xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 77ab8d5d)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
58 
59 #include <trace/events/kvm.h>
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.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
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
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 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
139 unsigned int __read_mostly lapic_timer_advance_ns = 0;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141 
142 static bool __read_mostly vector_hashing = true;
143 module_param(vector_hashing, bool, S_IRUGO);
144 
145 bool __read_mostly enable_vmware_backdoor = false;
146 module_param(enable_vmware_backdoor, bool, S_IRUGO);
147 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
148 
149 static bool __read_mostly force_emulation_prefix = false;
150 module_param(force_emulation_prefix, bool, S_IRUGO);
151 
152 #define KVM_NR_SHARED_MSRS 16
153 
154 struct kvm_shared_msrs_global {
155 	int nr;
156 	u32 msrs[KVM_NR_SHARED_MSRS];
157 };
158 
159 struct kvm_shared_msrs {
160 	struct user_return_notifier urn;
161 	bool registered;
162 	struct kvm_shared_msr_values {
163 		u64 host;
164 		u64 curr;
165 	} values[KVM_NR_SHARED_MSRS];
166 };
167 
168 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
169 static struct kvm_shared_msrs __percpu *shared_msrs;
170 
171 struct kvm_stats_debugfs_item debugfs_entries[] = {
172 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
173 	{ "pf_guest", VCPU_STAT(pf_guest) },
174 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
175 	{ "invlpg", VCPU_STAT(invlpg) },
176 	{ "exits", VCPU_STAT(exits) },
177 	{ "io_exits", VCPU_STAT(io_exits) },
178 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
179 	{ "signal_exits", VCPU_STAT(signal_exits) },
180 	{ "irq_window", VCPU_STAT(irq_window_exits) },
181 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
182 	{ "halt_exits", VCPU_STAT(halt_exits) },
183 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
184 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
185 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
186 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
187 	{ "hypercalls", VCPU_STAT(hypercalls) },
188 	{ "request_irq", VCPU_STAT(request_irq_exits) },
189 	{ "irq_exits", VCPU_STAT(irq_exits) },
190 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
191 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
192 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
193 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
194 	{ "irq_injections", VCPU_STAT(irq_injections) },
195 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
196 	{ "req_event", VCPU_STAT(req_event) },
197 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
198 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
199 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
200 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
201 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
202 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
203 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
204 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
205 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
206 	{ "largepages", VM_STAT(lpages) },
207 	{ "max_mmu_page_hash_collisions",
208 		VM_STAT(max_mmu_page_hash_collisions) },
209 	{ NULL }
210 };
211 
212 u64 __read_mostly host_xcr0;
213 
214 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
215 
216 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
217 {
218 	int i;
219 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
220 		vcpu->arch.apf.gfns[i] = ~0;
221 }
222 
223 static void kvm_on_user_return(struct user_return_notifier *urn)
224 {
225 	unsigned slot;
226 	struct kvm_shared_msrs *locals
227 		= container_of(urn, struct kvm_shared_msrs, urn);
228 	struct kvm_shared_msr_values *values;
229 	unsigned long flags;
230 
231 	/*
232 	 * Disabling irqs at this point since the following code could be
233 	 * interrupted and executed through kvm_arch_hardware_disable()
234 	 */
235 	local_irq_save(flags);
236 	if (locals->registered) {
237 		locals->registered = false;
238 		user_return_notifier_unregister(urn);
239 	}
240 	local_irq_restore(flags);
241 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
242 		values = &locals->values[slot];
243 		if (values->host != values->curr) {
244 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
245 			values->curr = values->host;
246 		}
247 	}
248 }
249 
250 static void shared_msr_update(unsigned slot, u32 msr)
251 {
252 	u64 value;
253 	unsigned int cpu = smp_processor_id();
254 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
255 
256 	/* only read, and nobody should modify it at this time,
257 	 * so don't need lock */
258 	if (slot >= shared_msrs_global.nr) {
259 		printk(KERN_ERR "kvm: invalid MSR slot!");
260 		return;
261 	}
262 	rdmsrl_safe(msr, &value);
263 	smsr->values[slot].host = value;
264 	smsr->values[slot].curr = value;
265 }
266 
267 void kvm_define_shared_msr(unsigned slot, u32 msr)
268 {
269 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
270 	shared_msrs_global.msrs[slot] = msr;
271 	if (slot >= shared_msrs_global.nr)
272 		shared_msrs_global.nr = slot + 1;
273 }
274 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
275 
276 static void kvm_shared_msr_cpu_online(void)
277 {
278 	unsigned i;
279 
280 	for (i = 0; i < shared_msrs_global.nr; ++i)
281 		shared_msr_update(i, shared_msrs_global.msrs[i]);
282 }
283 
284 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
285 {
286 	unsigned int cpu = smp_processor_id();
287 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
288 	int err;
289 
290 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
291 		return 0;
292 	smsr->values[slot].curr = value;
293 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
294 	if (err)
295 		return 1;
296 
297 	if (!smsr->registered) {
298 		smsr->urn.on_user_return = kvm_on_user_return;
299 		user_return_notifier_register(&smsr->urn);
300 		smsr->registered = true;
301 	}
302 	return 0;
303 }
304 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
305 
306 static void drop_user_return_notifiers(void)
307 {
308 	unsigned int cpu = smp_processor_id();
309 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
310 
311 	if (smsr->registered)
312 		kvm_on_user_return(&smsr->urn);
313 }
314 
315 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
316 {
317 	return vcpu->arch.apic_base;
318 }
319 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
320 
321 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
322 {
323 	u64 old_state = vcpu->arch.apic_base &
324 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
325 	u64 new_state = msr_info->data &
326 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
327 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
328 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
329 
330 	if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
331 		return 1;
332 	if (!msr_info->host_initiated &&
333 	    ((new_state == MSR_IA32_APICBASE_ENABLE &&
334 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
335 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
336 	      old_state == 0)))
337 		return 1;
338 
339 	kvm_lapic_set_base(vcpu, msr_info->data);
340 	return 0;
341 }
342 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
343 
344 asmlinkage __visible void kvm_spurious_fault(void)
345 {
346 	/* Fault while not rebooting.  We want the trace. */
347 	BUG();
348 }
349 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
350 
351 #define EXCPT_BENIGN		0
352 #define EXCPT_CONTRIBUTORY	1
353 #define EXCPT_PF		2
354 
355 static int exception_class(int vector)
356 {
357 	switch (vector) {
358 	case PF_VECTOR:
359 		return EXCPT_PF;
360 	case DE_VECTOR:
361 	case TS_VECTOR:
362 	case NP_VECTOR:
363 	case SS_VECTOR:
364 	case GP_VECTOR:
365 		return EXCPT_CONTRIBUTORY;
366 	default:
367 		break;
368 	}
369 	return EXCPT_BENIGN;
370 }
371 
372 #define EXCPT_FAULT		0
373 #define EXCPT_TRAP		1
374 #define EXCPT_ABORT		2
375 #define EXCPT_INTERRUPT		3
376 
377 static int exception_type(int vector)
378 {
379 	unsigned int mask;
380 
381 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
382 		return EXCPT_INTERRUPT;
383 
384 	mask = 1 << vector;
385 
386 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
387 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
388 		return EXCPT_TRAP;
389 
390 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
391 		return EXCPT_ABORT;
392 
393 	/* Reserved exceptions will result in fault */
394 	return EXCPT_FAULT;
395 }
396 
397 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
398 		unsigned nr, bool has_error, u32 error_code,
399 		bool reinject)
400 {
401 	u32 prev_nr;
402 	int class1, class2;
403 
404 	kvm_make_request(KVM_REQ_EVENT, vcpu);
405 
406 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
407 	queue:
408 		if (has_error && !is_protmode(vcpu))
409 			has_error = false;
410 		if (reinject) {
411 			/*
412 			 * On vmentry, vcpu->arch.exception.pending is only
413 			 * true if an event injection was blocked by
414 			 * nested_run_pending.  In that case, however,
415 			 * vcpu_enter_guest requests an immediate exit,
416 			 * and the guest shouldn't proceed far enough to
417 			 * need reinjection.
418 			 */
419 			WARN_ON_ONCE(vcpu->arch.exception.pending);
420 			vcpu->arch.exception.injected = true;
421 		} else {
422 			vcpu->arch.exception.pending = true;
423 			vcpu->arch.exception.injected = false;
424 		}
425 		vcpu->arch.exception.has_error_code = has_error;
426 		vcpu->arch.exception.nr = nr;
427 		vcpu->arch.exception.error_code = error_code;
428 		return;
429 	}
430 
431 	/* to check exception */
432 	prev_nr = vcpu->arch.exception.nr;
433 	if (prev_nr == DF_VECTOR) {
434 		/* triple fault -> shutdown */
435 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
436 		return;
437 	}
438 	class1 = exception_class(prev_nr);
439 	class2 = exception_class(nr);
440 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
441 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
442 		/*
443 		 * Generate double fault per SDM Table 5-5.  Set
444 		 * exception.pending = true so that the double fault
445 		 * can trigger a nested vmexit.
446 		 */
447 		vcpu->arch.exception.pending = true;
448 		vcpu->arch.exception.injected = false;
449 		vcpu->arch.exception.has_error_code = true;
450 		vcpu->arch.exception.nr = DF_VECTOR;
451 		vcpu->arch.exception.error_code = 0;
452 	} else
453 		/* replace previous exception with a new one in a hope
454 		   that instruction re-execution will regenerate lost
455 		   exception */
456 		goto queue;
457 }
458 
459 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
460 {
461 	kvm_multiple_exception(vcpu, nr, false, 0, false);
462 }
463 EXPORT_SYMBOL_GPL(kvm_queue_exception);
464 
465 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
466 {
467 	kvm_multiple_exception(vcpu, nr, false, 0, true);
468 }
469 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
470 
471 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
472 {
473 	if (err)
474 		kvm_inject_gp(vcpu, 0);
475 	else
476 		return kvm_skip_emulated_instruction(vcpu);
477 
478 	return 1;
479 }
480 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
481 
482 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
483 {
484 	++vcpu->stat.pf_guest;
485 	vcpu->arch.exception.nested_apf =
486 		is_guest_mode(vcpu) && fault->async_page_fault;
487 	if (vcpu->arch.exception.nested_apf)
488 		vcpu->arch.apf.nested_apf_token = fault->address;
489 	else
490 		vcpu->arch.cr2 = fault->address;
491 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
492 }
493 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
494 
495 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
496 {
497 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
498 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
499 	else
500 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
501 
502 	return fault->nested_page_fault;
503 }
504 
505 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
506 {
507 	atomic_inc(&vcpu->arch.nmi_queued);
508 	kvm_make_request(KVM_REQ_NMI, vcpu);
509 }
510 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
511 
512 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
513 {
514 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
515 }
516 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
517 
518 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
519 {
520 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
521 }
522 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
523 
524 /*
525  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
526  * a #GP and return false.
527  */
528 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
529 {
530 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
531 		return true;
532 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
533 	return false;
534 }
535 EXPORT_SYMBOL_GPL(kvm_require_cpl);
536 
537 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
538 {
539 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
540 		return true;
541 
542 	kvm_queue_exception(vcpu, UD_VECTOR);
543 	return false;
544 }
545 EXPORT_SYMBOL_GPL(kvm_require_dr);
546 
547 /*
548  * This function will be used to read from the physical memory of the currently
549  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
550  * can read from guest physical or from the guest's guest physical memory.
551  */
552 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
553 			    gfn_t ngfn, void *data, int offset, int len,
554 			    u32 access)
555 {
556 	struct x86_exception exception;
557 	gfn_t real_gfn;
558 	gpa_t ngpa;
559 
560 	ngpa     = gfn_to_gpa(ngfn);
561 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
562 	if (real_gfn == UNMAPPED_GVA)
563 		return -EFAULT;
564 
565 	real_gfn = gpa_to_gfn(real_gfn);
566 
567 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
568 }
569 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
570 
571 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
572 			       void *data, int offset, int len, u32 access)
573 {
574 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
575 				       data, offset, len, access);
576 }
577 
578 /*
579  * Load the pae pdptrs.  Return true is they are all valid.
580  */
581 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
582 {
583 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
584 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
585 	int i;
586 	int ret;
587 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
588 
589 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
590 				      offset * sizeof(u64), sizeof(pdpte),
591 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
592 	if (ret < 0) {
593 		ret = 0;
594 		goto out;
595 	}
596 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
597 		if ((pdpte[i] & PT_PRESENT_MASK) &&
598 		    (pdpte[i] &
599 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
600 			ret = 0;
601 			goto out;
602 		}
603 	}
604 	ret = 1;
605 
606 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
607 	__set_bit(VCPU_EXREG_PDPTR,
608 		  (unsigned long *)&vcpu->arch.regs_avail);
609 	__set_bit(VCPU_EXREG_PDPTR,
610 		  (unsigned long *)&vcpu->arch.regs_dirty);
611 out:
612 
613 	return ret;
614 }
615 EXPORT_SYMBOL_GPL(load_pdptrs);
616 
617 bool pdptrs_changed(struct kvm_vcpu *vcpu)
618 {
619 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
620 	bool changed = true;
621 	int offset;
622 	gfn_t gfn;
623 	int r;
624 
625 	if (is_long_mode(vcpu) || !is_pae(vcpu))
626 		return false;
627 
628 	if (!test_bit(VCPU_EXREG_PDPTR,
629 		      (unsigned long *)&vcpu->arch.regs_avail))
630 		return true;
631 
632 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
633 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
634 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
635 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
636 	if (r < 0)
637 		goto out;
638 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
639 out:
640 
641 	return changed;
642 }
643 EXPORT_SYMBOL_GPL(pdptrs_changed);
644 
645 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
646 {
647 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
648 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
649 
650 	cr0 |= X86_CR0_ET;
651 
652 #ifdef CONFIG_X86_64
653 	if (cr0 & 0xffffffff00000000UL)
654 		return 1;
655 #endif
656 
657 	cr0 &= ~CR0_RESERVED_BITS;
658 
659 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
660 		return 1;
661 
662 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
663 		return 1;
664 
665 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
666 #ifdef CONFIG_X86_64
667 		if ((vcpu->arch.efer & EFER_LME)) {
668 			int cs_db, cs_l;
669 
670 			if (!is_pae(vcpu))
671 				return 1;
672 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
673 			if (cs_l)
674 				return 1;
675 		} else
676 #endif
677 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
678 						 kvm_read_cr3(vcpu)))
679 			return 1;
680 	}
681 
682 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
683 		return 1;
684 
685 	kvm_x86_ops->set_cr0(vcpu, cr0);
686 
687 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
688 		kvm_clear_async_pf_completion_queue(vcpu);
689 		kvm_async_pf_hash_reset(vcpu);
690 	}
691 
692 	if ((cr0 ^ old_cr0) & update_bits)
693 		kvm_mmu_reset_context(vcpu);
694 
695 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
696 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
697 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
698 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
699 
700 	return 0;
701 }
702 EXPORT_SYMBOL_GPL(kvm_set_cr0);
703 
704 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
705 {
706 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
707 }
708 EXPORT_SYMBOL_GPL(kvm_lmsw);
709 
710 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
711 {
712 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
713 			!vcpu->guest_xcr0_loaded) {
714 		/* kvm_set_xcr() also depends on this */
715 		if (vcpu->arch.xcr0 != host_xcr0)
716 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
717 		vcpu->guest_xcr0_loaded = 1;
718 	}
719 }
720 
721 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
722 {
723 	if (vcpu->guest_xcr0_loaded) {
724 		if (vcpu->arch.xcr0 != host_xcr0)
725 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
726 		vcpu->guest_xcr0_loaded = 0;
727 	}
728 }
729 
730 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
731 {
732 	u64 xcr0 = xcr;
733 	u64 old_xcr0 = vcpu->arch.xcr0;
734 	u64 valid_bits;
735 
736 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
737 	if (index != XCR_XFEATURE_ENABLED_MASK)
738 		return 1;
739 	if (!(xcr0 & XFEATURE_MASK_FP))
740 		return 1;
741 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
742 		return 1;
743 
744 	/*
745 	 * Do not allow the guest to set bits that we do not support
746 	 * saving.  However, xcr0 bit 0 is always set, even if the
747 	 * emulated CPU does not support XSAVE (see fx_init).
748 	 */
749 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
750 	if (xcr0 & ~valid_bits)
751 		return 1;
752 
753 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
754 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
755 		return 1;
756 
757 	if (xcr0 & XFEATURE_MASK_AVX512) {
758 		if (!(xcr0 & XFEATURE_MASK_YMM))
759 			return 1;
760 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
761 			return 1;
762 	}
763 	vcpu->arch.xcr0 = xcr0;
764 
765 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
766 		kvm_update_cpuid(vcpu);
767 	return 0;
768 }
769 
770 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
771 {
772 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
773 	    __kvm_set_xcr(vcpu, index, xcr)) {
774 		kvm_inject_gp(vcpu, 0);
775 		return 1;
776 	}
777 	return 0;
778 }
779 EXPORT_SYMBOL_GPL(kvm_set_xcr);
780 
781 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
782 {
783 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
784 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
785 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
786 
787 	if (cr4 & CR4_RESERVED_BITS)
788 		return 1;
789 
790 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
791 		return 1;
792 
793 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
794 		return 1;
795 
796 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
797 		return 1;
798 
799 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
800 		return 1;
801 
802 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
803 		return 1;
804 
805 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
806 		return 1;
807 
808 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
809 		return 1;
810 
811 	if (is_long_mode(vcpu)) {
812 		if (!(cr4 & X86_CR4_PAE))
813 			return 1;
814 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
815 		   && ((cr4 ^ old_cr4) & pdptr_bits)
816 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
817 				   kvm_read_cr3(vcpu)))
818 		return 1;
819 
820 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
821 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
822 			return 1;
823 
824 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
825 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
826 			return 1;
827 	}
828 
829 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
830 		return 1;
831 
832 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
833 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
834 		kvm_mmu_reset_context(vcpu);
835 
836 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
837 		kvm_update_cpuid(vcpu);
838 
839 	return 0;
840 }
841 EXPORT_SYMBOL_GPL(kvm_set_cr4);
842 
843 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
844 {
845 #ifdef CONFIG_X86_64
846 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
847 
848 	if (pcid_enabled)
849 		cr3 &= ~CR3_PCID_INVD;
850 #endif
851 
852 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
853 		kvm_mmu_sync_roots(vcpu);
854 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
855 		return 0;
856 	}
857 
858 	if (is_long_mode(vcpu) &&
859 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
860 		return 1;
861 	else if (is_pae(vcpu) && is_paging(vcpu) &&
862 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
863 		return 1;
864 
865 	vcpu->arch.cr3 = cr3;
866 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
867 	kvm_mmu_new_cr3(vcpu);
868 	return 0;
869 }
870 EXPORT_SYMBOL_GPL(kvm_set_cr3);
871 
872 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
873 {
874 	if (cr8 & CR8_RESERVED_BITS)
875 		return 1;
876 	if (lapic_in_kernel(vcpu))
877 		kvm_lapic_set_tpr(vcpu, cr8);
878 	else
879 		vcpu->arch.cr8 = cr8;
880 	return 0;
881 }
882 EXPORT_SYMBOL_GPL(kvm_set_cr8);
883 
884 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
885 {
886 	if (lapic_in_kernel(vcpu))
887 		return kvm_lapic_get_cr8(vcpu);
888 	else
889 		return vcpu->arch.cr8;
890 }
891 EXPORT_SYMBOL_GPL(kvm_get_cr8);
892 
893 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
894 {
895 	int i;
896 
897 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
898 		for (i = 0; i < KVM_NR_DB_REGS; i++)
899 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
900 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
901 	}
902 }
903 
904 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
905 {
906 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
907 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
908 }
909 
910 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
911 {
912 	unsigned long dr7;
913 
914 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
915 		dr7 = vcpu->arch.guest_debug_dr7;
916 	else
917 		dr7 = vcpu->arch.dr7;
918 	kvm_x86_ops->set_dr7(vcpu, dr7);
919 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
920 	if (dr7 & DR7_BP_EN_MASK)
921 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
922 }
923 
924 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
925 {
926 	u64 fixed = DR6_FIXED_1;
927 
928 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
929 		fixed |= DR6_RTM;
930 	return fixed;
931 }
932 
933 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
934 {
935 	switch (dr) {
936 	case 0 ... 3:
937 		vcpu->arch.db[dr] = val;
938 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
939 			vcpu->arch.eff_db[dr] = val;
940 		break;
941 	case 4:
942 		/* fall through */
943 	case 6:
944 		if (val & 0xffffffff00000000ULL)
945 			return -1; /* #GP */
946 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
947 		kvm_update_dr6(vcpu);
948 		break;
949 	case 5:
950 		/* fall through */
951 	default: /* 7 */
952 		if (val & 0xffffffff00000000ULL)
953 			return -1; /* #GP */
954 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
955 		kvm_update_dr7(vcpu);
956 		break;
957 	}
958 
959 	return 0;
960 }
961 
962 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
963 {
964 	if (__kvm_set_dr(vcpu, dr, val)) {
965 		kvm_inject_gp(vcpu, 0);
966 		return 1;
967 	}
968 	return 0;
969 }
970 EXPORT_SYMBOL_GPL(kvm_set_dr);
971 
972 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
973 {
974 	switch (dr) {
975 	case 0 ... 3:
976 		*val = vcpu->arch.db[dr];
977 		break;
978 	case 4:
979 		/* fall through */
980 	case 6:
981 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
982 			*val = vcpu->arch.dr6;
983 		else
984 			*val = kvm_x86_ops->get_dr6(vcpu);
985 		break;
986 	case 5:
987 		/* fall through */
988 	default: /* 7 */
989 		*val = vcpu->arch.dr7;
990 		break;
991 	}
992 	return 0;
993 }
994 EXPORT_SYMBOL_GPL(kvm_get_dr);
995 
996 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
997 {
998 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
999 	u64 data;
1000 	int err;
1001 
1002 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1003 	if (err)
1004 		return err;
1005 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1006 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1007 	return err;
1008 }
1009 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1010 
1011 /*
1012  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1013  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1014  *
1015  * This list is modified at module load time to reflect the
1016  * capabilities of the host cpu. This capabilities test skips MSRs that are
1017  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1018  * may depend on host virtualization features rather than host cpu features.
1019  */
1020 
1021 static u32 msrs_to_save[] = {
1022 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1023 	MSR_STAR,
1024 #ifdef CONFIG_X86_64
1025 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1026 #endif
1027 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1028 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1029 	MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1030 };
1031 
1032 static unsigned num_msrs_to_save;
1033 
1034 static u32 emulated_msrs[] = {
1035 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1036 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1037 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1038 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1039 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1040 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1041 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1042 	HV_X64_MSR_RESET,
1043 	HV_X64_MSR_VP_INDEX,
1044 	HV_X64_MSR_VP_RUNTIME,
1045 	HV_X64_MSR_SCONTROL,
1046 	HV_X64_MSR_STIMER0_CONFIG,
1047 	HV_X64_MSR_VP_ASSIST_PAGE,
1048 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1049 	HV_X64_MSR_TSC_EMULATION_STATUS,
1050 
1051 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1052 	MSR_KVM_PV_EOI_EN,
1053 
1054 	MSR_IA32_TSC_ADJUST,
1055 	MSR_IA32_TSCDEADLINE,
1056 	MSR_IA32_MISC_ENABLE,
1057 	MSR_IA32_MCG_STATUS,
1058 	MSR_IA32_MCG_CTL,
1059 	MSR_IA32_MCG_EXT_CTL,
1060 	MSR_IA32_SMBASE,
1061 	MSR_SMI_COUNT,
1062 	MSR_PLATFORM_INFO,
1063 	MSR_MISC_FEATURES_ENABLES,
1064 	MSR_AMD64_VIRT_SPEC_CTRL,
1065 };
1066 
1067 static unsigned num_emulated_msrs;
1068 
1069 /*
1070  * List of msr numbers which are used to expose MSR-based features that
1071  * can be used by a hypervisor to validate requested CPU features.
1072  */
1073 static u32 msr_based_features[] = {
1074 	MSR_IA32_VMX_BASIC,
1075 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1076 	MSR_IA32_VMX_PINBASED_CTLS,
1077 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1078 	MSR_IA32_VMX_PROCBASED_CTLS,
1079 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1080 	MSR_IA32_VMX_EXIT_CTLS,
1081 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1082 	MSR_IA32_VMX_ENTRY_CTLS,
1083 	MSR_IA32_VMX_MISC,
1084 	MSR_IA32_VMX_CR0_FIXED0,
1085 	MSR_IA32_VMX_CR0_FIXED1,
1086 	MSR_IA32_VMX_CR4_FIXED0,
1087 	MSR_IA32_VMX_CR4_FIXED1,
1088 	MSR_IA32_VMX_VMCS_ENUM,
1089 	MSR_IA32_VMX_PROCBASED_CTLS2,
1090 	MSR_IA32_VMX_EPT_VPID_CAP,
1091 	MSR_IA32_VMX_VMFUNC,
1092 
1093 	MSR_F10H_DECFG,
1094 	MSR_IA32_UCODE_REV,
1095 };
1096 
1097 static unsigned int num_msr_based_features;
1098 
1099 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1100 {
1101 	switch (msr->index) {
1102 	case MSR_IA32_UCODE_REV:
1103 		rdmsrl(msr->index, msr->data);
1104 		break;
1105 	default:
1106 		if (kvm_x86_ops->get_msr_feature(msr))
1107 			return 1;
1108 	}
1109 	return 0;
1110 }
1111 
1112 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1113 {
1114 	struct kvm_msr_entry msr;
1115 	int r;
1116 
1117 	msr.index = index;
1118 	r = kvm_get_msr_feature(&msr);
1119 	if (r)
1120 		return r;
1121 
1122 	*data = msr.data;
1123 
1124 	return 0;
1125 }
1126 
1127 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1128 {
1129 	if (efer & efer_reserved_bits)
1130 		return false;
1131 
1132 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1133 			return false;
1134 
1135 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1136 			return false;
1137 
1138 	return true;
1139 }
1140 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1141 
1142 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1143 {
1144 	u64 old_efer = vcpu->arch.efer;
1145 
1146 	if (!kvm_valid_efer(vcpu, efer))
1147 		return 1;
1148 
1149 	if (is_paging(vcpu)
1150 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1151 		return 1;
1152 
1153 	efer &= ~EFER_LMA;
1154 	efer |= vcpu->arch.efer & EFER_LMA;
1155 
1156 	kvm_x86_ops->set_efer(vcpu, efer);
1157 
1158 	/* Update reserved bits */
1159 	if ((efer ^ old_efer) & EFER_NX)
1160 		kvm_mmu_reset_context(vcpu);
1161 
1162 	return 0;
1163 }
1164 
1165 void kvm_enable_efer_bits(u64 mask)
1166 {
1167        efer_reserved_bits &= ~mask;
1168 }
1169 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1170 
1171 /*
1172  * Writes msr value into into the appropriate "register".
1173  * Returns 0 on success, non-0 otherwise.
1174  * Assumes vcpu_load() was already called.
1175  */
1176 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1177 {
1178 	switch (msr->index) {
1179 	case MSR_FS_BASE:
1180 	case MSR_GS_BASE:
1181 	case MSR_KERNEL_GS_BASE:
1182 	case MSR_CSTAR:
1183 	case MSR_LSTAR:
1184 		if (is_noncanonical_address(msr->data, vcpu))
1185 			return 1;
1186 		break;
1187 	case MSR_IA32_SYSENTER_EIP:
1188 	case MSR_IA32_SYSENTER_ESP:
1189 		/*
1190 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1191 		 * non-canonical address is written on Intel but not on
1192 		 * AMD (which ignores the top 32-bits, because it does
1193 		 * not implement 64-bit SYSENTER).
1194 		 *
1195 		 * 64-bit code should hence be able to write a non-canonical
1196 		 * value on AMD.  Making the address canonical ensures that
1197 		 * vmentry does not fail on Intel after writing a non-canonical
1198 		 * value, and that something deterministic happens if the guest
1199 		 * invokes 64-bit SYSENTER.
1200 		 */
1201 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1202 	}
1203 	return kvm_x86_ops->set_msr(vcpu, msr);
1204 }
1205 EXPORT_SYMBOL_GPL(kvm_set_msr);
1206 
1207 /*
1208  * Adapt set_msr() to msr_io()'s calling convention
1209  */
1210 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1211 {
1212 	struct msr_data msr;
1213 	int r;
1214 
1215 	msr.index = index;
1216 	msr.host_initiated = true;
1217 	r = kvm_get_msr(vcpu, &msr);
1218 	if (r)
1219 		return r;
1220 
1221 	*data = msr.data;
1222 	return 0;
1223 }
1224 
1225 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1226 {
1227 	struct msr_data msr;
1228 
1229 	msr.data = *data;
1230 	msr.index = index;
1231 	msr.host_initiated = true;
1232 	return kvm_set_msr(vcpu, &msr);
1233 }
1234 
1235 #ifdef CONFIG_X86_64
1236 struct pvclock_gtod_data {
1237 	seqcount_t	seq;
1238 
1239 	struct { /* extract of a clocksource struct */
1240 		int vclock_mode;
1241 		u64	cycle_last;
1242 		u64	mask;
1243 		u32	mult;
1244 		u32	shift;
1245 	} clock;
1246 
1247 	u64		boot_ns;
1248 	u64		nsec_base;
1249 	u64		wall_time_sec;
1250 };
1251 
1252 static struct pvclock_gtod_data pvclock_gtod_data;
1253 
1254 static void update_pvclock_gtod(struct timekeeper *tk)
1255 {
1256 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1257 	u64 boot_ns;
1258 
1259 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1260 
1261 	write_seqcount_begin(&vdata->seq);
1262 
1263 	/* copy pvclock gtod data */
1264 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1265 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1266 	vdata->clock.mask		= tk->tkr_mono.mask;
1267 	vdata->clock.mult		= tk->tkr_mono.mult;
1268 	vdata->clock.shift		= tk->tkr_mono.shift;
1269 
1270 	vdata->boot_ns			= boot_ns;
1271 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1272 
1273 	vdata->wall_time_sec            = tk->xtime_sec;
1274 
1275 	write_seqcount_end(&vdata->seq);
1276 }
1277 #endif
1278 
1279 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1280 {
1281 	/*
1282 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1283 	 * vcpu_enter_guest.  This function is only called from
1284 	 * the physical CPU that is running vcpu.
1285 	 */
1286 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1287 }
1288 
1289 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1290 {
1291 	int version;
1292 	int r;
1293 	struct pvclock_wall_clock wc;
1294 	struct timespec64 boot;
1295 
1296 	if (!wall_clock)
1297 		return;
1298 
1299 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1300 	if (r)
1301 		return;
1302 
1303 	if (version & 1)
1304 		++version;  /* first time write, random junk */
1305 
1306 	++version;
1307 
1308 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1309 		return;
1310 
1311 	/*
1312 	 * The guest calculates current wall clock time by adding
1313 	 * system time (updated by kvm_guest_time_update below) to the
1314 	 * wall clock specified here.  guest system time equals host
1315 	 * system time for us, thus we must fill in host boot time here.
1316 	 */
1317 	getboottime64(&boot);
1318 
1319 	if (kvm->arch.kvmclock_offset) {
1320 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1321 		boot = timespec64_sub(boot, ts);
1322 	}
1323 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1324 	wc.nsec = boot.tv_nsec;
1325 	wc.version = version;
1326 
1327 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1328 
1329 	version++;
1330 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1331 }
1332 
1333 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1334 {
1335 	do_shl32_div32(dividend, divisor);
1336 	return dividend;
1337 }
1338 
1339 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1340 			       s8 *pshift, u32 *pmultiplier)
1341 {
1342 	uint64_t scaled64;
1343 	int32_t  shift = 0;
1344 	uint64_t tps64;
1345 	uint32_t tps32;
1346 
1347 	tps64 = base_hz;
1348 	scaled64 = scaled_hz;
1349 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1350 		tps64 >>= 1;
1351 		shift--;
1352 	}
1353 
1354 	tps32 = (uint32_t)tps64;
1355 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1356 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1357 			scaled64 >>= 1;
1358 		else
1359 			tps32 <<= 1;
1360 		shift++;
1361 	}
1362 
1363 	*pshift = shift;
1364 	*pmultiplier = div_frac(scaled64, tps32);
1365 
1366 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1367 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1368 }
1369 
1370 #ifdef CONFIG_X86_64
1371 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1372 #endif
1373 
1374 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1375 static unsigned long max_tsc_khz;
1376 
1377 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1378 {
1379 	u64 v = (u64)khz * (1000000 + ppm);
1380 	do_div(v, 1000000);
1381 	return v;
1382 }
1383 
1384 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1385 {
1386 	u64 ratio;
1387 
1388 	/* Guest TSC same frequency as host TSC? */
1389 	if (!scale) {
1390 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1391 		return 0;
1392 	}
1393 
1394 	/* TSC scaling supported? */
1395 	if (!kvm_has_tsc_control) {
1396 		if (user_tsc_khz > tsc_khz) {
1397 			vcpu->arch.tsc_catchup = 1;
1398 			vcpu->arch.tsc_always_catchup = 1;
1399 			return 0;
1400 		} else {
1401 			WARN(1, "user requested TSC rate below hardware speed\n");
1402 			return -1;
1403 		}
1404 	}
1405 
1406 	/* TSC scaling required  - calculate ratio */
1407 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1408 				user_tsc_khz, tsc_khz);
1409 
1410 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1411 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1412 			  user_tsc_khz);
1413 		return -1;
1414 	}
1415 
1416 	vcpu->arch.tsc_scaling_ratio = ratio;
1417 	return 0;
1418 }
1419 
1420 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1421 {
1422 	u32 thresh_lo, thresh_hi;
1423 	int use_scaling = 0;
1424 
1425 	/* tsc_khz can be zero if TSC calibration fails */
1426 	if (user_tsc_khz == 0) {
1427 		/* set tsc_scaling_ratio to a safe value */
1428 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1429 		return -1;
1430 	}
1431 
1432 	/* Compute a scale to convert nanoseconds in TSC cycles */
1433 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1434 			   &vcpu->arch.virtual_tsc_shift,
1435 			   &vcpu->arch.virtual_tsc_mult);
1436 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1437 
1438 	/*
1439 	 * Compute the variation in TSC rate which is acceptable
1440 	 * within the range of tolerance and decide if the
1441 	 * rate being applied is within that bounds of the hardware
1442 	 * rate.  If so, no scaling or compensation need be done.
1443 	 */
1444 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1445 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1446 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1447 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1448 		use_scaling = 1;
1449 	}
1450 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1451 }
1452 
1453 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1454 {
1455 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1456 				      vcpu->arch.virtual_tsc_mult,
1457 				      vcpu->arch.virtual_tsc_shift);
1458 	tsc += vcpu->arch.this_tsc_write;
1459 	return tsc;
1460 }
1461 
1462 static inline int gtod_is_based_on_tsc(int mode)
1463 {
1464 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1465 }
1466 
1467 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1468 {
1469 #ifdef CONFIG_X86_64
1470 	bool vcpus_matched;
1471 	struct kvm_arch *ka = &vcpu->kvm->arch;
1472 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1473 
1474 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1475 			 atomic_read(&vcpu->kvm->online_vcpus));
1476 
1477 	/*
1478 	 * Once the masterclock is enabled, always perform request in
1479 	 * order to update it.
1480 	 *
1481 	 * In order to enable masterclock, the host clocksource must be TSC
1482 	 * and the vcpus need to have matched TSCs.  When that happens,
1483 	 * perform request to enable masterclock.
1484 	 */
1485 	if (ka->use_master_clock ||
1486 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1487 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1488 
1489 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1490 			    atomic_read(&vcpu->kvm->online_vcpus),
1491 		            ka->use_master_clock, gtod->clock.vclock_mode);
1492 #endif
1493 }
1494 
1495 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1496 {
1497 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1498 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1499 }
1500 
1501 /*
1502  * Multiply tsc by a fixed point number represented by ratio.
1503  *
1504  * The most significant 64-N bits (mult) of ratio represent the
1505  * integral part of the fixed point number; the remaining N bits
1506  * (frac) represent the fractional part, ie. ratio represents a fixed
1507  * point number (mult + frac * 2^(-N)).
1508  *
1509  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1510  */
1511 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1512 {
1513 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1514 }
1515 
1516 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1517 {
1518 	u64 _tsc = tsc;
1519 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1520 
1521 	if (ratio != kvm_default_tsc_scaling_ratio)
1522 		_tsc = __scale_tsc(ratio, tsc);
1523 
1524 	return _tsc;
1525 }
1526 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1527 
1528 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1529 {
1530 	u64 tsc;
1531 
1532 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1533 
1534 	return target_tsc - tsc;
1535 }
1536 
1537 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1538 {
1539 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1540 
1541 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1542 }
1543 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1544 
1545 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1546 {
1547 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1548 	vcpu->arch.tsc_offset = offset;
1549 }
1550 
1551 static inline bool kvm_check_tsc_unstable(void)
1552 {
1553 #ifdef CONFIG_X86_64
1554 	/*
1555 	 * TSC is marked unstable when we're running on Hyper-V,
1556 	 * 'TSC page' clocksource is good.
1557 	 */
1558 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1559 		return false;
1560 #endif
1561 	return check_tsc_unstable();
1562 }
1563 
1564 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1565 {
1566 	struct kvm *kvm = vcpu->kvm;
1567 	u64 offset, ns, elapsed;
1568 	unsigned long flags;
1569 	bool matched;
1570 	bool already_matched;
1571 	u64 data = msr->data;
1572 	bool synchronizing = false;
1573 
1574 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1575 	offset = kvm_compute_tsc_offset(vcpu, data);
1576 	ns = ktime_get_boot_ns();
1577 	elapsed = ns - kvm->arch.last_tsc_nsec;
1578 
1579 	if (vcpu->arch.virtual_tsc_khz) {
1580 		if (data == 0 && msr->host_initiated) {
1581 			/*
1582 			 * detection of vcpu initialization -- need to sync
1583 			 * with other vCPUs. This particularly helps to keep
1584 			 * kvm_clock stable after CPU hotplug
1585 			 */
1586 			synchronizing = true;
1587 		} else {
1588 			u64 tsc_exp = kvm->arch.last_tsc_write +
1589 						nsec_to_cycles(vcpu, elapsed);
1590 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1591 			/*
1592 			 * Special case: TSC write with a small delta (1 second)
1593 			 * of virtual cycle time against real time is
1594 			 * interpreted as an attempt to synchronize the CPU.
1595 			 */
1596 			synchronizing = data < tsc_exp + tsc_hz &&
1597 					data + tsc_hz > tsc_exp;
1598 		}
1599 	}
1600 
1601 	/*
1602 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1603 	 * TSC, we add elapsed time in this computation.  We could let the
1604 	 * compensation code attempt to catch up if we fall behind, but
1605 	 * it's better to try to match offsets from the beginning.
1606          */
1607 	if (synchronizing &&
1608 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1609 		if (!kvm_check_tsc_unstable()) {
1610 			offset = kvm->arch.cur_tsc_offset;
1611 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1612 		} else {
1613 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1614 			data += delta;
1615 			offset = kvm_compute_tsc_offset(vcpu, data);
1616 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1617 		}
1618 		matched = true;
1619 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1620 	} else {
1621 		/*
1622 		 * We split periods of matched TSC writes into generations.
1623 		 * For each generation, we track the original measured
1624 		 * nanosecond time, offset, and write, so if TSCs are in
1625 		 * sync, we can match exact offset, and if not, we can match
1626 		 * exact software computation in compute_guest_tsc()
1627 		 *
1628 		 * These values are tracked in kvm->arch.cur_xxx variables.
1629 		 */
1630 		kvm->arch.cur_tsc_generation++;
1631 		kvm->arch.cur_tsc_nsec = ns;
1632 		kvm->arch.cur_tsc_write = data;
1633 		kvm->arch.cur_tsc_offset = offset;
1634 		matched = false;
1635 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1636 			 kvm->arch.cur_tsc_generation, data);
1637 	}
1638 
1639 	/*
1640 	 * We also track th most recent recorded KHZ, write and time to
1641 	 * allow the matching interval to be extended at each write.
1642 	 */
1643 	kvm->arch.last_tsc_nsec = ns;
1644 	kvm->arch.last_tsc_write = data;
1645 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1646 
1647 	vcpu->arch.last_guest_tsc = data;
1648 
1649 	/* Keep track of which generation this VCPU has synchronized to */
1650 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1651 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1652 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1653 
1654 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1655 		update_ia32_tsc_adjust_msr(vcpu, offset);
1656 
1657 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1658 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1659 
1660 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1661 	if (!matched) {
1662 		kvm->arch.nr_vcpus_matched_tsc = 0;
1663 	} else if (!already_matched) {
1664 		kvm->arch.nr_vcpus_matched_tsc++;
1665 	}
1666 
1667 	kvm_track_tsc_matching(vcpu);
1668 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1669 }
1670 
1671 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1672 
1673 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1674 					   s64 adjustment)
1675 {
1676 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1677 }
1678 
1679 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1680 {
1681 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1682 		WARN_ON(adjustment < 0);
1683 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1684 	adjust_tsc_offset_guest(vcpu, adjustment);
1685 }
1686 
1687 #ifdef CONFIG_X86_64
1688 
1689 static u64 read_tsc(void)
1690 {
1691 	u64 ret = (u64)rdtsc_ordered();
1692 	u64 last = pvclock_gtod_data.clock.cycle_last;
1693 
1694 	if (likely(ret >= last))
1695 		return ret;
1696 
1697 	/*
1698 	 * GCC likes to generate cmov here, but this branch is extremely
1699 	 * predictable (it's just a function of time and the likely is
1700 	 * very likely) and there's a data dependence, so force GCC
1701 	 * to generate a branch instead.  I don't barrier() because
1702 	 * we don't actually need a barrier, and if this function
1703 	 * ever gets inlined it will generate worse code.
1704 	 */
1705 	asm volatile ("");
1706 	return last;
1707 }
1708 
1709 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1710 {
1711 	long v;
1712 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1713 	u64 tsc_pg_val;
1714 
1715 	switch (gtod->clock.vclock_mode) {
1716 	case VCLOCK_HVCLOCK:
1717 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1718 						  tsc_timestamp);
1719 		if (tsc_pg_val != U64_MAX) {
1720 			/* TSC page valid */
1721 			*mode = VCLOCK_HVCLOCK;
1722 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1723 				gtod->clock.mask;
1724 		} else {
1725 			/* TSC page invalid */
1726 			*mode = VCLOCK_NONE;
1727 		}
1728 		break;
1729 	case VCLOCK_TSC:
1730 		*mode = VCLOCK_TSC;
1731 		*tsc_timestamp = read_tsc();
1732 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1733 			gtod->clock.mask;
1734 		break;
1735 	default:
1736 		*mode = VCLOCK_NONE;
1737 	}
1738 
1739 	if (*mode == VCLOCK_NONE)
1740 		*tsc_timestamp = v = 0;
1741 
1742 	return v * gtod->clock.mult;
1743 }
1744 
1745 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1746 {
1747 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1748 	unsigned long seq;
1749 	int mode;
1750 	u64 ns;
1751 
1752 	do {
1753 		seq = read_seqcount_begin(&gtod->seq);
1754 		ns = gtod->nsec_base;
1755 		ns += vgettsc(tsc_timestamp, &mode);
1756 		ns >>= gtod->clock.shift;
1757 		ns += gtod->boot_ns;
1758 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1759 	*t = ns;
1760 
1761 	return mode;
1762 }
1763 
1764 static int do_realtime(struct timespec *ts, u64 *tsc_timestamp)
1765 {
1766 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1767 	unsigned long seq;
1768 	int mode;
1769 	u64 ns;
1770 
1771 	do {
1772 		seq = read_seqcount_begin(&gtod->seq);
1773 		ts->tv_sec = gtod->wall_time_sec;
1774 		ns = gtod->nsec_base;
1775 		ns += vgettsc(tsc_timestamp, &mode);
1776 		ns >>= gtod->clock.shift;
1777 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1778 
1779 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1780 	ts->tv_nsec = ns;
1781 
1782 	return mode;
1783 }
1784 
1785 /* returns true if host is using TSC based clocksource */
1786 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1787 {
1788 	/* checked again under seqlock below */
1789 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1790 		return false;
1791 
1792 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1793 						      tsc_timestamp));
1794 }
1795 
1796 /* returns true if host is using TSC based clocksource */
1797 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1798 					   u64 *tsc_timestamp)
1799 {
1800 	/* checked again under seqlock below */
1801 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1802 		return false;
1803 
1804 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1805 }
1806 #endif
1807 
1808 /*
1809  *
1810  * Assuming a stable TSC across physical CPUS, and a stable TSC
1811  * across virtual CPUs, the following condition is possible.
1812  * Each numbered line represents an event visible to both
1813  * CPUs at the next numbered event.
1814  *
1815  * "timespecX" represents host monotonic time. "tscX" represents
1816  * RDTSC value.
1817  *
1818  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1819  *
1820  * 1.  read timespec0,tsc0
1821  * 2.					| timespec1 = timespec0 + N
1822  * 					| tsc1 = tsc0 + M
1823  * 3. transition to guest		| transition to guest
1824  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1825  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1826  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1827  *
1828  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1829  *
1830  * 	- ret0 < ret1
1831  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1832  *		...
1833  *	- 0 < N - M => M < N
1834  *
1835  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1836  * always the case (the difference between two distinct xtime instances
1837  * might be smaller then the difference between corresponding TSC reads,
1838  * when updating guest vcpus pvclock areas).
1839  *
1840  * To avoid that problem, do not allow visibility of distinct
1841  * system_timestamp/tsc_timestamp values simultaneously: use a master
1842  * copy of host monotonic time values. Update that master copy
1843  * in lockstep.
1844  *
1845  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1846  *
1847  */
1848 
1849 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1850 {
1851 #ifdef CONFIG_X86_64
1852 	struct kvm_arch *ka = &kvm->arch;
1853 	int vclock_mode;
1854 	bool host_tsc_clocksource, vcpus_matched;
1855 
1856 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1857 			atomic_read(&kvm->online_vcpus));
1858 
1859 	/*
1860 	 * If the host uses TSC clock, then passthrough TSC as stable
1861 	 * to the guest.
1862 	 */
1863 	host_tsc_clocksource = kvm_get_time_and_clockread(
1864 					&ka->master_kernel_ns,
1865 					&ka->master_cycle_now);
1866 
1867 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1868 				&& !ka->backwards_tsc_observed
1869 				&& !ka->boot_vcpu_runs_old_kvmclock;
1870 
1871 	if (ka->use_master_clock)
1872 		atomic_set(&kvm_guest_has_master_clock, 1);
1873 
1874 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1875 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1876 					vcpus_matched);
1877 #endif
1878 }
1879 
1880 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1881 {
1882 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1883 }
1884 
1885 static void kvm_gen_update_masterclock(struct kvm *kvm)
1886 {
1887 #ifdef CONFIG_X86_64
1888 	int i;
1889 	struct kvm_vcpu *vcpu;
1890 	struct kvm_arch *ka = &kvm->arch;
1891 
1892 	spin_lock(&ka->pvclock_gtod_sync_lock);
1893 	kvm_make_mclock_inprogress_request(kvm);
1894 	/* no guest entries from this point */
1895 	pvclock_update_vm_gtod_copy(kvm);
1896 
1897 	kvm_for_each_vcpu(i, vcpu, kvm)
1898 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1899 
1900 	/* guest entries allowed */
1901 	kvm_for_each_vcpu(i, vcpu, kvm)
1902 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1903 
1904 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1905 #endif
1906 }
1907 
1908 u64 get_kvmclock_ns(struct kvm *kvm)
1909 {
1910 	struct kvm_arch *ka = &kvm->arch;
1911 	struct pvclock_vcpu_time_info hv_clock;
1912 	u64 ret;
1913 
1914 	spin_lock(&ka->pvclock_gtod_sync_lock);
1915 	if (!ka->use_master_clock) {
1916 		spin_unlock(&ka->pvclock_gtod_sync_lock);
1917 		return ktime_get_boot_ns() + ka->kvmclock_offset;
1918 	}
1919 
1920 	hv_clock.tsc_timestamp = ka->master_cycle_now;
1921 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1922 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1923 
1924 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
1925 	get_cpu();
1926 
1927 	if (__this_cpu_read(cpu_tsc_khz)) {
1928 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1929 				   &hv_clock.tsc_shift,
1930 				   &hv_clock.tsc_to_system_mul);
1931 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1932 	} else
1933 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1934 
1935 	put_cpu();
1936 
1937 	return ret;
1938 }
1939 
1940 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1941 {
1942 	struct kvm_vcpu_arch *vcpu = &v->arch;
1943 	struct pvclock_vcpu_time_info guest_hv_clock;
1944 
1945 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1946 		&guest_hv_clock, sizeof(guest_hv_clock))))
1947 		return;
1948 
1949 	/* This VCPU is paused, but it's legal for a guest to read another
1950 	 * VCPU's kvmclock, so we really have to follow the specification where
1951 	 * it says that version is odd if data is being modified, and even after
1952 	 * it is consistent.
1953 	 *
1954 	 * Version field updates must be kept separate.  This is because
1955 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1956 	 * writes within a string instruction are weakly ordered.  So there
1957 	 * are three writes overall.
1958 	 *
1959 	 * As a small optimization, only write the version field in the first
1960 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1961 	 * version field is the first in the struct.
1962 	 */
1963 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1964 
1965 	if (guest_hv_clock.version & 1)
1966 		++guest_hv_clock.version;  /* first time write, random junk */
1967 
1968 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1969 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1970 				&vcpu->hv_clock,
1971 				sizeof(vcpu->hv_clock.version));
1972 
1973 	smp_wmb();
1974 
1975 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1976 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1977 
1978 	if (vcpu->pvclock_set_guest_stopped_request) {
1979 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1980 		vcpu->pvclock_set_guest_stopped_request = false;
1981 	}
1982 
1983 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1984 
1985 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1986 				&vcpu->hv_clock,
1987 				sizeof(vcpu->hv_clock));
1988 
1989 	smp_wmb();
1990 
1991 	vcpu->hv_clock.version++;
1992 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1993 				&vcpu->hv_clock,
1994 				sizeof(vcpu->hv_clock.version));
1995 }
1996 
1997 static int kvm_guest_time_update(struct kvm_vcpu *v)
1998 {
1999 	unsigned long flags, tgt_tsc_khz;
2000 	struct kvm_vcpu_arch *vcpu = &v->arch;
2001 	struct kvm_arch *ka = &v->kvm->arch;
2002 	s64 kernel_ns;
2003 	u64 tsc_timestamp, host_tsc;
2004 	u8 pvclock_flags;
2005 	bool use_master_clock;
2006 
2007 	kernel_ns = 0;
2008 	host_tsc = 0;
2009 
2010 	/*
2011 	 * If the host uses TSC clock, then passthrough TSC as stable
2012 	 * to the guest.
2013 	 */
2014 	spin_lock(&ka->pvclock_gtod_sync_lock);
2015 	use_master_clock = ka->use_master_clock;
2016 	if (use_master_clock) {
2017 		host_tsc = ka->master_cycle_now;
2018 		kernel_ns = ka->master_kernel_ns;
2019 	}
2020 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2021 
2022 	/* Keep irq disabled to prevent changes to the clock */
2023 	local_irq_save(flags);
2024 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2025 	if (unlikely(tgt_tsc_khz == 0)) {
2026 		local_irq_restore(flags);
2027 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2028 		return 1;
2029 	}
2030 	if (!use_master_clock) {
2031 		host_tsc = rdtsc();
2032 		kernel_ns = ktime_get_boot_ns();
2033 	}
2034 
2035 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2036 
2037 	/*
2038 	 * We may have to catch up the TSC to match elapsed wall clock
2039 	 * time for two reasons, even if kvmclock is used.
2040 	 *   1) CPU could have been running below the maximum TSC rate
2041 	 *   2) Broken TSC compensation resets the base at each VCPU
2042 	 *      entry to avoid unknown leaps of TSC even when running
2043 	 *      again on the same CPU.  This may cause apparent elapsed
2044 	 *      time to disappear, and the guest to stand still or run
2045 	 *	very slowly.
2046 	 */
2047 	if (vcpu->tsc_catchup) {
2048 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2049 		if (tsc > tsc_timestamp) {
2050 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2051 			tsc_timestamp = tsc;
2052 		}
2053 	}
2054 
2055 	local_irq_restore(flags);
2056 
2057 	/* With all the info we got, fill in the values */
2058 
2059 	if (kvm_has_tsc_control)
2060 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2061 
2062 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2063 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2064 				   &vcpu->hv_clock.tsc_shift,
2065 				   &vcpu->hv_clock.tsc_to_system_mul);
2066 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2067 	}
2068 
2069 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2070 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2071 	vcpu->last_guest_tsc = tsc_timestamp;
2072 
2073 	/* If the host uses TSC clocksource, then it is stable */
2074 	pvclock_flags = 0;
2075 	if (use_master_clock)
2076 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2077 
2078 	vcpu->hv_clock.flags = pvclock_flags;
2079 
2080 	if (vcpu->pv_time_enabled)
2081 		kvm_setup_pvclock_page(v);
2082 	if (v == kvm_get_vcpu(v->kvm, 0))
2083 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2084 	return 0;
2085 }
2086 
2087 /*
2088  * kvmclock updates which are isolated to a given vcpu, such as
2089  * vcpu->cpu migration, should not allow system_timestamp from
2090  * the rest of the vcpus to remain static. Otherwise ntp frequency
2091  * correction applies to one vcpu's system_timestamp but not
2092  * the others.
2093  *
2094  * So in those cases, request a kvmclock update for all vcpus.
2095  * We need to rate-limit these requests though, as they can
2096  * considerably slow guests that have a large number of vcpus.
2097  * The time for a remote vcpu to update its kvmclock is bound
2098  * by the delay we use to rate-limit the updates.
2099  */
2100 
2101 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2102 
2103 static void kvmclock_update_fn(struct work_struct *work)
2104 {
2105 	int i;
2106 	struct delayed_work *dwork = to_delayed_work(work);
2107 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2108 					   kvmclock_update_work);
2109 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2110 	struct kvm_vcpu *vcpu;
2111 
2112 	kvm_for_each_vcpu(i, vcpu, kvm) {
2113 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2114 		kvm_vcpu_kick(vcpu);
2115 	}
2116 }
2117 
2118 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2119 {
2120 	struct kvm *kvm = v->kvm;
2121 
2122 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2123 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2124 					KVMCLOCK_UPDATE_DELAY);
2125 }
2126 
2127 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2128 
2129 static void kvmclock_sync_fn(struct work_struct *work)
2130 {
2131 	struct delayed_work *dwork = to_delayed_work(work);
2132 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2133 					   kvmclock_sync_work);
2134 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2135 
2136 	if (!kvmclock_periodic_sync)
2137 		return;
2138 
2139 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2140 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2141 					KVMCLOCK_SYNC_PERIOD);
2142 }
2143 
2144 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2145 {
2146 	u64 mcg_cap = vcpu->arch.mcg_cap;
2147 	unsigned bank_num = mcg_cap & 0xff;
2148 	u32 msr = msr_info->index;
2149 	u64 data = msr_info->data;
2150 
2151 	switch (msr) {
2152 	case MSR_IA32_MCG_STATUS:
2153 		vcpu->arch.mcg_status = data;
2154 		break;
2155 	case MSR_IA32_MCG_CTL:
2156 		if (!(mcg_cap & MCG_CTL_P))
2157 			return 1;
2158 		if (data != 0 && data != ~(u64)0)
2159 			return -1;
2160 		vcpu->arch.mcg_ctl = data;
2161 		break;
2162 	default:
2163 		if (msr >= MSR_IA32_MC0_CTL &&
2164 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2165 			u32 offset = msr - MSR_IA32_MC0_CTL;
2166 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2167 			 * some Linux kernels though clear bit 10 in bank 4 to
2168 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2169 			 * this to avoid an uncatched #GP in the guest
2170 			 */
2171 			if ((offset & 0x3) == 0 &&
2172 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2173 				return -1;
2174 			if (!msr_info->host_initiated &&
2175 				(offset & 0x3) == 1 && data != 0)
2176 				return -1;
2177 			vcpu->arch.mce_banks[offset] = data;
2178 			break;
2179 		}
2180 		return 1;
2181 	}
2182 	return 0;
2183 }
2184 
2185 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2186 {
2187 	struct kvm *kvm = vcpu->kvm;
2188 	int lm = is_long_mode(vcpu);
2189 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2190 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2191 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2192 		: kvm->arch.xen_hvm_config.blob_size_32;
2193 	u32 page_num = data & ~PAGE_MASK;
2194 	u64 page_addr = data & PAGE_MASK;
2195 	u8 *page;
2196 	int r;
2197 
2198 	r = -E2BIG;
2199 	if (page_num >= blob_size)
2200 		goto out;
2201 	r = -ENOMEM;
2202 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2203 	if (IS_ERR(page)) {
2204 		r = PTR_ERR(page);
2205 		goto out;
2206 	}
2207 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2208 		goto out_free;
2209 	r = 0;
2210 out_free:
2211 	kfree(page);
2212 out:
2213 	return r;
2214 }
2215 
2216 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2217 {
2218 	gpa_t gpa = data & ~0x3f;
2219 
2220 	/* Bits 3:5 are reserved, Should be zero */
2221 	if (data & 0x38)
2222 		return 1;
2223 
2224 	vcpu->arch.apf.msr_val = data;
2225 
2226 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2227 		kvm_clear_async_pf_completion_queue(vcpu);
2228 		kvm_async_pf_hash_reset(vcpu);
2229 		return 0;
2230 	}
2231 
2232 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2233 					sizeof(u32)))
2234 		return 1;
2235 
2236 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2237 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2238 	kvm_async_pf_wakeup_all(vcpu);
2239 	return 0;
2240 }
2241 
2242 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2243 {
2244 	vcpu->arch.pv_time_enabled = false;
2245 }
2246 
2247 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2248 {
2249 	++vcpu->stat.tlb_flush;
2250 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2251 }
2252 
2253 static void record_steal_time(struct kvm_vcpu *vcpu)
2254 {
2255 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2256 		return;
2257 
2258 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2259 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2260 		return;
2261 
2262 	/*
2263 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2264 	 * expensive IPIs.
2265 	 */
2266 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2267 		kvm_vcpu_flush_tlb(vcpu, false);
2268 
2269 	if (vcpu->arch.st.steal.version & 1)
2270 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2271 
2272 	vcpu->arch.st.steal.version += 1;
2273 
2274 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2275 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2276 
2277 	smp_wmb();
2278 
2279 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2280 		vcpu->arch.st.last_steal;
2281 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2282 
2283 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2284 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2285 
2286 	smp_wmb();
2287 
2288 	vcpu->arch.st.steal.version += 1;
2289 
2290 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2291 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2292 }
2293 
2294 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2295 {
2296 	bool pr = false;
2297 	u32 msr = msr_info->index;
2298 	u64 data = msr_info->data;
2299 
2300 	switch (msr) {
2301 	case MSR_AMD64_NB_CFG:
2302 	case MSR_IA32_UCODE_WRITE:
2303 	case MSR_VM_HSAVE_PA:
2304 	case MSR_AMD64_PATCH_LOADER:
2305 	case MSR_AMD64_BU_CFG2:
2306 	case MSR_AMD64_DC_CFG:
2307 		break;
2308 
2309 	case MSR_IA32_UCODE_REV:
2310 		if (msr_info->host_initiated)
2311 			vcpu->arch.microcode_version = data;
2312 		break;
2313 	case MSR_EFER:
2314 		return set_efer(vcpu, data);
2315 	case MSR_K7_HWCR:
2316 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2317 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2318 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2319 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2320 		if (data != 0) {
2321 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2322 				    data);
2323 			return 1;
2324 		}
2325 		break;
2326 	case MSR_FAM10H_MMIO_CONF_BASE:
2327 		if (data != 0) {
2328 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2329 				    "0x%llx\n", data);
2330 			return 1;
2331 		}
2332 		break;
2333 	case MSR_IA32_DEBUGCTLMSR:
2334 		if (!data) {
2335 			/* We support the non-activated case already */
2336 			break;
2337 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2338 			/* Values other than LBR and BTF are vendor-specific,
2339 			   thus reserved and should throw a #GP */
2340 			return 1;
2341 		}
2342 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2343 			    __func__, data);
2344 		break;
2345 	case 0x200 ... 0x2ff:
2346 		return kvm_mtrr_set_msr(vcpu, msr, data);
2347 	case MSR_IA32_APICBASE:
2348 		return kvm_set_apic_base(vcpu, msr_info);
2349 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2350 		return kvm_x2apic_msr_write(vcpu, msr, data);
2351 	case MSR_IA32_TSCDEADLINE:
2352 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2353 		break;
2354 	case MSR_IA32_TSC_ADJUST:
2355 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2356 			if (!msr_info->host_initiated) {
2357 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2358 				adjust_tsc_offset_guest(vcpu, adj);
2359 			}
2360 			vcpu->arch.ia32_tsc_adjust_msr = data;
2361 		}
2362 		break;
2363 	case MSR_IA32_MISC_ENABLE:
2364 		vcpu->arch.ia32_misc_enable_msr = data;
2365 		break;
2366 	case MSR_IA32_SMBASE:
2367 		if (!msr_info->host_initiated)
2368 			return 1;
2369 		vcpu->arch.smbase = data;
2370 		break;
2371 	case MSR_IA32_TSC:
2372 		kvm_write_tsc(vcpu, msr_info);
2373 		break;
2374 	case MSR_SMI_COUNT:
2375 		if (!msr_info->host_initiated)
2376 			return 1;
2377 		vcpu->arch.smi_count = data;
2378 		break;
2379 	case MSR_KVM_WALL_CLOCK_NEW:
2380 	case MSR_KVM_WALL_CLOCK:
2381 		vcpu->kvm->arch.wall_clock = data;
2382 		kvm_write_wall_clock(vcpu->kvm, data);
2383 		break;
2384 	case MSR_KVM_SYSTEM_TIME_NEW:
2385 	case MSR_KVM_SYSTEM_TIME: {
2386 		struct kvm_arch *ka = &vcpu->kvm->arch;
2387 
2388 		kvmclock_reset(vcpu);
2389 
2390 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2391 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2392 
2393 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2394 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2395 
2396 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2397 		}
2398 
2399 		vcpu->arch.time = data;
2400 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2401 
2402 		/* we verify if the enable bit is set... */
2403 		if (!(data & 1))
2404 			break;
2405 
2406 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2407 		     &vcpu->arch.pv_time, data & ~1ULL,
2408 		     sizeof(struct pvclock_vcpu_time_info)))
2409 			vcpu->arch.pv_time_enabled = false;
2410 		else
2411 			vcpu->arch.pv_time_enabled = true;
2412 
2413 		break;
2414 	}
2415 	case MSR_KVM_ASYNC_PF_EN:
2416 		if (kvm_pv_enable_async_pf(vcpu, data))
2417 			return 1;
2418 		break;
2419 	case MSR_KVM_STEAL_TIME:
2420 
2421 		if (unlikely(!sched_info_on()))
2422 			return 1;
2423 
2424 		if (data & KVM_STEAL_RESERVED_MASK)
2425 			return 1;
2426 
2427 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2428 						data & KVM_STEAL_VALID_BITS,
2429 						sizeof(struct kvm_steal_time)))
2430 			return 1;
2431 
2432 		vcpu->arch.st.msr_val = data;
2433 
2434 		if (!(data & KVM_MSR_ENABLED))
2435 			break;
2436 
2437 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2438 
2439 		break;
2440 	case MSR_KVM_PV_EOI_EN:
2441 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2442 			return 1;
2443 		break;
2444 
2445 	case MSR_IA32_MCG_CTL:
2446 	case MSR_IA32_MCG_STATUS:
2447 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2448 		return set_msr_mce(vcpu, msr_info);
2449 
2450 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2451 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2452 		pr = true; /* fall through */
2453 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2454 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2455 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2456 			return kvm_pmu_set_msr(vcpu, msr_info);
2457 
2458 		if (pr || data != 0)
2459 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2460 				    "0x%x data 0x%llx\n", msr, data);
2461 		break;
2462 	case MSR_K7_CLK_CTL:
2463 		/*
2464 		 * Ignore all writes to this no longer documented MSR.
2465 		 * Writes are only relevant for old K7 processors,
2466 		 * all pre-dating SVM, but a recommended workaround from
2467 		 * AMD for these chips. It is possible to specify the
2468 		 * affected processor models on the command line, hence
2469 		 * the need to ignore the workaround.
2470 		 */
2471 		break;
2472 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2473 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2474 	case HV_X64_MSR_CRASH_CTL:
2475 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2476 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2477 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2478 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2479 		return kvm_hv_set_msr_common(vcpu, msr, data,
2480 					     msr_info->host_initiated);
2481 	case MSR_IA32_BBL_CR_CTL3:
2482 		/* Drop writes to this legacy MSR -- see rdmsr
2483 		 * counterpart for further detail.
2484 		 */
2485 		if (report_ignored_msrs)
2486 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2487 				msr, data);
2488 		break;
2489 	case MSR_AMD64_OSVW_ID_LENGTH:
2490 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2491 			return 1;
2492 		vcpu->arch.osvw.length = data;
2493 		break;
2494 	case MSR_AMD64_OSVW_STATUS:
2495 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2496 			return 1;
2497 		vcpu->arch.osvw.status = data;
2498 		break;
2499 	case MSR_PLATFORM_INFO:
2500 		if (!msr_info->host_initiated ||
2501 		    data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2502 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2503 		     cpuid_fault_enabled(vcpu)))
2504 			return 1;
2505 		vcpu->arch.msr_platform_info = data;
2506 		break;
2507 	case MSR_MISC_FEATURES_ENABLES:
2508 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2509 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2510 		     !supports_cpuid_fault(vcpu)))
2511 			return 1;
2512 		vcpu->arch.msr_misc_features_enables = data;
2513 		break;
2514 	default:
2515 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2516 			return xen_hvm_config(vcpu, data);
2517 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2518 			return kvm_pmu_set_msr(vcpu, msr_info);
2519 		if (!ignore_msrs) {
2520 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2521 				    msr, data);
2522 			return 1;
2523 		} else {
2524 			if (report_ignored_msrs)
2525 				vcpu_unimpl(vcpu,
2526 					"ignored wrmsr: 0x%x data 0x%llx\n",
2527 					msr, data);
2528 			break;
2529 		}
2530 	}
2531 	return 0;
2532 }
2533 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2534 
2535 
2536 /*
2537  * Reads an msr value (of 'msr_index') into 'pdata'.
2538  * Returns 0 on success, non-0 otherwise.
2539  * Assumes vcpu_load() was already called.
2540  */
2541 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2542 {
2543 	return kvm_x86_ops->get_msr(vcpu, msr);
2544 }
2545 EXPORT_SYMBOL_GPL(kvm_get_msr);
2546 
2547 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2548 {
2549 	u64 data;
2550 	u64 mcg_cap = vcpu->arch.mcg_cap;
2551 	unsigned bank_num = mcg_cap & 0xff;
2552 
2553 	switch (msr) {
2554 	case MSR_IA32_P5_MC_ADDR:
2555 	case MSR_IA32_P5_MC_TYPE:
2556 		data = 0;
2557 		break;
2558 	case MSR_IA32_MCG_CAP:
2559 		data = vcpu->arch.mcg_cap;
2560 		break;
2561 	case MSR_IA32_MCG_CTL:
2562 		if (!(mcg_cap & MCG_CTL_P))
2563 			return 1;
2564 		data = vcpu->arch.mcg_ctl;
2565 		break;
2566 	case MSR_IA32_MCG_STATUS:
2567 		data = vcpu->arch.mcg_status;
2568 		break;
2569 	default:
2570 		if (msr >= MSR_IA32_MC0_CTL &&
2571 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2572 			u32 offset = msr - MSR_IA32_MC0_CTL;
2573 			data = vcpu->arch.mce_banks[offset];
2574 			break;
2575 		}
2576 		return 1;
2577 	}
2578 	*pdata = data;
2579 	return 0;
2580 }
2581 
2582 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2583 {
2584 	switch (msr_info->index) {
2585 	case MSR_IA32_PLATFORM_ID:
2586 	case MSR_IA32_EBL_CR_POWERON:
2587 	case MSR_IA32_DEBUGCTLMSR:
2588 	case MSR_IA32_LASTBRANCHFROMIP:
2589 	case MSR_IA32_LASTBRANCHTOIP:
2590 	case MSR_IA32_LASTINTFROMIP:
2591 	case MSR_IA32_LASTINTTOIP:
2592 	case MSR_K8_SYSCFG:
2593 	case MSR_K8_TSEG_ADDR:
2594 	case MSR_K8_TSEG_MASK:
2595 	case MSR_K7_HWCR:
2596 	case MSR_VM_HSAVE_PA:
2597 	case MSR_K8_INT_PENDING_MSG:
2598 	case MSR_AMD64_NB_CFG:
2599 	case MSR_FAM10H_MMIO_CONF_BASE:
2600 	case MSR_AMD64_BU_CFG2:
2601 	case MSR_IA32_PERF_CTL:
2602 	case MSR_AMD64_DC_CFG:
2603 		msr_info->data = 0;
2604 		break;
2605 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2606 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2607 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2608 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2609 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2610 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2611 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2612 		msr_info->data = 0;
2613 		break;
2614 	case MSR_IA32_UCODE_REV:
2615 		msr_info->data = vcpu->arch.microcode_version;
2616 		break;
2617 	case MSR_IA32_TSC:
2618 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2619 		break;
2620 	case MSR_MTRRcap:
2621 	case 0x200 ... 0x2ff:
2622 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2623 	case 0xcd: /* fsb frequency */
2624 		msr_info->data = 3;
2625 		break;
2626 		/*
2627 		 * MSR_EBC_FREQUENCY_ID
2628 		 * Conservative value valid for even the basic CPU models.
2629 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2630 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2631 		 * and 266MHz for model 3, or 4. Set Core Clock
2632 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2633 		 * 31:24) even though these are only valid for CPU
2634 		 * models > 2, however guests may end up dividing or
2635 		 * multiplying by zero otherwise.
2636 		 */
2637 	case MSR_EBC_FREQUENCY_ID:
2638 		msr_info->data = 1 << 24;
2639 		break;
2640 	case MSR_IA32_APICBASE:
2641 		msr_info->data = kvm_get_apic_base(vcpu);
2642 		break;
2643 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2644 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2645 		break;
2646 	case MSR_IA32_TSCDEADLINE:
2647 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2648 		break;
2649 	case MSR_IA32_TSC_ADJUST:
2650 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2651 		break;
2652 	case MSR_IA32_MISC_ENABLE:
2653 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2654 		break;
2655 	case MSR_IA32_SMBASE:
2656 		if (!msr_info->host_initiated)
2657 			return 1;
2658 		msr_info->data = vcpu->arch.smbase;
2659 		break;
2660 	case MSR_SMI_COUNT:
2661 		msr_info->data = vcpu->arch.smi_count;
2662 		break;
2663 	case MSR_IA32_PERF_STATUS:
2664 		/* TSC increment by tick */
2665 		msr_info->data = 1000ULL;
2666 		/* CPU multiplier */
2667 		msr_info->data |= (((uint64_t)4ULL) << 40);
2668 		break;
2669 	case MSR_EFER:
2670 		msr_info->data = vcpu->arch.efer;
2671 		break;
2672 	case MSR_KVM_WALL_CLOCK:
2673 	case MSR_KVM_WALL_CLOCK_NEW:
2674 		msr_info->data = vcpu->kvm->arch.wall_clock;
2675 		break;
2676 	case MSR_KVM_SYSTEM_TIME:
2677 	case MSR_KVM_SYSTEM_TIME_NEW:
2678 		msr_info->data = vcpu->arch.time;
2679 		break;
2680 	case MSR_KVM_ASYNC_PF_EN:
2681 		msr_info->data = vcpu->arch.apf.msr_val;
2682 		break;
2683 	case MSR_KVM_STEAL_TIME:
2684 		msr_info->data = vcpu->arch.st.msr_val;
2685 		break;
2686 	case MSR_KVM_PV_EOI_EN:
2687 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2688 		break;
2689 	case MSR_IA32_P5_MC_ADDR:
2690 	case MSR_IA32_P5_MC_TYPE:
2691 	case MSR_IA32_MCG_CAP:
2692 	case MSR_IA32_MCG_CTL:
2693 	case MSR_IA32_MCG_STATUS:
2694 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2695 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2696 	case MSR_K7_CLK_CTL:
2697 		/*
2698 		 * Provide expected ramp-up count for K7. All other
2699 		 * are set to zero, indicating minimum divisors for
2700 		 * every field.
2701 		 *
2702 		 * This prevents guest kernels on AMD host with CPU
2703 		 * type 6, model 8 and higher from exploding due to
2704 		 * the rdmsr failing.
2705 		 */
2706 		msr_info->data = 0x20000000;
2707 		break;
2708 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2709 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2710 	case HV_X64_MSR_CRASH_CTL:
2711 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2712 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2713 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2714 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2715 		return kvm_hv_get_msr_common(vcpu,
2716 					     msr_info->index, &msr_info->data);
2717 		break;
2718 	case MSR_IA32_BBL_CR_CTL3:
2719 		/* This legacy MSR exists but isn't fully documented in current
2720 		 * silicon.  It is however accessed by winxp in very narrow
2721 		 * scenarios where it sets bit #19, itself documented as
2722 		 * a "reserved" bit.  Best effort attempt to source coherent
2723 		 * read data here should the balance of the register be
2724 		 * interpreted by the guest:
2725 		 *
2726 		 * L2 cache control register 3: 64GB range, 256KB size,
2727 		 * enabled, latency 0x1, configured
2728 		 */
2729 		msr_info->data = 0xbe702111;
2730 		break;
2731 	case MSR_AMD64_OSVW_ID_LENGTH:
2732 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2733 			return 1;
2734 		msr_info->data = vcpu->arch.osvw.length;
2735 		break;
2736 	case MSR_AMD64_OSVW_STATUS:
2737 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2738 			return 1;
2739 		msr_info->data = vcpu->arch.osvw.status;
2740 		break;
2741 	case MSR_PLATFORM_INFO:
2742 		msr_info->data = vcpu->arch.msr_platform_info;
2743 		break;
2744 	case MSR_MISC_FEATURES_ENABLES:
2745 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2746 		break;
2747 	default:
2748 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2749 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2750 		if (!ignore_msrs) {
2751 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2752 					       msr_info->index);
2753 			return 1;
2754 		} else {
2755 			if (report_ignored_msrs)
2756 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2757 					msr_info->index);
2758 			msr_info->data = 0;
2759 		}
2760 		break;
2761 	}
2762 	return 0;
2763 }
2764 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2765 
2766 /*
2767  * Read or write a bunch of msrs. All parameters are kernel addresses.
2768  *
2769  * @return number of msrs set successfully.
2770  */
2771 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2772 		    struct kvm_msr_entry *entries,
2773 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2774 				  unsigned index, u64 *data))
2775 {
2776 	int i;
2777 
2778 	for (i = 0; i < msrs->nmsrs; ++i)
2779 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2780 			break;
2781 
2782 	return i;
2783 }
2784 
2785 /*
2786  * Read or write a bunch of msrs. Parameters are user addresses.
2787  *
2788  * @return number of msrs set successfully.
2789  */
2790 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2791 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2792 				unsigned index, u64 *data),
2793 		  int writeback)
2794 {
2795 	struct kvm_msrs msrs;
2796 	struct kvm_msr_entry *entries;
2797 	int r, n;
2798 	unsigned size;
2799 
2800 	r = -EFAULT;
2801 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2802 		goto out;
2803 
2804 	r = -E2BIG;
2805 	if (msrs.nmsrs >= MAX_IO_MSRS)
2806 		goto out;
2807 
2808 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2809 	entries = memdup_user(user_msrs->entries, size);
2810 	if (IS_ERR(entries)) {
2811 		r = PTR_ERR(entries);
2812 		goto out;
2813 	}
2814 
2815 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2816 	if (r < 0)
2817 		goto out_free;
2818 
2819 	r = -EFAULT;
2820 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2821 		goto out_free;
2822 
2823 	r = n;
2824 
2825 out_free:
2826 	kfree(entries);
2827 out:
2828 	return r;
2829 }
2830 
2831 static inline bool kvm_can_mwait_in_guest(void)
2832 {
2833 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2834 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2835 		boot_cpu_has(X86_FEATURE_ARAT);
2836 }
2837 
2838 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2839 {
2840 	int r = 0;
2841 
2842 	switch (ext) {
2843 	case KVM_CAP_IRQCHIP:
2844 	case KVM_CAP_HLT:
2845 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2846 	case KVM_CAP_SET_TSS_ADDR:
2847 	case KVM_CAP_EXT_CPUID:
2848 	case KVM_CAP_EXT_EMUL_CPUID:
2849 	case KVM_CAP_CLOCKSOURCE:
2850 	case KVM_CAP_PIT:
2851 	case KVM_CAP_NOP_IO_DELAY:
2852 	case KVM_CAP_MP_STATE:
2853 	case KVM_CAP_SYNC_MMU:
2854 	case KVM_CAP_USER_NMI:
2855 	case KVM_CAP_REINJECT_CONTROL:
2856 	case KVM_CAP_IRQ_INJECT_STATUS:
2857 	case KVM_CAP_IOEVENTFD:
2858 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2859 	case KVM_CAP_PIT2:
2860 	case KVM_CAP_PIT_STATE2:
2861 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2862 	case KVM_CAP_XEN_HVM:
2863 	case KVM_CAP_VCPU_EVENTS:
2864 	case KVM_CAP_HYPERV:
2865 	case KVM_CAP_HYPERV_VAPIC:
2866 	case KVM_CAP_HYPERV_SPIN:
2867 	case KVM_CAP_HYPERV_SYNIC:
2868 	case KVM_CAP_HYPERV_SYNIC2:
2869 	case KVM_CAP_HYPERV_VP_INDEX:
2870 	case KVM_CAP_HYPERV_EVENTFD:
2871 	case KVM_CAP_PCI_SEGMENT:
2872 	case KVM_CAP_DEBUGREGS:
2873 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2874 	case KVM_CAP_XSAVE:
2875 	case KVM_CAP_ASYNC_PF:
2876 	case KVM_CAP_GET_TSC_KHZ:
2877 	case KVM_CAP_KVMCLOCK_CTRL:
2878 	case KVM_CAP_READONLY_MEM:
2879 	case KVM_CAP_HYPERV_TIME:
2880 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2881 	case KVM_CAP_TSC_DEADLINE_TIMER:
2882 	case KVM_CAP_ENABLE_CAP_VM:
2883 	case KVM_CAP_DISABLE_QUIRKS:
2884 	case KVM_CAP_SET_BOOT_CPU_ID:
2885  	case KVM_CAP_SPLIT_IRQCHIP:
2886 	case KVM_CAP_IMMEDIATE_EXIT:
2887 	case KVM_CAP_GET_MSR_FEATURES:
2888 		r = 1;
2889 		break;
2890 	case KVM_CAP_SYNC_REGS:
2891 		r = KVM_SYNC_X86_VALID_FIELDS;
2892 		break;
2893 	case KVM_CAP_ADJUST_CLOCK:
2894 		r = KVM_CLOCK_TSC_STABLE;
2895 		break;
2896 	case KVM_CAP_X86_DISABLE_EXITS:
2897 		r |=  KVM_X86_DISABLE_EXITS_HTL | KVM_X86_DISABLE_EXITS_PAUSE;
2898 		if(kvm_can_mwait_in_guest())
2899 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
2900 		break;
2901 	case KVM_CAP_X86_SMM:
2902 		/* SMBASE is usually relocated above 1M on modern chipsets,
2903 		 * and SMM handlers might indeed rely on 4G segment limits,
2904 		 * so do not report SMM to be available if real mode is
2905 		 * emulated via vm86 mode.  Still, do not go to great lengths
2906 		 * to avoid userspace's usage of the feature, because it is a
2907 		 * fringe case that is not enabled except via specific settings
2908 		 * of the module parameters.
2909 		 */
2910 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
2911 		break;
2912 	case KVM_CAP_VAPIC:
2913 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2914 		break;
2915 	case KVM_CAP_NR_VCPUS:
2916 		r = KVM_SOFT_MAX_VCPUS;
2917 		break;
2918 	case KVM_CAP_MAX_VCPUS:
2919 		r = KVM_MAX_VCPUS;
2920 		break;
2921 	case KVM_CAP_NR_MEMSLOTS:
2922 		r = KVM_USER_MEM_SLOTS;
2923 		break;
2924 	case KVM_CAP_PV_MMU:	/* obsolete */
2925 		r = 0;
2926 		break;
2927 	case KVM_CAP_MCE:
2928 		r = KVM_MAX_MCE_BANKS;
2929 		break;
2930 	case KVM_CAP_XCRS:
2931 		r = boot_cpu_has(X86_FEATURE_XSAVE);
2932 		break;
2933 	case KVM_CAP_TSC_CONTROL:
2934 		r = kvm_has_tsc_control;
2935 		break;
2936 	case KVM_CAP_X2APIC_API:
2937 		r = KVM_X2APIC_API_VALID_FLAGS;
2938 		break;
2939 	default:
2940 		break;
2941 	}
2942 	return r;
2943 
2944 }
2945 
2946 long kvm_arch_dev_ioctl(struct file *filp,
2947 			unsigned int ioctl, unsigned long arg)
2948 {
2949 	void __user *argp = (void __user *)arg;
2950 	long r;
2951 
2952 	switch (ioctl) {
2953 	case KVM_GET_MSR_INDEX_LIST: {
2954 		struct kvm_msr_list __user *user_msr_list = argp;
2955 		struct kvm_msr_list msr_list;
2956 		unsigned n;
2957 
2958 		r = -EFAULT;
2959 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2960 			goto out;
2961 		n = msr_list.nmsrs;
2962 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2963 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2964 			goto out;
2965 		r = -E2BIG;
2966 		if (n < msr_list.nmsrs)
2967 			goto out;
2968 		r = -EFAULT;
2969 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2970 				 num_msrs_to_save * sizeof(u32)))
2971 			goto out;
2972 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2973 				 &emulated_msrs,
2974 				 num_emulated_msrs * sizeof(u32)))
2975 			goto out;
2976 		r = 0;
2977 		break;
2978 	}
2979 	case KVM_GET_SUPPORTED_CPUID:
2980 	case KVM_GET_EMULATED_CPUID: {
2981 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2982 		struct kvm_cpuid2 cpuid;
2983 
2984 		r = -EFAULT;
2985 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2986 			goto out;
2987 
2988 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2989 					    ioctl);
2990 		if (r)
2991 			goto out;
2992 
2993 		r = -EFAULT;
2994 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2995 			goto out;
2996 		r = 0;
2997 		break;
2998 	}
2999 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3000 		r = -EFAULT;
3001 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3002 				 sizeof(kvm_mce_cap_supported)))
3003 			goto out;
3004 		r = 0;
3005 		break;
3006 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3007 		struct kvm_msr_list __user *user_msr_list = argp;
3008 		struct kvm_msr_list msr_list;
3009 		unsigned int n;
3010 
3011 		r = -EFAULT;
3012 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3013 			goto out;
3014 		n = msr_list.nmsrs;
3015 		msr_list.nmsrs = num_msr_based_features;
3016 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3017 			goto out;
3018 		r = -E2BIG;
3019 		if (n < msr_list.nmsrs)
3020 			goto out;
3021 		r = -EFAULT;
3022 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3023 				 num_msr_based_features * sizeof(u32)))
3024 			goto out;
3025 		r = 0;
3026 		break;
3027 	}
3028 	case KVM_GET_MSRS:
3029 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3030 		break;
3031 	}
3032 	default:
3033 		r = -EINVAL;
3034 	}
3035 out:
3036 	return r;
3037 }
3038 
3039 static void wbinvd_ipi(void *garbage)
3040 {
3041 	wbinvd();
3042 }
3043 
3044 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3045 {
3046 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3047 }
3048 
3049 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3050 {
3051 	/* Address WBINVD may be executed by guest */
3052 	if (need_emulate_wbinvd(vcpu)) {
3053 		if (kvm_x86_ops->has_wbinvd_exit())
3054 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3055 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3056 			smp_call_function_single(vcpu->cpu,
3057 					wbinvd_ipi, NULL, 1);
3058 	}
3059 
3060 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3061 
3062 	/* Apply any externally detected TSC adjustments (due to suspend) */
3063 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3064 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3065 		vcpu->arch.tsc_offset_adjustment = 0;
3066 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3067 	}
3068 
3069 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3070 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3071 				rdtsc() - vcpu->arch.last_host_tsc;
3072 		if (tsc_delta < 0)
3073 			mark_tsc_unstable("KVM discovered backwards TSC");
3074 
3075 		if (kvm_check_tsc_unstable()) {
3076 			u64 offset = kvm_compute_tsc_offset(vcpu,
3077 						vcpu->arch.last_guest_tsc);
3078 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3079 			vcpu->arch.tsc_catchup = 1;
3080 		}
3081 
3082 		if (kvm_lapic_hv_timer_in_use(vcpu))
3083 			kvm_lapic_restart_hv_timer(vcpu);
3084 
3085 		/*
3086 		 * On a host with synchronized TSC, there is no need to update
3087 		 * kvmclock on vcpu->cpu migration
3088 		 */
3089 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3090 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3091 		if (vcpu->cpu != cpu)
3092 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3093 		vcpu->cpu = cpu;
3094 	}
3095 
3096 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3097 }
3098 
3099 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3100 {
3101 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3102 		return;
3103 
3104 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3105 
3106 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3107 			&vcpu->arch.st.steal.preempted,
3108 			offsetof(struct kvm_steal_time, preempted),
3109 			sizeof(vcpu->arch.st.steal.preempted));
3110 }
3111 
3112 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3113 {
3114 	int idx;
3115 
3116 	if (vcpu->preempted)
3117 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3118 
3119 	/*
3120 	 * Disable page faults because we're in atomic context here.
3121 	 * kvm_write_guest_offset_cached() would call might_fault()
3122 	 * that relies on pagefault_disable() to tell if there's a
3123 	 * bug. NOTE: the write to guest memory may not go through if
3124 	 * during postcopy live migration or if there's heavy guest
3125 	 * paging.
3126 	 */
3127 	pagefault_disable();
3128 	/*
3129 	 * kvm_memslots() will be called by
3130 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3131 	 */
3132 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3133 	kvm_steal_time_set_preempted(vcpu);
3134 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3135 	pagefault_enable();
3136 	kvm_x86_ops->vcpu_put(vcpu);
3137 	vcpu->arch.last_host_tsc = rdtsc();
3138 	/*
3139 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3140 	 * on every vmexit, but if not, we might have a stale dr6 from the
3141 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3142 	 */
3143 	set_debugreg(0, 6);
3144 }
3145 
3146 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3147 				    struct kvm_lapic_state *s)
3148 {
3149 	if (vcpu->arch.apicv_active)
3150 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3151 
3152 	return kvm_apic_get_state(vcpu, s);
3153 }
3154 
3155 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3156 				    struct kvm_lapic_state *s)
3157 {
3158 	int r;
3159 
3160 	r = kvm_apic_set_state(vcpu, s);
3161 	if (r)
3162 		return r;
3163 	update_cr8_intercept(vcpu);
3164 
3165 	return 0;
3166 }
3167 
3168 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3169 {
3170 	return (!lapic_in_kernel(vcpu) ||
3171 		kvm_apic_accept_pic_intr(vcpu));
3172 }
3173 
3174 /*
3175  * if userspace requested an interrupt window, check that the
3176  * interrupt window is open.
3177  *
3178  * No need to exit to userspace if we already have an interrupt queued.
3179  */
3180 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3181 {
3182 	return kvm_arch_interrupt_allowed(vcpu) &&
3183 		!kvm_cpu_has_interrupt(vcpu) &&
3184 		!kvm_event_needs_reinjection(vcpu) &&
3185 		kvm_cpu_accept_dm_intr(vcpu);
3186 }
3187 
3188 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3189 				    struct kvm_interrupt *irq)
3190 {
3191 	if (irq->irq >= KVM_NR_INTERRUPTS)
3192 		return -EINVAL;
3193 
3194 	if (!irqchip_in_kernel(vcpu->kvm)) {
3195 		kvm_queue_interrupt(vcpu, irq->irq, false);
3196 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3197 		return 0;
3198 	}
3199 
3200 	/*
3201 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3202 	 * fail for in-kernel 8259.
3203 	 */
3204 	if (pic_in_kernel(vcpu->kvm))
3205 		return -ENXIO;
3206 
3207 	if (vcpu->arch.pending_external_vector != -1)
3208 		return -EEXIST;
3209 
3210 	vcpu->arch.pending_external_vector = irq->irq;
3211 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3212 	return 0;
3213 }
3214 
3215 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3216 {
3217 	kvm_inject_nmi(vcpu);
3218 
3219 	return 0;
3220 }
3221 
3222 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3223 {
3224 	kvm_make_request(KVM_REQ_SMI, vcpu);
3225 
3226 	return 0;
3227 }
3228 
3229 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3230 					   struct kvm_tpr_access_ctl *tac)
3231 {
3232 	if (tac->flags)
3233 		return -EINVAL;
3234 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3235 	return 0;
3236 }
3237 
3238 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3239 					u64 mcg_cap)
3240 {
3241 	int r;
3242 	unsigned bank_num = mcg_cap & 0xff, bank;
3243 
3244 	r = -EINVAL;
3245 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3246 		goto out;
3247 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3248 		goto out;
3249 	r = 0;
3250 	vcpu->arch.mcg_cap = mcg_cap;
3251 	/* Init IA32_MCG_CTL to all 1s */
3252 	if (mcg_cap & MCG_CTL_P)
3253 		vcpu->arch.mcg_ctl = ~(u64)0;
3254 	/* Init IA32_MCi_CTL to all 1s */
3255 	for (bank = 0; bank < bank_num; bank++)
3256 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3257 
3258 	if (kvm_x86_ops->setup_mce)
3259 		kvm_x86_ops->setup_mce(vcpu);
3260 out:
3261 	return r;
3262 }
3263 
3264 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3265 				      struct kvm_x86_mce *mce)
3266 {
3267 	u64 mcg_cap = vcpu->arch.mcg_cap;
3268 	unsigned bank_num = mcg_cap & 0xff;
3269 	u64 *banks = vcpu->arch.mce_banks;
3270 
3271 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3272 		return -EINVAL;
3273 	/*
3274 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3275 	 * reporting is disabled
3276 	 */
3277 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3278 	    vcpu->arch.mcg_ctl != ~(u64)0)
3279 		return 0;
3280 	banks += 4 * mce->bank;
3281 	/*
3282 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3283 	 * reporting is disabled for the bank
3284 	 */
3285 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3286 		return 0;
3287 	if (mce->status & MCI_STATUS_UC) {
3288 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3289 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3290 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3291 			return 0;
3292 		}
3293 		if (banks[1] & MCI_STATUS_VAL)
3294 			mce->status |= MCI_STATUS_OVER;
3295 		banks[2] = mce->addr;
3296 		banks[3] = mce->misc;
3297 		vcpu->arch.mcg_status = mce->mcg_status;
3298 		banks[1] = mce->status;
3299 		kvm_queue_exception(vcpu, MC_VECTOR);
3300 	} else if (!(banks[1] & MCI_STATUS_VAL)
3301 		   || !(banks[1] & MCI_STATUS_UC)) {
3302 		if (banks[1] & MCI_STATUS_VAL)
3303 			mce->status |= MCI_STATUS_OVER;
3304 		banks[2] = mce->addr;
3305 		banks[3] = mce->misc;
3306 		banks[1] = mce->status;
3307 	} else
3308 		banks[1] |= MCI_STATUS_OVER;
3309 	return 0;
3310 }
3311 
3312 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3313 					       struct kvm_vcpu_events *events)
3314 {
3315 	process_nmi(vcpu);
3316 	/*
3317 	 * FIXME: pass injected and pending separately.  This is only
3318 	 * needed for nested virtualization, whose state cannot be
3319 	 * migrated yet.  For now we can combine them.
3320 	 */
3321 	events->exception.injected =
3322 		(vcpu->arch.exception.pending ||
3323 		 vcpu->arch.exception.injected) &&
3324 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3325 	events->exception.nr = vcpu->arch.exception.nr;
3326 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3327 	events->exception.pad = 0;
3328 	events->exception.error_code = vcpu->arch.exception.error_code;
3329 
3330 	events->interrupt.injected =
3331 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3332 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3333 	events->interrupt.soft = 0;
3334 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3335 
3336 	events->nmi.injected = vcpu->arch.nmi_injected;
3337 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3338 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3339 	events->nmi.pad = 0;
3340 
3341 	events->sipi_vector = 0; /* never valid when reporting to user space */
3342 
3343 	events->smi.smm = is_smm(vcpu);
3344 	events->smi.pending = vcpu->arch.smi_pending;
3345 	events->smi.smm_inside_nmi =
3346 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3347 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3348 
3349 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3350 			 | KVM_VCPUEVENT_VALID_SHADOW
3351 			 | KVM_VCPUEVENT_VALID_SMM);
3352 	memset(&events->reserved, 0, sizeof(events->reserved));
3353 }
3354 
3355 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3356 
3357 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3358 					      struct kvm_vcpu_events *events)
3359 {
3360 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3361 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3362 			      | KVM_VCPUEVENT_VALID_SHADOW
3363 			      | KVM_VCPUEVENT_VALID_SMM))
3364 		return -EINVAL;
3365 
3366 	if (events->exception.injected &&
3367 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3368 	     is_guest_mode(vcpu)))
3369 		return -EINVAL;
3370 
3371 	/* INITs are latched while in SMM */
3372 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3373 	    (events->smi.smm || events->smi.pending) &&
3374 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3375 		return -EINVAL;
3376 
3377 	process_nmi(vcpu);
3378 	vcpu->arch.exception.injected = false;
3379 	vcpu->arch.exception.pending = events->exception.injected;
3380 	vcpu->arch.exception.nr = events->exception.nr;
3381 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3382 	vcpu->arch.exception.error_code = events->exception.error_code;
3383 
3384 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3385 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3386 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3387 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3388 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3389 						  events->interrupt.shadow);
3390 
3391 	vcpu->arch.nmi_injected = events->nmi.injected;
3392 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3393 		vcpu->arch.nmi_pending = events->nmi.pending;
3394 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3395 
3396 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3397 	    lapic_in_kernel(vcpu))
3398 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3399 
3400 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3401 		u32 hflags = vcpu->arch.hflags;
3402 		if (events->smi.smm)
3403 			hflags |= HF_SMM_MASK;
3404 		else
3405 			hflags &= ~HF_SMM_MASK;
3406 		kvm_set_hflags(vcpu, hflags);
3407 
3408 		vcpu->arch.smi_pending = events->smi.pending;
3409 
3410 		if (events->smi.smm) {
3411 			if (events->smi.smm_inside_nmi)
3412 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3413 			else
3414 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3415 			if (lapic_in_kernel(vcpu)) {
3416 				if (events->smi.latched_init)
3417 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3418 				else
3419 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3420 			}
3421 		}
3422 	}
3423 
3424 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3425 
3426 	return 0;
3427 }
3428 
3429 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3430 					     struct kvm_debugregs *dbgregs)
3431 {
3432 	unsigned long val;
3433 
3434 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3435 	kvm_get_dr(vcpu, 6, &val);
3436 	dbgregs->dr6 = val;
3437 	dbgregs->dr7 = vcpu->arch.dr7;
3438 	dbgregs->flags = 0;
3439 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3440 }
3441 
3442 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3443 					    struct kvm_debugregs *dbgregs)
3444 {
3445 	if (dbgregs->flags)
3446 		return -EINVAL;
3447 
3448 	if (dbgregs->dr6 & ~0xffffffffull)
3449 		return -EINVAL;
3450 	if (dbgregs->dr7 & ~0xffffffffull)
3451 		return -EINVAL;
3452 
3453 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3454 	kvm_update_dr0123(vcpu);
3455 	vcpu->arch.dr6 = dbgregs->dr6;
3456 	kvm_update_dr6(vcpu);
3457 	vcpu->arch.dr7 = dbgregs->dr7;
3458 	kvm_update_dr7(vcpu);
3459 
3460 	return 0;
3461 }
3462 
3463 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3464 
3465 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3466 {
3467 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3468 	u64 xstate_bv = xsave->header.xfeatures;
3469 	u64 valid;
3470 
3471 	/*
3472 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3473 	 * leaves 0 and 1 in the loop below.
3474 	 */
3475 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3476 
3477 	/* Set XSTATE_BV */
3478 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3479 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3480 
3481 	/*
3482 	 * Copy each region from the possibly compacted offset to the
3483 	 * non-compacted offset.
3484 	 */
3485 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3486 	while (valid) {
3487 		u64 feature = valid & -valid;
3488 		int index = fls64(feature) - 1;
3489 		void *src = get_xsave_addr(xsave, feature);
3490 
3491 		if (src) {
3492 			u32 size, offset, ecx, edx;
3493 			cpuid_count(XSTATE_CPUID, index,
3494 				    &size, &offset, &ecx, &edx);
3495 			if (feature == XFEATURE_MASK_PKRU)
3496 				memcpy(dest + offset, &vcpu->arch.pkru,
3497 				       sizeof(vcpu->arch.pkru));
3498 			else
3499 				memcpy(dest + offset, src, size);
3500 
3501 		}
3502 
3503 		valid -= feature;
3504 	}
3505 }
3506 
3507 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3508 {
3509 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3510 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3511 	u64 valid;
3512 
3513 	/*
3514 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3515 	 * leaves 0 and 1 in the loop below.
3516 	 */
3517 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3518 
3519 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3520 	xsave->header.xfeatures = xstate_bv;
3521 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3522 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3523 
3524 	/*
3525 	 * Copy each region from the non-compacted offset to the
3526 	 * possibly compacted offset.
3527 	 */
3528 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3529 	while (valid) {
3530 		u64 feature = valid & -valid;
3531 		int index = fls64(feature) - 1;
3532 		void *dest = get_xsave_addr(xsave, feature);
3533 
3534 		if (dest) {
3535 			u32 size, offset, ecx, edx;
3536 			cpuid_count(XSTATE_CPUID, index,
3537 				    &size, &offset, &ecx, &edx);
3538 			if (feature == XFEATURE_MASK_PKRU)
3539 				memcpy(&vcpu->arch.pkru, src + offset,
3540 				       sizeof(vcpu->arch.pkru));
3541 			else
3542 				memcpy(dest, src + offset, size);
3543 		}
3544 
3545 		valid -= feature;
3546 	}
3547 }
3548 
3549 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3550 					 struct kvm_xsave *guest_xsave)
3551 {
3552 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3553 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3554 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3555 	} else {
3556 		memcpy(guest_xsave->region,
3557 			&vcpu->arch.guest_fpu.state.fxsave,
3558 			sizeof(struct fxregs_state));
3559 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3560 			XFEATURE_MASK_FPSSE;
3561 	}
3562 }
3563 
3564 #define XSAVE_MXCSR_OFFSET 24
3565 
3566 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3567 					struct kvm_xsave *guest_xsave)
3568 {
3569 	u64 xstate_bv =
3570 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3571 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3572 
3573 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3574 		/*
3575 		 * Here we allow setting states that are not present in
3576 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3577 		 * with old userspace.
3578 		 */
3579 		if (xstate_bv & ~kvm_supported_xcr0() ||
3580 			mxcsr & ~mxcsr_feature_mask)
3581 			return -EINVAL;
3582 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3583 	} else {
3584 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3585 			mxcsr & ~mxcsr_feature_mask)
3586 			return -EINVAL;
3587 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3588 			guest_xsave->region, sizeof(struct fxregs_state));
3589 	}
3590 	return 0;
3591 }
3592 
3593 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3594 					struct kvm_xcrs *guest_xcrs)
3595 {
3596 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3597 		guest_xcrs->nr_xcrs = 0;
3598 		return;
3599 	}
3600 
3601 	guest_xcrs->nr_xcrs = 1;
3602 	guest_xcrs->flags = 0;
3603 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3604 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3605 }
3606 
3607 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3608 				       struct kvm_xcrs *guest_xcrs)
3609 {
3610 	int i, r = 0;
3611 
3612 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3613 		return -EINVAL;
3614 
3615 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3616 		return -EINVAL;
3617 
3618 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3619 		/* Only support XCR0 currently */
3620 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3621 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3622 				guest_xcrs->xcrs[i].value);
3623 			break;
3624 		}
3625 	if (r)
3626 		r = -EINVAL;
3627 	return r;
3628 }
3629 
3630 /*
3631  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3632  * stopped by the hypervisor.  This function will be called from the host only.
3633  * EINVAL is returned when the host attempts to set the flag for a guest that
3634  * does not support pv clocks.
3635  */
3636 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3637 {
3638 	if (!vcpu->arch.pv_time_enabled)
3639 		return -EINVAL;
3640 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3641 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3642 	return 0;
3643 }
3644 
3645 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3646 				     struct kvm_enable_cap *cap)
3647 {
3648 	if (cap->flags)
3649 		return -EINVAL;
3650 
3651 	switch (cap->cap) {
3652 	case KVM_CAP_HYPERV_SYNIC2:
3653 		if (cap->args[0])
3654 			return -EINVAL;
3655 	case KVM_CAP_HYPERV_SYNIC:
3656 		if (!irqchip_in_kernel(vcpu->kvm))
3657 			return -EINVAL;
3658 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3659 					     KVM_CAP_HYPERV_SYNIC2);
3660 	default:
3661 		return -EINVAL;
3662 	}
3663 }
3664 
3665 long kvm_arch_vcpu_ioctl(struct file *filp,
3666 			 unsigned int ioctl, unsigned long arg)
3667 {
3668 	struct kvm_vcpu *vcpu = filp->private_data;
3669 	void __user *argp = (void __user *)arg;
3670 	int r;
3671 	union {
3672 		struct kvm_lapic_state *lapic;
3673 		struct kvm_xsave *xsave;
3674 		struct kvm_xcrs *xcrs;
3675 		void *buffer;
3676 	} u;
3677 
3678 	vcpu_load(vcpu);
3679 
3680 	u.buffer = NULL;
3681 	switch (ioctl) {
3682 	case KVM_GET_LAPIC: {
3683 		r = -EINVAL;
3684 		if (!lapic_in_kernel(vcpu))
3685 			goto out;
3686 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3687 
3688 		r = -ENOMEM;
3689 		if (!u.lapic)
3690 			goto out;
3691 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3692 		if (r)
3693 			goto out;
3694 		r = -EFAULT;
3695 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3696 			goto out;
3697 		r = 0;
3698 		break;
3699 	}
3700 	case KVM_SET_LAPIC: {
3701 		r = -EINVAL;
3702 		if (!lapic_in_kernel(vcpu))
3703 			goto out;
3704 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3705 		if (IS_ERR(u.lapic)) {
3706 			r = PTR_ERR(u.lapic);
3707 			goto out_nofree;
3708 		}
3709 
3710 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3711 		break;
3712 	}
3713 	case KVM_INTERRUPT: {
3714 		struct kvm_interrupt irq;
3715 
3716 		r = -EFAULT;
3717 		if (copy_from_user(&irq, argp, sizeof irq))
3718 			goto out;
3719 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3720 		break;
3721 	}
3722 	case KVM_NMI: {
3723 		r = kvm_vcpu_ioctl_nmi(vcpu);
3724 		break;
3725 	}
3726 	case KVM_SMI: {
3727 		r = kvm_vcpu_ioctl_smi(vcpu);
3728 		break;
3729 	}
3730 	case KVM_SET_CPUID: {
3731 		struct kvm_cpuid __user *cpuid_arg = argp;
3732 		struct kvm_cpuid cpuid;
3733 
3734 		r = -EFAULT;
3735 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3736 			goto out;
3737 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3738 		break;
3739 	}
3740 	case KVM_SET_CPUID2: {
3741 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3742 		struct kvm_cpuid2 cpuid;
3743 
3744 		r = -EFAULT;
3745 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3746 			goto out;
3747 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3748 					      cpuid_arg->entries);
3749 		break;
3750 	}
3751 	case KVM_GET_CPUID2: {
3752 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3753 		struct kvm_cpuid2 cpuid;
3754 
3755 		r = -EFAULT;
3756 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3757 			goto out;
3758 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3759 					      cpuid_arg->entries);
3760 		if (r)
3761 			goto out;
3762 		r = -EFAULT;
3763 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3764 			goto out;
3765 		r = 0;
3766 		break;
3767 	}
3768 	case KVM_GET_MSRS: {
3769 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3770 		r = msr_io(vcpu, argp, do_get_msr, 1);
3771 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3772 		break;
3773 	}
3774 	case KVM_SET_MSRS: {
3775 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3776 		r = msr_io(vcpu, argp, do_set_msr, 0);
3777 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3778 		break;
3779 	}
3780 	case KVM_TPR_ACCESS_REPORTING: {
3781 		struct kvm_tpr_access_ctl tac;
3782 
3783 		r = -EFAULT;
3784 		if (copy_from_user(&tac, argp, sizeof tac))
3785 			goto out;
3786 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3787 		if (r)
3788 			goto out;
3789 		r = -EFAULT;
3790 		if (copy_to_user(argp, &tac, sizeof tac))
3791 			goto out;
3792 		r = 0;
3793 		break;
3794 	};
3795 	case KVM_SET_VAPIC_ADDR: {
3796 		struct kvm_vapic_addr va;
3797 		int idx;
3798 
3799 		r = -EINVAL;
3800 		if (!lapic_in_kernel(vcpu))
3801 			goto out;
3802 		r = -EFAULT;
3803 		if (copy_from_user(&va, argp, sizeof va))
3804 			goto out;
3805 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3806 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3807 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3808 		break;
3809 	}
3810 	case KVM_X86_SETUP_MCE: {
3811 		u64 mcg_cap;
3812 
3813 		r = -EFAULT;
3814 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3815 			goto out;
3816 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3817 		break;
3818 	}
3819 	case KVM_X86_SET_MCE: {
3820 		struct kvm_x86_mce mce;
3821 
3822 		r = -EFAULT;
3823 		if (copy_from_user(&mce, argp, sizeof mce))
3824 			goto out;
3825 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3826 		break;
3827 	}
3828 	case KVM_GET_VCPU_EVENTS: {
3829 		struct kvm_vcpu_events events;
3830 
3831 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3832 
3833 		r = -EFAULT;
3834 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3835 			break;
3836 		r = 0;
3837 		break;
3838 	}
3839 	case KVM_SET_VCPU_EVENTS: {
3840 		struct kvm_vcpu_events events;
3841 
3842 		r = -EFAULT;
3843 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3844 			break;
3845 
3846 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3847 		break;
3848 	}
3849 	case KVM_GET_DEBUGREGS: {
3850 		struct kvm_debugregs dbgregs;
3851 
3852 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3853 
3854 		r = -EFAULT;
3855 		if (copy_to_user(argp, &dbgregs,
3856 				 sizeof(struct kvm_debugregs)))
3857 			break;
3858 		r = 0;
3859 		break;
3860 	}
3861 	case KVM_SET_DEBUGREGS: {
3862 		struct kvm_debugregs dbgregs;
3863 
3864 		r = -EFAULT;
3865 		if (copy_from_user(&dbgregs, argp,
3866 				   sizeof(struct kvm_debugregs)))
3867 			break;
3868 
3869 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3870 		break;
3871 	}
3872 	case KVM_GET_XSAVE: {
3873 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3874 		r = -ENOMEM;
3875 		if (!u.xsave)
3876 			break;
3877 
3878 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3879 
3880 		r = -EFAULT;
3881 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3882 			break;
3883 		r = 0;
3884 		break;
3885 	}
3886 	case KVM_SET_XSAVE: {
3887 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3888 		if (IS_ERR(u.xsave)) {
3889 			r = PTR_ERR(u.xsave);
3890 			goto out_nofree;
3891 		}
3892 
3893 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3894 		break;
3895 	}
3896 	case KVM_GET_XCRS: {
3897 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3898 		r = -ENOMEM;
3899 		if (!u.xcrs)
3900 			break;
3901 
3902 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3903 
3904 		r = -EFAULT;
3905 		if (copy_to_user(argp, u.xcrs,
3906 				 sizeof(struct kvm_xcrs)))
3907 			break;
3908 		r = 0;
3909 		break;
3910 	}
3911 	case KVM_SET_XCRS: {
3912 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3913 		if (IS_ERR(u.xcrs)) {
3914 			r = PTR_ERR(u.xcrs);
3915 			goto out_nofree;
3916 		}
3917 
3918 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3919 		break;
3920 	}
3921 	case KVM_SET_TSC_KHZ: {
3922 		u32 user_tsc_khz;
3923 
3924 		r = -EINVAL;
3925 		user_tsc_khz = (u32)arg;
3926 
3927 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3928 			goto out;
3929 
3930 		if (user_tsc_khz == 0)
3931 			user_tsc_khz = tsc_khz;
3932 
3933 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3934 			r = 0;
3935 
3936 		goto out;
3937 	}
3938 	case KVM_GET_TSC_KHZ: {
3939 		r = vcpu->arch.virtual_tsc_khz;
3940 		goto out;
3941 	}
3942 	case KVM_KVMCLOCK_CTRL: {
3943 		r = kvm_set_guest_paused(vcpu);
3944 		goto out;
3945 	}
3946 	case KVM_ENABLE_CAP: {
3947 		struct kvm_enable_cap cap;
3948 
3949 		r = -EFAULT;
3950 		if (copy_from_user(&cap, argp, sizeof(cap)))
3951 			goto out;
3952 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3953 		break;
3954 	}
3955 	default:
3956 		r = -EINVAL;
3957 	}
3958 out:
3959 	kfree(u.buffer);
3960 out_nofree:
3961 	vcpu_put(vcpu);
3962 	return r;
3963 }
3964 
3965 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3966 {
3967 	return VM_FAULT_SIGBUS;
3968 }
3969 
3970 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3971 {
3972 	int ret;
3973 
3974 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3975 		return -EINVAL;
3976 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3977 	return ret;
3978 }
3979 
3980 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3981 					      u64 ident_addr)
3982 {
3983 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
3984 }
3985 
3986 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3987 					  u32 kvm_nr_mmu_pages)
3988 {
3989 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3990 		return -EINVAL;
3991 
3992 	mutex_lock(&kvm->slots_lock);
3993 
3994 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3995 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3996 
3997 	mutex_unlock(&kvm->slots_lock);
3998 	return 0;
3999 }
4000 
4001 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4002 {
4003 	return kvm->arch.n_max_mmu_pages;
4004 }
4005 
4006 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4007 {
4008 	struct kvm_pic *pic = kvm->arch.vpic;
4009 	int r;
4010 
4011 	r = 0;
4012 	switch (chip->chip_id) {
4013 	case KVM_IRQCHIP_PIC_MASTER:
4014 		memcpy(&chip->chip.pic, &pic->pics[0],
4015 			sizeof(struct kvm_pic_state));
4016 		break;
4017 	case KVM_IRQCHIP_PIC_SLAVE:
4018 		memcpy(&chip->chip.pic, &pic->pics[1],
4019 			sizeof(struct kvm_pic_state));
4020 		break;
4021 	case KVM_IRQCHIP_IOAPIC:
4022 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4023 		break;
4024 	default:
4025 		r = -EINVAL;
4026 		break;
4027 	}
4028 	return r;
4029 }
4030 
4031 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4032 {
4033 	struct kvm_pic *pic = kvm->arch.vpic;
4034 	int r;
4035 
4036 	r = 0;
4037 	switch (chip->chip_id) {
4038 	case KVM_IRQCHIP_PIC_MASTER:
4039 		spin_lock(&pic->lock);
4040 		memcpy(&pic->pics[0], &chip->chip.pic,
4041 			sizeof(struct kvm_pic_state));
4042 		spin_unlock(&pic->lock);
4043 		break;
4044 	case KVM_IRQCHIP_PIC_SLAVE:
4045 		spin_lock(&pic->lock);
4046 		memcpy(&pic->pics[1], &chip->chip.pic,
4047 			sizeof(struct kvm_pic_state));
4048 		spin_unlock(&pic->lock);
4049 		break;
4050 	case KVM_IRQCHIP_IOAPIC:
4051 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4052 		break;
4053 	default:
4054 		r = -EINVAL;
4055 		break;
4056 	}
4057 	kvm_pic_update_irq(pic);
4058 	return r;
4059 }
4060 
4061 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4062 {
4063 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4064 
4065 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4066 
4067 	mutex_lock(&kps->lock);
4068 	memcpy(ps, &kps->channels, sizeof(*ps));
4069 	mutex_unlock(&kps->lock);
4070 	return 0;
4071 }
4072 
4073 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4074 {
4075 	int i;
4076 	struct kvm_pit *pit = kvm->arch.vpit;
4077 
4078 	mutex_lock(&pit->pit_state.lock);
4079 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4080 	for (i = 0; i < 3; i++)
4081 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4082 	mutex_unlock(&pit->pit_state.lock);
4083 	return 0;
4084 }
4085 
4086 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4087 {
4088 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4089 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4090 		sizeof(ps->channels));
4091 	ps->flags = kvm->arch.vpit->pit_state.flags;
4092 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4093 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4094 	return 0;
4095 }
4096 
4097 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4098 {
4099 	int start = 0;
4100 	int i;
4101 	u32 prev_legacy, cur_legacy;
4102 	struct kvm_pit *pit = kvm->arch.vpit;
4103 
4104 	mutex_lock(&pit->pit_state.lock);
4105 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4106 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4107 	if (!prev_legacy && cur_legacy)
4108 		start = 1;
4109 	memcpy(&pit->pit_state.channels, &ps->channels,
4110 	       sizeof(pit->pit_state.channels));
4111 	pit->pit_state.flags = ps->flags;
4112 	for (i = 0; i < 3; i++)
4113 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4114 				   start && i == 0);
4115 	mutex_unlock(&pit->pit_state.lock);
4116 	return 0;
4117 }
4118 
4119 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4120 				 struct kvm_reinject_control *control)
4121 {
4122 	struct kvm_pit *pit = kvm->arch.vpit;
4123 
4124 	if (!pit)
4125 		return -ENXIO;
4126 
4127 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4128 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4129 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4130 	 */
4131 	mutex_lock(&pit->pit_state.lock);
4132 	kvm_pit_set_reinject(pit, control->pit_reinject);
4133 	mutex_unlock(&pit->pit_state.lock);
4134 
4135 	return 0;
4136 }
4137 
4138 /**
4139  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4140  * @kvm: kvm instance
4141  * @log: slot id and address to which we copy the log
4142  *
4143  * Steps 1-4 below provide general overview of dirty page logging. See
4144  * kvm_get_dirty_log_protect() function description for additional details.
4145  *
4146  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4147  * always flush the TLB (step 4) even if previous step failed  and the dirty
4148  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4149  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4150  * writes will be marked dirty for next log read.
4151  *
4152  *   1. Take a snapshot of the bit and clear it if needed.
4153  *   2. Write protect the corresponding page.
4154  *   3. Copy the snapshot to the userspace.
4155  *   4. Flush TLB's if needed.
4156  */
4157 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4158 {
4159 	bool is_dirty = false;
4160 	int r;
4161 
4162 	mutex_lock(&kvm->slots_lock);
4163 
4164 	/*
4165 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4166 	 */
4167 	if (kvm_x86_ops->flush_log_dirty)
4168 		kvm_x86_ops->flush_log_dirty(kvm);
4169 
4170 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4171 
4172 	/*
4173 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4174 	 * kvm_mmu_slot_remove_write_access().
4175 	 */
4176 	lockdep_assert_held(&kvm->slots_lock);
4177 	if (is_dirty)
4178 		kvm_flush_remote_tlbs(kvm);
4179 
4180 	mutex_unlock(&kvm->slots_lock);
4181 	return r;
4182 }
4183 
4184 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4185 			bool line_status)
4186 {
4187 	if (!irqchip_in_kernel(kvm))
4188 		return -ENXIO;
4189 
4190 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4191 					irq_event->irq, irq_event->level,
4192 					line_status);
4193 	return 0;
4194 }
4195 
4196 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4197 				   struct kvm_enable_cap *cap)
4198 {
4199 	int r;
4200 
4201 	if (cap->flags)
4202 		return -EINVAL;
4203 
4204 	switch (cap->cap) {
4205 	case KVM_CAP_DISABLE_QUIRKS:
4206 		kvm->arch.disabled_quirks = cap->args[0];
4207 		r = 0;
4208 		break;
4209 	case KVM_CAP_SPLIT_IRQCHIP: {
4210 		mutex_lock(&kvm->lock);
4211 		r = -EINVAL;
4212 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4213 			goto split_irqchip_unlock;
4214 		r = -EEXIST;
4215 		if (irqchip_in_kernel(kvm))
4216 			goto split_irqchip_unlock;
4217 		if (kvm->created_vcpus)
4218 			goto split_irqchip_unlock;
4219 		r = kvm_setup_empty_irq_routing(kvm);
4220 		if (r)
4221 			goto split_irqchip_unlock;
4222 		/* Pairs with irqchip_in_kernel. */
4223 		smp_wmb();
4224 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4225 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4226 		r = 0;
4227 split_irqchip_unlock:
4228 		mutex_unlock(&kvm->lock);
4229 		break;
4230 	}
4231 	case KVM_CAP_X2APIC_API:
4232 		r = -EINVAL;
4233 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4234 			break;
4235 
4236 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4237 			kvm->arch.x2apic_format = true;
4238 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4239 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4240 
4241 		r = 0;
4242 		break;
4243 	case KVM_CAP_X86_DISABLE_EXITS:
4244 		r = -EINVAL;
4245 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4246 			break;
4247 
4248 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4249 			kvm_can_mwait_in_guest())
4250 			kvm->arch.mwait_in_guest = true;
4251 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
4252 			kvm->arch.hlt_in_guest = true;
4253 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4254 			kvm->arch.pause_in_guest = true;
4255 		r = 0;
4256 		break;
4257 	default:
4258 		r = -EINVAL;
4259 		break;
4260 	}
4261 	return r;
4262 }
4263 
4264 long kvm_arch_vm_ioctl(struct file *filp,
4265 		       unsigned int ioctl, unsigned long arg)
4266 {
4267 	struct kvm *kvm = filp->private_data;
4268 	void __user *argp = (void __user *)arg;
4269 	int r = -ENOTTY;
4270 	/*
4271 	 * This union makes it completely explicit to gcc-3.x
4272 	 * that these two variables' stack usage should be
4273 	 * combined, not added together.
4274 	 */
4275 	union {
4276 		struct kvm_pit_state ps;
4277 		struct kvm_pit_state2 ps2;
4278 		struct kvm_pit_config pit_config;
4279 	} u;
4280 
4281 	switch (ioctl) {
4282 	case KVM_SET_TSS_ADDR:
4283 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4284 		break;
4285 	case KVM_SET_IDENTITY_MAP_ADDR: {
4286 		u64 ident_addr;
4287 
4288 		mutex_lock(&kvm->lock);
4289 		r = -EINVAL;
4290 		if (kvm->created_vcpus)
4291 			goto set_identity_unlock;
4292 		r = -EFAULT;
4293 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4294 			goto set_identity_unlock;
4295 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4296 set_identity_unlock:
4297 		mutex_unlock(&kvm->lock);
4298 		break;
4299 	}
4300 	case KVM_SET_NR_MMU_PAGES:
4301 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4302 		break;
4303 	case KVM_GET_NR_MMU_PAGES:
4304 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4305 		break;
4306 	case KVM_CREATE_IRQCHIP: {
4307 		mutex_lock(&kvm->lock);
4308 
4309 		r = -EEXIST;
4310 		if (irqchip_in_kernel(kvm))
4311 			goto create_irqchip_unlock;
4312 
4313 		r = -EINVAL;
4314 		if (kvm->created_vcpus)
4315 			goto create_irqchip_unlock;
4316 
4317 		r = kvm_pic_init(kvm);
4318 		if (r)
4319 			goto create_irqchip_unlock;
4320 
4321 		r = kvm_ioapic_init(kvm);
4322 		if (r) {
4323 			kvm_pic_destroy(kvm);
4324 			goto create_irqchip_unlock;
4325 		}
4326 
4327 		r = kvm_setup_default_irq_routing(kvm);
4328 		if (r) {
4329 			kvm_ioapic_destroy(kvm);
4330 			kvm_pic_destroy(kvm);
4331 			goto create_irqchip_unlock;
4332 		}
4333 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4334 		smp_wmb();
4335 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4336 	create_irqchip_unlock:
4337 		mutex_unlock(&kvm->lock);
4338 		break;
4339 	}
4340 	case KVM_CREATE_PIT:
4341 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4342 		goto create_pit;
4343 	case KVM_CREATE_PIT2:
4344 		r = -EFAULT;
4345 		if (copy_from_user(&u.pit_config, argp,
4346 				   sizeof(struct kvm_pit_config)))
4347 			goto out;
4348 	create_pit:
4349 		mutex_lock(&kvm->lock);
4350 		r = -EEXIST;
4351 		if (kvm->arch.vpit)
4352 			goto create_pit_unlock;
4353 		r = -ENOMEM;
4354 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4355 		if (kvm->arch.vpit)
4356 			r = 0;
4357 	create_pit_unlock:
4358 		mutex_unlock(&kvm->lock);
4359 		break;
4360 	case KVM_GET_IRQCHIP: {
4361 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4362 		struct kvm_irqchip *chip;
4363 
4364 		chip = memdup_user(argp, sizeof(*chip));
4365 		if (IS_ERR(chip)) {
4366 			r = PTR_ERR(chip);
4367 			goto out;
4368 		}
4369 
4370 		r = -ENXIO;
4371 		if (!irqchip_kernel(kvm))
4372 			goto get_irqchip_out;
4373 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4374 		if (r)
4375 			goto get_irqchip_out;
4376 		r = -EFAULT;
4377 		if (copy_to_user(argp, chip, sizeof *chip))
4378 			goto get_irqchip_out;
4379 		r = 0;
4380 	get_irqchip_out:
4381 		kfree(chip);
4382 		break;
4383 	}
4384 	case KVM_SET_IRQCHIP: {
4385 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4386 		struct kvm_irqchip *chip;
4387 
4388 		chip = memdup_user(argp, sizeof(*chip));
4389 		if (IS_ERR(chip)) {
4390 			r = PTR_ERR(chip);
4391 			goto out;
4392 		}
4393 
4394 		r = -ENXIO;
4395 		if (!irqchip_kernel(kvm))
4396 			goto set_irqchip_out;
4397 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4398 		if (r)
4399 			goto set_irqchip_out;
4400 		r = 0;
4401 	set_irqchip_out:
4402 		kfree(chip);
4403 		break;
4404 	}
4405 	case KVM_GET_PIT: {
4406 		r = -EFAULT;
4407 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4408 			goto out;
4409 		r = -ENXIO;
4410 		if (!kvm->arch.vpit)
4411 			goto out;
4412 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4413 		if (r)
4414 			goto out;
4415 		r = -EFAULT;
4416 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4417 			goto out;
4418 		r = 0;
4419 		break;
4420 	}
4421 	case KVM_SET_PIT: {
4422 		r = -EFAULT;
4423 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4424 			goto out;
4425 		r = -ENXIO;
4426 		if (!kvm->arch.vpit)
4427 			goto out;
4428 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4429 		break;
4430 	}
4431 	case KVM_GET_PIT2: {
4432 		r = -ENXIO;
4433 		if (!kvm->arch.vpit)
4434 			goto out;
4435 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4436 		if (r)
4437 			goto out;
4438 		r = -EFAULT;
4439 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4440 			goto out;
4441 		r = 0;
4442 		break;
4443 	}
4444 	case KVM_SET_PIT2: {
4445 		r = -EFAULT;
4446 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4447 			goto out;
4448 		r = -ENXIO;
4449 		if (!kvm->arch.vpit)
4450 			goto out;
4451 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4452 		break;
4453 	}
4454 	case KVM_REINJECT_CONTROL: {
4455 		struct kvm_reinject_control control;
4456 		r =  -EFAULT;
4457 		if (copy_from_user(&control, argp, sizeof(control)))
4458 			goto out;
4459 		r = kvm_vm_ioctl_reinject(kvm, &control);
4460 		break;
4461 	}
4462 	case KVM_SET_BOOT_CPU_ID:
4463 		r = 0;
4464 		mutex_lock(&kvm->lock);
4465 		if (kvm->created_vcpus)
4466 			r = -EBUSY;
4467 		else
4468 			kvm->arch.bsp_vcpu_id = arg;
4469 		mutex_unlock(&kvm->lock);
4470 		break;
4471 	case KVM_XEN_HVM_CONFIG: {
4472 		struct kvm_xen_hvm_config xhc;
4473 		r = -EFAULT;
4474 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4475 			goto out;
4476 		r = -EINVAL;
4477 		if (xhc.flags)
4478 			goto out;
4479 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4480 		r = 0;
4481 		break;
4482 	}
4483 	case KVM_SET_CLOCK: {
4484 		struct kvm_clock_data user_ns;
4485 		u64 now_ns;
4486 
4487 		r = -EFAULT;
4488 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4489 			goto out;
4490 
4491 		r = -EINVAL;
4492 		if (user_ns.flags)
4493 			goto out;
4494 
4495 		r = 0;
4496 		/*
4497 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4498 		 * kvm_gen_update_masterclock() can be cut down to locked
4499 		 * pvclock_update_vm_gtod_copy().
4500 		 */
4501 		kvm_gen_update_masterclock(kvm);
4502 		now_ns = get_kvmclock_ns(kvm);
4503 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4504 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4505 		break;
4506 	}
4507 	case KVM_GET_CLOCK: {
4508 		struct kvm_clock_data user_ns;
4509 		u64 now_ns;
4510 
4511 		now_ns = get_kvmclock_ns(kvm);
4512 		user_ns.clock = now_ns;
4513 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4514 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4515 
4516 		r = -EFAULT;
4517 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4518 			goto out;
4519 		r = 0;
4520 		break;
4521 	}
4522 	case KVM_ENABLE_CAP: {
4523 		struct kvm_enable_cap cap;
4524 
4525 		r = -EFAULT;
4526 		if (copy_from_user(&cap, argp, sizeof(cap)))
4527 			goto out;
4528 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4529 		break;
4530 	}
4531 	case KVM_MEMORY_ENCRYPT_OP: {
4532 		r = -ENOTTY;
4533 		if (kvm_x86_ops->mem_enc_op)
4534 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4535 		break;
4536 	}
4537 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4538 		struct kvm_enc_region region;
4539 
4540 		r = -EFAULT;
4541 		if (copy_from_user(&region, argp, sizeof(region)))
4542 			goto out;
4543 
4544 		r = -ENOTTY;
4545 		if (kvm_x86_ops->mem_enc_reg_region)
4546 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4547 		break;
4548 	}
4549 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4550 		struct kvm_enc_region region;
4551 
4552 		r = -EFAULT;
4553 		if (copy_from_user(&region, argp, sizeof(region)))
4554 			goto out;
4555 
4556 		r = -ENOTTY;
4557 		if (kvm_x86_ops->mem_enc_unreg_region)
4558 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4559 		break;
4560 	}
4561 	case KVM_HYPERV_EVENTFD: {
4562 		struct kvm_hyperv_eventfd hvevfd;
4563 
4564 		r = -EFAULT;
4565 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4566 			goto out;
4567 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4568 		break;
4569 	}
4570 	default:
4571 		r = -ENOTTY;
4572 	}
4573 out:
4574 	return r;
4575 }
4576 
4577 static void kvm_init_msr_list(void)
4578 {
4579 	u32 dummy[2];
4580 	unsigned i, j;
4581 
4582 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4583 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4584 			continue;
4585 
4586 		/*
4587 		 * Even MSRs that are valid in the host may not be exposed
4588 		 * to the guests in some cases.
4589 		 */
4590 		switch (msrs_to_save[i]) {
4591 		case MSR_IA32_BNDCFGS:
4592 			if (!kvm_x86_ops->mpx_supported())
4593 				continue;
4594 			break;
4595 		case MSR_TSC_AUX:
4596 			if (!kvm_x86_ops->rdtscp_supported())
4597 				continue;
4598 			break;
4599 		default:
4600 			break;
4601 		}
4602 
4603 		if (j < i)
4604 			msrs_to_save[j] = msrs_to_save[i];
4605 		j++;
4606 	}
4607 	num_msrs_to_save = j;
4608 
4609 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4610 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4611 			continue;
4612 
4613 		if (j < i)
4614 			emulated_msrs[j] = emulated_msrs[i];
4615 		j++;
4616 	}
4617 	num_emulated_msrs = j;
4618 
4619 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4620 		struct kvm_msr_entry msr;
4621 
4622 		msr.index = msr_based_features[i];
4623 		if (kvm_get_msr_feature(&msr))
4624 			continue;
4625 
4626 		if (j < i)
4627 			msr_based_features[j] = msr_based_features[i];
4628 		j++;
4629 	}
4630 	num_msr_based_features = j;
4631 }
4632 
4633 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4634 			   const void *v)
4635 {
4636 	int handled = 0;
4637 	int n;
4638 
4639 	do {
4640 		n = min(len, 8);
4641 		if (!(lapic_in_kernel(vcpu) &&
4642 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4643 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4644 			break;
4645 		handled += n;
4646 		addr += n;
4647 		len -= n;
4648 		v += n;
4649 	} while (len);
4650 
4651 	return handled;
4652 }
4653 
4654 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4655 {
4656 	int handled = 0;
4657 	int n;
4658 
4659 	do {
4660 		n = min(len, 8);
4661 		if (!(lapic_in_kernel(vcpu) &&
4662 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4663 					 addr, n, v))
4664 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4665 			break;
4666 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4667 		handled += n;
4668 		addr += n;
4669 		len -= n;
4670 		v += n;
4671 	} while (len);
4672 
4673 	return handled;
4674 }
4675 
4676 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4677 			struct kvm_segment *var, int seg)
4678 {
4679 	kvm_x86_ops->set_segment(vcpu, var, seg);
4680 }
4681 
4682 void kvm_get_segment(struct kvm_vcpu *vcpu,
4683 		     struct kvm_segment *var, int seg)
4684 {
4685 	kvm_x86_ops->get_segment(vcpu, var, seg);
4686 }
4687 
4688 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4689 			   struct x86_exception *exception)
4690 {
4691 	gpa_t t_gpa;
4692 
4693 	BUG_ON(!mmu_is_nested(vcpu));
4694 
4695 	/* NPT walks are always user-walks */
4696 	access |= PFERR_USER_MASK;
4697 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4698 
4699 	return t_gpa;
4700 }
4701 
4702 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4703 			      struct x86_exception *exception)
4704 {
4705 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4706 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4707 }
4708 
4709  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4710 				struct x86_exception *exception)
4711 {
4712 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4713 	access |= PFERR_FETCH_MASK;
4714 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4715 }
4716 
4717 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4718 			       struct x86_exception *exception)
4719 {
4720 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4721 	access |= PFERR_WRITE_MASK;
4722 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4723 }
4724 
4725 /* uses this to access any guest's mapped memory without checking CPL */
4726 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4727 				struct x86_exception *exception)
4728 {
4729 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4730 }
4731 
4732 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4733 				      struct kvm_vcpu *vcpu, u32 access,
4734 				      struct x86_exception *exception)
4735 {
4736 	void *data = val;
4737 	int r = X86EMUL_CONTINUE;
4738 
4739 	while (bytes) {
4740 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4741 							    exception);
4742 		unsigned offset = addr & (PAGE_SIZE-1);
4743 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4744 		int ret;
4745 
4746 		if (gpa == UNMAPPED_GVA)
4747 			return X86EMUL_PROPAGATE_FAULT;
4748 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4749 					       offset, toread);
4750 		if (ret < 0) {
4751 			r = X86EMUL_IO_NEEDED;
4752 			goto out;
4753 		}
4754 
4755 		bytes -= toread;
4756 		data += toread;
4757 		addr += toread;
4758 	}
4759 out:
4760 	return r;
4761 }
4762 
4763 /* used for instruction fetching */
4764 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4765 				gva_t addr, void *val, unsigned int bytes,
4766 				struct x86_exception *exception)
4767 {
4768 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4769 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4770 	unsigned offset;
4771 	int ret;
4772 
4773 	/* Inline kvm_read_guest_virt_helper for speed.  */
4774 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4775 						    exception);
4776 	if (unlikely(gpa == UNMAPPED_GVA))
4777 		return X86EMUL_PROPAGATE_FAULT;
4778 
4779 	offset = addr & (PAGE_SIZE-1);
4780 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4781 		bytes = (unsigned)PAGE_SIZE - offset;
4782 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4783 				       offset, bytes);
4784 	if (unlikely(ret < 0))
4785 		return X86EMUL_IO_NEEDED;
4786 
4787 	return X86EMUL_CONTINUE;
4788 }
4789 
4790 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4791 			       gva_t addr, void *val, unsigned int bytes,
4792 			       struct x86_exception *exception)
4793 {
4794 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4795 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4796 
4797 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4798 					  exception);
4799 }
4800 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4801 
4802 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4803 				      gva_t addr, void *val, unsigned int bytes,
4804 				      struct x86_exception *exception)
4805 {
4806 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4807 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4808 }
4809 
4810 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4811 		unsigned long addr, void *val, unsigned int bytes)
4812 {
4813 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4814 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4815 
4816 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4817 }
4818 
4819 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4820 				       gva_t addr, void *val,
4821 				       unsigned int bytes,
4822 				       struct x86_exception *exception)
4823 {
4824 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4825 	void *data = val;
4826 	int r = X86EMUL_CONTINUE;
4827 
4828 	while (bytes) {
4829 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4830 							     PFERR_WRITE_MASK,
4831 							     exception);
4832 		unsigned offset = addr & (PAGE_SIZE-1);
4833 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4834 		int ret;
4835 
4836 		if (gpa == UNMAPPED_GVA)
4837 			return X86EMUL_PROPAGATE_FAULT;
4838 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4839 		if (ret < 0) {
4840 			r = X86EMUL_IO_NEEDED;
4841 			goto out;
4842 		}
4843 
4844 		bytes -= towrite;
4845 		data += towrite;
4846 		addr += towrite;
4847 	}
4848 out:
4849 	return r;
4850 }
4851 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4852 
4853 int handle_ud(struct kvm_vcpu *vcpu)
4854 {
4855 	int emul_type = EMULTYPE_TRAP_UD;
4856 	enum emulation_result er;
4857 	char sig[5]; /* ud2; .ascii "kvm" */
4858 	struct x86_exception e;
4859 
4860 	if (force_emulation_prefix &&
4861 	    kvm_read_guest_virt(&vcpu->arch.emulate_ctxt,
4862 				kvm_get_linear_rip(vcpu), sig, sizeof(sig), &e) == 0 &&
4863 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
4864 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
4865 		emul_type = 0;
4866 	}
4867 
4868 	er = emulate_instruction(vcpu, emul_type);
4869 	if (er == EMULATE_USER_EXIT)
4870 		return 0;
4871 	if (er != EMULATE_DONE)
4872 		kvm_queue_exception(vcpu, UD_VECTOR);
4873 	return 1;
4874 }
4875 EXPORT_SYMBOL_GPL(handle_ud);
4876 
4877 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4878 			    gpa_t gpa, bool write)
4879 {
4880 	/* For APIC access vmexit */
4881 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4882 		return 1;
4883 
4884 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4885 		trace_vcpu_match_mmio(gva, gpa, write, true);
4886 		return 1;
4887 	}
4888 
4889 	return 0;
4890 }
4891 
4892 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4893 				gpa_t *gpa, struct x86_exception *exception,
4894 				bool write)
4895 {
4896 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4897 		| (write ? PFERR_WRITE_MASK : 0);
4898 
4899 	/*
4900 	 * currently PKRU is only applied to ept enabled guest so
4901 	 * there is no pkey in EPT page table for L1 guest or EPT
4902 	 * shadow page table for L2 guest.
4903 	 */
4904 	if (vcpu_match_mmio_gva(vcpu, gva)
4905 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4906 				 vcpu->arch.access, 0, access)) {
4907 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4908 					(gva & (PAGE_SIZE - 1));
4909 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4910 		return 1;
4911 	}
4912 
4913 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4914 
4915 	if (*gpa == UNMAPPED_GVA)
4916 		return -1;
4917 
4918 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4919 }
4920 
4921 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4922 			const void *val, int bytes)
4923 {
4924 	int ret;
4925 
4926 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4927 	if (ret < 0)
4928 		return 0;
4929 	kvm_page_track_write(vcpu, gpa, val, bytes);
4930 	return 1;
4931 }
4932 
4933 struct read_write_emulator_ops {
4934 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4935 				  int bytes);
4936 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4937 				  void *val, int bytes);
4938 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4939 			       int bytes, void *val);
4940 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4941 				    void *val, int bytes);
4942 	bool write;
4943 };
4944 
4945 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4946 {
4947 	if (vcpu->mmio_read_completed) {
4948 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4949 			       vcpu->mmio_fragments[0].gpa, val);
4950 		vcpu->mmio_read_completed = 0;
4951 		return 1;
4952 	}
4953 
4954 	return 0;
4955 }
4956 
4957 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4958 			void *val, int bytes)
4959 {
4960 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4961 }
4962 
4963 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4964 			 void *val, int bytes)
4965 {
4966 	return emulator_write_phys(vcpu, gpa, val, bytes);
4967 }
4968 
4969 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4970 {
4971 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
4972 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4973 }
4974 
4975 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4976 			  void *val, int bytes)
4977 {
4978 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
4979 	return X86EMUL_IO_NEEDED;
4980 }
4981 
4982 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4983 			   void *val, int bytes)
4984 {
4985 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4986 
4987 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4988 	return X86EMUL_CONTINUE;
4989 }
4990 
4991 static const struct read_write_emulator_ops read_emultor = {
4992 	.read_write_prepare = read_prepare,
4993 	.read_write_emulate = read_emulate,
4994 	.read_write_mmio = vcpu_mmio_read,
4995 	.read_write_exit_mmio = read_exit_mmio,
4996 };
4997 
4998 static const struct read_write_emulator_ops write_emultor = {
4999 	.read_write_emulate = write_emulate,
5000 	.read_write_mmio = write_mmio,
5001 	.read_write_exit_mmio = write_exit_mmio,
5002 	.write = true,
5003 };
5004 
5005 static int emulator_read_write_onepage(unsigned long addr, void *val,
5006 				       unsigned int bytes,
5007 				       struct x86_exception *exception,
5008 				       struct kvm_vcpu *vcpu,
5009 				       const struct read_write_emulator_ops *ops)
5010 {
5011 	gpa_t gpa;
5012 	int handled, ret;
5013 	bool write = ops->write;
5014 	struct kvm_mmio_fragment *frag;
5015 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5016 
5017 	/*
5018 	 * If the exit was due to a NPF we may already have a GPA.
5019 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5020 	 * Note, this cannot be used on string operations since string
5021 	 * operation using rep will only have the initial GPA from the NPF
5022 	 * occurred.
5023 	 */
5024 	if (vcpu->arch.gpa_available &&
5025 	    emulator_can_use_gpa(ctxt) &&
5026 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5027 		gpa = vcpu->arch.gpa_val;
5028 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5029 	} else {
5030 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5031 		if (ret < 0)
5032 			return X86EMUL_PROPAGATE_FAULT;
5033 	}
5034 
5035 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5036 		return X86EMUL_CONTINUE;
5037 
5038 	/*
5039 	 * Is this MMIO handled locally?
5040 	 */
5041 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5042 	if (handled == bytes)
5043 		return X86EMUL_CONTINUE;
5044 
5045 	gpa += handled;
5046 	bytes -= handled;
5047 	val += handled;
5048 
5049 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5050 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5051 	frag->gpa = gpa;
5052 	frag->data = val;
5053 	frag->len = bytes;
5054 	return X86EMUL_CONTINUE;
5055 }
5056 
5057 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5058 			unsigned long addr,
5059 			void *val, unsigned int bytes,
5060 			struct x86_exception *exception,
5061 			const struct read_write_emulator_ops *ops)
5062 {
5063 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5064 	gpa_t gpa;
5065 	int rc;
5066 
5067 	if (ops->read_write_prepare &&
5068 		  ops->read_write_prepare(vcpu, val, bytes))
5069 		return X86EMUL_CONTINUE;
5070 
5071 	vcpu->mmio_nr_fragments = 0;
5072 
5073 	/* Crossing a page boundary? */
5074 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5075 		int now;
5076 
5077 		now = -addr & ~PAGE_MASK;
5078 		rc = emulator_read_write_onepage(addr, val, now, exception,
5079 						 vcpu, ops);
5080 
5081 		if (rc != X86EMUL_CONTINUE)
5082 			return rc;
5083 		addr += now;
5084 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5085 			addr = (u32)addr;
5086 		val += now;
5087 		bytes -= now;
5088 	}
5089 
5090 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5091 					 vcpu, ops);
5092 	if (rc != X86EMUL_CONTINUE)
5093 		return rc;
5094 
5095 	if (!vcpu->mmio_nr_fragments)
5096 		return rc;
5097 
5098 	gpa = vcpu->mmio_fragments[0].gpa;
5099 
5100 	vcpu->mmio_needed = 1;
5101 	vcpu->mmio_cur_fragment = 0;
5102 
5103 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5104 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5105 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5106 	vcpu->run->mmio.phys_addr = gpa;
5107 
5108 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5109 }
5110 
5111 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5112 				  unsigned long addr,
5113 				  void *val,
5114 				  unsigned int bytes,
5115 				  struct x86_exception *exception)
5116 {
5117 	return emulator_read_write(ctxt, addr, val, bytes,
5118 				   exception, &read_emultor);
5119 }
5120 
5121 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5122 			    unsigned long addr,
5123 			    const void *val,
5124 			    unsigned int bytes,
5125 			    struct x86_exception *exception)
5126 {
5127 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5128 				   exception, &write_emultor);
5129 }
5130 
5131 #define CMPXCHG_TYPE(t, ptr, old, new) \
5132 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5133 
5134 #ifdef CONFIG_X86_64
5135 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5136 #else
5137 #  define CMPXCHG64(ptr, old, new) \
5138 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5139 #endif
5140 
5141 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5142 				     unsigned long addr,
5143 				     const void *old,
5144 				     const void *new,
5145 				     unsigned int bytes,
5146 				     struct x86_exception *exception)
5147 {
5148 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5149 	gpa_t gpa;
5150 	struct page *page;
5151 	char *kaddr;
5152 	bool exchanged;
5153 
5154 	/* guests cmpxchg8b have to be emulated atomically */
5155 	if (bytes > 8 || (bytes & (bytes - 1)))
5156 		goto emul_write;
5157 
5158 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5159 
5160 	if (gpa == UNMAPPED_GVA ||
5161 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5162 		goto emul_write;
5163 
5164 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5165 		goto emul_write;
5166 
5167 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5168 	if (is_error_page(page))
5169 		goto emul_write;
5170 
5171 	kaddr = kmap_atomic(page);
5172 	kaddr += offset_in_page(gpa);
5173 	switch (bytes) {
5174 	case 1:
5175 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5176 		break;
5177 	case 2:
5178 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5179 		break;
5180 	case 4:
5181 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5182 		break;
5183 	case 8:
5184 		exchanged = CMPXCHG64(kaddr, old, new);
5185 		break;
5186 	default:
5187 		BUG();
5188 	}
5189 	kunmap_atomic(kaddr);
5190 	kvm_release_page_dirty(page);
5191 
5192 	if (!exchanged)
5193 		return X86EMUL_CMPXCHG_FAILED;
5194 
5195 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5196 	kvm_page_track_write(vcpu, gpa, new, bytes);
5197 
5198 	return X86EMUL_CONTINUE;
5199 
5200 emul_write:
5201 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5202 
5203 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5204 }
5205 
5206 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5207 {
5208 	int r = 0, i;
5209 
5210 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5211 		if (vcpu->arch.pio.in)
5212 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5213 					    vcpu->arch.pio.size, pd);
5214 		else
5215 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5216 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5217 					     pd);
5218 		if (r)
5219 			break;
5220 		pd += vcpu->arch.pio.size;
5221 	}
5222 	return r;
5223 }
5224 
5225 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5226 			       unsigned short port, void *val,
5227 			       unsigned int count, bool in)
5228 {
5229 	vcpu->arch.pio.port = port;
5230 	vcpu->arch.pio.in = in;
5231 	vcpu->arch.pio.count  = count;
5232 	vcpu->arch.pio.size = size;
5233 
5234 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5235 		vcpu->arch.pio.count = 0;
5236 		return 1;
5237 	}
5238 
5239 	vcpu->run->exit_reason = KVM_EXIT_IO;
5240 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5241 	vcpu->run->io.size = size;
5242 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5243 	vcpu->run->io.count = count;
5244 	vcpu->run->io.port = port;
5245 
5246 	return 0;
5247 }
5248 
5249 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5250 				    int size, unsigned short port, void *val,
5251 				    unsigned int count)
5252 {
5253 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5254 	int ret;
5255 
5256 	if (vcpu->arch.pio.count)
5257 		goto data_avail;
5258 
5259 	memset(vcpu->arch.pio_data, 0, size * count);
5260 
5261 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5262 	if (ret) {
5263 data_avail:
5264 		memcpy(val, vcpu->arch.pio_data, size * count);
5265 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5266 		vcpu->arch.pio.count = 0;
5267 		return 1;
5268 	}
5269 
5270 	return 0;
5271 }
5272 
5273 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5274 				     int size, unsigned short port,
5275 				     const void *val, unsigned int count)
5276 {
5277 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5278 
5279 	memcpy(vcpu->arch.pio_data, val, size * count);
5280 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5281 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5282 }
5283 
5284 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5285 {
5286 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5287 }
5288 
5289 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5290 {
5291 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5292 }
5293 
5294 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5295 {
5296 	if (!need_emulate_wbinvd(vcpu))
5297 		return X86EMUL_CONTINUE;
5298 
5299 	if (kvm_x86_ops->has_wbinvd_exit()) {
5300 		int cpu = get_cpu();
5301 
5302 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5303 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5304 				wbinvd_ipi, NULL, 1);
5305 		put_cpu();
5306 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5307 	} else
5308 		wbinvd();
5309 	return X86EMUL_CONTINUE;
5310 }
5311 
5312 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5313 {
5314 	kvm_emulate_wbinvd_noskip(vcpu);
5315 	return kvm_skip_emulated_instruction(vcpu);
5316 }
5317 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5318 
5319 
5320 
5321 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5322 {
5323 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5324 }
5325 
5326 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5327 			   unsigned long *dest)
5328 {
5329 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5330 }
5331 
5332 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5333 			   unsigned long value)
5334 {
5335 
5336 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5337 }
5338 
5339 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5340 {
5341 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5342 }
5343 
5344 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5345 {
5346 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5347 	unsigned long value;
5348 
5349 	switch (cr) {
5350 	case 0:
5351 		value = kvm_read_cr0(vcpu);
5352 		break;
5353 	case 2:
5354 		value = vcpu->arch.cr2;
5355 		break;
5356 	case 3:
5357 		value = kvm_read_cr3(vcpu);
5358 		break;
5359 	case 4:
5360 		value = kvm_read_cr4(vcpu);
5361 		break;
5362 	case 8:
5363 		value = kvm_get_cr8(vcpu);
5364 		break;
5365 	default:
5366 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5367 		return 0;
5368 	}
5369 
5370 	return value;
5371 }
5372 
5373 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5374 {
5375 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5376 	int res = 0;
5377 
5378 	switch (cr) {
5379 	case 0:
5380 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5381 		break;
5382 	case 2:
5383 		vcpu->arch.cr2 = val;
5384 		break;
5385 	case 3:
5386 		res = kvm_set_cr3(vcpu, val);
5387 		break;
5388 	case 4:
5389 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5390 		break;
5391 	case 8:
5392 		res = kvm_set_cr8(vcpu, val);
5393 		break;
5394 	default:
5395 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5396 		res = -1;
5397 	}
5398 
5399 	return res;
5400 }
5401 
5402 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5403 {
5404 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5405 }
5406 
5407 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5408 {
5409 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5410 }
5411 
5412 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5413 {
5414 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5415 }
5416 
5417 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5418 {
5419 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5420 }
5421 
5422 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5423 {
5424 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5425 }
5426 
5427 static unsigned long emulator_get_cached_segment_base(
5428 	struct x86_emulate_ctxt *ctxt, int seg)
5429 {
5430 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5431 }
5432 
5433 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5434 				 struct desc_struct *desc, u32 *base3,
5435 				 int seg)
5436 {
5437 	struct kvm_segment var;
5438 
5439 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5440 	*selector = var.selector;
5441 
5442 	if (var.unusable) {
5443 		memset(desc, 0, sizeof(*desc));
5444 		if (base3)
5445 			*base3 = 0;
5446 		return false;
5447 	}
5448 
5449 	if (var.g)
5450 		var.limit >>= 12;
5451 	set_desc_limit(desc, var.limit);
5452 	set_desc_base(desc, (unsigned long)var.base);
5453 #ifdef CONFIG_X86_64
5454 	if (base3)
5455 		*base3 = var.base >> 32;
5456 #endif
5457 	desc->type = var.type;
5458 	desc->s = var.s;
5459 	desc->dpl = var.dpl;
5460 	desc->p = var.present;
5461 	desc->avl = var.avl;
5462 	desc->l = var.l;
5463 	desc->d = var.db;
5464 	desc->g = var.g;
5465 
5466 	return true;
5467 }
5468 
5469 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5470 				 struct desc_struct *desc, u32 base3,
5471 				 int seg)
5472 {
5473 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5474 	struct kvm_segment var;
5475 
5476 	var.selector = selector;
5477 	var.base = get_desc_base(desc);
5478 #ifdef CONFIG_X86_64
5479 	var.base |= ((u64)base3) << 32;
5480 #endif
5481 	var.limit = get_desc_limit(desc);
5482 	if (desc->g)
5483 		var.limit = (var.limit << 12) | 0xfff;
5484 	var.type = desc->type;
5485 	var.dpl = desc->dpl;
5486 	var.db = desc->d;
5487 	var.s = desc->s;
5488 	var.l = desc->l;
5489 	var.g = desc->g;
5490 	var.avl = desc->avl;
5491 	var.present = desc->p;
5492 	var.unusable = !var.present;
5493 	var.padding = 0;
5494 
5495 	kvm_set_segment(vcpu, &var, seg);
5496 	return;
5497 }
5498 
5499 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5500 			    u32 msr_index, u64 *pdata)
5501 {
5502 	struct msr_data msr;
5503 	int r;
5504 
5505 	msr.index = msr_index;
5506 	msr.host_initiated = false;
5507 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5508 	if (r)
5509 		return r;
5510 
5511 	*pdata = msr.data;
5512 	return 0;
5513 }
5514 
5515 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5516 			    u32 msr_index, u64 data)
5517 {
5518 	struct msr_data msr;
5519 
5520 	msr.data = data;
5521 	msr.index = msr_index;
5522 	msr.host_initiated = false;
5523 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5524 }
5525 
5526 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5527 {
5528 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5529 
5530 	return vcpu->arch.smbase;
5531 }
5532 
5533 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5534 {
5535 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5536 
5537 	vcpu->arch.smbase = smbase;
5538 }
5539 
5540 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5541 			      u32 pmc)
5542 {
5543 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5544 }
5545 
5546 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5547 			     u32 pmc, u64 *pdata)
5548 {
5549 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5550 }
5551 
5552 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5553 {
5554 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5555 }
5556 
5557 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5558 			      struct x86_instruction_info *info,
5559 			      enum x86_intercept_stage stage)
5560 {
5561 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5562 }
5563 
5564 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5565 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5566 {
5567 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5568 }
5569 
5570 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5571 {
5572 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5573 }
5574 
5575 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5576 {
5577 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5578 }
5579 
5580 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5581 {
5582 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5583 }
5584 
5585 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5586 {
5587 	return emul_to_vcpu(ctxt)->arch.hflags;
5588 }
5589 
5590 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5591 {
5592 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5593 }
5594 
5595 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5596 {
5597 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5598 }
5599 
5600 static const struct x86_emulate_ops emulate_ops = {
5601 	.read_gpr            = emulator_read_gpr,
5602 	.write_gpr           = emulator_write_gpr,
5603 	.read_std            = kvm_read_guest_virt_system,
5604 	.write_std           = kvm_write_guest_virt_system,
5605 	.read_phys           = kvm_read_guest_phys_system,
5606 	.fetch               = kvm_fetch_guest_virt,
5607 	.read_emulated       = emulator_read_emulated,
5608 	.write_emulated      = emulator_write_emulated,
5609 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5610 	.invlpg              = emulator_invlpg,
5611 	.pio_in_emulated     = emulator_pio_in_emulated,
5612 	.pio_out_emulated    = emulator_pio_out_emulated,
5613 	.get_segment         = emulator_get_segment,
5614 	.set_segment         = emulator_set_segment,
5615 	.get_cached_segment_base = emulator_get_cached_segment_base,
5616 	.get_gdt             = emulator_get_gdt,
5617 	.get_idt	     = emulator_get_idt,
5618 	.set_gdt             = emulator_set_gdt,
5619 	.set_idt	     = emulator_set_idt,
5620 	.get_cr              = emulator_get_cr,
5621 	.set_cr              = emulator_set_cr,
5622 	.cpl                 = emulator_get_cpl,
5623 	.get_dr              = emulator_get_dr,
5624 	.set_dr              = emulator_set_dr,
5625 	.get_smbase          = emulator_get_smbase,
5626 	.set_smbase          = emulator_set_smbase,
5627 	.set_msr             = emulator_set_msr,
5628 	.get_msr             = emulator_get_msr,
5629 	.check_pmc	     = emulator_check_pmc,
5630 	.read_pmc            = emulator_read_pmc,
5631 	.halt                = emulator_halt,
5632 	.wbinvd              = emulator_wbinvd,
5633 	.fix_hypercall       = emulator_fix_hypercall,
5634 	.intercept           = emulator_intercept,
5635 	.get_cpuid           = emulator_get_cpuid,
5636 	.set_nmi_mask        = emulator_set_nmi_mask,
5637 	.get_hflags          = emulator_get_hflags,
5638 	.set_hflags          = emulator_set_hflags,
5639 	.pre_leave_smm       = emulator_pre_leave_smm,
5640 };
5641 
5642 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5643 {
5644 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5645 	/*
5646 	 * an sti; sti; sequence only disable interrupts for the first
5647 	 * instruction. So, if the last instruction, be it emulated or
5648 	 * not, left the system with the INT_STI flag enabled, it
5649 	 * means that the last instruction is an sti. We should not
5650 	 * leave the flag on in this case. The same goes for mov ss
5651 	 */
5652 	if (int_shadow & mask)
5653 		mask = 0;
5654 	if (unlikely(int_shadow || mask)) {
5655 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5656 		if (!mask)
5657 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5658 	}
5659 }
5660 
5661 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5662 {
5663 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5664 	if (ctxt->exception.vector == PF_VECTOR)
5665 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5666 
5667 	if (ctxt->exception.error_code_valid)
5668 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5669 				      ctxt->exception.error_code);
5670 	else
5671 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5672 	return false;
5673 }
5674 
5675 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5676 {
5677 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5678 	int cs_db, cs_l;
5679 
5680 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5681 
5682 	ctxt->eflags = kvm_get_rflags(vcpu);
5683 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5684 
5685 	ctxt->eip = kvm_rip_read(vcpu);
5686 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5687 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5688 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5689 		     cs_db				? X86EMUL_MODE_PROT32 :
5690 							  X86EMUL_MODE_PROT16;
5691 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5692 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5693 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5694 
5695 	init_decode_cache(ctxt);
5696 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5697 }
5698 
5699 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5700 {
5701 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5702 	int ret;
5703 
5704 	init_emulate_ctxt(vcpu);
5705 
5706 	ctxt->op_bytes = 2;
5707 	ctxt->ad_bytes = 2;
5708 	ctxt->_eip = ctxt->eip + inc_eip;
5709 	ret = emulate_int_real(ctxt, irq);
5710 
5711 	if (ret != X86EMUL_CONTINUE)
5712 		return EMULATE_FAIL;
5713 
5714 	ctxt->eip = ctxt->_eip;
5715 	kvm_rip_write(vcpu, ctxt->eip);
5716 	kvm_set_rflags(vcpu, ctxt->eflags);
5717 
5718 	return EMULATE_DONE;
5719 }
5720 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5721 
5722 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5723 {
5724 	int r = EMULATE_DONE;
5725 
5726 	++vcpu->stat.insn_emulation_fail;
5727 	trace_kvm_emulate_insn_failed(vcpu);
5728 
5729 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
5730 		return EMULATE_FAIL;
5731 
5732 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5733 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5734 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5735 		vcpu->run->internal.ndata = 0;
5736 		r = EMULATE_USER_EXIT;
5737 	}
5738 
5739 	kvm_queue_exception(vcpu, UD_VECTOR);
5740 
5741 	return r;
5742 }
5743 
5744 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5745 				  bool write_fault_to_shadow_pgtable,
5746 				  int emulation_type)
5747 {
5748 	gpa_t gpa = cr2;
5749 	kvm_pfn_t pfn;
5750 
5751 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5752 		return false;
5753 
5754 	if (!vcpu->arch.mmu.direct_map) {
5755 		/*
5756 		 * Write permission should be allowed since only
5757 		 * write access need to be emulated.
5758 		 */
5759 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5760 
5761 		/*
5762 		 * If the mapping is invalid in guest, let cpu retry
5763 		 * it to generate fault.
5764 		 */
5765 		if (gpa == UNMAPPED_GVA)
5766 			return true;
5767 	}
5768 
5769 	/*
5770 	 * Do not retry the unhandleable instruction if it faults on the
5771 	 * readonly host memory, otherwise it will goto a infinite loop:
5772 	 * retry instruction -> write #PF -> emulation fail -> retry
5773 	 * instruction -> ...
5774 	 */
5775 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5776 
5777 	/*
5778 	 * If the instruction failed on the error pfn, it can not be fixed,
5779 	 * report the error to userspace.
5780 	 */
5781 	if (is_error_noslot_pfn(pfn))
5782 		return false;
5783 
5784 	kvm_release_pfn_clean(pfn);
5785 
5786 	/* The instructions are well-emulated on direct mmu. */
5787 	if (vcpu->arch.mmu.direct_map) {
5788 		unsigned int indirect_shadow_pages;
5789 
5790 		spin_lock(&vcpu->kvm->mmu_lock);
5791 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5792 		spin_unlock(&vcpu->kvm->mmu_lock);
5793 
5794 		if (indirect_shadow_pages)
5795 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5796 
5797 		return true;
5798 	}
5799 
5800 	/*
5801 	 * if emulation was due to access to shadowed page table
5802 	 * and it failed try to unshadow page and re-enter the
5803 	 * guest to let CPU execute the instruction.
5804 	 */
5805 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5806 
5807 	/*
5808 	 * If the access faults on its page table, it can not
5809 	 * be fixed by unprotecting shadow page and it should
5810 	 * be reported to userspace.
5811 	 */
5812 	return !write_fault_to_shadow_pgtable;
5813 }
5814 
5815 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5816 			      unsigned long cr2,  int emulation_type)
5817 {
5818 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5819 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5820 
5821 	last_retry_eip = vcpu->arch.last_retry_eip;
5822 	last_retry_addr = vcpu->arch.last_retry_addr;
5823 
5824 	/*
5825 	 * If the emulation is caused by #PF and it is non-page_table
5826 	 * writing instruction, it means the VM-EXIT is caused by shadow
5827 	 * page protected, we can zap the shadow page and retry this
5828 	 * instruction directly.
5829 	 *
5830 	 * Note: if the guest uses a non-page-table modifying instruction
5831 	 * on the PDE that points to the instruction, then we will unmap
5832 	 * the instruction and go to an infinite loop. So, we cache the
5833 	 * last retried eip and the last fault address, if we meet the eip
5834 	 * and the address again, we can break out of the potential infinite
5835 	 * loop.
5836 	 */
5837 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5838 
5839 	if (!(emulation_type & EMULTYPE_RETRY))
5840 		return false;
5841 
5842 	if (x86_page_table_writing_insn(ctxt))
5843 		return false;
5844 
5845 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5846 		return false;
5847 
5848 	vcpu->arch.last_retry_eip = ctxt->eip;
5849 	vcpu->arch.last_retry_addr = cr2;
5850 
5851 	if (!vcpu->arch.mmu.direct_map)
5852 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5853 
5854 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5855 
5856 	return true;
5857 }
5858 
5859 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5860 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5861 
5862 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5863 {
5864 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5865 		/* This is a good place to trace that we are exiting SMM.  */
5866 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5867 
5868 		/* Process a latched INIT or SMI, if any.  */
5869 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5870 	}
5871 
5872 	kvm_mmu_reset_context(vcpu);
5873 }
5874 
5875 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5876 {
5877 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5878 
5879 	vcpu->arch.hflags = emul_flags;
5880 
5881 	if (changed & HF_SMM_MASK)
5882 		kvm_smm_changed(vcpu);
5883 }
5884 
5885 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5886 				unsigned long *db)
5887 {
5888 	u32 dr6 = 0;
5889 	int i;
5890 	u32 enable, rwlen;
5891 
5892 	enable = dr7;
5893 	rwlen = dr7 >> 16;
5894 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5895 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5896 			dr6 |= (1 << i);
5897 	return dr6;
5898 }
5899 
5900 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5901 {
5902 	struct kvm_run *kvm_run = vcpu->run;
5903 
5904 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5905 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5906 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5907 		kvm_run->debug.arch.exception = DB_VECTOR;
5908 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5909 		*r = EMULATE_USER_EXIT;
5910 	} else {
5911 		/*
5912 		 * "Certain debug exceptions may clear bit 0-3.  The
5913 		 * remaining contents of the DR6 register are never
5914 		 * cleared by the processor".
5915 		 */
5916 		vcpu->arch.dr6 &= ~15;
5917 		vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5918 		kvm_queue_exception(vcpu, DB_VECTOR);
5919 	}
5920 }
5921 
5922 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5923 {
5924 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5925 	int r = EMULATE_DONE;
5926 
5927 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5928 
5929 	/*
5930 	 * rflags is the old, "raw" value of the flags.  The new value has
5931 	 * not been saved yet.
5932 	 *
5933 	 * This is correct even for TF set by the guest, because "the
5934 	 * processor will not generate this exception after the instruction
5935 	 * that sets the TF flag".
5936 	 */
5937 	if (unlikely(rflags & X86_EFLAGS_TF))
5938 		kvm_vcpu_do_singlestep(vcpu, &r);
5939 	return r == EMULATE_DONE;
5940 }
5941 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5942 
5943 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5944 {
5945 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5946 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5947 		struct kvm_run *kvm_run = vcpu->run;
5948 		unsigned long eip = kvm_get_linear_rip(vcpu);
5949 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5950 					   vcpu->arch.guest_debug_dr7,
5951 					   vcpu->arch.eff_db);
5952 
5953 		if (dr6 != 0) {
5954 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5955 			kvm_run->debug.arch.pc = eip;
5956 			kvm_run->debug.arch.exception = DB_VECTOR;
5957 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5958 			*r = EMULATE_USER_EXIT;
5959 			return true;
5960 		}
5961 	}
5962 
5963 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5964 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5965 		unsigned long eip = kvm_get_linear_rip(vcpu);
5966 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5967 					   vcpu->arch.dr7,
5968 					   vcpu->arch.db);
5969 
5970 		if (dr6 != 0) {
5971 			vcpu->arch.dr6 &= ~15;
5972 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5973 			kvm_queue_exception(vcpu, DB_VECTOR);
5974 			*r = EMULATE_DONE;
5975 			return true;
5976 		}
5977 	}
5978 
5979 	return false;
5980 }
5981 
5982 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
5983 {
5984 	switch (ctxt->opcode_len) {
5985 	case 1:
5986 		switch (ctxt->b) {
5987 		case 0xe4:	/* IN */
5988 		case 0xe5:
5989 		case 0xec:
5990 		case 0xed:
5991 		case 0xe6:	/* OUT */
5992 		case 0xe7:
5993 		case 0xee:
5994 		case 0xef:
5995 		case 0x6c:	/* INS */
5996 		case 0x6d:
5997 		case 0x6e:	/* OUTS */
5998 		case 0x6f:
5999 			return true;
6000 		}
6001 		break;
6002 	case 2:
6003 		switch (ctxt->b) {
6004 		case 0x33:	/* RDPMC */
6005 			return true;
6006 		}
6007 		break;
6008 	}
6009 
6010 	return false;
6011 }
6012 
6013 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6014 			    unsigned long cr2,
6015 			    int emulation_type,
6016 			    void *insn,
6017 			    int insn_len)
6018 {
6019 	int r;
6020 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6021 	bool writeback = true;
6022 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6023 
6024 	/*
6025 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6026 	 * never reused.
6027 	 */
6028 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6029 	kvm_clear_exception_queue(vcpu);
6030 
6031 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6032 		init_emulate_ctxt(vcpu);
6033 
6034 		/*
6035 		 * We will reenter on the same instruction since
6036 		 * we do not set complete_userspace_io.  This does not
6037 		 * handle watchpoints yet, those would be handled in
6038 		 * the emulate_ops.
6039 		 */
6040 		if (!(emulation_type & EMULTYPE_SKIP) &&
6041 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6042 			return r;
6043 
6044 		ctxt->interruptibility = 0;
6045 		ctxt->have_exception = false;
6046 		ctxt->exception.vector = -1;
6047 		ctxt->perm_ok = false;
6048 
6049 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6050 
6051 		r = x86_decode_insn(ctxt, insn, insn_len);
6052 
6053 		trace_kvm_emulate_insn_start(vcpu);
6054 		++vcpu->stat.insn_emulation;
6055 		if (r != EMULATION_OK)  {
6056 			if (emulation_type & EMULTYPE_TRAP_UD)
6057 				return EMULATE_FAIL;
6058 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6059 						emulation_type))
6060 				return EMULATE_DONE;
6061 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6062 				return EMULATE_DONE;
6063 			if (emulation_type & EMULTYPE_SKIP)
6064 				return EMULATE_FAIL;
6065 			return handle_emulation_failure(vcpu, emulation_type);
6066 		}
6067 	}
6068 
6069 	if ((emulation_type & EMULTYPE_VMWARE) &&
6070 	    !is_vmware_backdoor_opcode(ctxt))
6071 		return EMULATE_FAIL;
6072 
6073 	if (emulation_type & EMULTYPE_SKIP) {
6074 		kvm_rip_write(vcpu, ctxt->_eip);
6075 		if (ctxt->eflags & X86_EFLAGS_RF)
6076 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6077 		return EMULATE_DONE;
6078 	}
6079 
6080 	if (retry_instruction(ctxt, cr2, emulation_type))
6081 		return EMULATE_DONE;
6082 
6083 	/* this is needed for vmware backdoor interface to work since it
6084 	   changes registers values  during IO operation */
6085 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6086 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6087 		emulator_invalidate_register_cache(ctxt);
6088 	}
6089 
6090 restart:
6091 	/* Save the faulting GPA (cr2) in the address field */
6092 	ctxt->exception.address = cr2;
6093 
6094 	r = x86_emulate_insn(ctxt);
6095 
6096 	if (r == EMULATION_INTERCEPTED)
6097 		return EMULATE_DONE;
6098 
6099 	if (r == EMULATION_FAILED) {
6100 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6101 					emulation_type))
6102 			return EMULATE_DONE;
6103 
6104 		return handle_emulation_failure(vcpu, emulation_type);
6105 	}
6106 
6107 	if (ctxt->have_exception) {
6108 		r = EMULATE_DONE;
6109 		if (inject_emulated_exception(vcpu))
6110 			return r;
6111 	} else if (vcpu->arch.pio.count) {
6112 		if (!vcpu->arch.pio.in) {
6113 			/* FIXME: return into emulator if single-stepping.  */
6114 			vcpu->arch.pio.count = 0;
6115 		} else {
6116 			writeback = false;
6117 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6118 		}
6119 		r = EMULATE_USER_EXIT;
6120 	} else if (vcpu->mmio_needed) {
6121 		if (!vcpu->mmio_is_write)
6122 			writeback = false;
6123 		r = EMULATE_USER_EXIT;
6124 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6125 	} else if (r == EMULATION_RESTART)
6126 		goto restart;
6127 	else
6128 		r = EMULATE_DONE;
6129 
6130 	if (writeback) {
6131 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6132 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6133 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6134 		kvm_rip_write(vcpu, ctxt->eip);
6135 		if (r == EMULATE_DONE &&
6136 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6137 			kvm_vcpu_do_singlestep(vcpu, &r);
6138 		if (!ctxt->have_exception ||
6139 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6140 			__kvm_set_rflags(vcpu, ctxt->eflags);
6141 
6142 		/*
6143 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6144 		 * do nothing, and it will be requested again as soon as
6145 		 * the shadow expires.  But we still need to check here,
6146 		 * because POPF has no interrupt shadow.
6147 		 */
6148 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6149 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6150 	} else
6151 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6152 
6153 	return r;
6154 }
6155 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
6156 
6157 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6158 			    unsigned short port)
6159 {
6160 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6161 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6162 					    size, port, &val, 1);
6163 	/* do not return to emulator after return from userspace */
6164 	vcpu->arch.pio.count = 0;
6165 	return ret;
6166 }
6167 
6168 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6169 {
6170 	unsigned long val;
6171 
6172 	/* We should only ever be called with arch.pio.count equal to 1 */
6173 	BUG_ON(vcpu->arch.pio.count != 1);
6174 
6175 	/* For size less than 4 we merge, else we zero extend */
6176 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6177 					: 0;
6178 
6179 	/*
6180 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6181 	 * the copy and tracing
6182 	 */
6183 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6184 				 vcpu->arch.pio.port, &val, 1);
6185 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6186 
6187 	return 1;
6188 }
6189 
6190 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6191 			   unsigned short port)
6192 {
6193 	unsigned long val;
6194 	int ret;
6195 
6196 	/* For size less than 4 we merge, else we zero extend */
6197 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6198 
6199 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6200 				       &val, 1);
6201 	if (ret) {
6202 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6203 		return ret;
6204 	}
6205 
6206 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6207 
6208 	return 0;
6209 }
6210 
6211 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6212 {
6213 	int ret = kvm_skip_emulated_instruction(vcpu);
6214 
6215 	/*
6216 	 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6217 	 * KVM_EXIT_DEBUG here.
6218 	 */
6219 	if (in)
6220 		return kvm_fast_pio_in(vcpu, size, port) && ret;
6221 	else
6222 		return kvm_fast_pio_out(vcpu, size, port) && ret;
6223 }
6224 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6225 
6226 static int kvmclock_cpu_down_prep(unsigned int cpu)
6227 {
6228 	__this_cpu_write(cpu_tsc_khz, 0);
6229 	return 0;
6230 }
6231 
6232 static void tsc_khz_changed(void *data)
6233 {
6234 	struct cpufreq_freqs *freq = data;
6235 	unsigned long khz = 0;
6236 
6237 	if (data)
6238 		khz = freq->new;
6239 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6240 		khz = cpufreq_quick_get(raw_smp_processor_id());
6241 	if (!khz)
6242 		khz = tsc_khz;
6243 	__this_cpu_write(cpu_tsc_khz, khz);
6244 }
6245 
6246 #ifdef CONFIG_X86_64
6247 static void kvm_hyperv_tsc_notifier(void)
6248 {
6249 	struct kvm *kvm;
6250 	struct kvm_vcpu *vcpu;
6251 	int cpu;
6252 
6253 	spin_lock(&kvm_lock);
6254 	list_for_each_entry(kvm, &vm_list, vm_list)
6255 		kvm_make_mclock_inprogress_request(kvm);
6256 
6257 	hyperv_stop_tsc_emulation();
6258 
6259 	/* TSC frequency always matches when on Hyper-V */
6260 	for_each_present_cpu(cpu)
6261 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6262 	kvm_max_guest_tsc_khz = tsc_khz;
6263 
6264 	list_for_each_entry(kvm, &vm_list, vm_list) {
6265 		struct kvm_arch *ka = &kvm->arch;
6266 
6267 		spin_lock(&ka->pvclock_gtod_sync_lock);
6268 
6269 		pvclock_update_vm_gtod_copy(kvm);
6270 
6271 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6272 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6273 
6274 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6275 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6276 
6277 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6278 	}
6279 	spin_unlock(&kvm_lock);
6280 }
6281 #endif
6282 
6283 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6284 				     void *data)
6285 {
6286 	struct cpufreq_freqs *freq = data;
6287 	struct kvm *kvm;
6288 	struct kvm_vcpu *vcpu;
6289 	int i, send_ipi = 0;
6290 
6291 	/*
6292 	 * We allow guests to temporarily run on slowing clocks,
6293 	 * provided we notify them after, or to run on accelerating
6294 	 * clocks, provided we notify them before.  Thus time never
6295 	 * goes backwards.
6296 	 *
6297 	 * However, we have a problem.  We can't atomically update
6298 	 * the frequency of a given CPU from this function; it is
6299 	 * merely a notifier, which can be called from any CPU.
6300 	 * Changing the TSC frequency at arbitrary points in time
6301 	 * requires a recomputation of local variables related to
6302 	 * the TSC for each VCPU.  We must flag these local variables
6303 	 * to be updated and be sure the update takes place with the
6304 	 * new frequency before any guests proceed.
6305 	 *
6306 	 * Unfortunately, the combination of hotplug CPU and frequency
6307 	 * change creates an intractable locking scenario; the order
6308 	 * of when these callouts happen is undefined with respect to
6309 	 * CPU hotplug, and they can race with each other.  As such,
6310 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6311 	 * undefined; you can actually have a CPU frequency change take
6312 	 * place in between the computation of X and the setting of the
6313 	 * variable.  To protect against this problem, all updates of
6314 	 * the per_cpu tsc_khz variable are done in an interrupt
6315 	 * protected IPI, and all callers wishing to update the value
6316 	 * must wait for a synchronous IPI to complete (which is trivial
6317 	 * if the caller is on the CPU already).  This establishes the
6318 	 * necessary total order on variable updates.
6319 	 *
6320 	 * Note that because a guest time update may take place
6321 	 * anytime after the setting of the VCPU's request bit, the
6322 	 * correct TSC value must be set before the request.  However,
6323 	 * to ensure the update actually makes it to any guest which
6324 	 * starts running in hardware virtualization between the set
6325 	 * and the acquisition of the spinlock, we must also ping the
6326 	 * CPU after setting the request bit.
6327 	 *
6328 	 */
6329 
6330 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6331 		return 0;
6332 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6333 		return 0;
6334 
6335 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6336 
6337 	spin_lock(&kvm_lock);
6338 	list_for_each_entry(kvm, &vm_list, vm_list) {
6339 		kvm_for_each_vcpu(i, vcpu, kvm) {
6340 			if (vcpu->cpu != freq->cpu)
6341 				continue;
6342 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6343 			if (vcpu->cpu != smp_processor_id())
6344 				send_ipi = 1;
6345 		}
6346 	}
6347 	spin_unlock(&kvm_lock);
6348 
6349 	if (freq->old < freq->new && send_ipi) {
6350 		/*
6351 		 * We upscale the frequency.  Must make the guest
6352 		 * doesn't see old kvmclock values while running with
6353 		 * the new frequency, otherwise we risk the guest sees
6354 		 * time go backwards.
6355 		 *
6356 		 * In case we update the frequency for another cpu
6357 		 * (which might be in guest context) send an interrupt
6358 		 * to kick the cpu out of guest context.  Next time
6359 		 * guest context is entered kvmclock will be updated,
6360 		 * so the guest will not see stale values.
6361 		 */
6362 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6363 	}
6364 	return 0;
6365 }
6366 
6367 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6368 	.notifier_call  = kvmclock_cpufreq_notifier
6369 };
6370 
6371 static int kvmclock_cpu_online(unsigned int cpu)
6372 {
6373 	tsc_khz_changed(NULL);
6374 	return 0;
6375 }
6376 
6377 static void kvm_timer_init(void)
6378 {
6379 	max_tsc_khz = tsc_khz;
6380 
6381 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6382 #ifdef CONFIG_CPU_FREQ
6383 		struct cpufreq_policy policy;
6384 		int cpu;
6385 
6386 		memset(&policy, 0, sizeof(policy));
6387 		cpu = get_cpu();
6388 		cpufreq_get_policy(&policy, cpu);
6389 		if (policy.cpuinfo.max_freq)
6390 			max_tsc_khz = policy.cpuinfo.max_freq;
6391 		put_cpu();
6392 #endif
6393 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6394 					  CPUFREQ_TRANSITION_NOTIFIER);
6395 	}
6396 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6397 
6398 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6399 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6400 }
6401 
6402 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6403 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6404 
6405 int kvm_is_in_guest(void)
6406 {
6407 	return __this_cpu_read(current_vcpu) != NULL;
6408 }
6409 
6410 static int kvm_is_user_mode(void)
6411 {
6412 	int user_mode = 3;
6413 
6414 	if (__this_cpu_read(current_vcpu))
6415 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6416 
6417 	return user_mode != 0;
6418 }
6419 
6420 static unsigned long kvm_get_guest_ip(void)
6421 {
6422 	unsigned long ip = 0;
6423 
6424 	if (__this_cpu_read(current_vcpu))
6425 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6426 
6427 	return ip;
6428 }
6429 
6430 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6431 	.is_in_guest		= kvm_is_in_guest,
6432 	.is_user_mode		= kvm_is_user_mode,
6433 	.get_guest_ip		= kvm_get_guest_ip,
6434 };
6435 
6436 static void kvm_set_mmio_spte_mask(void)
6437 {
6438 	u64 mask;
6439 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6440 
6441 	/*
6442 	 * Set the reserved bits and the present bit of an paging-structure
6443 	 * entry to generate page fault with PFER.RSV = 1.
6444 	 */
6445 	 /* Mask the reserved physical address bits. */
6446 	mask = rsvd_bits(maxphyaddr, 51);
6447 
6448 	/* Set the present bit. */
6449 	mask |= 1ull;
6450 
6451 #ifdef CONFIG_X86_64
6452 	/*
6453 	 * If reserved bit is not supported, clear the present bit to disable
6454 	 * mmio page fault.
6455 	 */
6456 	if (maxphyaddr == 52)
6457 		mask &= ~1ull;
6458 #endif
6459 
6460 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6461 }
6462 
6463 #ifdef CONFIG_X86_64
6464 static void pvclock_gtod_update_fn(struct work_struct *work)
6465 {
6466 	struct kvm *kvm;
6467 
6468 	struct kvm_vcpu *vcpu;
6469 	int i;
6470 
6471 	spin_lock(&kvm_lock);
6472 	list_for_each_entry(kvm, &vm_list, vm_list)
6473 		kvm_for_each_vcpu(i, vcpu, kvm)
6474 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6475 	atomic_set(&kvm_guest_has_master_clock, 0);
6476 	spin_unlock(&kvm_lock);
6477 }
6478 
6479 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6480 
6481 /*
6482  * Notification about pvclock gtod data update.
6483  */
6484 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6485 			       void *priv)
6486 {
6487 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6488 	struct timekeeper *tk = priv;
6489 
6490 	update_pvclock_gtod(tk);
6491 
6492 	/* disable master clock if host does not trust, or does not
6493 	 * use, TSC based clocksource.
6494 	 */
6495 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6496 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6497 		queue_work(system_long_wq, &pvclock_gtod_work);
6498 
6499 	return 0;
6500 }
6501 
6502 static struct notifier_block pvclock_gtod_notifier = {
6503 	.notifier_call = pvclock_gtod_notify,
6504 };
6505 #endif
6506 
6507 int kvm_arch_init(void *opaque)
6508 {
6509 	int r;
6510 	struct kvm_x86_ops *ops = opaque;
6511 
6512 	if (kvm_x86_ops) {
6513 		printk(KERN_ERR "kvm: already loaded the other module\n");
6514 		r = -EEXIST;
6515 		goto out;
6516 	}
6517 
6518 	if (!ops->cpu_has_kvm_support()) {
6519 		printk(KERN_ERR "kvm: no hardware support\n");
6520 		r = -EOPNOTSUPP;
6521 		goto out;
6522 	}
6523 	if (ops->disabled_by_bios()) {
6524 		printk(KERN_ERR "kvm: disabled by bios\n");
6525 		r = -EOPNOTSUPP;
6526 		goto out;
6527 	}
6528 
6529 	r = -ENOMEM;
6530 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6531 	if (!shared_msrs) {
6532 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6533 		goto out;
6534 	}
6535 
6536 	r = kvm_mmu_module_init();
6537 	if (r)
6538 		goto out_free_percpu;
6539 
6540 	kvm_set_mmio_spte_mask();
6541 
6542 	kvm_x86_ops = ops;
6543 
6544 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6545 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6546 			PT_PRESENT_MASK, 0, sme_me_mask);
6547 	kvm_timer_init();
6548 
6549 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6550 
6551 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6552 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6553 
6554 	kvm_lapic_init();
6555 #ifdef CONFIG_X86_64
6556 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6557 
6558 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6559 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6560 #endif
6561 
6562 	return 0;
6563 
6564 out_free_percpu:
6565 	free_percpu(shared_msrs);
6566 out:
6567 	return r;
6568 }
6569 
6570 void kvm_arch_exit(void)
6571 {
6572 #ifdef CONFIG_X86_64
6573 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6574 		clear_hv_tscchange_cb();
6575 #endif
6576 	kvm_lapic_exit();
6577 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6578 
6579 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6580 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6581 					    CPUFREQ_TRANSITION_NOTIFIER);
6582 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6583 #ifdef CONFIG_X86_64
6584 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6585 #endif
6586 	kvm_x86_ops = NULL;
6587 	kvm_mmu_module_exit();
6588 	free_percpu(shared_msrs);
6589 }
6590 
6591 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6592 {
6593 	++vcpu->stat.halt_exits;
6594 	if (lapic_in_kernel(vcpu)) {
6595 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6596 		return 1;
6597 	} else {
6598 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6599 		return 0;
6600 	}
6601 }
6602 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6603 
6604 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6605 {
6606 	int ret = kvm_skip_emulated_instruction(vcpu);
6607 	/*
6608 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6609 	 * KVM_EXIT_DEBUG here.
6610 	 */
6611 	return kvm_vcpu_halt(vcpu) && ret;
6612 }
6613 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6614 
6615 #ifdef CONFIG_X86_64
6616 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6617 			        unsigned long clock_type)
6618 {
6619 	struct kvm_clock_pairing clock_pairing;
6620 	struct timespec ts;
6621 	u64 cycle;
6622 	int ret;
6623 
6624 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6625 		return -KVM_EOPNOTSUPP;
6626 
6627 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6628 		return -KVM_EOPNOTSUPP;
6629 
6630 	clock_pairing.sec = ts.tv_sec;
6631 	clock_pairing.nsec = ts.tv_nsec;
6632 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6633 	clock_pairing.flags = 0;
6634 
6635 	ret = 0;
6636 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6637 			    sizeof(struct kvm_clock_pairing)))
6638 		ret = -KVM_EFAULT;
6639 
6640 	return ret;
6641 }
6642 #endif
6643 
6644 /*
6645  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6646  *
6647  * @apicid - apicid of vcpu to be kicked.
6648  */
6649 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6650 {
6651 	struct kvm_lapic_irq lapic_irq;
6652 
6653 	lapic_irq.shorthand = 0;
6654 	lapic_irq.dest_mode = 0;
6655 	lapic_irq.level = 0;
6656 	lapic_irq.dest_id = apicid;
6657 	lapic_irq.msi_redir_hint = false;
6658 
6659 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6660 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6661 }
6662 
6663 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6664 {
6665 	vcpu->arch.apicv_active = false;
6666 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6667 }
6668 
6669 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6670 {
6671 	unsigned long nr, a0, a1, a2, a3, ret;
6672 	int op_64_bit;
6673 
6674 	if (kvm_hv_hypercall_enabled(vcpu->kvm)) {
6675 		if (!kvm_hv_hypercall(vcpu))
6676 			return 0;
6677 		goto out;
6678 	}
6679 
6680 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6681 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6682 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6683 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6684 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6685 
6686 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6687 
6688 	op_64_bit = is_64_bit_mode(vcpu);
6689 	if (!op_64_bit) {
6690 		nr &= 0xFFFFFFFF;
6691 		a0 &= 0xFFFFFFFF;
6692 		a1 &= 0xFFFFFFFF;
6693 		a2 &= 0xFFFFFFFF;
6694 		a3 &= 0xFFFFFFFF;
6695 	}
6696 
6697 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6698 		ret = -KVM_EPERM;
6699 		goto out_error;
6700 	}
6701 
6702 	switch (nr) {
6703 	case KVM_HC_VAPIC_POLL_IRQ:
6704 		ret = 0;
6705 		break;
6706 	case KVM_HC_KICK_CPU:
6707 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6708 		ret = 0;
6709 		break;
6710 #ifdef CONFIG_X86_64
6711 	case KVM_HC_CLOCK_PAIRING:
6712 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6713 		break;
6714 #endif
6715 	default:
6716 		ret = -KVM_ENOSYS;
6717 		break;
6718 	}
6719 out_error:
6720 	if (!op_64_bit)
6721 		ret = (u32)ret;
6722 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6723 
6724 out:
6725 	++vcpu->stat.hypercalls;
6726 	return kvm_skip_emulated_instruction(vcpu);
6727 }
6728 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6729 
6730 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6731 {
6732 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6733 	char instruction[3];
6734 	unsigned long rip = kvm_rip_read(vcpu);
6735 
6736 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6737 
6738 	return emulator_write_emulated(ctxt, rip, instruction, 3,
6739 		&ctxt->exception);
6740 }
6741 
6742 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6743 {
6744 	return vcpu->run->request_interrupt_window &&
6745 		likely(!pic_in_kernel(vcpu->kvm));
6746 }
6747 
6748 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6749 {
6750 	struct kvm_run *kvm_run = vcpu->run;
6751 
6752 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6753 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6754 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6755 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6756 	kvm_run->ready_for_interrupt_injection =
6757 		pic_in_kernel(vcpu->kvm) ||
6758 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6759 }
6760 
6761 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6762 {
6763 	int max_irr, tpr;
6764 
6765 	if (!kvm_x86_ops->update_cr8_intercept)
6766 		return;
6767 
6768 	if (!lapic_in_kernel(vcpu))
6769 		return;
6770 
6771 	if (vcpu->arch.apicv_active)
6772 		return;
6773 
6774 	if (!vcpu->arch.apic->vapic_addr)
6775 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6776 	else
6777 		max_irr = -1;
6778 
6779 	if (max_irr != -1)
6780 		max_irr >>= 4;
6781 
6782 	tpr = kvm_lapic_get_cr8(vcpu);
6783 
6784 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6785 }
6786 
6787 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6788 {
6789 	int r;
6790 
6791 	/* try to reinject previous events if any */
6792 
6793 	if (vcpu->arch.exception.injected)
6794 		kvm_x86_ops->queue_exception(vcpu);
6795 	/*
6796 	 * Do not inject an NMI or interrupt if there is a pending
6797 	 * exception.  Exceptions and interrupts are recognized at
6798 	 * instruction boundaries, i.e. the start of an instruction.
6799 	 * Trap-like exceptions, e.g. #DB, have higher priority than
6800 	 * NMIs and interrupts, i.e. traps are recognized before an
6801 	 * NMI/interrupt that's pending on the same instruction.
6802 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
6803 	 * priority, but are only generated (pended) during instruction
6804 	 * execution, i.e. a pending fault-like exception means the
6805 	 * fault occurred on the *previous* instruction and must be
6806 	 * serviced prior to recognizing any new events in order to
6807 	 * fully complete the previous instruction.
6808 	 */
6809 	else if (!vcpu->arch.exception.pending) {
6810 		if (vcpu->arch.nmi_injected)
6811 			kvm_x86_ops->set_nmi(vcpu);
6812 		else if (vcpu->arch.interrupt.injected)
6813 			kvm_x86_ops->set_irq(vcpu);
6814 	}
6815 
6816 	/*
6817 	 * Call check_nested_events() even if we reinjected a previous event
6818 	 * in order for caller to determine if it should require immediate-exit
6819 	 * from L2 to L1 due to pending L1 events which require exit
6820 	 * from L2 to L1.
6821 	 */
6822 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6823 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6824 		if (r != 0)
6825 			return r;
6826 	}
6827 
6828 	/* try to inject new event if pending */
6829 	if (vcpu->arch.exception.pending) {
6830 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6831 					vcpu->arch.exception.has_error_code,
6832 					vcpu->arch.exception.error_code);
6833 
6834 		WARN_ON_ONCE(vcpu->arch.exception.injected);
6835 		vcpu->arch.exception.pending = false;
6836 		vcpu->arch.exception.injected = true;
6837 
6838 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6839 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6840 					     X86_EFLAGS_RF);
6841 
6842 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6843 		    (vcpu->arch.dr7 & DR7_GD)) {
6844 			vcpu->arch.dr7 &= ~DR7_GD;
6845 			kvm_update_dr7(vcpu);
6846 		}
6847 
6848 		kvm_x86_ops->queue_exception(vcpu);
6849 	}
6850 
6851 	/* Don't consider new event if we re-injected an event */
6852 	if (kvm_event_needs_reinjection(vcpu))
6853 		return 0;
6854 
6855 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
6856 	    kvm_x86_ops->smi_allowed(vcpu)) {
6857 		vcpu->arch.smi_pending = false;
6858 		++vcpu->arch.smi_count;
6859 		enter_smm(vcpu);
6860 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6861 		--vcpu->arch.nmi_pending;
6862 		vcpu->arch.nmi_injected = true;
6863 		kvm_x86_ops->set_nmi(vcpu);
6864 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6865 		/*
6866 		 * Because interrupts can be injected asynchronously, we are
6867 		 * calling check_nested_events again here to avoid a race condition.
6868 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6869 		 * proposal and current concerns.  Perhaps we should be setting
6870 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6871 		 */
6872 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6873 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6874 			if (r != 0)
6875 				return r;
6876 		}
6877 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6878 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6879 					    false);
6880 			kvm_x86_ops->set_irq(vcpu);
6881 		}
6882 	}
6883 
6884 	return 0;
6885 }
6886 
6887 static void process_nmi(struct kvm_vcpu *vcpu)
6888 {
6889 	unsigned limit = 2;
6890 
6891 	/*
6892 	 * x86 is limited to one NMI running, and one NMI pending after it.
6893 	 * If an NMI is already in progress, limit further NMIs to just one.
6894 	 * Otherwise, allow two (and we'll inject the first one immediately).
6895 	 */
6896 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6897 		limit = 1;
6898 
6899 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6900 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6901 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6902 }
6903 
6904 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6905 {
6906 	u32 flags = 0;
6907 	flags |= seg->g       << 23;
6908 	flags |= seg->db      << 22;
6909 	flags |= seg->l       << 21;
6910 	flags |= seg->avl     << 20;
6911 	flags |= seg->present << 15;
6912 	flags |= seg->dpl     << 13;
6913 	flags |= seg->s       << 12;
6914 	flags |= seg->type    << 8;
6915 	return flags;
6916 }
6917 
6918 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6919 {
6920 	struct kvm_segment seg;
6921 	int offset;
6922 
6923 	kvm_get_segment(vcpu, &seg, n);
6924 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6925 
6926 	if (n < 3)
6927 		offset = 0x7f84 + n * 12;
6928 	else
6929 		offset = 0x7f2c + (n - 3) * 12;
6930 
6931 	put_smstate(u32, buf, offset + 8, seg.base);
6932 	put_smstate(u32, buf, offset + 4, seg.limit);
6933 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6934 }
6935 
6936 #ifdef CONFIG_X86_64
6937 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6938 {
6939 	struct kvm_segment seg;
6940 	int offset;
6941 	u16 flags;
6942 
6943 	kvm_get_segment(vcpu, &seg, n);
6944 	offset = 0x7e00 + n * 16;
6945 
6946 	flags = enter_smm_get_segment_flags(&seg) >> 8;
6947 	put_smstate(u16, buf, offset, seg.selector);
6948 	put_smstate(u16, buf, offset + 2, flags);
6949 	put_smstate(u32, buf, offset + 4, seg.limit);
6950 	put_smstate(u64, buf, offset + 8, seg.base);
6951 }
6952 #endif
6953 
6954 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6955 {
6956 	struct desc_ptr dt;
6957 	struct kvm_segment seg;
6958 	unsigned long val;
6959 	int i;
6960 
6961 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6962 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6963 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6964 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6965 
6966 	for (i = 0; i < 8; i++)
6967 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6968 
6969 	kvm_get_dr(vcpu, 6, &val);
6970 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6971 	kvm_get_dr(vcpu, 7, &val);
6972 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6973 
6974 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6975 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6976 	put_smstate(u32, buf, 0x7f64, seg.base);
6977 	put_smstate(u32, buf, 0x7f60, seg.limit);
6978 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6979 
6980 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6981 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6982 	put_smstate(u32, buf, 0x7f80, seg.base);
6983 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6984 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6985 
6986 	kvm_x86_ops->get_gdt(vcpu, &dt);
6987 	put_smstate(u32, buf, 0x7f74, dt.address);
6988 	put_smstate(u32, buf, 0x7f70, dt.size);
6989 
6990 	kvm_x86_ops->get_idt(vcpu, &dt);
6991 	put_smstate(u32, buf, 0x7f58, dt.address);
6992 	put_smstate(u32, buf, 0x7f54, dt.size);
6993 
6994 	for (i = 0; i < 6; i++)
6995 		enter_smm_save_seg_32(vcpu, buf, i);
6996 
6997 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6998 
6999 	/* revision id */
7000 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7001 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7002 }
7003 
7004 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7005 {
7006 #ifdef CONFIG_X86_64
7007 	struct desc_ptr dt;
7008 	struct kvm_segment seg;
7009 	unsigned long val;
7010 	int i;
7011 
7012 	for (i = 0; i < 16; i++)
7013 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7014 
7015 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7016 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7017 
7018 	kvm_get_dr(vcpu, 6, &val);
7019 	put_smstate(u64, buf, 0x7f68, val);
7020 	kvm_get_dr(vcpu, 7, &val);
7021 	put_smstate(u64, buf, 0x7f60, val);
7022 
7023 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7024 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7025 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7026 
7027 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7028 
7029 	/* revision id */
7030 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7031 
7032 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7033 
7034 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7035 	put_smstate(u16, buf, 0x7e90, seg.selector);
7036 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7037 	put_smstate(u32, buf, 0x7e94, seg.limit);
7038 	put_smstate(u64, buf, 0x7e98, seg.base);
7039 
7040 	kvm_x86_ops->get_idt(vcpu, &dt);
7041 	put_smstate(u32, buf, 0x7e84, dt.size);
7042 	put_smstate(u64, buf, 0x7e88, dt.address);
7043 
7044 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7045 	put_smstate(u16, buf, 0x7e70, seg.selector);
7046 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7047 	put_smstate(u32, buf, 0x7e74, seg.limit);
7048 	put_smstate(u64, buf, 0x7e78, seg.base);
7049 
7050 	kvm_x86_ops->get_gdt(vcpu, &dt);
7051 	put_smstate(u32, buf, 0x7e64, dt.size);
7052 	put_smstate(u64, buf, 0x7e68, dt.address);
7053 
7054 	for (i = 0; i < 6; i++)
7055 		enter_smm_save_seg_64(vcpu, buf, i);
7056 #else
7057 	WARN_ON_ONCE(1);
7058 #endif
7059 }
7060 
7061 static void enter_smm(struct kvm_vcpu *vcpu)
7062 {
7063 	struct kvm_segment cs, ds;
7064 	struct desc_ptr dt;
7065 	char buf[512];
7066 	u32 cr0;
7067 
7068 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7069 	memset(buf, 0, 512);
7070 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7071 		enter_smm_save_state_64(vcpu, buf);
7072 	else
7073 		enter_smm_save_state_32(vcpu, buf);
7074 
7075 	/*
7076 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7077 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7078 	 * the SMM state-save area.
7079 	 */
7080 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7081 
7082 	vcpu->arch.hflags |= HF_SMM_MASK;
7083 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7084 
7085 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7086 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7087 	else
7088 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7089 
7090 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7091 	kvm_rip_write(vcpu, 0x8000);
7092 
7093 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7094 	kvm_x86_ops->set_cr0(vcpu, cr0);
7095 	vcpu->arch.cr0 = cr0;
7096 
7097 	kvm_x86_ops->set_cr4(vcpu, 0);
7098 
7099 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7100 	dt.address = dt.size = 0;
7101 	kvm_x86_ops->set_idt(vcpu, &dt);
7102 
7103 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7104 
7105 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7106 	cs.base = vcpu->arch.smbase;
7107 
7108 	ds.selector = 0;
7109 	ds.base = 0;
7110 
7111 	cs.limit    = ds.limit = 0xffffffff;
7112 	cs.type     = ds.type = 0x3;
7113 	cs.dpl      = ds.dpl = 0;
7114 	cs.db       = ds.db = 0;
7115 	cs.s        = ds.s = 1;
7116 	cs.l        = ds.l = 0;
7117 	cs.g        = ds.g = 1;
7118 	cs.avl      = ds.avl = 0;
7119 	cs.present  = ds.present = 1;
7120 	cs.unusable = ds.unusable = 0;
7121 	cs.padding  = ds.padding = 0;
7122 
7123 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7124 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7125 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7126 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7127 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7128 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7129 
7130 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7131 		kvm_x86_ops->set_efer(vcpu, 0);
7132 
7133 	kvm_update_cpuid(vcpu);
7134 	kvm_mmu_reset_context(vcpu);
7135 }
7136 
7137 static void process_smi(struct kvm_vcpu *vcpu)
7138 {
7139 	vcpu->arch.smi_pending = true;
7140 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7141 }
7142 
7143 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7144 {
7145 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7146 }
7147 
7148 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7149 {
7150 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7151 		return;
7152 
7153 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7154 
7155 	if (irqchip_split(vcpu->kvm))
7156 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7157 	else {
7158 		if (vcpu->arch.apicv_active)
7159 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7160 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7161 	}
7162 
7163 	if (is_guest_mode(vcpu))
7164 		vcpu->arch.load_eoi_exitmap_pending = true;
7165 	else
7166 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7167 }
7168 
7169 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7170 {
7171 	u64 eoi_exit_bitmap[4];
7172 
7173 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7174 		return;
7175 
7176 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7177 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7178 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7179 }
7180 
7181 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7182 		unsigned long start, unsigned long end)
7183 {
7184 	unsigned long apic_address;
7185 
7186 	/*
7187 	 * The physical address of apic access page is stored in the VMCS.
7188 	 * Update it when it becomes invalid.
7189 	 */
7190 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7191 	if (start <= apic_address && apic_address < end)
7192 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7193 }
7194 
7195 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7196 {
7197 	struct page *page = NULL;
7198 
7199 	if (!lapic_in_kernel(vcpu))
7200 		return;
7201 
7202 	if (!kvm_x86_ops->set_apic_access_page_addr)
7203 		return;
7204 
7205 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7206 	if (is_error_page(page))
7207 		return;
7208 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7209 
7210 	/*
7211 	 * Do not pin apic access page in memory, the MMU notifier
7212 	 * will call us again if it is migrated or swapped out.
7213 	 */
7214 	put_page(page);
7215 }
7216 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7217 
7218 /*
7219  * Returns 1 to let vcpu_run() continue the guest execution loop without
7220  * exiting to the userspace.  Otherwise, the value will be returned to the
7221  * userspace.
7222  */
7223 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7224 {
7225 	int r;
7226 	bool req_int_win =
7227 		dm_request_for_irq_injection(vcpu) &&
7228 		kvm_cpu_accept_dm_intr(vcpu);
7229 
7230 	bool req_immediate_exit = false;
7231 
7232 	if (kvm_request_pending(vcpu)) {
7233 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7234 			kvm_mmu_unload(vcpu);
7235 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7236 			__kvm_migrate_timers(vcpu);
7237 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7238 			kvm_gen_update_masterclock(vcpu->kvm);
7239 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7240 			kvm_gen_kvmclock_update(vcpu);
7241 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7242 			r = kvm_guest_time_update(vcpu);
7243 			if (unlikely(r))
7244 				goto out;
7245 		}
7246 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7247 			kvm_mmu_sync_roots(vcpu);
7248 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7249 			kvm_vcpu_flush_tlb(vcpu, true);
7250 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7251 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7252 			r = 0;
7253 			goto out;
7254 		}
7255 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7256 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7257 			vcpu->mmio_needed = 0;
7258 			r = 0;
7259 			goto out;
7260 		}
7261 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7262 			/* Page is swapped out. Do synthetic halt */
7263 			vcpu->arch.apf.halted = true;
7264 			r = 1;
7265 			goto out;
7266 		}
7267 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7268 			record_steal_time(vcpu);
7269 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7270 			process_smi(vcpu);
7271 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7272 			process_nmi(vcpu);
7273 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7274 			kvm_pmu_handle_event(vcpu);
7275 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7276 			kvm_pmu_deliver_pmi(vcpu);
7277 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7278 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7279 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7280 				     vcpu->arch.ioapic_handled_vectors)) {
7281 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7282 				vcpu->run->eoi.vector =
7283 						vcpu->arch.pending_ioapic_eoi;
7284 				r = 0;
7285 				goto out;
7286 			}
7287 		}
7288 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7289 			vcpu_scan_ioapic(vcpu);
7290 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7291 			vcpu_load_eoi_exitmap(vcpu);
7292 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7293 			kvm_vcpu_reload_apic_access_page(vcpu);
7294 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7295 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7296 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7297 			r = 0;
7298 			goto out;
7299 		}
7300 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7301 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7302 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7303 			r = 0;
7304 			goto out;
7305 		}
7306 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7307 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7308 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7309 			r = 0;
7310 			goto out;
7311 		}
7312 
7313 		/*
7314 		 * KVM_REQ_HV_STIMER has to be processed after
7315 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7316 		 * depend on the guest clock being up-to-date
7317 		 */
7318 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7319 			kvm_hv_process_stimers(vcpu);
7320 	}
7321 
7322 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7323 		++vcpu->stat.req_event;
7324 		kvm_apic_accept_events(vcpu);
7325 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7326 			r = 1;
7327 			goto out;
7328 		}
7329 
7330 		if (inject_pending_event(vcpu, req_int_win) != 0)
7331 			req_immediate_exit = true;
7332 		else {
7333 			/* Enable SMI/NMI/IRQ window open exits if needed.
7334 			 *
7335 			 * SMIs have three cases:
7336 			 * 1) They can be nested, and then there is nothing to
7337 			 *    do here because RSM will cause a vmexit anyway.
7338 			 * 2) There is an ISA-specific reason why SMI cannot be
7339 			 *    injected, and the moment when this changes can be
7340 			 *    intercepted.
7341 			 * 3) Or the SMI can be pending because
7342 			 *    inject_pending_event has completed the injection
7343 			 *    of an IRQ or NMI from the previous vmexit, and
7344 			 *    then we request an immediate exit to inject the
7345 			 *    SMI.
7346 			 */
7347 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7348 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7349 					req_immediate_exit = true;
7350 			if (vcpu->arch.nmi_pending)
7351 				kvm_x86_ops->enable_nmi_window(vcpu);
7352 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7353 				kvm_x86_ops->enable_irq_window(vcpu);
7354 			WARN_ON(vcpu->arch.exception.pending);
7355 		}
7356 
7357 		if (kvm_lapic_enabled(vcpu)) {
7358 			update_cr8_intercept(vcpu);
7359 			kvm_lapic_sync_to_vapic(vcpu);
7360 		}
7361 	}
7362 
7363 	r = kvm_mmu_reload(vcpu);
7364 	if (unlikely(r)) {
7365 		goto cancel_injection;
7366 	}
7367 
7368 	preempt_disable();
7369 
7370 	kvm_x86_ops->prepare_guest_switch(vcpu);
7371 
7372 	/*
7373 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7374 	 * IPI are then delayed after guest entry, which ensures that they
7375 	 * result in virtual interrupt delivery.
7376 	 */
7377 	local_irq_disable();
7378 	vcpu->mode = IN_GUEST_MODE;
7379 
7380 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7381 
7382 	/*
7383 	 * 1) We should set ->mode before checking ->requests.  Please see
7384 	 * the comment in kvm_vcpu_exiting_guest_mode().
7385 	 *
7386 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
7387 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7388 	 * (see vmx_deliver_posted_interrupt).
7389 	 *
7390 	 * 3) This also orders the write to mode from any reads to the page
7391 	 * tables done while the VCPU is running.  Please see the comment
7392 	 * in kvm_flush_remote_tlbs.
7393 	 */
7394 	smp_mb__after_srcu_read_unlock();
7395 
7396 	/*
7397 	 * This handles the case where a posted interrupt was
7398 	 * notified with kvm_vcpu_kick.
7399 	 */
7400 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7401 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7402 
7403 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7404 	    || need_resched() || signal_pending(current)) {
7405 		vcpu->mode = OUTSIDE_GUEST_MODE;
7406 		smp_wmb();
7407 		local_irq_enable();
7408 		preempt_enable();
7409 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7410 		r = 1;
7411 		goto cancel_injection;
7412 	}
7413 
7414 	kvm_load_guest_xcr0(vcpu);
7415 
7416 	if (req_immediate_exit) {
7417 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7418 		smp_send_reschedule(vcpu->cpu);
7419 	}
7420 
7421 	trace_kvm_entry(vcpu->vcpu_id);
7422 	if (lapic_timer_advance_ns)
7423 		wait_lapic_expire(vcpu);
7424 	guest_enter_irqoff();
7425 
7426 	if (unlikely(vcpu->arch.switch_db_regs)) {
7427 		set_debugreg(0, 7);
7428 		set_debugreg(vcpu->arch.eff_db[0], 0);
7429 		set_debugreg(vcpu->arch.eff_db[1], 1);
7430 		set_debugreg(vcpu->arch.eff_db[2], 2);
7431 		set_debugreg(vcpu->arch.eff_db[3], 3);
7432 		set_debugreg(vcpu->arch.dr6, 6);
7433 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7434 	}
7435 
7436 	kvm_x86_ops->run(vcpu);
7437 
7438 	/*
7439 	 * Do this here before restoring debug registers on the host.  And
7440 	 * since we do this before handling the vmexit, a DR access vmexit
7441 	 * can (a) read the correct value of the debug registers, (b) set
7442 	 * KVM_DEBUGREG_WONT_EXIT again.
7443 	 */
7444 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7445 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7446 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7447 		kvm_update_dr0123(vcpu);
7448 		kvm_update_dr6(vcpu);
7449 		kvm_update_dr7(vcpu);
7450 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7451 	}
7452 
7453 	/*
7454 	 * If the guest has used debug registers, at least dr7
7455 	 * will be disabled while returning to the host.
7456 	 * If we don't have active breakpoints in the host, we don't
7457 	 * care about the messed up debug address registers. But if
7458 	 * we have some of them active, restore the old state.
7459 	 */
7460 	if (hw_breakpoint_active())
7461 		hw_breakpoint_restore();
7462 
7463 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7464 
7465 	vcpu->mode = OUTSIDE_GUEST_MODE;
7466 	smp_wmb();
7467 
7468 	kvm_put_guest_xcr0(vcpu);
7469 
7470 	kvm_before_interrupt(vcpu);
7471 	kvm_x86_ops->handle_external_intr(vcpu);
7472 	kvm_after_interrupt(vcpu);
7473 
7474 	++vcpu->stat.exits;
7475 
7476 	guest_exit_irqoff();
7477 
7478 	local_irq_enable();
7479 	preempt_enable();
7480 
7481 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7482 
7483 	/*
7484 	 * Profile KVM exit RIPs:
7485 	 */
7486 	if (unlikely(prof_on == KVM_PROFILING)) {
7487 		unsigned long rip = kvm_rip_read(vcpu);
7488 		profile_hit(KVM_PROFILING, (void *)rip);
7489 	}
7490 
7491 	if (unlikely(vcpu->arch.tsc_always_catchup))
7492 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7493 
7494 	if (vcpu->arch.apic_attention)
7495 		kvm_lapic_sync_from_vapic(vcpu);
7496 
7497 	vcpu->arch.gpa_available = false;
7498 	r = kvm_x86_ops->handle_exit(vcpu);
7499 	return r;
7500 
7501 cancel_injection:
7502 	kvm_x86_ops->cancel_injection(vcpu);
7503 	if (unlikely(vcpu->arch.apic_attention))
7504 		kvm_lapic_sync_from_vapic(vcpu);
7505 out:
7506 	return r;
7507 }
7508 
7509 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7510 {
7511 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7512 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7513 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7514 		kvm_vcpu_block(vcpu);
7515 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7516 
7517 		if (kvm_x86_ops->post_block)
7518 			kvm_x86_ops->post_block(vcpu);
7519 
7520 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7521 			return 1;
7522 	}
7523 
7524 	kvm_apic_accept_events(vcpu);
7525 	switch(vcpu->arch.mp_state) {
7526 	case KVM_MP_STATE_HALTED:
7527 		vcpu->arch.pv.pv_unhalted = false;
7528 		vcpu->arch.mp_state =
7529 			KVM_MP_STATE_RUNNABLE;
7530 	case KVM_MP_STATE_RUNNABLE:
7531 		vcpu->arch.apf.halted = false;
7532 		break;
7533 	case KVM_MP_STATE_INIT_RECEIVED:
7534 		break;
7535 	default:
7536 		return -EINTR;
7537 		break;
7538 	}
7539 	return 1;
7540 }
7541 
7542 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7543 {
7544 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7545 		kvm_x86_ops->check_nested_events(vcpu, false);
7546 
7547 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7548 		!vcpu->arch.apf.halted);
7549 }
7550 
7551 static int vcpu_run(struct kvm_vcpu *vcpu)
7552 {
7553 	int r;
7554 	struct kvm *kvm = vcpu->kvm;
7555 
7556 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7557 
7558 	for (;;) {
7559 		if (kvm_vcpu_running(vcpu)) {
7560 			r = vcpu_enter_guest(vcpu);
7561 		} else {
7562 			r = vcpu_block(kvm, vcpu);
7563 		}
7564 
7565 		if (r <= 0)
7566 			break;
7567 
7568 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7569 		if (kvm_cpu_has_pending_timer(vcpu))
7570 			kvm_inject_pending_timer_irqs(vcpu);
7571 
7572 		if (dm_request_for_irq_injection(vcpu) &&
7573 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7574 			r = 0;
7575 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7576 			++vcpu->stat.request_irq_exits;
7577 			break;
7578 		}
7579 
7580 		kvm_check_async_pf_completion(vcpu);
7581 
7582 		if (signal_pending(current)) {
7583 			r = -EINTR;
7584 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7585 			++vcpu->stat.signal_exits;
7586 			break;
7587 		}
7588 		if (need_resched()) {
7589 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7590 			cond_resched();
7591 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7592 		}
7593 	}
7594 
7595 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7596 
7597 	return r;
7598 }
7599 
7600 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7601 {
7602 	int r;
7603 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7604 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7605 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7606 	if (r != EMULATE_DONE)
7607 		return 0;
7608 	return 1;
7609 }
7610 
7611 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7612 {
7613 	BUG_ON(!vcpu->arch.pio.count);
7614 
7615 	return complete_emulated_io(vcpu);
7616 }
7617 
7618 /*
7619  * Implements the following, as a state machine:
7620  *
7621  * read:
7622  *   for each fragment
7623  *     for each mmio piece in the fragment
7624  *       write gpa, len
7625  *       exit
7626  *       copy data
7627  *   execute insn
7628  *
7629  * write:
7630  *   for each fragment
7631  *     for each mmio piece in the fragment
7632  *       write gpa, len
7633  *       copy data
7634  *       exit
7635  */
7636 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7637 {
7638 	struct kvm_run *run = vcpu->run;
7639 	struct kvm_mmio_fragment *frag;
7640 	unsigned len;
7641 
7642 	BUG_ON(!vcpu->mmio_needed);
7643 
7644 	/* Complete previous fragment */
7645 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7646 	len = min(8u, frag->len);
7647 	if (!vcpu->mmio_is_write)
7648 		memcpy(frag->data, run->mmio.data, len);
7649 
7650 	if (frag->len <= 8) {
7651 		/* Switch to the next fragment. */
7652 		frag++;
7653 		vcpu->mmio_cur_fragment++;
7654 	} else {
7655 		/* Go forward to the next mmio piece. */
7656 		frag->data += len;
7657 		frag->gpa += len;
7658 		frag->len -= len;
7659 	}
7660 
7661 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7662 		vcpu->mmio_needed = 0;
7663 
7664 		/* FIXME: return into emulator if single-stepping.  */
7665 		if (vcpu->mmio_is_write)
7666 			return 1;
7667 		vcpu->mmio_read_completed = 1;
7668 		return complete_emulated_io(vcpu);
7669 	}
7670 
7671 	run->exit_reason = KVM_EXIT_MMIO;
7672 	run->mmio.phys_addr = frag->gpa;
7673 	if (vcpu->mmio_is_write)
7674 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7675 	run->mmio.len = min(8u, frag->len);
7676 	run->mmio.is_write = vcpu->mmio_is_write;
7677 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7678 	return 0;
7679 }
7680 
7681 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7682 {
7683 	int r;
7684 
7685 	vcpu_load(vcpu);
7686 	kvm_sigset_activate(vcpu);
7687 	kvm_load_guest_fpu(vcpu);
7688 
7689 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7690 		if (kvm_run->immediate_exit) {
7691 			r = -EINTR;
7692 			goto out;
7693 		}
7694 		kvm_vcpu_block(vcpu);
7695 		kvm_apic_accept_events(vcpu);
7696 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7697 		r = -EAGAIN;
7698 		if (signal_pending(current)) {
7699 			r = -EINTR;
7700 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7701 			++vcpu->stat.signal_exits;
7702 		}
7703 		goto out;
7704 	}
7705 
7706 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
7707 		r = -EINVAL;
7708 		goto out;
7709 	}
7710 
7711 	if (vcpu->run->kvm_dirty_regs) {
7712 		r = sync_regs(vcpu);
7713 		if (r != 0)
7714 			goto out;
7715 	}
7716 
7717 	/* re-sync apic's tpr */
7718 	if (!lapic_in_kernel(vcpu)) {
7719 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7720 			r = -EINVAL;
7721 			goto out;
7722 		}
7723 	}
7724 
7725 	if (unlikely(vcpu->arch.complete_userspace_io)) {
7726 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7727 		vcpu->arch.complete_userspace_io = NULL;
7728 		r = cui(vcpu);
7729 		if (r <= 0)
7730 			goto out;
7731 	} else
7732 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7733 
7734 	if (kvm_run->immediate_exit)
7735 		r = -EINTR;
7736 	else
7737 		r = vcpu_run(vcpu);
7738 
7739 out:
7740 	kvm_put_guest_fpu(vcpu);
7741 	if (vcpu->run->kvm_valid_regs)
7742 		store_regs(vcpu);
7743 	post_kvm_run_save(vcpu);
7744 	kvm_sigset_deactivate(vcpu);
7745 
7746 	vcpu_put(vcpu);
7747 	return r;
7748 }
7749 
7750 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7751 {
7752 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7753 		/*
7754 		 * We are here if userspace calls get_regs() in the middle of
7755 		 * instruction emulation. Registers state needs to be copied
7756 		 * back from emulation context to vcpu. Userspace shouldn't do
7757 		 * that usually, but some bad designed PV devices (vmware
7758 		 * backdoor interface) need this to work
7759 		 */
7760 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7761 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7762 	}
7763 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7764 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7765 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7766 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7767 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7768 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7769 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7770 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7771 #ifdef CONFIG_X86_64
7772 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7773 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7774 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7775 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7776 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7777 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7778 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7779 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7780 #endif
7781 
7782 	regs->rip = kvm_rip_read(vcpu);
7783 	regs->rflags = kvm_get_rflags(vcpu);
7784 }
7785 
7786 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7787 {
7788 	vcpu_load(vcpu);
7789 	__get_regs(vcpu, regs);
7790 	vcpu_put(vcpu);
7791 	return 0;
7792 }
7793 
7794 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7795 {
7796 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7797 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7798 
7799 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7800 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7801 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7802 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7803 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7804 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7805 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7806 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7807 #ifdef CONFIG_X86_64
7808 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7809 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7810 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7811 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7812 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7813 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7814 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7815 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7816 #endif
7817 
7818 	kvm_rip_write(vcpu, regs->rip);
7819 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7820 
7821 	vcpu->arch.exception.pending = false;
7822 
7823 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7824 }
7825 
7826 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7827 {
7828 	vcpu_load(vcpu);
7829 	__set_regs(vcpu, regs);
7830 	vcpu_put(vcpu);
7831 	return 0;
7832 }
7833 
7834 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7835 {
7836 	struct kvm_segment cs;
7837 
7838 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7839 	*db = cs.db;
7840 	*l = cs.l;
7841 }
7842 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7843 
7844 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7845 {
7846 	struct desc_ptr dt;
7847 
7848 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7849 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7850 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7851 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7852 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7853 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7854 
7855 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7856 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7857 
7858 	kvm_x86_ops->get_idt(vcpu, &dt);
7859 	sregs->idt.limit = dt.size;
7860 	sregs->idt.base = dt.address;
7861 	kvm_x86_ops->get_gdt(vcpu, &dt);
7862 	sregs->gdt.limit = dt.size;
7863 	sregs->gdt.base = dt.address;
7864 
7865 	sregs->cr0 = kvm_read_cr0(vcpu);
7866 	sregs->cr2 = vcpu->arch.cr2;
7867 	sregs->cr3 = kvm_read_cr3(vcpu);
7868 	sregs->cr4 = kvm_read_cr4(vcpu);
7869 	sregs->cr8 = kvm_get_cr8(vcpu);
7870 	sregs->efer = vcpu->arch.efer;
7871 	sregs->apic_base = kvm_get_apic_base(vcpu);
7872 
7873 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7874 
7875 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
7876 		set_bit(vcpu->arch.interrupt.nr,
7877 			(unsigned long *)sregs->interrupt_bitmap);
7878 }
7879 
7880 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7881 				  struct kvm_sregs *sregs)
7882 {
7883 	vcpu_load(vcpu);
7884 	__get_sregs(vcpu, sregs);
7885 	vcpu_put(vcpu);
7886 	return 0;
7887 }
7888 
7889 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7890 				    struct kvm_mp_state *mp_state)
7891 {
7892 	vcpu_load(vcpu);
7893 
7894 	kvm_apic_accept_events(vcpu);
7895 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7896 					vcpu->arch.pv.pv_unhalted)
7897 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7898 	else
7899 		mp_state->mp_state = vcpu->arch.mp_state;
7900 
7901 	vcpu_put(vcpu);
7902 	return 0;
7903 }
7904 
7905 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7906 				    struct kvm_mp_state *mp_state)
7907 {
7908 	int ret = -EINVAL;
7909 
7910 	vcpu_load(vcpu);
7911 
7912 	if (!lapic_in_kernel(vcpu) &&
7913 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7914 		goto out;
7915 
7916 	/* INITs are latched while in SMM */
7917 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7918 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7919 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7920 		goto out;
7921 
7922 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7923 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7924 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7925 	} else
7926 		vcpu->arch.mp_state = mp_state->mp_state;
7927 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7928 
7929 	ret = 0;
7930 out:
7931 	vcpu_put(vcpu);
7932 	return ret;
7933 }
7934 
7935 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7936 		    int reason, bool has_error_code, u32 error_code)
7937 {
7938 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7939 	int ret;
7940 
7941 	init_emulate_ctxt(vcpu);
7942 
7943 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7944 				   has_error_code, error_code);
7945 
7946 	if (ret)
7947 		return EMULATE_FAIL;
7948 
7949 	kvm_rip_write(vcpu, ctxt->eip);
7950 	kvm_set_rflags(vcpu, ctxt->eflags);
7951 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7952 	return EMULATE_DONE;
7953 }
7954 EXPORT_SYMBOL_GPL(kvm_task_switch);
7955 
7956 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7957 {
7958 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
7959 		/*
7960 		 * When EFER.LME and CR0.PG are set, the processor is in
7961 		 * 64-bit mode (though maybe in a 32-bit code segment).
7962 		 * CR4.PAE and EFER.LMA must be set.
7963 		 */
7964 		if (!(sregs->cr4 & X86_CR4_PAE)
7965 		    || !(sregs->efer & EFER_LMA))
7966 			return -EINVAL;
7967 	} else {
7968 		/*
7969 		 * Not in 64-bit mode: EFER.LMA is clear and the code
7970 		 * segment cannot be 64-bit.
7971 		 */
7972 		if (sregs->efer & EFER_LMA || sregs->cs.l)
7973 			return -EINVAL;
7974 	}
7975 
7976 	return 0;
7977 }
7978 
7979 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7980 {
7981 	struct msr_data apic_base_msr;
7982 	int mmu_reset_needed = 0;
7983 	int pending_vec, max_bits, idx;
7984 	struct desc_ptr dt;
7985 	int ret = -EINVAL;
7986 
7987 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7988 			(sregs->cr4 & X86_CR4_OSXSAVE))
7989 		goto out;
7990 
7991 	if (kvm_valid_sregs(vcpu, sregs))
7992 		goto out;
7993 
7994 	apic_base_msr.data = sregs->apic_base;
7995 	apic_base_msr.host_initiated = true;
7996 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
7997 		goto out;
7998 
7999 	dt.size = sregs->idt.limit;
8000 	dt.address = sregs->idt.base;
8001 	kvm_x86_ops->set_idt(vcpu, &dt);
8002 	dt.size = sregs->gdt.limit;
8003 	dt.address = sregs->gdt.base;
8004 	kvm_x86_ops->set_gdt(vcpu, &dt);
8005 
8006 	vcpu->arch.cr2 = sregs->cr2;
8007 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8008 	vcpu->arch.cr3 = sregs->cr3;
8009 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8010 
8011 	kvm_set_cr8(vcpu, sregs->cr8);
8012 
8013 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8014 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8015 
8016 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8017 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8018 	vcpu->arch.cr0 = sregs->cr0;
8019 
8020 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8021 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8022 	if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
8023 		kvm_update_cpuid(vcpu);
8024 
8025 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8026 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
8027 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8028 		mmu_reset_needed = 1;
8029 	}
8030 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8031 
8032 	if (mmu_reset_needed)
8033 		kvm_mmu_reset_context(vcpu);
8034 
8035 	max_bits = KVM_NR_INTERRUPTS;
8036 	pending_vec = find_first_bit(
8037 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8038 	if (pending_vec < max_bits) {
8039 		kvm_queue_interrupt(vcpu, pending_vec, false);
8040 		pr_debug("Set back pending irq %d\n", pending_vec);
8041 	}
8042 
8043 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8044 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8045 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8046 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8047 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8048 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8049 
8050 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8051 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8052 
8053 	update_cr8_intercept(vcpu);
8054 
8055 	/* Older userspace won't unhalt the vcpu on reset. */
8056 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8057 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8058 	    !is_protmode(vcpu))
8059 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8060 
8061 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8062 
8063 	ret = 0;
8064 out:
8065 	return ret;
8066 }
8067 
8068 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8069 				  struct kvm_sregs *sregs)
8070 {
8071 	int ret;
8072 
8073 	vcpu_load(vcpu);
8074 	ret = __set_sregs(vcpu, sregs);
8075 	vcpu_put(vcpu);
8076 	return ret;
8077 }
8078 
8079 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8080 					struct kvm_guest_debug *dbg)
8081 {
8082 	unsigned long rflags;
8083 	int i, r;
8084 
8085 	vcpu_load(vcpu);
8086 
8087 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8088 		r = -EBUSY;
8089 		if (vcpu->arch.exception.pending)
8090 			goto out;
8091 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8092 			kvm_queue_exception(vcpu, DB_VECTOR);
8093 		else
8094 			kvm_queue_exception(vcpu, BP_VECTOR);
8095 	}
8096 
8097 	/*
8098 	 * Read rflags as long as potentially injected trace flags are still
8099 	 * filtered out.
8100 	 */
8101 	rflags = kvm_get_rflags(vcpu);
8102 
8103 	vcpu->guest_debug = dbg->control;
8104 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8105 		vcpu->guest_debug = 0;
8106 
8107 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8108 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8109 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8110 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8111 	} else {
8112 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8113 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8114 	}
8115 	kvm_update_dr7(vcpu);
8116 
8117 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8118 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8119 			get_segment_base(vcpu, VCPU_SREG_CS);
8120 
8121 	/*
8122 	 * Trigger an rflags update that will inject or remove the trace
8123 	 * flags.
8124 	 */
8125 	kvm_set_rflags(vcpu, rflags);
8126 
8127 	kvm_x86_ops->update_bp_intercept(vcpu);
8128 
8129 	r = 0;
8130 
8131 out:
8132 	vcpu_put(vcpu);
8133 	return r;
8134 }
8135 
8136 /*
8137  * Translate a guest virtual address to a guest physical address.
8138  */
8139 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8140 				    struct kvm_translation *tr)
8141 {
8142 	unsigned long vaddr = tr->linear_address;
8143 	gpa_t gpa;
8144 	int idx;
8145 
8146 	vcpu_load(vcpu);
8147 
8148 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8149 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8150 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8151 	tr->physical_address = gpa;
8152 	tr->valid = gpa != UNMAPPED_GVA;
8153 	tr->writeable = 1;
8154 	tr->usermode = 0;
8155 
8156 	vcpu_put(vcpu);
8157 	return 0;
8158 }
8159 
8160 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8161 {
8162 	struct fxregs_state *fxsave;
8163 
8164 	vcpu_load(vcpu);
8165 
8166 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8167 	memcpy(fpu->fpr, fxsave->st_space, 128);
8168 	fpu->fcw = fxsave->cwd;
8169 	fpu->fsw = fxsave->swd;
8170 	fpu->ftwx = fxsave->twd;
8171 	fpu->last_opcode = fxsave->fop;
8172 	fpu->last_ip = fxsave->rip;
8173 	fpu->last_dp = fxsave->rdp;
8174 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
8175 
8176 	vcpu_put(vcpu);
8177 	return 0;
8178 }
8179 
8180 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8181 {
8182 	struct fxregs_state *fxsave;
8183 
8184 	vcpu_load(vcpu);
8185 
8186 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8187 
8188 	memcpy(fxsave->st_space, fpu->fpr, 128);
8189 	fxsave->cwd = fpu->fcw;
8190 	fxsave->swd = fpu->fsw;
8191 	fxsave->twd = fpu->ftwx;
8192 	fxsave->fop = fpu->last_opcode;
8193 	fxsave->rip = fpu->last_ip;
8194 	fxsave->rdp = fpu->last_dp;
8195 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
8196 
8197 	vcpu_put(vcpu);
8198 	return 0;
8199 }
8200 
8201 static void store_regs(struct kvm_vcpu *vcpu)
8202 {
8203 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8204 
8205 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8206 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8207 
8208 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8209 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8210 
8211 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8212 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8213 				vcpu, &vcpu->run->s.regs.events);
8214 }
8215 
8216 static int sync_regs(struct kvm_vcpu *vcpu)
8217 {
8218 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8219 		return -EINVAL;
8220 
8221 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8222 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8223 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8224 	}
8225 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8226 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8227 			return -EINVAL;
8228 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8229 	}
8230 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8231 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8232 				vcpu, &vcpu->run->s.regs.events))
8233 			return -EINVAL;
8234 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8235 	}
8236 
8237 	return 0;
8238 }
8239 
8240 static void fx_init(struct kvm_vcpu *vcpu)
8241 {
8242 	fpstate_init(&vcpu->arch.guest_fpu.state);
8243 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8244 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8245 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8246 
8247 	/*
8248 	 * Ensure guest xcr0 is valid for loading
8249 	 */
8250 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8251 
8252 	vcpu->arch.cr0 |= X86_CR0_ET;
8253 }
8254 
8255 /* Swap (qemu) user FPU context for the guest FPU context. */
8256 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8257 {
8258 	preempt_disable();
8259 	copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
8260 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8261 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8262 				~XFEATURE_MASK_PKRU);
8263 	preempt_enable();
8264 	trace_kvm_fpu(1);
8265 }
8266 
8267 /* When vcpu_run ends, restore user space FPU context. */
8268 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8269 {
8270 	preempt_disable();
8271 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8272 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8273 	preempt_enable();
8274 	++vcpu->stat.fpu_reload;
8275 	trace_kvm_fpu(0);
8276 }
8277 
8278 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8279 {
8280 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8281 
8282 	kvmclock_reset(vcpu);
8283 
8284 	kvm_x86_ops->vcpu_free(vcpu);
8285 	free_cpumask_var(wbinvd_dirty_mask);
8286 }
8287 
8288 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8289 						unsigned int id)
8290 {
8291 	struct kvm_vcpu *vcpu;
8292 
8293 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8294 		printk_once(KERN_WARNING
8295 		"kvm: SMP vm created on host with unstable TSC; "
8296 		"guest TSC will not be reliable\n");
8297 
8298 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8299 
8300 	return vcpu;
8301 }
8302 
8303 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8304 {
8305 	kvm_vcpu_mtrr_init(vcpu);
8306 	vcpu_load(vcpu);
8307 	kvm_vcpu_reset(vcpu, false);
8308 	kvm_mmu_setup(vcpu);
8309 	vcpu_put(vcpu);
8310 	return 0;
8311 }
8312 
8313 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8314 {
8315 	struct msr_data msr;
8316 	struct kvm *kvm = vcpu->kvm;
8317 
8318 	kvm_hv_vcpu_postcreate(vcpu);
8319 
8320 	if (mutex_lock_killable(&vcpu->mutex))
8321 		return;
8322 	vcpu_load(vcpu);
8323 	msr.data = 0x0;
8324 	msr.index = MSR_IA32_TSC;
8325 	msr.host_initiated = true;
8326 	kvm_write_tsc(vcpu, &msr);
8327 	vcpu_put(vcpu);
8328 	mutex_unlock(&vcpu->mutex);
8329 
8330 	if (!kvmclock_periodic_sync)
8331 		return;
8332 
8333 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8334 					KVMCLOCK_SYNC_PERIOD);
8335 }
8336 
8337 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8338 {
8339 	vcpu->arch.apf.msr_val = 0;
8340 
8341 	vcpu_load(vcpu);
8342 	kvm_mmu_unload(vcpu);
8343 	vcpu_put(vcpu);
8344 
8345 	kvm_x86_ops->vcpu_free(vcpu);
8346 }
8347 
8348 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8349 {
8350 	kvm_lapic_reset(vcpu, init_event);
8351 
8352 	vcpu->arch.hflags = 0;
8353 
8354 	vcpu->arch.smi_pending = 0;
8355 	vcpu->arch.smi_count = 0;
8356 	atomic_set(&vcpu->arch.nmi_queued, 0);
8357 	vcpu->arch.nmi_pending = 0;
8358 	vcpu->arch.nmi_injected = false;
8359 	kvm_clear_interrupt_queue(vcpu);
8360 	kvm_clear_exception_queue(vcpu);
8361 	vcpu->arch.exception.pending = false;
8362 
8363 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8364 	kvm_update_dr0123(vcpu);
8365 	vcpu->arch.dr6 = DR6_INIT;
8366 	kvm_update_dr6(vcpu);
8367 	vcpu->arch.dr7 = DR7_FIXED_1;
8368 	kvm_update_dr7(vcpu);
8369 
8370 	vcpu->arch.cr2 = 0;
8371 
8372 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8373 	vcpu->arch.apf.msr_val = 0;
8374 	vcpu->arch.st.msr_val = 0;
8375 
8376 	kvmclock_reset(vcpu);
8377 
8378 	kvm_clear_async_pf_completion_queue(vcpu);
8379 	kvm_async_pf_hash_reset(vcpu);
8380 	vcpu->arch.apf.halted = false;
8381 
8382 	if (kvm_mpx_supported()) {
8383 		void *mpx_state_buffer;
8384 
8385 		/*
8386 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8387 		 * called with loaded FPU and does not let userspace fix the state.
8388 		 */
8389 		if (init_event)
8390 			kvm_put_guest_fpu(vcpu);
8391 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8392 					XFEATURE_MASK_BNDREGS);
8393 		if (mpx_state_buffer)
8394 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8395 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8396 					XFEATURE_MASK_BNDCSR);
8397 		if (mpx_state_buffer)
8398 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8399 		if (init_event)
8400 			kvm_load_guest_fpu(vcpu);
8401 	}
8402 
8403 	if (!init_event) {
8404 		kvm_pmu_reset(vcpu);
8405 		vcpu->arch.smbase = 0x30000;
8406 
8407 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8408 		vcpu->arch.msr_misc_features_enables = 0;
8409 
8410 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8411 	}
8412 
8413 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8414 	vcpu->arch.regs_avail = ~0;
8415 	vcpu->arch.regs_dirty = ~0;
8416 
8417 	vcpu->arch.ia32_xss = 0;
8418 
8419 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8420 }
8421 
8422 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8423 {
8424 	struct kvm_segment cs;
8425 
8426 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8427 	cs.selector = vector << 8;
8428 	cs.base = vector << 12;
8429 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8430 	kvm_rip_write(vcpu, 0);
8431 }
8432 
8433 int kvm_arch_hardware_enable(void)
8434 {
8435 	struct kvm *kvm;
8436 	struct kvm_vcpu *vcpu;
8437 	int i;
8438 	int ret;
8439 	u64 local_tsc;
8440 	u64 max_tsc = 0;
8441 	bool stable, backwards_tsc = false;
8442 
8443 	kvm_shared_msr_cpu_online();
8444 	ret = kvm_x86_ops->hardware_enable();
8445 	if (ret != 0)
8446 		return ret;
8447 
8448 	local_tsc = rdtsc();
8449 	stable = !kvm_check_tsc_unstable();
8450 	list_for_each_entry(kvm, &vm_list, vm_list) {
8451 		kvm_for_each_vcpu(i, vcpu, kvm) {
8452 			if (!stable && vcpu->cpu == smp_processor_id())
8453 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8454 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8455 				backwards_tsc = true;
8456 				if (vcpu->arch.last_host_tsc > max_tsc)
8457 					max_tsc = vcpu->arch.last_host_tsc;
8458 			}
8459 		}
8460 	}
8461 
8462 	/*
8463 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8464 	 * platforms that reset TSC during suspend or hibernate actions, but
8465 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8466 	 * detect that condition here, which happens early in CPU bringup,
8467 	 * before any KVM threads can be running.  Unfortunately, we can't
8468 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8469 	 * enough into CPU bringup that we know how much real time has actually
8470 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8471 	 * variables that haven't been updated yet.
8472 	 *
8473 	 * So we simply find the maximum observed TSC above, then record the
8474 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8475 	 * the adjustment will be applied.  Note that we accumulate
8476 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8477 	 * gets a chance to run again.  In the event that no KVM threads get a
8478 	 * chance to run, we will miss the entire elapsed period, as we'll have
8479 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8480 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8481 	 * uniform across all VCPUs (not to mention the scenario is extremely
8482 	 * unlikely). It is possible that a second hibernate recovery happens
8483 	 * much faster than a first, causing the observed TSC here to be
8484 	 * smaller; this would require additional padding adjustment, which is
8485 	 * why we set last_host_tsc to the local tsc observed here.
8486 	 *
8487 	 * N.B. - this code below runs only on platforms with reliable TSC,
8488 	 * as that is the only way backwards_tsc is set above.  Also note
8489 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8490 	 * have the same delta_cyc adjustment applied if backwards_tsc
8491 	 * is detected.  Note further, this adjustment is only done once,
8492 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8493 	 * called multiple times (one for each physical CPU bringup).
8494 	 *
8495 	 * Platforms with unreliable TSCs don't have to deal with this, they
8496 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8497 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8498 	 * guarantee that they stay in perfect synchronization.
8499 	 */
8500 	if (backwards_tsc) {
8501 		u64 delta_cyc = max_tsc - local_tsc;
8502 		list_for_each_entry(kvm, &vm_list, vm_list) {
8503 			kvm->arch.backwards_tsc_observed = true;
8504 			kvm_for_each_vcpu(i, vcpu, kvm) {
8505 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8506 				vcpu->arch.last_host_tsc = local_tsc;
8507 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8508 			}
8509 
8510 			/*
8511 			 * We have to disable TSC offset matching.. if you were
8512 			 * booting a VM while issuing an S4 host suspend....
8513 			 * you may have some problem.  Solving this issue is
8514 			 * left as an exercise to the reader.
8515 			 */
8516 			kvm->arch.last_tsc_nsec = 0;
8517 			kvm->arch.last_tsc_write = 0;
8518 		}
8519 
8520 	}
8521 	return 0;
8522 }
8523 
8524 void kvm_arch_hardware_disable(void)
8525 {
8526 	kvm_x86_ops->hardware_disable();
8527 	drop_user_return_notifiers();
8528 }
8529 
8530 int kvm_arch_hardware_setup(void)
8531 {
8532 	int r;
8533 
8534 	r = kvm_x86_ops->hardware_setup();
8535 	if (r != 0)
8536 		return r;
8537 
8538 	if (kvm_has_tsc_control) {
8539 		/*
8540 		 * Make sure the user can only configure tsc_khz values that
8541 		 * fit into a signed integer.
8542 		 * A min value is not calculated needed because it will always
8543 		 * be 1 on all machines.
8544 		 */
8545 		u64 max = min(0x7fffffffULL,
8546 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8547 		kvm_max_guest_tsc_khz = max;
8548 
8549 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8550 	}
8551 
8552 	kvm_init_msr_list();
8553 	return 0;
8554 }
8555 
8556 void kvm_arch_hardware_unsetup(void)
8557 {
8558 	kvm_x86_ops->hardware_unsetup();
8559 }
8560 
8561 void kvm_arch_check_processor_compat(void *rtn)
8562 {
8563 	kvm_x86_ops->check_processor_compatibility(rtn);
8564 }
8565 
8566 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8567 {
8568 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8569 }
8570 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8571 
8572 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8573 {
8574 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8575 }
8576 
8577 struct static_key kvm_no_apic_vcpu __read_mostly;
8578 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8579 
8580 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8581 {
8582 	struct page *page;
8583 	int r;
8584 
8585 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8586 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8587 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8588 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8589 	else
8590 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8591 
8592 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8593 	if (!page) {
8594 		r = -ENOMEM;
8595 		goto fail;
8596 	}
8597 	vcpu->arch.pio_data = page_address(page);
8598 
8599 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
8600 
8601 	r = kvm_mmu_create(vcpu);
8602 	if (r < 0)
8603 		goto fail_free_pio_data;
8604 
8605 	if (irqchip_in_kernel(vcpu->kvm)) {
8606 		r = kvm_create_lapic(vcpu);
8607 		if (r < 0)
8608 			goto fail_mmu_destroy;
8609 	} else
8610 		static_key_slow_inc(&kvm_no_apic_vcpu);
8611 
8612 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8613 				       GFP_KERNEL);
8614 	if (!vcpu->arch.mce_banks) {
8615 		r = -ENOMEM;
8616 		goto fail_free_lapic;
8617 	}
8618 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8619 
8620 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8621 		r = -ENOMEM;
8622 		goto fail_free_mce_banks;
8623 	}
8624 
8625 	fx_init(vcpu);
8626 
8627 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8628 
8629 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8630 
8631 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8632 
8633 	kvm_async_pf_hash_reset(vcpu);
8634 	kvm_pmu_init(vcpu);
8635 
8636 	vcpu->arch.pending_external_vector = -1;
8637 	vcpu->arch.preempted_in_kernel = false;
8638 
8639 	kvm_hv_vcpu_init(vcpu);
8640 
8641 	return 0;
8642 
8643 fail_free_mce_banks:
8644 	kfree(vcpu->arch.mce_banks);
8645 fail_free_lapic:
8646 	kvm_free_lapic(vcpu);
8647 fail_mmu_destroy:
8648 	kvm_mmu_destroy(vcpu);
8649 fail_free_pio_data:
8650 	free_page((unsigned long)vcpu->arch.pio_data);
8651 fail:
8652 	return r;
8653 }
8654 
8655 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8656 {
8657 	int idx;
8658 
8659 	kvm_hv_vcpu_uninit(vcpu);
8660 	kvm_pmu_destroy(vcpu);
8661 	kfree(vcpu->arch.mce_banks);
8662 	kvm_free_lapic(vcpu);
8663 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8664 	kvm_mmu_destroy(vcpu);
8665 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8666 	free_page((unsigned long)vcpu->arch.pio_data);
8667 	if (!lapic_in_kernel(vcpu))
8668 		static_key_slow_dec(&kvm_no_apic_vcpu);
8669 }
8670 
8671 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8672 {
8673 	kvm_x86_ops->sched_in(vcpu, cpu);
8674 }
8675 
8676 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8677 {
8678 	if (type)
8679 		return -EINVAL;
8680 
8681 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8682 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8683 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8684 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8685 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8686 
8687 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8688 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8689 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8690 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8691 		&kvm->arch.irq_sources_bitmap);
8692 
8693 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8694 	mutex_init(&kvm->arch.apic_map_lock);
8695 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8696 
8697 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8698 	pvclock_update_vm_gtod_copy(kvm);
8699 
8700 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8701 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8702 
8703 	kvm_hv_init_vm(kvm);
8704 	kvm_page_track_init(kvm);
8705 	kvm_mmu_init_vm(kvm);
8706 
8707 	if (kvm_x86_ops->vm_init)
8708 		return kvm_x86_ops->vm_init(kvm);
8709 
8710 	return 0;
8711 }
8712 
8713 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8714 {
8715 	vcpu_load(vcpu);
8716 	kvm_mmu_unload(vcpu);
8717 	vcpu_put(vcpu);
8718 }
8719 
8720 static void kvm_free_vcpus(struct kvm *kvm)
8721 {
8722 	unsigned int i;
8723 	struct kvm_vcpu *vcpu;
8724 
8725 	/*
8726 	 * Unpin any mmu pages first.
8727 	 */
8728 	kvm_for_each_vcpu(i, vcpu, kvm) {
8729 		kvm_clear_async_pf_completion_queue(vcpu);
8730 		kvm_unload_vcpu_mmu(vcpu);
8731 	}
8732 	kvm_for_each_vcpu(i, vcpu, kvm)
8733 		kvm_arch_vcpu_free(vcpu);
8734 
8735 	mutex_lock(&kvm->lock);
8736 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8737 		kvm->vcpus[i] = NULL;
8738 
8739 	atomic_set(&kvm->online_vcpus, 0);
8740 	mutex_unlock(&kvm->lock);
8741 }
8742 
8743 void kvm_arch_sync_events(struct kvm *kvm)
8744 {
8745 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8746 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8747 	kvm_free_pit(kvm);
8748 }
8749 
8750 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8751 {
8752 	int i, r;
8753 	unsigned long hva;
8754 	struct kvm_memslots *slots = kvm_memslots(kvm);
8755 	struct kvm_memory_slot *slot, old;
8756 
8757 	/* Called with kvm->slots_lock held.  */
8758 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8759 		return -EINVAL;
8760 
8761 	slot = id_to_memslot(slots, id);
8762 	if (size) {
8763 		if (slot->npages)
8764 			return -EEXIST;
8765 
8766 		/*
8767 		 * MAP_SHARED to prevent internal slot pages from being moved
8768 		 * by fork()/COW.
8769 		 */
8770 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8771 			      MAP_SHARED | MAP_ANONYMOUS, 0);
8772 		if (IS_ERR((void *)hva))
8773 			return PTR_ERR((void *)hva);
8774 	} else {
8775 		if (!slot->npages)
8776 			return 0;
8777 
8778 		hva = 0;
8779 	}
8780 
8781 	old = *slot;
8782 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8783 		struct kvm_userspace_memory_region m;
8784 
8785 		m.slot = id | (i << 16);
8786 		m.flags = 0;
8787 		m.guest_phys_addr = gpa;
8788 		m.userspace_addr = hva;
8789 		m.memory_size = size;
8790 		r = __kvm_set_memory_region(kvm, &m);
8791 		if (r < 0)
8792 			return r;
8793 	}
8794 
8795 	if (!size)
8796 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8797 
8798 	return 0;
8799 }
8800 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8801 
8802 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8803 {
8804 	int r;
8805 
8806 	mutex_lock(&kvm->slots_lock);
8807 	r = __x86_set_memory_region(kvm, id, gpa, size);
8808 	mutex_unlock(&kvm->slots_lock);
8809 
8810 	return r;
8811 }
8812 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8813 
8814 void kvm_arch_destroy_vm(struct kvm *kvm)
8815 {
8816 	if (current->mm == kvm->mm) {
8817 		/*
8818 		 * Free memory regions allocated on behalf of userspace,
8819 		 * unless the the memory map has changed due to process exit
8820 		 * or fd copying.
8821 		 */
8822 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8823 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8824 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8825 	}
8826 	if (kvm_x86_ops->vm_destroy)
8827 		kvm_x86_ops->vm_destroy(kvm);
8828 	kvm_pic_destroy(kvm);
8829 	kvm_ioapic_destroy(kvm);
8830 	kvm_free_vcpus(kvm);
8831 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8832 	kvm_mmu_uninit_vm(kvm);
8833 	kvm_page_track_cleanup(kvm);
8834 	kvm_hv_destroy_vm(kvm);
8835 }
8836 
8837 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8838 			   struct kvm_memory_slot *dont)
8839 {
8840 	int i;
8841 
8842 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8843 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8844 			kvfree(free->arch.rmap[i]);
8845 			free->arch.rmap[i] = NULL;
8846 		}
8847 		if (i == 0)
8848 			continue;
8849 
8850 		if (!dont || free->arch.lpage_info[i - 1] !=
8851 			     dont->arch.lpage_info[i - 1]) {
8852 			kvfree(free->arch.lpage_info[i - 1]);
8853 			free->arch.lpage_info[i - 1] = NULL;
8854 		}
8855 	}
8856 
8857 	kvm_page_track_free_memslot(free, dont);
8858 }
8859 
8860 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8861 			    unsigned long npages)
8862 {
8863 	int i;
8864 
8865 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8866 		struct kvm_lpage_info *linfo;
8867 		unsigned long ugfn;
8868 		int lpages;
8869 		int level = i + 1;
8870 
8871 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
8872 				      slot->base_gfn, level) + 1;
8873 
8874 		slot->arch.rmap[i] =
8875 			kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8876 		if (!slot->arch.rmap[i])
8877 			goto out_free;
8878 		if (i == 0)
8879 			continue;
8880 
8881 		linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8882 		if (!linfo)
8883 			goto out_free;
8884 
8885 		slot->arch.lpage_info[i - 1] = linfo;
8886 
8887 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8888 			linfo[0].disallow_lpage = 1;
8889 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8890 			linfo[lpages - 1].disallow_lpage = 1;
8891 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
8892 		/*
8893 		 * If the gfn and userspace address are not aligned wrt each
8894 		 * other, or if explicitly asked to, disable large page
8895 		 * support for this slot
8896 		 */
8897 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8898 		    !kvm_largepages_enabled()) {
8899 			unsigned long j;
8900 
8901 			for (j = 0; j < lpages; ++j)
8902 				linfo[j].disallow_lpage = 1;
8903 		}
8904 	}
8905 
8906 	if (kvm_page_track_create_memslot(slot, npages))
8907 		goto out_free;
8908 
8909 	return 0;
8910 
8911 out_free:
8912 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8913 		kvfree(slot->arch.rmap[i]);
8914 		slot->arch.rmap[i] = NULL;
8915 		if (i == 0)
8916 			continue;
8917 
8918 		kvfree(slot->arch.lpage_info[i - 1]);
8919 		slot->arch.lpage_info[i - 1] = NULL;
8920 	}
8921 	return -ENOMEM;
8922 }
8923 
8924 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8925 {
8926 	/*
8927 	 * memslots->generation has been incremented.
8928 	 * mmio generation may have reached its maximum value.
8929 	 */
8930 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8931 }
8932 
8933 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8934 				struct kvm_memory_slot *memslot,
8935 				const struct kvm_userspace_memory_region *mem,
8936 				enum kvm_mr_change change)
8937 {
8938 	return 0;
8939 }
8940 
8941 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8942 				     struct kvm_memory_slot *new)
8943 {
8944 	/* Still write protect RO slot */
8945 	if (new->flags & KVM_MEM_READONLY) {
8946 		kvm_mmu_slot_remove_write_access(kvm, new);
8947 		return;
8948 	}
8949 
8950 	/*
8951 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
8952 	 *
8953 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
8954 	 *
8955 	 *  - KVM_MR_CREATE with dirty logging is disabled
8956 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8957 	 *
8958 	 * The reason is, in case of PML, we need to set D-bit for any slots
8959 	 * with dirty logging disabled in order to eliminate unnecessary GPA
8960 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
8961 	 * guarantees leaving PML enabled during guest's lifetime won't have
8962 	 * any additonal overhead from PML when guest is running with dirty
8963 	 * logging disabled for memory slots.
8964 	 *
8965 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8966 	 * to dirty logging mode.
8967 	 *
8968 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8969 	 *
8970 	 * In case of write protect:
8971 	 *
8972 	 * Write protect all pages for dirty logging.
8973 	 *
8974 	 * All the sptes including the large sptes which point to this
8975 	 * slot are set to readonly. We can not create any new large
8976 	 * spte on this slot until the end of the logging.
8977 	 *
8978 	 * See the comments in fast_page_fault().
8979 	 */
8980 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8981 		if (kvm_x86_ops->slot_enable_log_dirty)
8982 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8983 		else
8984 			kvm_mmu_slot_remove_write_access(kvm, new);
8985 	} else {
8986 		if (kvm_x86_ops->slot_disable_log_dirty)
8987 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8988 	}
8989 }
8990 
8991 void kvm_arch_commit_memory_region(struct kvm *kvm,
8992 				const struct kvm_userspace_memory_region *mem,
8993 				const struct kvm_memory_slot *old,
8994 				const struct kvm_memory_slot *new,
8995 				enum kvm_mr_change change)
8996 {
8997 	int nr_mmu_pages = 0;
8998 
8999 	if (!kvm->arch.n_requested_mmu_pages)
9000 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9001 
9002 	if (nr_mmu_pages)
9003 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9004 
9005 	/*
9006 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9007 	 * sptes have to be split.  If live migration is successful, the guest
9008 	 * in the source machine will be destroyed and large sptes will be
9009 	 * created in the destination. However, if the guest continues to run
9010 	 * in the source machine (for example if live migration fails), small
9011 	 * sptes will remain around and cause bad performance.
9012 	 *
9013 	 * Scan sptes if dirty logging has been stopped, dropping those
9014 	 * which can be collapsed into a single large-page spte.  Later
9015 	 * page faults will create the large-page sptes.
9016 	 */
9017 	if ((change != KVM_MR_DELETE) &&
9018 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9019 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9020 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9021 
9022 	/*
9023 	 * Set up write protection and/or dirty logging for the new slot.
9024 	 *
9025 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9026 	 * been zapped so no dirty logging staff is needed for old slot. For
9027 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9028 	 * new and it's also covered when dealing with the new slot.
9029 	 *
9030 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9031 	 */
9032 	if (change != KVM_MR_DELETE)
9033 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9034 }
9035 
9036 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9037 {
9038 	kvm_mmu_invalidate_zap_all_pages(kvm);
9039 }
9040 
9041 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9042 				   struct kvm_memory_slot *slot)
9043 {
9044 	kvm_page_track_flush_slot(kvm, slot);
9045 }
9046 
9047 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9048 {
9049 	if (!list_empty_careful(&vcpu->async_pf.done))
9050 		return true;
9051 
9052 	if (kvm_apic_has_events(vcpu))
9053 		return true;
9054 
9055 	if (vcpu->arch.pv.pv_unhalted)
9056 		return true;
9057 
9058 	if (vcpu->arch.exception.pending)
9059 		return true;
9060 
9061 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9062 	    (vcpu->arch.nmi_pending &&
9063 	     kvm_x86_ops->nmi_allowed(vcpu)))
9064 		return true;
9065 
9066 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9067 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9068 		return true;
9069 
9070 	if (kvm_arch_interrupt_allowed(vcpu) &&
9071 	    kvm_cpu_has_interrupt(vcpu))
9072 		return true;
9073 
9074 	if (kvm_hv_has_stimer_pending(vcpu))
9075 		return true;
9076 
9077 	return false;
9078 }
9079 
9080 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9081 {
9082 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9083 }
9084 
9085 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9086 {
9087 	return vcpu->arch.preempted_in_kernel;
9088 }
9089 
9090 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9091 {
9092 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9093 }
9094 
9095 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9096 {
9097 	return kvm_x86_ops->interrupt_allowed(vcpu);
9098 }
9099 
9100 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9101 {
9102 	if (is_64_bit_mode(vcpu))
9103 		return kvm_rip_read(vcpu);
9104 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9105 		     kvm_rip_read(vcpu));
9106 }
9107 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9108 
9109 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9110 {
9111 	return kvm_get_linear_rip(vcpu) == linear_rip;
9112 }
9113 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9114 
9115 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9116 {
9117 	unsigned long rflags;
9118 
9119 	rflags = kvm_x86_ops->get_rflags(vcpu);
9120 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9121 		rflags &= ~X86_EFLAGS_TF;
9122 	return rflags;
9123 }
9124 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9125 
9126 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9127 {
9128 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9129 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9130 		rflags |= X86_EFLAGS_TF;
9131 	kvm_x86_ops->set_rflags(vcpu, rflags);
9132 }
9133 
9134 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9135 {
9136 	__kvm_set_rflags(vcpu, rflags);
9137 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9138 }
9139 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9140 
9141 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9142 {
9143 	int r;
9144 
9145 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
9146 	      work->wakeup_all)
9147 		return;
9148 
9149 	r = kvm_mmu_reload(vcpu);
9150 	if (unlikely(r))
9151 		return;
9152 
9153 	if (!vcpu->arch.mmu.direct_map &&
9154 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
9155 		return;
9156 
9157 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
9158 }
9159 
9160 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9161 {
9162 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9163 }
9164 
9165 static inline u32 kvm_async_pf_next_probe(u32 key)
9166 {
9167 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9168 }
9169 
9170 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9171 {
9172 	u32 key = kvm_async_pf_hash_fn(gfn);
9173 
9174 	while (vcpu->arch.apf.gfns[key] != ~0)
9175 		key = kvm_async_pf_next_probe(key);
9176 
9177 	vcpu->arch.apf.gfns[key] = gfn;
9178 }
9179 
9180 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9181 {
9182 	int i;
9183 	u32 key = kvm_async_pf_hash_fn(gfn);
9184 
9185 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9186 		     (vcpu->arch.apf.gfns[key] != gfn &&
9187 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9188 		key = kvm_async_pf_next_probe(key);
9189 
9190 	return key;
9191 }
9192 
9193 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9194 {
9195 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9196 }
9197 
9198 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9199 {
9200 	u32 i, j, k;
9201 
9202 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9203 	while (true) {
9204 		vcpu->arch.apf.gfns[i] = ~0;
9205 		do {
9206 			j = kvm_async_pf_next_probe(j);
9207 			if (vcpu->arch.apf.gfns[j] == ~0)
9208 				return;
9209 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9210 			/*
9211 			 * k lies cyclically in ]i,j]
9212 			 * |    i.k.j |
9213 			 * |....j i.k.| or  |.k..j i...|
9214 			 */
9215 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9216 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9217 		i = j;
9218 	}
9219 }
9220 
9221 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9222 {
9223 
9224 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9225 				      sizeof(val));
9226 }
9227 
9228 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9229 {
9230 
9231 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9232 				      sizeof(u32));
9233 }
9234 
9235 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9236 				     struct kvm_async_pf *work)
9237 {
9238 	struct x86_exception fault;
9239 
9240 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9241 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9242 
9243 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9244 	    (vcpu->arch.apf.send_user_only &&
9245 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9246 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9247 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9248 		fault.vector = PF_VECTOR;
9249 		fault.error_code_valid = true;
9250 		fault.error_code = 0;
9251 		fault.nested_page_fault = false;
9252 		fault.address = work->arch.token;
9253 		fault.async_page_fault = true;
9254 		kvm_inject_page_fault(vcpu, &fault);
9255 	}
9256 }
9257 
9258 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9259 				 struct kvm_async_pf *work)
9260 {
9261 	struct x86_exception fault;
9262 	u32 val;
9263 
9264 	if (work->wakeup_all)
9265 		work->arch.token = ~0; /* broadcast wakeup */
9266 	else
9267 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9268 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9269 
9270 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9271 	    !apf_get_user(vcpu, &val)) {
9272 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9273 		    vcpu->arch.exception.pending &&
9274 		    vcpu->arch.exception.nr == PF_VECTOR &&
9275 		    !apf_put_user(vcpu, 0)) {
9276 			vcpu->arch.exception.injected = false;
9277 			vcpu->arch.exception.pending = false;
9278 			vcpu->arch.exception.nr = 0;
9279 			vcpu->arch.exception.has_error_code = false;
9280 			vcpu->arch.exception.error_code = 0;
9281 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9282 			fault.vector = PF_VECTOR;
9283 			fault.error_code_valid = true;
9284 			fault.error_code = 0;
9285 			fault.nested_page_fault = false;
9286 			fault.address = work->arch.token;
9287 			fault.async_page_fault = true;
9288 			kvm_inject_page_fault(vcpu, &fault);
9289 		}
9290 	}
9291 	vcpu->arch.apf.halted = false;
9292 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9293 }
9294 
9295 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9296 {
9297 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9298 		return true;
9299 	else
9300 		return kvm_can_do_async_pf(vcpu);
9301 }
9302 
9303 void kvm_arch_start_assignment(struct kvm *kvm)
9304 {
9305 	atomic_inc(&kvm->arch.assigned_device_count);
9306 }
9307 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9308 
9309 void kvm_arch_end_assignment(struct kvm *kvm)
9310 {
9311 	atomic_dec(&kvm->arch.assigned_device_count);
9312 }
9313 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9314 
9315 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9316 {
9317 	return atomic_read(&kvm->arch.assigned_device_count);
9318 }
9319 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9320 
9321 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9322 {
9323 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9324 }
9325 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9326 
9327 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9328 {
9329 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9330 }
9331 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9332 
9333 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9334 {
9335 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9336 }
9337 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9338 
9339 bool kvm_arch_has_irq_bypass(void)
9340 {
9341 	return kvm_x86_ops->update_pi_irte != NULL;
9342 }
9343 
9344 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9345 				      struct irq_bypass_producer *prod)
9346 {
9347 	struct kvm_kernel_irqfd *irqfd =
9348 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9349 
9350 	irqfd->producer = prod;
9351 
9352 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9353 					   prod->irq, irqfd->gsi, 1);
9354 }
9355 
9356 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9357 				      struct irq_bypass_producer *prod)
9358 {
9359 	int ret;
9360 	struct kvm_kernel_irqfd *irqfd =
9361 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9362 
9363 	WARN_ON(irqfd->producer != prod);
9364 	irqfd->producer = NULL;
9365 
9366 	/*
9367 	 * When producer of consumer is unregistered, we change back to
9368 	 * remapped mode, so we can re-use the current implementation
9369 	 * when the irq is masked/disabled or the consumer side (KVM
9370 	 * int this case doesn't want to receive the interrupts.
9371 	*/
9372 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9373 	if (ret)
9374 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9375 		       " fails: %d\n", irqfd->consumer.token, ret);
9376 }
9377 
9378 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9379 				   uint32_t guest_irq, bool set)
9380 {
9381 	if (!kvm_x86_ops->update_pi_irte)
9382 		return -EINVAL;
9383 
9384 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9385 }
9386 
9387 bool kvm_vector_hashing_enabled(void)
9388 {
9389 	return vector_hashing;
9390 }
9391 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9392 
9393 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9394 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9395 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9396 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9397 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9399 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9400 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9401 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9402 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9403 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9404 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9405 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9406 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9407 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9408 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9409 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9412