5132 			if (!kvm_x86_ops->pt_supported() ||
5133 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5134 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5135 				continue;
5136 			break;
5137 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5138 			if (!kvm_x86_ops->pt_supported() ||
5139 				msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5140 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5141 				continue;
5142 			break;
5143 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5144 			if (msrs_to_save[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5145 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5146 				continue;
5147 			break;
5148 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5149 			if (msrs_to_save[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5150 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5151 				continue;
5152 		}
5153 		default:
5154 			break;
5155 		}
5156 
5157 		if (j < i)
5158 			msrs_to_save[j] = msrs_to_save[i];
5159 		j++;
5160 	}
5161 	num_msrs_to_save = j;
5162 
5163 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5164 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5165 			continue;
5166 
5167 		if (j < i)
5168 			emulated_msrs[j] = emulated_msrs[i];
5169 		j++;
5170 	}
5171 	num_emulated_msrs = j;
5172 
5173 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5174 		struct kvm_msr_entry msr;
5175 
5176 		msr.index = msr_based_features[i];
5177 		if (kvm_get_msr_feature(&msr))
5178 			continue;
5179 
5180 		if (j < i)
5181 			msr_based_features[j] = msr_based_features[i];
5182 		j++;
5183 	}
5184 	num_msr_based_features = j;
5185 }
5186 
5187 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5188 			   const void *v)
5189 {
5190 	int handled = 0;
5191 	int n;
5192 
5193 	do {
5194 		n = min(len, 8);
5195 		if (!(lapic_in_kernel(vcpu) &&
5196 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5197 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5198 			break;
5199 		handled += n;
5200 		addr += n;
5201 		len -= n;
5202 		v += n;
5203 	} while (len);
5204 
5205 	return handled;
5206 }
5207 
5208 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5209 {
5210 	int handled = 0;
5211 	int n;
5212 
5213 	do {
5214 		n = min(len, 8);
5215 		if (!(lapic_in_kernel(vcpu) &&
5216 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5217 					 addr, n, v))
5218 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5219 			break;
5220 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5221 		handled += n;
5222 		addr += n;
5223 		len -= n;
5224 		v += n;
5225 	} while (len);
5226 
5227 	return handled;
5228 }
5229 
5230 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5231 			struct kvm_segment *var, int seg)
5232 {
5233 	kvm_x86_ops->set_segment(vcpu, var, seg);
5234 }
5235 
5236 void kvm_get_segment(struct kvm_vcpu *vcpu,
5237 		     struct kvm_segment *var, int seg)
5238 {
5239 	kvm_x86_ops->get_segment(vcpu, var, seg);
5240 }
5241 
5242 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5243 			   struct x86_exception *exception)
5244 {
5245 	gpa_t t_gpa;
5246 
5247 	BUG_ON(!mmu_is_nested(vcpu));
5248 
5249 	/* NPT walks are always user-walks */
5250 	access |= PFERR_USER_MASK;
5251 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5252 
5253 	return t_gpa;
5254 }
5255 
5256 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5257 			      struct x86_exception *exception)
5258 {
5259 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5260 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5261 }
5262 
5263  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5264 				struct x86_exception *exception)
5265 {
5266 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5267 	access |= PFERR_FETCH_MASK;
5268 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5269 }
5270 
5271 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5272 			       struct x86_exception *exception)
5273 {
5274 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5275 	access |= PFERR_WRITE_MASK;
5276 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5277 }
5278 
5279 /* uses this to access any guest's mapped memory without checking CPL */
5280 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5281 				struct x86_exception *exception)
5282 {
5283 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5284 }
5285 
5286 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5287 				      struct kvm_vcpu *vcpu, u32 access,
5288 				      struct x86_exception *exception)
5289 {
5290 	void *data = val;
5291 	int r = X86EMUL_CONTINUE;
5292 
5293 	while (bytes) {
5294 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5295 							    exception);
5296 		unsigned offset = addr & (PAGE_SIZE-1);
5297 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5298 		int ret;
5299 
5300 		if (gpa == UNMAPPED_GVA)
5301 			return X86EMUL_PROPAGATE_FAULT;
5302 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5303 					       offset, toread);
5304 		if (ret < 0) {
5305 			r = X86EMUL_IO_NEEDED;
5306 			goto out;
5307 		}
5308 
5309 		bytes -= toread;
5310 		data += toread;
5311 		addr += toread;
5312 	}
5313 out:
5314 	return r;
5315 }
5316 
5317 /* used for instruction fetching */
5318 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5319 				gva_t addr, void *val, unsigned int bytes,
5320 				struct x86_exception *exception)
5321 {
5322 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5323 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5324 	unsigned offset;
5325 	int ret;
5326 
5327 	/* Inline kvm_read_guest_virt_helper for speed.  */
5328 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5329 						    exception);
5330 	if (unlikely(gpa == UNMAPPED_GVA))
5331 		return X86EMUL_PROPAGATE_FAULT;
5332 
5333 	offset = addr & (PAGE_SIZE-1);
5334 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5335 		bytes = (unsigned)PAGE_SIZE - offset;
5336 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5337 				       offset, bytes);
5338 	if (unlikely(ret < 0))
5339 		return X86EMUL_IO_NEEDED;
5340 
5341 	return X86EMUL_CONTINUE;
5342 }
5343 
5344 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5345 			       gva_t addr, void *val, unsigned int bytes,
5346 			       struct x86_exception *exception)
5347 {
5348 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5349 
5350 	/*
5351 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5352 	 * is returned, but our callers are not ready for that and they blindly
5353 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5354 	 * uninitialized kernel stack memory into cr2 and error code.
5355 	 */
5356 	memset(exception, 0, sizeof(*exception));
5357 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5358 					  exception);
5359 }
5360 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5361 
5362 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5363 			     gva_t addr, void *val, unsigned int bytes,
5364 			     struct x86_exception *exception, bool system)
5365 {
5366 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5367 	u32 access = 0;
5368 
5369 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5370 		access |= PFERR_USER_MASK;
5371 
5372 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5373 }
5374 
5375 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5376 		unsigned long addr, void *val, unsigned int bytes)
5377 {
5378 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5379 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5380 
5381 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5382 }
5383 
5384 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5385 				      struct kvm_vcpu *vcpu, u32 access,
5386 				      struct x86_exception *exception)
5387 {
5388 	void *data = val;
5389 	int r = X86EMUL_CONTINUE;
5390 
5391 	while (bytes) {
5392 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5393 							     access,
5394 							     exception);
5395 		unsigned offset = addr & (PAGE_SIZE-1);
5396 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5397 		int ret;
5398 
5399 		if (gpa == UNMAPPED_GVA)
5400 			return X86EMUL_PROPAGATE_FAULT;
5401 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5402 		if (ret < 0) {
5403 			r = X86EMUL_IO_NEEDED;
5404 			goto out;
5405 		}
5406 
5407 		bytes -= towrite;
5408 		data += towrite;
5409 		addr += towrite;
5410 	}
5411 out:
5412 	return r;
5413 }
5414 
5415 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5416 			      unsigned int bytes, struct x86_exception *exception,
5417 			      bool system)
5418 {
5419 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5420 	u32 access = PFERR_WRITE_MASK;
5421 
5422 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5423 		access |= PFERR_USER_MASK;
5424 
5425 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5426 					   access, exception);
5427 }
5428 
5429 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5430 				unsigned int bytes, struct x86_exception *exception)
5431 {
5432 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5433 	vcpu->arch.l1tf_flush_l1d = true;
5434 
5435 	/*
5436 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5437 	 * is returned, but our callers are not ready for that and they blindly
5438 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5439 	 * uninitialized kernel stack memory into cr2 and error code.
5440 	 */
5441 	memset(exception, 0, sizeof(*exception));
5442 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5443 					   PFERR_WRITE_MASK, exception);
5444 }
5445 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5446 
5447 int handle_ud(struct kvm_vcpu *vcpu)
5448 {
5449 	int emul_type = EMULTYPE_TRAP_UD;
5450 	char sig[5]; /* ud2; .ascii "kvm" */
5451 	struct x86_exception e;
5452 
5453 	if (force_emulation_prefix &&
5454 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5455 				sig, sizeof(sig), &e) == 0 &&
5456 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5457 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5458 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5459 	}
5460 
5461 	return kvm_emulate_instruction(vcpu, emul_type);
5462 }
5463 EXPORT_SYMBOL_GPL(handle_ud);
5464 
5465 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5466 			    gpa_t gpa, bool write)
5467 {
5468 	/* For APIC access vmexit */
5469 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5470 		return 1;
5471 
5472 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5473 		trace_vcpu_match_mmio(gva, gpa, write, true);
5474 		return 1;
5475 	}
5476 
5477 	return 0;
5478 }
5479 
5480 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5481 				gpa_t *gpa, struct x86_exception *exception,
5482 				bool write)
5483 {
5484 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5485 		| (write ? PFERR_WRITE_MASK : 0);
5486 
5487 	/*
5488 	 * currently PKRU is only applied to ept enabled guest so
5489 	 * there is no pkey in EPT page table for L1 guest or EPT
5490 	 * shadow page table for L2 guest.
5491 	 */
5492 	if (vcpu_match_mmio_gva(vcpu, gva)
5493 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5494 				 vcpu->arch.mmio_access, 0, access)) {
5495 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5496 					(gva & (PAGE_SIZE - 1));
5497 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5498 		return 1;
5499 	}
5500 
5501 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5502 
5503 	if (*gpa == UNMAPPED_GVA)
5504 		return -1;
5505 
5506 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5507 }
5508 
5509 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5510 			const void *val, int bytes)
5511 {
5512 	int ret;
5513 
5514 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5515 	if (ret < 0)
5516 		return 0;
5517 	kvm_page_track_write(vcpu, gpa, val, bytes);
5518 	return 1;
5519 }
5520 
5521 struct read_write_emulator_ops {
5522 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5523 				  int bytes);
5524 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5525 				  void *val, int bytes);
5526 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5527 			       int bytes, void *val);
5528 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5529 				    void *val, int bytes);
5530 	bool write;
5531 };
5532 
5533 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5534 {
5535 	if (vcpu->mmio_read_completed) {
5536 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5537 			       vcpu->mmio_fragments[0].gpa, val);
5538 		vcpu->mmio_read_completed = 0;
5539 		return 1;
5540 	}
5541 
5542 	return 0;
5543 }
5544 
5545 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5546 			void *val, int bytes)
5547 {
5548 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5549 }
5550 
5551 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5552 			 void *val, int bytes)
5553 {
5554 	return emulator_write_phys(vcpu, gpa, val, bytes);
5555 }
5556 
5557 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5558 {
5559 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5560 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5561 }
5562 
5563 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5564 			  void *val, int bytes)
5565 {
5566 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5567 	return X86EMUL_IO_NEEDED;
5568 }
5569 
5570 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5571 			   void *val, int bytes)
5572 {
5573 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5574 
5575 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5576 	return X86EMUL_CONTINUE;
5577 }
5578 
5579 static const struct read_write_emulator_ops read_emultor = {
5580 	.read_write_prepare = read_prepare,
5581 	.read_write_emulate = read_emulate,
5582 	.read_write_mmio = vcpu_mmio_read,
5583 	.read_write_exit_mmio = read_exit_mmio,
5584 };
5585 
5586 static const struct read_write_emulator_ops write_emultor = {
5587 	.read_write_emulate = write_emulate,
5588 	.read_write_mmio = write_mmio,
5589 	.read_write_exit_mmio = write_exit_mmio,
5590 	.write = true,
5591 };
5592 
5593 static int emulator_read_write_onepage(unsigned long addr, void *val,
5594 				       unsigned int bytes,
5595 				       struct x86_exception *exception,
5596 				       struct kvm_vcpu *vcpu,
5597 				       const struct read_write_emulator_ops *ops)
5598 {
5599 	gpa_t gpa;
5600 	int handled, ret;
5601 	bool write = ops->write;
5602 	struct kvm_mmio_fragment *frag;
5603 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5604 
5605 	/*
5606 	 * If the exit was due to a NPF we may already have a GPA.
5607 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5608 	 * Note, this cannot be used on string operations since string
5609 	 * operation using rep will only have the initial GPA from the NPF
5610 	 * occurred.
5611 	 */
5612 	if (vcpu->arch.gpa_available &&
5613 	    emulator_can_use_gpa(ctxt) &&
5614 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5615 		gpa = vcpu->arch.gpa_val;
5616 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5617 	} else {
5618 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5619 		if (ret < 0)
5620 			return X86EMUL_PROPAGATE_FAULT;
5621 	}
5622 
5623 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5624 		return X86EMUL_CONTINUE;
5625 
5626 	/*
5627 	 * Is this MMIO handled locally?
5628 	 */
5629 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5630 	if (handled == bytes)
5631 		return X86EMUL_CONTINUE;
5632 
5633 	gpa += handled;
5634 	bytes -= handled;
5635 	val += handled;
5636 
5637 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5638 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5639 	frag->gpa = gpa;
5640 	frag->data = val;
5641 	frag->len = bytes;
5642 	return X86EMUL_CONTINUE;
5643 }
5644 
5645 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5646 			unsigned long addr,
5647 			void *val, unsigned int bytes,
5648 			struct x86_exception *exception,
5649 			const struct read_write_emulator_ops *ops)
5650 {
5651 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5652 	gpa_t gpa;
5653 	int rc;
5654 
5655 	if (ops->read_write_prepare &&
5656 		  ops->read_write_prepare(vcpu, val, bytes))
5657 		return X86EMUL_CONTINUE;
5658 
5659 	vcpu->mmio_nr_fragments = 0;
5660 
5661 	/* Crossing a page boundary? */
5662 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5663 		int now;
5664 
5665 		now = -addr & ~PAGE_MASK;
5666 		rc = emulator_read_write_onepage(addr, val, now, exception,
5667 						 vcpu, ops);
5668 
5669 		if (rc != X86EMUL_CONTINUE)
5670 			return rc;
5671 		addr += now;
5672 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5673 			addr = (u32)addr;
5674 		val += now;
5675 		bytes -= now;
5676 	}
5677 
5678 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5679 					 vcpu, ops);
5680 	if (rc != X86EMUL_CONTINUE)
5681 		return rc;
5682 
5683 	if (!vcpu->mmio_nr_fragments)
5684 		return rc;
5685 
5686 	gpa = vcpu->mmio_fragments[0].gpa;
5687 
5688 	vcpu->mmio_needed = 1;
5689 	vcpu->mmio_cur_fragment = 0;
5690 
5691 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5692 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5693 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5694 	vcpu->run->mmio.phys_addr = gpa;
5695 
5696 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5697 }
5698 
5699 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5700 				  unsigned long addr,
5701 				  void *val,
5702 				  unsigned int bytes,
5703 				  struct x86_exception *exception)
5704 {
5705 	return emulator_read_write(ctxt, addr, val, bytes,
5706 				   exception, &read_emultor);
5707 }
5708 
5709 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5710 			    unsigned long addr,
5711 			    const void *val,
5712 			    unsigned int bytes,
5713 			    struct x86_exception *exception)
5714 {
5715 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5716 				   exception, &write_emultor);
5717 }
5718 
5719 #define CMPXCHG_TYPE(t, ptr, old, new) \
5720 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5721 
5722 #ifdef CONFIG_X86_64
5723 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5724 #else
5725 #  define CMPXCHG64(ptr, old, new) \
5726 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5727 #endif
5728 
5729 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5730 				     unsigned long addr,
5731 				     const void *old,
5732 				     const void *new,
5733 				     unsigned int bytes,
5734 				     struct x86_exception *exception)
5735 {
5736 	struct kvm_host_map map;
5737 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5738 	gpa_t gpa;
5739 	char *kaddr;
5740 	bool exchanged;
5741 
5742 	/* guests cmpxchg8b have to be emulated atomically */
5743 	if (bytes > 8 || (bytes & (bytes - 1)))
5744 		goto emul_write;
5745 
5746 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5747 
5748 	if (gpa == UNMAPPED_GVA ||
5749 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5750 		goto emul_write;
5751 
5752 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5753 		goto emul_write;
5754 
5755 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5756 		goto emul_write;
5757 
5758 	kaddr = map.hva + offset_in_page(gpa);
5759 
5760 	switch (bytes) {
5761 	case 1:
5762 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5763 		break;
5764 	case 2:
5765 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5766 		break;
5767 	case 4:
5768 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5769 		break;
5770 	case 8:
5771 		exchanged = CMPXCHG64(kaddr, old, new);
5772 		break;
5773 	default:
5774 		BUG();
5775 	}
5776 
5777 	kvm_vcpu_unmap(vcpu, &map, true);
5778 
5779 	if (!exchanged)
5780 		return X86EMUL_CMPXCHG_FAILED;
5781 
5782 	kvm_page_track_write(vcpu, gpa, new, bytes);
5783 
5784 	return X86EMUL_CONTINUE;
5785 
5786 emul_write:
5787 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5788 
5789 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5790 }
5791 
5792 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5793 {
5794 	int r = 0, i;
5795 
5796 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5797 		if (vcpu->arch.pio.in)
5798 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5799 					    vcpu->arch.pio.size, pd);
5800 		else
5801 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5802 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5803 					     pd);
5804 		if (r)
5805 			break;
5806 		pd += vcpu->arch.pio.size;
5807 	}
5808 	return r;
5809 }
5810 
5811 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5812 			       unsigned short port, void *val,
5813 			       unsigned int count, bool in)
5814 {
5815 	vcpu->arch.pio.port = port;
5816 	vcpu->arch.pio.in = in;
5817 	vcpu->arch.pio.count  = count;
5818 	vcpu->arch.pio.size = size;
5819 
5820 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5821 		vcpu->arch.pio.count = 0;
5822 		return 1;
5823 	}
5824 
5825 	vcpu->run->exit_reason = KVM_EXIT_IO;
5826 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5827 	vcpu->run->io.size = size;
5828 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5829 	vcpu->run->io.count = count;
5830 	vcpu->run->io.port = port;
5831 
5832 	return 0;
5833 }
5834 
5835 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5836 				    int size, unsigned short port, void *val,
5837 				    unsigned int count)
5838 {
5839 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5840 	int ret;
5841 
5842 	if (vcpu->arch.pio.count)
5843 		goto data_avail;
5844 
5845 	memset(vcpu->arch.pio_data, 0, size * count);
5846 
5847 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5848 	if (ret) {
5849 data_avail:
5850 		memcpy(val, vcpu->arch.pio_data, size * count);
5851 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5852 		vcpu->arch.pio.count = 0;
5853 		return 1;
5854 	}
5855 
5856 	return 0;
5857 }
5858 
5859 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5860 				     int size, unsigned short port,
5861 				     const void *val, unsigned int count)
5862 {
5863 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5864 
5865 	memcpy(vcpu->arch.pio_data, val, size * count);
5866 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5867 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5868 }
5869 
5870 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5871 {
5872 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5873 }
5874 
5875 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5876 {
5877 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5878 }
5879 
5880 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5881 {
5882 	if (!need_emulate_wbinvd(vcpu))
5883 		return X86EMUL_CONTINUE;
5884 
5885 	if (kvm_x86_ops->has_wbinvd_exit()) {
5886 		int cpu = get_cpu();
5887 
5888 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5889 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5890 				wbinvd_ipi, NULL, 1);
5891 		put_cpu();
5892 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5893 	} else
5894 		wbinvd();
5895 	return X86EMUL_CONTINUE;
5896 }
5897 
5898 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5899 {
5900 	kvm_emulate_wbinvd_noskip(vcpu);
5901 	return kvm_skip_emulated_instruction(vcpu);
5902 }
5903 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5904 
5905 
5906 
5907 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5908 {
5909 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5910 }
5911 
5912 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5913 			   unsigned long *dest)
5914 {
5915 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5916 }
5917 
5918 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5919 			   unsigned long value)
5920 {
5921 
5922 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5923 }
5924 
5925 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5926 {
5927 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5928 }
5929 
5930 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5931 {
5932 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5933 	unsigned long value;
5934 
5935 	switch (cr) {
5936 	case 0:
5937 		value = kvm_read_cr0(vcpu);
5938 		break;
5939 	case 2:
5940 		value = vcpu->arch.cr2;
5941 		break;
5942 	case 3:
5943 		value = kvm_read_cr3(vcpu);
5944 		break;
5945 	case 4:
5946 		value = kvm_read_cr4(vcpu);
5947 		break;
5948 	case 8:
5949 		value = kvm_get_cr8(vcpu);
5950 		break;
5951 	default:
5952 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5953 		return 0;
5954 	}
5955 
5956 	return value;
5957 }
5958 
5959 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5960 {
5961 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5962 	int res = 0;
5963 
5964 	switch (cr) {
5965 	case 0:
5966 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5967 		break;
5968 	case 2:
5969 		vcpu->arch.cr2 = val;
5970 		break;
5971 	case 3:
5972 		res = kvm_set_cr3(vcpu, val);
5973 		break;
5974 	case 4:
5975 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5976 		break;
5977 	case 8:
5978 		res = kvm_set_cr8(vcpu, val);
5979 		break;
5980 	default:
5981 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5982 		res = -1;
5983 	}
5984 
5985 	return res;
5986 }
5987 
5988 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5989 {
5990 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5991 }
5992 
5993 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5994 {
5995 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5996 }
5997 
5998 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5999 {
6000 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6001 }
6002 
6003 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6004 {
6005 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6006 }
6007 
6008 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6009 {
6010 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6011 }
6012 
6013 static unsigned long emulator_get_cached_segment_base(
6014 	struct x86_emulate_ctxt *ctxt, int seg)
6015 {
6016 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6017 }
6018 
6019 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6020 				 struct desc_struct *desc, u32 *base3,
6021 				 int seg)
6022 {
6023 	struct kvm_segment var;
6024 
6025 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6026 	*selector = var.selector;
6027 
6028 	if (var.unusable) {
6029 		memset(desc, 0, sizeof(*desc));
6030 		if (base3)
6031 			*base3 = 0;
6032 		return false;
6033 	}
6034 
6035 	if (var.g)
6036 		var.limit >>= 12;
6037 	set_desc_limit(desc, var.limit);
6038 	set_desc_base(desc, (unsigned long)var.base);
6039 #ifdef CONFIG_X86_64
6040 	if (base3)
6041 		*base3 = var.base >> 32;
6042 #endif
6043 	desc->type = var.type;
6044 	desc->s = var.s;
6045 	desc->dpl = var.dpl;
6046 	desc->p = var.present;
6047 	desc->avl = var.avl;
6048 	desc->l = var.l;
6049 	desc->d = var.db;
6050 	desc->g = var.g;
6051 
6052 	return true;
6053 }
6054 
6055 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6056 				 struct desc_struct *desc, u32 base3,
6057 				 int seg)
6058 {
6059 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6060 	struct kvm_segment var;
6061 
6062 	var.selector = selector;
6063 	var.base = get_desc_base(desc);
6064 #ifdef CONFIG_X86_64
6065 	var.base |= ((u64)base3) << 32;
6066 #endif
6067 	var.limit = get_desc_limit(desc);
6068 	if (desc->g)
6069 		var.limit = (var.limit << 12) | 0xfff;
6070 	var.type = desc->type;
6071 	var.dpl = desc->dpl;
6072 	var.db = desc->d;
6073 	var.s = desc->s;
6074 	var.l = desc->l;
6075 	var.g = desc->g;
6076 	var.avl = desc->avl;
6077 	var.present = desc->p;
6078 	var.unusable = !var.present;
6079 	var.padding = 0;
6080 
6081 	kvm_set_segment(vcpu, &var, seg);
6082 	return;
6083 }
6084 
6085 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6086 			    u32 msr_index, u64 *pdata)
6087 {
6088 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6089 }
6090 
6091 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6092 			    u32 msr_index, u64 data)
6093 {
6094 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6095 }
6096 
6097 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6098 {
6099 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6100 
6101 	return vcpu->arch.smbase;
6102 }
6103 
6104 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6105 {
6106 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6107 
6108 	vcpu->arch.smbase = smbase;
6109 }
6110 
6111 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6112 			      u32 pmc)
6113 {
6114 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6115 }
6116 
6117 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6118 			     u32 pmc, u64 *pdata)
6119 {
6120 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6121 }
6122 
6123 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6124 {
6125 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6126 }
6127 
6128 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6129 			      struct x86_instruction_info *info,
6130 			      enum x86_intercept_stage stage)
6131 {
6132 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6133 }
6134 
6135 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6136 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6137 {
6138 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6139 }
6140 
6141 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6142 {
6143 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6144 }
6145 
6146 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6147 {
6148 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6149 }
6150 
6151 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6152 {
6153 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6154 }
6155 
6156 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6157 {
6158 	return emul_to_vcpu(ctxt)->arch.hflags;
6159 }
6160 
6161 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6162 {
6163 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6164 }
6165 
6166 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6167 				  const char *smstate)
6168 {
6169 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6170 }
6171 
6172 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6173 {
6174 	kvm_smm_changed(emul_to_vcpu(ctxt));
6175 }
6176 
6177 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6178 {
6179 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6180 }
6181 
6182 static const struct x86_emulate_ops emulate_ops = {
6183 	.read_gpr            = emulator_read_gpr,
6184 	.write_gpr           = emulator_write_gpr,
6185 	.read_std            = emulator_read_std,
6186 	.write_std           = emulator_write_std,
6187 	.read_phys           = kvm_read_guest_phys_system,
6188 	.fetch               = kvm_fetch_guest_virt,
6189 	.read_emulated       = emulator_read_emulated,
6190 	.write_emulated      = emulator_write_emulated,
6191 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6192 	.invlpg              = emulator_invlpg,
6193 	.pio_in_emulated     = emulator_pio_in_emulated,
6194 	.pio_out_emulated    = emulator_pio_out_emulated,
6195 	.get_segment         = emulator_get_segment,
6196 	.set_segment         = emulator_set_segment,
6197 	.get_cached_segment_base = emulator_get_cached_segment_base,
6198 	.get_gdt             = emulator_get_gdt,
6199 	.get_idt	     = emulator_get_idt,
6200 	.set_gdt             = emulator_set_gdt,
6201 	.set_idt	     = emulator_set_idt,
6202 	.get_cr              = emulator_get_cr,
6203 	.set_cr              = emulator_set_cr,
6204 	.cpl                 = emulator_get_cpl,
6205 	.get_dr              = emulator_get_dr,
6206 	.set_dr              = emulator_set_dr,
6207 	.get_smbase          = emulator_get_smbase,
6208 	.set_smbase          = emulator_set_smbase,
6209 	.set_msr             = emulator_set_msr,
6210 	.get_msr             = emulator_get_msr,
6211 	.check_pmc	     = emulator_check_pmc,
6212 	.read_pmc            = emulator_read_pmc,
6213 	.halt                = emulator_halt,
6214 	.wbinvd              = emulator_wbinvd,
6215 	.fix_hypercall       = emulator_fix_hypercall,
6216 	.intercept           = emulator_intercept,
6217 	.get_cpuid           = emulator_get_cpuid,
6218 	.set_nmi_mask        = emulator_set_nmi_mask,
6219 	.get_hflags          = emulator_get_hflags,
6220 	.set_hflags          = emulator_set_hflags,
6221 	.pre_leave_smm       = emulator_pre_leave_smm,
6222 	.post_leave_smm      = emulator_post_leave_smm,
6223 	.set_xcr             = emulator_set_xcr,
6224 };
6225 
6226 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6227 {
6228 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6229 	/*
6230 	 * an sti; sti; sequence only disable interrupts for the first
6231 	 * instruction. So, if the last instruction, be it emulated or
6232 	 * not, left the system with the INT_STI flag enabled, it
6233 	 * means that the last instruction is an sti. We should not
6234 	 * leave the flag on in this case. The same goes for mov ss
6235 	 */
6236 	if (int_shadow & mask)
6237 		mask = 0;
6238 	if (unlikely(int_shadow || mask)) {
6239 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6240 		if (!mask)
6241 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6242 	}
6243 }
6244 
6245 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6246 {
6247 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6248 	if (ctxt->exception.vector == PF_VECTOR)
6249 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6250 
6251 	if (ctxt->exception.error_code_valid)
6252 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6253 				      ctxt->exception.error_code);
6254 	else
6255 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6256 	return false;
6257 }
6258 
6259 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6260 {
6261 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6262 	int cs_db, cs_l;
6263 
6264 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6265 
6266 	ctxt->eflags = kvm_get_rflags(vcpu);
6267 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6268 
6269 	ctxt->eip = kvm_rip_read(vcpu);
6270 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6271 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6272 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6273 		     cs_db				? X86EMUL_MODE_PROT32 :
6274 							  X86EMUL_MODE_PROT16;
6275 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6276 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6277 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6278 
6279 	init_decode_cache(ctxt);
6280 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6281 }
6282 
6283 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6284 {
6285 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6286 	int ret;
6287 
6288 	init_emulate_ctxt(vcpu);
6289 
6290 	ctxt->op_bytes = 2;
6291 	ctxt->ad_bytes = 2;
6292 	ctxt->_eip = ctxt->eip + inc_eip;
6293 	ret = emulate_int_real(ctxt, irq);
6294 
6295 	if (ret != X86EMUL_CONTINUE) {
6296 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6297 	} else {
6298 		ctxt->eip = ctxt->_eip;
6299 		kvm_rip_write(vcpu, ctxt->eip);
6300 		kvm_set_rflags(vcpu, ctxt->eflags);
6301 	}
6302 }
6303 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6304 
6305 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6306 {
6307 	++vcpu->stat.insn_emulation_fail;
6308 	trace_kvm_emulate_insn_failed(vcpu);
6309 
6310 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6311 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6312 		return 1;
6313 	}
6314 
6315 	if (emulation_type & EMULTYPE_SKIP) {
6316 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6317 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6318 		vcpu->run->internal.ndata = 0;
6319 		return 0;
6320 	}
6321 
6322 	kvm_queue_exception(vcpu, UD_VECTOR);
6323 
6324 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6325 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6326 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6327 		vcpu->run->internal.ndata = 0;
6328 		return 0;
6329 	}
6330 
6331 	return 1;
6332 }
6333 
6334 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6335 				  bool write_fault_to_shadow_pgtable,
6336 				  int emulation_type)
6337 {
6338 	gpa_t gpa = cr2;
6339 	kvm_pfn_t pfn;
6340 
6341 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6342 		return false;
6343 
6344 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6345 		return false;
6346 
6347 	if (!vcpu->arch.mmu->direct_map) {
6348 		/*
6349 		 * Write permission should be allowed since only
6350 		 * write access need to be emulated.
6351 		 */
6352 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6353 
6354 		/*
6355 		 * If the mapping is invalid in guest, let cpu retry
6356 		 * it to generate fault.
6357 		 */
6358 		if (gpa == UNMAPPED_GVA)
6359 			return true;
6360 	}
6361 
6362 	/*
6363 	 * Do not retry the unhandleable instruction if it faults on the
6364 	 * readonly host memory, otherwise it will goto a infinite loop:
6365 	 * retry instruction -> write #PF -> emulation fail -> retry
6366 	 * instruction -> ...
6367 	 */
6368 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6369 
6370 	/*
6371 	 * If the instruction failed on the error pfn, it can not be fixed,
6372 	 * report the error to userspace.
6373 	 */
6374 	if (is_error_noslot_pfn(pfn))
6375 		return false;
6376 
6377 	kvm_release_pfn_clean(pfn);
6378 
6379 	/* The instructions are well-emulated on direct mmu. */
6380 	if (vcpu->arch.mmu->direct_map) {
6381 		unsigned int indirect_shadow_pages;
6382 
6383 		spin_lock(&vcpu->kvm->mmu_lock);
6384 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6385 		spin_unlock(&vcpu->kvm->mmu_lock);
6386 
6387 		if (indirect_shadow_pages)
6388 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6389 
6390 		return true;
6391 	}
6392 
6393 	/*
6394 	 * if emulation was due to access to shadowed page table
6395 	 * and it failed try to unshadow page and re-enter the
6396 	 * guest to let CPU execute the instruction.
6397 	 */
6398 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6399 
6400 	/*
6401 	 * If the access faults on its page table, it can not
6402 	 * be fixed by unprotecting shadow page and it should
6403 	 * be reported to userspace.
6404 	 */
6405 	return !write_fault_to_shadow_pgtable;
6406 }
6407 
6408 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6409 			      unsigned long cr2,  int emulation_type)
6410 {
6411 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6412 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6413 
6414 	last_retry_eip = vcpu->arch.last_retry_eip;
6415 	last_retry_addr = vcpu->arch.last_retry_addr;
6416 
6417 	/*
6418 	 * If the emulation is caused by #PF and it is non-page_table
6419 	 * writing instruction, it means the VM-EXIT is caused by shadow
6420 	 * page protected, we can zap the shadow page and retry this
6421 	 * instruction directly.
6422 	 *
6423 	 * Note: if the guest uses a non-page-table modifying instruction
6424 	 * on the PDE that points to the instruction, then we will unmap
6425 	 * the instruction and go to an infinite loop. So, we cache the
6426 	 * last retried eip and the last fault address, if we meet the eip
6427 	 * and the address again, we can break out of the potential infinite
6428 	 * loop.
6429 	 */
6430 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6431 
6432 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6433 		return false;
6434 
6435 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6436 		return false;
6437 
6438 	if (x86_page_table_writing_insn(ctxt))
6439 		return false;
6440 
6441 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6442 		return false;
6443 
6444 	vcpu->arch.last_retry_eip = ctxt->eip;
6445 	vcpu->arch.last_retry_addr = cr2;
6446 
6447 	if (!vcpu->arch.mmu->direct_map)
6448 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6449 
6450 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6451 
6452 	return true;
6453 }
6454 
6455 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6456 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6457 
6458 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6459 {
6460 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6461 		/* This is a good place to trace that we are exiting SMM.  */
6462 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6463 
6464 		/* Process a latched INIT or SMI, if any.  */
6465 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6466 	}
6467 
6468 	kvm_mmu_reset_context(vcpu);
6469 }
6470 
6471 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6472 				unsigned long *db)
6473 {
6474 	u32 dr6 = 0;
6475 	int i;
6476 	u32 enable, rwlen;
6477 
6478 	enable = dr7;
6479 	rwlen = dr7 >> 16;
6480 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6481 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6482 			dr6 |= (1 << i);
6483 	return dr6;
6484 }
6485 
6486 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6487 {
6488 	struct kvm_run *kvm_run = vcpu->run;
6489 
6490 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6491 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6492 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6493 		kvm_run->debug.arch.exception = DB_VECTOR;
6494 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6495 		return 0;
6496 	}
6497 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6498 	return 1;
6499 }
6500 
6501 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6502 {
6503 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6504 	int r;
6505 
6506 	r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6507 	if (unlikely(!r))
6508 		return 0;
6509 
6510 	/*
6511 	 * rflags is the old, "raw" value of the flags.  The new value has
6512 	 * not been saved yet.
6513 	 *
6514 	 * This is correct even for TF set by the guest, because "the
6515 	 * processor will not generate this exception after the instruction
6516 	 * that sets the TF flag".
6517 	 */
6518 	if (unlikely(rflags & X86_EFLAGS_TF))
6519 		r = kvm_vcpu_do_singlestep(vcpu);
6520 	return r;
6521 }
6522 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6523 
6524 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6525 {
6526 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6527 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6528 		struct kvm_run *kvm_run = vcpu->run;
6529 		unsigned long eip = kvm_get_linear_rip(vcpu);
6530 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6531 					   vcpu->arch.guest_debug_dr7,
6532 					   vcpu->arch.eff_db);
6533 
6534 		if (dr6 != 0) {
6535 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6536 			kvm_run->debug.arch.pc = eip;
6537 			kvm_run->debug.arch.exception = DB_VECTOR;
6538 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6539 			*r = 0;
6540 			return true;
6541 		}
6542 	}
6543 
6544 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6545 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6546 		unsigned long eip = kvm_get_linear_rip(vcpu);
6547 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6548 					   vcpu->arch.dr7,
6549 					   vcpu->arch.db);
6550 
6551 		if (dr6 != 0) {
6552 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6553 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6554 			kvm_queue_exception(vcpu, DB_VECTOR);
6555 			*r = 1;
6556 			return true;
6557 		}
6558 	}
6559 
6560 	return false;
6561 }
6562 
6563 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6564 {
6565 	switch (ctxt->opcode_len) {
6566 	case 1:
6567 		switch (ctxt->b) {
6568 		case 0xe4:	/* IN */
6569 		case 0xe5:
6570 		case 0xec:
6571 		case 0xed:
6572 		case 0xe6:	/* OUT */
6573 		case 0xe7:
6574 		case 0xee:
6575 		case 0xef:
6576 		case 0x6c:	/* INS */
6577 		case 0x6d:
6578 		case 0x6e:	/* OUTS */
6579 		case 0x6f:
6580 			return true;
6581 		}
6582 		break;
6583 	case 2:
6584 		switch (ctxt->b) {
6585 		case 0x33:	/* RDPMC */
6586 			return true;
6587 		}
6588 		break;
6589 	}
6590 
6591 	return false;
6592 }
6593 
6594 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6595 			    unsigned long cr2,
6596 			    int emulation_type,
6597 			    void *insn,
6598 			    int insn_len)
6599 {
6600 	int r;
6601 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6602 	bool writeback = true;
6603 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6604 
6605 	vcpu->arch.l1tf_flush_l1d = true;
6606 
6607 	/*
6608 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6609 	 * never reused.
6610 	 */
6611 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6612 	kvm_clear_exception_queue(vcpu);
6613 
6614 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6615 		init_emulate_ctxt(vcpu);
6616 
6617 		/*
6618 		 * We will reenter on the same instruction since
6619 		 * we do not set complete_userspace_io.  This does not
6620 		 * handle watchpoints yet, those would be handled in
6621 		 * the emulate_ops.
6622 		 */
6623 		if (!(emulation_type & EMULTYPE_SKIP) &&
6624 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6625 			return r;
6626 
6627 		ctxt->interruptibility = 0;
6628 		ctxt->have_exception = false;
6629 		ctxt->exception.vector = -1;
6630 		ctxt->perm_ok = false;
6631 
6632 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6633 
6634 		r = x86_decode_insn(ctxt, insn, insn_len);
6635 
6636 		trace_kvm_emulate_insn_start(vcpu);
6637 		++vcpu->stat.insn_emulation;
6638 		if (r != EMULATION_OK)  {
6639 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6640 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6641 				kvm_queue_exception(vcpu, UD_VECTOR);
6642 				return 1;
6643 			}
6644 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6645 						emulation_type))
6646 				return 1;
6647 			if (ctxt->have_exception) {
6648 				/*
6649 				 * #UD should result in just EMULATION_FAILED, and trap-like
6650 				 * exception should not be encountered during decode.
6651 				 */
6652 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6653 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6654 				inject_emulated_exception(vcpu);
6655 				return 1;
6656 			}
6657 			return handle_emulation_failure(vcpu, emulation_type);
6658 		}
6659 	}
6660 
6661 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6662 	    !is_vmware_backdoor_opcode(ctxt)) {
6663 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6664 		return 1;
6665 	}
6666 
6667 	/*
6668 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6669 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6670 	 * updating interruptibility state and injecting single-step #DBs.
6671 	 */
6672 	if (emulation_type & EMULTYPE_SKIP) {
6673 		kvm_rip_write(vcpu, ctxt->_eip);
6674 		if (ctxt->eflags & X86_EFLAGS_RF)
6675 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6676 		return 1;
6677 	}
6678 
6679 	if (retry_instruction(ctxt, cr2, emulation_type))
6680 		return 1;
6681 
6682 	/* this is needed for vmware backdoor interface to work since it
6683 	   changes registers values  during IO operation */
6684 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6685 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6686 		emulator_invalidate_register_cache(ctxt);
6687 	}
6688 
6689 restart:
6690 	/* Save the faulting GPA (cr2) in the address field */
6691 	ctxt->exception.address = cr2;
6692 
6693 	r = x86_emulate_insn(ctxt);
6694 
6695 	if (r == EMULATION_INTERCEPTED)
6696 		return 1;
6697 
6698 	if (r == EMULATION_FAILED) {
6699 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6700 					emulation_type))
6701 			return 1;
6702 
6703 		return handle_emulation_failure(vcpu, emulation_type);
6704 	}
6705 
6706 	if (ctxt->have_exception) {
6707 		r = 1;
6708 		if (inject_emulated_exception(vcpu))
6709 			return r;
6710 	} else if (vcpu->arch.pio.count) {
6711 		if (!vcpu->arch.pio.in) {
6712 			/* FIXME: return into emulator if single-stepping.  */
6713 			vcpu->arch.pio.count = 0;
6714 		} else {
6715 			writeback = false;
6716 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6717 		}
6718 		r = 0;
6719 	} else if (vcpu->mmio_needed) {
6720 		++vcpu->stat.mmio_exits;
6721 
6722 		if (!vcpu->mmio_is_write)
6723 			writeback = false;
6724 		r = 0;
6725 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6726 	} else if (r == EMULATION_RESTART)
6727 		goto restart;
6728 	else
6729 		r = 1;
6730 
6731 	if (writeback) {
6732 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6733 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6734 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6735 		if (!ctxt->have_exception ||
6736 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6737 			kvm_rip_write(vcpu, ctxt->eip);
6738 			if (r && ctxt->tf)
6739 				r = kvm_vcpu_do_singlestep(vcpu);
6740 			__kvm_set_rflags(vcpu, ctxt->eflags);
6741 		}
6742 
6743 		/*
6744 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6745 		 * do nothing, and it will be requested again as soon as
6746 		 * the shadow expires.  But we still need to check here,
6747 		 * because POPF has no interrupt shadow.
6748 		 */
6749 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6750 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6751 	} else
6752 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6753 
6754 	return r;
6755 }
6756 
6757 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6758 {
6759 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6760 }
6761 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6762 
6763 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6764 					void *insn, int insn_len)
6765 {
6766 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6767 }
6768 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6769 
6770 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6771 {
6772 	vcpu->arch.pio.count = 0;
6773 	return 1;
6774 }
6775 
6776 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6777 {
6778 	vcpu->arch.pio.count = 0;
6779 
6780 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6781 		return 1;
6782 
6783 	return kvm_skip_emulated_instruction(vcpu);
6784 }
6785 
6786 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6787 			    unsigned short port)
6788 {
6789 	unsigned long val = kvm_rax_read(vcpu);
6790 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6791 					    size, port, &val, 1);
6792 	if (ret)
6793 		return ret;
6794 
6795 	/*
6796 	 * Workaround userspace that relies on old KVM behavior of %rip being
6797 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6798 	 */
6799 	if (port == 0x7e &&
6800 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6801 		vcpu->arch.complete_userspace_io =
6802 			complete_fast_pio_out_port_0x7e;
6803 		kvm_skip_emulated_instruction(vcpu);
6804 	} else {
6805 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6806 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6807 	}
6808 	return 0;
6809 }
6810 
6811 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6812 {
6813 	unsigned long val;
6814 
6815 	/* We should only ever be called with arch.pio.count equal to 1 */
6816 	BUG_ON(vcpu->arch.pio.count != 1);
6817 
6818 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6819 		vcpu->arch.pio.count = 0;
6820 		return 1;
6821 	}
6822 
6823 	/* For size less than 4 we merge, else we zero extend */
6824 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6825 
6826 	/*
6827 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6828 	 * the copy and tracing
6829 	 */
6830 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6831 				 vcpu->arch.pio.port, &val, 1);
6832 	kvm_rax_write(vcpu, val);
6833 
6834 	return kvm_skip_emulated_instruction(vcpu);
6835 }
6836 
6837 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6838 			   unsigned short port)
6839 {
6840 	unsigned long val;
6841 	int ret;
6842 
6843 	/* For size less than 4 we merge, else we zero extend */
6844 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6845 
6846 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6847 				       &val, 1);
6848 	if (ret) {
6849 		kvm_rax_write(vcpu, val);
6850 		return ret;
6851 	}
6852 
6853 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6854 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6855 
6856 	return 0;
6857 }
6858 
6859 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6860 {
6861 	int ret;
6862 
6863 	if (in)
6864 		ret = kvm_fast_pio_in(vcpu, size, port);
6865 	else
6866 		ret = kvm_fast_pio_out(vcpu, size, port);
6867 	return ret && kvm_skip_emulated_instruction(vcpu);
6868 }
6869 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6870 
6871 static int kvmclock_cpu_down_prep(unsigned int cpu)
6872 {
6873 	__this_cpu_write(cpu_tsc_khz, 0);
6874 	return 0;
6875 }
6876 
6877 static void tsc_khz_changed(void *data)
6878 {
6879 	struct cpufreq_freqs *freq = data;
6880 	unsigned long khz = 0;
6881 
6882 	if (data)
6883 		khz = freq->new;
6884 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6885 		khz = cpufreq_quick_get(raw_smp_processor_id());
6886 	if (!khz)
6887 		khz = tsc_khz;
6888 	__this_cpu_write(cpu_tsc_khz, khz);
6889 }
6890 
6891 #ifdef CONFIG_X86_64
6892 static void kvm_hyperv_tsc_notifier(void)
6893 {
6894 	struct kvm *kvm;
6895 	struct kvm_vcpu *vcpu;
6896 	int cpu;
6897 
6898 	mutex_lock(&kvm_lock);
6899 	list_for_each_entry(kvm, &vm_list, vm_list)
6900 		kvm_make_mclock_inprogress_request(kvm);
6901 
6902 	hyperv_stop_tsc_emulation();
6903 
6904 	/* TSC frequency always matches when on Hyper-V */
6905 	for_each_present_cpu(cpu)
6906 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6907 	kvm_max_guest_tsc_khz = tsc_khz;
6908 
6909 	list_for_each_entry(kvm, &vm_list, vm_list) {
6910 		struct kvm_arch *ka = &kvm->arch;
6911 
6912 		spin_lock(&ka->pvclock_gtod_sync_lock);
6913 
6914 		pvclock_update_vm_gtod_copy(kvm);
6915 
6916 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6917 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6918 
6919 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6920 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6921 
6922 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6923 	}
6924 	mutex_unlock(&kvm_lock);
6925 }
6926 #endif
6927 
6928 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6929 {
6930 	struct kvm *kvm;
6931 	struct kvm_vcpu *vcpu;
6932 	int i, send_ipi = 0;
6933 
6934 	/*
6935 	 * We allow guests to temporarily run on slowing clocks,
6936 	 * provided we notify them after, or to run on accelerating
6937 	 * clocks, provided we notify them before.  Thus time never
6938 	 * goes backwards.
6939 	 *
6940 	 * However, we have a problem.  We can't atomically update
6941 	 * the frequency of a given CPU from this function; it is
6942 	 * merely a notifier, which can be called from any CPU.
6943 	 * Changing the TSC frequency at arbitrary points in time
6944 	 * requires a recomputation of local variables related to
6945 	 * the TSC for each VCPU.  We must flag these local variables
6946 	 * to be updated and be sure the update takes place with the
6947 	 * new frequency before any guests proceed.
6948 	 *
6949 	 * Unfortunately, the combination of hotplug CPU and frequency
6950 	 * change creates an intractable locking scenario; the order
6951 	 * of when these callouts happen is undefined with respect to
6952 	 * CPU hotplug, and they can race with each other.  As such,
6953 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6954 	 * undefined; you can actually have a CPU frequency change take
6955 	 * place in between the computation of X and the setting of the
6956 	 * variable.  To protect against this problem, all updates of
6957 	 * the per_cpu tsc_khz variable are done in an interrupt
6958 	 * protected IPI, and all callers wishing to update the value
6959 	 * must wait for a synchronous IPI to complete (which is trivial
6960 	 * if the caller is on the CPU already).  This establishes the
6961 	 * necessary total order on variable updates.
6962 	 *
6963 	 * Note that because a guest time update may take place
6964 	 * anytime after the setting of the VCPU's request bit, the
6965 	 * correct TSC value must be set before the request.  However,
6966 	 * to ensure the update actually makes it to any guest which
6967 	 * starts running in hardware virtualization between the set
6968 	 * and the acquisition of the spinlock, we must also ping the
6969 	 * CPU after setting the request bit.
6970 	 *
6971 	 */
6972 
6973 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6974 
6975 	mutex_lock(&kvm_lock);
6976 	list_for_each_entry(kvm, &vm_list, vm_list) {
6977 		kvm_for_each_vcpu(i, vcpu, kvm) {
6978 			if (vcpu->cpu != cpu)
6979 				continue;
6980 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6981 			if (vcpu->cpu != raw_smp_processor_id())
6982 				send_ipi = 1;
6983 		}
6984 	}
6985 	mutex_unlock(&kvm_lock);
6986 
6987 	if (freq->old < freq->new && send_ipi) {
6988 		/*
6989 		 * We upscale the frequency.  Must make the guest
6990 		 * doesn't see old kvmclock values while running with
6991 		 * the new frequency, otherwise we risk the guest sees
6992 		 * time go backwards.
6993 		 *
6994 		 * In case we update the frequency for another cpu
6995 		 * (which might be in guest context) send an interrupt
6996 		 * to kick the cpu out of guest context.  Next time
6997 		 * guest context is entered kvmclock will be updated,
6998 		 * so the guest will not see stale values.
6999 		 */
7000 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7001 	}
7002 }
7003 
7004 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7005 				     void *data)
7006 {
7007 	struct cpufreq_freqs *freq = data;
7008 	int cpu;
7009 
7010 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7011 		return 0;
7012 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7013 		return 0;
7014 
7015 	for_each_cpu(cpu, freq->policy->cpus)
7016 		__kvmclock_cpufreq_notifier(freq, cpu);
7017 
7018 	return 0;
7019 }
7020 
7021 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7022 	.notifier_call  = kvmclock_cpufreq_notifier
7023 };
7024 
7025 static int kvmclock_cpu_online(unsigned int cpu)
7026 {
7027 	tsc_khz_changed(NULL);
7028 	return 0;
7029 }
7030 
7031 static void kvm_timer_init(void)
7032 {
7033 	max_tsc_khz = tsc_khz;
7034 
7035 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7036 #ifdef CONFIG_CPU_FREQ
7037 		struct cpufreq_policy policy;
7038 		int cpu;
7039 
7040 		memset(&policy, 0, sizeof(policy));
7041 		cpu = get_cpu();
7042 		cpufreq_get_policy(&policy, cpu);
7043 		if (policy.cpuinfo.max_freq)
7044 			max_tsc_khz = policy.cpuinfo.max_freq;
7045 		put_cpu();
7046 #endif
7047 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7048 					  CPUFREQ_TRANSITION_NOTIFIER);
7049 	}
7050 
7051 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7052 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7053 }
7054 
7055 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7056 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7057 
7058 int kvm_is_in_guest(void)
7059 {
7060 	return __this_cpu_read(current_vcpu) != NULL;
7061 }
7062 
7063 static int kvm_is_user_mode(void)
7064 {
7065 	int user_mode = 3;
7066 
7067 	if (__this_cpu_read(current_vcpu))
7068 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7069 
7070 	return user_mode != 0;
7071 }
7072 
7073 static unsigned long kvm_get_guest_ip(void)
7074 {
7075 	unsigned long ip = 0;
7076 
7077 	if (__this_cpu_read(current_vcpu))
7078 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7079 
7080 	return ip;
7081 }
7082 
7083 static void kvm_handle_intel_pt_intr(void)
7084 {
7085 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7086 
7087 	kvm_make_request(KVM_REQ_PMI, vcpu);
7088 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7089 			(unsigned long *)&vcpu->arch.pmu.global_status);
7090 }
7091 
7092 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7093 	.is_in_guest		= kvm_is_in_guest,
7094 	.is_user_mode		= kvm_is_user_mode,
7095 	.get_guest_ip		= kvm_get_guest_ip,
7096 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7097 };
7098 
7099 #ifdef CONFIG_X86_64
7100 static void pvclock_gtod_update_fn(struct work_struct *work)
7101 {
7102 	struct kvm *kvm;
7103 
7104 	struct kvm_vcpu *vcpu;
7105 	int i;
7106 
7107 	mutex_lock(&kvm_lock);
7108 	list_for_each_entry(kvm, &vm_list, vm_list)
7109 		kvm_for_each_vcpu(i, vcpu, kvm)
7110 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7111 	atomic_set(&kvm_guest_has_master_clock, 0);
7112 	mutex_unlock(&kvm_lock);
7113 }
7114 
7115 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7116 
7117 /*
7118  * Notification about pvclock gtod data update.
7119  */
7120 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7121 			       void *priv)
7122 {
7123 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7124 	struct timekeeper *tk = priv;
7125 
7126 	update_pvclock_gtod(tk);
7127 
7128 	/* disable master clock if host does not trust, or does not
7129 	 * use, TSC based clocksource.
7130 	 */
7131 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7132 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7133 		queue_work(system_long_wq, &pvclock_gtod_work);
7134 
7135 	return 0;
7136 }
7137 
7138 static struct notifier_block pvclock_gtod_notifier = {
7139 	.notifier_call = pvclock_gtod_notify,
7140 };
7141 #endif
7142 
7143 int kvm_arch_init(void *opaque)
7144 {
7145 	int r;
7146 	struct kvm_x86_ops *ops = opaque;
7147 
7148 	if (kvm_x86_ops) {
7149 		printk(KERN_ERR "kvm: already loaded the other module\n");
7150 		r = -EEXIST;
7151 		goto out;
7152 	}
7153 
7154 	if (!ops->cpu_has_kvm_support()) {
7155 		printk(KERN_ERR "kvm: no hardware support\n");
7156 		r = -EOPNOTSUPP;
7157 		goto out;
7158 	}
7159 	if (ops->disabled_by_bios()) {
7160 		printk(KERN_ERR "kvm: disabled by bios\n");
7161 		r = -EOPNOTSUPP;
7162 		goto out;
7163 	}
7164 
7165 	/*
7166 	 * KVM explicitly assumes that the guest has an FPU and
7167 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7168 	 * vCPU's FPU state as a fxregs_state struct.
7169 	 */
7170 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7171 		printk(KERN_ERR "kvm: inadequate fpu\n");
7172 		r = -EOPNOTSUPP;
7173 		goto out;
7174 	}
7175 
7176 	r = -ENOMEM;
7177 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7178 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7179 					  NULL);
7180 	if (!x86_fpu_cache) {
7181 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7182 		goto out;
7183 	}
7184 
7185 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7186 	if (!shared_msrs) {
7187 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7188 		goto out_free_x86_fpu_cache;
7189 	}
7190 
7191 	r = kvm_mmu_module_init();
7192 	if (r)
7193 		goto out_free_percpu;
7194 
7195 	kvm_x86_ops = ops;
7196 
7197 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7198 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7199 			PT_PRESENT_MASK, 0, sme_me_mask);
7200 	kvm_timer_init();
7201 
7202 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7203 
7204 	if (boot_cpu_has(X86_FEATURE_XSAVE))
7205 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7206 
7207 	kvm_lapic_init();
7208 	if (pi_inject_timer == -1)
7209 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7210 #ifdef CONFIG_X86_64
7211 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7212 
7213 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7214 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7215 #endif
7216 
7217 	return 0;
7218 
7219 out_free_percpu:
7220 	free_percpu(shared_msrs);
7221 out_free_x86_fpu_cache:
7222 	kmem_cache_destroy(x86_fpu_cache);
7223 out:
7224 	return r;
7225 }
7226 
7227 void kvm_arch_exit(void)
7228 {
7229 #ifdef CONFIG_X86_64
7230 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7231 		clear_hv_tscchange_cb();
7232 #endif
7233 	kvm_lapic_exit();
7234 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7235 
7236 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7237 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7238 					    CPUFREQ_TRANSITION_NOTIFIER);
7239 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7240 #ifdef CONFIG_X86_64
7241 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7242 #endif
7243 	kvm_x86_ops = NULL;
7244 	kvm_mmu_module_exit();
7245 	free_percpu(shared_msrs);
7246 	kmem_cache_destroy(x86_fpu_cache);
7247 }
7248 
7249 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7250 {
7251 	++vcpu->stat.halt_exits;
7252 	if (lapic_in_kernel(vcpu)) {
7253 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7254 		return 1;
7255 	} else {
7256 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7257 		return 0;
7258 	}
7259 }
7260 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7261 
7262 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7263 {
7264 	int ret = kvm_skip_emulated_instruction(vcpu);
7265 	/*
7266 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7267 	 * KVM_EXIT_DEBUG here.
7268 	 */
7269 	return kvm_vcpu_halt(vcpu) && ret;
7270 }
7271 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7272 
7273 #ifdef CONFIG_X86_64
7274 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7275 			        unsigned long clock_type)
7276 {
7277 	struct kvm_clock_pairing clock_pairing;
7278 	struct timespec64 ts;
7279 	u64 cycle;
7280 	int ret;
7281 
7282 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7283 		return -KVM_EOPNOTSUPP;
7284 
7285 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7286 		return -KVM_EOPNOTSUPP;
7287 
7288 	clock_pairing.sec = ts.tv_sec;
7289 	clock_pairing.nsec = ts.tv_nsec;
7290 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7291 	clock_pairing.flags = 0;
7292 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7293 
7294 	ret = 0;
7295 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7296 			    sizeof(struct kvm_clock_pairing)))
7297 		ret = -KVM_EFAULT;
7298 
7299 	return ret;
7300 }
7301 #endif
7302 
7303 /*
7304  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7305  *
7306  * @apicid - apicid of vcpu to be kicked.
7307  */
7308 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7309 {
7310 	struct kvm_lapic_irq lapic_irq;
7311 
7312 	lapic_irq.shorthand = 0;
7313 	lapic_irq.dest_mode = 0;
7314 	lapic_irq.level = 0;
7315 	lapic_irq.dest_id = apicid;
7316 	lapic_irq.msi_redir_hint = false;
7317 
7318 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7319 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7320 }
7321 
7322 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7323 {
7324 	if (!lapic_in_kernel(vcpu)) {
7325 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7326 		return;
7327 	}
7328 	if (!vcpu->arch.apicv_active)
7329 		return;
7330 
7331 	vcpu->arch.apicv_active = false;
7332 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7333 }
7334 
7335 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7336 {
7337 	struct kvm_vcpu *target = NULL;
7338 	struct kvm_apic_map *map;
7339 
7340 	rcu_read_lock();
7341 	map = rcu_dereference(kvm->arch.apic_map);
7342 
7343 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7344 		target = map->phys_map[dest_id]->vcpu;
7345 
7346 	rcu_read_unlock();
7347 
7348 	if (target && READ_ONCE(target->ready))
7349 		kvm_vcpu_yield_to(target);
7350 }
7351 
7352 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7353 {
7354 	unsigned long nr, a0, a1, a2, a3, ret;
7355 	int op_64_bit;
7356 
7357 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7358 		return kvm_hv_hypercall(vcpu);
7359 
7360 	nr = kvm_rax_read(vcpu);
7361 	a0 = kvm_rbx_read(vcpu);
7362 	a1 = kvm_rcx_read(vcpu);
7363 	a2 = kvm_rdx_read(vcpu);
7364 	a3 = kvm_rsi_read(vcpu);
7365 
7366 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7367 
7368 	op_64_bit = is_64_bit_mode(vcpu);
7369 	if (!op_64_bit) {
7370 		nr &= 0xFFFFFFFF;
7371 		a0 &= 0xFFFFFFFF;
7372 		a1 &= 0xFFFFFFFF;
7373 		a2 &= 0xFFFFFFFF;
7374 		a3 &= 0xFFFFFFFF;
7375 	}
7376 
7377 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7378 		ret = -KVM_EPERM;
7379 		goto out;
7380 	}
7381 
7382 	switch (nr) {
7383 	case KVM_HC_VAPIC_POLL_IRQ:
7384 		ret = 0;
7385 		break;
7386 	case KVM_HC_KICK_CPU:
7387 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7388 		kvm_sched_yield(vcpu->kvm, a1);
7389 		ret = 0;
7390 		break;
7391 #ifdef CONFIG_X86_64
7392 	case KVM_HC_CLOCK_PAIRING:
7393 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7394 		break;
7395 #endif
7396 	case KVM_HC_SEND_IPI:
7397 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7398 		break;
7399 	case KVM_HC_SCHED_YIELD:
7400 		kvm_sched_yield(vcpu->kvm, a0);
7401 		ret = 0;
7402 		break;
7403 	default:
7404 		ret = -KVM_ENOSYS;
7405 		break;
7406 	}
7407 out:
7408 	if (!op_64_bit)
7409 		ret = (u32)ret;
7410 	kvm_rax_write(vcpu, ret);
7411 
7412 	++vcpu->stat.hypercalls;
7413 	return kvm_skip_emulated_instruction(vcpu);
7414 }
7415 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7416 
7417 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7418 {
7419 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7420 	char instruction[3];
7421 	unsigned long rip = kvm_rip_read(vcpu);
7422 
7423 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7424 
7425 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7426 		&ctxt->exception);
7427 }
7428 
7429 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7430 {
7431 	return vcpu->run->request_interrupt_window &&
7432 		likely(!pic_in_kernel(vcpu->kvm));
7433 }
7434 
7435 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7436 {
7437 	struct kvm_run *kvm_run = vcpu->run;
7438 
7439 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7440 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7441 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7442 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7443 	kvm_run->ready_for_interrupt_injection =
7444 		pic_in_kernel(vcpu->kvm) ||
7445 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7446 }
7447 
7448 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7449 {
7450 	int max_irr, tpr;
7451 
7452 	if (!kvm_x86_ops->update_cr8_intercept)
7453 		return;
7454 
7455 	if (!lapic_in_kernel(vcpu))
7456 		return;
7457 
7458 	if (vcpu->arch.apicv_active)
7459 		return;
7460 
7461 	if (!vcpu->arch.apic->vapic_addr)
7462 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7463 	else
7464 		max_irr = -1;
7465 
7466 	if (max_irr != -1)
7467 		max_irr >>= 4;
7468 
7469 	tpr = kvm_lapic_get_cr8(vcpu);
7470 
7471 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7472 }
7473 
7474 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7475 {
7476 	int r;
7477 
7478 	/* try to reinject previous events if any */
7479 
7480 	if (vcpu->arch.exception.injected)
7481 		kvm_x86_ops->queue_exception(vcpu);
7482 	/*
7483 	 * Do not inject an NMI or interrupt if there is a pending
7484 	 * exception.  Exceptions and interrupts are recognized at
7485 	 * instruction boundaries, i.e. the start of an instruction.
7486 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7487 	 * NMIs and interrupts, i.e. traps are recognized before an
7488 	 * NMI/interrupt that's pending on the same instruction.
7489 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7490 	 * priority, but are only generated (pended) during instruction
7491 	 * execution, i.e. a pending fault-like exception means the
7492 	 * fault occurred on the *previous* instruction and must be
7493 	 * serviced prior to recognizing any new events in order to
7494 	 * fully complete the previous instruction.
7495 	 */
7496 	else if (!vcpu->arch.exception.pending) {
7497 		if (vcpu->arch.nmi_injected)
7498 			kvm_x86_ops->set_nmi(vcpu);
7499 		else if (vcpu->arch.interrupt.injected)
7500 			kvm_x86_ops->set_irq(vcpu);
7501 	}
7502 
7503 	/*
7504 	 * Call check_nested_events() even if we reinjected a previous event
7505 	 * in order for caller to determine if it should require immediate-exit
7506 	 * from L2 to L1 due to pending L1 events which require exit
7507 	 * from L2 to L1.
7508 	 */
7509 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7510 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7511 		if (r != 0)
7512 			return r;
7513 	}
7514 
7515 	/* try to inject new event if pending */
7516 	if (vcpu->arch.exception.pending) {
7517 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7518 					vcpu->arch.exception.has_error_code,
7519 					vcpu->arch.exception.error_code);
7520 
7521 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7522 		vcpu->arch.exception.pending = false;
7523 		vcpu->arch.exception.injected = true;
7524 
7525 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7526 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7527 					     X86_EFLAGS_RF);
7528 
7529 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7530 			/*
7531 			 * This code assumes that nSVM doesn't use
7532 			 * check_nested_events(). If it does, the
7533 			 * DR6/DR7 changes should happen before L1
7534 			 * gets a #VMEXIT for an intercepted #DB in
7535 			 * L2.  (Under VMX, on the other hand, the
7536 			 * DR6/DR7 changes should not happen in the
7537 			 * event of a VM-exit to L1 for an intercepted
7538 			 * #DB in L2.)
7539 			 */
7540 			kvm_deliver_exception_payload(vcpu);
7541 			if (vcpu->arch.dr7 & DR7_GD) {
7542 				vcpu->arch.dr7 &= ~DR7_GD;
7543 				kvm_update_dr7(vcpu);
7544 			}
7545 		}
7546 
7547 		kvm_x86_ops->queue_exception(vcpu);
7548 	}
7549 
7550 	/* Don't consider new event if we re-injected an event */
7551 	if (kvm_event_needs_reinjection(vcpu))
7552 		return 0;
7553 
7554 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7555 	    kvm_x86_ops->smi_allowed(vcpu)) {
7556 		vcpu->arch.smi_pending = false;
7557 		++vcpu->arch.smi_count;
7558 		enter_smm(vcpu);
7559 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7560 		--vcpu->arch.nmi_pending;
7561 		vcpu->arch.nmi_injected = true;
7562 		kvm_x86_ops->set_nmi(vcpu);
7563 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7564 		/*
7565 		 * Because interrupts can be injected asynchronously, we are
7566 		 * calling check_nested_events again here to avoid a race condition.
7567 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7568 		 * proposal and current concerns.  Perhaps we should be setting
7569 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7570 		 */
7571 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7572 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7573 			if (r != 0)
7574 				return r;
7575 		}
7576 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7577 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7578 					    false);
7579 			kvm_x86_ops->set_irq(vcpu);
7580 		}
7581 	}
7582 
7583 	return 0;
7584 }
7585 
7586 static void process_nmi(struct kvm_vcpu *vcpu)
7587 {
7588 	unsigned limit = 2;
7589 
7590 	/*
7591 	 * x86 is limited to one NMI running, and one NMI pending after it.
7592 	 * If an NMI is already in progress, limit further NMIs to just one.
7593 	 * Otherwise, allow two (and we'll inject the first one immediately).
7594 	 */
7595 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7596 		limit = 1;
7597 
7598 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7599 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7600 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7601 }
7602 
7603 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7604 {
7605 	u32 flags = 0;
7606 	flags |= seg->g       << 23;
7607 	flags |= seg->db      << 22;
7608 	flags |= seg->l       << 21;
7609 	flags |= seg->avl     << 20;
7610 	flags |= seg->present << 15;
7611 	flags |= seg->dpl     << 13;
7612 	flags |= seg->s       << 12;
7613 	flags |= seg->type    << 8;
7614 	return flags;
7615 }
7616 
7617 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7618 {
7619 	struct kvm_segment seg;
7620 	int offset;
7621 
7622 	kvm_get_segment(vcpu, &seg, n);
7623 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7624 
7625 	if (n < 3)
7626 		offset = 0x7f84 + n * 12;
7627 	else
7628 		offset = 0x7f2c + (n - 3) * 12;
7629 
7630 	put_smstate(u32, buf, offset + 8, seg.base);
7631 	put_smstate(u32, buf, offset + 4, seg.limit);
7632 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7633 }
7634 
7635 #ifdef CONFIG_X86_64
7636 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7637 {
7638 	struct kvm_segment seg;
7639 	int offset;
7640 	u16 flags;
7641 
7642 	kvm_get_segment(vcpu, &seg, n);
7643 	offset = 0x7e00 + n * 16;
7644 
7645 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7646 	put_smstate(u16, buf, offset, seg.selector);
7647 	put_smstate(u16, buf, offset + 2, flags);
7648 	put_smstate(u32, buf, offset + 4, seg.limit);
7649 	put_smstate(u64, buf, offset + 8, seg.base);
7650 }
7651 #endif
7652 
7653 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7654 {
7655 	struct desc_ptr dt;
7656 	struct kvm_segment seg;
7657 	unsigned long val;
7658 	int i;
7659 
7660 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7661 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7662 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7663 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7664 
7665 	for (i = 0; i < 8; i++)
7666 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7667 
7668 	kvm_get_dr(vcpu, 6, &val);
7669 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7670 	kvm_get_dr(vcpu, 7, &val);
7671 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7672 
7673 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7674 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7675 	put_smstate(u32, buf, 0x7f64, seg.base);
7676 	put_smstate(u32, buf, 0x7f60, seg.limit);
7677 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7678 
7679 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7680 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7681 	put_smstate(u32, buf, 0x7f80, seg.base);
7682 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7683 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7684 
7685 	kvm_x86_ops->get_gdt(vcpu, &dt);
7686 	put_smstate(u32, buf, 0x7f74, dt.address);
7687 	put_smstate(u32, buf, 0x7f70, dt.size);
7688 
7689 	kvm_x86_ops->get_idt(vcpu, &dt);
7690 	put_smstate(u32, buf, 0x7f58, dt.address);
7691 	put_smstate(u32, buf, 0x7f54, dt.size);
7692 
7693 	for (i = 0; i < 6; i++)
7694 		enter_smm_save_seg_32(vcpu, buf, i);
7695 
7696 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7697 
7698 	/* revision id */
7699 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7700 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7701 }
7702 
7703 #ifdef CONFIG_X86_64
7704 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7705 {
7706 	struct desc_ptr dt;
7707 	struct kvm_segment seg;
7708 	unsigned long val;
7709 	int i;
7710 
7711 	for (i = 0; i < 16; i++)
7712 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7713 
7714 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7715 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7716 
7717 	kvm_get_dr(vcpu, 6, &val);
7718 	put_smstate(u64, buf, 0x7f68, val);
7719 	kvm_get_dr(vcpu, 7, &val);
7720 	put_smstate(u64, buf, 0x7f60, val);
7721 
7722 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7723 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7724 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7725 
7726 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7727 
7728 	/* revision id */
7729 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7730 
7731 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7732 
7733 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7734 	put_smstate(u16, buf, 0x7e90, seg.selector);
7735 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7736 	put_smstate(u32, buf, 0x7e94, seg.limit);
7737 	put_smstate(u64, buf, 0x7e98, seg.base);
7738 
7739 	kvm_x86_ops->get_idt(vcpu, &dt);
7740 	put_smstate(u32, buf, 0x7e84, dt.size);
7741 	put_smstate(u64, buf, 0x7e88, dt.address);
7742 
7743 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7744 	put_smstate(u16, buf, 0x7e70, seg.selector);
7745 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7746 	put_smstate(u32, buf, 0x7e74, seg.limit);
7747 	put_smstate(u64, buf, 0x7e78, seg.base);
7748 
7749 	kvm_x86_ops->get_gdt(vcpu, &dt);
7750 	put_smstate(u32, buf, 0x7e64, dt.size);
7751 	put_smstate(u64, buf, 0x7e68, dt.address);
7752 
7753 	for (i = 0; i < 6; i++)
7754 		enter_smm_save_seg_64(vcpu, buf, i);
7755 }
7756 #endif
7757 
7758 static void enter_smm(struct kvm_vcpu *vcpu)
7759 {
7760 	struct kvm_segment cs, ds;
7761 	struct desc_ptr dt;
7762 	char buf[512];
7763 	u32 cr0;
7764 
7765 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7766 	memset(buf, 0, 512);
7767 #ifdef CONFIG_X86_64
7768 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7769 		enter_smm_save_state_64(vcpu, buf);
7770 	else
7771 #endif
7772 		enter_smm_save_state_32(vcpu, buf);
7773 
7774 	/*
7775 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7776 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7777 	 * the SMM state-save area.
7778 	 */
7779 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7780 
7781 	vcpu->arch.hflags |= HF_SMM_MASK;
7782 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7783 
7784 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7785 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7786 	else
7787 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7788 
7789 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7790 	kvm_rip_write(vcpu, 0x8000);
7791 
7792 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7793 	kvm_x86_ops->set_cr0(vcpu, cr0);
7794 	vcpu->arch.cr0 = cr0;
7795 
7796 	kvm_x86_ops->set_cr4(vcpu, 0);
7797 
7798 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7799 	dt.address = dt.size = 0;
7800 	kvm_x86_ops->set_idt(vcpu, &dt);
7801 
7802 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7803 
7804 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7805 	cs.base = vcpu->arch.smbase;
7806 
7807 	ds.selector = 0;
7808 	ds.base = 0;
7809 
7810 	cs.limit    = ds.limit = 0xffffffff;
7811 	cs.type     = ds.type = 0x3;
7812 	cs.dpl      = ds.dpl = 0;
7813 	cs.db       = ds.db = 0;
7814 	cs.s        = ds.s = 1;
7815 	cs.l        = ds.l = 0;
7816 	cs.g        = ds.g = 1;
7817 	cs.avl      = ds.avl = 0;
7818 	cs.present  = ds.present = 1;
7819 	cs.unusable = ds.unusable = 0;
7820 	cs.padding  = ds.padding = 0;
7821 
7822 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7823 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7824 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7825 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7826 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7827 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7828 
7829 #ifdef CONFIG_X86_64
7830 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7831 		kvm_x86_ops->set_efer(vcpu, 0);
7832 #endif
7833 
7834 	kvm_update_cpuid(vcpu);
7835 	kvm_mmu_reset_context(vcpu);
7836 }
7837 
7838 static void process_smi(struct kvm_vcpu *vcpu)
7839 {
7840 	vcpu->arch.smi_pending = true;
7841 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7842 }
7843 
7844 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7845 {
7846 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7847 }
7848 
7849 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7850 {
7851 	if (!kvm_apic_present(vcpu))
7852 		return;
7853 
7854 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7855 
7856 	if (irqchip_split(vcpu->kvm))
7857 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7858 	else {
7859 		if (vcpu->arch.apicv_active)
7860 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7861 		if (ioapic_in_kernel(vcpu->kvm))
7862 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7863 	}
7864 
7865 	if (is_guest_mode(vcpu))
7866 		vcpu->arch.load_eoi_exitmap_pending = true;
7867 	else
7868 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7869 }
7870 
7871 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7872 {
7873 	u64 eoi_exit_bitmap[4];
7874 
7875 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7876 		return;
7877 
7878 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7879 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7880 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7881 }
7882 
7883 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7884 		unsigned long start, unsigned long end,
7885 		bool blockable)
7886 {
7887 	unsigned long apic_address;
7888 
7889 	/*
7890 	 * The physical address of apic access page is stored in the VMCS.
7891 	 * Update it when it becomes invalid.
7892 	 */
7893 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7894 	if (start <= apic_address && apic_address < end)
7895 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7896 
7897 	return 0;
7898 }
7899 
7900 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7901 {
7902 	struct page *page = NULL;
7903 
7904 	if (!lapic_in_kernel(vcpu))
7905 		return;
7906 
7907 	if (!kvm_x86_ops->set_apic_access_page_addr)
7908 		return;
7909 
7910 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7911 	if (is_error_page(page))
7912 		return;
7913 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7914 
7915 	/*
7916 	 * Do not pin apic access page in memory, the MMU notifier
7917 	 * will call us again if it is migrated or swapped out.
7918 	 */
7919 	put_page(page);
7920 }
7921 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7922 
7923 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7924 {
7925 	smp_send_reschedule(vcpu->cpu);
7926 }
7927 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7928 
7929 /*
7930  * Returns 1 to let vcpu_run() continue the guest execution loop without
7931  * exiting to the userspace.  Otherwise, the value will be returned to the
7932  * userspace.
7933  */
7934 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7935 {
7936 	int r;
7937 	bool req_int_win =
7938 		dm_request_for_irq_injection(vcpu) &&
7939 		kvm_cpu_accept_dm_intr(vcpu);
7940 
7941 	bool req_immediate_exit = false;
7942 
7943 	if (kvm_request_pending(vcpu)) {
7944 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7945 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7946 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7947 			kvm_mmu_unload(vcpu);
7948 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7949 			__kvm_migrate_timers(vcpu);
7950 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7951 			kvm_gen_update_masterclock(vcpu->kvm);
7952 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7953 			kvm_gen_kvmclock_update(vcpu);
7954 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7955 			r = kvm_guest_time_update(vcpu);
7956 			if (unlikely(r))
7957 				goto out;
7958 		}
7959 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7960 			kvm_mmu_sync_roots(vcpu);
7961 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7962 			kvm_mmu_load_cr3(vcpu);
7963 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7964 			kvm_vcpu_flush_tlb(vcpu, true);
7965 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7966 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7967 			r = 0;
7968 			goto out;
7969 		}
7970 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7971 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7972 			vcpu->mmio_needed = 0;
7973 			r = 0;
7974 			goto out;
7975 		}
7976 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7977 			/* Page is swapped out. Do synthetic halt */
7978 			vcpu->arch.apf.halted = true;
7979 			r = 1;
7980 			goto out;
7981 		}
7982 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7983 			record_steal_time(vcpu);
7984 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7985 			process_smi(vcpu);
7986 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7987 			process_nmi(vcpu);
7988 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7989 			kvm_pmu_handle_event(vcpu);
7990 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7991 			kvm_pmu_deliver_pmi(vcpu);
7992 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7993 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7994 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7995 				     vcpu->arch.ioapic_handled_vectors)) {
7996 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7997 				vcpu->run->eoi.vector =
7998 						vcpu->arch.pending_ioapic_eoi;
7999 				r = 0;
8000 				goto out;
8001 			}
8002 		}
8003 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8004 			vcpu_scan_ioapic(vcpu);
8005 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8006 			vcpu_load_eoi_exitmap(vcpu);
8007 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8008 			kvm_vcpu_reload_apic_access_page(vcpu);
8009 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8010 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8011 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8012 			r = 0;
8013 			goto out;
8014 		}
8015 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8016 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8017 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8018 			r = 0;
8019 			goto out;
8020 		}
8021 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8022 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8023 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8024 			r = 0;
8025 			goto out;
8026 		}
8027 
8028 		/*
8029 		 * KVM_REQ_HV_STIMER has to be processed after
8030 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8031 		 * depend on the guest clock being up-to-date
8032 		 */
8033 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8034 			kvm_hv_process_stimers(vcpu);
8035 	}
8036 
8037 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8038 		++vcpu->stat.req_event;
8039 		kvm_apic_accept_events(vcpu);
8040 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8041 			r = 1;
8042 			goto out;
8043 		}
8044 
8045 		if (inject_pending_event(vcpu, req_int_win) != 0)
8046 			req_immediate_exit = true;
8047 		else {
8048 			/* Enable SMI/NMI/IRQ window open exits if needed.
8049 			 *
8050 			 * SMIs have three cases:
8051 			 * 1) They can be nested, and then there is nothing to
8052 			 *    do here because RSM will cause a vmexit anyway.
8053 			 * 2) There is an ISA-specific reason why SMI cannot be
8054 			 *    injected, and the moment when this changes can be
8055 			 *    intercepted.
8056 			 * 3) Or the SMI can be pending because
8057 			 *    inject_pending_event has completed the injection
8058 			 *    of an IRQ or NMI from the previous vmexit, and
8059 			 *    then we request an immediate exit to inject the
8060 			 *    SMI.
8061 			 */
8062 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
8063 				if (!kvm_x86_ops->enable_smi_window(vcpu))
8064 					req_immediate_exit = true;
8065 			if (vcpu->arch.nmi_pending)
8066 				kvm_x86_ops->enable_nmi_window(vcpu);
8067 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8068 				kvm_x86_ops->enable_irq_window(vcpu);
8069 			WARN_ON(vcpu->arch.exception.pending);
8070 		}
8071 
8072 		if (kvm_lapic_enabled(vcpu)) {
8073 			update_cr8_intercept(vcpu);
8074 			kvm_lapic_sync_to_vapic(vcpu);
8075 		}
8076 	}
8077 
8078 	r = kvm_mmu_reload(vcpu);
8079 	if (unlikely(r)) {
8080 		goto cancel_injection;
8081 	}
8082 
8083 	preempt_disable();
8084 
8085 	kvm_x86_ops->prepare_guest_switch(vcpu);
8086 
8087 	/*
8088 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8089 	 * IPI are then delayed after guest entry, which ensures that they
8090 	 * result in virtual interrupt delivery.
8091 	 */
8092 	local_irq_disable();
8093 	vcpu->mode = IN_GUEST_MODE;
8094 
8095 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8096 
8097 	/*
8098 	 * 1) We should set ->mode before checking ->requests.  Please see
8099 	 * the comment in kvm_vcpu_exiting_guest_mode().
8100 	 *
8101 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8102 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8103 	 * (see vmx_deliver_posted_interrupt).
8104 	 *
8105 	 * 3) This also orders the write to mode from any reads to the page
8106 	 * tables done while the VCPU is running.  Please see the comment
8107 	 * in kvm_flush_remote_tlbs.
8108 	 */
8109 	smp_mb__after_srcu_read_unlock();
8110 
8111 	/*
8112 	 * This handles the case where a posted interrupt was
8113 	 * notified with kvm_vcpu_kick.
8114 	 */
8115 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8116 		kvm_x86_ops->sync_pir_to_irr(vcpu);
8117 
8118 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8119 	    || need_resched() || signal_pending(current)) {
8120 		vcpu->mode = OUTSIDE_GUEST_MODE;
8121 		smp_wmb();
8122 		local_irq_enable();
8123 		preempt_enable();
8124 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8125 		r = 1;
8126 		goto cancel_injection;
8127 	}
8128 
8129 	if (req_immediate_exit) {
8130 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8131 		kvm_x86_ops->request_immediate_exit(vcpu);
8132 	}
8133 
8134 	trace_kvm_entry(vcpu->vcpu_id);
8135 	guest_enter_irqoff();
8136 
8137 	/* The preempt notifier should have taken care of the FPU already.  */
8138 	WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8139 
8140 	if (unlikely(vcpu->arch.switch_db_regs)) {
8141 		set_debugreg(0, 7);
8142 		set_debugreg(vcpu->arch.eff_db[0], 0);
8143 		set_debugreg(vcpu->arch.eff_db[1], 1);
8144 		set_debugreg(vcpu->arch.eff_db[2], 2);
8145 		set_debugreg(vcpu->arch.eff_db[3], 3);
8146 		set_debugreg(vcpu->arch.dr6, 6);
8147 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8148 	}
8149 
8150 	kvm_x86_ops->run(vcpu);
8151 
8152 	/*
8153 	 * Do this here before restoring debug registers on the host.  And
8154 	 * since we do this before handling the vmexit, a DR access vmexit
8155 	 * can (a) read the correct value of the debug registers, (b) set
8156 	 * KVM_DEBUGREG_WONT_EXIT again.
8157 	 */
8158 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8159 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8160 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8161 		kvm_update_dr0123(vcpu);
8162 		kvm_update_dr6(vcpu);
8163 		kvm_update_dr7(vcpu);
8164 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8165 	}
8166 
8167 	/*
8168 	 * If the guest has used debug registers, at least dr7
8169 	 * will be disabled while returning to the host.
8170 	 * If we don't have active breakpoints in the host, we don't
8171 	 * care about the messed up debug address registers. But if
8172 	 * we have some of them active, restore the old state.
8173 	 */
8174 	if (hw_breakpoint_active())
8175 		hw_breakpoint_restore();
8176 
8177 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8178 
8179 	vcpu->mode = OUTSIDE_GUEST_MODE;
8180 	smp_wmb();
8181 
8182 	kvm_x86_ops->handle_exit_irqoff(vcpu);
8183 
8184 	/*
8185 	 * Consume any pending interrupts, including the possible source of
8186 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8187 	 * An instruction is required after local_irq_enable() to fully unblock
8188 	 * interrupts on processors that implement an interrupt shadow, the
8189 	 * stat.exits increment will do nicely.
8190 	 */
8191 	kvm_before_interrupt(vcpu);
8192 	local_irq_enable();
8193 	++vcpu->stat.exits;
8194 	local_irq_disable();
8195 	kvm_after_interrupt(vcpu);
8196 
8197 	guest_exit_irqoff();
8198 	if (lapic_in_kernel(vcpu)) {
8199 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8200 		if (delta != S64_MIN) {
8201 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8202 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8203 		}
8204 	}
8205 
8206 	local_irq_enable();
8207 	preempt_enable();
8208 
8209 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8210 
8211 	/*
8212 	 * Profile KVM exit RIPs:
8213 	 */
8214 	if (unlikely(prof_on == KVM_PROFILING)) {
8215 		unsigned long rip = kvm_rip_read(vcpu);
8216 		profile_hit(KVM_PROFILING, (void *)rip);
8217 	}
8218 
8219 	if (unlikely(vcpu->arch.tsc_always_catchup))
8220 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8221 
8222 	if (vcpu->arch.apic_attention)
8223 		kvm_lapic_sync_from_vapic(vcpu);
8224 
8225 	vcpu->arch.gpa_available = false;
8226 	r = kvm_x86_ops->handle_exit(vcpu);
8227 	return r;
8228 
8229 cancel_injection:
8230 	kvm_x86_ops->cancel_injection(vcpu);
8231 	if (unlikely(vcpu->arch.apic_attention))
8232 		kvm_lapic_sync_from_vapic(vcpu);
8233 out:
8234 	return r;
8235 }
8236 
8237 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8238 {
8239 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8240 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8241 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8242 		kvm_vcpu_block(vcpu);
8243 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8244 
8245 		if (kvm_x86_ops->post_block)
8246 			kvm_x86_ops->post_block(vcpu);
8247 
8248 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8249 			return 1;
8250 	}
8251 
8252 	kvm_apic_accept_events(vcpu);
8253 	switch(vcpu->arch.mp_state) {
8254 	case KVM_MP_STATE_HALTED:
8255 		vcpu->arch.pv.pv_unhalted = false;
8256 		vcpu->arch.mp_state =
8257 			KVM_MP_STATE_RUNNABLE;
8258 		/* fall through */
8259 	case KVM_MP_STATE_RUNNABLE:
8260 		vcpu->arch.apf.halted = false;
8261 		break;
8262 	case KVM_MP_STATE_INIT_RECEIVED:
8263 		break;
8264 	default:
8265 		return -EINTR;
8266 		break;
8267 	}
8268 	return 1;
8269 }
8270 
8271 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8272 {
8273 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8274 		kvm_x86_ops->check_nested_events(vcpu, false);
8275 
8276 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8277 		!vcpu->arch.apf.halted);
8278 }
8279 
8280 static int vcpu_run(struct kvm_vcpu *vcpu)
8281 {
8282 	int r;
8283 	struct kvm *kvm = vcpu->kvm;
8284 
8285 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8286 	vcpu->arch.l1tf_flush_l1d = true;
8287 
8288 	for (;;) {
8289 		if (kvm_vcpu_running(vcpu)) {
8290 			r = vcpu_enter_guest(vcpu);
8291 		} else {
8292 			r = vcpu_block(kvm, vcpu);
8293 		}
8294 
8295 		if (r <= 0)
8296 			break;
8297 
8298 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8299 		if (kvm_cpu_has_pending_timer(vcpu))
8300 			kvm_inject_pending_timer_irqs(vcpu);
8301 
8302 		if (dm_request_for_irq_injection(vcpu) &&
8303 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8304 			r = 0;
8305 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8306 			++vcpu->stat.request_irq_exits;
8307 			break;
8308 		}
8309 
8310 		kvm_check_async_pf_completion(vcpu);
8311 
8312 		if (signal_pending(current)) {
8313 			r = -EINTR;
8314 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8315 			++vcpu->stat.signal_exits;
8316 			break;
8317 		}
8318 		if (need_resched()) {
8319 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8320 			cond_resched();
8321 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8322 		}
8323 	}
8324 
8325 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8326 
8327 	return r;
8328 }
8329 
8330 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8331 {
8332 	int r;
8333 
8334 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8335 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8336 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8337 	return r;
8338 }
8339 
8340 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8341 {
8342 	BUG_ON(!vcpu->arch.pio.count);
8343 
8344 	return complete_emulated_io(vcpu);
8345 }
8346 
8347 /*
8348  * Implements the following, as a state machine:
8349  *
8350  * read:
8351  *   for each fragment
8352  *     for each mmio piece in the fragment
8353  *       write gpa, len
8354  *       exit
8355  *       copy data
8356  *   execute insn
8357  *
8358  * write:
8359  *   for each fragment
8360  *     for each mmio piece in the fragment
8361  *       write gpa, len
8362  *       copy data
8363  *       exit
8364  */
8365 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8366 {
8367 	struct kvm_run *run = vcpu->run;
8368 	struct kvm_mmio_fragment *frag;
8369 	unsigned len;
8370 
8371 	BUG_ON(!vcpu->mmio_needed);
8372 
8373 	/* Complete previous fragment */
8374 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8375 	len = min(8u, frag->len);
8376 	if (!vcpu->mmio_is_write)
8377 		memcpy(frag->data, run->mmio.data, len);
8378 
8379 	if (frag->len <= 8) {
8380 		/* Switch to the next fragment. */
8381 		frag++;
8382 		vcpu->mmio_cur_fragment++;
8383 	} else {
8384 		/* Go forward to the next mmio piece. */
8385 		frag->data += len;
8386 		frag->gpa += len;
8387 		frag->len -= len;
8388 	}
8389 
8390 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8391 		vcpu->mmio_needed = 0;
8392 
8393 		/* FIXME: return into emulator if single-stepping.  */
8394 		if (vcpu->mmio_is_write)
8395 			return 1;
8396 		vcpu->mmio_read_completed = 1;
8397 		return complete_emulated_io(vcpu);
8398 	}
8399 
8400 	run->exit_reason = KVM_EXIT_MMIO;
8401 	run->mmio.phys_addr = frag->gpa;
8402 	if (vcpu->mmio_is_write)
8403 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8404 	run->mmio.len = min(8u, frag->len);
8405 	run->mmio.is_write = vcpu->mmio_is_write;
8406 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8407 	return 0;
8408 }
8409 
8410 /* Swap (qemu) user FPU context for the guest FPU context. */
8411 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8412 {
8413 	fpregs_lock();
8414 
8415 	copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8416 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8417 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8418 				~XFEATURE_MASK_PKRU);
8419 
8420 	fpregs_mark_activate();
8421 	fpregs_unlock();
8422 
8423 	trace_kvm_fpu(1);
8424 }
8425 
8426 /* When vcpu_run ends, restore user space FPU context. */
8427 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8428 {
8429 	fpregs_lock();
8430 
8431 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8432 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8433 
8434 	fpregs_mark_activate();
8435 	fpregs_unlock();
8436 
8437 	++vcpu->stat.fpu_reload;
8438 	trace_kvm_fpu(0);
8439 }
8440 
8441 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8442 {
8443 	int r;
8444 
8445 	vcpu_load(vcpu);
8446 	kvm_sigset_activate(vcpu);
8447 	kvm_load_guest_fpu(vcpu);
8448 
8449 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8450 		if (kvm_run->immediate_exit) {
8451 			r = -EINTR;
8452 			goto out;
8453 		}
8454 		kvm_vcpu_block(vcpu);
8455 		kvm_apic_accept_events(vcpu);
8456 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8457 		r = -EAGAIN;
8458 		if (signal_pending(current)) {
8459 			r = -EINTR;
8460 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8461 			++vcpu->stat.signal_exits;
8462 		}
8463 		goto out;
8464 	}
8465 
8466 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8467 		r = -EINVAL;
8468 		goto out;
8469 	}
8470 
8471 	if (vcpu->run->kvm_dirty_regs) {
8472 		r = sync_regs(vcpu);
8473 		if (r != 0)
8474 			goto out;
8475 	}
8476 
8477 	/* re-sync apic's tpr */
8478 	if (!lapic_in_kernel(vcpu)) {
8479 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8480 			r = -EINVAL;
8481 			goto out;
8482 		}
8483 	}
8484 
8485 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8486 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8487 		vcpu->arch.complete_userspace_io = NULL;
8488 		r = cui(vcpu);
8489 		if (r <= 0)
8490 			goto out;
8491 	} else
8492 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8493 
8494 	if (kvm_run->immediate_exit)
8495 		r = -EINTR;
8496 	else
8497 		r = vcpu_run(vcpu);
8498 
8499 out:
8500 	kvm_put_guest_fpu(vcpu);
8501 	if (vcpu->run->kvm_valid_regs)
8502 		store_regs(vcpu);
8503 	post_kvm_run_save(vcpu);
8504 	kvm_sigset_deactivate(vcpu);
8505 
8506 	vcpu_put(vcpu);
8507 	return r;
8508 }
8509 
8510 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8511 {
8512 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8513 		/*
8514 		 * We are here if userspace calls get_regs() in the middle of
8515 		 * instruction emulation. Registers state needs to be copied
8516 		 * back from emulation context to vcpu. Userspace shouldn't do
8517 		 * that usually, but some bad designed PV devices (vmware
8518 		 * backdoor interface) need this to work
8519 		 */
8520 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8521 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8522 	}
8523 	regs->rax = kvm_rax_read(vcpu);
8524 	regs->rbx = kvm_rbx_read(vcpu);
8525 	regs->rcx = kvm_rcx_read(vcpu);
8526 	regs->rdx = kvm_rdx_read(vcpu);
8527 	regs->rsi = kvm_rsi_read(vcpu);
8528 	regs->rdi = kvm_rdi_read(vcpu);
8529 	regs->rsp = kvm_rsp_read(vcpu);
8530 	regs->rbp = kvm_rbp_read(vcpu);
8531 #ifdef CONFIG_X86_64
8532 	regs->r8 = kvm_r8_read(vcpu);
8533 	regs->r9 = kvm_r9_read(vcpu);
8534 	regs->r10 = kvm_r10_read(vcpu);
8535 	regs->r11 = kvm_r11_read(vcpu);
8536 	regs->r12 = kvm_r12_read(vcpu);
8537 	regs->r13 = kvm_r13_read(vcpu);
8538 	regs->r14 = kvm_r14_read(vcpu);
8539 	regs->r15 = kvm_r15_read(vcpu);
8540 #endif
8541 
8542 	regs->rip = kvm_rip_read(vcpu);
8543 	regs->rflags = kvm_get_rflags(vcpu);
8544 }
8545 
8546 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8547 {
8548 	vcpu_load(vcpu);
8549 	__get_regs(vcpu, regs);
8550 	vcpu_put(vcpu);
8551 	return 0;
8552 }
8553 
8554 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8555 {
8556 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8557 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8558 
8559 	kvm_rax_write(vcpu, regs->rax);
8560 	kvm_rbx_write(vcpu, regs->rbx);
8561 	kvm_rcx_write(vcpu, regs->rcx);
8562 	kvm_rdx_write(vcpu, regs->rdx);
8563 	kvm_rsi_write(vcpu, regs->rsi);
8564 	kvm_rdi_write(vcpu, regs->rdi);
8565 	kvm_rsp_write(vcpu, regs->rsp);
8566 	kvm_rbp_write(vcpu, regs->rbp);
8567 #ifdef CONFIG_X86_64
8568 	kvm_r8_write(vcpu, regs->r8);
8569 	kvm_r9_write(vcpu, regs->r9);
8570 	kvm_r10_write(vcpu, regs->r10);
8571 	kvm_r11_write(vcpu, regs->r11);
8572 	kvm_r12_write(vcpu, regs->r12);
8573 	kvm_r13_write(vcpu, regs->r13);
8574 	kvm_r14_write(vcpu, regs->r14);
8575 	kvm_r15_write(vcpu, regs->r15);
8576 #endif
8577 
8578 	kvm_rip_write(vcpu, regs->rip);
8579 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8580 
8581 	vcpu->arch.exception.pending = false;
8582 
8583 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8584 }
8585 
8586 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8587 {
8588 	vcpu_load(vcpu);
8589 	__set_regs(vcpu, regs);
8590 	vcpu_put(vcpu);
8591 	return 0;
8592 }
8593 
8594 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8595 {
8596 	struct kvm_segment cs;
8597 
8598 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8599 	*db = cs.db;
8600 	*l = cs.l;
8601 }
8602 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8603 
8604 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8605 {
8606 	struct desc_ptr dt;
8607 
8608 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8609 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8610 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8611 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8612 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8613 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8614 
8615 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8616 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8617 
8618 	kvm_x86_ops->get_idt(vcpu, &dt);
8619 	sregs->idt.limit = dt.size;
8620 	sregs->idt.base = dt.address;
8621 	kvm_x86_ops->get_gdt(vcpu, &dt);
8622 	sregs->gdt.limit = dt.size;
8623 	sregs->gdt.base = dt.address;
8624 
8625 	sregs->cr0 = kvm_read_cr0(vcpu);
8626 	sregs->cr2 = vcpu->arch.cr2;
8627 	sregs->cr3 = kvm_read_cr3(vcpu);
8628 	sregs->cr4 = kvm_read_cr4(vcpu);
8629 	sregs->cr8 = kvm_get_cr8(vcpu);
8630 	sregs->efer = vcpu->arch.efer;
8631 	sregs->apic_base = kvm_get_apic_base(vcpu);
8632 
8633 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8634 
8635 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8636 		set_bit(vcpu->arch.interrupt.nr,
8637 			(unsigned long *)sregs->interrupt_bitmap);
8638 }
8639 
8640 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8641 				  struct kvm_sregs *sregs)
8642 {
8643 	vcpu_load(vcpu);
8644 	__get_sregs(vcpu, sregs);
8645 	vcpu_put(vcpu);
8646 	return 0;
8647 }
8648 
8649 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8650 				    struct kvm_mp_state *mp_state)
8651 {
8652 	vcpu_load(vcpu);
8653 
8654 	kvm_apic_accept_events(vcpu);
8655 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8656 					vcpu->arch.pv.pv_unhalted)
8657 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8658 	else
8659 		mp_state->mp_state = vcpu->arch.mp_state;
8660 
8661 	vcpu_put(vcpu);
8662 	return 0;
8663 }
8664 
8665 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8666 				    struct kvm_mp_state *mp_state)
8667 {
8668 	int ret = -EINVAL;
8669 
8670 	vcpu_load(vcpu);
8671 
8672 	if (!lapic_in_kernel(vcpu) &&
8673 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8674 		goto out;
8675 
8676 	/* INITs are latched while in SMM */
8677 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8678 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8679 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8680 		goto out;
8681 
8682 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8683 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8684 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8685 	} else
8686 		vcpu->arch.mp_state = mp_state->mp_state;
8687 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8688 
8689 	ret = 0;
8690 out:
8691 	vcpu_put(vcpu);
8692 	return ret;
8693 }
8694 
8695 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8696 		    int reason, bool has_error_code, u32 error_code)
8697 {
8698 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8699 	int ret;
8700 
8701 	init_emulate_ctxt(vcpu);
8702 
8703 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8704 				   has_error_code, error_code);
8705 	if (ret) {
8706 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8707 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8708 		vcpu->run->internal.ndata = 0;
8709 		return 0;
8710 	}
8711 
8712 	kvm_rip_write(vcpu, ctxt->eip);
8713 	kvm_set_rflags(vcpu, ctxt->eflags);
8714 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8715 	return 1;
8716 }
8717 EXPORT_SYMBOL_GPL(kvm_task_switch);
8718 
8719 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8720 {
8721 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8722 		/*
8723 		 * When EFER.LME and CR0.PG are set, the processor is in
8724 		 * 64-bit mode (though maybe in a 32-bit code segment).
8725 		 * CR4.PAE and EFER.LMA must be set.
8726 		 */
8727 		if (!(sregs->cr4 & X86_CR4_PAE)
8728 		    || !(sregs->efer & EFER_LMA))
8729 			return -EINVAL;
8730 	} else {
8731 		/*
8732 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8733 		 * segment cannot be 64-bit.
8734 		 */
8735 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8736 			return -EINVAL;
8737 	}
8738 
8739 	return kvm_valid_cr4(vcpu, sregs->cr4);
8740 }
8741 
8742 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8743 {
8744 	struct msr_data apic_base_msr;
8745 	int mmu_reset_needed = 0;
8746 	int cpuid_update_needed = 0;
8747 	int pending_vec, max_bits, idx;
8748 	struct desc_ptr dt;
8749 	int ret = -EINVAL;
8750 
8751 	if (kvm_valid_sregs(vcpu, sregs))
8752 		goto out;
8753 
8754 	apic_base_msr.data = sregs->apic_base;
8755 	apic_base_msr.host_initiated = true;
8756 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8757 		goto out;
8758 
8759 	dt.size = sregs->idt.limit;
8760 	dt.address = sregs->idt.base;
8761 	kvm_x86_ops->set_idt(vcpu, &dt);
8762 	dt.size = sregs->gdt.limit;
8763 	dt.address = sregs->gdt.base;
8764 	kvm_x86_ops->set_gdt(vcpu, &dt);
8765 
8766 	vcpu->arch.cr2 = sregs->cr2;
8767 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8768 	vcpu->arch.cr3 = sregs->cr3;
8769 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8770 
8771 	kvm_set_cr8(vcpu, sregs->cr8);
8772 
8773 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8774 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8775 
8776 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8777 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8778 	vcpu->arch.cr0 = sregs->cr0;
8779 
8780 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8781 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8782 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8783 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8784 	if (cpuid_update_needed)
8785 		kvm_update_cpuid(vcpu);
8786 
8787 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8788 	if (is_pae_paging(vcpu)) {
8789 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8790 		mmu_reset_needed = 1;
8791 	}
8792 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8793 
8794 	if (mmu_reset_needed)
8795 		kvm_mmu_reset_context(vcpu);
8796 
8797 	max_bits = KVM_NR_INTERRUPTS;
8798 	pending_vec = find_first_bit(
8799 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8800 	if (pending_vec < max_bits) {
8801 		kvm_queue_interrupt(vcpu, pending_vec, false);
8802 		pr_debug("Set back pending irq %d\n", pending_vec);
8803 	}
8804 
8805 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8806 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8807 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8808 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8809 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8810 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8811 
8812 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8813 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8814 
8815 	update_cr8_intercept(vcpu);
8816 
8817 	/* Older userspace won't unhalt the vcpu on reset. */
8818 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8819 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8820 	    !is_protmode(vcpu))
8821 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8822 
8823 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8824 
8825 	ret = 0;
8826 out:
8827 	return ret;
8828 }
8829 
8830 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8831 				  struct kvm_sregs *sregs)
8832 {
8833 	int ret;
8834 
8835 	vcpu_load(vcpu);
8836 	ret = __set_sregs(vcpu, sregs);
8837 	vcpu_put(vcpu);
8838 	return ret;
8839 }
8840 
8841 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8842 					struct kvm_guest_debug *dbg)
8843 {
8844 	unsigned long rflags;
8845 	int i, r;
8846 
8847 	vcpu_load(vcpu);
8848 
8849 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8850 		r = -EBUSY;
8851 		if (vcpu->arch.exception.pending)
8852 			goto out;
8853 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8854 			kvm_queue_exception(vcpu, DB_VECTOR);
8855 		else
8856 			kvm_queue_exception(vcpu, BP_VECTOR);
8857 	}
8858 
8859 	/*
8860 	 * Read rflags as long as potentially injected trace flags are still
8861 	 * filtered out.
8862 	 */
8863 	rflags = kvm_get_rflags(vcpu);
8864 
8865 	vcpu->guest_debug = dbg->control;
8866 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8867 		vcpu->guest_debug = 0;
8868 
8869 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8870 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8871 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8872 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8873 	} else {
8874 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8875 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8876 	}
8877 	kvm_update_dr7(vcpu);
8878 
8879 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8880 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8881 			get_segment_base(vcpu, VCPU_SREG_CS);
8882 
8883 	/*
8884 	 * Trigger an rflags update that will inject or remove the trace
8885 	 * flags.
8886 	 */
8887 	kvm_set_rflags(vcpu, rflags);
8888 
8889 	kvm_x86_ops->update_bp_intercept(vcpu);
8890 
8891 	r = 0;
8892 
8893 out:
8894 	vcpu_put(vcpu);
8895 	return r;
8896 }
8897 
8898 /*
8899  * Translate a guest virtual address to a guest physical address.
8900  */
8901 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8902 				    struct kvm_translation *tr)
8903 {
8904 	unsigned long vaddr = tr->linear_address;
8905 	gpa_t gpa;
8906 	int idx;
8907 
8908 	vcpu_load(vcpu);
8909 
8910 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8911 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8912 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8913 	tr->physical_address = gpa;
8914 	tr->valid = gpa != UNMAPPED_GVA;
8915 	tr->writeable = 1;
8916 	tr->usermode = 0;
8917 
8918 	vcpu_put(vcpu);
8919 	return 0;
8920 }
8921 
8922 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8923 {
8924 	struct fxregs_state *fxsave;
8925 
8926 	vcpu_load(vcpu);
8927 
8928 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8929 	memcpy(fpu->fpr, fxsave->st_space, 128);
8930 	fpu->fcw = fxsave->cwd;
8931 	fpu->fsw = fxsave->swd;
8932 	fpu->ftwx = fxsave->twd;
8933 	fpu->last_opcode = fxsave->fop;
8934 	fpu->last_ip = fxsave->rip;
8935 	fpu->last_dp = fxsave->rdp;
8936 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8937 
8938 	vcpu_put(vcpu);
8939 	return 0;
8940 }
8941 
8942 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8943 {
8944 	struct fxregs_state *fxsave;
8945 
8946 	vcpu_load(vcpu);
8947 
8948 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8949 
8950 	memcpy(fxsave->st_space, fpu->fpr, 128);
8951 	fxsave->cwd = fpu->fcw;
8952 	fxsave->swd = fpu->fsw;
8953 	fxsave->twd = fpu->ftwx;
8954 	fxsave->fop = fpu->last_opcode;
8955 	fxsave->rip = fpu->last_ip;
8956 	fxsave->rdp = fpu->last_dp;
8957 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8958 
8959 	vcpu_put(vcpu);
8960 	return 0;
8961 }
8962 
8963 static void store_regs(struct kvm_vcpu *vcpu)
8964 {
8965 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8966 
8967 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8968 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8969 
8970 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8971 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8972 
8973 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8974 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8975 				vcpu, &vcpu->run->s.regs.events);
8976 }
8977 
8978 static int sync_regs(struct kvm_vcpu *vcpu)
8979 {
8980 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8981 		return -EINVAL;
8982 
8983 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8984 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8985 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8986 	}
8987 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8988 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8989 			return -EINVAL;
8990 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8991 	}
8992 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8993 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8994 				vcpu, &vcpu->run->s.regs.events))
8995 			return -EINVAL;
8996 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8997 	}
8998 
8999 	return 0;
9000 }
9001 
9002 static void fx_init(struct kvm_vcpu *vcpu)
9003 {
9004 	fpstate_init(&vcpu->arch.guest_fpu->state);
9005 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9006 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9007 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9008 
9009 	/*
9010 	 * Ensure guest xcr0 is valid for loading
9011 	 */
9012 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9013 
9014 	vcpu->arch.cr0 |= X86_CR0_ET;
9015 }
9016 
9017 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9018 {
9019 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9020 
9021 	kvmclock_reset(vcpu);
9022 
9023 	kvm_x86_ops->vcpu_free(vcpu);
9024 	free_cpumask_var(wbinvd_dirty_mask);
9025 }
9026 
9027 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9028 						unsigned int id)
9029 {
9030 	struct kvm_vcpu *vcpu;
9031 
9032 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9033 		printk_once(KERN_WARNING
9034 		"kvm: SMP vm created on host with unstable TSC; "
9035 		"guest TSC will not be reliable\n");
9036 
9037 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9038 
9039 	return vcpu;
9040 }
9041 
9042 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9043 {
9044 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9045 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9046 	kvm_vcpu_mtrr_init(vcpu);
9047 	vcpu_load(vcpu);
9048 	kvm_vcpu_reset(vcpu, false);
9049 	kvm_init_mmu(vcpu, false);
9050 	vcpu_put(vcpu);
9051 	return 0;
9052 }
9053 
9054 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9055 {
9056 	struct msr_data msr;
9057 	struct kvm *kvm = vcpu->kvm;
9058 
9059 	kvm_hv_vcpu_postcreate(vcpu);
9060 
9061 	if (mutex_lock_killable(&vcpu->mutex))
9062 		return;
9063 	vcpu_load(vcpu);
9064 	msr.data = 0x0;
9065 	msr.index = MSR_IA32_TSC;
9066 	msr.host_initiated = true;
9067 	kvm_write_tsc(vcpu, &msr);
9068 	vcpu_put(vcpu);
9069 
9070 	/* poll control enabled by default */
9071 	vcpu->arch.msr_kvm_poll_control = 1;
9072 
9073 	mutex_unlock(&vcpu->mutex);
9074 
9075 	if (!kvmclock_periodic_sync)
9076 		return;
9077 
9078 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9079 					KVMCLOCK_SYNC_PERIOD);
9080 }
9081 
9082 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9083 {
9084 	vcpu->arch.apf.msr_val = 0;
9085 
9086 	vcpu_load(vcpu);
9087 	kvm_mmu_unload(vcpu);
9088 	vcpu_put(vcpu);
9089 
9090 	kvm_x86_ops->vcpu_free(vcpu);
9091 }
9092 
9093 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9094 {
9095 	kvm_lapic_reset(vcpu, init_event);
9096 
9097 	vcpu->arch.hflags = 0;
9098 
9099 	vcpu->arch.smi_pending = 0;
9100 	vcpu->arch.smi_count = 0;
9101 	atomic_set(&vcpu->arch.nmi_queued, 0);
9102 	vcpu->arch.nmi_pending = 0;
9103 	vcpu->arch.nmi_injected = false;
9104 	kvm_clear_interrupt_queue(vcpu);
9105 	kvm_clear_exception_queue(vcpu);
9106 	vcpu->arch.exception.pending = false;
9107 
9108 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9109 	kvm_update_dr0123(vcpu);
9110 	vcpu->arch.dr6 = DR6_INIT;
9111 	kvm_update_dr6(vcpu);
9112 	vcpu->arch.dr7 = DR7_FIXED_1;
9113 	kvm_update_dr7(vcpu);
9114 
9115 	vcpu->arch.cr2 = 0;
9116 
9117 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9118 	vcpu->arch.apf.msr_val = 0;
9119 	vcpu->arch.st.msr_val = 0;
9120 
9121 	kvmclock_reset(vcpu);
9122 
9123 	kvm_clear_async_pf_completion_queue(vcpu);
9124 	kvm_async_pf_hash_reset(vcpu);
9125 	vcpu->arch.apf.halted = false;
9126 
9127 	if (kvm_mpx_supported()) {
9128 		void *mpx_state_buffer;
9129 
9130 		/*
9131 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9132 		 * called with loaded FPU and does not let userspace fix the state.
9133 		 */
9134 		if (init_event)
9135 			kvm_put_guest_fpu(vcpu);
9136 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9137 					XFEATURE_BNDREGS);
9138 		if (mpx_state_buffer)
9139 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9140 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9141 					XFEATURE_BNDCSR);
9142 		if (mpx_state_buffer)
9143 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9144 		if (init_event)
9145 			kvm_load_guest_fpu(vcpu);
9146 	}
9147 
9148 	if (!init_event) {
9149 		kvm_pmu_reset(vcpu);
9150 		vcpu->arch.smbase = 0x30000;
9151 
9152 		vcpu->arch.msr_misc_features_enables = 0;
9153 
9154 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9155 	}
9156 
9157 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9158 	vcpu->arch.regs_avail = ~0;
9159 	vcpu->arch.regs_dirty = ~0;
9160 
9161 	vcpu->arch.ia32_xss = 0;
9162 
9163 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
9164 }
9165 
9166 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9167 {
9168 	struct kvm_segment cs;
9169 
9170 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9171 	cs.selector = vector << 8;
9172 	cs.base = vector << 12;
9173 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9174 	kvm_rip_write(vcpu, 0);
9175 }
9176 
9177 int kvm_arch_hardware_enable(void)
9178 {
9179 	struct kvm *kvm;
9180 	struct kvm_vcpu *vcpu;
9181 	int i;
9182 	int ret;
9183 	u64 local_tsc;
9184 	u64 max_tsc = 0;
9185 	bool stable, backwards_tsc = false;
9186 
9187 	kvm_shared_msr_cpu_online();
9188 	ret = kvm_x86_ops->hardware_enable();
9189 	if (ret != 0)
9190 		return ret;
9191 
9192 	local_tsc = rdtsc();
9193 	stable = !kvm_check_tsc_unstable();
9194 	list_for_each_entry(kvm, &vm_list, vm_list) {
9195 		kvm_for_each_vcpu(i, vcpu, kvm) {
9196 			if (!stable && vcpu->cpu == smp_processor_id())
9197 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9198 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9199 				backwards_tsc = true;
9200 				if (vcpu->arch.last_host_tsc > max_tsc)
9201 					max_tsc = vcpu->arch.last_host_tsc;
9202 			}
9203 		}
9204 	}
9205 
9206 	/*
9207 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9208 	 * platforms that reset TSC during suspend or hibernate actions, but
9209 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9210 	 * detect that condition here, which happens early in CPU bringup,
9211 	 * before any KVM threads can be running.  Unfortunately, we can't
9212 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9213 	 * enough into CPU bringup that we know how much real time has actually
9214 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9215 	 * variables that haven't been updated yet.
9216 	 *
9217 	 * So we simply find the maximum observed TSC above, then record the
9218 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9219 	 * the adjustment will be applied.  Note that we accumulate
9220 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9221 	 * gets a chance to run again.  In the event that no KVM threads get a
9222 	 * chance to run, we will miss the entire elapsed period, as we'll have
9223 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9224 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9225 	 * uniform across all VCPUs (not to mention the scenario is extremely
9226 	 * unlikely). It is possible that a second hibernate recovery happens
9227 	 * much faster than a first, causing the observed TSC here to be
9228 	 * smaller; this would require additional padding adjustment, which is
9229 	 * why we set last_host_tsc to the local tsc observed here.
9230 	 *
9231 	 * N.B. - this code below runs only on platforms with reliable TSC,
9232 	 * as that is the only way backwards_tsc is set above.  Also note
9233 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9234 	 * have the same delta_cyc adjustment applied if backwards_tsc
9235 	 * is detected.  Note further, this adjustment is only done once,
9236 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9237 	 * called multiple times (one for each physical CPU bringup).
9238 	 *
9239 	 * Platforms with unreliable TSCs don't have to deal with this, they
9240 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9241 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9242 	 * guarantee that they stay in perfect synchronization.
9243 	 */
9244 	if (backwards_tsc) {
9245 		u64 delta_cyc = max_tsc - local_tsc;
9246 		list_for_each_entry(kvm, &vm_list, vm_list) {
9247 			kvm->arch.backwards_tsc_observed = true;
9248 			kvm_for_each_vcpu(i, vcpu, kvm) {
9249 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9250 				vcpu->arch.last_host_tsc = local_tsc;
9251 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9252 			}
9253 
9254 			/*
9255 			 * We have to disable TSC offset matching.. if you were
9256 			 * booting a VM while issuing an S4 host suspend....
9257 			 * you may have some problem.  Solving this issue is
9258 			 * left as an exercise to the reader.
9259 			 */
9260 			kvm->arch.last_tsc_nsec = 0;
9261 			kvm->arch.last_tsc_write = 0;
9262 		}
9263 
9264 	}
9265 	return 0;
9266 }
9267 
9268 void kvm_arch_hardware_disable(void)
9269 {
9270 	kvm_x86_ops->hardware_disable();
9271 	drop_user_return_notifiers();
9272 }
9273 
9274 int kvm_arch_hardware_setup(void)
9275 {
9276 	int r;
9277 
9278 	r = kvm_x86_ops->hardware_setup();
9279 	if (r != 0)
9280 		return r;
9281 
9282 	if (kvm_has_tsc_control) {
9283 		/*
9284 		 * Make sure the user can only configure tsc_khz values that
9285 		 * fit into a signed integer.
9286 		 * A min value is not calculated because it will always
9287 		 * be 1 on all machines.
9288 		 */
9289 		u64 max = min(0x7fffffffULL,
9290 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9291 		kvm_max_guest_tsc_khz = max;
9292 
9293 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9294 	}
9295 
9296 	kvm_init_msr_list();
9297 	return 0;
9298 }
9299 
9300 void kvm_arch_hardware_unsetup(void)
9301 {
9302 	kvm_x86_ops->hardware_unsetup();
9303 }
9304 
9305 int kvm_arch_check_processor_compat(void)
9306 {
9307 	return kvm_x86_ops->check_processor_compatibility();
9308 }
9309 
9310 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9311 {
9312 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9313 }
9314 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9315 
9316 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9317 {
9318 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9319 }
9320 
9321 struct static_key kvm_no_apic_vcpu __read_mostly;
9322 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9323 
9324 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9325 {
9326 	struct page *page;
9327 	int r;
9328 
9329 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9330 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9331 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9332 	else
9333 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9334 
9335 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9336 	if (!page) {
9337 		r = -ENOMEM;
9338 		goto fail;
9339 	}
9340 	vcpu->arch.pio_data = page_address(page);
9341 
9342 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9343 
9344 	r = kvm_mmu_create(vcpu);
9345 	if (r < 0)
9346 		goto fail_free_pio_data;
9347 
9348 	if (irqchip_in_kernel(vcpu->kvm)) {
9349 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9350 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9351 		if (r < 0)
9352 			goto fail_mmu_destroy;
9353 	} else
9354 		static_key_slow_inc(&kvm_no_apic_vcpu);
9355 
9356 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9357 				       GFP_KERNEL_ACCOUNT);
9358 	if (!vcpu->arch.mce_banks) {
9359 		r = -ENOMEM;
9360 		goto fail_free_lapic;
9361 	}
9362 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9363 
9364 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9365 				GFP_KERNEL_ACCOUNT)) {
9366 		r = -ENOMEM;
9367 		goto fail_free_mce_banks;
9368 	}
9369 
9370 	fx_init(vcpu);
9371 
9372 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9373 
9374 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9375 
9376 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9377 
9378 	kvm_async_pf_hash_reset(vcpu);
9379 	kvm_pmu_init(vcpu);
9380 
9381 	vcpu->arch.pending_external_vector = -1;
9382 	vcpu->arch.preempted_in_kernel = false;
9383 
9384 	kvm_hv_vcpu_init(vcpu);
9385 
9386 	return 0;
9387 
9388 fail_free_mce_banks:
9389 	kfree(vcpu->arch.mce_banks);
9390 fail_free_lapic:
9391 	kvm_free_lapic(vcpu);
9392 fail_mmu_destroy:
9393 	kvm_mmu_destroy(vcpu);
9394 fail_free_pio_data:
9395 	free_page((unsigned long)vcpu->arch.pio_data);
9396 fail:
9397 	return r;
9398 }
9399 
9400 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9401 {
9402 	int idx;
9403 
9404 	kvm_hv_vcpu_uninit(vcpu);
9405 	kvm_pmu_destroy(vcpu);
9406 	kfree(vcpu->arch.mce_banks);
9407 	kvm_free_lapic(vcpu);
9408 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9409 	kvm_mmu_destroy(vcpu);
9410 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9411 	free_page((unsigned long)vcpu->arch.pio_data);
9412 	if (!lapic_in_kernel(vcpu))
9413 		static_key_slow_dec(&kvm_no_apic_vcpu);
9414 }
9415 
9416 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9417 {
9418 	vcpu->arch.l1tf_flush_l1d = true;
9419 	kvm_x86_ops->sched_in(vcpu, cpu);
9420 }
9421 
9422 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9423 {
9424 	if (type)
9425 		return -EINVAL;
9426 
9427 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9428 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9429 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9430 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9431 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9432 
9433 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9434 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9435 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9436 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9437 		&kvm->arch.irq_sources_bitmap);
9438 
9439 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9440 	mutex_init(&kvm->arch.apic_map_lock);
9441 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9442 
9443 	kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9444 	pvclock_update_vm_gtod_copy(kvm);
9445 
9446 	kvm->arch.guest_can_read_msr_platform_info = true;
9447 
9448 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9449 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9450 
9451 	kvm_hv_init_vm(kvm);
9452 	kvm_page_track_init(kvm);
9453 	kvm_mmu_init_vm(kvm);
9454 
9455 	return kvm_x86_ops->vm_init(kvm);
9456 }
9457 
9458 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9459 {
9460 	vcpu_load(vcpu);
9461 	kvm_mmu_unload(vcpu);
9462 	vcpu_put(vcpu);
9463 }
9464 
9465 static void kvm_free_vcpus(struct kvm *kvm)
9466 {
9467 	unsigned int i;
9468 	struct kvm_vcpu *vcpu;
9469 
9470 	/*
9471 	 * Unpin any mmu pages first.
9472 	 */
9473 	kvm_for_each_vcpu(i, vcpu, kvm) {
9474 		kvm_clear_async_pf_completion_queue(vcpu);
9475 		kvm_unload_vcpu_mmu(vcpu);
9476 	}
9477 	kvm_for_each_vcpu(i, vcpu, kvm)
9478 		kvm_arch_vcpu_free(vcpu);
9479 
9480 	mutex_lock(&kvm->lock);
9481 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9482 		kvm->vcpus[i] = NULL;
9483 
9484 	atomic_set(&kvm->online_vcpus, 0);
9485 	mutex_unlock(&kvm->lock);
9486 }
9487 
9488 void kvm_arch_sync_events(struct kvm *kvm)
9489 {
9490 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9491 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9492 	kvm_free_pit(kvm);
9493 }
9494 
9495 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9496 {
9497 	int i, r;
9498 	unsigned long hva;
9499 	struct kvm_memslots *slots = kvm_memslots(kvm);
9500 	struct kvm_memory_slot *slot, old;
9501 
9502 	/* Called with kvm->slots_lock held.  */
9503 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9504 		return -EINVAL;
9505 
9506 	slot = id_to_memslot(slots, id);
9507 	if (size) {
9508 		if (slot->npages)
9509 			return -EEXIST;
9510 
9511 		/*
9512 		 * MAP_SHARED to prevent internal slot pages from being moved
9513 		 * by fork()/COW.
9514 		 */
9515 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9516 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9517 		if (IS_ERR((void *)hva))
9518 			return PTR_ERR((void *)hva);
9519 	} else {
9520 		if (!slot->npages)
9521 			return 0;
9522 
9523 		hva = 0;
9524 	}
9525 
9526 	old = *slot;
9527 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9528 		struct kvm_userspace_memory_region m;
9529 
9530 		m.slot = id | (i << 16);
9531 		m.flags = 0;
9532 		m.guest_phys_addr = gpa;
9533 		m.userspace_addr = hva;
9534 		m.memory_size = size;
9535 		r = __kvm_set_memory_region(kvm, &m);
9536 		if (r < 0)
9537 			return r;
9538 	}
9539 
9540 	if (!size)
9541 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9542 
9543 	return 0;
9544 }
9545 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9546 
9547 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9548 {
9549 	int r;
9550 
9551 	mutex_lock(&kvm->slots_lock);
9552 	r = __x86_set_memory_region(kvm, id, gpa, size);
9553 	mutex_unlock(&kvm->slots_lock);
9554 
9555 	return r;
9556 }
9557 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9558 
9559 void kvm_arch_destroy_vm(struct kvm *kvm)
9560 {
9561 	if (current->mm == kvm->mm) {
9562 		/*
9563 		 * Free memory regions allocated on behalf of userspace,
9564 		 * unless the the memory map has changed due to process exit
9565 		 * or fd copying.
9566 		 */
9567 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9568 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9569 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9570 	}
9571 	if (kvm_x86_ops->vm_destroy)
9572 		kvm_x86_ops->vm_destroy(kvm);
9573 	kvm_pic_destroy(kvm);
9574 	kvm_ioapic_destroy(kvm);
9575 	kvm_free_vcpus(kvm);
9576 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9577 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9578 	kvm_mmu_uninit_vm(kvm);
9579 	kvm_page_track_cleanup(kvm);
9580 	kvm_hv_destroy_vm(kvm);
9581 }
9582 
9583 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9584 			   struct kvm_memory_slot *dont)
9585 {
9586 	int i;
9587 
9588 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9589 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9590 			kvfree(free->arch.rmap[i]);
9591 			free->arch.rmap[i] = NULL;
9592 		}
9593 		if (i == 0)
9594 			continue;
9595 
9596 		if (!dont || free->arch.lpage_info[i - 1] !=
9597 			     dont->arch.lpage_info[i - 1]) {
9598 			kvfree(free->arch.lpage_info[i - 1]);
9599 			free->arch.lpage_info[i - 1] = NULL;
9600 		}
9601 	}
9602 
9603 	kvm_page_track_free_memslot(free, dont);
9604 }
9605 
9606 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9607 			    unsigned long npages)
9608 {
9609 	int i;
9610 
9611 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9612 		struct kvm_lpage_info *linfo;
9613 		unsigned long ugfn;
9614 		int lpages;
9615 		int level = i + 1;
9616 
9617 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9618 				      slot->base_gfn, level) + 1;
9619 
9620 		slot->arch.rmap[i] =
9621 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9622 				 GFP_KERNEL_ACCOUNT);
9623 		if (!slot->arch.rmap[i])
9624 			goto out_free;
9625 		if (i == 0)
9626 			continue;
9627 
9628 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9629 		if (!linfo)
9630 			goto out_free;
9631 
9632 		slot->arch.lpage_info[i - 1] = linfo;
9633 
9634 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9635 			linfo[0].disallow_lpage = 1;
9636 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9637 			linfo[lpages - 1].disallow_lpage = 1;
9638 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9639 		/*
9640 		 * If the gfn and userspace address are not aligned wrt each
9641 		 * other, or if explicitly asked to, disable large page
9642 		 * support for this slot
9643 		 */
9644 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9645 		    !kvm_largepages_enabled()) {
9646 			unsigned long j;
9647 
9648 			for (j = 0; j < lpages; ++j)
9649 				linfo[j].disallow_lpage = 1;
9650 		}
9651 	}
9652 
9653 	if (kvm_page_track_create_memslot(slot, npages))
9654 		goto out_free;
9655 
9656 	return 0;
9657 
9658 out_free:
9659 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9660 		kvfree(slot->arch.rmap[i]);
9661 		slot->arch.rmap[i] = NULL;
9662 		if (i == 0)
9663 			continue;
9664 
9665 		kvfree(slot->arch.lpage_info[i - 1]);
9666 		slot->arch.lpage_info[i - 1] = NULL;
9667 	}
9668 	return -ENOMEM;
9669 }
9670 
9671 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9672 {
9673 	/*
9674 	 * memslots->generation has been incremented.
9675 	 * mmio generation may have reached its maximum value.
9676 	 */
9677 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9678 }
9679 
9680 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9681 				struct kvm_memory_slot *memslot,
9682 				const struct kvm_userspace_memory_region *mem,
9683 				enum kvm_mr_change change)
9684 {
9685 	return 0;
9686 }
9687 
9688 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9689 				     struct kvm_memory_slot *new)
9690 {
9691 	/* Still write protect RO slot */
9692 	if (new->flags & KVM_MEM_READONLY) {
9693 		kvm_mmu_slot_remove_write_access(kvm, new);
9694 		return;
9695 	}
9696 
9697 	/*
9698 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9699 	 *
9700 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9701 	 *
9702 	 *  - KVM_MR_CREATE with dirty logging is disabled
9703 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9704 	 *
9705 	 * The reason is, in case of PML, we need to set D-bit for any slots
9706 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9707 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9708 	 * guarantees leaving PML enabled during guest's lifetime won't have
9709 	 * any additional overhead from PML when guest is running with dirty
9710 	 * logging disabled for memory slots.
9711 	 *
9712 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9713 	 * to dirty logging mode.
9714 	 *
9715 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9716 	 *
9717 	 * In case of write protect:
9718 	 *
9719 	 * Write protect all pages for dirty logging.
9720 	 *
9721 	 * All the sptes including the large sptes which point to this
9722 	 * slot are set to readonly. We can not create any new large
9723 	 * spte on this slot until the end of the logging.
9724 	 *
9725 	 * See the comments in fast_page_fault().
9726 	 */
9727 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9728 		if (kvm_x86_ops->slot_enable_log_dirty)
9729 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9730 		else
9731 			kvm_mmu_slot_remove_write_access(kvm, new);
9732 	} else {
9733 		if (kvm_x86_ops->slot_disable_log_dirty)
9734 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9735 	}
9736 }
9737 
9738 void kvm_arch_commit_memory_region(struct kvm *kvm,
9739 				const struct kvm_userspace_memory_region *mem,
9740 				const struct kvm_memory_slot *old,
9741 				const struct kvm_memory_slot *new,
9742 				enum kvm_mr_change change)
9743 {
9744 	if (!kvm->arch.n_requested_mmu_pages)
9745 		kvm_mmu_change_mmu_pages(kvm,
9746 				kvm_mmu_calculate_default_mmu_pages(kvm));
9747 
9748 	/*
9749 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9750 	 * sptes have to be split.  If live migration is successful, the guest
9751 	 * in the source machine will be destroyed and large sptes will be
9752 	 * created in the destination. However, if the guest continues to run
9753 	 * in the source machine (for example if live migration fails), small
9754 	 * sptes will remain around and cause bad performance.
9755 	 *
9756 	 * Scan sptes if dirty logging has been stopped, dropping those
9757 	 * which can be collapsed into a single large-page spte.  Later
9758 	 * page faults will create the large-page sptes.
9759 	 *
9760 	 * There is no need to do this in any of the following cases:
9761 	 * CREATE:	No dirty mappings will already exist.
9762 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
9763 	 *		kvm_arch_flush_shadow_memslot()
9764 	 */
9765 	if (change == KVM_MR_FLAGS_ONLY &&
9766 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9767 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9768 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9769 
9770 	/*
9771 	 * Set up write protection and/or dirty logging for the new slot.
9772 	 *
9773 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9774 	 * been zapped so no dirty logging staff is needed for old slot. For
9775 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9776 	 * new and it's also covered when dealing with the new slot.
9777 	 *
9778 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9779 	 */
9780 	if (change != KVM_MR_DELETE)
9781 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9782 }
9783 
9784 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9785 {
9786 	kvm_mmu_zap_all(kvm);
9787 }
9788 
9789 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9790 				   struct kvm_memory_slot *slot)
9791 {
9792 	kvm_page_track_flush_slot(kvm, slot);
9793 }
9794 
9795 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9796 {
9797 	return (is_guest_mode(vcpu) &&
9798 			kvm_x86_ops->guest_apic_has_interrupt &&
9799 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9800 }
9801 
9802 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9803 {
9804 	if (!list_empty_careful(&vcpu->async_pf.done))
9805 		return true;
9806 
9807 	if (kvm_apic_has_events(vcpu))
9808 		return true;
9809 
9810 	if (vcpu->arch.pv.pv_unhalted)
9811 		return true;
9812 
9813 	if (vcpu->arch.exception.pending)
9814 		return true;
9815 
9816 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9817 	    (vcpu->arch.nmi_pending &&
9818 	     kvm_x86_ops->nmi_allowed(vcpu)))
9819 		return true;
9820 
9821 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9822 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9823 		return true;
9824 
9825 	if (kvm_arch_interrupt_allowed(vcpu) &&
9826 	    (kvm_cpu_has_interrupt(vcpu) ||
9827 	    kvm_guest_apic_has_interrupt(vcpu)))
9828 		return true;
9829 
9830 	if (kvm_hv_has_stimer_pending(vcpu))
9831 		return true;
9832 
9833 	return false;
9834 }
9835 
9836 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9837 {
9838 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9839 }
9840 
9841 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9842 {
9843 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9844 		return true;
9845 
9846 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9847 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
9848 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
9849 		return true;
9850 
9851 	if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9852 		return true;
9853 
9854 	return false;
9855 }
9856 
9857 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9858 {
9859 	return vcpu->arch.preempted_in_kernel;
9860 }
9861 
9862 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9863 {
9864 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9865 }
9866 
9867 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9868 {
9869 	return kvm_x86_ops->interrupt_allowed(vcpu);
9870 }
9871 
9872 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9873 {
9874 	if (is_64_bit_mode(vcpu))
9875 		return kvm_rip_read(vcpu);
9876 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9877 		     kvm_rip_read(vcpu));
9878 }
9879 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9880 
9881 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9882 {
9883 	return kvm_get_linear_rip(vcpu) == linear_rip;
9884 }
9885 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9886 
9887 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9888 {
9889 	unsigned long rflags;
9890 
9891 	rflags = kvm_x86_ops->get_rflags(vcpu);
9892 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9893 		rflags &= ~X86_EFLAGS_TF;
9894 	return rflags;
9895 }
9896 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9897 
9898 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9899 {
9900 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9901 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9902 		rflags |= X86_EFLAGS_TF;
9903 	kvm_x86_ops->set_rflags(vcpu, rflags);
9904 }
9905 
9906 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9907 {
9908 	__kvm_set_rflags(vcpu, rflags);
9909 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9910 }
9911 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9912 
9913 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9914 {
9915 	int r;
9916 
9917 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9918 	      work->wakeup_all)
9919 		return;
9920 
9921 	r = kvm_mmu_reload(vcpu);
9922 	if (unlikely(r))
9923 		return;
9924 
9925 	if (!vcpu->arch.mmu->direct_map &&
9926 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9927 		return;
9928 
9929 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9930 }
9931 
9932 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9933 {
9934 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9935 }
9936 
9937 static inline u32 kvm_async_pf_next_probe(u32 key)
9938 {
9939 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9940 }
9941 
9942 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9943 {
9944 	u32 key = kvm_async_pf_hash_fn(gfn);
9945 
9946 	while (vcpu->arch.apf.gfns[key] != ~0)
9947 		key = kvm_async_pf_next_probe(key);
9948 
9949 	vcpu->arch.apf.gfns[key] = gfn;
9950 }
9951 
9952 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9953 {
9954 	int i;
9955 	u32 key = kvm_async_pf_hash_fn(gfn);
9956 
9957 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9958 		     (vcpu->arch.apf.gfns[key] != gfn &&
9959 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9960 		key = kvm_async_pf_next_probe(key);
9961 
9962 	return key;
9963 }
9964 
9965 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9966 {
9967 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9968 }
9969 
9970 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9971 {
9972 	u32 i, j, k;
9973 
9974 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9975 	while (true) {
9976 		vcpu->arch.apf.gfns[i] = ~0;
9977 		do {
9978 			j = kvm_async_pf_next_probe(j);
9979 			if (vcpu->arch.apf.gfns[j] == ~0)
9980 				return;
9981 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9982 			/*
9983 			 * k lies cyclically in ]i,j]
9984 			 * |    i.k.j |
9985 			 * |....j i.k.| or  |.k..j i...|
9986 			 */
9987 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9988 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9989 		i = j;
9990 	}
9991 }
9992 
9993 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9994 {
9995 
9996 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9997 				      sizeof(val));
9998 }
9999 
10000 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10001 {
10002 
10003 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10004 				      sizeof(u32));
10005 }
10006 
10007 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10008 {
10009 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10010 		return false;
10011 
10012 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10013 	    (vcpu->arch.apf.send_user_only &&
10014 	     kvm_x86_ops->get_cpl(vcpu) == 0))
10015 		return false;
10016 
10017 	return true;
10018 }
10019 
10020 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10021 {
10022 	if (unlikely(!lapic_in_kernel(vcpu) ||
10023 		     kvm_event_needs_reinjection(vcpu) ||
10024 		     vcpu->arch.exception.pending))
10025 		return false;
10026 
10027 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10028 		return false;
10029 
10030 	/*
10031 	 * If interrupts are off we cannot even use an artificial
10032 	 * halt state.
10033 	 */
10034 	return kvm_x86_ops->interrupt_allowed(vcpu);
10035 }
10036 
10037 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10038 				     struct kvm_async_pf *work)
10039 {
10040 	struct x86_exception fault;
10041 
10042 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10043 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10044 
10045 	if (kvm_can_deliver_async_pf(vcpu) &&
10046 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10047 		fault.vector = PF_VECTOR;
10048 		fault.error_code_valid = true;
10049 		fault.error_code = 0;
10050 		fault.nested_page_fault = false;
10051 		fault.address = work->arch.token;
10052 		fault.async_page_fault = true;
10053 		kvm_inject_page_fault(vcpu, &fault);
10054 	} else {
10055 		/*
10056 		 * It is not possible to deliver a paravirtualized asynchronous
10057 		 * page fault, but putting the guest in an artificial halt state
10058 		 * can be beneficial nevertheless: if an interrupt arrives, we
10059 		 * can deliver it timely and perhaps the guest will schedule
10060 		 * another process.  When the instruction that triggered a page
10061 		 * fault is retried, hopefully the page will be ready in the host.
10062 		 */
10063 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10064 	}
10065 }
10066 
10067 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10068 				 struct kvm_async_pf *work)
10069 {
10070 	struct x86_exception fault;
10071 	u32 val;
10072 
10073 	if (work->wakeup_all)
10074 		work->arch.token = ~0; /* broadcast wakeup */
10075 	else
10076 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10077 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
10078 
10079 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10080 	    !apf_get_user(vcpu, &val)) {
10081 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10082 		    vcpu->arch.exception.pending &&
10083 		    vcpu->arch.exception.nr == PF_VECTOR &&
10084 		    !apf_put_user(vcpu, 0)) {
10085 			vcpu->arch.exception.injected = false;
10086 			vcpu->arch.exception.pending = false;
10087 			vcpu->arch.exception.nr = 0;
10088 			vcpu->arch.exception.has_error_code = false;
10089 			vcpu->arch.exception.error_code = 0;
10090 			vcpu->arch.exception.has_payload = false;
10091 			vcpu->arch.exception.payload = 0;
10092 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10093 			fault.vector = PF_VECTOR;
10094 			fault.error_code_valid = true;
10095 			fault.error_code = 0;
10096 			fault.nested_page_fault = false;
10097 			fault.address = work->arch.token;
10098 			fault.async_page_fault = true;
10099 			kvm_inject_page_fault(vcpu, &fault);
10100 		}
10101 	}
10102 	vcpu->arch.apf.halted = false;
10103 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10104 }
10105 
10106 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10107 {
10108 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10109 		return true;
10110 	else
10111 		return kvm_can_do_async_pf(vcpu);
10112 }
10113 
10114 void kvm_arch_start_assignment(struct kvm *kvm)
10115 {
10116 	atomic_inc(&kvm->arch.assigned_device_count);
10117 }
10118 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10119 
10120 void kvm_arch_end_assignment(struct kvm *kvm)
10121 {
10122 	atomic_dec(&kvm->arch.assigned_device_count);
10123 }
10124 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10125 
10126 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10127 {
10128 	return atomic_read(&kvm->arch.assigned_device_count);
10129 }
10130 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10131 
10132 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10133 {
10134 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10135 }
10136 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10137 
10138 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10139 {
10140 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10141 }
10142 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10143 
10144 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10145 {
10146 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10147 }
10148 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10149 
10150 bool kvm_arch_has_irq_bypass(void)
10151 {
10152 	return true;
10153 }
10154 
10155 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10156 				      struct irq_bypass_producer *prod)
10157 {
10158 	struct kvm_kernel_irqfd *irqfd =
10159 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10160 
10161 	irqfd->producer = prod;
10162 
10163 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10164 					   prod->irq, irqfd->gsi, 1);
10165 }
10166 
10167 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10168 				      struct irq_bypass_producer *prod)
10169 {
10170 	int ret;
10171 	struct kvm_kernel_irqfd *irqfd =
10172 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10173 
10174 	WARN_ON(irqfd->producer != prod);
10175 	irqfd->producer = NULL;
10176 
10177 	/*
10178 	 * When producer of consumer is unregistered, we change back to
10179 	 * remapped mode, so we can re-use the current implementation
10180 	 * when the irq is masked/disabled or the consumer side (KVM
10181 	 * int this case doesn't want to receive the interrupts.
10182 	*/
10183 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10184 	if (ret)
10185 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10186 		       " fails: %d\n", irqfd->consumer.token, ret);
10187 }
10188 
10189 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10190 				   uint32_t guest_irq, bool set)
10191 {
10192 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10193 }
10194 
10195 bool kvm_vector_hashing_enabled(void)
10196 {
10197 	return vector_hashing;
10198 }
10199 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10200 
10201 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10202 {
10203 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10204 }
10205 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10206 
10207 
10208 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10209 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10210 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10211 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10212 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10213 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10214 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10215 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10216 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10217 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10218 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10219 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10220 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10221 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10222 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10223 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10224 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10225 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10226 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10227 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10228