xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 4cff79e9)
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 = 500;
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 	cr3 &= ~CR3_PCID_INVD;
847 #endif
848 
849 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
850 		kvm_mmu_sync_roots(vcpu);
851 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
852 		return 0;
853 	}
854 
855 	if (is_long_mode(vcpu) &&
856 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
857 		return 1;
858 	else if (is_pae(vcpu) && is_paging(vcpu) &&
859 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
860 		return 1;
861 
862 	vcpu->arch.cr3 = cr3;
863 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
864 	kvm_mmu_new_cr3(vcpu);
865 	return 0;
866 }
867 EXPORT_SYMBOL_GPL(kvm_set_cr3);
868 
869 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
870 {
871 	if (cr8 & CR8_RESERVED_BITS)
872 		return 1;
873 	if (lapic_in_kernel(vcpu))
874 		kvm_lapic_set_tpr(vcpu, cr8);
875 	else
876 		vcpu->arch.cr8 = cr8;
877 	return 0;
878 }
879 EXPORT_SYMBOL_GPL(kvm_set_cr8);
880 
881 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
882 {
883 	if (lapic_in_kernel(vcpu))
884 		return kvm_lapic_get_cr8(vcpu);
885 	else
886 		return vcpu->arch.cr8;
887 }
888 EXPORT_SYMBOL_GPL(kvm_get_cr8);
889 
890 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
891 {
892 	int i;
893 
894 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
895 		for (i = 0; i < KVM_NR_DB_REGS; i++)
896 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
897 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
898 	}
899 }
900 
901 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
902 {
903 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
904 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
905 }
906 
907 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
908 {
909 	unsigned long dr7;
910 
911 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
912 		dr7 = vcpu->arch.guest_debug_dr7;
913 	else
914 		dr7 = vcpu->arch.dr7;
915 	kvm_x86_ops->set_dr7(vcpu, dr7);
916 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
917 	if (dr7 & DR7_BP_EN_MASK)
918 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
919 }
920 
921 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
922 {
923 	u64 fixed = DR6_FIXED_1;
924 
925 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
926 		fixed |= DR6_RTM;
927 	return fixed;
928 }
929 
930 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
931 {
932 	switch (dr) {
933 	case 0 ... 3:
934 		vcpu->arch.db[dr] = val;
935 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
936 			vcpu->arch.eff_db[dr] = val;
937 		break;
938 	case 4:
939 		/* fall through */
940 	case 6:
941 		if (val & 0xffffffff00000000ULL)
942 			return -1; /* #GP */
943 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
944 		kvm_update_dr6(vcpu);
945 		break;
946 	case 5:
947 		/* fall through */
948 	default: /* 7 */
949 		if (val & 0xffffffff00000000ULL)
950 			return -1; /* #GP */
951 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
952 		kvm_update_dr7(vcpu);
953 		break;
954 	}
955 
956 	return 0;
957 }
958 
959 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
960 {
961 	if (__kvm_set_dr(vcpu, dr, val)) {
962 		kvm_inject_gp(vcpu, 0);
963 		return 1;
964 	}
965 	return 0;
966 }
967 EXPORT_SYMBOL_GPL(kvm_set_dr);
968 
969 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
970 {
971 	switch (dr) {
972 	case 0 ... 3:
973 		*val = vcpu->arch.db[dr];
974 		break;
975 	case 4:
976 		/* fall through */
977 	case 6:
978 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
979 			*val = vcpu->arch.dr6;
980 		else
981 			*val = kvm_x86_ops->get_dr6(vcpu);
982 		break;
983 	case 5:
984 		/* fall through */
985 	default: /* 7 */
986 		*val = vcpu->arch.dr7;
987 		break;
988 	}
989 	return 0;
990 }
991 EXPORT_SYMBOL_GPL(kvm_get_dr);
992 
993 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
994 {
995 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
996 	u64 data;
997 	int err;
998 
999 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1000 	if (err)
1001 		return err;
1002 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1003 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1004 	return err;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1007 
1008 /*
1009  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1010  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1011  *
1012  * This list is modified at module load time to reflect the
1013  * capabilities of the host cpu. This capabilities test skips MSRs that are
1014  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1015  * may depend on host virtualization features rather than host cpu features.
1016  */
1017 
1018 static u32 msrs_to_save[] = {
1019 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1020 	MSR_STAR,
1021 #ifdef CONFIG_X86_64
1022 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1023 #endif
1024 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1025 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1026 	MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1027 };
1028 
1029 static unsigned num_msrs_to_save;
1030 
1031 static u32 emulated_msrs[] = {
1032 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1033 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1034 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1035 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1036 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1037 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1038 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1039 	HV_X64_MSR_RESET,
1040 	HV_X64_MSR_VP_INDEX,
1041 	HV_X64_MSR_VP_RUNTIME,
1042 	HV_X64_MSR_SCONTROL,
1043 	HV_X64_MSR_STIMER0_CONFIG,
1044 	HV_X64_MSR_VP_ASSIST_PAGE,
1045 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1046 	HV_X64_MSR_TSC_EMULATION_STATUS,
1047 
1048 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1049 	MSR_KVM_PV_EOI_EN,
1050 
1051 	MSR_IA32_TSC_ADJUST,
1052 	MSR_IA32_TSCDEADLINE,
1053 	MSR_IA32_MISC_ENABLE,
1054 	MSR_IA32_MCG_STATUS,
1055 	MSR_IA32_MCG_CTL,
1056 	MSR_IA32_MCG_EXT_CTL,
1057 	MSR_IA32_SMBASE,
1058 	MSR_SMI_COUNT,
1059 	MSR_PLATFORM_INFO,
1060 	MSR_MISC_FEATURES_ENABLES,
1061 };
1062 
1063 static unsigned num_emulated_msrs;
1064 
1065 /*
1066  * List of msr numbers which are used to expose MSR-based features that
1067  * can be used by a hypervisor to validate requested CPU features.
1068  */
1069 static u32 msr_based_features[] = {
1070 	MSR_IA32_VMX_BASIC,
1071 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1072 	MSR_IA32_VMX_PINBASED_CTLS,
1073 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1074 	MSR_IA32_VMX_PROCBASED_CTLS,
1075 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1076 	MSR_IA32_VMX_EXIT_CTLS,
1077 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1078 	MSR_IA32_VMX_ENTRY_CTLS,
1079 	MSR_IA32_VMX_MISC,
1080 	MSR_IA32_VMX_CR0_FIXED0,
1081 	MSR_IA32_VMX_CR0_FIXED1,
1082 	MSR_IA32_VMX_CR4_FIXED0,
1083 	MSR_IA32_VMX_CR4_FIXED1,
1084 	MSR_IA32_VMX_VMCS_ENUM,
1085 	MSR_IA32_VMX_PROCBASED_CTLS2,
1086 	MSR_IA32_VMX_EPT_VPID_CAP,
1087 	MSR_IA32_VMX_VMFUNC,
1088 
1089 	MSR_F10H_DECFG,
1090 	MSR_IA32_UCODE_REV,
1091 };
1092 
1093 static unsigned int num_msr_based_features;
1094 
1095 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1096 {
1097 	switch (msr->index) {
1098 	case MSR_IA32_UCODE_REV:
1099 		rdmsrl(msr->index, msr->data);
1100 		break;
1101 	default:
1102 		if (kvm_x86_ops->get_msr_feature(msr))
1103 			return 1;
1104 	}
1105 	return 0;
1106 }
1107 
1108 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1109 {
1110 	struct kvm_msr_entry msr;
1111 	int r;
1112 
1113 	msr.index = index;
1114 	r = kvm_get_msr_feature(&msr);
1115 	if (r)
1116 		return r;
1117 
1118 	*data = msr.data;
1119 
1120 	return 0;
1121 }
1122 
1123 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1124 {
1125 	if (efer & efer_reserved_bits)
1126 		return false;
1127 
1128 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1129 			return false;
1130 
1131 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1132 			return false;
1133 
1134 	return true;
1135 }
1136 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1137 
1138 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1139 {
1140 	u64 old_efer = vcpu->arch.efer;
1141 
1142 	if (!kvm_valid_efer(vcpu, efer))
1143 		return 1;
1144 
1145 	if (is_paging(vcpu)
1146 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1147 		return 1;
1148 
1149 	efer &= ~EFER_LMA;
1150 	efer |= vcpu->arch.efer & EFER_LMA;
1151 
1152 	kvm_x86_ops->set_efer(vcpu, efer);
1153 
1154 	/* Update reserved bits */
1155 	if ((efer ^ old_efer) & EFER_NX)
1156 		kvm_mmu_reset_context(vcpu);
1157 
1158 	return 0;
1159 }
1160 
1161 void kvm_enable_efer_bits(u64 mask)
1162 {
1163        efer_reserved_bits &= ~mask;
1164 }
1165 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1166 
1167 /*
1168  * Writes msr value into into the appropriate "register".
1169  * Returns 0 on success, non-0 otherwise.
1170  * Assumes vcpu_load() was already called.
1171  */
1172 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1173 {
1174 	switch (msr->index) {
1175 	case MSR_FS_BASE:
1176 	case MSR_GS_BASE:
1177 	case MSR_KERNEL_GS_BASE:
1178 	case MSR_CSTAR:
1179 	case MSR_LSTAR:
1180 		if (is_noncanonical_address(msr->data, vcpu))
1181 			return 1;
1182 		break;
1183 	case MSR_IA32_SYSENTER_EIP:
1184 	case MSR_IA32_SYSENTER_ESP:
1185 		/*
1186 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1187 		 * non-canonical address is written on Intel but not on
1188 		 * AMD (which ignores the top 32-bits, because it does
1189 		 * not implement 64-bit SYSENTER).
1190 		 *
1191 		 * 64-bit code should hence be able to write a non-canonical
1192 		 * value on AMD.  Making the address canonical ensures that
1193 		 * vmentry does not fail on Intel after writing a non-canonical
1194 		 * value, and that something deterministic happens if the guest
1195 		 * invokes 64-bit SYSENTER.
1196 		 */
1197 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1198 	}
1199 	return kvm_x86_ops->set_msr(vcpu, msr);
1200 }
1201 EXPORT_SYMBOL_GPL(kvm_set_msr);
1202 
1203 /*
1204  * Adapt set_msr() to msr_io()'s calling convention
1205  */
1206 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1207 {
1208 	struct msr_data msr;
1209 	int r;
1210 
1211 	msr.index = index;
1212 	msr.host_initiated = true;
1213 	r = kvm_get_msr(vcpu, &msr);
1214 	if (r)
1215 		return r;
1216 
1217 	*data = msr.data;
1218 	return 0;
1219 }
1220 
1221 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1222 {
1223 	struct msr_data msr;
1224 
1225 	msr.data = *data;
1226 	msr.index = index;
1227 	msr.host_initiated = true;
1228 	return kvm_set_msr(vcpu, &msr);
1229 }
1230 
1231 #ifdef CONFIG_X86_64
1232 struct pvclock_gtod_data {
1233 	seqcount_t	seq;
1234 
1235 	struct { /* extract of a clocksource struct */
1236 		int vclock_mode;
1237 		u64	cycle_last;
1238 		u64	mask;
1239 		u32	mult;
1240 		u32	shift;
1241 	} clock;
1242 
1243 	u64		boot_ns;
1244 	u64		nsec_base;
1245 	u64		wall_time_sec;
1246 };
1247 
1248 static struct pvclock_gtod_data pvclock_gtod_data;
1249 
1250 static void update_pvclock_gtod(struct timekeeper *tk)
1251 {
1252 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1253 	u64 boot_ns;
1254 
1255 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1256 
1257 	write_seqcount_begin(&vdata->seq);
1258 
1259 	/* copy pvclock gtod data */
1260 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1261 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1262 	vdata->clock.mask		= tk->tkr_mono.mask;
1263 	vdata->clock.mult		= tk->tkr_mono.mult;
1264 	vdata->clock.shift		= tk->tkr_mono.shift;
1265 
1266 	vdata->boot_ns			= boot_ns;
1267 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1268 
1269 	vdata->wall_time_sec            = tk->xtime_sec;
1270 
1271 	write_seqcount_end(&vdata->seq);
1272 }
1273 #endif
1274 
1275 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1276 {
1277 	/*
1278 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1279 	 * vcpu_enter_guest.  This function is only called from
1280 	 * the physical CPU that is running vcpu.
1281 	 */
1282 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1283 }
1284 
1285 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1286 {
1287 	int version;
1288 	int r;
1289 	struct pvclock_wall_clock wc;
1290 	struct timespec64 boot;
1291 
1292 	if (!wall_clock)
1293 		return;
1294 
1295 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1296 	if (r)
1297 		return;
1298 
1299 	if (version & 1)
1300 		++version;  /* first time write, random junk */
1301 
1302 	++version;
1303 
1304 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1305 		return;
1306 
1307 	/*
1308 	 * The guest calculates current wall clock time by adding
1309 	 * system time (updated by kvm_guest_time_update below) to the
1310 	 * wall clock specified here.  guest system time equals host
1311 	 * system time for us, thus we must fill in host boot time here.
1312 	 */
1313 	getboottime64(&boot);
1314 
1315 	if (kvm->arch.kvmclock_offset) {
1316 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1317 		boot = timespec64_sub(boot, ts);
1318 	}
1319 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1320 	wc.nsec = boot.tv_nsec;
1321 	wc.version = version;
1322 
1323 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1324 
1325 	version++;
1326 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1327 }
1328 
1329 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1330 {
1331 	do_shl32_div32(dividend, divisor);
1332 	return dividend;
1333 }
1334 
1335 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1336 			       s8 *pshift, u32 *pmultiplier)
1337 {
1338 	uint64_t scaled64;
1339 	int32_t  shift = 0;
1340 	uint64_t tps64;
1341 	uint32_t tps32;
1342 
1343 	tps64 = base_hz;
1344 	scaled64 = scaled_hz;
1345 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1346 		tps64 >>= 1;
1347 		shift--;
1348 	}
1349 
1350 	tps32 = (uint32_t)tps64;
1351 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1352 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1353 			scaled64 >>= 1;
1354 		else
1355 			tps32 <<= 1;
1356 		shift++;
1357 	}
1358 
1359 	*pshift = shift;
1360 	*pmultiplier = div_frac(scaled64, tps32);
1361 
1362 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1363 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1364 }
1365 
1366 #ifdef CONFIG_X86_64
1367 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1368 #endif
1369 
1370 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1371 static unsigned long max_tsc_khz;
1372 
1373 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1374 {
1375 	u64 v = (u64)khz * (1000000 + ppm);
1376 	do_div(v, 1000000);
1377 	return v;
1378 }
1379 
1380 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1381 {
1382 	u64 ratio;
1383 
1384 	/* Guest TSC same frequency as host TSC? */
1385 	if (!scale) {
1386 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1387 		return 0;
1388 	}
1389 
1390 	/* TSC scaling supported? */
1391 	if (!kvm_has_tsc_control) {
1392 		if (user_tsc_khz > tsc_khz) {
1393 			vcpu->arch.tsc_catchup = 1;
1394 			vcpu->arch.tsc_always_catchup = 1;
1395 			return 0;
1396 		} else {
1397 			WARN(1, "user requested TSC rate below hardware speed\n");
1398 			return -1;
1399 		}
1400 	}
1401 
1402 	/* TSC scaling required  - calculate ratio */
1403 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1404 				user_tsc_khz, tsc_khz);
1405 
1406 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1407 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1408 			  user_tsc_khz);
1409 		return -1;
1410 	}
1411 
1412 	vcpu->arch.tsc_scaling_ratio = ratio;
1413 	return 0;
1414 }
1415 
1416 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1417 {
1418 	u32 thresh_lo, thresh_hi;
1419 	int use_scaling = 0;
1420 
1421 	/* tsc_khz can be zero if TSC calibration fails */
1422 	if (user_tsc_khz == 0) {
1423 		/* set tsc_scaling_ratio to a safe value */
1424 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1425 		return -1;
1426 	}
1427 
1428 	/* Compute a scale to convert nanoseconds in TSC cycles */
1429 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1430 			   &vcpu->arch.virtual_tsc_shift,
1431 			   &vcpu->arch.virtual_tsc_mult);
1432 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1433 
1434 	/*
1435 	 * Compute the variation in TSC rate which is acceptable
1436 	 * within the range of tolerance and decide if the
1437 	 * rate being applied is within that bounds of the hardware
1438 	 * rate.  If so, no scaling or compensation need be done.
1439 	 */
1440 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1441 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1442 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1443 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1444 		use_scaling = 1;
1445 	}
1446 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1447 }
1448 
1449 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1450 {
1451 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1452 				      vcpu->arch.virtual_tsc_mult,
1453 				      vcpu->arch.virtual_tsc_shift);
1454 	tsc += vcpu->arch.this_tsc_write;
1455 	return tsc;
1456 }
1457 
1458 static inline int gtod_is_based_on_tsc(int mode)
1459 {
1460 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1461 }
1462 
1463 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1464 {
1465 #ifdef CONFIG_X86_64
1466 	bool vcpus_matched;
1467 	struct kvm_arch *ka = &vcpu->kvm->arch;
1468 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1469 
1470 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1471 			 atomic_read(&vcpu->kvm->online_vcpus));
1472 
1473 	/*
1474 	 * Once the masterclock is enabled, always perform request in
1475 	 * order to update it.
1476 	 *
1477 	 * In order to enable masterclock, the host clocksource must be TSC
1478 	 * and the vcpus need to have matched TSCs.  When that happens,
1479 	 * perform request to enable masterclock.
1480 	 */
1481 	if (ka->use_master_clock ||
1482 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1483 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1484 
1485 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1486 			    atomic_read(&vcpu->kvm->online_vcpus),
1487 		            ka->use_master_clock, gtod->clock.vclock_mode);
1488 #endif
1489 }
1490 
1491 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1492 {
1493 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1494 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1495 }
1496 
1497 /*
1498  * Multiply tsc by a fixed point number represented by ratio.
1499  *
1500  * The most significant 64-N bits (mult) of ratio represent the
1501  * integral part of the fixed point number; the remaining N bits
1502  * (frac) represent the fractional part, ie. ratio represents a fixed
1503  * point number (mult + frac * 2^(-N)).
1504  *
1505  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1506  */
1507 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1508 {
1509 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1510 }
1511 
1512 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1513 {
1514 	u64 _tsc = tsc;
1515 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1516 
1517 	if (ratio != kvm_default_tsc_scaling_ratio)
1518 		_tsc = __scale_tsc(ratio, tsc);
1519 
1520 	return _tsc;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1523 
1524 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1525 {
1526 	u64 tsc;
1527 
1528 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1529 
1530 	return target_tsc - tsc;
1531 }
1532 
1533 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1534 {
1535 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1536 
1537 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1538 }
1539 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1540 
1541 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1542 {
1543 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1544 	vcpu->arch.tsc_offset = offset;
1545 }
1546 
1547 static inline bool kvm_check_tsc_unstable(void)
1548 {
1549 #ifdef CONFIG_X86_64
1550 	/*
1551 	 * TSC is marked unstable when we're running on Hyper-V,
1552 	 * 'TSC page' clocksource is good.
1553 	 */
1554 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1555 		return false;
1556 #endif
1557 	return check_tsc_unstable();
1558 }
1559 
1560 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1561 {
1562 	struct kvm *kvm = vcpu->kvm;
1563 	u64 offset, ns, elapsed;
1564 	unsigned long flags;
1565 	bool matched;
1566 	bool already_matched;
1567 	u64 data = msr->data;
1568 	bool synchronizing = false;
1569 
1570 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1571 	offset = kvm_compute_tsc_offset(vcpu, data);
1572 	ns = ktime_get_boot_ns();
1573 	elapsed = ns - kvm->arch.last_tsc_nsec;
1574 
1575 	if (vcpu->arch.virtual_tsc_khz) {
1576 		if (data == 0 && msr->host_initiated) {
1577 			/*
1578 			 * detection of vcpu initialization -- need to sync
1579 			 * with other vCPUs. This particularly helps to keep
1580 			 * kvm_clock stable after CPU hotplug
1581 			 */
1582 			synchronizing = true;
1583 		} else {
1584 			u64 tsc_exp = kvm->arch.last_tsc_write +
1585 						nsec_to_cycles(vcpu, elapsed);
1586 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1587 			/*
1588 			 * Special case: TSC write with a small delta (1 second)
1589 			 * of virtual cycle time against real time is
1590 			 * interpreted as an attempt to synchronize the CPU.
1591 			 */
1592 			synchronizing = data < tsc_exp + tsc_hz &&
1593 					data + tsc_hz > tsc_exp;
1594 		}
1595 	}
1596 
1597 	/*
1598 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1599 	 * TSC, we add elapsed time in this computation.  We could let the
1600 	 * compensation code attempt to catch up if we fall behind, but
1601 	 * it's better to try to match offsets from the beginning.
1602          */
1603 	if (synchronizing &&
1604 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1605 		if (!kvm_check_tsc_unstable()) {
1606 			offset = kvm->arch.cur_tsc_offset;
1607 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1608 		} else {
1609 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1610 			data += delta;
1611 			offset = kvm_compute_tsc_offset(vcpu, data);
1612 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1613 		}
1614 		matched = true;
1615 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1616 	} else {
1617 		/*
1618 		 * We split periods of matched TSC writes into generations.
1619 		 * For each generation, we track the original measured
1620 		 * nanosecond time, offset, and write, so if TSCs are in
1621 		 * sync, we can match exact offset, and if not, we can match
1622 		 * exact software computation in compute_guest_tsc()
1623 		 *
1624 		 * These values are tracked in kvm->arch.cur_xxx variables.
1625 		 */
1626 		kvm->arch.cur_tsc_generation++;
1627 		kvm->arch.cur_tsc_nsec = ns;
1628 		kvm->arch.cur_tsc_write = data;
1629 		kvm->arch.cur_tsc_offset = offset;
1630 		matched = false;
1631 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1632 			 kvm->arch.cur_tsc_generation, data);
1633 	}
1634 
1635 	/*
1636 	 * We also track th most recent recorded KHZ, write and time to
1637 	 * allow the matching interval to be extended at each write.
1638 	 */
1639 	kvm->arch.last_tsc_nsec = ns;
1640 	kvm->arch.last_tsc_write = data;
1641 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1642 
1643 	vcpu->arch.last_guest_tsc = data;
1644 
1645 	/* Keep track of which generation this VCPU has synchronized to */
1646 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1647 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1648 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1649 
1650 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1651 		update_ia32_tsc_adjust_msr(vcpu, offset);
1652 
1653 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1654 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1655 
1656 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1657 	if (!matched) {
1658 		kvm->arch.nr_vcpus_matched_tsc = 0;
1659 	} else if (!already_matched) {
1660 		kvm->arch.nr_vcpus_matched_tsc++;
1661 	}
1662 
1663 	kvm_track_tsc_matching(vcpu);
1664 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1665 }
1666 
1667 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1668 
1669 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1670 					   s64 adjustment)
1671 {
1672 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1673 }
1674 
1675 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1676 {
1677 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1678 		WARN_ON(adjustment < 0);
1679 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1680 	adjust_tsc_offset_guest(vcpu, adjustment);
1681 }
1682 
1683 #ifdef CONFIG_X86_64
1684 
1685 static u64 read_tsc(void)
1686 {
1687 	u64 ret = (u64)rdtsc_ordered();
1688 	u64 last = pvclock_gtod_data.clock.cycle_last;
1689 
1690 	if (likely(ret >= last))
1691 		return ret;
1692 
1693 	/*
1694 	 * GCC likes to generate cmov here, but this branch is extremely
1695 	 * predictable (it's just a function of time and the likely is
1696 	 * very likely) and there's a data dependence, so force GCC
1697 	 * to generate a branch instead.  I don't barrier() because
1698 	 * we don't actually need a barrier, and if this function
1699 	 * ever gets inlined it will generate worse code.
1700 	 */
1701 	asm volatile ("");
1702 	return last;
1703 }
1704 
1705 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1706 {
1707 	long v;
1708 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1709 	u64 tsc_pg_val;
1710 
1711 	switch (gtod->clock.vclock_mode) {
1712 	case VCLOCK_HVCLOCK:
1713 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1714 						  tsc_timestamp);
1715 		if (tsc_pg_val != U64_MAX) {
1716 			/* TSC page valid */
1717 			*mode = VCLOCK_HVCLOCK;
1718 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1719 				gtod->clock.mask;
1720 		} else {
1721 			/* TSC page invalid */
1722 			*mode = VCLOCK_NONE;
1723 		}
1724 		break;
1725 	case VCLOCK_TSC:
1726 		*mode = VCLOCK_TSC;
1727 		*tsc_timestamp = read_tsc();
1728 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1729 			gtod->clock.mask;
1730 		break;
1731 	default:
1732 		*mode = VCLOCK_NONE;
1733 	}
1734 
1735 	if (*mode == VCLOCK_NONE)
1736 		*tsc_timestamp = v = 0;
1737 
1738 	return v * gtod->clock.mult;
1739 }
1740 
1741 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1742 {
1743 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1744 	unsigned long seq;
1745 	int mode;
1746 	u64 ns;
1747 
1748 	do {
1749 		seq = read_seqcount_begin(&gtod->seq);
1750 		ns = gtod->nsec_base;
1751 		ns += vgettsc(tsc_timestamp, &mode);
1752 		ns >>= gtod->clock.shift;
1753 		ns += gtod->boot_ns;
1754 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1755 	*t = ns;
1756 
1757 	return mode;
1758 }
1759 
1760 static int do_realtime(struct timespec *ts, u64 *tsc_timestamp)
1761 {
1762 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1763 	unsigned long seq;
1764 	int mode;
1765 	u64 ns;
1766 
1767 	do {
1768 		seq = read_seqcount_begin(&gtod->seq);
1769 		ts->tv_sec = gtod->wall_time_sec;
1770 		ns = gtod->nsec_base;
1771 		ns += vgettsc(tsc_timestamp, &mode);
1772 		ns >>= gtod->clock.shift;
1773 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1774 
1775 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1776 	ts->tv_nsec = ns;
1777 
1778 	return mode;
1779 }
1780 
1781 /* returns true if host is using TSC based clocksource */
1782 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1783 {
1784 	/* checked again under seqlock below */
1785 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1786 		return false;
1787 
1788 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1789 						      tsc_timestamp));
1790 }
1791 
1792 /* returns true if host is using TSC based clocksource */
1793 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1794 					   u64 *tsc_timestamp)
1795 {
1796 	/* checked again under seqlock below */
1797 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1798 		return false;
1799 
1800 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1801 }
1802 #endif
1803 
1804 /*
1805  *
1806  * Assuming a stable TSC across physical CPUS, and a stable TSC
1807  * across virtual CPUs, the following condition is possible.
1808  * Each numbered line represents an event visible to both
1809  * CPUs at the next numbered event.
1810  *
1811  * "timespecX" represents host monotonic time. "tscX" represents
1812  * RDTSC value.
1813  *
1814  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1815  *
1816  * 1.  read timespec0,tsc0
1817  * 2.					| timespec1 = timespec0 + N
1818  * 					| tsc1 = tsc0 + M
1819  * 3. transition to guest		| transition to guest
1820  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1821  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1822  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1823  *
1824  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1825  *
1826  * 	- ret0 < ret1
1827  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1828  *		...
1829  *	- 0 < N - M => M < N
1830  *
1831  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1832  * always the case (the difference between two distinct xtime instances
1833  * might be smaller then the difference between corresponding TSC reads,
1834  * when updating guest vcpus pvclock areas).
1835  *
1836  * To avoid that problem, do not allow visibility of distinct
1837  * system_timestamp/tsc_timestamp values simultaneously: use a master
1838  * copy of host monotonic time values. Update that master copy
1839  * in lockstep.
1840  *
1841  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1842  *
1843  */
1844 
1845 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1846 {
1847 #ifdef CONFIG_X86_64
1848 	struct kvm_arch *ka = &kvm->arch;
1849 	int vclock_mode;
1850 	bool host_tsc_clocksource, vcpus_matched;
1851 
1852 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1853 			atomic_read(&kvm->online_vcpus));
1854 
1855 	/*
1856 	 * If the host uses TSC clock, then passthrough TSC as stable
1857 	 * to the guest.
1858 	 */
1859 	host_tsc_clocksource = kvm_get_time_and_clockread(
1860 					&ka->master_kernel_ns,
1861 					&ka->master_cycle_now);
1862 
1863 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1864 				&& !ka->backwards_tsc_observed
1865 				&& !ka->boot_vcpu_runs_old_kvmclock;
1866 
1867 	if (ka->use_master_clock)
1868 		atomic_set(&kvm_guest_has_master_clock, 1);
1869 
1870 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1871 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1872 					vcpus_matched);
1873 #endif
1874 }
1875 
1876 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1877 {
1878 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1879 }
1880 
1881 static void kvm_gen_update_masterclock(struct kvm *kvm)
1882 {
1883 #ifdef CONFIG_X86_64
1884 	int i;
1885 	struct kvm_vcpu *vcpu;
1886 	struct kvm_arch *ka = &kvm->arch;
1887 
1888 	spin_lock(&ka->pvclock_gtod_sync_lock);
1889 	kvm_make_mclock_inprogress_request(kvm);
1890 	/* no guest entries from this point */
1891 	pvclock_update_vm_gtod_copy(kvm);
1892 
1893 	kvm_for_each_vcpu(i, vcpu, kvm)
1894 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1895 
1896 	/* guest entries allowed */
1897 	kvm_for_each_vcpu(i, vcpu, kvm)
1898 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1899 
1900 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1901 #endif
1902 }
1903 
1904 u64 get_kvmclock_ns(struct kvm *kvm)
1905 {
1906 	struct kvm_arch *ka = &kvm->arch;
1907 	struct pvclock_vcpu_time_info hv_clock;
1908 	u64 ret;
1909 
1910 	spin_lock(&ka->pvclock_gtod_sync_lock);
1911 	if (!ka->use_master_clock) {
1912 		spin_unlock(&ka->pvclock_gtod_sync_lock);
1913 		return ktime_get_boot_ns() + ka->kvmclock_offset;
1914 	}
1915 
1916 	hv_clock.tsc_timestamp = ka->master_cycle_now;
1917 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1918 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1919 
1920 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
1921 	get_cpu();
1922 
1923 	if (__this_cpu_read(cpu_tsc_khz)) {
1924 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1925 				   &hv_clock.tsc_shift,
1926 				   &hv_clock.tsc_to_system_mul);
1927 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1928 	} else
1929 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1930 
1931 	put_cpu();
1932 
1933 	return ret;
1934 }
1935 
1936 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1937 {
1938 	struct kvm_vcpu_arch *vcpu = &v->arch;
1939 	struct pvclock_vcpu_time_info guest_hv_clock;
1940 
1941 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1942 		&guest_hv_clock, sizeof(guest_hv_clock))))
1943 		return;
1944 
1945 	/* This VCPU is paused, but it's legal for a guest to read another
1946 	 * VCPU's kvmclock, so we really have to follow the specification where
1947 	 * it says that version is odd if data is being modified, and even after
1948 	 * it is consistent.
1949 	 *
1950 	 * Version field updates must be kept separate.  This is because
1951 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1952 	 * writes within a string instruction are weakly ordered.  So there
1953 	 * are three writes overall.
1954 	 *
1955 	 * As a small optimization, only write the version field in the first
1956 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1957 	 * version field is the first in the struct.
1958 	 */
1959 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1960 
1961 	if (guest_hv_clock.version & 1)
1962 		++guest_hv_clock.version;  /* first time write, random junk */
1963 
1964 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1965 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1966 				&vcpu->hv_clock,
1967 				sizeof(vcpu->hv_clock.version));
1968 
1969 	smp_wmb();
1970 
1971 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1972 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1973 
1974 	if (vcpu->pvclock_set_guest_stopped_request) {
1975 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1976 		vcpu->pvclock_set_guest_stopped_request = false;
1977 	}
1978 
1979 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1980 
1981 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1982 				&vcpu->hv_clock,
1983 				sizeof(vcpu->hv_clock));
1984 
1985 	smp_wmb();
1986 
1987 	vcpu->hv_clock.version++;
1988 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1989 				&vcpu->hv_clock,
1990 				sizeof(vcpu->hv_clock.version));
1991 }
1992 
1993 static int kvm_guest_time_update(struct kvm_vcpu *v)
1994 {
1995 	unsigned long flags, tgt_tsc_khz;
1996 	struct kvm_vcpu_arch *vcpu = &v->arch;
1997 	struct kvm_arch *ka = &v->kvm->arch;
1998 	s64 kernel_ns;
1999 	u64 tsc_timestamp, host_tsc;
2000 	u8 pvclock_flags;
2001 	bool use_master_clock;
2002 
2003 	kernel_ns = 0;
2004 	host_tsc = 0;
2005 
2006 	/*
2007 	 * If the host uses TSC clock, then passthrough TSC as stable
2008 	 * to the guest.
2009 	 */
2010 	spin_lock(&ka->pvclock_gtod_sync_lock);
2011 	use_master_clock = ka->use_master_clock;
2012 	if (use_master_clock) {
2013 		host_tsc = ka->master_cycle_now;
2014 		kernel_ns = ka->master_kernel_ns;
2015 	}
2016 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2017 
2018 	/* Keep irq disabled to prevent changes to the clock */
2019 	local_irq_save(flags);
2020 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2021 	if (unlikely(tgt_tsc_khz == 0)) {
2022 		local_irq_restore(flags);
2023 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2024 		return 1;
2025 	}
2026 	if (!use_master_clock) {
2027 		host_tsc = rdtsc();
2028 		kernel_ns = ktime_get_boot_ns();
2029 	}
2030 
2031 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2032 
2033 	/*
2034 	 * We may have to catch up the TSC to match elapsed wall clock
2035 	 * time for two reasons, even if kvmclock is used.
2036 	 *   1) CPU could have been running below the maximum TSC rate
2037 	 *   2) Broken TSC compensation resets the base at each VCPU
2038 	 *      entry to avoid unknown leaps of TSC even when running
2039 	 *      again on the same CPU.  This may cause apparent elapsed
2040 	 *      time to disappear, and the guest to stand still or run
2041 	 *	very slowly.
2042 	 */
2043 	if (vcpu->tsc_catchup) {
2044 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2045 		if (tsc > tsc_timestamp) {
2046 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2047 			tsc_timestamp = tsc;
2048 		}
2049 	}
2050 
2051 	local_irq_restore(flags);
2052 
2053 	/* With all the info we got, fill in the values */
2054 
2055 	if (kvm_has_tsc_control)
2056 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2057 
2058 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2059 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2060 				   &vcpu->hv_clock.tsc_shift,
2061 				   &vcpu->hv_clock.tsc_to_system_mul);
2062 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2063 	}
2064 
2065 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2066 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2067 	vcpu->last_guest_tsc = tsc_timestamp;
2068 
2069 	/* If the host uses TSC clocksource, then it is stable */
2070 	pvclock_flags = 0;
2071 	if (use_master_clock)
2072 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2073 
2074 	vcpu->hv_clock.flags = pvclock_flags;
2075 
2076 	if (vcpu->pv_time_enabled)
2077 		kvm_setup_pvclock_page(v);
2078 	if (v == kvm_get_vcpu(v->kvm, 0))
2079 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2080 	return 0;
2081 }
2082 
2083 /*
2084  * kvmclock updates which are isolated to a given vcpu, such as
2085  * vcpu->cpu migration, should not allow system_timestamp from
2086  * the rest of the vcpus to remain static. Otherwise ntp frequency
2087  * correction applies to one vcpu's system_timestamp but not
2088  * the others.
2089  *
2090  * So in those cases, request a kvmclock update for all vcpus.
2091  * We need to rate-limit these requests though, as they can
2092  * considerably slow guests that have a large number of vcpus.
2093  * The time for a remote vcpu to update its kvmclock is bound
2094  * by the delay we use to rate-limit the updates.
2095  */
2096 
2097 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2098 
2099 static void kvmclock_update_fn(struct work_struct *work)
2100 {
2101 	int i;
2102 	struct delayed_work *dwork = to_delayed_work(work);
2103 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2104 					   kvmclock_update_work);
2105 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2106 	struct kvm_vcpu *vcpu;
2107 
2108 	kvm_for_each_vcpu(i, vcpu, kvm) {
2109 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2110 		kvm_vcpu_kick(vcpu);
2111 	}
2112 }
2113 
2114 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2115 {
2116 	struct kvm *kvm = v->kvm;
2117 
2118 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2119 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2120 					KVMCLOCK_UPDATE_DELAY);
2121 }
2122 
2123 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2124 
2125 static void kvmclock_sync_fn(struct work_struct *work)
2126 {
2127 	struct delayed_work *dwork = to_delayed_work(work);
2128 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2129 					   kvmclock_sync_work);
2130 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2131 
2132 	if (!kvmclock_periodic_sync)
2133 		return;
2134 
2135 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2136 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2137 					KVMCLOCK_SYNC_PERIOD);
2138 }
2139 
2140 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2141 {
2142 	u64 mcg_cap = vcpu->arch.mcg_cap;
2143 	unsigned bank_num = mcg_cap & 0xff;
2144 	u32 msr = msr_info->index;
2145 	u64 data = msr_info->data;
2146 
2147 	switch (msr) {
2148 	case MSR_IA32_MCG_STATUS:
2149 		vcpu->arch.mcg_status = data;
2150 		break;
2151 	case MSR_IA32_MCG_CTL:
2152 		if (!(mcg_cap & MCG_CTL_P))
2153 			return 1;
2154 		if (data != 0 && data != ~(u64)0)
2155 			return -1;
2156 		vcpu->arch.mcg_ctl = data;
2157 		break;
2158 	default:
2159 		if (msr >= MSR_IA32_MC0_CTL &&
2160 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2161 			u32 offset = msr - MSR_IA32_MC0_CTL;
2162 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2163 			 * some Linux kernels though clear bit 10 in bank 4 to
2164 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2165 			 * this to avoid an uncatched #GP in the guest
2166 			 */
2167 			if ((offset & 0x3) == 0 &&
2168 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2169 				return -1;
2170 			if (!msr_info->host_initiated &&
2171 				(offset & 0x3) == 1 && data != 0)
2172 				return -1;
2173 			vcpu->arch.mce_banks[offset] = data;
2174 			break;
2175 		}
2176 		return 1;
2177 	}
2178 	return 0;
2179 }
2180 
2181 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2182 {
2183 	struct kvm *kvm = vcpu->kvm;
2184 	int lm = is_long_mode(vcpu);
2185 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2186 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2187 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2188 		: kvm->arch.xen_hvm_config.blob_size_32;
2189 	u32 page_num = data & ~PAGE_MASK;
2190 	u64 page_addr = data & PAGE_MASK;
2191 	u8 *page;
2192 	int r;
2193 
2194 	r = -E2BIG;
2195 	if (page_num >= blob_size)
2196 		goto out;
2197 	r = -ENOMEM;
2198 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2199 	if (IS_ERR(page)) {
2200 		r = PTR_ERR(page);
2201 		goto out;
2202 	}
2203 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2204 		goto out_free;
2205 	r = 0;
2206 out_free:
2207 	kfree(page);
2208 out:
2209 	return r;
2210 }
2211 
2212 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2213 {
2214 	gpa_t gpa = data & ~0x3f;
2215 
2216 	/* Bits 3:5 are reserved, Should be zero */
2217 	if (data & 0x38)
2218 		return 1;
2219 
2220 	vcpu->arch.apf.msr_val = data;
2221 
2222 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2223 		kvm_clear_async_pf_completion_queue(vcpu);
2224 		kvm_async_pf_hash_reset(vcpu);
2225 		return 0;
2226 	}
2227 
2228 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2229 					sizeof(u32)))
2230 		return 1;
2231 
2232 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2233 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2234 	kvm_async_pf_wakeup_all(vcpu);
2235 	return 0;
2236 }
2237 
2238 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2239 {
2240 	vcpu->arch.pv_time_enabled = false;
2241 }
2242 
2243 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2244 {
2245 	++vcpu->stat.tlb_flush;
2246 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2247 }
2248 
2249 static void record_steal_time(struct kvm_vcpu *vcpu)
2250 {
2251 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2252 		return;
2253 
2254 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2255 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2256 		return;
2257 
2258 	/*
2259 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2260 	 * expensive IPIs.
2261 	 */
2262 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2263 		kvm_vcpu_flush_tlb(vcpu, false);
2264 
2265 	if (vcpu->arch.st.steal.version & 1)
2266 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2267 
2268 	vcpu->arch.st.steal.version += 1;
2269 
2270 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2271 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2272 
2273 	smp_wmb();
2274 
2275 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2276 		vcpu->arch.st.last_steal;
2277 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2278 
2279 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2280 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2281 
2282 	smp_wmb();
2283 
2284 	vcpu->arch.st.steal.version += 1;
2285 
2286 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2287 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2288 }
2289 
2290 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2291 {
2292 	bool pr = false;
2293 	u32 msr = msr_info->index;
2294 	u64 data = msr_info->data;
2295 
2296 	switch (msr) {
2297 	case MSR_AMD64_NB_CFG:
2298 	case MSR_IA32_UCODE_WRITE:
2299 	case MSR_VM_HSAVE_PA:
2300 	case MSR_AMD64_PATCH_LOADER:
2301 	case MSR_AMD64_BU_CFG2:
2302 	case MSR_AMD64_DC_CFG:
2303 		break;
2304 
2305 	case MSR_IA32_UCODE_REV:
2306 		if (msr_info->host_initiated)
2307 			vcpu->arch.microcode_version = data;
2308 		break;
2309 	case MSR_EFER:
2310 		return set_efer(vcpu, data);
2311 	case MSR_K7_HWCR:
2312 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2313 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2314 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2315 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2316 		if (data != 0) {
2317 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2318 				    data);
2319 			return 1;
2320 		}
2321 		break;
2322 	case MSR_FAM10H_MMIO_CONF_BASE:
2323 		if (data != 0) {
2324 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2325 				    "0x%llx\n", data);
2326 			return 1;
2327 		}
2328 		break;
2329 	case MSR_IA32_DEBUGCTLMSR:
2330 		if (!data) {
2331 			/* We support the non-activated case already */
2332 			break;
2333 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2334 			/* Values other than LBR and BTF are vendor-specific,
2335 			   thus reserved and should throw a #GP */
2336 			return 1;
2337 		}
2338 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2339 			    __func__, data);
2340 		break;
2341 	case 0x200 ... 0x2ff:
2342 		return kvm_mtrr_set_msr(vcpu, msr, data);
2343 	case MSR_IA32_APICBASE:
2344 		return kvm_set_apic_base(vcpu, msr_info);
2345 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2346 		return kvm_x2apic_msr_write(vcpu, msr, data);
2347 	case MSR_IA32_TSCDEADLINE:
2348 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2349 		break;
2350 	case MSR_IA32_TSC_ADJUST:
2351 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2352 			if (!msr_info->host_initiated) {
2353 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2354 				adjust_tsc_offset_guest(vcpu, adj);
2355 			}
2356 			vcpu->arch.ia32_tsc_adjust_msr = data;
2357 		}
2358 		break;
2359 	case MSR_IA32_MISC_ENABLE:
2360 		vcpu->arch.ia32_misc_enable_msr = data;
2361 		break;
2362 	case MSR_IA32_SMBASE:
2363 		if (!msr_info->host_initiated)
2364 			return 1;
2365 		vcpu->arch.smbase = data;
2366 		break;
2367 	case MSR_IA32_TSC:
2368 		kvm_write_tsc(vcpu, msr_info);
2369 		break;
2370 	case MSR_SMI_COUNT:
2371 		if (!msr_info->host_initiated)
2372 			return 1;
2373 		vcpu->arch.smi_count = data;
2374 		break;
2375 	case MSR_KVM_WALL_CLOCK_NEW:
2376 	case MSR_KVM_WALL_CLOCK:
2377 		vcpu->kvm->arch.wall_clock = data;
2378 		kvm_write_wall_clock(vcpu->kvm, data);
2379 		break;
2380 	case MSR_KVM_SYSTEM_TIME_NEW:
2381 	case MSR_KVM_SYSTEM_TIME: {
2382 		struct kvm_arch *ka = &vcpu->kvm->arch;
2383 
2384 		kvmclock_reset(vcpu);
2385 
2386 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2387 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2388 
2389 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2390 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2391 
2392 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2393 		}
2394 
2395 		vcpu->arch.time = data;
2396 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2397 
2398 		/* we verify if the enable bit is set... */
2399 		if (!(data & 1))
2400 			break;
2401 
2402 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2403 		     &vcpu->arch.pv_time, data & ~1ULL,
2404 		     sizeof(struct pvclock_vcpu_time_info)))
2405 			vcpu->arch.pv_time_enabled = false;
2406 		else
2407 			vcpu->arch.pv_time_enabled = true;
2408 
2409 		break;
2410 	}
2411 	case MSR_KVM_ASYNC_PF_EN:
2412 		if (kvm_pv_enable_async_pf(vcpu, data))
2413 			return 1;
2414 		break;
2415 	case MSR_KVM_STEAL_TIME:
2416 
2417 		if (unlikely(!sched_info_on()))
2418 			return 1;
2419 
2420 		if (data & KVM_STEAL_RESERVED_MASK)
2421 			return 1;
2422 
2423 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2424 						data & KVM_STEAL_VALID_BITS,
2425 						sizeof(struct kvm_steal_time)))
2426 			return 1;
2427 
2428 		vcpu->arch.st.msr_val = data;
2429 
2430 		if (!(data & KVM_MSR_ENABLED))
2431 			break;
2432 
2433 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2434 
2435 		break;
2436 	case MSR_KVM_PV_EOI_EN:
2437 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2438 			return 1;
2439 		break;
2440 
2441 	case MSR_IA32_MCG_CTL:
2442 	case MSR_IA32_MCG_STATUS:
2443 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2444 		return set_msr_mce(vcpu, msr_info);
2445 
2446 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2447 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2448 		pr = true; /* fall through */
2449 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2450 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2451 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2452 			return kvm_pmu_set_msr(vcpu, msr_info);
2453 
2454 		if (pr || data != 0)
2455 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2456 				    "0x%x data 0x%llx\n", msr, data);
2457 		break;
2458 	case MSR_K7_CLK_CTL:
2459 		/*
2460 		 * Ignore all writes to this no longer documented MSR.
2461 		 * Writes are only relevant for old K7 processors,
2462 		 * all pre-dating SVM, but a recommended workaround from
2463 		 * AMD for these chips. It is possible to specify the
2464 		 * affected processor models on the command line, hence
2465 		 * the need to ignore the workaround.
2466 		 */
2467 		break;
2468 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2469 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2470 	case HV_X64_MSR_CRASH_CTL:
2471 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2472 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2473 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2474 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2475 		return kvm_hv_set_msr_common(vcpu, msr, data,
2476 					     msr_info->host_initiated);
2477 	case MSR_IA32_BBL_CR_CTL3:
2478 		/* Drop writes to this legacy MSR -- see rdmsr
2479 		 * counterpart for further detail.
2480 		 */
2481 		if (report_ignored_msrs)
2482 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2483 				msr, data);
2484 		break;
2485 	case MSR_AMD64_OSVW_ID_LENGTH:
2486 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2487 			return 1;
2488 		vcpu->arch.osvw.length = data;
2489 		break;
2490 	case MSR_AMD64_OSVW_STATUS:
2491 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2492 			return 1;
2493 		vcpu->arch.osvw.status = data;
2494 		break;
2495 	case MSR_PLATFORM_INFO:
2496 		if (!msr_info->host_initiated ||
2497 		    data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2498 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2499 		     cpuid_fault_enabled(vcpu)))
2500 			return 1;
2501 		vcpu->arch.msr_platform_info = data;
2502 		break;
2503 	case MSR_MISC_FEATURES_ENABLES:
2504 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2505 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2506 		     !supports_cpuid_fault(vcpu)))
2507 			return 1;
2508 		vcpu->arch.msr_misc_features_enables = data;
2509 		break;
2510 	default:
2511 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2512 			return xen_hvm_config(vcpu, data);
2513 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2514 			return kvm_pmu_set_msr(vcpu, msr_info);
2515 		if (!ignore_msrs) {
2516 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2517 				    msr, data);
2518 			return 1;
2519 		} else {
2520 			if (report_ignored_msrs)
2521 				vcpu_unimpl(vcpu,
2522 					"ignored wrmsr: 0x%x data 0x%llx\n",
2523 					msr, data);
2524 			break;
2525 		}
2526 	}
2527 	return 0;
2528 }
2529 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2530 
2531 
2532 /*
2533  * Reads an msr value (of 'msr_index') into 'pdata'.
2534  * Returns 0 on success, non-0 otherwise.
2535  * Assumes vcpu_load() was already called.
2536  */
2537 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2538 {
2539 	return kvm_x86_ops->get_msr(vcpu, msr);
2540 }
2541 EXPORT_SYMBOL_GPL(kvm_get_msr);
2542 
2543 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2544 {
2545 	u64 data;
2546 	u64 mcg_cap = vcpu->arch.mcg_cap;
2547 	unsigned bank_num = mcg_cap & 0xff;
2548 
2549 	switch (msr) {
2550 	case MSR_IA32_P5_MC_ADDR:
2551 	case MSR_IA32_P5_MC_TYPE:
2552 		data = 0;
2553 		break;
2554 	case MSR_IA32_MCG_CAP:
2555 		data = vcpu->arch.mcg_cap;
2556 		break;
2557 	case MSR_IA32_MCG_CTL:
2558 		if (!(mcg_cap & MCG_CTL_P))
2559 			return 1;
2560 		data = vcpu->arch.mcg_ctl;
2561 		break;
2562 	case MSR_IA32_MCG_STATUS:
2563 		data = vcpu->arch.mcg_status;
2564 		break;
2565 	default:
2566 		if (msr >= MSR_IA32_MC0_CTL &&
2567 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2568 			u32 offset = msr - MSR_IA32_MC0_CTL;
2569 			data = vcpu->arch.mce_banks[offset];
2570 			break;
2571 		}
2572 		return 1;
2573 	}
2574 	*pdata = data;
2575 	return 0;
2576 }
2577 
2578 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2579 {
2580 	switch (msr_info->index) {
2581 	case MSR_IA32_PLATFORM_ID:
2582 	case MSR_IA32_EBL_CR_POWERON:
2583 	case MSR_IA32_DEBUGCTLMSR:
2584 	case MSR_IA32_LASTBRANCHFROMIP:
2585 	case MSR_IA32_LASTBRANCHTOIP:
2586 	case MSR_IA32_LASTINTFROMIP:
2587 	case MSR_IA32_LASTINTTOIP:
2588 	case MSR_K8_SYSCFG:
2589 	case MSR_K8_TSEG_ADDR:
2590 	case MSR_K8_TSEG_MASK:
2591 	case MSR_K7_HWCR:
2592 	case MSR_VM_HSAVE_PA:
2593 	case MSR_K8_INT_PENDING_MSG:
2594 	case MSR_AMD64_NB_CFG:
2595 	case MSR_FAM10H_MMIO_CONF_BASE:
2596 	case MSR_AMD64_BU_CFG2:
2597 	case MSR_IA32_PERF_CTL:
2598 	case MSR_AMD64_DC_CFG:
2599 		msr_info->data = 0;
2600 		break;
2601 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2602 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2603 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2604 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2605 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2606 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2607 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2608 		msr_info->data = 0;
2609 		break;
2610 	case MSR_IA32_UCODE_REV:
2611 		msr_info->data = vcpu->arch.microcode_version;
2612 		break;
2613 	case MSR_IA32_TSC:
2614 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2615 		break;
2616 	case MSR_MTRRcap:
2617 	case 0x200 ... 0x2ff:
2618 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2619 	case 0xcd: /* fsb frequency */
2620 		msr_info->data = 3;
2621 		break;
2622 		/*
2623 		 * MSR_EBC_FREQUENCY_ID
2624 		 * Conservative value valid for even the basic CPU models.
2625 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2626 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2627 		 * and 266MHz for model 3, or 4. Set Core Clock
2628 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2629 		 * 31:24) even though these are only valid for CPU
2630 		 * models > 2, however guests may end up dividing or
2631 		 * multiplying by zero otherwise.
2632 		 */
2633 	case MSR_EBC_FREQUENCY_ID:
2634 		msr_info->data = 1 << 24;
2635 		break;
2636 	case MSR_IA32_APICBASE:
2637 		msr_info->data = kvm_get_apic_base(vcpu);
2638 		break;
2639 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2640 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2641 		break;
2642 	case MSR_IA32_TSCDEADLINE:
2643 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2644 		break;
2645 	case MSR_IA32_TSC_ADJUST:
2646 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2647 		break;
2648 	case MSR_IA32_MISC_ENABLE:
2649 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2650 		break;
2651 	case MSR_IA32_SMBASE:
2652 		if (!msr_info->host_initiated)
2653 			return 1;
2654 		msr_info->data = vcpu->arch.smbase;
2655 		break;
2656 	case MSR_SMI_COUNT:
2657 		msr_info->data = vcpu->arch.smi_count;
2658 		break;
2659 	case MSR_IA32_PERF_STATUS:
2660 		/* TSC increment by tick */
2661 		msr_info->data = 1000ULL;
2662 		/* CPU multiplier */
2663 		msr_info->data |= (((uint64_t)4ULL) << 40);
2664 		break;
2665 	case MSR_EFER:
2666 		msr_info->data = vcpu->arch.efer;
2667 		break;
2668 	case MSR_KVM_WALL_CLOCK:
2669 	case MSR_KVM_WALL_CLOCK_NEW:
2670 		msr_info->data = vcpu->kvm->arch.wall_clock;
2671 		break;
2672 	case MSR_KVM_SYSTEM_TIME:
2673 	case MSR_KVM_SYSTEM_TIME_NEW:
2674 		msr_info->data = vcpu->arch.time;
2675 		break;
2676 	case MSR_KVM_ASYNC_PF_EN:
2677 		msr_info->data = vcpu->arch.apf.msr_val;
2678 		break;
2679 	case MSR_KVM_STEAL_TIME:
2680 		msr_info->data = vcpu->arch.st.msr_val;
2681 		break;
2682 	case MSR_KVM_PV_EOI_EN:
2683 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2684 		break;
2685 	case MSR_IA32_P5_MC_ADDR:
2686 	case MSR_IA32_P5_MC_TYPE:
2687 	case MSR_IA32_MCG_CAP:
2688 	case MSR_IA32_MCG_CTL:
2689 	case MSR_IA32_MCG_STATUS:
2690 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2691 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2692 	case MSR_K7_CLK_CTL:
2693 		/*
2694 		 * Provide expected ramp-up count for K7. All other
2695 		 * are set to zero, indicating minimum divisors for
2696 		 * every field.
2697 		 *
2698 		 * This prevents guest kernels on AMD host with CPU
2699 		 * type 6, model 8 and higher from exploding due to
2700 		 * the rdmsr failing.
2701 		 */
2702 		msr_info->data = 0x20000000;
2703 		break;
2704 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2705 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2706 	case HV_X64_MSR_CRASH_CTL:
2707 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2708 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2709 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2710 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2711 		return kvm_hv_get_msr_common(vcpu,
2712 					     msr_info->index, &msr_info->data);
2713 		break;
2714 	case MSR_IA32_BBL_CR_CTL3:
2715 		/* This legacy MSR exists but isn't fully documented in current
2716 		 * silicon.  It is however accessed by winxp in very narrow
2717 		 * scenarios where it sets bit #19, itself documented as
2718 		 * a "reserved" bit.  Best effort attempt to source coherent
2719 		 * read data here should the balance of the register be
2720 		 * interpreted by the guest:
2721 		 *
2722 		 * L2 cache control register 3: 64GB range, 256KB size,
2723 		 * enabled, latency 0x1, configured
2724 		 */
2725 		msr_info->data = 0xbe702111;
2726 		break;
2727 	case MSR_AMD64_OSVW_ID_LENGTH:
2728 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2729 			return 1;
2730 		msr_info->data = vcpu->arch.osvw.length;
2731 		break;
2732 	case MSR_AMD64_OSVW_STATUS:
2733 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2734 			return 1;
2735 		msr_info->data = vcpu->arch.osvw.status;
2736 		break;
2737 	case MSR_PLATFORM_INFO:
2738 		msr_info->data = vcpu->arch.msr_platform_info;
2739 		break;
2740 	case MSR_MISC_FEATURES_ENABLES:
2741 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2742 		break;
2743 	default:
2744 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2745 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2746 		if (!ignore_msrs) {
2747 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2748 					       msr_info->index);
2749 			return 1;
2750 		} else {
2751 			if (report_ignored_msrs)
2752 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2753 					msr_info->index);
2754 			msr_info->data = 0;
2755 		}
2756 		break;
2757 	}
2758 	return 0;
2759 }
2760 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2761 
2762 /*
2763  * Read or write a bunch of msrs. All parameters are kernel addresses.
2764  *
2765  * @return number of msrs set successfully.
2766  */
2767 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2768 		    struct kvm_msr_entry *entries,
2769 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2770 				  unsigned index, u64 *data))
2771 {
2772 	int i;
2773 
2774 	for (i = 0; i < msrs->nmsrs; ++i)
2775 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2776 			break;
2777 
2778 	return i;
2779 }
2780 
2781 /*
2782  * Read or write a bunch of msrs. Parameters are user addresses.
2783  *
2784  * @return number of msrs set successfully.
2785  */
2786 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2787 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2788 				unsigned index, u64 *data),
2789 		  int writeback)
2790 {
2791 	struct kvm_msrs msrs;
2792 	struct kvm_msr_entry *entries;
2793 	int r, n;
2794 	unsigned size;
2795 
2796 	r = -EFAULT;
2797 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2798 		goto out;
2799 
2800 	r = -E2BIG;
2801 	if (msrs.nmsrs >= MAX_IO_MSRS)
2802 		goto out;
2803 
2804 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2805 	entries = memdup_user(user_msrs->entries, size);
2806 	if (IS_ERR(entries)) {
2807 		r = PTR_ERR(entries);
2808 		goto out;
2809 	}
2810 
2811 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2812 	if (r < 0)
2813 		goto out_free;
2814 
2815 	r = -EFAULT;
2816 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2817 		goto out_free;
2818 
2819 	r = n;
2820 
2821 out_free:
2822 	kfree(entries);
2823 out:
2824 	return r;
2825 }
2826 
2827 static inline bool kvm_can_mwait_in_guest(void)
2828 {
2829 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2830 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2831 		boot_cpu_has(X86_FEATURE_ARAT);
2832 }
2833 
2834 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2835 {
2836 	int r = 0;
2837 
2838 	switch (ext) {
2839 	case KVM_CAP_IRQCHIP:
2840 	case KVM_CAP_HLT:
2841 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2842 	case KVM_CAP_SET_TSS_ADDR:
2843 	case KVM_CAP_EXT_CPUID:
2844 	case KVM_CAP_EXT_EMUL_CPUID:
2845 	case KVM_CAP_CLOCKSOURCE:
2846 	case KVM_CAP_PIT:
2847 	case KVM_CAP_NOP_IO_DELAY:
2848 	case KVM_CAP_MP_STATE:
2849 	case KVM_CAP_SYNC_MMU:
2850 	case KVM_CAP_USER_NMI:
2851 	case KVM_CAP_REINJECT_CONTROL:
2852 	case KVM_CAP_IRQ_INJECT_STATUS:
2853 	case KVM_CAP_IOEVENTFD:
2854 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2855 	case KVM_CAP_PIT2:
2856 	case KVM_CAP_PIT_STATE2:
2857 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2858 	case KVM_CAP_XEN_HVM:
2859 	case KVM_CAP_VCPU_EVENTS:
2860 	case KVM_CAP_HYPERV:
2861 	case KVM_CAP_HYPERV_VAPIC:
2862 	case KVM_CAP_HYPERV_SPIN:
2863 	case KVM_CAP_HYPERV_SYNIC:
2864 	case KVM_CAP_HYPERV_SYNIC2:
2865 	case KVM_CAP_HYPERV_VP_INDEX:
2866 	case KVM_CAP_HYPERV_EVENTFD:
2867 	case KVM_CAP_PCI_SEGMENT:
2868 	case KVM_CAP_DEBUGREGS:
2869 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2870 	case KVM_CAP_XSAVE:
2871 	case KVM_CAP_ASYNC_PF:
2872 	case KVM_CAP_GET_TSC_KHZ:
2873 	case KVM_CAP_KVMCLOCK_CTRL:
2874 	case KVM_CAP_READONLY_MEM:
2875 	case KVM_CAP_HYPERV_TIME:
2876 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2877 	case KVM_CAP_TSC_DEADLINE_TIMER:
2878 	case KVM_CAP_ENABLE_CAP_VM:
2879 	case KVM_CAP_DISABLE_QUIRKS:
2880 	case KVM_CAP_SET_BOOT_CPU_ID:
2881  	case KVM_CAP_SPLIT_IRQCHIP:
2882 	case KVM_CAP_IMMEDIATE_EXIT:
2883 	case KVM_CAP_GET_MSR_FEATURES:
2884 		r = 1;
2885 		break;
2886 	case KVM_CAP_SYNC_REGS:
2887 		r = KVM_SYNC_X86_VALID_FIELDS;
2888 		break;
2889 	case KVM_CAP_ADJUST_CLOCK:
2890 		r = KVM_CLOCK_TSC_STABLE;
2891 		break;
2892 	case KVM_CAP_X86_DISABLE_EXITS:
2893 		r |=  KVM_X86_DISABLE_EXITS_HTL | KVM_X86_DISABLE_EXITS_PAUSE;
2894 		if(kvm_can_mwait_in_guest())
2895 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
2896 		break;
2897 	case KVM_CAP_X86_SMM:
2898 		/* SMBASE is usually relocated above 1M on modern chipsets,
2899 		 * and SMM handlers might indeed rely on 4G segment limits,
2900 		 * so do not report SMM to be available if real mode is
2901 		 * emulated via vm86 mode.  Still, do not go to great lengths
2902 		 * to avoid userspace's usage of the feature, because it is a
2903 		 * fringe case that is not enabled except via specific settings
2904 		 * of the module parameters.
2905 		 */
2906 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2907 		break;
2908 	case KVM_CAP_VAPIC:
2909 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2910 		break;
2911 	case KVM_CAP_NR_VCPUS:
2912 		r = KVM_SOFT_MAX_VCPUS;
2913 		break;
2914 	case KVM_CAP_MAX_VCPUS:
2915 		r = KVM_MAX_VCPUS;
2916 		break;
2917 	case KVM_CAP_NR_MEMSLOTS:
2918 		r = KVM_USER_MEM_SLOTS;
2919 		break;
2920 	case KVM_CAP_PV_MMU:	/* obsolete */
2921 		r = 0;
2922 		break;
2923 	case KVM_CAP_MCE:
2924 		r = KVM_MAX_MCE_BANKS;
2925 		break;
2926 	case KVM_CAP_XCRS:
2927 		r = boot_cpu_has(X86_FEATURE_XSAVE);
2928 		break;
2929 	case KVM_CAP_TSC_CONTROL:
2930 		r = kvm_has_tsc_control;
2931 		break;
2932 	case KVM_CAP_X2APIC_API:
2933 		r = KVM_X2APIC_API_VALID_FLAGS;
2934 		break;
2935 	default:
2936 		break;
2937 	}
2938 	return r;
2939 
2940 }
2941 
2942 long kvm_arch_dev_ioctl(struct file *filp,
2943 			unsigned int ioctl, unsigned long arg)
2944 {
2945 	void __user *argp = (void __user *)arg;
2946 	long r;
2947 
2948 	switch (ioctl) {
2949 	case KVM_GET_MSR_INDEX_LIST: {
2950 		struct kvm_msr_list __user *user_msr_list = argp;
2951 		struct kvm_msr_list msr_list;
2952 		unsigned n;
2953 
2954 		r = -EFAULT;
2955 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2956 			goto out;
2957 		n = msr_list.nmsrs;
2958 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2959 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2960 			goto out;
2961 		r = -E2BIG;
2962 		if (n < msr_list.nmsrs)
2963 			goto out;
2964 		r = -EFAULT;
2965 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2966 				 num_msrs_to_save * sizeof(u32)))
2967 			goto out;
2968 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2969 				 &emulated_msrs,
2970 				 num_emulated_msrs * sizeof(u32)))
2971 			goto out;
2972 		r = 0;
2973 		break;
2974 	}
2975 	case KVM_GET_SUPPORTED_CPUID:
2976 	case KVM_GET_EMULATED_CPUID: {
2977 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2978 		struct kvm_cpuid2 cpuid;
2979 
2980 		r = -EFAULT;
2981 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2982 			goto out;
2983 
2984 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2985 					    ioctl);
2986 		if (r)
2987 			goto out;
2988 
2989 		r = -EFAULT;
2990 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2991 			goto out;
2992 		r = 0;
2993 		break;
2994 	}
2995 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2996 		r = -EFAULT;
2997 		if (copy_to_user(argp, &kvm_mce_cap_supported,
2998 				 sizeof(kvm_mce_cap_supported)))
2999 			goto out;
3000 		r = 0;
3001 		break;
3002 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3003 		struct kvm_msr_list __user *user_msr_list = argp;
3004 		struct kvm_msr_list msr_list;
3005 		unsigned int n;
3006 
3007 		r = -EFAULT;
3008 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3009 			goto out;
3010 		n = msr_list.nmsrs;
3011 		msr_list.nmsrs = num_msr_based_features;
3012 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3013 			goto out;
3014 		r = -E2BIG;
3015 		if (n < msr_list.nmsrs)
3016 			goto out;
3017 		r = -EFAULT;
3018 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3019 				 num_msr_based_features * sizeof(u32)))
3020 			goto out;
3021 		r = 0;
3022 		break;
3023 	}
3024 	case KVM_GET_MSRS:
3025 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3026 		break;
3027 	}
3028 	default:
3029 		r = -EINVAL;
3030 	}
3031 out:
3032 	return r;
3033 }
3034 
3035 static void wbinvd_ipi(void *garbage)
3036 {
3037 	wbinvd();
3038 }
3039 
3040 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3041 {
3042 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3043 }
3044 
3045 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3046 {
3047 	/* Address WBINVD may be executed by guest */
3048 	if (need_emulate_wbinvd(vcpu)) {
3049 		if (kvm_x86_ops->has_wbinvd_exit())
3050 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3051 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3052 			smp_call_function_single(vcpu->cpu,
3053 					wbinvd_ipi, NULL, 1);
3054 	}
3055 
3056 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3057 
3058 	/* Apply any externally detected TSC adjustments (due to suspend) */
3059 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3060 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3061 		vcpu->arch.tsc_offset_adjustment = 0;
3062 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3063 	}
3064 
3065 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3066 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3067 				rdtsc() - vcpu->arch.last_host_tsc;
3068 		if (tsc_delta < 0)
3069 			mark_tsc_unstable("KVM discovered backwards TSC");
3070 
3071 		if (kvm_check_tsc_unstable()) {
3072 			u64 offset = kvm_compute_tsc_offset(vcpu,
3073 						vcpu->arch.last_guest_tsc);
3074 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3075 			vcpu->arch.tsc_catchup = 1;
3076 		}
3077 
3078 		if (kvm_lapic_hv_timer_in_use(vcpu))
3079 			kvm_lapic_restart_hv_timer(vcpu);
3080 
3081 		/*
3082 		 * On a host with synchronized TSC, there is no need to update
3083 		 * kvmclock on vcpu->cpu migration
3084 		 */
3085 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3086 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3087 		if (vcpu->cpu != cpu)
3088 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3089 		vcpu->cpu = cpu;
3090 	}
3091 
3092 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3093 }
3094 
3095 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3096 {
3097 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3098 		return;
3099 
3100 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3101 
3102 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3103 			&vcpu->arch.st.steal.preempted,
3104 			offsetof(struct kvm_steal_time, preempted),
3105 			sizeof(vcpu->arch.st.steal.preempted));
3106 }
3107 
3108 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3109 {
3110 	int idx;
3111 
3112 	if (vcpu->preempted)
3113 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3114 
3115 	/*
3116 	 * Disable page faults because we're in atomic context here.
3117 	 * kvm_write_guest_offset_cached() would call might_fault()
3118 	 * that relies on pagefault_disable() to tell if there's a
3119 	 * bug. NOTE: the write to guest memory may not go through if
3120 	 * during postcopy live migration or if there's heavy guest
3121 	 * paging.
3122 	 */
3123 	pagefault_disable();
3124 	/*
3125 	 * kvm_memslots() will be called by
3126 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3127 	 */
3128 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3129 	kvm_steal_time_set_preempted(vcpu);
3130 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3131 	pagefault_enable();
3132 	kvm_x86_ops->vcpu_put(vcpu);
3133 	vcpu->arch.last_host_tsc = rdtsc();
3134 	/*
3135 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3136 	 * on every vmexit, but if not, we might have a stale dr6 from the
3137 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3138 	 */
3139 	set_debugreg(0, 6);
3140 }
3141 
3142 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3143 				    struct kvm_lapic_state *s)
3144 {
3145 	if (vcpu->arch.apicv_active)
3146 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3147 
3148 	return kvm_apic_get_state(vcpu, s);
3149 }
3150 
3151 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3152 				    struct kvm_lapic_state *s)
3153 {
3154 	int r;
3155 
3156 	r = kvm_apic_set_state(vcpu, s);
3157 	if (r)
3158 		return r;
3159 	update_cr8_intercept(vcpu);
3160 
3161 	return 0;
3162 }
3163 
3164 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3165 {
3166 	return (!lapic_in_kernel(vcpu) ||
3167 		kvm_apic_accept_pic_intr(vcpu));
3168 }
3169 
3170 /*
3171  * if userspace requested an interrupt window, check that the
3172  * interrupt window is open.
3173  *
3174  * No need to exit to userspace if we already have an interrupt queued.
3175  */
3176 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3177 {
3178 	return kvm_arch_interrupt_allowed(vcpu) &&
3179 		!kvm_cpu_has_interrupt(vcpu) &&
3180 		!kvm_event_needs_reinjection(vcpu) &&
3181 		kvm_cpu_accept_dm_intr(vcpu);
3182 }
3183 
3184 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3185 				    struct kvm_interrupt *irq)
3186 {
3187 	if (irq->irq >= KVM_NR_INTERRUPTS)
3188 		return -EINVAL;
3189 
3190 	if (!irqchip_in_kernel(vcpu->kvm)) {
3191 		kvm_queue_interrupt(vcpu, irq->irq, false);
3192 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3193 		return 0;
3194 	}
3195 
3196 	/*
3197 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3198 	 * fail for in-kernel 8259.
3199 	 */
3200 	if (pic_in_kernel(vcpu->kvm))
3201 		return -ENXIO;
3202 
3203 	if (vcpu->arch.pending_external_vector != -1)
3204 		return -EEXIST;
3205 
3206 	vcpu->arch.pending_external_vector = irq->irq;
3207 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3208 	return 0;
3209 }
3210 
3211 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3212 {
3213 	kvm_inject_nmi(vcpu);
3214 
3215 	return 0;
3216 }
3217 
3218 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3219 {
3220 	kvm_make_request(KVM_REQ_SMI, vcpu);
3221 
3222 	return 0;
3223 }
3224 
3225 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3226 					   struct kvm_tpr_access_ctl *tac)
3227 {
3228 	if (tac->flags)
3229 		return -EINVAL;
3230 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3231 	return 0;
3232 }
3233 
3234 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3235 					u64 mcg_cap)
3236 {
3237 	int r;
3238 	unsigned bank_num = mcg_cap & 0xff, bank;
3239 
3240 	r = -EINVAL;
3241 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3242 		goto out;
3243 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3244 		goto out;
3245 	r = 0;
3246 	vcpu->arch.mcg_cap = mcg_cap;
3247 	/* Init IA32_MCG_CTL to all 1s */
3248 	if (mcg_cap & MCG_CTL_P)
3249 		vcpu->arch.mcg_ctl = ~(u64)0;
3250 	/* Init IA32_MCi_CTL to all 1s */
3251 	for (bank = 0; bank < bank_num; bank++)
3252 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3253 
3254 	if (kvm_x86_ops->setup_mce)
3255 		kvm_x86_ops->setup_mce(vcpu);
3256 out:
3257 	return r;
3258 }
3259 
3260 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3261 				      struct kvm_x86_mce *mce)
3262 {
3263 	u64 mcg_cap = vcpu->arch.mcg_cap;
3264 	unsigned bank_num = mcg_cap & 0xff;
3265 	u64 *banks = vcpu->arch.mce_banks;
3266 
3267 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3268 		return -EINVAL;
3269 	/*
3270 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3271 	 * reporting is disabled
3272 	 */
3273 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3274 	    vcpu->arch.mcg_ctl != ~(u64)0)
3275 		return 0;
3276 	banks += 4 * mce->bank;
3277 	/*
3278 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3279 	 * reporting is disabled for the bank
3280 	 */
3281 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3282 		return 0;
3283 	if (mce->status & MCI_STATUS_UC) {
3284 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3285 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3286 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3287 			return 0;
3288 		}
3289 		if (banks[1] & MCI_STATUS_VAL)
3290 			mce->status |= MCI_STATUS_OVER;
3291 		banks[2] = mce->addr;
3292 		banks[3] = mce->misc;
3293 		vcpu->arch.mcg_status = mce->mcg_status;
3294 		banks[1] = mce->status;
3295 		kvm_queue_exception(vcpu, MC_VECTOR);
3296 	} else if (!(banks[1] & MCI_STATUS_VAL)
3297 		   || !(banks[1] & MCI_STATUS_UC)) {
3298 		if (banks[1] & MCI_STATUS_VAL)
3299 			mce->status |= MCI_STATUS_OVER;
3300 		banks[2] = mce->addr;
3301 		banks[3] = mce->misc;
3302 		banks[1] = mce->status;
3303 	} else
3304 		banks[1] |= MCI_STATUS_OVER;
3305 	return 0;
3306 }
3307 
3308 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3309 					       struct kvm_vcpu_events *events)
3310 {
3311 	process_nmi(vcpu);
3312 	/*
3313 	 * FIXME: pass injected and pending separately.  This is only
3314 	 * needed for nested virtualization, whose state cannot be
3315 	 * migrated yet.  For now we can combine them.
3316 	 */
3317 	events->exception.injected =
3318 		(vcpu->arch.exception.pending ||
3319 		 vcpu->arch.exception.injected) &&
3320 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3321 	events->exception.nr = vcpu->arch.exception.nr;
3322 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3323 	events->exception.pad = 0;
3324 	events->exception.error_code = vcpu->arch.exception.error_code;
3325 
3326 	events->interrupt.injected =
3327 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3328 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3329 	events->interrupt.soft = 0;
3330 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3331 
3332 	events->nmi.injected = vcpu->arch.nmi_injected;
3333 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3334 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3335 	events->nmi.pad = 0;
3336 
3337 	events->sipi_vector = 0; /* never valid when reporting to user space */
3338 
3339 	events->smi.smm = is_smm(vcpu);
3340 	events->smi.pending = vcpu->arch.smi_pending;
3341 	events->smi.smm_inside_nmi =
3342 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3343 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3344 
3345 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3346 			 | KVM_VCPUEVENT_VALID_SHADOW
3347 			 | KVM_VCPUEVENT_VALID_SMM);
3348 	memset(&events->reserved, 0, sizeof(events->reserved));
3349 }
3350 
3351 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3352 
3353 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3354 					      struct kvm_vcpu_events *events)
3355 {
3356 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3357 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3358 			      | KVM_VCPUEVENT_VALID_SHADOW
3359 			      | KVM_VCPUEVENT_VALID_SMM))
3360 		return -EINVAL;
3361 
3362 	if (events->exception.injected &&
3363 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3364 	     is_guest_mode(vcpu)))
3365 		return -EINVAL;
3366 
3367 	/* INITs are latched while in SMM */
3368 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3369 	    (events->smi.smm || events->smi.pending) &&
3370 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3371 		return -EINVAL;
3372 
3373 	process_nmi(vcpu);
3374 	vcpu->arch.exception.injected = false;
3375 	vcpu->arch.exception.pending = events->exception.injected;
3376 	vcpu->arch.exception.nr = events->exception.nr;
3377 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3378 	vcpu->arch.exception.error_code = events->exception.error_code;
3379 
3380 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3381 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3382 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3383 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3384 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3385 						  events->interrupt.shadow);
3386 
3387 	vcpu->arch.nmi_injected = events->nmi.injected;
3388 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3389 		vcpu->arch.nmi_pending = events->nmi.pending;
3390 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3391 
3392 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3393 	    lapic_in_kernel(vcpu))
3394 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3395 
3396 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3397 		u32 hflags = vcpu->arch.hflags;
3398 		if (events->smi.smm)
3399 			hflags |= HF_SMM_MASK;
3400 		else
3401 			hflags &= ~HF_SMM_MASK;
3402 		kvm_set_hflags(vcpu, hflags);
3403 
3404 		vcpu->arch.smi_pending = events->smi.pending;
3405 
3406 		if (events->smi.smm) {
3407 			if (events->smi.smm_inside_nmi)
3408 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3409 			else
3410 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3411 			if (lapic_in_kernel(vcpu)) {
3412 				if (events->smi.latched_init)
3413 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3414 				else
3415 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3416 			}
3417 		}
3418 	}
3419 
3420 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3421 
3422 	return 0;
3423 }
3424 
3425 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3426 					     struct kvm_debugregs *dbgregs)
3427 {
3428 	unsigned long val;
3429 
3430 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3431 	kvm_get_dr(vcpu, 6, &val);
3432 	dbgregs->dr6 = val;
3433 	dbgregs->dr7 = vcpu->arch.dr7;
3434 	dbgregs->flags = 0;
3435 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3436 }
3437 
3438 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3439 					    struct kvm_debugregs *dbgregs)
3440 {
3441 	if (dbgregs->flags)
3442 		return -EINVAL;
3443 
3444 	if (dbgregs->dr6 & ~0xffffffffull)
3445 		return -EINVAL;
3446 	if (dbgregs->dr7 & ~0xffffffffull)
3447 		return -EINVAL;
3448 
3449 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3450 	kvm_update_dr0123(vcpu);
3451 	vcpu->arch.dr6 = dbgregs->dr6;
3452 	kvm_update_dr6(vcpu);
3453 	vcpu->arch.dr7 = dbgregs->dr7;
3454 	kvm_update_dr7(vcpu);
3455 
3456 	return 0;
3457 }
3458 
3459 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3460 
3461 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3462 {
3463 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3464 	u64 xstate_bv = xsave->header.xfeatures;
3465 	u64 valid;
3466 
3467 	/*
3468 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3469 	 * leaves 0 and 1 in the loop below.
3470 	 */
3471 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3472 
3473 	/* Set XSTATE_BV */
3474 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3475 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3476 
3477 	/*
3478 	 * Copy each region from the possibly compacted offset to the
3479 	 * non-compacted offset.
3480 	 */
3481 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3482 	while (valid) {
3483 		u64 feature = valid & -valid;
3484 		int index = fls64(feature) - 1;
3485 		void *src = get_xsave_addr(xsave, feature);
3486 
3487 		if (src) {
3488 			u32 size, offset, ecx, edx;
3489 			cpuid_count(XSTATE_CPUID, index,
3490 				    &size, &offset, &ecx, &edx);
3491 			if (feature == XFEATURE_MASK_PKRU)
3492 				memcpy(dest + offset, &vcpu->arch.pkru,
3493 				       sizeof(vcpu->arch.pkru));
3494 			else
3495 				memcpy(dest + offset, src, size);
3496 
3497 		}
3498 
3499 		valid -= feature;
3500 	}
3501 }
3502 
3503 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3504 {
3505 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3506 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3507 	u64 valid;
3508 
3509 	/*
3510 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3511 	 * leaves 0 and 1 in the loop below.
3512 	 */
3513 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3514 
3515 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3516 	xsave->header.xfeatures = xstate_bv;
3517 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3518 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3519 
3520 	/*
3521 	 * Copy each region from the non-compacted offset to the
3522 	 * possibly compacted offset.
3523 	 */
3524 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3525 	while (valid) {
3526 		u64 feature = valid & -valid;
3527 		int index = fls64(feature) - 1;
3528 		void *dest = get_xsave_addr(xsave, feature);
3529 
3530 		if (dest) {
3531 			u32 size, offset, ecx, edx;
3532 			cpuid_count(XSTATE_CPUID, index,
3533 				    &size, &offset, &ecx, &edx);
3534 			if (feature == XFEATURE_MASK_PKRU)
3535 				memcpy(&vcpu->arch.pkru, src + offset,
3536 				       sizeof(vcpu->arch.pkru));
3537 			else
3538 				memcpy(dest, src + offset, size);
3539 		}
3540 
3541 		valid -= feature;
3542 	}
3543 }
3544 
3545 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3546 					 struct kvm_xsave *guest_xsave)
3547 {
3548 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3549 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3550 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3551 	} else {
3552 		memcpy(guest_xsave->region,
3553 			&vcpu->arch.guest_fpu.state.fxsave,
3554 			sizeof(struct fxregs_state));
3555 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3556 			XFEATURE_MASK_FPSSE;
3557 	}
3558 }
3559 
3560 #define XSAVE_MXCSR_OFFSET 24
3561 
3562 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3563 					struct kvm_xsave *guest_xsave)
3564 {
3565 	u64 xstate_bv =
3566 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3567 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3568 
3569 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3570 		/*
3571 		 * Here we allow setting states that are not present in
3572 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3573 		 * with old userspace.
3574 		 */
3575 		if (xstate_bv & ~kvm_supported_xcr0() ||
3576 			mxcsr & ~mxcsr_feature_mask)
3577 			return -EINVAL;
3578 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3579 	} else {
3580 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3581 			mxcsr & ~mxcsr_feature_mask)
3582 			return -EINVAL;
3583 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3584 			guest_xsave->region, sizeof(struct fxregs_state));
3585 	}
3586 	return 0;
3587 }
3588 
3589 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3590 					struct kvm_xcrs *guest_xcrs)
3591 {
3592 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3593 		guest_xcrs->nr_xcrs = 0;
3594 		return;
3595 	}
3596 
3597 	guest_xcrs->nr_xcrs = 1;
3598 	guest_xcrs->flags = 0;
3599 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3600 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3601 }
3602 
3603 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3604 				       struct kvm_xcrs *guest_xcrs)
3605 {
3606 	int i, r = 0;
3607 
3608 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3609 		return -EINVAL;
3610 
3611 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3612 		return -EINVAL;
3613 
3614 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3615 		/* Only support XCR0 currently */
3616 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3617 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3618 				guest_xcrs->xcrs[i].value);
3619 			break;
3620 		}
3621 	if (r)
3622 		r = -EINVAL;
3623 	return r;
3624 }
3625 
3626 /*
3627  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3628  * stopped by the hypervisor.  This function will be called from the host only.
3629  * EINVAL is returned when the host attempts to set the flag for a guest that
3630  * does not support pv clocks.
3631  */
3632 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3633 {
3634 	if (!vcpu->arch.pv_time_enabled)
3635 		return -EINVAL;
3636 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3637 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3638 	return 0;
3639 }
3640 
3641 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3642 				     struct kvm_enable_cap *cap)
3643 {
3644 	if (cap->flags)
3645 		return -EINVAL;
3646 
3647 	switch (cap->cap) {
3648 	case KVM_CAP_HYPERV_SYNIC2:
3649 		if (cap->args[0])
3650 			return -EINVAL;
3651 	case KVM_CAP_HYPERV_SYNIC:
3652 		if (!irqchip_in_kernel(vcpu->kvm))
3653 			return -EINVAL;
3654 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3655 					     KVM_CAP_HYPERV_SYNIC2);
3656 	default:
3657 		return -EINVAL;
3658 	}
3659 }
3660 
3661 long kvm_arch_vcpu_ioctl(struct file *filp,
3662 			 unsigned int ioctl, unsigned long arg)
3663 {
3664 	struct kvm_vcpu *vcpu = filp->private_data;
3665 	void __user *argp = (void __user *)arg;
3666 	int r;
3667 	union {
3668 		struct kvm_lapic_state *lapic;
3669 		struct kvm_xsave *xsave;
3670 		struct kvm_xcrs *xcrs;
3671 		void *buffer;
3672 	} u;
3673 
3674 	vcpu_load(vcpu);
3675 
3676 	u.buffer = NULL;
3677 	switch (ioctl) {
3678 	case KVM_GET_LAPIC: {
3679 		r = -EINVAL;
3680 		if (!lapic_in_kernel(vcpu))
3681 			goto out;
3682 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3683 
3684 		r = -ENOMEM;
3685 		if (!u.lapic)
3686 			goto out;
3687 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3688 		if (r)
3689 			goto out;
3690 		r = -EFAULT;
3691 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3692 			goto out;
3693 		r = 0;
3694 		break;
3695 	}
3696 	case KVM_SET_LAPIC: {
3697 		r = -EINVAL;
3698 		if (!lapic_in_kernel(vcpu))
3699 			goto out;
3700 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3701 		if (IS_ERR(u.lapic)) {
3702 			r = PTR_ERR(u.lapic);
3703 			goto out_nofree;
3704 		}
3705 
3706 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3707 		break;
3708 	}
3709 	case KVM_INTERRUPT: {
3710 		struct kvm_interrupt irq;
3711 
3712 		r = -EFAULT;
3713 		if (copy_from_user(&irq, argp, sizeof irq))
3714 			goto out;
3715 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3716 		break;
3717 	}
3718 	case KVM_NMI: {
3719 		r = kvm_vcpu_ioctl_nmi(vcpu);
3720 		break;
3721 	}
3722 	case KVM_SMI: {
3723 		r = kvm_vcpu_ioctl_smi(vcpu);
3724 		break;
3725 	}
3726 	case KVM_SET_CPUID: {
3727 		struct kvm_cpuid __user *cpuid_arg = argp;
3728 		struct kvm_cpuid cpuid;
3729 
3730 		r = -EFAULT;
3731 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3732 			goto out;
3733 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3734 		break;
3735 	}
3736 	case KVM_SET_CPUID2: {
3737 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3738 		struct kvm_cpuid2 cpuid;
3739 
3740 		r = -EFAULT;
3741 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3742 			goto out;
3743 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3744 					      cpuid_arg->entries);
3745 		break;
3746 	}
3747 	case KVM_GET_CPUID2: {
3748 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3749 		struct kvm_cpuid2 cpuid;
3750 
3751 		r = -EFAULT;
3752 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3753 			goto out;
3754 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3755 					      cpuid_arg->entries);
3756 		if (r)
3757 			goto out;
3758 		r = -EFAULT;
3759 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3760 			goto out;
3761 		r = 0;
3762 		break;
3763 	}
3764 	case KVM_GET_MSRS: {
3765 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3766 		r = msr_io(vcpu, argp, do_get_msr, 1);
3767 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3768 		break;
3769 	}
3770 	case KVM_SET_MSRS: {
3771 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3772 		r = msr_io(vcpu, argp, do_set_msr, 0);
3773 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3774 		break;
3775 	}
3776 	case KVM_TPR_ACCESS_REPORTING: {
3777 		struct kvm_tpr_access_ctl tac;
3778 
3779 		r = -EFAULT;
3780 		if (copy_from_user(&tac, argp, sizeof tac))
3781 			goto out;
3782 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3783 		if (r)
3784 			goto out;
3785 		r = -EFAULT;
3786 		if (copy_to_user(argp, &tac, sizeof tac))
3787 			goto out;
3788 		r = 0;
3789 		break;
3790 	};
3791 	case KVM_SET_VAPIC_ADDR: {
3792 		struct kvm_vapic_addr va;
3793 		int idx;
3794 
3795 		r = -EINVAL;
3796 		if (!lapic_in_kernel(vcpu))
3797 			goto out;
3798 		r = -EFAULT;
3799 		if (copy_from_user(&va, argp, sizeof va))
3800 			goto out;
3801 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3802 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3803 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3804 		break;
3805 	}
3806 	case KVM_X86_SETUP_MCE: {
3807 		u64 mcg_cap;
3808 
3809 		r = -EFAULT;
3810 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3811 			goto out;
3812 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3813 		break;
3814 	}
3815 	case KVM_X86_SET_MCE: {
3816 		struct kvm_x86_mce mce;
3817 
3818 		r = -EFAULT;
3819 		if (copy_from_user(&mce, argp, sizeof mce))
3820 			goto out;
3821 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3822 		break;
3823 	}
3824 	case KVM_GET_VCPU_EVENTS: {
3825 		struct kvm_vcpu_events events;
3826 
3827 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3828 
3829 		r = -EFAULT;
3830 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3831 			break;
3832 		r = 0;
3833 		break;
3834 	}
3835 	case KVM_SET_VCPU_EVENTS: {
3836 		struct kvm_vcpu_events events;
3837 
3838 		r = -EFAULT;
3839 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3840 			break;
3841 
3842 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3843 		break;
3844 	}
3845 	case KVM_GET_DEBUGREGS: {
3846 		struct kvm_debugregs dbgregs;
3847 
3848 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3849 
3850 		r = -EFAULT;
3851 		if (copy_to_user(argp, &dbgregs,
3852 				 sizeof(struct kvm_debugregs)))
3853 			break;
3854 		r = 0;
3855 		break;
3856 	}
3857 	case KVM_SET_DEBUGREGS: {
3858 		struct kvm_debugregs dbgregs;
3859 
3860 		r = -EFAULT;
3861 		if (copy_from_user(&dbgregs, argp,
3862 				   sizeof(struct kvm_debugregs)))
3863 			break;
3864 
3865 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3866 		break;
3867 	}
3868 	case KVM_GET_XSAVE: {
3869 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3870 		r = -ENOMEM;
3871 		if (!u.xsave)
3872 			break;
3873 
3874 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3875 
3876 		r = -EFAULT;
3877 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3878 			break;
3879 		r = 0;
3880 		break;
3881 	}
3882 	case KVM_SET_XSAVE: {
3883 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3884 		if (IS_ERR(u.xsave)) {
3885 			r = PTR_ERR(u.xsave);
3886 			goto out_nofree;
3887 		}
3888 
3889 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3890 		break;
3891 	}
3892 	case KVM_GET_XCRS: {
3893 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3894 		r = -ENOMEM;
3895 		if (!u.xcrs)
3896 			break;
3897 
3898 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3899 
3900 		r = -EFAULT;
3901 		if (copy_to_user(argp, u.xcrs,
3902 				 sizeof(struct kvm_xcrs)))
3903 			break;
3904 		r = 0;
3905 		break;
3906 	}
3907 	case KVM_SET_XCRS: {
3908 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3909 		if (IS_ERR(u.xcrs)) {
3910 			r = PTR_ERR(u.xcrs);
3911 			goto out_nofree;
3912 		}
3913 
3914 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3915 		break;
3916 	}
3917 	case KVM_SET_TSC_KHZ: {
3918 		u32 user_tsc_khz;
3919 
3920 		r = -EINVAL;
3921 		user_tsc_khz = (u32)arg;
3922 
3923 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3924 			goto out;
3925 
3926 		if (user_tsc_khz == 0)
3927 			user_tsc_khz = tsc_khz;
3928 
3929 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3930 			r = 0;
3931 
3932 		goto out;
3933 	}
3934 	case KVM_GET_TSC_KHZ: {
3935 		r = vcpu->arch.virtual_tsc_khz;
3936 		goto out;
3937 	}
3938 	case KVM_KVMCLOCK_CTRL: {
3939 		r = kvm_set_guest_paused(vcpu);
3940 		goto out;
3941 	}
3942 	case KVM_ENABLE_CAP: {
3943 		struct kvm_enable_cap cap;
3944 
3945 		r = -EFAULT;
3946 		if (copy_from_user(&cap, argp, sizeof(cap)))
3947 			goto out;
3948 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3949 		break;
3950 	}
3951 	default:
3952 		r = -EINVAL;
3953 	}
3954 out:
3955 	kfree(u.buffer);
3956 out_nofree:
3957 	vcpu_put(vcpu);
3958 	return r;
3959 }
3960 
3961 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3962 {
3963 	return VM_FAULT_SIGBUS;
3964 }
3965 
3966 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3967 {
3968 	int ret;
3969 
3970 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3971 		return -EINVAL;
3972 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3973 	return ret;
3974 }
3975 
3976 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3977 					      u64 ident_addr)
3978 {
3979 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
3980 }
3981 
3982 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3983 					  u32 kvm_nr_mmu_pages)
3984 {
3985 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3986 		return -EINVAL;
3987 
3988 	mutex_lock(&kvm->slots_lock);
3989 
3990 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3991 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3992 
3993 	mutex_unlock(&kvm->slots_lock);
3994 	return 0;
3995 }
3996 
3997 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3998 {
3999 	return kvm->arch.n_max_mmu_pages;
4000 }
4001 
4002 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4003 {
4004 	struct kvm_pic *pic = kvm->arch.vpic;
4005 	int r;
4006 
4007 	r = 0;
4008 	switch (chip->chip_id) {
4009 	case KVM_IRQCHIP_PIC_MASTER:
4010 		memcpy(&chip->chip.pic, &pic->pics[0],
4011 			sizeof(struct kvm_pic_state));
4012 		break;
4013 	case KVM_IRQCHIP_PIC_SLAVE:
4014 		memcpy(&chip->chip.pic, &pic->pics[1],
4015 			sizeof(struct kvm_pic_state));
4016 		break;
4017 	case KVM_IRQCHIP_IOAPIC:
4018 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4019 		break;
4020 	default:
4021 		r = -EINVAL;
4022 		break;
4023 	}
4024 	return r;
4025 }
4026 
4027 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4028 {
4029 	struct kvm_pic *pic = kvm->arch.vpic;
4030 	int r;
4031 
4032 	r = 0;
4033 	switch (chip->chip_id) {
4034 	case KVM_IRQCHIP_PIC_MASTER:
4035 		spin_lock(&pic->lock);
4036 		memcpy(&pic->pics[0], &chip->chip.pic,
4037 			sizeof(struct kvm_pic_state));
4038 		spin_unlock(&pic->lock);
4039 		break;
4040 	case KVM_IRQCHIP_PIC_SLAVE:
4041 		spin_lock(&pic->lock);
4042 		memcpy(&pic->pics[1], &chip->chip.pic,
4043 			sizeof(struct kvm_pic_state));
4044 		spin_unlock(&pic->lock);
4045 		break;
4046 	case KVM_IRQCHIP_IOAPIC:
4047 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4048 		break;
4049 	default:
4050 		r = -EINVAL;
4051 		break;
4052 	}
4053 	kvm_pic_update_irq(pic);
4054 	return r;
4055 }
4056 
4057 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4058 {
4059 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4060 
4061 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4062 
4063 	mutex_lock(&kps->lock);
4064 	memcpy(ps, &kps->channels, sizeof(*ps));
4065 	mutex_unlock(&kps->lock);
4066 	return 0;
4067 }
4068 
4069 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4070 {
4071 	int i;
4072 	struct kvm_pit *pit = kvm->arch.vpit;
4073 
4074 	mutex_lock(&pit->pit_state.lock);
4075 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4076 	for (i = 0; i < 3; i++)
4077 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4078 	mutex_unlock(&pit->pit_state.lock);
4079 	return 0;
4080 }
4081 
4082 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4083 {
4084 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4085 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4086 		sizeof(ps->channels));
4087 	ps->flags = kvm->arch.vpit->pit_state.flags;
4088 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4089 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4090 	return 0;
4091 }
4092 
4093 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4094 {
4095 	int start = 0;
4096 	int i;
4097 	u32 prev_legacy, cur_legacy;
4098 	struct kvm_pit *pit = kvm->arch.vpit;
4099 
4100 	mutex_lock(&pit->pit_state.lock);
4101 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4102 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4103 	if (!prev_legacy && cur_legacy)
4104 		start = 1;
4105 	memcpy(&pit->pit_state.channels, &ps->channels,
4106 	       sizeof(pit->pit_state.channels));
4107 	pit->pit_state.flags = ps->flags;
4108 	for (i = 0; i < 3; i++)
4109 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4110 				   start && i == 0);
4111 	mutex_unlock(&pit->pit_state.lock);
4112 	return 0;
4113 }
4114 
4115 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4116 				 struct kvm_reinject_control *control)
4117 {
4118 	struct kvm_pit *pit = kvm->arch.vpit;
4119 
4120 	if (!pit)
4121 		return -ENXIO;
4122 
4123 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4124 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4125 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4126 	 */
4127 	mutex_lock(&pit->pit_state.lock);
4128 	kvm_pit_set_reinject(pit, control->pit_reinject);
4129 	mutex_unlock(&pit->pit_state.lock);
4130 
4131 	return 0;
4132 }
4133 
4134 /**
4135  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4136  * @kvm: kvm instance
4137  * @log: slot id and address to which we copy the log
4138  *
4139  * Steps 1-4 below provide general overview of dirty page logging. See
4140  * kvm_get_dirty_log_protect() function description for additional details.
4141  *
4142  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4143  * always flush the TLB (step 4) even if previous step failed  and the dirty
4144  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4145  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4146  * writes will be marked dirty for next log read.
4147  *
4148  *   1. Take a snapshot of the bit and clear it if needed.
4149  *   2. Write protect the corresponding page.
4150  *   3. Copy the snapshot to the userspace.
4151  *   4. Flush TLB's if needed.
4152  */
4153 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4154 {
4155 	bool is_dirty = false;
4156 	int r;
4157 
4158 	mutex_lock(&kvm->slots_lock);
4159 
4160 	/*
4161 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4162 	 */
4163 	if (kvm_x86_ops->flush_log_dirty)
4164 		kvm_x86_ops->flush_log_dirty(kvm);
4165 
4166 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4167 
4168 	/*
4169 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4170 	 * kvm_mmu_slot_remove_write_access().
4171 	 */
4172 	lockdep_assert_held(&kvm->slots_lock);
4173 	if (is_dirty)
4174 		kvm_flush_remote_tlbs(kvm);
4175 
4176 	mutex_unlock(&kvm->slots_lock);
4177 	return r;
4178 }
4179 
4180 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4181 			bool line_status)
4182 {
4183 	if (!irqchip_in_kernel(kvm))
4184 		return -ENXIO;
4185 
4186 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4187 					irq_event->irq, irq_event->level,
4188 					line_status);
4189 	return 0;
4190 }
4191 
4192 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4193 				   struct kvm_enable_cap *cap)
4194 {
4195 	int r;
4196 
4197 	if (cap->flags)
4198 		return -EINVAL;
4199 
4200 	switch (cap->cap) {
4201 	case KVM_CAP_DISABLE_QUIRKS:
4202 		kvm->arch.disabled_quirks = cap->args[0];
4203 		r = 0;
4204 		break;
4205 	case KVM_CAP_SPLIT_IRQCHIP: {
4206 		mutex_lock(&kvm->lock);
4207 		r = -EINVAL;
4208 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4209 			goto split_irqchip_unlock;
4210 		r = -EEXIST;
4211 		if (irqchip_in_kernel(kvm))
4212 			goto split_irqchip_unlock;
4213 		if (kvm->created_vcpus)
4214 			goto split_irqchip_unlock;
4215 		r = kvm_setup_empty_irq_routing(kvm);
4216 		if (r)
4217 			goto split_irqchip_unlock;
4218 		/* Pairs with irqchip_in_kernel. */
4219 		smp_wmb();
4220 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4221 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4222 		r = 0;
4223 split_irqchip_unlock:
4224 		mutex_unlock(&kvm->lock);
4225 		break;
4226 	}
4227 	case KVM_CAP_X2APIC_API:
4228 		r = -EINVAL;
4229 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4230 			break;
4231 
4232 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4233 			kvm->arch.x2apic_format = true;
4234 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4235 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4236 
4237 		r = 0;
4238 		break;
4239 	case KVM_CAP_X86_DISABLE_EXITS:
4240 		r = -EINVAL;
4241 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4242 			break;
4243 
4244 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4245 			kvm_can_mwait_in_guest())
4246 			kvm->arch.mwait_in_guest = true;
4247 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
4248 			kvm->arch.hlt_in_guest = true;
4249 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4250 			kvm->arch.pause_in_guest = true;
4251 		r = 0;
4252 		break;
4253 	default:
4254 		r = -EINVAL;
4255 		break;
4256 	}
4257 	return r;
4258 }
4259 
4260 long kvm_arch_vm_ioctl(struct file *filp,
4261 		       unsigned int ioctl, unsigned long arg)
4262 {
4263 	struct kvm *kvm = filp->private_data;
4264 	void __user *argp = (void __user *)arg;
4265 	int r = -ENOTTY;
4266 	/*
4267 	 * This union makes it completely explicit to gcc-3.x
4268 	 * that these two variables' stack usage should be
4269 	 * combined, not added together.
4270 	 */
4271 	union {
4272 		struct kvm_pit_state ps;
4273 		struct kvm_pit_state2 ps2;
4274 		struct kvm_pit_config pit_config;
4275 	} u;
4276 
4277 	switch (ioctl) {
4278 	case KVM_SET_TSS_ADDR:
4279 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4280 		break;
4281 	case KVM_SET_IDENTITY_MAP_ADDR: {
4282 		u64 ident_addr;
4283 
4284 		mutex_lock(&kvm->lock);
4285 		r = -EINVAL;
4286 		if (kvm->created_vcpus)
4287 			goto set_identity_unlock;
4288 		r = -EFAULT;
4289 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4290 			goto set_identity_unlock;
4291 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4292 set_identity_unlock:
4293 		mutex_unlock(&kvm->lock);
4294 		break;
4295 	}
4296 	case KVM_SET_NR_MMU_PAGES:
4297 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4298 		break;
4299 	case KVM_GET_NR_MMU_PAGES:
4300 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4301 		break;
4302 	case KVM_CREATE_IRQCHIP: {
4303 		mutex_lock(&kvm->lock);
4304 
4305 		r = -EEXIST;
4306 		if (irqchip_in_kernel(kvm))
4307 			goto create_irqchip_unlock;
4308 
4309 		r = -EINVAL;
4310 		if (kvm->created_vcpus)
4311 			goto create_irqchip_unlock;
4312 
4313 		r = kvm_pic_init(kvm);
4314 		if (r)
4315 			goto create_irqchip_unlock;
4316 
4317 		r = kvm_ioapic_init(kvm);
4318 		if (r) {
4319 			kvm_pic_destroy(kvm);
4320 			goto create_irqchip_unlock;
4321 		}
4322 
4323 		r = kvm_setup_default_irq_routing(kvm);
4324 		if (r) {
4325 			kvm_ioapic_destroy(kvm);
4326 			kvm_pic_destroy(kvm);
4327 			goto create_irqchip_unlock;
4328 		}
4329 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4330 		smp_wmb();
4331 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4332 	create_irqchip_unlock:
4333 		mutex_unlock(&kvm->lock);
4334 		break;
4335 	}
4336 	case KVM_CREATE_PIT:
4337 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4338 		goto create_pit;
4339 	case KVM_CREATE_PIT2:
4340 		r = -EFAULT;
4341 		if (copy_from_user(&u.pit_config, argp,
4342 				   sizeof(struct kvm_pit_config)))
4343 			goto out;
4344 	create_pit:
4345 		mutex_lock(&kvm->lock);
4346 		r = -EEXIST;
4347 		if (kvm->arch.vpit)
4348 			goto create_pit_unlock;
4349 		r = -ENOMEM;
4350 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4351 		if (kvm->arch.vpit)
4352 			r = 0;
4353 	create_pit_unlock:
4354 		mutex_unlock(&kvm->lock);
4355 		break;
4356 	case KVM_GET_IRQCHIP: {
4357 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4358 		struct kvm_irqchip *chip;
4359 
4360 		chip = memdup_user(argp, sizeof(*chip));
4361 		if (IS_ERR(chip)) {
4362 			r = PTR_ERR(chip);
4363 			goto out;
4364 		}
4365 
4366 		r = -ENXIO;
4367 		if (!irqchip_kernel(kvm))
4368 			goto get_irqchip_out;
4369 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4370 		if (r)
4371 			goto get_irqchip_out;
4372 		r = -EFAULT;
4373 		if (copy_to_user(argp, chip, sizeof *chip))
4374 			goto get_irqchip_out;
4375 		r = 0;
4376 	get_irqchip_out:
4377 		kfree(chip);
4378 		break;
4379 	}
4380 	case KVM_SET_IRQCHIP: {
4381 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4382 		struct kvm_irqchip *chip;
4383 
4384 		chip = memdup_user(argp, sizeof(*chip));
4385 		if (IS_ERR(chip)) {
4386 			r = PTR_ERR(chip);
4387 			goto out;
4388 		}
4389 
4390 		r = -ENXIO;
4391 		if (!irqchip_kernel(kvm))
4392 			goto set_irqchip_out;
4393 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4394 		if (r)
4395 			goto set_irqchip_out;
4396 		r = 0;
4397 	set_irqchip_out:
4398 		kfree(chip);
4399 		break;
4400 	}
4401 	case KVM_GET_PIT: {
4402 		r = -EFAULT;
4403 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4404 			goto out;
4405 		r = -ENXIO;
4406 		if (!kvm->arch.vpit)
4407 			goto out;
4408 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4409 		if (r)
4410 			goto out;
4411 		r = -EFAULT;
4412 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4413 			goto out;
4414 		r = 0;
4415 		break;
4416 	}
4417 	case KVM_SET_PIT: {
4418 		r = -EFAULT;
4419 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4420 			goto out;
4421 		r = -ENXIO;
4422 		if (!kvm->arch.vpit)
4423 			goto out;
4424 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4425 		break;
4426 	}
4427 	case KVM_GET_PIT2: {
4428 		r = -ENXIO;
4429 		if (!kvm->arch.vpit)
4430 			goto out;
4431 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4432 		if (r)
4433 			goto out;
4434 		r = -EFAULT;
4435 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4436 			goto out;
4437 		r = 0;
4438 		break;
4439 	}
4440 	case KVM_SET_PIT2: {
4441 		r = -EFAULT;
4442 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4443 			goto out;
4444 		r = -ENXIO;
4445 		if (!kvm->arch.vpit)
4446 			goto out;
4447 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4448 		break;
4449 	}
4450 	case KVM_REINJECT_CONTROL: {
4451 		struct kvm_reinject_control control;
4452 		r =  -EFAULT;
4453 		if (copy_from_user(&control, argp, sizeof(control)))
4454 			goto out;
4455 		r = kvm_vm_ioctl_reinject(kvm, &control);
4456 		break;
4457 	}
4458 	case KVM_SET_BOOT_CPU_ID:
4459 		r = 0;
4460 		mutex_lock(&kvm->lock);
4461 		if (kvm->created_vcpus)
4462 			r = -EBUSY;
4463 		else
4464 			kvm->arch.bsp_vcpu_id = arg;
4465 		mutex_unlock(&kvm->lock);
4466 		break;
4467 	case KVM_XEN_HVM_CONFIG: {
4468 		struct kvm_xen_hvm_config xhc;
4469 		r = -EFAULT;
4470 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4471 			goto out;
4472 		r = -EINVAL;
4473 		if (xhc.flags)
4474 			goto out;
4475 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4476 		r = 0;
4477 		break;
4478 	}
4479 	case KVM_SET_CLOCK: {
4480 		struct kvm_clock_data user_ns;
4481 		u64 now_ns;
4482 
4483 		r = -EFAULT;
4484 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4485 			goto out;
4486 
4487 		r = -EINVAL;
4488 		if (user_ns.flags)
4489 			goto out;
4490 
4491 		r = 0;
4492 		/*
4493 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4494 		 * kvm_gen_update_masterclock() can be cut down to locked
4495 		 * pvclock_update_vm_gtod_copy().
4496 		 */
4497 		kvm_gen_update_masterclock(kvm);
4498 		now_ns = get_kvmclock_ns(kvm);
4499 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4500 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4501 		break;
4502 	}
4503 	case KVM_GET_CLOCK: {
4504 		struct kvm_clock_data user_ns;
4505 		u64 now_ns;
4506 
4507 		now_ns = get_kvmclock_ns(kvm);
4508 		user_ns.clock = now_ns;
4509 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4510 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4511 
4512 		r = -EFAULT;
4513 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4514 			goto out;
4515 		r = 0;
4516 		break;
4517 	}
4518 	case KVM_ENABLE_CAP: {
4519 		struct kvm_enable_cap cap;
4520 
4521 		r = -EFAULT;
4522 		if (copy_from_user(&cap, argp, sizeof(cap)))
4523 			goto out;
4524 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4525 		break;
4526 	}
4527 	case KVM_MEMORY_ENCRYPT_OP: {
4528 		r = -ENOTTY;
4529 		if (kvm_x86_ops->mem_enc_op)
4530 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4531 		break;
4532 	}
4533 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4534 		struct kvm_enc_region region;
4535 
4536 		r = -EFAULT;
4537 		if (copy_from_user(&region, argp, sizeof(region)))
4538 			goto out;
4539 
4540 		r = -ENOTTY;
4541 		if (kvm_x86_ops->mem_enc_reg_region)
4542 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4543 		break;
4544 	}
4545 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4546 		struct kvm_enc_region region;
4547 
4548 		r = -EFAULT;
4549 		if (copy_from_user(&region, argp, sizeof(region)))
4550 			goto out;
4551 
4552 		r = -ENOTTY;
4553 		if (kvm_x86_ops->mem_enc_unreg_region)
4554 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4555 		break;
4556 	}
4557 	case KVM_HYPERV_EVENTFD: {
4558 		struct kvm_hyperv_eventfd hvevfd;
4559 
4560 		r = -EFAULT;
4561 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4562 			goto out;
4563 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4564 		break;
4565 	}
4566 	default:
4567 		r = -ENOTTY;
4568 	}
4569 out:
4570 	return r;
4571 }
4572 
4573 static void kvm_init_msr_list(void)
4574 {
4575 	u32 dummy[2];
4576 	unsigned i, j;
4577 
4578 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4579 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4580 			continue;
4581 
4582 		/*
4583 		 * Even MSRs that are valid in the host may not be exposed
4584 		 * to the guests in some cases.
4585 		 */
4586 		switch (msrs_to_save[i]) {
4587 		case MSR_IA32_BNDCFGS:
4588 			if (!kvm_x86_ops->mpx_supported())
4589 				continue;
4590 			break;
4591 		case MSR_TSC_AUX:
4592 			if (!kvm_x86_ops->rdtscp_supported())
4593 				continue;
4594 			break;
4595 		default:
4596 			break;
4597 		}
4598 
4599 		if (j < i)
4600 			msrs_to_save[j] = msrs_to_save[i];
4601 		j++;
4602 	}
4603 	num_msrs_to_save = j;
4604 
4605 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4606 		switch (emulated_msrs[i]) {
4607 		case MSR_IA32_SMBASE:
4608 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4609 				continue;
4610 			break;
4611 		default:
4612 			break;
4613 		}
4614 
4615 		if (j < i)
4616 			emulated_msrs[j] = emulated_msrs[i];
4617 		j++;
4618 	}
4619 	num_emulated_msrs = j;
4620 
4621 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4622 		struct kvm_msr_entry msr;
4623 
4624 		msr.index = msr_based_features[i];
4625 		if (kvm_get_msr_feature(&msr))
4626 			continue;
4627 
4628 		if (j < i)
4629 			msr_based_features[j] = msr_based_features[i];
4630 		j++;
4631 	}
4632 	num_msr_based_features = j;
4633 }
4634 
4635 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4636 			   const void *v)
4637 {
4638 	int handled = 0;
4639 	int n;
4640 
4641 	do {
4642 		n = min(len, 8);
4643 		if (!(lapic_in_kernel(vcpu) &&
4644 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4645 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4646 			break;
4647 		handled += n;
4648 		addr += n;
4649 		len -= n;
4650 		v += n;
4651 	} while (len);
4652 
4653 	return handled;
4654 }
4655 
4656 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4657 {
4658 	int handled = 0;
4659 	int n;
4660 
4661 	do {
4662 		n = min(len, 8);
4663 		if (!(lapic_in_kernel(vcpu) &&
4664 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4665 					 addr, n, v))
4666 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4667 			break;
4668 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4669 		handled += n;
4670 		addr += n;
4671 		len -= n;
4672 		v += n;
4673 	} while (len);
4674 
4675 	return handled;
4676 }
4677 
4678 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4679 			struct kvm_segment *var, int seg)
4680 {
4681 	kvm_x86_ops->set_segment(vcpu, var, seg);
4682 }
4683 
4684 void kvm_get_segment(struct kvm_vcpu *vcpu,
4685 		     struct kvm_segment *var, int seg)
4686 {
4687 	kvm_x86_ops->get_segment(vcpu, var, seg);
4688 }
4689 
4690 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4691 			   struct x86_exception *exception)
4692 {
4693 	gpa_t t_gpa;
4694 
4695 	BUG_ON(!mmu_is_nested(vcpu));
4696 
4697 	/* NPT walks are always user-walks */
4698 	access |= PFERR_USER_MASK;
4699 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4700 
4701 	return t_gpa;
4702 }
4703 
4704 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4705 			      struct x86_exception *exception)
4706 {
4707 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4708 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4709 }
4710 
4711  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4712 				struct x86_exception *exception)
4713 {
4714 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4715 	access |= PFERR_FETCH_MASK;
4716 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4717 }
4718 
4719 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4720 			       struct x86_exception *exception)
4721 {
4722 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4723 	access |= PFERR_WRITE_MASK;
4724 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4725 }
4726 
4727 /* uses this to access any guest's mapped memory without checking CPL */
4728 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4729 				struct x86_exception *exception)
4730 {
4731 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4732 }
4733 
4734 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4735 				      struct kvm_vcpu *vcpu, u32 access,
4736 				      struct x86_exception *exception)
4737 {
4738 	void *data = val;
4739 	int r = X86EMUL_CONTINUE;
4740 
4741 	while (bytes) {
4742 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4743 							    exception);
4744 		unsigned offset = addr & (PAGE_SIZE-1);
4745 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4746 		int ret;
4747 
4748 		if (gpa == UNMAPPED_GVA)
4749 			return X86EMUL_PROPAGATE_FAULT;
4750 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4751 					       offset, toread);
4752 		if (ret < 0) {
4753 			r = X86EMUL_IO_NEEDED;
4754 			goto out;
4755 		}
4756 
4757 		bytes -= toread;
4758 		data += toread;
4759 		addr += toread;
4760 	}
4761 out:
4762 	return r;
4763 }
4764 
4765 /* used for instruction fetching */
4766 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4767 				gva_t addr, void *val, unsigned int bytes,
4768 				struct x86_exception *exception)
4769 {
4770 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4771 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4772 	unsigned offset;
4773 	int ret;
4774 
4775 	/* Inline kvm_read_guest_virt_helper for speed.  */
4776 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4777 						    exception);
4778 	if (unlikely(gpa == UNMAPPED_GVA))
4779 		return X86EMUL_PROPAGATE_FAULT;
4780 
4781 	offset = addr & (PAGE_SIZE-1);
4782 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4783 		bytes = (unsigned)PAGE_SIZE - offset;
4784 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4785 				       offset, bytes);
4786 	if (unlikely(ret < 0))
4787 		return X86EMUL_IO_NEEDED;
4788 
4789 	return X86EMUL_CONTINUE;
4790 }
4791 
4792 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4793 			       gva_t addr, void *val, unsigned int bytes,
4794 			       struct x86_exception *exception)
4795 {
4796 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4797 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4798 
4799 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4800 					  exception);
4801 }
4802 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4803 
4804 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4805 				      gva_t addr, void *val, unsigned int bytes,
4806 				      struct x86_exception *exception)
4807 {
4808 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4809 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4810 }
4811 
4812 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4813 		unsigned long addr, void *val, unsigned int bytes)
4814 {
4815 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4816 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4817 
4818 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4819 }
4820 
4821 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4822 				       gva_t addr, void *val,
4823 				       unsigned int bytes,
4824 				       struct x86_exception *exception)
4825 {
4826 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4827 	void *data = val;
4828 	int r = X86EMUL_CONTINUE;
4829 
4830 	while (bytes) {
4831 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4832 							     PFERR_WRITE_MASK,
4833 							     exception);
4834 		unsigned offset = addr & (PAGE_SIZE-1);
4835 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4836 		int ret;
4837 
4838 		if (gpa == UNMAPPED_GVA)
4839 			return X86EMUL_PROPAGATE_FAULT;
4840 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4841 		if (ret < 0) {
4842 			r = X86EMUL_IO_NEEDED;
4843 			goto out;
4844 		}
4845 
4846 		bytes -= towrite;
4847 		data += towrite;
4848 		addr += towrite;
4849 	}
4850 out:
4851 	return r;
4852 }
4853 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4854 
4855 int handle_ud(struct kvm_vcpu *vcpu)
4856 {
4857 	int emul_type = EMULTYPE_TRAP_UD;
4858 	enum emulation_result er;
4859 	char sig[5]; /* ud2; .ascii "kvm" */
4860 	struct x86_exception e;
4861 
4862 	if (force_emulation_prefix &&
4863 	    kvm_read_guest_virt(&vcpu->arch.emulate_ctxt,
4864 				kvm_get_linear_rip(vcpu), sig, sizeof(sig), &e) == 0 &&
4865 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
4866 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
4867 		emul_type = 0;
4868 	}
4869 
4870 	er = emulate_instruction(vcpu, emul_type);
4871 	if (er == EMULATE_USER_EXIT)
4872 		return 0;
4873 	if (er != EMULATE_DONE)
4874 		kvm_queue_exception(vcpu, UD_VECTOR);
4875 	return 1;
4876 }
4877 EXPORT_SYMBOL_GPL(handle_ud);
4878 
4879 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4880 			    gpa_t gpa, bool write)
4881 {
4882 	/* For APIC access vmexit */
4883 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4884 		return 1;
4885 
4886 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4887 		trace_vcpu_match_mmio(gva, gpa, write, true);
4888 		return 1;
4889 	}
4890 
4891 	return 0;
4892 }
4893 
4894 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4895 				gpa_t *gpa, struct x86_exception *exception,
4896 				bool write)
4897 {
4898 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4899 		| (write ? PFERR_WRITE_MASK : 0);
4900 
4901 	/*
4902 	 * currently PKRU is only applied to ept enabled guest so
4903 	 * there is no pkey in EPT page table for L1 guest or EPT
4904 	 * shadow page table for L2 guest.
4905 	 */
4906 	if (vcpu_match_mmio_gva(vcpu, gva)
4907 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4908 				 vcpu->arch.access, 0, access)) {
4909 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4910 					(gva & (PAGE_SIZE - 1));
4911 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4912 		return 1;
4913 	}
4914 
4915 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4916 
4917 	if (*gpa == UNMAPPED_GVA)
4918 		return -1;
4919 
4920 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4921 }
4922 
4923 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4924 			const void *val, int bytes)
4925 {
4926 	int ret;
4927 
4928 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4929 	if (ret < 0)
4930 		return 0;
4931 	kvm_page_track_write(vcpu, gpa, val, bytes);
4932 	return 1;
4933 }
4934 
4935 struct read_write_emulator_ops {
4936 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4937 				  int bytes);
4938 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4939 				  void *val, int bytes);
4940 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4941 			       int bytes, void *val);
4942 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4943 				    void *val, int bytes);
4944 	bool write;
4945 };
4946 
4947 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4948 {
4949 	if (vcpu->mmio_read_completed) {
4950 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4951 			       vcpu->mmio_fragments[0].gpa, val);
4952 		vcpu->mmio_read_completed = 0;
4953 		return 1;
4954 	}
4955 
4956 	return 0;
4957 }
4958 
4959 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4960 			void *val, int bytes)
4961 {
4962 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4963 }
4964 
4965 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4966 			 void *val, int bytes)
4967 {
4968 	return emulator_write_phys(vcpu, gpa, val, bytes);
4969 }
4970 
4971 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4972 {
4973 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
4974 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4975 }
4976 
4977 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4978 			  void *val, int bytes)
4979 {
4980 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
4981 	return X86EMUL_IO_NEEDED;
4982 }
4983 
4984 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4985 			   void *val, int bytes)
4986 {
4987 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4988 
4989 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4990 	return X86EMUL_CONTINUE;
4991 }
4992 
4993 static const struct read_write_emulator_ops read_emultor = {
4994 	.read_write_prepare = read_prepare,
4995 	.read_write_emulate = read_emulate,
4996 	.read_write_mmio = vcpu_mmio_read,
4997 	.read_write_exit_mmio = read_exit_mmio,
4998 };
4999 
5000 static const struct read_write_emulator_ops write_emultor = {
5001 	.read_write_emulate = write_emulate,
5002 	.read_write_mmio = write_mmio,
5003 	.read_write_exit_mmio = write_exit_mmio,
5004 	.write = true,
5005 };
5006 
5007 static int emulator_read_write_onepage(unsigned long addr, void *val,
5008 				       unsigned int bytes,
5009 				       struct x86_exception *exception,
5010 				       struct kvm_vcpu *vcpu,
5011 				       const struct read_write_emulator_ops *ops)
5012 {
5013 	gpa_t gpa;
5014 	int handled, ret;
5015 	bool write = ops->write;
5016 	struct kvm_mmio_fragment *frag;
5017 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5018 
5019 	/*
5020 	 * If the exit was due to a NPF we may already have a GPA.
5021 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5022 	 * Note, this cannot be used on string operations since string
5023 	 * operation using rep will only have the initial GPA from the NPF
5024 	 * occurred.
5025 	 */
5026 	if (vcpu->arch.gpa_available &&
5027 	    emulator_can_use_gpa(ctxt) &&
5028 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5029 		gpa = vcpu->arch.gpa_val;
5030 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5031 	} else {
5032 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5033 		if (ret < 0)
5034 			return X86EMUL_PROPAGATE_FAULT;
5035 	}
5036 
5037 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5038 		return X86EMUL_CONTINUE;
5039 
5040 	/*
5041 	 * Is this MMIO handled locally?
5042 	 */
5043 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5044 	if (handled == bytes)
5045 		return X86EMUL_CONTINUE;
5046 
5047 	gpa += handled;
5048 	bytes -= handled;
5049 	val += handled;
5050 
5051 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5052 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5053 	frag->gpa = gpa;
5054 	frag->data = val;
5055 	frag->len = bytes;
5056 	return X86EMUL_CONTINUE;
5057 }
5058 
5059 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5060 			unsigned long addr,
5061 			void *val, unsigned int bytes,
5062 			struct x86_exception *exception,
5063 			const struct read_write_emulator_ops *ops)
5064 {
5065 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5066 	gpa_t gpa;
5067 	int rc;
5068 
5069 	if (ops->read_write_prepare &&
5070 		  ops->read_write_prepare(vcpu, val, bytes))
5071 		return X86EMUL_CONTINUE;
5072 
5073 	vcpu->mmio_nr_fragments = 0;
5074 
5075 	/* Crossing a page boundary? */
5076 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5077 		int now;
5078 
5079 		now = -addr & ~PAGE_MASK;
5080 		rc = emulator_read_write_onepage(addr, val, now, exception,
5081 						 vcpu, ops);
5082 
5083 		if (rc != X86EMUL_CONTINUE)
5084 			return rc;
5085 		addr += now;
5086 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5087 			addr = (u32)addr;
5088 		val += now;
5089 		bytes -= now;
5090 	}
5091 
5092 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5093 					 vcpu, ops);
5094 	if (rc != X86EMUL_CONTINUE)
5095 		return rc;
5096 
5097 	if (!vcpu->mmio_nr_fragments)
5098 		return rc;
5099 
5100 	gpa = vcpu->mmio_fragments[0].gpa;
5101 
5102 	vcpu->mmio_needed = 1;
5103 	vcpu->mmio_cur_fragment = 0;
5104 
5105 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5106 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5107 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5108 	vcpu->run->mmio.phys_addr = gpa;
5109 
5110 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5111 }
5112 
5113 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5114 				  unsigned long addr,
5115 				  void *val,
5116 				  unsigned int bytes,
5117 				  struct x86_exception *exception)
5118 {
5119 	return emulator_read_write(ctxt, addr, val, bytes,
5120 				   exception, &read_emultor);
5121 }
5122 
5123 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5124 			    unsigned long addr,
5125 			    const void *val,
5126 			    unsigned int bytes,
5127 			    struct x86_exception *exception)
5128 {
5129 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5130 				   exception, &write_emultor);
5131 }
5132 
5133 #define CMPXCHG_TYPE(t, ptr, old, new) \
5134 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5135 
5136 #ifdef CONFIG_X86_64
5137 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5138 #else
5139 #  define CMPXCHG64(ptr, old, new) \
5140 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5141 #endif
5142 
5143 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5144 				     unsigned long addr,
5145 				     const void *old,
5146 				     const void *new,
5147 				     unsigned int bytes,
5148 				     struct x86_exception *exception)
5149 {
5150 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5151 	gpa_t gpa;
5152 	struct page *page;
5153 	char *kaddr;
5154 	bool exchanged;
5155 
5156 	/* guests cmpxchg8b have to be emulated atomically */
5157 	if (bytes > 8 || (bytes & (bytes - 1)))
5158 		goto emul_write;
5159 
5160 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5161 
5162 	if (gpa == UNMAPPED_GVA ||
5163 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5164 		goto emul_write;
5165 
5166 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5167 		goto emul_write;
5168 
5169 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5170 	if (is_error_page(page))
5171 		goto emul_write;
5172 
5173 	kaddr = kmap_atomic(page);
5174 	kaddr += offset_in_page(gpa);
5175 	switch (bytes) {
5176 	case 1:
5177 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5178 		break;
5179 	case 2:
5180 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5181 		break;
5182 	case 4:
5183 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5184 		break;
5185 	case 8:
5186 		exchanged = CMPXCHG64(kaddr, old, new);
5187 		break;
5188 	default:
5189 		BUG();
5190 	}
5191 	kunmap_atomic(kaddr);
5192 	kvm_release_page_dirty(page);
5193 
5194 	if (!exchanged)
5195 		return X86EMUL_CMPXCHG_FAILED;
5196 
5197 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5198 	kvm_page_track_write(vcpu, gpa, new, bytes);
5199 
5200 	return X86EMUL_CONTINUE;
5201 
5202 emul_write:
5203 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5204 
5205 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5206 }
5207 
5208 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5209 {
5210 	int r = 0, i;
5211 
5212 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5213 		if (vcpu->arch.pio.in)
5214 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5215 					    vcpu->arch.pio.size, pd);
5216 		else
5217 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5218 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5219 					     pd);
5220 		if (r)
5221 			break;
5222 		pd += vcpu->arch.pio.size;
5223 	}
5224 	return r;
5225 }
5226 
5227 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5228 			       unsigned short port, void *val,
5229 			       unsigned int count, bool in)
5230 {
5231 	vcpu->arch.pio.port = port;
5232 	vcpu->arch.pio.in = in;
5233 	vcpu->arch.pio.count  = count;
5234 	vcpu->arch.pio.size = size;
5235 
5236 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5237 		vcpu->arch.pio.count = 0;
5238 		return 1;
5239 	}
5240 
5241 	vcpu->run->exit_reason = KVM_EXIT_IO;
5242 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5243 	vcpu->run->io.size = size;
5244 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5245 	vcpu->run->io.count = count;
5246 	vcpu->run->io.port = port;
5247 
5248 	return 0;
5249 }
5250 
5251 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5252 				    int size, unsigned short port, void *val,
5253 				    unsigned int count)
5254 {
5255 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5256 	int ret;
5257 
5258 	if (vcpu->arch.pio.count)
5259 		goto data_avail;
5260 
5261 	memset(vcpu->arch.pio_data, 0, size * count);
5262 
5263 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5264 	if (ret) {
5265 data_avail:
5266 		memcpy(val, vcpu->arch.pio_data, size * count);
5267 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5268 		vcpu->arch.pio.count = 0;
5269 		return 1;
5270 	}
5271 
5272 	return 0;
5273 }
5274 
5275 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5276 				     int size, unsigned short port,
5277 				     const void *val, unsigned int count)
5278 {
5279 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5280 
5281 	memcpy(vcpu->arch.pio_data, val, size * count);
5282 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5283 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5284 }
5285 
5286 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5287 {
5288 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5289 }
5290 
5291 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5292 {
5293 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5294 }
5295 
5296 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5297 {
5298 	if (!need_emulate_wbinvd(vcpu))
5299 		return X86EMUL_CONTINUE;
5300 
5301 	if (kvm_x86_ops->has_wbinvd_exit()) {
5302 		int cpu = get_cpu();
5303 
5304 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5305 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5306 				wbinvd_ipi, NULL, 1);
5307 		put_cpu();
5308 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5309 	} else
5310 		wbinvd();
5311 	return X86EMUL_CONTINUE;
5312 }
5313 
5314 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5315 {
5316 	kvm_emulate_wbinvd_noskip(vcpu);
5317 	return kvm_skip_emulated_instruction(vcpu);
5318 }
5319 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5320 
5321 
5322 
5323 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5324 {
5325 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5326 }
5327 
5328 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5329 			   unsigned long *dest)
5330 {
5331 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5332 }
5333 
5334 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5335 			   unsigned long value)
5336 {
5337 
5338 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5339 }
5340 
5341 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5342 {
5343 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5344 }
5345 
5346 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5347 {
5348 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5349 	unsigned long value;
5350 
5351 	switch (cr) {
5352 	case 0:
5353 		value = kvm_read_cr0(vcpu);
5354 		break;
5355 	case 2:
5356 		value = vcpu->arch.cr2;
5357 		break;
5358 	case 3:
5359 		value = kvm_read_cr3(vcpu);
5360 		break;
5361 	case 4:
5362 		value = kvm_read_cr4(vcpu);
5363 		break;
5364 	case 8:
5365 		value = kvm_get_cr8(vcpu);
5366 		break;
5367 	default:
5368 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5369 		return 0;
5370 	}
5371 
5372 	return value;
5373 }
5374 
5375 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5376 {
5377 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5378 	int res = 0;
5379 
5380 	switch (cr) {
5381 	case 0:
5382 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5383 		break;
5384 	case 2:
5385 		vcpu->arch.cr2 = val;
5386 		break;
5387 	case 3:
5388 		res = kvm_set_cr3(vcpu, val);
5389 		break;
5390 	case 4:
5391 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5392 		break;
5393 	case 8:
5394 		res = kvm_set_cr8(vcpu, val);
5395 		break;
5396 	default:
5397 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5398 		res = -1;
5399 	}
5400 
5401 	return res;
5402 }
5403 
5404 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5405 {
5406 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5407 }
5408 
5409 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5410 {
5411 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5412 }
5413 
5414 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5415 {
5416 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5417 }
5418 
5419 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5420 {
5421 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5422 }
5423 
5424 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5425 {
5426 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5427 }
5428 
5429 static unsigned long emulator_get_cached_segment_base(
5430 	struct x86_emulate_ctxt *ctxt, int seg)
5431 {
5432 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5433 }
5434 
5435 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5436 				 struct desc_struct *desc, u32 *base3,
5437 				 int seg)
5438 {
5439 	struct kvm_segment var;
5440 
5441 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5442 	*selector = var.selector;
5443 
5444 	if (var.unusable) {
5445 		memset(desc, 0, sizeof(*desc));
5446 		if (base3)
5447 			*base3 = 0;
5448 		return false;
5449 	}
5450 
5451 	if (var.g)
5452 		var.limit >>= 12;
5453 	set_desc_limit(desc, var.limit);
5454 	set_desc_base(desc, (unsigned long)var.base);
5455 #ifdef CONFIG_X86_64
5456 	if (base3)
5457 		*base3 = var.base >> 32;
5458 #endif
5459 	desc->type = var.type;
5460 	desc->s = var.s;
5461 	desc->dpl = var.dpl;
5462 	desc->p = var.present;
5463 	desc->avl = var.avl;
5464 	desc->l = var.l;
5465 	desc->d = var.db;
5466 	desc->g = var.g;
5467 
5468 	return true;
5469 }
5470 
5471 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5472 				 struct desc_struct *desc, u32 base3,
5473 				 int seg)
5474 {
5475 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5476 	struct kvm_segment var;
5477 
5478 	var.selector = selector;
5479 	var.base = get_desc_base(desc);
5480 #ifdef CONFIG_X86_64
5481 	var.base |= ((u64)base3) << 32;
5482 #endif
5483 	var.limit = get_desc_limit(desc);
5484 	if (desc->g)
5485 		var.limit = (var.limit << 12) | 0xfff;
5486 	var.type = desc->type;
5487 	var.dpl = desc->dpl;
5488 	var.db = desc->d;
5489 	var.s = desc->s;
5490 	var.l = desc->l;
5491 	var.g = desc->g;
5492 	var.avl = desc->avl;
5493 	var.present = desc->p;
5494 	var.unusable = !var.present;
5495 	var.padding = 0;
5496 
5497 	kvm_set_segment(vcpu, &var, seg);
5498 	return;
5499 }
5500 
5501 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5502 			    u32 msr_index, u64 *pdata)
5503 {
5504 	struct msr_data msr;
5505 	int r;
5506 
5507 	msr.index = msr_index;
5508 	msr.host_initiated = false;
5509 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5510 	if (r)
5511 		return r;
5512 
5513 	*pdata = msr.data;
5514 	return 0;
5515 }
5516 
5517 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5518 			    u32 msr_index, u64 data)
5519 {
5520 	struct msr_data msr;
5521 
5522 	msr.data = data;
5523 	msr.index = msr_index;
5524 	msr.host_initiated = false;
5525 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5526 }
5527 
5528 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5529 {
5530 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5531 
5532 	return vcpu->arch.smbase;
5533 }
5534 
5535 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5536 {
5537 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5538 
5539 	vcpu->arch.smbase = smbase;
5540 }
5541 
5542 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5543 			      u32 pmc)
5544 {
5545 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5546 }
5547 
5548 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5549 			     u32 pmc, u64 *pdata)
5550 {
5551 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5552 }
5553 
5554 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5555 {
5556 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5557 }
5558 
5559 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5560 			      struct x86_instruction_info *info,
5561 			      enum x86_intercept_stage stage)
5562 {
5563 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5564 }
5565 
5566 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5567 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5568 {
5569 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5570 }
5571 
5572 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5573 {
5574 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5575 }
5576 
5577 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5578 {
5579 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5580 }
5581 
5582 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5583 {
5584 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5585 }
5586 
5587 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5588 {
5589 	return emul_to_vcpu(ctxt)->arch.hflags;
5590 }
5591 
5592 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5593 {
5594 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5595 }
5596 
5597 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5598 {
5599 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5600 }
5601 
5602 static const struct x86_emulate_ops emulate_ops = {
5603 	.read_gpr            = emulator_read_gpr,
5604 	.write_gpr           = emulator_write_gpr,
5605 	.read_std            = kvm_read_guest_virt_system,
5606 	.write_std           = kvm_write_guest_virt_system,
5607 	.read_phys           = kvm_read_guest_phys_system,
5608 	.fetch               = kvm_fetch_guest_virt,
5609 	.read_emulated       = emulator_read_emulated,
5610 	.write_emulated      = emulator_write_emulated,
5611 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5612 	.invlpg              = emulator_invlpg,
5613 	.pio_in_emulated     = emulator_pio_in_emulated,
5614 	.pio_out_emulated    = emulator_pio_out_emulated,
5615 	.get_segment         = emulator_get_segment,
5616 	.set_segment         = emulator_set_segment,
5617 	.get_cached_segment_base = emulator_get_cached_segment_base,
5618 	.get_gdt             = emulator_get_gdt,
5619 	.get_idt	     = emulator_get_idt,
5620 	.set_gdt             = emulator_set_gdt,
5621 	.set_idt	     = emulator_set_idt,
5622 	.get_cr              = emulator_get_cr,
5623 	.set_cr              = emulator_set_cr,
5624 	.cpl                 = emulator_get_cpl,
5625 	.get_dr              = emulator_get_dr,
5626 	.set_dr              = emulator_set_dr,
5627 	.get_smbase          = emulator_get_smbase,
5628 	.set_smbase          = emulator_set_smbase,
5629 	.set_msr             = emulator_set_msr,
5630 	.get_msr             = emulator_get_msr,
5631 	.check_pmc	     = emulator_check_pmc,
5632 	.read_pmc            = emulator_read_pmc,
5633 	.halt                = emulator_halt,
5634 	.wbinvd              = emulator_wbinvd,
5635 	.fix_hypercall       = emulator_fix_hypercall,
5636 	.intercept           = emulator_intercept,
5637 	.get_cpuid           = emulator_get_cpuid,
5638 	.set_nmi_mask        = emulator_set_nmi_mask,
5639 	.get_hflags          = emulator_get_hflags,
5640 	.set_hflags          = emulator_set_hflags,
5641 	.pre_leave_smm       = emulator_pre_leave_smm,
5642 };
5643 
5644 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5645 {
5646 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5647 	/*
5648 	 * an sti; sti; sequence only disable interrupts for the first
5649 	 * instruction. So, if the last instruction, be it emulated or
5650 	 * not, left the system with the INT_STI flag enabled, it
5651 	 * means that the last instruction is an sti. We should not
5652 	 * leave the flag on in this case. The same goes for mov ss
5653 	 */
5654 	if (int_shadow & mask)
5655 		mask = 0;
5656 	if (unlikely(int_shadow || mask)) {
5657 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5658 		if (!mask)
5659 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5660 	}
5661 }
5662 
5663 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5664 {
5665 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5666 	if (ctxt->exception.vector == PF_VECTOR)
5667 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5668 
5669 	if (ctxt->exception.error_code_valid)
5670 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5671 				      ctxt->exception.error_code);
5672 	else
5673 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5674 	return false;
5675 }
5676 
5677 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5678 {
5679 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5680 	int cs_db, cs_l;
5681 
5682 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5683 
5684 	ctxt->eflags = kvm_get_rflags(vcpu);
5685 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5686 
5687 	ctxt->eip = kvm_rip_read(vcpu);
5688 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5689 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5690 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5691 		     cs_db				? X86EMUL_MODE_PROT32 :
5692 							  X86EMUL_MODE_PROT16;
5693 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5694 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5695 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5696 
5697 	init_decode_cache(ctxt);
5698 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5699 }
5700 
5701 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5702 {
5703 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5704 	int ret;
5705 
5706 	init_emulate_ctxt(vcpu);
5707 
5708 	ctxt->op_bytes = 2;
5709 	ctxt->ad_bytes = 2;
5710 	ctxt->_eip = ctxt->eip + inc_eip;
5711 	ret = emulate_int_real(ctxt, irq);
5712 
5713 	if (ret != X86EMUL_CONTINUE)
5714 		return EMULATE_FAIL;
5715 
5716 	ctxt->eip = ctxt->_eip;
5717 	kvm_rip_write(vcpu, ctxt->eip);
5718 	kvm_set_rflags(vcpu, ctxt->eflags);
5719 
5720 	return EMULATE_DONE;
5721 }
5722 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5723 
5724 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5725 {
5726 	int r = EMULATE_DONE;
5727 
5728 	++vcpu->stat.insn_emulation_fail;
5729 	trace_kvm_emulate_insn_failed(vcpu);
5730 
5731 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
5732 		return EMULATE_FAIL;
5733 
5734 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5735 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5736 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5737 		vcpu->run->internal.ndata = 0;
5738 		r = EMULATE_USER_EXIT;
5739 	}
5740 
5741 	kvm_queue_exception(vcpu, UD_VECTOR);
5742 
5743 	return r;
5744 }
5745 
5746 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5747 				  bool write_fault_to_shadow_pgtable,
5748 				  int emulation_type)
5749 {
5750 	gpa_t gpa = cr2;
5751 	kvm_pfn_t pfn;
5752 
5753 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5754 		return false;
5755 
5756 	if (!vcpu->arch.mmu.direct_map) {
5757 		/*
5758 		 * Write permission should be allowed since only
5759 		 * write access need to be emulated.
5760 		 */
5761 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5762 
5763 		/*
5764 		 * If the mapping is invalid in guest, let cpu retry
5765 		 * it to generate fault.
5766 		 */
5767 		if (gpa == UNMAPPED_GVA)
5768 			return true;
5769 	}
5770 
5771 	/*
5772 	 * Do not retry the unhandleable instruction if it faults on the
5773 	 * readonly host memory, otherwise it will goto a infinite loop:
5774 	 * retry instruction -> write #PF -> emulation fail -> retry
5775 	 * instruction -> ...
5776 	 */
5777 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5778 
5779 	/*
5780 	 * If the instruction failed on the error pfn, it can not be fixed,
5781 	 * report the error to userspace.
5782 	 */
5783 	if (is_error_noslot_pfn(pfn))
5784 		return false;
5785 
5786 	kvm_release_pfn_clean(pfn);
5787 
5788 	/* The instructions are well-emulated on direct mmu. */
5789 	if (vcpu->arch.mmu.direct_map) {
5790 		unsigned int indirect_shadow_pages;
5791 
5792 		spin_lock(&vcpu->kvm->mmu_lock);
5793 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5794 		spin_unlock(&vcpu->kvm->mmu_lock);
5795 
5796 		if (indirect_shadow_pages)
5797 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5798 
5799 		return true;
5800 	}
5801 
5802 	/*
5803 	 * if emulation was due to access to shadowed page table
5804 	 * and it failed try to unshadow page and re-enter the
5805 	 * guest to let CPU execute the instruction.
5806 	 */
5807 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5808 
5809 	/*
5810 	 * If the access faults on its page table, it can not
5811 	 * be fixed by unprotecting shadow page and it should
5812 	 * be reported to userspace.
5813 	 */
5814 	return !write_fault_to_shadow_pgtable;
5815 }
5816 
5817 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5818 			      unsigned long cr2,  int emulation_type)
5819 {
5820 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5821 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5822 
5823 	last_retry_eip = vcpu->arch.last_retry_eip;
5824 	last_retry_addr = vcpu->arch.last_retry_addr;
5825 
5826 	/*
5827 	 * If the emulation is caused by #PF and it is non-page_table
5828 	 * writing instruction, it means the VM-EXIT is caused by shadow
5829 	 * page protected, we can zap the shadow page and retry this
5830 	 * instruction directly.
5831 	 *
5832 	 * Note: if the guest uses a non-page-table modifying instruction
5833 	 * on the PDE that points to the instruction, then we will unmap
5834 	 * the instruction and go to an infinite loop. So, we cache the
5835 	 * last retried eip and the last fault address, if we meet the eip
5836 	 * and the address again, we can break out of the potential infinite
5837 	 * loop.
5838 	 */
5839 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5840 
5841 	if (!(emulation_type & EMULTYPE_RETRY))
5842 		return false;
5843 
5844 	if (x86_page_table_writing_insn(ctxt))
5845 		return false;
5846 
5847 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5848 		return false;
5849 
5850 	vcpu->arch.last_retry_eip = ctxt->eip;
5851 	vcpu->arch.last_retry_addr = cr2;
5852 
5853 	if (!vcpu->arch.mmu.direct_map)
5854 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5855 
5856 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5857 
5858 	return true;
5859 }
5860 
5861 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5862 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5863 
5864 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5865 {
5866 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5867 		/* This is a good place to trace that we are exiting SMM.  */
5868 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5869 
5870 		/* Process a latched INIT or SMI, if any.  */
5871 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5872 	}
5873 
5874 	kvm_mmu_reset_context(vcpu);
5875 }
5876 
5877 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5878 {
5879 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5880 
5881 	vcpu->arch.hflags = emul_flags;
5882 
5883 	if (changed & HF_SMM_MASK)
5884 		kvm_smm_changed(vcpu);
5885 }
5886 
5887 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5888 				unsigned long *db)
5889 {
5890 	u32 dr6 = 0;
5891 	int i;
5892 	u32 enable, rwlen;
5893 
5894 	enable = dr7;
5895 	rwlen = dr7 >> 16;
5896 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5897 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5898 			dr6 |= (1 << i);
5899 	return dr6;
5900 }
5901 
5902 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5903 {
5904 	struct kvm_run *kvm_run = vcpu->run;
5905 
5906 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5907 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5908 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5909 		kvm_run->debug.arch.exception = DB_VECTOR;
5910 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5911 		*r = EMULATE_USER_EXIT;
5912 	} else {
5913 		/*
5914 		 * "Certain debug exceptions may clear bit 0-3.  The
5915 		 * remaining contents of the DR6 register are never
5916 		 * cleared by the processor".
5917 		 */
5918 		vcpu->arch.dr6 &= ~15;
5919 		vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5920 		kvm_queue_exception(vcpu, DB_VECTOR);
5921 	}
5922 }
5923 
5924 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5925 {
5926 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5927 	int r = EMULATE_DONE;
5928 
5929 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5930 
5931 	/*
5932 	 * rflags is the old, "raw" value of the flags.  The new value has
5933 	 * not been saved yet.
5934 	 *
5935 	 * This is correct even for TF set by the guest, because "the
5936 	 * processor will not generate this exception after the instruction
5937 	 * that sets the TF flag".
5938 	 */
5939 	if (unlikely(rflags & X86_EFLAGS_TF))
5940 		kvm_vcpu_do_singlestep(vcpu, &r);
5941 	return r == EMULATE_DONE;
5942 }
5943 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5944 
5945 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5946 {
5947 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5948 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5949 		struct kvm_run *kvm_run = vcpu->run;
5950 		unsigned long eip = kvm_get_linear_rip(vcpu);
5951 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5952 					   vcpu->arch.guest_debug_dr7,
5953 					   vcpu->arch.eff_db);
5954 
5955 		if (dr6 != 0) {
5956 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5957 			kvm_run->debug.arch.pc = eip;
5958 			kvm_run->debug.arch.exception = DB_VECTOR;
5959 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5960 			*r = EMULATE_USER_EXIT;
5961 			return true;
5962 		}
5963 	}
5964 
5965 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5966 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5967 		unsigned long eip = kvm_get_linear_rip(vcpu);
5968 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5969 					   vcpu->arch.dr7,
5970 					   vcpu->arch.db);
5971 
5972 		if (dr6 != 0) {
5973 			vcpu->arch.dr6 &= ~15;
5974 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5975 			kvm_queue_exception(vcpu, DB_VECTOR);
5976 			*r = EMULATE_DONE;
5977 			return true;
5978 		}
5979 	}
5980 
5981 	return false;
5982 }
5983 
5984 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
5985 {
5986 	switch (ctxt->opcode_len) {
5987 	case 1:
5988 		switch (ctxt->b) {
5989 		case 0xe4:	/* IN */
5990 		case 0xe5:
5991 		case 0xec:
5992 		case 0xed:
5993 		case 0xe6:	/* OUT */
5994 		case 0xe7:
5995 		case 0xee:
5996 		case 0xef:
5997 		case 0x6c:	/* INS */
5998 		case 0x6d:
5999 		case 0x6e:	/* OUTS */
6000 		case 0x6f:
6001 			return true;
6002 		}
6003 		break;
6004 	case 2:
6005 		switch (ctxt->b) {
6006 		case 0x33:	/* RDPMC */
6007 			return true;
6008 		}
6009 		break;
6010 	}
6011 
6012 	return false;
6013 }
6014 
6015 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6016 			    unsigned long cr2,
6017 			    int emulation_type,
6018 			    void *insn,
6019 			    int insn_len)
6020 {
6021 	int r;
6022 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6023 	bool writeback = true;
6024 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6025 
6026 	/*
6027 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6028 	 * never reused.
6029 	 */
6030 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6031 	kvm_clear_exception_queue(vcpu);
6032 
6033 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6034 		init_emulate_ctxt(vcpu);
6035 
6036 		/*
6037 		 * We will reenter on the same instruction since
6038 		 * we do not set complete_userspace_io.  This does not
6039 		 * handle watchpoints yet, those would be handled in
6040 		 * the emulate_ops.
6041 		 */
6042 		if (!(emulation_type & EMULTYPE_SKIP) &&
6043 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6044 			return r;
6045 
6046 		ctxt->interruptibility = 0;
6047 		ctxt->have_exception = false;
6048 		ctxt->exception.vector = -1;
6049 		ctxt->perm_ok = false;
6050 
6051 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6052 
6053 		r = x86_decode_insn(ctxt, insn, insn_len);
6054 
6055 		trace_kvm_emulate_insn_start(vcpu);
6056 		++vcpu->stat.insn_emulation;
6057 		if (r != EMULATION_OK)  {
6058 			if (emulation_type & EMULTYPE_TRAP_UD)
6059 				return EMULATE_FAIL;
6060 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6061 						emulation_type))
6062 				return EMULATE_DONE;
6063 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6064 				return EMULATE_DONE;
6065 			if (emulation_type & EMULTYPE_SKIP)
6066 				return EMULATE_FAIL;
6067 			return handle_emulation_failure(vcpu, emulation_type);
6068 		}
6069 	}
6070 
6071 	if ((emulation_type & EMULTYPE_VMWARE) &&
6072 	    !is_vmware_backdoor_opcode(ctxt))
6073 		return EMULATE_FAIL;
6074 
6075 	if (emulation_type & EMULTYPE_SKIP) {
6076 		kvm_rip_write(vcpu, ctxt->_eip);
6077 		if (ctxt->eflags & X86_EFLAGS_RF)
6078 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6079 		return EMULATE_DONE;
6080 	}
6081 
6082 	if (retry_instruction(ctxt, cr2, emulation_type))
6083 		return EMULATE_DONE;
6084 
6085 	/* this is needed for vmware backdoor interface to work since it
6086 	   changes registers values  during IO operation */
6087 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6088 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6089 		emulator_invalidate_register_cache(ctxt);
6090 	}
6091 
6092 restart:
6093 	/* Save the faulting GPA (cr2) in the address field */
6094 	ctxt->exception.address = cr2;
6095 
6096 	r = x86_emulate_insn(ctxt);
6097 
6098 	if (r == EMULATION_INTERCEPTED)
6099 		return EMULATE_DONE;
6100 
6101 	if (r == EMULATION_FAILED) {
6102 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6103 					emulation_type))
6104 			return EMULATE_DONE;
6105 
6106 		return handle_emulation_failure(vcpu, emulation_type);
6107 	}
6108 
6109 	if (ctxt->have_exception) {
6110 		r = EMULATE_DONE;
6111 		if (inject_emulated_exception(vcpu))
6112 			return r;
6113 	} else if (vcpu->arch.pio.count) {
6114 		if (!vcpu->arch.pio.in) {
6115 			/* FIXME: return into emulator if single-stepping.  */
6116 			vcpu->arch.pio.count = 0;
6117 		} else {
6118 			writeback = false;
6119 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6120 		}
6121 		r = EMULATE_USER_EXIT;
6122 	} else if (vcpu->mmio_needed) {
6123 		if (!vcpu->mmio_is_write)
6124 			writeback = false;
6125 		r = EMULATE_USER_EXIT;
6126 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6127 	} else if (r == EMULATION_RESTART)
6128 		goto restart;
6129 	else
6130 		r = EMULATE_DONE;
6131 
6132 	if (writeback) {
6133 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6134 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6135 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6136 		kvm_rip_write(vcpu, ctxt->eip);
6137 		if (r == EMULATE_DONE &&
6138 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6139 			kvm_vcpu_do_singlestep(vcpu, &r);
6140 		if (!ctxt->have_exception ||
6141 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6142 			__kvm_set_rflags(vcpu, ctxt->eflags);
6143 
6144 		/*
6145 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6146 		 * do nothing, and it will be requested again as soon as
6147 		 * the shadow expires.  But we still need to check here,
6148 		 * because POPF has no interrupt shadow.
6149 		 */
6150 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6151 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6152 	} else
6153 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6154 
6155 	return r;
6156 }
6157 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
6158 
6159 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6160 			    unsigned short port)
6161 {
6162 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6163 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6164 					    size, port, &val, 1);
6165 	/* do not return to emulator after return from userspace */
6166 	vcpu->arch.pio.count = 0;
6167 	return ret;
6168 }
6169 
6170 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6171 {
6172 	unsigned long val;
6173 
6174 	/* We should only ever be called with arch.pio.count equal to 1 */
6175 	BUG_ON(vcpu->arch.pio.count != 1);
6176 
6177 	/* For size less than 4 we merge, else we zero extend */
6178 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6179 					: 0;
6180 
6181 	/*
6182 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6183 	 * the copy and tracing
6184 	 */
6185 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6186 				 vcpu->arch.pio.port, &val, 1);
6187 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6188 
6189 	return 1;
6190 }
6191 
6192 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6193 			   unsigned short port)
6194 {
6195 	unsigned long val;
6196 	int ret;
6197 
6198 	/* For size less than 4 we merge, else we zero extend */
6199 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6200 
6201 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6202 				       &val, 1);
6203 	if (ret) {
6204 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6205 		return ret;
6206 	}
6207 
6208 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6209 
6210 	return 0;
6211 }
6212 
6213 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6214 {
6215 	int ret = kvm_skip_emulated_instruction(vcpu);
6216 
6217 	/*
6218 	 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6219 	 * KVM_EXIT_DEBUG here.
6220 	 */
6221 	if (in)
6222 		return kvm_fast_pio_in(vcpu, size, port) && ret;
6223 	else
6224 		return kvm_fast_pio_out(vcpu, size, port) && ret;
6225 }
6226 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6227 
6228 static int kvmclock_cpu_down_prep(unsigned int cpu)
6229 {
6230 	__this_cpu_write(cpu_tsc_khz, 0);
6231 	return 0;
6232 }
6233 
6234 static void tsc_khz_changed(void *data)
6235 {
6236 	struct cpufreq_freqs *freq = data;
6237 	unsigned long khz = 0;
6238 
6239 	if (data)
6240 		khz = freq->new;
6241 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6242 		khz = cpufreq_quick_get(raw_smp_processor_id());
6243 	if (!khz)
6244 		khz = tsc_khz;
6245 	__this_cpu_write(cpu_tsc_khz, khz);
6246 }
6247 
6248 #ifdef CONFIG_X86_64
6249 static void kvm_hyperv_tsc_notifier(void)
6250 {
6251 	struct kvm *kvm;
6252 	struct kvm_vcpu *vcpu;
6253 	int cpu;
6254 
6255 	spin_lock(&kvm_lock);
6256 	list_for_each_entry(kvm, &vm_list, vm_list)
6257 		kvm_make_mclock_inprogress_request(kvm);
6258 
6259 	hyperv_stop_tsc_emulation();
6260 
6261 	/* TSC frequency always matches when on Hyper-V */
6262 	for_each_present_cpu(cpu)
6263 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6264 	kvm_max_guest_tsc_khz = tsc_khz;
6265 
6266 	list_for_each_entry(kvm, &vm_list, vm_list) {
6267 		struct kvm_arch *ka = &kvm->arch;
6268 
6269 		spin_lock(&ka->pvclock_gtod_sync_lock);
6270 
6271 		pvclock_update_vm_gtod_copy(kvm);
6272 
6273 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6274 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6275 
6276 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6277 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6278 
6279 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6280 	}
6281 	spin_unlock(&kvm_lock);
6282 }
6283 #endif
6284 
6285 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6286 				     void *data)
6287 {
6288 	struct cpufreq_freqs *freq = data;
6289 	struct kvm *kvm;
6290 	struct kvm_vcpu *vcpu;
6291 	int i, send_ipi = 0;
6292 
6293 	/*
6294 	 * We allow guests to temporarily run on slowing clocks,
6295 	 * provided we notify them after, or to run on accelerating
6296 	 * clocks, provided we notify them before.  Thus time never
6297 	 * goes backwards.
6298 	 *
6299 	 * However, we have a problem.  We can't atomically update
6300 	 * the frequency of a given CPU from this function; it is
6301 	 * merely a notifier, which can be called from any CPU.
6302 	 * Changing the TSC frequency at arbitrary points in time
6303 	 * requires a recomputation of local variables related to
6304 	 * the TSC for each VCPU.  We must flag these local variables
6305 	 * to be updated and be sure the update takes place with the
6306 	 * new frequency before any guests proceed.
6307 	 *
6308 	 * Unfortunately, the combination of hotplug CPU and frequency
6309 	 * change creates an intractable locking scenario; the order
6310 	 * of when these callouts happen is undefined with respect to
6311 	 * CPU hotplug, and they can race with each other.  As such,
6312 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6313 	 * undefined; you can actually have a CPU frequency change take
6314 	 * place in between the computation of X and the setting of the
6315 	 * variable.  To protect against this problem, all updates of
6316 	 * the per_cpu tsc_khz variable are done in an interrupt
6317 	 * protected IPI, and all callers wishing to update the value
6318 	 * must wait for a synchronous IPI to complete (which is trivial
6319 	 * if the caller is on the CPU already).  This establishes the
6320 	 * necessary total order on variable updates.
6321 	 *
6322 	 * Note that because a guest time update may take place
6323 	 * anytime after the setting of the VCPU's request bit, the
6324 	 * correct TSC value must be set before the request.  However,
6325 	 * to ensure the update actually makes it to any guest which
6326 	 * starts running in hardware virtualization between the set
6327 	 * and the acquisition of the spinlock, we must also ping the
6328 	 * CPU after setting the request bit.
6329 	 *
6330 	 */
6331 
6332 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6333 		return 0;
6334 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6335 		return 0;
6336 
6337 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6338 
6339 	spin_lock(&kvm_lock);
6340 	list_for_each_entry(kvm, &vm_list, vm_list) {
6341 		kvm_for_each_vcpu(i, vcpu, kvm) {
6342 			if (vcpu->cpu != freq->cpu)
6343 				continue;
6344 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6345 			if (vcpu->cpu != smp_processor_id())
6346 				send_ipi = 1;
6347 		}
6348 	}
6349 	spin_unlock(&kvm_lock);
6350 
6351 	if (freq->old < freq->new && send_ipi) {
6352 		/*
6353 		 * We upscale the frequency.  Must make the guest
6354 		 * doesn't see old kvmclock values while running with
6355 		 * the new frequency, otherwise we risk the guest sees
6356 		 * time go backwards.
6357 		 *
6358 		 * In case we update the frequency for another cpu
6359 		 * (which might be in guest context) send an interrupt
6360 		 * to kick the cpu out of guest context.  Next time
6361 		 * guest context is entered kvmclock will be updated,
6362 		 * so the guest will not see stale values.
6363 		 */
6364 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6365 	}
6366 	return 0;
6367 }
6368 
6369 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6370 	.notifier_call  = kvmclock_cpufreq_notifier
6371 };
6372 
6373 static int kvmclock_cpu_online(unsigned int cpu)
6374 {
6375 	tsc_khz_changed(NULL);
6376 	return 0;
6377 }
6378 
6379 static void kvm_timer_init(void)
6380 {
6381 	max_tsc_khz = tsc_khz;
6382 
6383 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6384 #ifdef CONFIG_CPU_FREQ
6385 		struct cpufreq_policy policy;
6386 		int cpu;
6387 
6388 		memset(&policy, 0, sizeof(policy));
6389 		cpu = get_cpu();
6390 		cpufreq_get_policy(&policy, cpu);
6391 		if (policy.cpuinfo.max_freq)
6392 			max_tsc_khz = policy.cpuinfo.max_freq;
6393 		put_cpu();
6394 #endif
6395 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6396 					  CPUFREQ_TRANSITION_NOTIFIER);
6397 	}
6398 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6399 
6400 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6401 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6402 }
6403 
6404 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6405 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6406 
6407 int kvm_is_in_guest(void)
6408 {
6409 	return __this_cpu_read(current_vcpu) != NULL;
6410 }
6411 
6412 static int kvm_is_user_mode(void)
6413 {
6414 	int user_mode = 3;
6415 
6416 	if (__this_cpu_read(current_vcpu))
6417 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6418 
6419 	return user_mode != 0;
6420 }
6421 
6422 static unsigned long kvm_get_guest_ip(void)
6423 {
6424 	unsigned long ip = 0;
6425 
6426 	if (__this_cpu_read(current_vcpu))
6427 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6428 
6429 	return ip;
6430 }
6431 
6432 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6433 	.is_in_guest		= kvm_is_in_guest,
6434 	.is_user_mode		= kvm_is_user_mode,
6435 	.get_guest_ip		= kvm_get_guest_ip,
6436 };
6437 
6438 static void kvm_set_mmio_spte_mask(void)
6439 {
6440 	u64 mask;
6441 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6442 
6443 	/*
6444 	 * Set the reserved bits and the present bit of an paging-structure
6445 	 * entry to generate page fault with PFER.RSV = 1.
6446 	 */
6447 	 /* Mask the reserved physical address bits. */
6448 	mask = rsvd_bits(maxphyaddr, 51);
6449 
6450 	/* Set the present bit. */
6451 	mask |= 1ull;
6452 
6453 #ifdef CONFIG_X86_64
6454 	/*
6455 	 * If reserved bit is not supported, clear the present bit to disable
6456 	 * mmio page fault.
6457 	 */
6458 	if (maxphyaddr == 52)
6459 		mask &= ~1ull;
6460 #endif
6461 
6462 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6463 }
6464 
6465 #ifdef CONFIG_X86_64
6466 static void pvclock_gtod_update_fn(struct work_struct *work)
6467 {
6468 	struct kvm *kvm;
6469 
6470 	struct kvm_vcpu *vcpu;
6471 	int i;
6472 
6473 	spin_lock(&kvm_lock);
6474 	list_for_each_entry(kvm, &vm_list, vm_list)
6475 		kvm_for_each_vcpu(i, vcpu, kvm)
6476 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6477 	atomic_set(&kvm_guest_has_master_clock, 0);
6478 	spin_unlock(&kvm_lock);
6479 }
6480 
6481 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6482 
6483 /*
6484  * Notification about pvclock gtod data update.
6485  */
6486 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6487 			       void *priv)
6488 {
6489 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6490 	struct timekeeper *tk = priv;
6491 
6492 	update_pvclock_gtod(tk);
6493 
6494 	/* disable master clock if host does not trust, or does not
6495 	 * use, TSC based clocksource.
6496 	 */
6497 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6498 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6499 		queue_work(system_long_wq, &pvclock_gtod_work);
6500 
6501 	return 0;
6502 }
6503 
6504 static struct notifier_block pvclock_gtod_notifier = {
6505 	.notifier_call = pvclock_gtod_notify,
6506 };
6507 #endif
6508 
6509 int kvm_arch_init(void *opaque)
6510 {
6511 	int r;
6512 	struct kvm_x86_ops *ops = opaque;
6513 
6514 	if (kvm_x86_ops) {
6515 		printk(KERN_ERR "kvm: already loaded the other module\n");
6516 		r = -EEXIST;
6517 		goto out;
6518 	}
6519 
6520 	if (!ops->cpu_has_kvm_support()) {
6521 		printk(KERN_ERR "kvm: no hardware support\n");
6522 		r = -EOPNOTSUPP;
6523 		goto out;
6524 	}
6525 	if (ops->disabled_by_bios()) {
6526 		printk(KERN_ERR "kvm: disabled by bios\n");
6527 		r = -EOPNOTSUPP;
6528 		goto out;
6529 	}
6530 
6531 	r = -ENOMEM;
6532 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6533 	if (!shared_msrs) {
6534 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6535 		goto out;
6536 	}
6537 
6538 	r = kvm_mmu_module_init();
6539 	if (r)
6540 		goto out_free_percpu;
6541 
6542 	kvm_set_mmio_spte_mask();
6543 
6544 	kvm_x86_ops = ops;
6545 
6546 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6547 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6548 			PT_PRESENT_MASK, 0, sme_me_mask);
6549 	kvm_timer_init();
6550 
6551 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6552 
6553 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6554 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6555 
6556 	kvm_lapic_init();
6557 #ifdef CONFIG_X86_64
6558 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6559 
6560 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6561 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6562 #endif
6563 
6564 	return 0;
6565 
6566 out_free_percpu:
6567 	free_percpu(shared_msrs);
6568 out:
6569 	return r;
6570 }
6571 
6572 void kvm_arch_exit(void)
6573 {
6574 #ifdef CONFIG_X86_64
6575 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6576 		clear_hv_tscchange_cb();
6577 #endif
6578 	kvm_lapic_exit();
6579 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6580 
6581 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6582 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6583 					    CPUFREQ_TRANSITION_NOTIFIER);
6584 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6585 #ifdef CONFIG_X86_64
6586 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6587 #endif
6588 	kvm_x86_ops = NULL;
6589 	kvm_mmu_module_exit();
6590 	free_percpu(shared_msrs);
6591 }
6592 
6593 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6594 {
6595 	++vcpu->stat.halt_exits;
6596 	if (lapic_in_kernel(vcpu)) {
6597 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6598 		return 1;
6599 	} else {
6600 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6601 		return 0;
6602 	}
6603 }
6604 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6605 
6606 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6607 {
6608 	int ret = kvm_skip_emulated_instruction(vcpu);
6609 	/*
6610 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6611 	 * KVM_EXIT_DEBUG here.
6612 	 */
6613 	return kvm_vcpu_halt(vcpu) && ret;
6614 }
6615 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6616 
6617 #ifdef CONFIG_X86_64
6618 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6619 			        unsigned long clock_type)
6620 {
6621 	struct kvm_clock_pairing clock_pairing;
6622 	struct timespec ts;
6623 	u64 cycle;
6624 	int ret;
6625 
6626 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6627 		return -KVM_EOPNOTSUPP;
6628 
6629 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6630 		return -KVM_EOPNOTSUPP;
6631 
6632 	clock_pairing.sec = ts.tv_sec;
6633 	clock_pairing.nsec = ts.tv_nsec;
6634 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6635 	clock_pairing.flags = 0;
6636 
6637 	ret = 0;
6638 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6639 			    sizeof(struct kvm_clock_pairing)))
6640 		ret = -KVM_EFAULT;
6641 
6642 	return ret;
6643 }
6644 #endif
6645 
6646 /*
6647  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6648  *
6649  * @apicid - apicid of vcpu to be kicked.
6650  */
6651 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6652 {
6653 	struct kvm_lapic_irq lapic_irq;
6654 
6655 	lapic_irq.shorthand = 0;
6656 	lapic_irq.dest_mode = 0;
6657 	lapic_irq.level = 0;
6658 	lapic_irq.dest_id = apicid;
6659 	lapic_irq.msi_redir_hint = false;
6660 
6661 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6662 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6663 }
6664 
6665 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6666 {
6667 	vcpu->arch.apicv_active = false;
6668 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6669 }
6670 
6671 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6672 {
6673 	unsigned long nr, a0, a1, a2, a3, ret;
6674 	int op_64_bit, r;
6675 
6676 	r = kvm_skip_emulated_instruction(vcpu);
6677 
6678 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6679 		return kvm_hv_hypercall(vcpu);
6680 
6681 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6682 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6683 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6684 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6685 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6686 
6687 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6688 
6689 	op_64_bit = is_64_bit_mode(vcpu);
6690 	if (!op_64_bit) {
6691 		nr &= 0xFFFFFFFF;
6692 		a0 &= 0xFFFFFFFF;
6693 		a1 &= 0xFFFFFFFF;
6694 		a2 &= 0xFFFFFFFF;
6695 		a3 &= 0xFFFFFFFF;
6696 	}
6697 
6698 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6699 		ret = -KVM_EPERM;
6700 		goto out;
6701 	}
6702 
6703 	switch (nr) {
6704 	case KVM_HC_VAPIC_POLL_IRQ:
6705 		ret = 0;
6706 		break;
6707 	case KVM_HC_KICK_CPU:
6708 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6709 		ret = 0;
6710 		break;
6711 #ifdef CONFIG_X86_64
6712 	case KVM_HC_CLOCK_PAIRING:
6713 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6714 		break;
6715 #endif
6716 	default:
6717 		ret = -KVM_ENOSYS;
6718 		break;
6719 	}
6720 out:
6721 	if (!op_64_bit)
6722 		ret = (u32)ret;
6723 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6724 	++vcpu->stat.hypercalls;
6725 	return r;
6726 }
6727 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6728 
6729 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6730 {
6731 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6732 	char instruction[3];
6733 	unsigned long rip = kvm_rip_read(vcpu);
6734 
6735 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6736 
6737 	return emulator_write_emulated(ctxt, rip, instruction, 3,
6738 		&ctxt->exception);
6739 }
6740 
6741 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6742 {
6743 	return vcpu->run->request_interrupt_window &&
6744 		likely(!pic_in_kernel(vcpu->kvm));
6745 }
6746 
6747 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6748 {
6749 	struct kvm_run *kvm_run = vcpu->run;
6750 
6751 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6752 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6753 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6754 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6755 	kvm_run->ready_for_interrupt_injection =
6756 		pic_in_kernel(vcpu->kvm) ||
6757 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6758 }
6759 
6760 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6761 {
6762 	int max_irr, tpr;
6763 
6764 	if (!kvm_x86_ops->update_cr8_intercept)
6765 		return;
6766 
6767 	if (!lapic_in_kernel(vcpu))
6768 		return;
6769 
6770 	if (vcpu->arch.apicv_active)
6771 		return;
6772 
6773 	if (!vcpu->arch.apic->vapic_addr)
6774 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6775 	else
6776 		max_irr = -1;
6777 
6778 	if (max_irr != -1)
6779 		max_irr >>= 4;
6780 
6781 	tpr = kvm_lapic_get_cr8(vcpu);
6782 
6783 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6784 }
6785 
6786 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6787 {
6788 	int r;
6789 
6790 	/* try to reinject previous events if any */
6791 
6792 	if (vcpu->arch.exception.injected)
6793 		kvm_x86_ops->queue_exception(vcpu);
6794 	/*
6795 	 * Do not inject an NMI or interrupt if there is a pending
6796 	 * exception.  Exceptions and interrupts are recognized at
6797 	 * instruction boundaries, i.e. the start of an instruction.
6798 	 * Trap-like exceptions, e.g. #DB, have higher priority than
6799 	 * NMIs and interrupts, i.e. traps are recognized before an
6800 	 * NMI/interrupt that's pending on the same instruction.
6801 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
6802 	 * priority, but are only generated (pended) during instruction
6803 	 * execution, i.e. a pending fault-like exception means the
6804 	 * fault occurred on the *previous* instruction and must be
6805 	 * serviced prior to recognizing any new events in order to
6806 	 * fully complete the previous instruction.
6807 	 */
6808 	else if (!vcpu->arch.exception.pending) {
6809 		if (vcpu->arch.nmi_injected)
6810 			kvm_x86_ops->set_nmi(vcpu);
6811 		else if (vcpu->arch.interrupt.injected)
6812 			kvm_x86_ops->set_irq(vcpu);
6813 	}
6814 
6815 	/*
6816 	 * Call check_nested_events() even if we reinjected a previous event
6817 	 * in order for caller to determine if it should require immediate-exit
6818 	 * from L2 to L1 due to pending L1 events which require exit
6819 	 * from L2 to L1.
6820 	 */
6821 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6822 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6823 		if (r != 0)
6824 			return r;
6825 	}
6826 
6827 	/* try to inject new event if pending */
6828 	if (vcpu->arch.exception.pending) {
6829 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6830 					vcpu->arch.exception.has_error_code,
6831 					vcpu->arch.exception.error_code);
6832 
6833 		WARN_ON_ONCE(vcpu->arch.exception.injected);
6834 		vcpu->arch.exception.pending = false;
6835 		vcpu->arch.exception.injected = true;
6836 
6837 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6838 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6839 					     X86_EFLAGS_RF);
6840 
6841 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6842 		    (vcpu->arch.dr7 & DR7_GD)) {
6843 			vcpu->arch.dr7 &= ~DR7_GD;
6844 			kvm_update_dr7(vcpu);
6845 		}
6846 
6847 		kvm_x86_ops->queue_exception(vcpu);
6848 	}
6849 
6850 	/* Don't consider new event if we re-injected an event */
6851 	if (kvm_event_needs_reinjection(vcpu))
6852 		return 0;
6853 
6854 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
6855 	    kvm_x86_ops->smi_allowed(vcpu)) {
6856 		vcpu->arch.smi_pending = false;
6857 		++vcpu->arch.smi_count;
6858 		enter_smm(vcpu);
6859 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6860 		--vcpu->arch.nmi_pending;
6861 		vcpu->arch.nmi_injected = true;
6862 		kvm_x86_ops->set_nmi(vcpu);
6863 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6864 		/*
6865 		 * Because interrupts can be injected asynchronously, we are
6866 		 * calling check_nested_events again here to avoid a race condition.
6867 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6868 		 * proposal and current concerns.  Perhaps we should be setting
6869 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6870 		 */
6871 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6872 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6873 			if (r != 0)
6874 				return r;
6875 		}
6876 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6877 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6878 					    false);
6879 			kvm_x86_ops->set_irq(vcpu);
6880 		}
6881 	}
6882 
6883 	return 0;
6884 }
6885 
6886 static void process_nmi(struct kvm_vcpu *vcpu)
6887 {
6888 	unsigned limit = 2;
6889 
6890 	/*
6891 	 * x86 is limited to one NMI running, and one NMI pending after it.
6892 	 * If an NMI is already in progress, limit further NMIs to just one.
6893 	 * Otherwise, allow two (and we'll inject the first one immediately).
6894 	 */
6895 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6896 		limit = 1;
6897 
6898 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6899 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6900 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6901 }
6902 
6903 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6904 {
6905 	u32 flags = 0;
6906 	flags |= seg->g       << 23;
6907 	flags |= seg->db      << 22;
6908 	flags |= seg->l       << 21;
6909 	flags |= seg->avl     << 20;
6910 	flags |= seg->present << 15;
6911 	flags |= seg->dpl     << 13;
6912 	flags |= seg->s       << 12;
6913 	flags |= seg->type    << 8;
6914 	return flags;
6915 }
6916 
6917 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6918 {
6919 	struct kvm_segment seg;
6920 	int offset;
6921 
6922 	kvm_get_segment(vcpu, &seg, n);
6923 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6924 
6925 	if (n < 3)
6926 		offset = 0x7f84 + n * 12;
6927 	else
6928 		offset = 0x7f2c + (n - 3) * 12;
6929 
6930 	put_smstate(u32, buf, offset + 8, seg.base);
6931 	put_smstate(u32, buf, offset + 4, seg.limit);
6932 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6933 }
6934 
6935 #ifdef CONFIG_X86_64
6936 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6937 {
6938 	struct kvm_segment seg;
6939 	int offset;
6940 	u16 flags;
6941 
6942 	kvm_get_segment(vcpu, &seg, n);
6943 	offset = 0x7e00 + n * 16;
6944 
6945 	flags = enter_smm_get_segment_flags(&seg) >> 8;
6946 	put_smstate(u16, buf, offset, seg.selector);
6947 	put_smstate(u16, buf, offset + 2, flags);
6948 	put_smstate(u32, buf, offset + 4, seg.limit);
6949 	put_smstate(u64, buf, offset + 8, seg.base);
6950 }
6951 #endif
6952 
6953 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6954 {
6955 	struct desc_ptr dt;
6956 	struct kvm_segment seg;
6957 	unsigned long val;
6958 	int i;
6959 
6960 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6961 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6962 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6963 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6964 
6965 	for (i = 0; i < 8; i++)
6966 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6967 
6968 	kvm_get_dr(vcpu, 6, &val);
6969 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6970 	kvm_get_dr(vcpu, 7, &val);
6971 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6972 
6973 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6974 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6975 	put_smstate(u32, buf, 0x7f64, seg.base);
6976 	put_smstate(u32, buf, 0x7f60, seg.limit);
6977 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6978 
6979 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6980 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6981 	put_smstate(u32, buf, 0x7f80, seg.base);
6982 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6983 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6984 
6985 	kvm_x86_ops->get_gdt(vcpu, &dt);
6986 	put_smstate(u32, buf, 0x7f74, dt.address);
6987 	put_smstate(u32, buf, 0x7f70, dt.size);
6988 
6989 	kvm_x86_ops->get_idt(vcpu, &dt);
6990 	put_smstate(u32, buf, 0x7f58, dt.address);
6991 	put_smstate(u32, buf, 0x7f54, dt.size);
6992 
6993 	for (i = 0; i < 6; i++)
6994 		enter_smm_save_seg_32(vcpu, buf, i);
6995 
6996 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6997 
6998 	/* revision id */
6999 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7000 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7001 }
7002 
7003 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7004 {
7005 #ifdef CONFIG_X86_64
7006 	struct desc_ptr dt;
7007 	struct kvm_segment seg;
7008 	unsigned long val;
7009 	int i;
7010 
7011 	for (i = 0; i < 16; i++)
7012 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7013 
7014 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7015 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7016 
7017 	kvm_get_dr(vcpu, 6, &val);
7018 	put_smstate(u64, buf, 0x7f68, val);
7019 	kvm_get_dr(vcpu, 7, &val);
7020 	put_smstate(u64, buf, 0x7f60, val);
7021 
7022 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7023 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7024 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7025 
7026 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7027 
7028 	/* revision id */
7029 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7030 
7031 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7032 
7033 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7034 	put_smstate(u16, buf, 0x7e90, seg.selector);
7035 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7036 	put_smstate(u32, buf, 0x7e94, seg.limit);
7037 	put_smstate(u64, buf, 0x7e98, seg.base);
7038 
7039 	kvm_x86_ops->get_idt(vcpu, &dt);
7040 	put_smstate(u32, buf, 0x7e84, dt.size);
7041 	put_smstate(u64, buf, 0x7e88, dt.address);
7042 
7043 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7044 	put_smstate(u16, buf, 0x7e70, seg.selector);
7045 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7046 	put_smstate(u32, buf, 0x7e74, seg.limit);
7047 	put_smstate(u64, buf, 0x7e78, seg.base);
7048 
7049 	kvm_x86_ops->get_gdt(vcpu, &dt);
7050 	put_smstate(u32, buf, 0x7e64, dt.size);
7051 	put_smstate(u64, buf, 0x7e68, dt.address);
7052 
7053 	for (i = 0; i < 6; i++)
7054 		enter_smm_save_seg_64(vcpu, buf, i);
7055 #else
7056 	WARN_ON_ONCE(1);
7057 #endif
7058 }
7059 
7060 static void enter_smm(struct kvm_vcpu *vcpu)
7061 {
7062 	struct kvm_segment cs, ds;
7063 	struct desc_ptr dt;
7064 	char buf[512];
7065 	u32 cr0;
7066 
7067 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7068 	memset(buf, 0, 512);
7069 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7070 		enter_smm_save_state_64(vcpu, buf);
7071 	else
7072 		enter_smm_save_state_32(vcpu, buf);
7073 
7074 	/*
7075 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7076 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7077 	 * the SMM state-save area.
7078 	 */
7079 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7080 
7081 	vcpu->arch.hflags |= HF_SMM_MASK;
7082 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7083 
7084 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7085 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7086 	else
7087 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7088 
7089 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7090 	kvm_rip_write(vcpu, 0x8000);
7091 
7092 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7093 	kvm_x86_ops->set_cr0(vcpu, cr0);
7094 	vcpu->arch.cr0 = cr0;
7095 
7096 	kvm_x86_ops->set_cr4(vcpu, 0);
7097 
7098 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7099 	dt.address = dt.size = 0;
7100 	kvm_x86_ops->set_idt(vcpu, &dt);
7101 
7102 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7103 
7104 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7105 	cs.base = vcpu->arch.smbase;
7106 
7107 	ds.selector = 0;
7108 	ds.base = 0;
7109 
7110 	cs.limit    = ds.limit = 0xffffffff;
7111 	cs.type     = ds.type = 0x3;
7112 	cs.dpl      = ds.dpl = 0;
7113 	cs.db       = ds.db = 0;
7114 	cs.s        = ds.s = 1;
7115 	cs.l        = ds.l = 0;
7116 	cs.g        = ds.g = 1;
7117 	cs.avl      = ds.avl = 0;
7118 	cs.present  = ds.present = 1;
7119 	cs.unusable = ds.unusable = 0;
7120 	cs.padding  = ds.padding = 0;
7121 
7122 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7123 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7124 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7125 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7126 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7127 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7128 
7129 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7130 		kvm_x86_ops->set_efer(vcpu, 0);
7131 
7132 	kvm_update_cpuid(vcpu);
7133 	kvm_mmu_reset_context(vcpu);
7134 }
7135 
7136 static void process_smi(struct kvm_vcpu *vcpu)
7137 {
7138 	vcpu->arch.smi_pending = true;
7139 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7140 }
7141 
7142 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7143 {
7144 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7145 }
7146 
7147 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7148 {
7149 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7150 		return;
7151 
7152 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7153 
7154 	if (irqchip_split(vcpu->kvm))
7155 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7156 	else {
7157 		if (vcpu->arch.apicv_active)
7158 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7159 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7160 	}
7161 
7162 	if (is_guest_mode(vcpu))
7163 		vcpu->arch.load_eoi_exitmap_pending = true;
7164 	else
7165 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7166 }
7167 
7168 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7169 {
7170 	u64 eoi_exit_bitmap[4];
7171 
7172 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7173 		return;
7174 
7175 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7176 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7177 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7178 }
7179 
7180 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7181 		unsigned long start, unsigned long end)
7182 {
7183 	unsigned long apic_address;
7184 
7185 	/*
7186 	 * The physical address of apic access page is stored in the VMCS.
7187 	 * Update it when it becomes invalid.
7188 	 */
7189 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7190 	if (start <= apic_address && apic_address < end)
7191 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7192 }
7193 
7194 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7195 {
7196 	struct page *page = NULL;
7197 
7198 	if (!lapic_in_kernel(vcpu))
7199 		return;
7200 
7201 	if (!kvm_x86_ops->set_apic_access_page_addr)
7202 		return;
7203 
7204 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7205 	if (is_error_page(page))
7206 		return;
7207 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7208 
7209 	/*
7210 	 * Do not pin apic access page in memory, the MMU notifier
7211 	 * will call us again if it is migrated or swapped out.
7212 	 */
7213 	put_page(page);
7214 }
7215 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7216 
7217 /*
7218  * Returns 1 to let vcpu_run() continue the guest execution loop without
7219  * exiting to the userspace.  Otherwise, the value will be returned to the
7220  * userspace.
7221  */
7222 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7223 {
7224 	int r;
7225 	bool req_int_win =
7226 		dm_request_for_irq_injection(vcpu) &&
7227 		kvm_cpu_accept_dm_intr(vcpu);
7228 
7229 	bool req_immediate_exit = false;
7230 
7231 	if (kvm_request_pending(vcpu)) {
7232 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7233 			kvm_mmu_unload(vcpu);
7234 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7235 			__kvm_migrate_timers(vcpu);
7236 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7237 			kvm_gen_update_masterclock(vcpu->kvm);
7238 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7239 			kvm_gen_kvmclock_update(vcpu);
7240 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7241 			r = kvm_guest_time_update(vcpu);
7242 			if (unlikely(r))
7243 				goto out;
7244 		}
7245 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7246 			kvm_mmu_sync_roots(vcpu);
7247 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7248 			kvm_vcpu_flush_tlb(vcpu, true);
7249 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7250 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7251 			r = 0;
7252 			goto out;
7253 		}
7254 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7255 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7256 			vcpu->mmio_needed = 0;
7257 			r = 0;
7258 			goto out;
7259 		}
7260 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7261 			/* Page is swapped out. Do synthetic halt */
7262 			vcpu->arch.apf.halted = true;
7263 			r = 1;
7264 			goto out;
7265 		}
7266 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7267 			record_steal_time(vcpu);
7268 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7269 			process_smi(vcpu);
7270 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7271 			process_nmi(vcpu);
7272 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7273 			kvm_pmu_handle_event(vcpu);
7274 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7275 			kvm_pmu_deliver_pmi(vcpu);
7276 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7277 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7278 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7279 				     vcpu->arch.ioapic_handled_vectors)) {
7280 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7281 				vcpu->run->eoi.vector =
7282 						vcpu->arch.pending_ioapic_eoi;
7283 				r = 0;
7284 				goto out;
7285 			}
7286 		}
7287 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7288 			vcpu_scan_ioapic(vcpu);
7289 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7290 			vcpu_load_eoi_exitmap(vcpu);
7291 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7292 			kvm_vcpu_reload_apic_access_page(vcpu);
7293 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7294 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7295 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7296 			r = 0;
7297 			goto out;
7298 		}
7299 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7300 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7301 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7302 			r = 0;
7303 			goto out;
7304 		}
7305 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7306 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7307 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7308 			r = 0;
7309 			goto out;
7310 		}
7311 
7312 		/*
7313 		 * KVM_REQ_HV_STIMER has to be processed after
7314 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7315 		 * depend on the guest clock being up-to-date
7316 		 */
7317 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7318 			kvm_hv_process_stimers(vcpu);
7319 	}
7320 
7321 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7322 		++vcpu->stat.req_event;
7323 		kvm_apic_accept_events(vcpu);
7324 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7325 			r = 1;
7326 			goto out;
7327 		}
7328 
7329 		if (inject_pending_event(vcpu, req_int_win) != 0)
7330 			req_immediate_exit = true;
7331 		else {
7332 			/* Enable SMI/NMI/IRQ window open exits if needed.
7333 			 *
7334 			 * SMIs have three cases:
7335 			 * 1) They can be nested, and then there is nothing to
7336 			 *    do here because RSM will cause a vmexit anyway.
7337 			 * 2) There is an ISA-specific reason why SMI cannot be
7338 			 *    injected, and the moment when this changes can be
7339 			 *    intercepted.
7340 			 * 3) Or the SMI can be pending because
7341 			 *    inject_pending_event has completed the injection
7342 			 *    of an IRQ or NMI from the previous vmexit, and
7343 			 *    then we request an immediate exit to inject the
7344 			 *    SMI.
7345 			 */
7346 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7347 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7348 					req_immediate_exit = true;
7349 			if (vcpu->arch.nmi_pending)
7350 				kvm_x86_ops->enable_nmi_window(vcpu);
7351 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7352 				kvm_x86_ops->enable_irq_window(vcpu);
7353 			WARN_ON(vcpu->arch.exception.pending);
7354 		}
7355 
7356 		if (kvm_lapic_enabled(vcpu)) {
7357 			update_cr8_intercept(vcpu);
7358 			kvm_lapic_sync_to_vapic(vcpu);
7359 		}
7360 	}
7361 
7362 	r = kvm_mmu_reload(vcpu);
7363 	if (unlikely(r)) {
7364 		goto cancel_injection;
7365 	}
7366 
7367 	preempt_disable();
7368 
7369 	kvm_x86_ops->prepare_guest_switch(vcpu);
7370 
7371 	/*
7372 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7373 	 * IPI are then delayed after guest entry, which ensures that they
7374 	 * result in virtual interrupt delivery.
7375 	 */
7376 	local_irq_disable();
7377 	vcpu->mode = IN_GUEST_MODE;
7378 
7379 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7380 
7381 	/*
7382 	 * 1) We should set ->mode before checking ->requests.  Please see
7383 	 * the comment in kvm_vcpu_exiting_guest_mode().
7384 	 *
7385 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
7386 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7387 	 * (see vmx_deliver_posted_interrupt).
7388 	 *
7389 	 * 3) This also orders the write to mode from any reads to the page
7390 	 * tables done while the VCPU is running.  Please see the comment
7391 	 * in kvm_flush_remote_tlbs.
7392 	 */
7393 	smp_mb__after_srcu_read_unlock();
7394 
7395 	/*
7396 	 * This handles the case where a posted interrupt was
7397 	 * notified with kvm_vcpu_kick.
7398 	 */
7399 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7400 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7401 
7402 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7403 	    || need_resched() || signal_pending(current)) {
7404 		vcpu->mode = OUTSIDE_GUEST_MODE;
7405 		smp_wmb();
7406 		local_irq_enable();
7407 		preempt_enable();
7408 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7409 		r = 1;
7410 		goto cancel_injection;
7411 	}
7412 
7413 	kvm_load_guest_xcr0(vcpu);
7414 
7415 	if (req_immediate_exit) {
7416 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7417 		smp_send_reschedule(vcpu->cpu);
7418 	}
7419 
7420 	trace_kvm_entry(vcpu->vcpu_id);
7421 	if (lapic_timer_advance_ns)
7422 		wait_lapic_expire(vcpu);
7423 	guest_enter_irqoff();
7424 
7425 	if (unlikely(vcpu->arch.switch_db_regs)) {
7426 		set_debugreg(0, 7);
7427 		set_debugreg(vcpu->arch.eff_db[0], 0);
7428 		set_debugreg(vcpu->arch.eff_db[1], 1);
7429 		set_debugreg(vcpu->arch.eff_db[2], 2);
7430 		set_debugreg(vcpu->arch.eff_db[3], 3);
7431 		set_debugreg(vcpu->arch.dr6, 6);
7432 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7433 	}
7434 
7435 	kvm_x86_ops->run(vcpu);
7436 
7437 	/*
7438 	 * Do this here before restoring debug registers on the host.  And
7439 	 * since we do this before handling the vmexit, a DR access vmexit
7440 	 * can (a) read the correct value of the debug registers, (b) set
7441 	 * KVM_DEBUGREG_WONT_EXIT again.
7442 	 */
7443 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7444 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7445 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7446 		kvm_update_dr0123(vcpu);
7447 		kvm_update_dr6(vcpu);
7448 		kvm_update_dr7(vcpu);
7449 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7450 	}
7451 
7452 	/*
7453 	 * If the guest has used debug registers, at least dr7
7454 	 * will be disabled while returning to the host.
7455 	 * If we don't have active breakpoints in the host, we don't
7456 	 * care about the messed up debug address registers. But if
7457 	 * we have some of them active, restore the old state.
7458 	 */
7459 	if (hw_breakpoint_active())
7460 		hw_breakpoint_restore();
7461 
7462 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7463 
7464 	vcpu->mode = OUTSIDE_GUEST_MODE;
7465 	smp_wmb();
7466 
7467 	kvm_put_guest_xcr0(vcpu);
7468 
7469 	kvm_before_interrupt(vcpu);
7470 	kvm_x86_ops->handle_external_intr(vcpu);
7471 	kvm_after_interrupt(vcpu);
7472 
7473 	++vcpu->stat.exits;
7474 
7475 	guest_exit_irqoff();
7476 
7477 	local_irq_enable();
7478 	preempt_enable();
7479 
7480 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7481 
7482 	/*
7483 	 * Profile KVM exit RIPs:
7484 	 */
7485 	if (unlikely(prof_on == KVM_PROFILING)) {
7486 		unsigned long rip = kvm_rip_read(vcpu);
7487 		profile_hit(KVM_PROFILING, (void *)rip);
7488 	}
7489 
7490 	if (unlikely(vcpu->arch.tsc_always_catchup))
7491 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7492 
7493 	if (vcpu->arch.apic_attention)
7494 		kvm_lapic_sync_from_vapic(vcpu);
7495 
7496 	vcpu->arch.gpa_available = false;
7497 	r = kvm_x86_ops->handle_exit(vcpu);
7498 	return r;
7499 
7500 cancel_injection:
7501 	kvm_x86_ops->cancel_injection(vcpu);
7502 	if (unlikely(vcpu->arch.apic_attention))
7503 		kvm_lapic_sync_from_vapic(vcpu);
7504 out:
7505 	return r;
7506 }
7507 
7508 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7509 {
7510 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7511 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7512 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7513 		kvm_vcpu_block(vcpu);
7514 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7515 
7516 		if (kvm_x86_ops->post_block)
7517 			kvm_x86_ops->post_block(vcpu);
7518 
7519 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7520 			return 1;
7521 	}
7522 
7523 	kvm_apic_accept_events(vcpu);
7524 	switch(vcpu->arch.mp_state) {
7525 	case KVM_MP_STATE_HALTED:
7526 		vcpu->arch.pv.pv_unhalted = false;
7527 		vcpu->arch.mp_state =
7528 			KVM_MP_STATE_RUNNABLE;
7529 	case KVM_MP_STATE_RUNNABLE:
7530 		vcpu->arch.apf.halted = false;
7531 		break;
7532 	case KVM_MP_STATE_INIT_RECEIVED:
7533 		break;
7534 	default:
7535 		return -EINTR;
7536 		break;
7537 	}
7538 	return 1;
7539 }
7540 
7541 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7542 {
7543 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7544 		kvm_x86_ops->check_nested_events(vcpu, false);
7545 
7546 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7547 		!vcpu->arch.apf.halted);
7548 }
7549 
7550 static int vcpu_run(struct kvm_vcpu *vcpu)
7551 {
7552 	int r;
7553 	struct kvm *kvm = vcpu->kvm;
7554 
7555 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7556 
7557 	for (;;) {
7558 		if (kvm_vcpu_running(vcpu)) {
7559 			r = vcpu_enter_guest(vcpu);
7560 		} else {
7561 			r = vcpu_block(kvm, vcpu);
7562 		}
7563 
7564 		if (r <= 0)
7565 			break;
7566 
7567 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7568 		if (kvm_cpu_has_pending_timer(vcpu))
7569 			kvm_inject_pending_timer_irqs(vcpu);
7570 
7571 		if (dm_request_for_irq_injection(vcpu) &&
7572 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7573 			r = 0;
7574 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7575 			++vcpu->stat.request_irq_exits;
7576 			break;
7577 		}
7578 
7579 		kvm_check_async_pf_completion(vcpu);
7580 
7581 		if (signal_pending(current)) {
7582 			r = -EINTR;
7583 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7584 			++vcpu->stat.signal_exits;
7585 			break;
7586 		}
7587 		if (need_resched()) {
7588 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7589 			cond_resched();
7590 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7591 		}
7592 	}
7593 
7594 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7595 
7596 	return r;
7597 }
7598 
7599 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7600 {
7601 	int r;
7602 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7603 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7604 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7605 	if (r != EMULATE_DONE)
7606 		return 0;
7607 	return 1;
7608 }
7609 
7610 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7611 {
7612 	BUG_ON(!vcpu->arch.pio.count);
7613 
7614 	return complete_emulated_io(vcpu);
7615 }
7616 
7617 /*
7618  * Implements the following, as a state machine:
7619  *
7620  * read:
7621  *   for each fragment
7622  *     for each mmio piece in the fragment
7623  *       write gpa, len
7624  *       exit
7625  *       copy data
7626  *   execute insn
7627  *
7628  * write:
7629  *   for each fragment
7630  *     for each mmio piece in the fragment
7631  *       write gpa, len
7632  *       copy data
7633  *       exit
7634  */
7635 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7636 {
7637 	struct kvm_run *run = vcpu->run;
7638 	struct kvm_mmio_fragment *frag;
7639 	unsigned len;
7640 
7641 	BUG_ON(!vcpu->mmio_needed);
7642 
7643 	/* Complete previous fragment */
7644 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7645 	len = min(8u, frag->len);
7646 	if (!vcpu->mmio_is_write)
7647 		memcpy(frag->data, run->mmio.data, len);
7648 
7649 	if (frag->len <= 8) {
7650 		/* Switch to the next fragment. */
7651 		frag++;
7652 		vcpu->mmio_cur_fragment++;
7653 	} else {
7654 		/* Go forward to the next mmio piece. */
7655 		frag->data += len;
7656 		frag->gpa += len;
7657 		frag->len -= len;
7658 	}
7659 
7660 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7661 		vcpu->mmio_needed = 0;
7662 
7663 		/* FIXME: return into emulator if single-stepping.  */
7664 		if (vcpu->mmio_is_write)
7665 			return 1;
7666 		vcpu->mmio_read_completed = 1;
7667 		return complete_emulated_io(vcpu);
7668 	}
7669 
7670 	run->exit_reason = KVM_EXIT_MMIO;
7671 	run->mmio.phys_addr = frag->gpa;
7672 	if (vcpu->mmio_is_write)
7673 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7674 	run->mmio.len = min(8u, frag->len);
7675 	run->mmio.is_write = vcpu->mmio_is_write;
7676 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7677 	return 0;
7678 }
7679 
7680 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7681 {
7682 	int r;
7683 
7684 	vcpu_load(vcpu);
7685 	kvm_sigset_activate(vcpu);
7686 	kvm_load_guest_fpu(vcpu);
7687 
7688 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7689 		if (kvm_run->immediate_exit) {
7690 			r = -EINTR;
7691 			goto out;
7692 		}
7693 		kvm_vcpu_block(vcpu);
7694 		kvm_apic_accept_events(vcpu);
7695 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7696 		r = -EAGAIN;
7697 		if (signal_pending(current)) {
7698 			r = -EINTR;
7699 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7700 			++vcpu->stat.signal_exits;
7701 		}
7702 		goto out;
7703 	}
7704 
7705 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
7706 		r = -EINVAL;
7707 		goto out;
7708 	}
7709 
7710 	if (vcpu->run->kvm_dirty_regs) {
7711 		r = sync_regs(vcpu);
7712 		if (r != 0)
7713 			goto out;
7714 	}
7715 
7716 	/* re-sync apic's tpr */
7717 	if (!lapic_in_kernel(vcpu)) {
7718 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7719 			r = -EINVAL;
7720 			goto out;
7721 		}
7722 	}
7723 
7724 	if (unlikely(vcpu->arch.complete_userspace_io)) {
7725 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7726 		vcpu->arch.complete_userspace_io = NULL;
7727 		r = cui(vcpu);
7728 		if (r <= 0)
7729 			goto out;
7730 	} else
7731 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7732 
7733 	if (kvm_run->immediate_exit)
7734 		r = -EINTR;
7735 	else
7736 		r = vcpu_run(vcpu);
7737 
7738 out:
7739 	kvm_put_guest_fpu(vcpu);
7740 	if (vcpu->run->kvm_valid_regs)
7741 		store_regs(vcpu);
7742 	post_kvm_run_save(vcpu);
7743 	kvm_sigset_deactivate(vcpu);
7744 
7745 	vcpu_put(vcpu);
7746 	return r;
7747 }
7748 
7749 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7750 {
7751 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7752 		/*
7753 		 * We are here if userspace calls get_regs() in the middle of
7754 		 * instruction emulation. Registers state needs to be copied
7755 		 * back from emulation context to vcpu. Userspace shouldn't do
7756 		 * that usually, but some bad designed PV devices (vmware
7757 		 * backdoor interface) need this to work
7758 		 */
7759 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7760 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7761 	}
7762 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7763 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7764 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7765 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7766 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7767 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7768 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7769 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7770 #ifdef CONFIG_X86_64
7771 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7772 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7773 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7774 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7775 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7776 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7777 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7778 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7779 #endif
7780 
7781 	regs->rip = kvm_rip_read(vcpu);
7782 	regs->rflags = kvm_get_rflags(vcpu);
7783 }
7784 
7785 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7786 {
7787 	vcpu_load(vcpu);
7788 	__get_regs(vcpu, regs);
7789 	vcpu_put(vcpu);
7790 	return 0;
7791 }
7792 
7793 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7794 {
7795 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7796 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7797 
7798 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7799 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7800 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7801 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7802 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7803 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7804 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7805 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7806 #ifdef CONFIG_X86_64
7807 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7808 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7809 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7810 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7811 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7812 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7813 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7814 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7815 #endif
7816 
7817 	kvm_rip_write(vcpu, regs->rip);
7818 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7819 
7820 	vcpu->arch.exception.pending = false;
7821 
7822 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7823 }
7824 
7825 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7826 {
7827 	vcpu_load(vcpu);
7828 	__set_regs(vcpu, regs);
7829 	vcpu_put(vcpu);
7830 	return 0;
7831 }
7832 
7833 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7834 {
7835 	struct kvm_segment cs;
7836 
7837 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7838 	*db = cs.db;
7839 	*l = cs.l;
7840 }
7841 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7842 
7843 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7844 {
7845 	struct desc_ptr dt;
7846 
7847 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7848 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7849 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7850 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7851 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7852 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7853 
7854 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7855 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7856 
7857 	kvm_x86_ops->get_idt(vcpu, &dt);
7858 	sregs->idt.limit = dt.size;
7859 	sregs->idt.base = dt.address;
7860 	kvm_x86_ops->get_gdt(vcpu, &dt);
7861 	sregs->gdt.limit = dt.size;
7862 	sregs->gdt.base = dt.address;
7863 
7864 	sregs->cr0 = kvm_read_cr0(vcpu);
7865 	sregs->cr2 = vcpu->arch.cr2;
7866 	sregs->cr3 = kvm_read_cr3(vcpu);
7867 	sregs->cr4 = kvm_read_cr4(vcpu);
7868 	sregs->cr8 = kvm_get_cr8(vcpu);
7869 	sregs->efer = vcpu->arch.efer;
7870 	sregs->apic_base = kvm_get_apic_base(vcpu);
7871 
7872 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7873 
7874 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
7875 		set_bit(vcpu->arch.interrupt.nr,
7876 			(unsigned long *)sregs->interrupt_bitmap);
7877 }
7878 
7879 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7880 				  struct kvm_sregs *sregs)
7881 {
7882 	vcpu_load(vcpu);
7883 	__get_sregs(vcpu, sregs);
7884 	vcpu_put(vcpu);
7885 	return 0;
7886 }
7887 
7888 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7889 				    struct kvm_mp_state *mp_state)
7890 {
7891 	vcpu_load(vcpu);
7892 
7893 	kvm_apic_accept_events(vcpu);
7894 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7895 					vcpu->arch.pv.pv_unhalted)
7896 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7897 	else
7898 		mp_state->mp_state = vcpu->arch.mp_state;
7899 
7900 	vcpu_put(vcpu);
7901 	return 0;
7902 }
7903 
7904 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7905 				    struct kvm_mp_state *mp_state)
7906 {
7907 	int ret = -EINVAL;
7908 
7909 	vcpu_load(vcpu);
7910 
7911 	if (!lapic_in_kernel(vcpu) &&
7912 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7913 		goto out;
7914 
7915 	/* INITs are latched while in SMM */
7916 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7917 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7918 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7919 		goto out;
7920 
7921 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7922 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7923 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7924 	} else
7925 		vcpu->arch.mp_state = mp_state->mp_state;
7926 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7927 
7928 	ret = 0;
7929 out:
7930 	vcpu_put(vcpu);
7931 	return ret;
7932 }
7933 
7934 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7935 		    int reason, bool has_error_code, u32 error_code)
7936 {
7937 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7938 	int ret;
7939 
7940 	init_emulate_ctxt(vcpu);
7941 
7942 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7943 				   has_error_code, error_code);
7944 
7945 	if (ret)
7946 		return EMULATE_FAIL;
7947 
7948 	kvm_rip_write(vcpu, ctxt->eip);
7949 	kvm_set_rflags(vcpu, ctxt->eflags);
7950 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7951 	return EMULATE_DONE;
7952 }
7953 EXPORT_SYMBOL_GPL(kvm_task_switch);
7954 
7955 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7956 {
7957 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
7958 		/*
7959 		 * When EFER.LME and CR0.PG are set, the processor is in
7960 		 * 64-bit mode (though maybe in a 32-bit code segment).
7961 		 * CR4.PAE and EFER.LMA must be set.
7962 		 */
7963 		if (!(sregs->cr4 & X86_CR4_PAE)
7964 		    || !(sregs->efer & EFER_LMA))
7965 			return -EINVAL;
7966 	} else {
7967 		/*
7968 		 * Not in 64-bit mode: EFER.LMA is clear and the code
7969 		 * segment cannot be 64-bit.
7970 		 */
7971 		if (sregs->efer & EFER_LMA || sregs->cs.l)
7972 			return -EINVAL;
7973 	}
7974 
7975 	return 0;
7976 }
7977 
7978 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7979 {
7980 	struct msr_data apic_base_msr;
7981 	int mmu_reset_needed = 0;
7982 	int pending_vec, max_bits, idx;
7983 	struct desc_ptr dt;
7984 	int ret = -EINVAL;
7985 
7986 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7987 			(sregs->cr4 & X86_CR4_OSXSAVE))
7988 		goto out;
7989 
7990 	if (kvm_valid_sregs(vcpu, sregs))
7991 		goto out;
7992 
7993 	apic_base_msr.data = sregs->apic_base;
7994 	apic_base_msr.host_initiated = true;
7995 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
7996 		goto out;
7997 
7998 	dt.size = sregs->idt.limit;
7999 	dt.address = sregs->idt.base;
8000 	kvm_x86_ops->set_idt(vcpu, &dt);
8001 	dt.size = sregs->gdt.limit;
8002 	dt.address = sregs->gdt.base;
8003 	kvm_x86_ops->set_gdt(vcpu, &dt);
8004 
8005 	vcpu->arch.cr2 = sregs->cr2;
8006 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8007 	vcpu->arch.cr3 = sregs->cr3;
8008 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8009 
8010 	kvm_set_cr8(vcpu, sregs->cr8);
8011 
8012 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8013 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8014 
8015 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8016 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8017 	vcpu->arch.cr0 = sregs->cr0;
8018 
8019 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8020 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8021 	if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
8022 		kvm_update_cpuid(vcpu);
8023 
8024 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8025 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
8026 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8027 		mmu_reset_needed = 1;
8028 	}
8029 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8030 
8031 	if (mmu_reset_needed)
8032 		kvm_mmu_reset_context(vcpu);
8033 
8034 	max_bits = KVM_NR_INTERRUPTS;
8035 	pending_vec = find_first_bit(
8036 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8037 	if (pending_vec < max_bits) {
8038 		kvm_queue_interrupt(vcpu, pending_vec, false);
8039 		pr_debug("Set back pending irq %d\n", pending_vec);
8040 	}
8041 
8042 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8043 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8044 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8045 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8046 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8047 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8048 
8049 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8050 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8051 
8052 	update_cr8_intercept(vcpu);
8053 
8054 	/* Older userspace won't unhalt the vcpu on reset. */
8055 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8056 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8057 	    !is_protmode(vcpu))
8058 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8059 
8060 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8061 
8062 	ret = 0;
8063 out:
8064 	return ret;
8065 }
8066 
8067 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8068 				  struct kvm_sregs *sregs)
8069 {
8070 	int ret;
8071 
8072 	vcpu_load(vcpu);
8073 	ret = __set_sregs(vcpu, sregs);
8074 	vcpu_put(vcpu);
8075 	return ret;
8076 }
8077 
8078 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8079 					struct kvm_guest_debug *dbg)
8080 {
8081 	unsigned long rflags;
8082 	int i, r;
8083 
8084 	vcpu_load(vcpu);
8085 
8086 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8087 		r = -EBUSY;
8088 		if (vcpu->arch.exception.pending)
8089 			goto out;
8090 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8091 			kvm_queue_exception(vcpu, DB_VECTOR);
8092 		else
8093 			kvm_queue_exception(vcpu, BP_VECTOR);
8094 	}
8095 
8096 	/*
8097 	 * Read rflags as long as potentially injected trace flags are still
8098 	 * filtered out.
8099 	 */
8100 	rflags = kvm_get_rflags(vcpu);
8101 
8102 	vcpu->guest_debug = dbg->control;
8103 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8104 		vcpu->guest_debug = 0;
8105 
8106 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8107 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8108 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8109 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8110 	} else {
8111 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8112 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8113 	}
8114 	kvm_update_dr7(vcpu);
8115 
8116 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8117 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8118 			get_segment_base(vcpu, VCPU_SREG_CS);
8119 
8120 	/*
8121 	 * Trigger an rflags update that will inject or remove the trace
8122 	 * flags.
8123 	 */
8124 	kvm_set_rflags(vcpu, rflags);
8125 
8126 	kvm_x86_ops->update_bp_intercept(vcpu);
8127 
8128 	r = 0;
8129 
8130 out:
8131 	vcpu_put(vcpu);
8132 	return r;
8133 }
8134 
8135 /*
8136  * Translate a guest virtual address to a guest physical address.
8137  */
8138 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8139 				    struct kvm_translation *tr)
8140 {
8141 	unsigned long vaddr = tr->linear_address;
8142 	gpa_t gpa;
8143 	int idx;
8144 
8145 	vcpu_load(vcpu);
8146 
8147 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8148 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8149 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8150 	tr->physical_address = gpa;
8151 	tr->valid = gpa != UNMAPPED_GVA;
8152 	tr->writeable = 1;
8153 	tr->usermode = 0;
8154 
8155 	vcpu_put(vcpu);
8156 	return 0;
8157 }
8158 
8159 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8160 {
8161 	struct fxregs_state *fxsave;
8162 
8163 	vcpu_load(vcpu);
8164 
8165 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8166 	memcpy(fpu->fpr, fxsave->st_space, 128);
8167 	fpu->fcw = fxsave->cwd;
8168 	fpu->fsw = fxsave->swd;
8169 	fpu->ftwx = fxsave->twd;
8170 	fpu->last_opcode = fxsave->fop;
8171 	fpu->last_ip = fxsave->rip;
8172 	fpu->last_dp = fxsave->rdp;
8173 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
8174 
8175 	vcpu_put(vcpu);
8176 	return 0;
8177 }
8178 
8179 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8180 {
8181 	struct fxregs_state *fxsave;
8182 
8183 	vcpu_load(vcpu);
8184 
8185 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8186 
8187 	memcpy(fxsave->st_space, fpu->fpr, 128);
8188 	fxsave->cwd = fpu->fcw;
8189 	fxsave->swd = fpu->fsw;
8190 	fxsave->twd = fpu->ftwx;
8191 	fxsave->fop = fpu->last_opcode;
8192 	fxsave->rip = fpu->last_ip;
8193 	fxsave->rdp = fpu->last_dp;
8194 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
8195 
8196 	vcpu_put(vcpu);
8197 	return 0;
8198 }
8199 
8200 static void store_regs(struct kvm_vcpu *vcpu)
8201 {
8202 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8203 
8204 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8205 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8206 
8207 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8208 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8209 
8210 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8211 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8212 				vcpu, &vcpu->run->s.regs.events);
8213 }
8214 
8215 static int sync_regs(struct kvm_vcpu *vcpu)
8216 {
8217 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8218 		return -EINVAL;
8219 
8220 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8221 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8222 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8223 	}
8224 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8225 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8226 			return -EINVAL;
8227 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8228 	}
8229 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8230 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8231 				vcpu, &vcpu->run->s.regs.events))
8232 			return -EINVAL;
8233 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8234 	}
8235 
8236 	return 0;
8237 }
8238 
8239 static void fx_init(struct kvm_vcpu *vcpu)
8240 {
8241 	fpstate_init(&vcpu->arch.guest_fpu.state);
8242 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8243 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8244 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8245 
8246 	/*
8247 	 * Ensure guest xcr0 is valid for loading
8248 	 */
8249 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8250 
8251 	vcpu->arch.cr0 |= X86_CR0_ET;
8252 }
8253 
8254 /* Swap (qemu) user FPU context for the guest FPU context. */
8255 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8256 {
8257 	preempt_disable();
8258 	copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
8259 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8260 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8261 				~XFEATURE_MASK_PKRU);
8262 	preempt_enable();
8263 	trace_kvm_fpu(1);
8264 }
8265 
8266 /* When vcpu_run ends, restore user space FPU context. */
8267 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8268 {
8269 	preempt_disable();
8270 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8271 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8272 	preempt_enable();
8273 	++vcpu->stat.fpu_reload;
8274 	trace_kvm_fpu(0);
8275 }
8276 
8277 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8278 {
8279 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8280 
8281 	kvmclock_reset(vcpu);
8282 
8283 	kvm_x86_ops->vcpu_free(vcpu);
8284 	free_cpumask_var(wbinvd_dirty_mask);
8285 }
8286 
8287 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8288 						unsigned int id)
8289 {
8290 	struct kvm_vcpu *vcpu;
8291 
8292 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8293 		printk_once(KERN_WARNING
8294 		"kvm: SMP vm created on host with unstable TSC; "
8295 		"guest TSC will not be reliable\n");
8296 
8297 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8298 
8299 	return vcpu;
8300 }
8301 
8302 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8303 {
8304 	kvm_vcpu_mtrr_init(vcpu);
8305 	vcpu_load(vcpu);
8306 	kvm_vcpu_reset(vcpu, false);
8307 	kvm_mmu_setup(vcpu);
8308 	vcpu_put(vcpu);
8309 	return 0;
8310 }
8311 
8312 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8313 {
8314 	struct msr_data msr;
8315 	struct kvm *kvm = vcpu->kvm;
8316 
8317 	kvm_hv_vcpu_postcreate(vcpu);
8318 
8319 	if (mutex_lock_killable(&vcpu->mutex))
8320 		return;
8321 	vcpu_load(vcpu);
8322 	msr.data = 0x0;
8323 	msr.index = MSR_IA32_TSC;
8324 	msr.host_initiated = true;
8325 	kvm_write_tsc(vcpu, &msr);
8326 	vcpu_put(vcpu);
8327 	mutex_unlock(&vcpu->mutex);
8328 
8329 	if (!kvmclock_periodic_sync)
8330 		return;
8331 
8332 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8333 					KVMCLOCK_SYNC_PERIOD);
8334 }
8335 
8336 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8337 {
8338 	vcpu->arch.apf.msr_val = 0;
8339 
8340 	vcpu_load(vcpu);
8341 	kvm_mmu_unload(vcpu);
8342 	vcpu_put(vcpu);
8343 
8344 	kvm_x86_ops->vcpu_free(vcpu);
8345 }
8346 
8347 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8348 {
8349 	kvm_lapic_reset(vcpu, init_event);
8350 
8351 	vcpu->arch.hflags = 0;
8352 
8353 	vcpu->arch.smi_pending = 0;
8354 	vcpu->arch.smi_count = 0;
8355 	atomic_set(&vcpu->arch.nmi_queued, 0);
8356 	vcpu->arch.nmi_pending = 0;
8357 	vcpu->arch.nmi_injected = false;
8358 	kvm_clear_interrupt_queue(vcpu);
8359 	kvm_clear_exception_queue(vcpu);
8360 	vcpu->arch.exception.pending = false;
8361 
8362 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8363 	kvm_update_dr0123(vcpu);
8364 	vcpu->arch.dr6 = DR6_INIT;
8365 	kvm_update_dr6(vcpu);
8366 	vcpu->arch.dr7 = DR7_FIXED_1;
8367 	kvm_update_dr7(vcpu);
8368 
8369 	vcpu->arch.cr2 = 0;
8370 
8371 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8372 	vcpu->arch.apf.msr_val = 0;
8373 	vcpu->arch.st.msr_val = 0;
8374 
8375 	kvmclock_reset(vcpu);
8376 
8377 	kvm_clear_async_pf_completion_queue(vcpu);
8378 	kvm_async_pf_hash_reset(vcpu);
8379 	vcpu->arch.apf.halted = false;
8380 
8381 	if (kvm_mpx_supported()) {
8382 		void *mpx_state_buffer;
8383 
8384 		/*
8385 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8386 		 * called with loaded FPU and does not let userspace fix the state.
8387 		 */
8388 		if (init_event)
8389 			kvm_put_guest_fpu(vcpu);
8390 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8391 					XFEATURE_MASK_BNDREGS);
8392 		if (mpx_state_buffer)
8393 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8394 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8395 					XFEATURE_MASK_BNDCSR);
8396 		if (mpx_state_buffer)
8397 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8398 		if (init_event)
8399 			kvm_load_guest_fpu(vcpu);
8400 	}
8401 
8402 	if (!init_event) {
8403 		kvm_pmu_reset(vcpu);
8404 		vcpu->arch.smbase = 0x30000;
8405 
8406 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8407 		vcpu->arch.msr_misc_features_enables = 0;
8408 
8409 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8410 	}
8411 
8412 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8413 	vcpu->arch.regs_avail = ~0;
8414 	vcpu->arch.regs_dirty = ~0;
8415 
8416 	vcpu->arch.ia32_xss = 0;
8417 
8418 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8419 }
8420 
8421 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8422 {
8423 	struct kvm_segment cs;
8424 
8425 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8426 	cs.selector = vector << 8;
8427 	cs.base = vector << 12;
8428 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8429 	kvm_rip_write(vcpu, 0);
8430 }
8431 
8432 int kvm_arch_hardware_enable(void)
8433 {
8434 	struct kvm *kvm;
8435 	struct kvm_vcpu *vcpu;
8436 	int i;
8437 	int ret;
8438 	u64 local_tsc;
8439 	u64 max_tsc = 0;
8440 	bool stable, backwards_tsc = false;
8441 
8442 	kvm_shared_msr_cpu_online();
8443 	ret = kvm_x86_ops->hardware_enable();
8444 	if (ret != 0)
8445 		return ret;
8446 
8447 	local_tsc = rdtsc();
8448 	stable = !kvm_check_tsc_unstable();
8449 	list_for_each_entry(kvm, &vm_list, vm_list) {
8450 		kvm_for_each_vcpu(i, vcpu, kvm) {
8451 			if (!stable && vcpu->cpu == smp_processor_id())
8452 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8453 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8454 				backwards_tsc = true;
8455 				if (vcpu->arch.last_host_tsc > max_tsc)
8456 					max_tsc = vcpu->arch.last_host_tsc;
8457 			}
8458 		}
8459 	}
8460 
8461 	/*
8462 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8463 	 * platforms that reset TSC during suspend or hibernate actions, but
8464 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8465 	 * detect that condition here, which happens early in CPU bringup,
8466 	 * before any KVM threads can be running.  Unfortunately, we can't
8467 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8468 	 * enough into CPU bringup that we know how much real time has actually
8469 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8470 	 * variables that haven't been updated yet.
8471 	 *
8472 	 * So we simply find the maximum observed TSC above, then record the
8473 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8474 	 * the adjustment will be applied.  Note that we accumulate
8475 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8476 	 * gets a chance to run again.  In the event that no KVM threads get a
8477 	 * chance to run, we will miss the entire elapsed period, as we'll have
8478 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8479 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8480 	 * uniform across all VCPUs (not to mention the scenario is extremely
8481 	 * unlikely). It is possible that a second hibernate recovery happens
8482 	 * much faster than a first, causing the observed TSC here to be
8483 	 * smaller; this would require additional padding adjustment, which is
8484 	 * why we set last_host_tsc to the local tsc observed here.
8485 	 *
8486 	 * N.B. - this code below runs only on platforms with reliable TSC,
8487 	 * as that is the only way backwards_tsc is set above.  Also note
8488 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8489 	 * have the same delta_cyc adjustment applied if backwards_tsc
8490 	 * is detected.  Note further, this adjustment is only done once,
8491 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8492 	 * called multiple times (one for each physical CPU bringup).
8493 	 *
8494 	 * Platforms with unreliable TSCs don't have to deal with this, they
8495 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8496 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8497 	 * guarantee that they stay in perfect synchronization.
8498 	 */
8499 	if (backwards_tsc) {
8500 		u64 delta_cyc = max_tsc - local_tsc;
8501 		list_for_each_entry(kvm, &vm_list, vm_list) {
8502 			kvm->arch.backwards_tsc_observed = true;
8503 			kvm_for_each_vcpu(i, vcpu, kvm) {
8504 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8505 				vcpu->arch.last_host_tsc = local_tsc;
8506 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8507 			}
8508 
8509 			/*
8510 			 * We have to disable TSC offset matching.. if you were
8511 			 * booting a VM while issuing an S4 host suspend....
8512 			 * you may have some problem.  Solving this issue is
8513 			 * left as an exercise to the reader.
8514 			 */
8515 			kvm->arch.last_tsc_nsec = 0;
8516 			kvm->arch.last_tsc_write = 0;
8517 		}
8518 
8519 	}
8520 	return 0;
8521 }
8522 
8523 void kvm_arch_hardware_disable(void)
8524 {
8525 	kvm_x86_ops->hardware_disable();
8526 	drop_user_return_notifiers();
8527 }
8528 
8529 int kvm_arch_hardware_setup(void)
8530 {
8531 	int r;
8532 
8533 	r = kvm_x86_ops->hardware_setup();
8534 	if (r != 0)
8535 		return r;
8536 
8537 	if (kvm_has_tsc_control) {
8538 		/*
8539 		 * Make sure the user can only configure tsc_khz values that
8540 		 * fit into a signed integer.
8541 		 * A min value is not calculated needed because it will always
8542 		 * be 1 on all machines.
8543 		 */
8544 		u64 max = min(0x7fffffffULL,
8545 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8546 		kvm_max_guest_tsc_khz = max;
8547 
8548 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8549 	}
8550 
8551 	kvm_init_msr_list();
8552 	return 0;
8553 }
8554 
8555 void kvm_arch_hardware_unsetup(void)
8556 {
8557 	kvm_x86_ops->hardware_unsetup();
8558 }
8559 
8560 void kvm_arch_check_processor_compat(void *rtn)
8561 {
8562 	kvm_x86_ops->check_processor_compatibility(rtn);
8563 }
8564 
8565 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8566 {
8567 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8568 }
8569 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8570 
8571 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8572 {
8573 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8574 }
8575 
8576 struct static_key kvm_no_apic_vcpu __read_mostly;
8577 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8578 
8579 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8580 {
8581 	struct page *page;
8582 	int r;
8583 
8584 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8585 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8586 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8587 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8588 	else
8589 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8590 
8591 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8592 	if (!page) {
8593 		r = -ENOMEM;
8594 		goto fail;
8595 	}
8596 	vcpu->arch.pio_data = page_address(page);
8597 
8598 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
8599 
8600 	r = kvm_mmu_create(vcpu);
8601 	if (r < 0)
8602 		goto fail_free_pio_data;
8603 
8604 	if (irqchip_in_kernel(vcpu->kvm)) {
8605 		r = kvm_create_lapic(vcpu);
8606 		if (r < 0)
8607 			goto fail_mmu_destroy;
8608 	} else
8609 		static_key_slow_inc(&kvm_no_apic_vcpu);
8610 
8611 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8612 				       GFP_KERNEL);
8613 	if (!vcpu->arch.mce_banks) {
8614 		r = -ENOMEM;
8615 		goto fail_free_lapic;
8616 	}
8617 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8618 
8619 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8620 		r = -ENOMEM;
8621 		goto fail_free_mce_banks;
8622 	}
8623 
8624 	fx_init(vcpu);
8625 
8626 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8627 
8628 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8629 
8630 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8631 
8632 	kvm_async_pf_hash_reset(vcpu);
8633 	kvm_pmu_init(vcpu);
8634 
8635 	vcpu->arch.pending_external_vector = -1;
8636 	vcpu->arch.preempted_in_kernel = false;
8637 
8638 	kvm_hv_vcpu_init(vcpu);
8639 
8640 	return 0;
8641 
8642 fail_free_mce_banks:
8643 	kfree(vcpu->arch.mce_banks);
8644 fail_free_lapic:
8645 	kvm_free_lapic(vcpu);
8646 fail_mmu_destroy:
8647 	kvm_mmu_destroy(vcpu);
8648 fail_free_pio_data:
8649 	free_page((unsigned long)vcpu->arch.pio_data);
8650 fail:
8651 	return r;
8652 }
8653 
8654 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8655 {
8656 	int idx;
8657 
8658 	kvm_hv_vcpu_uninit(vcpu);
8659 	kvm_pmu_destroy(vcpu);
8660 	kfree(vcpu->arch.mce_banks);
8661 	kvm_free_lapic(vcpu);
8662 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8663 	kvm_mmu_destroy(vcpu);
8664 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8665 	free_page((unsigned long)vcpu->arch.pio_data);
8666 	if (!lapic_in_kernel(vcpu))
8667 		static_key_slow_dec(&kvm_no_apic_vcpu);
8668 }
8669 
8670 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8671 {
8672 	kvm_x86_ops->sched_in(vcpu, cpu);
8673 }
8674 
8675 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8676 {
8677 	if (type)
8678 		return -EINVAL;
8679 
8680 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8681 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8682 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8683 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8684 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8685 
8686 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8687 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8688 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8689 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8690 		&kvm->arch.irq_sources_bitmap);
8691 
8692 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8693 	mutex_init(&kvm->arch.apic_map_lock);
8694 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8695 
8696 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8697 	pvclock_update_vm_gtod_copy(kvm);
8698 
8699 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8700 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8701 
8702 	kvm_hv_init_vm(kvm);
8703 	kvm_page_track_init(kvm);
8704 	kvm_mmu_init_vm(kvm);
8705 
8706 	if (kvm_x86_ops->vm_init)
8707 		return kvm_x86_ops->vm_init(kvm);
8708 
8709 	return 0;
8710 }
8711 
8712 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8713 {
8714 	vcpu_load(vcpu);
8715 	kvm_mmu_unload(vcpu);
8716 	vcpu_put(vcpu);
8717 }
8718 
8719 static void kvm_free_vcpus(struct kvm *kvm)
8720 {
8721 	unsigned int i;
8722 	struct kvm_vcpu *vcpu;
8723 
8724 	/*
8725 	 * Unpin any mmu pages first.
8726 	 */
8727 	kvm_for_each_vcpu(i, vcpu, kvm) {
8728 		kvm_clear_async_pf_completion_queue(vcpu);
8729 		kvm_unload_vcpu_mmu(vcpu);
8730 	}
8731 	kvm_for_each_vcpu(i, vcpu, kvm)
8732 		kvm_arch_vcpu_free(vcpu);
8733 
8734 	mutex_lock(&kvm->lock);
8735 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8736 		kvm->vcpus[i] = NULL;
8737 
8738 	atomic_set(&kvm->online_vcpus, 0);
8739 	mutex_unlock(&kvm->lock);
8740 }
8741 
8742 void kvm_arch_sync_events(struct kvm *kvm)
8743 {
8744 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8745 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8746 	kvm_free_pit(kvm);
8747 }
8748 
8749 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8750 {
8751 	int i, r;
8752 	unsigned long hva;
8753 	struct kvm_memslots *slots = kvm_memslots(kvm);
8754 	struct kvm_memory_slot *slot, old;
8755 
8756 	/* Called with kvm->slots_lock held.  */
8757 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8758 		return -EINVAL;
8759 
8760 	slot = id_to_memslot(slots, id);
8761 	if (size) {
8762 		if (slot->npages)
8763 			return -EEXIST;
8764 
8765 		/*
8766 		 * MAP_SHARED to prevent internal slot pages from being moved
8767 		 * by fork()/COW.
8768 		 */
8769 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8770 			      MAP_SHARED | MAP_ANONYMOUS, 0);
8771 		if (IS_ERR((void *)hva))
8772 			return PTR_ERR((void *)hva);
8773 	} else {
8774 		if (!slot->npages)
8775 			return 0;
8776 
8777 		hva = 0;
8778 	}
8779 
8780 	old = *slot;
8781 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8782 		struct kvm_userspace_memory_region m;
8783 
8784 		m.slot = id | (i << 16);
8785 		m.flags = 0;
8786 		m.guest_phys_addr = gpa;
8787 		m.userspace_addr = hva;
8788 		m.memory_size = size;
8789 		r = __kvm_set_memory_region(kvm, &m);
8790 		if (r < 0)
8791 			return r;
8792 	}
8793 
8794 	if (!size)
8795 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8796 
8797 	return 0;
8798 }
8799 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8800 
8801 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8802 {
8803 	int r;
8804 
8805 	mutex_lock(&kvm->slots_lock);
8806 	r = __x86_set_memory_region(kvm, id, gpa, size);
8807 	mutex_unlock(&kvm->slots_lock);
8808 
8809 	return r;
8810 }
8811 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8812 
8813 void kvm_arch_destroy_vm(struct kvm *kvm)
8814 {
8815 	if (current->mm == kvm->mm) {
8816 		/*
8817 		 * Free memory regions allocated on behalf of userspace,
8818 		 * unless the the memory map has changed due to process exit
8819 		 * or fd copying.
8820 		 */
8821 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8822 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8823 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8824 	}
8825 	if (kvm_x86_ops->vm_destroy)
8826 		kvm_x86_ops->vm_destroy(kvm);
8827 	kvm_pic_destroy(kvm);
8828 	kvm_ioapic_destroy(kvm);
8829 	kvm_free_vcpus(kvm);
8830 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8831 	kvm_mmu_uninit_vm(kvm);
8832 	kvm_page_track_cleanup(kvm);
8833 	kvm_hv_destroy_vm(kvm);
8834 }
8835 
8836 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8837 			   struct kvm_memory_slot *dont)
8838 {
8839 	int i;
8840 
8841 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8842 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8843 			kvfree(free->arch.rmap[i]);
8844 			free->arch.rmap[i] = NULL;
8845 		}
8846 		if (i == 0)
8847 			continue;
8848 
8849 		if (!dont || free->arch.lpage_info[i - 1] !=
8850 			     dont->arch.lpage_info[i - 1]) {
8851 			kvfree(free->arch.lpage_info[i - 1]);
8852 			free->arch.lpage_info[i - 1] = NULL;
8853 		}
8854 	}
8855 
8856 	kvm_page_track_free_memslot(free, dont);
8857 }
8858 
8859 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8860 			    unsigned long npages)
8861 {
8862 	int i;
8863 
8864 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8865 		struct kvm_lpage_info *linfo;
8866 		unsigned long ugfn;
8867 		int lpages;
8868 		int level = i + 1;
8869 
8870 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
8871 				      slot->base_gfn, level) + 1;
8872 
8873 		slot->arch.rmap[i] =
8874 			kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8875 		if (!slot->arch.rmap[i])
8876 			goto out_free;
8877 		if (i == 0)
8878 			continue;
8879 
8880 		linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8881 		if (!linfo)
8882 			goto out_free;
8883 
8884 		slot->arch.lpage_info[i - 1] = linfo;
8885 
8886 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8887 			linfo[0].disallow_lpage = 1;
8888 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8889 			linfo[lpages - 1].disallow_lpage = 1;
8890 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
8891 		/*
8892 		 * If the gfn and userspace address are not aligned wrt each
8893 		 * other, or if explicitly asked to, disable large page
8894 		 * support for this slot
8895 		 */
8896 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8897 		    !kvm_largepages_enabled()) {
8898 			unsigned long j;
8899 
8900 			for (j = 0; j < lpages; ++j)
8901 				linfo[j].disallow_lpage = 1;
8902 		}
8903 	}
8904 
8905 	if (kvm_page_track_create_memslot(slot, npages))
8906 		goto out_free;
8907 
8908 	return 0;
8909 
8910 out_free:
8911 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8912 		kvfree(slot->arch.rmap[i]);
8913 		slot->arch.rmap[i] = NULL;
8914 		if (i == 0)
8915 			continue;
8916 
8917 		kvfree(slot->arch.lpage_info[i - 1]);
8918 		slot->arch.lpage_info[i - 1] = NULL;
8919 	}
8920 	return -ENOMEM;
8921 }
8922 
8923 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8924 {
8925 	/*
8926 	 * memslots->generation has been incremented.
8927 	 * mmio generation may have reached its maximum value.
8928 	 */
8929 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8930 }
8931 
8932 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8933 				struct kvm_memory_slot *memslot,
8934 				const struct kvm_userspace_memory_region *mem,
8935 				enum kvm_mr_change change)
8936 {
8937 	return 0;
8938 }
8939 
8940 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8941 				     struct kvm_memory_slot *new)
8942 {
8943 	/* Still write protect RO slot */
8944 	if (new->flags & KVM_MEM_READONLY) {
8945 		kvm_mmu_slot_remove_write_access(kvm, new);
8946 		return;
8947 	}
8948 
8949 	/*
8950 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
8951 	 *
8952 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
8953 	 *
8954 	 *  - KVM_MR_CREATE with dirty logging is disabled
8955 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8956 	 *
8957 	 * The reason is, in case of PML, we need to set D-bit for any slots
8958 	 * with dirty logging disabled in order to eliminate unnecessary GPA
8959 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
8960 	 * guarantees leaving PML enabled during guest's lifetime won't have
8961 	 * any additonal overhead from PML when guest is running with dirty
8962 	 * logging disabled for memory slots.
8963 	 *
8964 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8965 	 * to dirty logging mode.
8966 	 *
8967 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8968 	 *
8969 	 * In case of write protect:
8970 	 *
8971 	 * Write protect all pages for dirty logging.
8972 	 *
8973 	 * All the sptes including the large sptes which point to this
8974 	 * slot are set to readonly. We can not create any new large
8975 	 * spte on this slot until the end of the logging.
8976 	 *
8977 	 * See the comments in fast_page_fault().
8978 	 */
8979 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8980 		if (kvm_x86_ops->slot_enable_log_dirty)
8981 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8982 		else
8983 			kvm_mmu_slot_remove_write_access(kvm, new);
8984 	} else {
8985 		if (kvm_x86_ops->slot_disable_log_dirty)
8986 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8987 	}
8988 }
8989 
8990 void kvm_arch_commit_memory_region(struct kvm *kvm,
8991 				const struct kvm_userspace_memory_region *mem,
8992 				const struct kvm_memory_slot *old,
8993 				const struct kvm_memory_slot *new,
8994 				enum kvm_mr_change change)
8995 {
8996 	int nr_mmu_pages = 0;
8997 
8998 	if (!kvm->arch.n_requested_mmu_pages)
8999 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9000 
9001 	if (nr_mmu_pages)
9002 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9003 
9004 	/*
9005 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9006 	 * sptes have to be split.  If live migration is successful, the guest
9007 	 * in the source machine will be destroyed and large sptes will be
9008 	 * created in the destination. However, if the guest continues to run
9009 	 * in the source machine (for example if live migration fails), small
9010 	 * sptes will remain around and cause bad performance.
9011 	 *
9012 	 * Scan sptes if dirty logging has been stopped, dropping those
9013 	 * which can be collapsed into a single large-page spte.  Later
9014 	 * page faults will create the large-page sptes.
9015 	 */
9016 	if ((change != KVM_MR_DELETE) &&
9017 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9018 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9019 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9020 
9021 	/*
9022 	 * Set up write protection and/or dirty logging for the new slot.
9023 	 *
9024 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9025 	 * been zapped so no dirty logging staff is needed for old slot. For
9026 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9027 	 * new and it's also covered when dealing with the new slot.
9028 	 *
9029 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9030 	 */
9031 	if (change != KVM_MR_DELETE)
9032 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9033 }
9034 
9035 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9036 {
9037 	kvm_mmu_invalidate_zap_all_pages(kvm);
9038 }
9039 
9040 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9041 				   struct kvm_memory_slot *slot)
9042 {
9043 	kvm_page_track_flush_slot(kvm, slot);
9044 }
9045 
9046 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9047 {
9048 	if (!list_empty_careful(&vcpu->async_pf.done))
9049 		return true;
9050 
9051 	if (kvm_apic_has_events(vcpu))
9052 		return true;
9053 
9054 	if (vcpu->arch.pv.pv_unhalted)
9055 		return true;
9056 
9057 	if (vcpu->arch.exception.pending)
9058 		return true;
9059 
9060 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9061 	    (vcpu->arch.nmi_pending &&
9062 	     kvm_x86_ops->nmi_allowed(vcpu)))
9063 		return true;
9064 
9065 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9066 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9067 		return true;
9068 
9069 	if (kvm_arch_interrupt_allowed(vcpu) &&
9070 	    kvm_cpu_has_interrupt(vcpu))
9071 		return true;
9072 
9073 	if (kvm_hv_has_stimer_pending(vcpu))
9074 		return true;
9075 
9076 	return false;
9077 }
9078 
9079 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9080 {
9081 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9082 }
9083 
9084 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9085 {
9086 	return vcpu->arch.preempted_in_kernel;
9087 }
9088 
9089 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9090 {
9091 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9092 }
9093 
9094 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9095 {
9096 	return kvm_x86_ops->interrupt_allowed(vcpu);
9097 }
9098 
9099 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9100 {
9101 	if (is_64_bit_mode(vcpu))
9102 		return kvm_rip_read(vcpu);
9103 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9104 		     kvm_rip_read(vcpu));
9105 }
9106 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9107 
9108 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9109 {
9110 	return kvm_get_linear_rip(vcpu) == linear_rip;
9111 }
9112 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9113 
9114 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9115 {
9116 	unsigned long rflags;
9117 
9118 	rflags = kvm_x86_ops->get_rflags(vcpu);
9119 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9120 		rflags &= ~X86_EFLAGS_TF;
9121 	return rflags;
9122 }
9123 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9124 
9125 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9126 {
9127 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9128 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9129 		rflags |= X86_EFLAGS_TF;
9130 	kvm_x86_ops->set_rflags(vcpu, rflags);
9131 }
9132 
9133 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9134 {
9135 	__kvm_set_rflags(vcpu, rflags);
9136 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9137 }
9138 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9139 
9140 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9141 {
9142 	int r;
9143 
9144 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
9145 	      work->wakeup_all)
9146 		return;
9147 
9148 	r = kvm_mmu_reload(vcpu);
9149 	if (unlikely(r))
9150 		return;
9151 
9152 	if (!vcpu->arch.mmu.direct_map &&
9153 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
9154 		return;
9155 
9156 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
9157 }
9158 
9159 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9160 {
9161 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9162 }
9163 
9164 static inline u32 kvm_async_pf_next_probe(u32 key)
9165 {
9166 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9167 }
9168 
9169 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9170 {
9171 	u32 key = kvm_async_pf_hash_fn(gfn);
9172 
9173 	while (vcpu->arch.apf.gfns[key] != ~0)
9174 		key = kvm_async_pf_next_probe(key);
9175 
9176 	vcpu->arch.apf.gfns[key] = gfn;
9177 }
9178 
9179 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9180 {
9181 	int i;
9182 	u32 key = kvm_async_pf_hash_fn(gfn);
9183 
9184 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9185 		     (vcpu->arch.apf.gfns[key] != gfn &&
9186 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9187 		key = kvm_async_pf_next_probe(key);
9188 
9189 	return key;
9190 }
9191 
9192 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9193 {
9194 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9195 }
9196 
9197 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9198 {
9199 	u32 i, j, k;
9200 
9201 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9202 	while (true) {
9203 		vcpu->arch.apf.gfns[i] = ~0;
9204 		do {
9205 			j = kvm_async_pf_next_probe(j);
9206 			if (vcpu->arch.apf.gfns[j] == ~0)
9207 				return;
9208 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9209 			/*
9210 			 * k lies cyclically in ]i,j]
9211 			 * |    i.k.j |
9212 			 * |....j i.k.| or  |.k..j i...|
9213 			 */
9214 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9215 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9216 		i = j;
9217 	}
9218 }
9219 
9220 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9221 {
9222 
9223 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9224 				      sizeof(val));
9225 }
9226 
9227 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9228 {
9229 
9230 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9231 				      sizeof(u32));
9232 }
9233 
9234 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9235 				     struct kvm_async_pf *work)
9236 {
9237 	struct x86_exception fault;
9238 
9239 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9240 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9241 
9242 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9243 	    (vcpu->arch.apf.send_user_only &&
9244 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9245 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9246 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9247 		fault.vector = PF_VECTOR;
9248 		fault.error_code_valid = true;
9249 		fault.error_code = 0;
9250 		fault.nested_page_fault = false;
9251 		fault.address = work->arch.token;
9252 		fault.async_page_fault = true;
9253 		kvm_inject_page_fault(vcpu, &fault);
9254 	}
9255 }
9256 
9257 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9258 				 struct kvm_async_pf *work)
9259 {
9260 	struct x86_exception fault;
9261 	u32 val;
9262 
9263 	if (work->wakeup_all)
9264 		work->arch.token = ~0; /* broadcast wakeup */
9265 	else
9266 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9267 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9268 
9269 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9270 	    !apf_get_user(vcpu, &val)) {
9271 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9272 		    vcpu->arch.exception.pending &&
9273 		    vcpu->arch.exception.nr == PF_VECTOR &&
9274 		    !apf_put_user(vcpu, 0)) {
9275 			vcpu->arch.exception.injected = false;
9276 			vcpu->arch.exception.pending = false;
9277 			vcpu->arch.exception.nr = 0;
9278 			vcpu->arch.exception.has_error_code = false;
9279 			vcpu->arch.exception.error_code = 0;
9280 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9281 			fault.vector = PF_VECTOR;
9282 			fault.error_code_valid = true;
9283 			fault.error_code = 0;
9284 			fault.nested_page_fault = false;
9285 			fault.address = work->arch.token;
9286 			fault.async_page_fault = true;
9287 			kvm_inject_page_fault(vcpu, &fault);
9288 		}
9289 	}
9290 	vcpu->arch.apf.halted = false;
9291 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9292 }
9293 
9294 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9295 {
9296 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9297 		return true;
9298 	else
9299 		return kvm_can_do_async_pf(vcpu);
9300 }
9301 
9302 void kvm_arch_start_assignment(struct kvm *kvm)
9303 {
9304 	atomic_inc(&kvm->arch.assigned_device_count);
9305 }
9306 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9307 
9308 void kvm_arch_end_assignment(struct kvm *kvm)
9309 {
9310 	atomic_dec(&kvm->arch.assigned_device_count);
9311 }
9312 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9313 
9314 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9315 {
9316 	return atomic_read(&kvm->arch.assigned_device_count);
9317 }
9318 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9319 
9320 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9321 {
9322 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9323 }
9324 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9325 
9326 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9327 {
9328 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9329 }
9330 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9331 
9332 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9333 {
9334 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9335 }
9336 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9337 
9338 bool kvm_arch_has_irq_bypass(void)
9339 {
9340 	return kvm_x86_ops->update_pi_irte != NULL;
9341 }
9342 
9343 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9344 				      struct irq_bypass_producer *prod)
9345 {
9346 	struct kvm_kernel_irqfd *irqfd =
9347 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9348 
9349 	irqfd->producer = prod;
9350 
9351 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9352 					   prod->irq, irqfd->gsi, 1);
9353 }
9354 
9355 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9356 				      struct irq_bypass_producer *prod)
9357 {
9358 	int ret;
9359 	struct kvm_kernel_irqfd *irqfd =
9360 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9361 
9362 	WARN_ON(irqfd->producer != prod);
9363 	irqfd->producer = NULL;
9364 
9365 	/*
9366 	 * When producer of consumer is unregistered, we change back to
9367 	 * remapped mode, so we can re-use the current implementation
9368 	 * when the irq is masked/disabled or the consumer side (KVM
9369 	 * int this case doesn't want to receive the interrupts.
9370 	*/
9371 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9372 	if (ret)
9373 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9374 		       " fails: %d\n", irqfd->consumer.token, ret);
9375 }
9376 
9377 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9378 				   uint32_t guest_irq, bool set)
9379 {
9380 	if (!kvm_x86_ops->update_pi_irte)
9381 		return -EINVAL;
9382 
9383 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9384 }
9385 
9386 bool kvm_vector_hashing_enabled(void)
9387 {
9388 	return vector_hashing;
9389 }
9390 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9391 
9392 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9393 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9394 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9395 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9396 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9397 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9399 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9400 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9401 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9402 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9403 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9404 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9405 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9406 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9407 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9408 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9409 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9411