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 "assigned-dev.h"
31#include "pmu.h"
32#include "hyperv.h"
33
34#include <linux/clocksource.h>
35#include <linux/interrupt.h>
36#include <linux/kvm.h>
37#include <linux/fs.h>
38#include <linux/vmalloc.h>
39#include <linux/module.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 <trace/events/kvm.h>
57
58#define CREATE_TRACE_POINTS
59#include "trace.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
71#define MAX_IO_MSRS 256
72#define KVM_MAX_MCE_BANKS 32
73#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
74
75#define emul_to_vcpu(ctxt) \
76	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
77
78/* EFER defaults:
79 * - enable syscall per default because its emulated by KVM
80 * - enable LME and LMA per default on 64 bit KVM
81 */
82#ifdef CONFIG_X86_64
83static
84u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
85#else
86static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
87#endif
88
89#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
91
92static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93static void process_nmi(struct kvm_vcpu *vcpu);
94static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
95
96struct kvm_x86_ops *kvm_x86_ops __read_mostly;
97EXPORT_SYMBOL_GPL(kvm_x86_ops);
98
99static bool __read_mostly ignore_msrs = 0;
100module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
101
102unsigned int min_timer_period_us = 500;
103module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
104
105static bool __read_mostly kvmclock_periodic_sync = true;
106module_param(kvmclock_periodic_sync, bool, S_IRUGO);
107
108bool __read_mostly kvm_has_tsc_control;
109EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
110u32  __read_mostly kvm_max_guest_tsc_khz;
111EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
112u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
113EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
114u64  __read_mostly kvm_max_tsc_scaling_ratio;
115EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
116static u64 __read_mostly kvm_default_tsc_scaling_ratio;
117
118/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
119static u32 __read_mostly tsc_tolerance_ppm = 250;
120module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
121
122/* lapic timer advance (tscdeadline mode only) in nanoseconds */
123unsigned int __read_mostly lapic_timer_advance_ns = 0;
124module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
125
126static bool __read_mostly backwards_tsc_observed = false;
127
128#define KVM_NR_SHARED_MSRS 16
129
130struct kvm_shared_msrs_global {
131	int nr;
132	u32 msrs[KVM_NR_SHARED_MSRS];
133};
134
135struct kvm_shared_msrs {
136	struct user_return_notifier urn;
137	bool registered;
138	struct kvm_shared_msr_values {
139		u64 host;
140		u64 curr;
141	} values[KVM_NR_SHARED_MSRS];
142};
143
144static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
145static struct kvm_shared_msrs __percpu *shared_msrs;
146
147struct kvm_stats_debugfs_item debugfs_entries[] = {
148	{ "pf_fixed", VCPU_STAT(pf_fixed) },
149	{ "pf_guest", VCPU_STAT(pf_guest) },
150	{ "tlb_flush", VCPU_STAT(tlb_flush) },
151	{ "invlpg", VCPU_STAT(invlpg) },
152	{ "exits", VCPU_STAT(exits) },
153	{ "io_exits", VCPU_STAT(io_exits) },
154	{ "mmio_exits", VCPU_STAT(mmio_exits) },
155	{ "signal_exits", VCPU_STAT(signal_exits) },
156	{ "irq_window", VCPU_STAT(irq_window_exits) },
157	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
158	{ "halt_exits", VCPU_STAT(halt_exits) },
159	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
160	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
161	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
162	{ "hypercalls", VCPU_STAT(hypercalls) },
163	{ "request_irq", VCPU_STAT(request_irq_exits) },
164	{ "irq_exits", VCPU_STAT(irq_exits) },
165	{ "host_state_reload", VCPU_STAT(host_state_reload) },
166	{ "efer_reload", VCPU_STAT(efer_reload) },
167	{ "fpu_reload", VCPU_STAT(fpu_reload) },
168	{ "insn_emulation", VCPU_STAT(insn_emulation) },
169	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
170	{ "irq_injections", VCPU_STAT(irq_injections) },
171	{ "nmi_injections", VCPU_STAT(nmi_injections) },
172	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
173	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
174	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
175	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
176	{ "mmu_flooded", VM_STAT(mmu_flooded) },
177	{ "mmu_recycled", VM_STAT(mmu_recycled) },
178	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
179	{ "mmu_unsync", VM_STAT(mmu_unsync) },
180	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
181	{ "largepages", VM_STAT(lpages) },
182	{ NULL }
183};
184
185u64 __read_mostly host_xcr0;
186
187static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
188
189static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
190{
191	int i;
192	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
193		vcpu->arch.apf.gfns[i] = ~0;
194}
195
196static void kvm_on_user_return(struct user_return_notifier *urn)
197{
198	unsigned slot;
199	struct kvm_shared_msrs *locals
200		= container_of(urn, struct kvm_shared_msrs, urn);
201	struct kvm_shared_msr_values *values;
202
203	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
204		values = &locals->values[slot];
205		if (values->host != values->curr) {
206			wrmsrl(shared_msrs_global.msrs[slot], values->host);
207			values->curr = values->host;
208		}
209	}
210	locals->registered = false;
211	user_return_notifier_unregister(urn);
212}
213
214static void shared_msr_update(unsigned slot, u32 msr)
215{
216	u64 value;
217	unsigned int cpu = smp_processor_id();
218	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
219
220	/* only read, and nobody should modify it at this time,
221	 * so don't need lock */
222	if (slot >= shared_msrs_global.nr) {
223		printk(KERN_ERR "kvm: invalid MSR slot!");
224		return;
225	}
226	rdmsrl_safe(msr, &value);
227	smsr->values[slot].host = value;
228	smsr->values[slot].curr = value;
229}
230
231void kvm_define_shared_msr(unsigned slot, u32 msr)
232{
233	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
234	shared_msrs_global.msrs[slot] = msr;
235	if (slot >= shared_msrs_global.nr)
236		shared_msrs_global.nr = slot + 1;
237}
238EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
239
240static void kvm_shared_msr_cpu_online(void)
241{
242	unsigned i;
243
244	for (i = 0; i < shared_msrs_global.nr; ++i)
245		shared_msr_update(i, shared_msrs_global.msrs[i]);
246}
247
248int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
249{
250	unsigned int cpu = smp_processor_id();
251	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
252	int err;
253
254	if (((value ^ smsr->values[slot].curr) & mask) == 0)
255		return 0;
256	smsr->values[slot].curr = value;
257	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
258	if (err)
259		return 1;
260
261	if (!smsr->registered) {
262		smsr->urn.on_user_return = kvm_on_user_return;
263		user_return_notifier_register(&smsr->urn);
264		smsr->registered = true;
265	}
266	return 0;
267}
268EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
269
270static void drop_user_return_notifiers(void)
271{
272	unsigned int cpu = smp_processor_id();
273	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
274
275	if (smsr->registered)
276		kvm_on_user_return(&smsr->urn);
277}
278
279u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
280{
281	return vcpu->arch.apic_base;
282}
283EXPORT_SYMBOL_GPL(kvm_get_apic_base);
284
285int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
286{
287	u64 old_state = vcpu->arch.apic_base &
288		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
289	u64 new_state = msr_info->data &
290		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
291	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
292		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
293
294	if (!msr_info->host_initiated &&
295	    ((msr_info->data & reserved_bits) != 0 ||
296	     new_state == X2APIC_ENABLE ||
297	     (new_state == MSR_IA32_APICBASE_ENABLE &&
298	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
299	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
300	      old_state == 0)))
301		return 1;
302
303	kvm_lapic_set_base(vcpu, msr_info->data);
304	return 0;
305}
306EXPORT_SYMBOL_GPL(kvm_set_apic_base);
307
308asmlinkage __visible void kvm_spurious_fault(void)
309{
310	/* Fault while not rebooting.  We want the trace. */
311	BUG();
312}
313EXPORT_SYMBOL_GPL(kvm_spurious_fault);
314
315#define EXCPT_BENIGN		0
316#define EXCPT_CONTRIBUTORY	1
317#define EXCPT_PF		2
318
319static int exception_class(int vector)
320{
321	switch (vector) {
322	case PF_VECTOR:
323		return EXCPT_PF;
324	case DE_VECTOR:
325	case TS_VECTOR:
326	case NP_VECTOR:
327	case SS_VECTOR:
328	case GP_VECTOR:
329		return EXCPT_CONTRIBUTORY;
330	default:
331		break;
332	}
333	return EXCPT_BENIGN;
334}
335
336#define EXCPT_FAULT		0
337#define EXCPT_TRAP		1
338#define EXCPT_ABORT		2
339#define EXCPT_INTERRUPT		3
340
341static int exception_type(int vector)
342{
343	unsigned int mask;
344
345	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
346		return EXCPT_INTERRUPT;
347
348	mask = 1 << vector;
349
350	/* #DB is trap, as instruction watchpoints are handled elsewhere */
351	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
352		return EXCPT_TRAP;
353
354	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
355		return EXCPT_ABORT;
356
357	/* Reserved exceptions will result in fault */
358	return EXCPT_FAULT;
359}
360
361static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
362		unsigned nr, bool has_error, u32 error_code,
363		bool reinject)
364{
365	u32 prev_nr;
366	int class1, class2;
367
368	kvm_make_request(KVM_REQ_EVENT, vcpu);
369
370	if (!vcpu->arch.exception.pending) {
371	queue:
372		if (has_error && !is_protmode(vcpu))
373			has_error = false;
374		vcpu->arch.exception.pending = true;
375		vcpu->arch.exception.has_error_code = has_error;
376		vcpu->arch.exception.nr = nr;
377		vcpu->arch.exception.error_code = error_code;
378		vcpu->arch.exception.reinject = reinject;
379		return;
380	}
381
382	/* to check exception */
383	prev_nr = vcpu->arch.exception.nr;
384	if (prev_nr == DF_VECTOR) {
385		/* triple fault -> shutdown */
386		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
387		return;
388	}
389	class1 = exception_class(prev_nr);
390	class2 = exception_class(nr);
391	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
392		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
393		/* generate double fault per SDM Table 5-5 */
394		vcpu->arch.exception.pending = true;
395		vcpu->arch.exception.has_error_code = true;
396		vcpu->arch.exception.nr = DF_VECTOR;
397		vcpu->arch.exception.error_code = 0;
398	} else
399		/* replace previous exception with a new one in a hope
400		   that instruction re-execution will regenerate lost
401		   exception */
402		goto queue;
403}
404
405void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
406{
407	kvm_multiple_exception(vcpu, nr, false, 0, false);
408}
409EXPORT_SYMBOL_GPL(kvm_queue_exception);
410
411void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
412{
413	kvm_multiple_exception(vcpu, nr, false, 0, true);
414}
415EXPORT_SYMBOL_GPL(kvm_requeue_exception);
416
417void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
418{
419	if (err)
420		kvm_inject_gp(vcpu, 0);
421	else
422		kvm_x86_ops->skip_emulated_instruction(vcpu);
423}
424EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
425
426void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
427{
428	++vcpu->stat.pf_guest;
429	vcpu->arch.cr2 = fault->address;
430	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
431}
432EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
433
434static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
435{
436	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
437		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
438	else
439		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
440
441	return fault->nested_page_fault;
442}
443
444void kvm_inject_nmi(struct kvm_vcpu *vcpu)
445{
446	atomic_inc(&vcpu->arch.nmi_queued);
447	kvm_make_request(KVM_REQ_NMI, vcpu);
448}
449EXPORT_SYMBOL_GPL(kvm_inject_nmi);
450
451void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
452{
453	kvm_multiple_exception(vcpu, nr, true, error_code, false);
454}
455EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
456
457void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
458{
459	kvm_multiple_exception(vcpu, nr, true, error_code, true);
460}
461EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
462
463/*
464 * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
465 * a #GP and return false.
466 */
467bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
468{
469	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
470		return true;
471	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
472	return false;
473}
474EXPORT_SYMBOL_GPL(kvm_require_cpl);
475
476bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
477{
478	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
479		return true;
480
481	kvm_queue_exception(vcpu, UD_VECTOR);
482	return false;
483}
484EXPORT_SYMBOL_GPL(kvm_require_dr);
485
486/*
487 * This function will be used to read from the physical memory of the currently
488 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
489 * can read from guest physical or from the guest's guest physical memory.
490 */
491int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
492			    gfn_t ngfn, void *data, int offset, int len,
493			    u32 access)
494{
495	struct x86_exception exception;
496	gfn_t real_gfn;
497	gpa_t ngpa;
498
499	ngpa     = gfn_to_gpa(ngfn);
500	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
501	if (real_gfn == UNMAPPED_GVA)
502		return -EFAULT;
503
504	real_gfn = gpa_to_gfn(real_gfn);
505
506	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
507}
508EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
509
510static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
511			       void *data, int offset, int len, u32 access)
512{
513	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
514				       data, offset, len, access);
515}
516
517/*
518 * Load the pae pdptrs.  Return true is they are all valid.
519 */
520int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
521{
522	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
523	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
524	int i;
525	int ret;
526	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
527
528	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
529				      offset * sizeof(u64), sizeof(pdpte),
530				      PFERR_USER_MASK|PFERR_WRITE_MASK);
531	if (ret < 0) {
532		ret = 0;
533		goto out;
534	}
535	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
536		if (is_present_gpte(pdpte[i]) &&
537		    (pdpte[i] &
538		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
539			ret = 0;
540			goto out;
541		}
542	}
543	ret = 1;
544
545	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
546	__set_bit(VCPU_EXREG_PDPTR,
547		  (unsigned long *)&vcpu->arch.regs_avail);
548	__set_bit(VCPU_EXREG_PDPTR,
549		  (unsigned long *)&vcpu->arch.regs_dirty);
550out:
551
552	return ret;
553}
554EXPORT_SYMBOL_GPL(load_pdptrs);
555
556static bool pdptrs_changed(struct kvm_vcpu *vcpu)
557{
558	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
559	bool changed = true;
560	int offset;
561	gfn_t gfn;
562	int r;
563
564	if (is_long_mode(vcpu) || !is_pae(vcpu))
565		return false;
566
567	if (!test_bit(VCPU_EXREG_PDPTR,
568		      (unsigned long *)&vcpu->arch.regs_avail))
569		return true;
570
571	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
572	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
573	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
574				       PFERR_USER_MASK | PFERR_WRITE_MASK);
575	if (r < 0)
576		goto out;
577	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
578out:
579
580	return changed;
581}
582
583int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
584{
585	unsigned long old_cr0 = kvm_read_cr0(vcpu);
586	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
587
588	cr0 |= X86_CR0_ET;
589
590#ifdef CONFIG_X86_64
591	if (cr0 & 0xffffffff00000000UL)
592		return 1;
593#endif
594
595	cr0 &= ~CR0_RESERVED_BITS;
596
597	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
598		return 1;
599
600	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
601		return 1;
602
603	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
604#ifdef CONFIG_X86_64
605		if ((vcpu->arch.efer & EFER_LME)) {
606			int cs_db, cs_l;
607
608			if (!is_pae(vcpu))
609				return 1;
610			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
611			if (cs_l)
612				return 1;
613		} else
614#endif
615		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
616						 kvm_read_cr3(vcpu)))
617			return 1;
618	}
619
620	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
621		return 1;
622
623	kvm_x86_ops->set_cr0(vcpu, cr0);
624
625	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
626		kvm_clear_async_pf_completion_queue(vcpu);
627		kvm_async_pf_hash_reset(vcpu);
628	}
629
630	if ((cr0 ^ old_cr0) & update_bits)
631		kvm_mmu_reset_context(vcpu);
632
633	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
634	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
635	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
636		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
637
638	return 0;
639}
640EXPORT_SYMBOL_GPL(kvm_set_cr0);
641
642void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
643{
644	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
645}
646EXPORT_SYMBOL_GPL(kvm_lmsw);
647
648static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
649{
650	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
651			!vcpu->guest_xcr0_loaded) {
652		/* kvm_set_xcr() also depends on this */
653		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
654		vcpu->guest_xcr0_loaded = 1;
655	}
656}
657
658static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
659{
660	if (vcpu->guest_xcr0_loaded) {
661		if (vcpu->arch.xcr0 != host_xcr0)
662			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
663		vcpu->guest_xcr0_loaded = 0;
664	}
665}
666
667static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
668{
669	u64 xcr0 = xcr;
670	u64 old_xcr0 = vcpu->arch.xcr0;
671	u64 valid_bits;
672
673	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
674	if (index != XCR_XFEATURE_ENABLED_MASK)
675		return 1;
676	if (!(xcr0 & XFEATURE_MASK_FP))
677		return 1;
678	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
679		return 1;
680
681	/*
682	 * Do not allow the guest to set bits that we do not support
683	 * saving.  However, xcr0 bit 0 is always set, even if the
684	 * emulated CPU does not support XSAVE (see fx_init).
685	 */
686	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
687	if (xcr0 & ~valid_bits)
688		return 1;
689
690	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
691	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
692		return 1;
693
694	if (xcr0 & XFEATURE_MASK_AVX512) {
695		if (!(xcr0 & XFEATURE_MASK_YMM))
696			return 1;
697		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
698			return 1;
699	}
700	vcpu->arch.xcr0 = xcr0;
701
702	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
703		kvm_update_cpuid(vcpu);
704	return 0;
705}
706
707int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
708{
709	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
710	    __kvm_set_xcr(vcpu, index, xcr)) {
711		kvm_inject_gp(vcpu, 0);
712		return 1;
713	}
714	return 0;
715}
716EXPORT_SYMBOL_GPL(kvm_set_xcr);
717
718int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
719{
720	unsigned long old_cr4 = kvm_read_cr4(vcpu);
721	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
722				   X86_CR4_SMEP | X86_CR4_SMAP;
723
724	if (cr4 & CR4_RESERVED_BITS)
725		return 1;
726
727	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
728		return 1;
729
730	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
731		return 1;
732
733	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
734		return 1;
735
736	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
737		return 1;
738
739	if (is_long_mode(vcpu)) {
740		if (!(cr4 & X86_CR4_PAE))
741			return 1;
742	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
743		   && ((cr4 ^ old_cr4) & pdptr_bits)
744		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
745				   kvm_read_cr3(vcpu)))
746		return 1;
747
748	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
749		if (!guest_cpuid_has_pcid(vcpu))
750			return 1;
751
752		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
753		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
754			return 1;
755	}
756
757	if (kvm_x86_ops->set_cr4(vcpu, cr4))
758		return 1;
759
760	if (((cr4 ^ old_cr4) & pdptr_bits) ||
761	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
762		kvm_mmu_reset_context(vcpu);
763
764	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
765		kvm_update_cpuid(vcpu);
766
767	return 0;
768}
769EXPORT_SYMBOL_GPL(kvm_set_cr4);
770
771int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
772{
773#ifdef CONFIG_X86_64
774	cr3 &= ~CR3_PCID_INVD;
775#endif
776
777	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
778		kvm_mmu_sync_roots(vcpu);
779		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
780		return 0;
781	}
782
783	if (is_long_mode(vcpu)) {
784		if (cr3 & CR3_L_MODE_RESERVED_BITS)
785			return 1;
786	} else if (is_pae(vcpu) && is_paging(vcpu) &&
787		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
788		return 1;
789
790	vcpu->arch.cr3 = cr3;
791	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
792	kvm_mmu_new_cr3(vcpu);
793	return 0;
794}
795EXPORT_SYMBOL_GPL(kvm_set_cr3);
796
797int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
798{
799	if (cr8 & CR8_RESERVED_BITS)
800		return 1;
801	if (lapic_in_kernel(vcpu))
802		kvm_lapic_set_tpr(vcpu, cr8);
803	else
804		vcpu->arch.cr8 = cr8;
805	return 0;
806}
807EXPORT_SYMBOL_GPL(kvm_set_cr8);
808
809unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
810{
811	if (lapic_in_kernel(vcpu))
812		return kvm_lapic_get_cr8(vcpu);
813	else
814		return vcpu->arch.cr8;
815}
816EXPORT_SYMBOL_GPL(kvm_get_cr8);
817
818static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
819{
820	int i;
821
822	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
823		for (i = 0; i < KVM_NR_DB_REGS; i++)
824			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
825		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
826	}
827}
828
829static void kvm_update_dr6(struct kvm_vcpu *vcpu)
830{
831	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
832		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
833}
834
835static void kvm_update_dr7(struct kvm_vcpu *vcpu)
836{
837	unsigned long dr7;
838
839	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
840		dr7 = vcpu->arch.guest_debug_dr7;
841	else
842		dr7 = vcpu->arch.dr7;
843	kvm_x86_ops->set_dr7(vcpu, dr7);
844	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
845	if (dr7 & DR7_BP_EN_MASK)
846		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
847}
848
849static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
850{
851	u64 fixed = DR6_FIXED_1;
852
853	if (!guest_cpuid_has_rtm(vcpu))
854		fixed |= DR6_RTM;
855	return fixed;
856}
857
858static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
859{
860	switch (dr) {
861	case 0 ... 3:
862		vcpu->arch.db[dr] = val;
863		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
864			vcpu->arch.eff_db[dr] = val;
865		break;
866	case 4:
867		/* fall through */
868	case 6:
869		if (val & 0xffffffff00000000ULL)
870			return -1; /* #GP */
871		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
872		kvm_update_dr6(vcpu);
873		break;
874	case 5:
875		/* fall through */
876	default: /* 7 */
877		if (val & 0xffffffff00000000ULL)
878			return -1; /* #GP */
879		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
880		kvm_update_dr7(vcpu);
881		break;
882	}
883
884	return 0;
885}
886
887int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
888{
889	if (__kvm_set_dr(vcpu, dr, val)) {
890		kvm_inject_gp(vcpu, 0);
891		return 1;
892	}
893	return 0;
894}
895EXPORT_SYMBOL_GPL(kvm_set_dr);
896
897int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
898{
899	switch (dr) {
900	case 0 ... 3:
901		*val = vcpu->arch.db[dr];
902		break;
903	case 4:
904		/* fall through */
905	case 6:
906		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
907			*val = vcpu->arch.dr6;
908		else
909			*val = kvm_x86_ops->get_dr6(vcpu);
910		break;
911	case 5:
912		/* fall through */
913	default: /* 7 */
914		*val = vcpu->arch.dr7;
915		break;
916	}
917	return 0;
918}
919EXPORT_SYMBOL_GPL(kvm_get_dr);
920
921bool kvm_rdpmc(struct kvm_vcpu *vcpu)
922{
923	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
924	u64 data;
925	int err;
926
927	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
928	if (err)
929		return err;
930	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
931	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
932	return err;
933}
934EXPORT_SYMBOL_GPL(kvm_rdpmc);
935
936/*
937 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
938 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
939 *
940 * This list is modified at module load time to reflect the
941 * capabilities of the host cpu. This capabilities test skips MSRs that are
942 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
943 * may depend on host virtualization features rather than host cpu features.
944 */
945
946static u32 msrs_to_save[] = {
947	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
948	MSR_STAR,
949#ifdef CONFIG_X86_64
950	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
951#endif
952	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
953	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
954};
955
956static unsigned num_msrs_to_save;
957
958static u32 emulated_msrs[] = {
959	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
960	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
961	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
962	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
963	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
964	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
965	HV_X64_MSR_RESET,
966	HV_X64_MSR_VP_INDEX,
967	HV_X64_MSR_VP_RUNTIME,
968	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
969	MSR_KVM_PV_EOI_EN,
970
971	MSR_IA32_TSC_ADJUST,
972	MSR_IA32_TSCDEADLINE,
973	MSR_IA32_MISC_ENABLE,
974	MSR_IA32_MCG_STATUS,
975	MSR_IA32_MCG_CTL,
976	MSR_IA32_SMBASE,
977};
978
979static unsigned num_emulated_msrs;
980
981bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
982{
983	if (efer & efer_reserved_bits)
984		return false;
985
986	if (efer & EFER_FFXSR) {
987		struct kvm_cpuid_entry2 *feat;
988
989		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
990		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
991			return false;
992	}
993
994	if (efer & EFER_SVME) {
995		struct kvm_cpuid_entry2 *feat;
996
997		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
998		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
999			return false;
1000	}
1001
1002	return true;
1003}
1004EXPORT_SYMBOL_GPL(kvm_valid_efer);
1005
1006static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1007{
1008	u64 old_efer = vcpu->arch.efer;
1009
1010	if (!kvm_valid_efer(vcpu, efer))
1011		return 1;
1012
1013	if (is_paging(vcpu)
1014	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1015		return 1;
1016
1017	efer &= ~EFER_LMA;
1018	efer |= vcpu->arch.efer & EFER_LMA;
1019
1020	kvm_x86_ops->set_efer(vcpu, efer);
1021
1022	/* Update reserved bits */
1023	if ((efer ^ old_efer) & EFER_NX)
1024		kvm_mmu_reset_context(vcpu);
1025
1026	return 0;
1027}
1028
1029void kvm_enable_efer_bits(u64 mask)
1030{
1031       efer_reserved_bits &= ~mask;
1032}
1033EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1034
1035/*
1036 * Writes msr value into into the appropriate "register".
1037 * Returns 0 on success, non-0 otherwise.
1038 * Assumes vcpu_load() was already called.
1039 */
1040int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1041{
1042	switch (msr->index) {
1043	case MSR_FS_BASE:
1044	case MSR_GS_BASE:
1045	case MSR_KERNEL_GS_BASE:
1046	case MSR_CSTAR:
1047	case MSR_LSTAR:
1048		if (is_noncanonical_address(msr->data))
1049			return 1;
1050		break;
1051	case MSR_IA32_SYSENTER_EIP:
1052	case MSR_IA32_SYSENTER_ESP:
1053		/*
1054		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1055		 * non-canonical address is written on Intel but not on
1056		 * AMD (which ignores the top 32-bits, because it does
1057		 * not implement 64-bit SYSENTER).
1058		 *
1059		 * 64-bit code should hence be able to write a non-canonical
1060		 * value on AMD.  Making the address canonical ensures that
1061		 * vmentry does not fail on Intel after writing a non-canonical
1062		 * value, and that something deterministic happens if the guest
1063		 * invokes 64-bit SYSENTER.
1064		 */
1065		msr->data = get_canonical(msr->data);
1066	}
1067	return kvm_x86_ops->set_msr(vcpu, msr);
1068}
1069EXPORT_SYMBOL_GPL(kvm_set_msr);
1070
1071/*
1072 * Adapt set_msr() to msr_io()'s calling convention
1073 */
1074static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1075{
1076	struct msr_data msr;
1077	int r;
1078
1079	msr.index = index;
1080	msr.host_initiated = true;
1081	r = kvm_get_msr(vcpu, &msr);
1082	if (r)
1083		return r;
1084
1085	*data = msr.data;
1086	return 0;
1087}
1088
1089static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1090{
1091	struct msr_data msr;
1092
1093	msr.data = *data;
1094	msr.index = index;
1095	msr.host_initiated = true;
1096	return kvm_set_msr(vcpu, &msr);
1097}
1098
1099#ifdef CONFIG_X86_64
1100struct pvclock_gtod_data {
1101	seqcount_t	seq;
1102
1103	struct { /* extract of a clocksource struct */
1104		int vclock_mode;
1105		cycle_t	cycle_last;
1106		cycle_t	mask;
1107		u32	mult;
1108		u32	shift;
1109	} clock;
1110
1111	u64		boot_ns;
1112	u64		nsec_base;
1113};
1114
1115static struct pvclock_gtod_data pvclock_gtod_data;
1116
1117static void update_pvclock_gtod(struct timekeeper *tk)
1118{
1119	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1120	u64 boot_ns;
1121
1122	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1123
1124	write_seqcount_begin(&vdata->seq);
1125
1126	/* copy pvclock gtod data */
1127	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1128	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1129	vdata->clock.mask		= tk->tkr_mono.mask;
1130	vdata->clock.mult		= tk->tkr_mono.mult;
1131	vdata->clock.shift		= tk->tkr_mono.shift;
1132
1133	vdata->boot_ns			= boot_ns;
1134	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1135
1136	write_seqcount_end(&vdata->seq);
1137}
1138#endif
1139
1140void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1141{
1142	/*
1143	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1144	 * vcpu_enter_guest.  This function is only called from
1145	 * the physical CPU that is running vcpu.
1146	 */
1147	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1148}
1149
1150static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1151{
1152	int version;
1153	int r;
1154	struct pvclock_wall_clock wc;
1155	struct timespec boot;
1156
1157	if (!wall_clock)
1158		return;
1159
1160	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1161	if (r)
1162		return;
1163
1164	if (version & 1)
1165		++version;  /* first time write, random junk */
1166
1167	++version;
1168
1169	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1170
1171	/*
1172	 * The guest calculates current wall clock time by adding
1173	 * system time (updated by kvm_guest_time_update below) to the
1174	 * wall clock specified here.  guest system time equals host
1175	 * system time for us, thus we must fill in host boot time here.
1176	 */
1177	getboottime(&boot);
1178
1179	if (kvm->arch.kvmclock_offset) {
1180		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1181		boot = timespec_sub(boot, ts);
1182	}
1183	wc.sec = boot.tv_sec;
1184	wc.nsec = boot.tv_nsec;
1185	wc.version = version;
1186
1187	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1188
1189	version++;
1190	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1191}
1192
1193static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1194{
1195	uint32_t quotient, remainder;
1196
1197	/* Don't try to replace with do_div(), this one calculates
1198	 * "(dividend << 32) / divisor" */
1199	__asm__ ( "divl %4"
1200		  : "=a" (quotient), "=d" (remainder)
1201		  : "0" (0), "1" (dividend), "r" (divisor) );
1202	return quotient;
1203}
1204
1205static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1206			       s8 *pshift, u32 *pmultiplier)
1207{
1208	uint64_t scaled64;
1209	int32_t  shift = 0;
1210	uint64_t tps64;
1211	uint32_t tps32;
1212
1213	tps64 = base_khz * 1000LL;
1214	scaled64 = scaled_khz * 1000LL;
1215	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1216		tps64 >>= 1;
1217		shift--;
1218	}
1219
1220	tps32 = (uint32_t)tps64;
1221	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1222		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1223			scaled64 >>= 1;
1224		else
1225			tps32 <<= 1;
1226		shift++;
1227	}
1228
1229	*pshift = shift;
1230	*pmultiplier = div_frac(scaled64, tps32);
1231
1232	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1233		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1234}
1235
1236#ifdef CONFIG_X86_64
1237static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1238#endif
1239
1240static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1241static unsigned long max_tsc_khz;
1242
1243static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1244{
1245	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1246				   vcpu->arch.virtual_tsc_shift);
1247}
1248
1249static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1250{
1251	u64 v = (u64)khz * (1000000 + ppm);
1252	do_div(v, 1000000);
1253	return v;
1254}
1255
1256static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1257{
1258	u64 ratio;
1259
1260	/* Guest TSC same frequency as host TSC? */
1261	if (!scale) {
1262		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1263		return 0;
1264	}
1265
1266	/* TSC scaling supported? */
1267	if (!kvm_has_tsc_control) {
1268		if (user_tsc_khz > tsc_khz) {
1269			vcpu->arch.tsc_catchup = 1;
1270			vcpu->arch.tsc_always_catchup = 1;
1271			return 0;
1272		} else {
1273			WARN(1, "user requested TSC rate below hardware speed\n");
1274			return -1;
1275		}
1276	}
1277
1278	/* TSC scaling required  - calculate ratio */
1279	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1280				user_tsc_khz, tsc_khz);
1281
1282	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1283		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1284			  user_tsc_khz);
1285		return -1;
1286	}
1287
1288	vcpu->arch.tsc_scaling_ratio = ratio;
1289	return 0;
1290}
1291
1292static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1293{
1294	u32 thresh_lo, thresh_hi;
1295	int use_scaling = 0;
1296
1297	/* tsc_khz can be zero if TSC calibration fails */
1298	if (this_tsc_khz == 0) {
1299		/* set tsc_scaling_ratio to a safe value */
1300		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1301		return -1;
1302	}
1303
1304	/* Compute a scale to convert nanoseconds in TSC cycles */
1305	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1306			   &vcpu->arch.virtual_tsc_shift,
1307			   &vcpu->arch.virtual_tsc_mult);
1308	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1309
1310	/*
1311	 * Compute the variation in TSC rate which is acceptable
1312	 * within the range of tolerance and decide if the
1313	 * rate being applied is within that bounds of the hardware
1314	 * rate.  If so, no scaling or compensation need be done.
1315	 */
1316	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1317	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1318	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1319		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1320		use_scaling = 1;
1321	}
1322	return set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1323}
1324
1325static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1326{
1327	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1328				      vcpu->arch.virtual_tsc_mult,
1329				      vcpu->arch.virtual_tsc_shift);
1330	tsc += vcpu->arch.this_tsc_write;
1331	return tsc;
1332}
1333
1334static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1335{
1336#ifdef CONFIG_X86_64
1337	bool vcpus_matched;
1338	struct kvm_arch *ka = &vcpu->kvm->arch;
1339	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1340
1341	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1342			 atomic_read(&vcpu->kvm->online_vcpus));
1343
1344	/*
1345	 * Once the masterclock is enabled, always perform request in
1346	 * order to update it.
1347	 *
1348	 * In order to enable masterclock, the host clocksource must be TSC
1349	 * and the vcpus need to have matched TSCs.  When that happens,
1350	 * perform request to enable masterclock.
1351	 */
1352	if (ka->use_master_clock ||
1353	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1354		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1355
1356	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1357			    atomic_read(&vcpu->kvm->online_vcpus),
1358		            ka->use_master_clock, gtod->clock.vclock_mode);
1359#endif
1360}
1361
1362static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1363{
1364	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1365	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1366}
1367
1368/*
1369 * Multiply tsc by a fixed point number represented by ratio.
1370 *
1371 * The most significant 64-N bits (mult) of ratio represent the
1372 * integral part of the fixed point number; the remaining N bits
1373 * (frac) represent the fractional part, ie. ratio represents a fixed
1374 * point number (mult + frac * 2^(-N)).
1375 *
1376 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1377 */
1378static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1379{
1380	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1381}
1382
1383u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1384{
1385	u64 _tsc = tsc;
1386	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1387
1388	if (ratio != kvm_default_tsc_scaling_ratio)
1389		_tsc = __scale_tsc(ratio, tsc);
1390
1391	return _tsc;
1392}
1393EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1394
1395static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1396{
1397	u64 tsc;
1398
1399	tsc = kvm_scale_tsc(vcpu, rdtsc());
1400
1401	return target_tsc - tsc;
1402}
1403
1404u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1405{
1406	return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1407}
1408EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1409
1410void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1411{
1412	struct kvm *kvm = vcpu->kvm;
1413	u64 offset, ns, elapsed;
1414	unsigned long flags;
1415	s64 usdiff;
1416	bool matched;
1417	bool already_matched;
1418	u64 data = msr->data;
1419
1420	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1421	offset = kvm_compute_tsc_offset(vcpu, data);
1422	ns = get_kernel_ns();
1423	elapsed = ns - kvm->arch.last_tsc_nsec;
1424
1425	if (vcpu->arch.virtual_tsc_khz) {
1426		int faulted = 0;
1427
1428		/* n.b - signed multiplication and division required */
1429		usdiff = data - kvm->arch.last_tsc_write;
1430#ifdef CONFIG_X86_64
1431		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1432#else
1433		/* do_div() only does unsigned */
1434		asm("1: idivl %[divisor]\n"
1435		    "2: xor %%edx, %%edx\n"
1436		    "   movl $0, %[faulted]\n"
1437		    "3:\n"
1438		    ".section .fixup,\"ax\"\n"
1439		    "4: movl $1, %[faulted]\n"
1440		    "   jmp  3b\n"
1441		    ".previous\n"
1442
1443		_ASM_EXTABLE(1b, 4b)
1444
1445		: "=A"(usdiff), [faulted] "=r" (faulted)
1446		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1447
1448#endif
1449		do_div(elapsed, 1000);
1450		usdiff -= elapsed;
1451		if (usdiff < 0)
1452			usdiff = -usdiff;
1453
1454		/* idivl overflow => difference is larger than USEC_PER_SEC */
1455		if (faulted)
1456			usdiff = USEC_PER_SEC;
1457	} else
1458		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1459
1460	/*
1461	 * Special case: TSC write with a small delta (1 second) of virtual
1462	 * cycle time against real time is interpreted as an attempt to
1463	 * synchronize the CPU.
1464         *
1465	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1466	 * TSC, we add elapsed time in this computation.  We could let the
1467	 * compensation code attempt to catch up if we fall behind, but
1468	 * it's better to try to match offsets from the beginning.
1469         */
1470	if (usdiff < USEC_PER_SEC &&
1471	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1472		if (!check_tsc_unstable()) {
1473			offset = kvm->arch.cur_tsc_offset;
1474			pr_debug("kvm: matched tsc offset for %llu\n", data);
1475		} else {
1476			u64 delta = nsec_to_cycles(vcpu, elapsed);
1477			data += delta;
1478			offset = kvm_compute_tsc_offset(vcpu, data);
1479			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1480		}
1481		matched = true;
1482		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1483	} else {
1484		/*
1485		 * We split periods of matched TSC writes into generations.
1486		 * For each generation, we track the original measured
1487		 * nanosecond time, offset, and write, so if TSCs are in
1488		 * sync, we can match exact offset, and if not, we can match
1489		 * exact software computation in compute_guest_tsc()
1490		 *
1491		 * These values are tracked in kvm->arch.cur_xxx variables.
1492		 */
1493		kvm->arch.cur_tsc_generation++;
1494		kvm->arch.cur_tsc_nsec = ns;
1495		kvm->arch.cur_tsc_write = data;
1496		kvm->arch.cur_tsc_offset = offset;
1497		matched = false;
1498		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1499			 kvm->arch.cur_tsc_generation, data);
1500	}
1501
1502	/*
1503	 * We also track th most recent recorded KHZ, write and time to
1504	 * allow the matching interval to be extended at each write.
1505	 */
1506	kvm->arch.last_tsc_nsec = ns;
1507	kvm->arch.last_tsc_write = data;
1508	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1509
1510	vcpu->arch.last_guest_tsc = data;
1511
1512	/* Keep track of which generation this VCPU has synchronized to */
1513	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1514	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1515	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1516
1517	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1518		update_ia32_tsc_adjust_msr(vcpu, offset);
1519	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1520	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1521
1522	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1523	if (!matched) {
1524		kvm->arch.nr_vcpus_matched_tsc = 0;
1525	} else if (!already_matched) {
1526		kvm->arch.nr_vcpus_matched_tsc++;
1527	}
1528
1529	kvm_track_tsc_matching(vcpu);
1530	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1531}
1532
1533EXPORT_SYMBOL_GPL(kvm_write_tsc);
1534
1535static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1536					   s64 adjustment)
1537{
1538	kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1539}
1540
1541static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1542{
1543	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1544		WARN_ON(adjustment < 0);
1545	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1546	kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1547}
1548
1549#ifdef CONFIG_X86_64
1550
1551static cycle_t read_tsc(void)
1552{
1553	cycle_t ret = (cycle_t)rdtsc_ordered();
1554	u64 last = pvclock_gtod_data.clock.cycle_last;
1555
1556	if (likely(ret >= last))
1557		return ret;
1558
1559	/*
1560	 * GCC likes to generate cmov here, but this branch is extremely
1561	 * predictable (it's just a funciton of time and the likely is
1562	 * very likely) and there's a data dependence, so force GCC
1563	 * to generate a branch instead.  I don't barrier() because
1564	 * we don't actually need a barrier, and if this function
1565	 * ever gets inlined it will generate worse code.
1566	 */
1567	asm volatile ("");
1568	return last;
1569}
1570
1571static inline u64 vgettsc(cycle_t *cycle_now)
1572{
1573	long v;
1574	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1575
1576	*cycle_now = read_tsc();
1577
1578	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1579	return v * gtod->clock.mult;
1580}
1581
1582static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1583{
1584	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1585	unsigned long seq;
1586	int mode;
1587	u64 ns;
1588
1589	do {
1590		seq = read_seqcount_begin(&gtod->seq);
1591		mode = gtod->clock.vclock_mode;
1592		ns = gtod->nsec_base;
1593		ns += vgettsc(cycle_now);
1594		ns >>= gtod->clock.shift;
1595		ns += gtod->boot_ns;
1596	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1597	*t = ns;
1598
1599	return mode;
1600}
1601
1602/* returns true if host is using tsc clocksource */
1603static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1604{
1605	/* checked again under seqlock below */
1606	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1607		return false;
1608
1609	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1610}
1611#endif
1612
1613/*
1614 *
1615 * Assuming a stable TSC across physical CPUS, and a stable TSC
1616 * across virtual CPUs, the following condition is possible.
1617 * Each numbered line represents an event visible to both
1618 * CPUs at the next numbered event.
1619 *
1620 * "timespecX" represents host monotonic time. "tscX" represents
1621 * RDTSC value.
1622 *
1623 * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1624 *
1625 * 1.  read timespec0,tsc0
1626 * 2.					| timespec1 = timespec0 + N
1627 * 					| tsc1 = tsc0 + M
1628 * 3. transition to guest		| transition to guest
1629 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1630 * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1631 * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1632 *
1633 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1634 *
1635 * 	- ret0 < ret1
1636 *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1637 *		...
1638 *	- 0 < N - M => M < N
1639 *
1640 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1641 * always the case (the difference between two distinct xtime instances
1642 * might be smaller then the difference between corresponding TSC reads,
1643 * when updating guest vcpus pvclock areas).
1644 *
1645 * To avoid that problem, do not allow visibility of distinct
1646 * system_timestamp/tsc_timestamp values simultaneously: use a master
1647 * copy of host monotonic time values. Update that master copy
1648 * in lockstep.
1649 *
1650 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1651 *
1652 */
1653
1654static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1655{
1656#ifdef CONFIG_X86_64
1657	struct kvm_arch *ka = &kvm->arch;
1658	int vclock_mode;
1659	bool host_tsc_clocksource, vcpus_matched;
1660
1661	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1662			atomic_read(&kvm->online_vcpus));
1663
1664	/*
1665	 * If the host uses TSC clock, then passthrough TSC as stable
1666	 * to the guest.
1667	 */
1668	host_tsc_clocksource = kvm_get_time_and_clockread(
1669					&ka->master_kernel_ns,
1670					&ka->master_cycle_now);
1671
1672	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1673				&& !backwards_tsc_observed
1674				&& !ka->boot_vcpu_runs_old_kvmclock;
1675
1676	if (ka->use_master_clock)
1677		atomic_set(&kvm_guest_has_master_clock, 1);
1678
1679	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1680	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1681					vcpus_matched);
1682#endif
1683}
1684
1685static void kvm_gen_update_masterclock(struct kvm *kvm)
1686{
1687#ifdef CONFIG_X86_64
1688	int i;
1689	struct kvm_vcpu *vcpu;
1690	struct kvm_arch *ka = &kvm->arch;
1691
1692	spin_lock(&ka->pvclock_gtod_sync_lock);
1693	kvm_make_mclock_inprogress_request(kvm);
1694	/* no guest entries from this point */
1695	pvclock_update_vm_gtod_copy(kvm);
1696
1697	kvm_for_each_vcpu(i, vcpu, kvm)
1698		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1699
1700	/* guest entries allowed */
1701	kvm_for_each_vcpu(i, vcpu, kvm)
1702		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1703
1704	spin_unlock(&ka->pvclock_gtod_sync_lock);
1705#endif
1706}
1707
1708static int kvm_guest_time_update(struct kvm_vcpu *v)
1709{
1710	unsigned long flags, this_tsc_khz, tgt_tsc_khz;
1711	struct kvm_vcpu_arch *vcpu = &v->arch;
1712	struct kvm_arch *ka = &v->kvm->arch;
1713	s64 kernel_ns;
1714	u64 tsc_timestamp, host_tsc;
1715	struct pvclock_vcpu_time_info guest_hv_clock;
1716	u8 pvclock_flags;
1717	bool use_master_clock;
1718
1719	kernel_ns = 0;
1720	host_tsc = 0;
1721
1722	/*
1723	 * If the host uses TSC clock, then passthrough TSC as stable
1724	 * to the guest.
1725	 */
1726	spin_lock(&ka->pvclock_gtod_sync_lock);
1727	use_master_clock = ka->use_master_clock;
1728	if (use_master_clock) {
1729		host_tsc = ka->master_cycle_now;
1730		kernel_ns = ka->master_kernel_ns;
1731	}
1732	spin_unlock(&ka->pvclock_gtod_sync_lock);
1733
1734	/* Keep irq disabled to prevent changes to the clock */
1735	local_irq_save(flags);
1736	this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1737	if (unlikely(this_tsc_khz == 0)) {
1738		local_irq_restore(flags);
1739		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1740		return 1;
1741	}
1742	if (!use_master_clock) {
1743		host_tsc = rdtsc();
1744		kernel_ns = get_kernel_ns();
1745	}
1746
1747	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1748
1749	/*
1750	 * We may have to catch up the TSC to match elapsed wall clock
1751	 * time for two reasons, even if kvmclock is used.
1752	 *   1) CPU could have been running below the maximum TSC rate
1753	 *   2) Broken TSC compensation resets the base at each VCPU
1754	 *      entry to avoid unknown leaps of TSC even when running
1755	 *      again on the same CPU.  This may cause apparent elapsed
1756	 *      time to disappear, and the guest to stand still or run
1757	 *	very slowly.
1758	 */
1759	if (vcpu->tsc_catchup) {
1760		u64 tsc = compute_guest_tsc(v, kernel_ns);
1761		if (tsc > tsc_timestamp) {
1762			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1763			tsc_timestamp = tsc;
1764		}
1765	}
1766
1767	local_irq_restore(flags);
1768
1769	if (!vcpu->pv_time_enabled)
1770		return 0;
1771
1772	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1773		tgt_tsc_khz = kvm_has_tsc_control ?
1774			vcpu->virtual_tsc_khz : this_tsc_khz;
1775		kvm_get_time_scale(NSEC_PER_SEC / 1000, tgt_tsc_khz,
1776				   &vcpu->hv_clock.tsc_shift,
1777				   &vcpu->hv_clock.tsc_to_system_mul);
1778		vcpu->hw_tsc_khz = this_tsc_khz;
1779	}
1780
1781	/* With all the info we got, fill in the values */
1782	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1783	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1784	vcpu->last_guest_tsc = tsc_timestamp;
1785
1786	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1787		&guest_hv_clock, sizeof(guest_hv_clock))))
1788		return 0;
1789
1790	/* This VCPU is paused, but it's legal for a guest to read another
1791	 * VCPU's kvmclock, so we really have to follow the specification where
1792	 * it says that version is odd if data is being modified, and even after
1793	 * it is consistent.
1794	 *
1795	 * Version field updates must be kept separate.  This is because
1796	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1797	 * writes within a string instruction are weakly ordered.  So there
1798	 * are three writes overall.
1799	 *
1800	 * As a small optimization, only write the version field in the first
1801	 * and third write.  The vcpu->pv_time cache is still valid, because the
1802	 * version field is the first in the struct.
1803	 */
1804	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1805
1806	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1807	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1808				&vcpu->hv_clock,
1809				sizeof(vcpu->hv_clock.version));
1810
1811	smp_wmb();
1812
1813	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1814	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1815
1816	if (vcpu->pvclock_set_guest_stopped_request) {
1817		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1818		vcpu->pvclock_set_guest_stopped_request = false;
1819	}
1820
1821	/* If the host uses TSC clocksource, then it is stable */
1822	if (use_master_clock)
1823		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1824
1825	vcpu->hv_clock.flags = pvclock_flags;
1826
1827	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1828
1829	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1830				&vcpu->hv_clock,
1831				sizeof(vcpu->hv_clock));
1832
1833	smp_wmb();
1834
1835	vcpu->hv_clock.version++;
1836	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1837				&vcpu->hv_clock,
1838				sizeof(vcpu->hv_clock.version));
1839	return 0;
1840}
1841
1842/*
1843 * kvmclock updates which are isolated to a given vcpu, such as
1844 * vcpu->cpu migration, should not allow system_timestamp from
1845 * the rest of the vcpus to remain static. Otherwise ntp frequency
1846 * correction applies to one vcpu's system_timestamp but not
1847 * the others.
1848 *
1849 * So in those cases, request a kvmclock update for all vcpus.
1850 * We need to rate-limit these requests though, as they can
1851 * considerably slow guests that have a large number of vcpus.
1852 * The time for a remote vcpu to update its kvmclock is bound
1853 * by the delay we use to rate-limit the updates.
1854 */
1855
1856#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1857
1858static void kvmclock_update_fn(struct work_struct *work)
1859{
1860	int i;
1861	struct delayed_work *dwork = to_delayed_work(work);
1862	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1863					   kvmclock_update_work);
1864	struct kvm *kvm = container_of(ka, struct kvm, arch);
1865	struct kvm_vcpu *vcpu;
1866
1867	kvm_for_each_vcpu(i, vcpu, kvm) {
1868		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1869		kvm_vcpu_kick(vcpu);
1870	}
1871}
1872
1873static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1874{
1875	struct kvm *kvm = v->kvm;
1876
1877	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1878	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1879					KVMCLOCK_UPDATE_DELAY);
1880}
1881
1882#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1883
1884static void kvmclock_sync_fn(struct work_struct *work)
1885{
1886	struct delayed_work *dwork = to_delayed_work(work);
1887	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1888					   kvmclock_sync_work);
1889	struct kvm *kvm = container_of(ka, struct kvm, arch);
1890
1891	if (!kvmclock_periodic_sync)
1892		return;
1893
1894	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1895	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1896					KVMCLOCK_SYNC_PERIOD);
1897}
1898
1899static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1900{
1901	u64 mcg_cap = vcpu->arch.mcg_cap;
1902	unsigned bank_num = mcg_cap & 0xff;
1903
1904	switch (msr) {
1905	case MSR_IA32_MCG_STATUS:
1906		vcpu->arch.mcg_status = data;
1907		break;
1908	case MSR_IA32_MCG_CTL:
1909		if (!(mcg_cap & MCG_CTL_P))
1910			return 1;
1911		if (data != 0 && data != ~(u64)0)
1912			return -1;
1913		vcpu->arch.mcg_ctl = data;
1914		break;
1915	default:
1916		if (msr >= MSR_IA32_MC0_CTL &&
1917		    msr < MSR_IA32_MCx_CTL(bank_num)) {
1918			u32 offset = msr - MSR_IA32_MC0_CTL;
1919			/* only 0 or all 1s can be written to IA32_MCi_CTL
1920			 * some Linux kernels though clear bit 10 in bank 4 to
1921			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1922			 * this to avoid an uncatched #GP in the guest
1923			 */
1924			if ((offset & 0x3) == 0 &&
1925			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1926				return -1;
1927			vcpu->arch.mce_banks[offset] = data;
1928			break;
1929		}
1930		return 1;
1931	}
1932	return 0;
1933}
1934
1935static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1936{
1937	struct kvm *kvm = vcpu->kvm;
1938	int lm = is_long_mode(vcpu);
1939	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1940		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1941	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1942		: kvm->arch.xen_hvm_config.blob_size_32;
1943	u32 page_num = data & ~PAGE_MASK;
1944	u64 page_addr = data & PAGE_MASK;
1945	u8 *page;
1946	int r;
1947
1948	r = -E2BIG;
1949	if (page_num >= blob_size)
1950		goto out;
1951	r = -ENOMEM;
1952	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1953	if (IS_ERR(page)) {
1954		r = PTR_ERR(page);
1955		goto out;
1956	}
1957	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1958		goto out_free;
1959	r = 0;
1960out_free:
1961	kfree(page);
1962out:
1963	return r;
1964}
1965
1966static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1967{
1968	gpa_t gpa = data & ~0x3f;
1969
1970	/* Bits 2:5 are reserved, Should be zero */
1971	if (data & 0x3c)
1972		return 1;
1973
1974	vcpu->arch.apf.msr_val = data;
1975
1976	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1977		kvm_clear_async_pf_completion_queue(vcpu);
1978		kvm_async_pf_hash_reset(vcpu);
1979		return 0;
1980	}
1981
1982	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1983					sizeof(u32)))
1984		return 1;
1985
1986	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1987	kvm_async_pf_wakeup_all(vcpu);
1988	return 0;
1989}
1990
1991static void kvmclock_reset(struct kvm_vcpu *vcpu)
1992{
1993	vcpu->arch.pv_time_enabled = false;
1994}
1995
1996static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1997{
1998	u64 delta;
1999
2000	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2001		return;
2002
2003	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2004	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2005	vcpu->arch.st.accum_steal = delta;
2006}
2007
2008static void record_steal_time(struct kvm_vcpu *vcpu)
2009{
2010	accumulate_steal_time(vcpu);
2011
2012	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2013		return;
2014
2015	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2016		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2017		return;
2018
2019	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2020	vcpu->arch.st.steal.version += 2;
2021	vcpu->arch.st.accum_steal = 0;
2022
2023	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2024		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2025}
2026
2027int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2028{
2029	bool pr = false;
2030	u32 msr = msr_info->index;
2031	u64 data = msr_info->data;
2032
2033	switch (msr) {
2034	case MSR_AMD64_NB_CFG:
2035	case MSR_IA32_UCODE_REV:
2036	case MSR_IA32_UCODE_WRITE:
2037	case MSR_VM_HSAVE_PA:
2038	case MSR_AMD64_PATCH_LOADER:
2039	case MSR_AMD64_BU_CFG2:
2040		break;
2041
2042	case MSR_EFER:
2043		return set_efer(vcpu, data);
2044	case MSR_K7_HWCR:
2045		data &= ~(u64)0x40;	/* ignore flush filter disable */
2046		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2047		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2048		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2049		if (data != 0) {
2050			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2051				    data);
2052			return 1;
2053		}
2054		break;
2055	case MSR_FAM10H_MMIO_CONF_BASE:
2056		if (data != 0) {
2057			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2058				    "0x%llx\n", data);
2059			return 1;
2060		}
2061		break;
2062	case MSR_IA32_DEBUGCTLMSR:
2063		if (!data) {
2064			/* We support the non-activated case already */
2065			break;
2066		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2067			/* Values other than LBR and BTF are vendor-specific,
2068			   thus reserved and should throw a #GP */
2069			return 1;
2070		}
2071		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2072			    __func__, data);
2073		break;
2074	case 0x200 ... 0x2ff:
2075		return kvm_mtrr_set_msr(vcpu, msr, data);
2076	case MSR_IA32_APICBASE:
2077		return kvm_set_apic_base(vcpu, msr_info);
2078	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2079		return kvm_x2apic_msr_write(vcpu, msr, data);
2080	case MSR_IA32_TSCDEADLINE:
2081		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2082		break;
2083	case MSR_IA32_TSC_ADJUST:
2084		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2085			if (!msr_info->host_initiated) {
2086				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2087				adjust_tsc_offset_guest(vcpu, adj);
2088			}
2089			vcpu->arch.ia32_tsc_adjust_msr = data;
2090		}
2091		break;
2092	case MSR_IA32_MISC_ENABLE:
2093		vcpu->arch.ia32_misc_enable_msr = data;
2094		break;
2095	case MSR_IA32_SMBASE:
2096		if (!msr_info->host_initiated)
2097			return 1;
2098		vcpu->arch.smbase = data;
2099		break;
2100	case MSR_KVM_WALL_CLOCK_NEW:
2101	case MSR_KVM_WALL_CLOCK:
2102		vcpu->kvm->arch.wall_clock = data;
2103		kvm_write_wall_clock(vcpu->kvm, data);
2104		break;
2105	case MSR_KVM_SYSTEM_TIME_NEW:
2106	case MSR_KVM_SYSTEM_TIME: {
2107		u64 gpa_offset;
2108		struct kvm_arch *ka = &vcpu->kvm->arch;
2109
2110		kvmclock_reset(vcpu);
2111
2112		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2113			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2114
2115			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2116				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2117					&vcpu->requests);
2118
2119			ka->boot_vcpu_runs_old_kvmclock = tmp;
2120		}
2121
2122		vcpu->arch.time = data;
2123		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2124
2125		/* we verify if the enable bit is set... */
2126		if (!(data & 1))
2127			break;
2128
2129		gpa_offset = data & ~(PAGE_MASK | 1);
2130
2131		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2132		     &vcpu->arch.pv_time, data & ~1ULL,
2133		     sizeof(struct pvclock_vcpu_time_info)))
2134			vcpu->arch.pv_time_enabled = false;
2135		else
2136			vcpu->arch.pv_time_enabled = true;
2137
2138		break;
2139	}
2140	case MSR_KVM_ASYNC_PF_EN:
2141		if (kvm_pv_enable_async_pf(vcpu, data))
2142			return 1;
2143		break;
2144	case MSR_KVM_STEAL_TIME:
2145
2146		if (unlikely(!sched_info_on()))
2147			return 1;
2148
2149		if (data & KVM_STEAL_RESERVED_MASK)
2150			return 1;
2151
2152		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2153						data & KVM_STEAL_VALID_BITS,
2154						sizeof(struct kvm_steal_time)))
2155			return 1;
2156
2157		vcpu->arch.st.msr_val = data;
2158
2159		if (!(data & KVM_MSR_ENABLED))
2160			break;
2161
2162		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2163
2164		break;
2165	case MSR_KVM_PV_EOI_EN:
2166		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2167			return 1;
2168		break;
2169
2170	case MSR_IA32_MCG_CTL:
2171	case MSR_IA32_MCG_STATUS:
2172	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2173		return set_msr_mce(vcpu, msr, data);
2174
2175	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2176	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2177		pr = true; /* fall through */
2178	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2179	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2180		if (kvm_pmu_is_valid_msr(vcpu, msr))
2181			return kvm_pmu_set_msr(vcpu, msr_info);
2182
2183		if (pr || data != 0)
2184			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2185				    "0x%x data 0x%llx\n", msr, data);
2186		break;
2187	case MSR_K7_CLK_CTL:
2188		/*
2189		 * Ignore all writes to this no longer documented MSR.
2190		 * Writes are only relevant for old K7 processors,
2191		 * all pre-dating SVM, but a recommended workaround from
2192		 * AMD for these chips. It is possible to specify the
2193		 * affected processor models on the command line, hence
2194		 * the need to ignore the workaround.
2195		 */
2196		break;
2197	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2198	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2199	case HV_X64_MSR_CRASH_CTL:
2200		return kvm_hv_set_msr_common(vcpu, msr, data,
2201					     msr_info->host_initiated);
2202	case MSR_IA32_BBL_CR_CTL3:
2203		/* Drop writes to this legacy MSR -- see rdmsr
2204		 * counterpart for further detail.
2205		 */
2206		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2207		break;
2208	case MSR_AMD64_OSVW_ID_LENGTH:
2209		if (!guest_cpuid_has_osvw(vcpu))
2210			return 1;
2211		vcpu->arch.osvw.length = data;
2212		break;
2213	case MSR_AMD64_OSVW_STATUS:
2214		if (!guest_cpuid_has_osvw(vcpu))
2215			return 1;
2216		vcpu->arch.osvw.status = data;
2217		break;
2218	default:
2219		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2220			return xen_hvm_config(vcpu, data);
2221		if (kvm_pmu_is_valid_msr(vcpu, msr))
2222			return kvm_pmu_set_msr(vcpu, msr_info);
2223		if (!ignore_msrs) {
2224			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2225				    msr, data);
2226			return 1;
2227		} else {
2228			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2229				    msr, data);
2230			break;
2231		}
2232	}
2233	return 0;
2234}
2235EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2236
2237
2238/*
2239 * Reads an msr value (of 'msr_index') into 'pdata'.
2240 * Returns 0 on success, non-0 otherwise.
2241 * Assumes vcpu_load() was already called.
2242 */
2243int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2244{
2245	return kvm_x86_ops->get_msr(vcpu, msr);
2246}
2247EXPORT_SYMBOL_GPL(kvm_get_msr);
2248
2249static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2250{
2251	u64 data;
2252	u64 mcg_cap = vcpu->arch.mcg_cap;
2253	unsigned bank_num = mcg_cap & 0xff;
2254
2255	switch (msr) {
2256	case MSR_IA32_P5_MC_ADDR:
2257	case MSR_IA32_P5_MC_TYPE:
2258		data = 0;
2259		break;
2260	case MSR_IA32_MCG_CAP:
2261		data = vcpu->arch.mcg_cap;
2262		break;
2263	case MSR_IA32_MCG_CTL:
2264		if (!(mcg_cap & MCG_CTL_P))
2265			return 1;
2266		data = vcpu->arch.mcg_ctl;
2267		break;
2268	case MSR_IA32_MCG_STATUS:
2269		data = vcpu->arch.mcg_status;
2270		break;
2271	default:
2272		if (msr >= MSR_IA32_MC0_CTL &&
2273		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2274			u32 offset = msr - MSR_IA32_MC0_CTL;
2275			data = vcpu->arch.mce_banks[offset];
2276			break;
2277		}
2278		return 1;
2279	}
2280	*pdata = data;
2281	return 0;
2282}
2283
2284int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2285{
2286	switch (msr_info->index) {
2287	case MSR_IA32_PLATFORM_ID:
2288	case MSR_IA32_EBL_CR_POWERON:
2289	case MSR_IA32_DEBUGCTLMSR:
2290	case MSR_IA32_LASTBRANCHFROMIP:
2291	case MSR_IA32_LASTBRANCHTOIP:
2292	case MSR_IA32_LASTINTFROMIP:
2293	case MSR_IA32_LASTINTTOIP:
2294	case MSR_K8_SYSCFG:
2295	case MSR_K8_TSEG_ADDR:
2296	case MSR_K8_TSEG_MASK:
2297	case MSR_K7_HWCR:
2298	case MSR_VM_HSAVE_PA:
2299	case MSR_K8_INT_PENDING_MSG:
2300	case MSR_AMD64_NB_CFG:
2301	case MSR_FAM10H_MMIO_CONF_BASE:
2302	case MSR_AMD64_BU_CFG2:
2303		msr_info->data = 0;
2304		break;
2305	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2306	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2307	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2308	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2309		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2310			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2311		msr_info->data = 0;
2312		break;
2313	case MSR_IA32_UCODE_REV:
2314		msr_info->data = 0x100000000ULL;
2315		break;
2316	case MSR_MTRRcap:
2317	case 0x200 ... 0x2ff:
2318		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2319	case 0xcd: /* fsb frequency */
2320		msr_info->data = 3;
2321		break;
2322		/*
2323		 * MSR_EBC_FREQUENCY_ID
2324		 * Conservative value valid for even the basic CPU models.
2325		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2326		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2327		 * and 266MHz for model 3, or 4. Set Core Clock
2328		 * Frequency to System Bus Frequency Ratio to 1 (bits
2329		 * 31:24) even though these are only valid for CPU
2330		 * models > 2, however guests may end up dividing or
2331		 * multiplying by zero otherwise.
2332		 */
2333	case MSR_EBC_FREQUENCY_ID:
2334		msr_info->data = 1 << 24;
2335		break;
2336	case MSR_IA32_APICBASE:
2337		msr_info->data = kvm_get_apic_base(vcpu);
2338		break;
2339	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2340		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2341		break;
2342	case MSR_IA32_TSCDEADLINE:
2343		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2344		break;
2345	case MSR_IA32_TSC_ADJUST:
2346		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2347		break;
2348	case MSR_IA32_MISC_ENABLE:
2349		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2350		break;
2351	case MSR_IA32_SMBASE:
2352		if (!msr_info->host_initiated)
2353			return 1;
2354		msr_info->data = vcpu->arch.smbase;
2355		break;
2356	case MSR_IA32_PERF_STATUS:
2357		/* TSC increment by tick */
2358		msr_info->data = 1000ULL;
2359		/* CPU multiplier */
2360		msr_info->data |= (((uint64_t)4ULL) << 40);
2361		break;
2362	case MSR_EFER:
2363		msr_info->data = vcpu->arch.efer;
2364		break;
2365	case MSR_KVM_WALL_CLOCK:
2366	case MSR_KVM_WALL_CLOCK_NEW:
2367		msr_info->data = vcpu->kvm->arch.wall_clock;
2368		break;
2369	case MSR_KVM_SYSTEM_TIME:
2370	case MSR_KVM_SYSTEM_TIME_NEW:
2371		msr_info->data = vcpu->arch.time;
2372		break;
2373	case MSR_KVM_ASYNC_PF_EN:
2374		msr_info->data = vcpu->arch.apf.msr_val;
2375		break;
2376	case MSR_KVM_STEAL_TIME:
2377		msr_info->data = vcpu->arch.st.msr_val;
2378		break;
2379	case MSR_KVM_PV_EOI_EN:
2380		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2381		break;
2382	case MSR_IA32_P5_MC_ADDR:
2383	case MSR_IA32_P5_MC_TYPE:
2384	case MSR_IA32_MCG_CAP:
2385	case MSR_IA32_MCG_CTL:
2386	case MSR_IA32_MCG_STATUS:
2387	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2388		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2389	case MSR_K7_CLK_CTL:
2390		/*
2391		 * Provide expected ramp-up count for K7. All other
2392		 * are set to zero, indicating minimum divisors for
2393		 * every field.
2394		 *
2395		 * This prevents guest kernels on AMD host with CPU
2396		 * type 6, model 8 and higher from exploding due to
2397		 * the rdmsr failing.
2398		 */
2399		msr_info->data = 0x20000000;
2400		break;
2401	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2402	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2403	case HV_X64_MSR_CRASH_CTL:
2404		return kvm_hv_get_msr_common(vcpu,
2405					     msr_info->index, &msr_info->data);
2406		break;
2407	case MSR_IA32_BBL_CR_CTL3:
2408		/* This legacy MSR exists but isn't fully documented in current
2409		 * silicon.  It is however accessed by winxp in very narrow
2410		 * scenarios where it sets bit #19, itself documented as
2411		 * a "reserved" bit.  Best effort attempt to source coherent
2412		 * read data here should the balance of the register be
2413		 * interpreted by the guest:
2414		 *
2415		 * L2 cache control register 3: 64GB range, 256KB size,
2416		 * enabled, latency 0x1, configured
2417		 */
2418		msr_info->data = 0xbe702111;
2419		break;
2420	case MSR_AMD64_OSVW_ID_LENGTH:
2421		if (!guest_cpuid_has_osvw(vcpu))
2422			return 1;
2423		msr_info->data = vcpu->arch.osvw.length;
2424		break;
2425	case MSR_AMD64_OSVW_STATUS:
2426		if (!guest_cpuid_has_osvw(vcpu))
2427			return 1;
2428		msr_info->data = vcpu->arch.osvw.status;
2429		break;
2430	default:
2431		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2432			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2433		if (!ignore_msrs) {
2434			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2435			return 1;
2436		} else {
2437			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2438			msr_info->data = 0;
2439		}
2440		break;
2441	}
2442	return 0;
2443}
2444EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2445
2446/*
2447 * Read or write a bunch of msrs. All parameters are kernel addresses.
2448 *
2449 * @return number of msrs set successfully.
2450 */
2451static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2452		    struct kvm_msr_entry *entries,
2453		    int (*do_msr)(struct kvm_vcpu *vcpu,
2454				  unsigned index, u64 *data))
2455{
2456	int i, idx;
2457
2458	idx = srcu_read_lock(&vcpu->kvm->srcu);
2459	for (i = 0; i < msrs->nmsrs; ++i)
2460		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2461			break;
2462	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2463
2464	return i;
2465}
2466
2467/*
2468 * Read or write a bunch of msrs. Parameters are user addresses.
2469 *
2470 * @return number of msrs set successfully.
2471 */
2472static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2473		  int (*do_msr)(struct kvm_vcpu *vcpu,
2474				unsigned index, u64 *data),
2475		  int writeback)
2476{
2477	struct kvm_msrs msrs;
2478	struct kvm_msr_entry *entries;
2479	int r, n;
2480	unsigned size;
2481
2482	r = -EFAULT;
2483	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2484		goto out;
2485
2486	r = -E2BIG;
2487	if (msrs.nmsrs >= MAX_IO_MSRS)
2488		goto out;
2489
2490	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2491	entries = memdup_user(user_msrs->entries, size);
2492	if (IS_ERR(entries)) {
2493		r = PTR_ERR(entries);
2494		goto out;
2495	}
2496
2497	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2498	if (r < 0)
2499		goto out_free;
2500
2501	r = -EFAULT;
2502	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2503		goto out_free;
2504
2505	r = n;
2506
2507out_free:
2508	kfree(entries);
2509out:
2510	return r;
2511}
2512
2513int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2514{
2515	int r;
2516
2517	switch (ext) {
2518	case KVM_CAP_IRQCHIP:
2519	case KVM_CAP_HLT:
2520	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2521	case KVM_CAP_SET_TSS_ADDR:
2522	case KVM_CAP_EXT_CPUID:
2523	case KVM_CAP_EXT_EMUL_CPUID:
2524	case KVM_CAP_CLOCKSOURCE:
2525	case KVM_CAP_PIT:
2526	case KVM_CAP_NOP_IO_DELAY:
2527	case KVM_CAP_MP_STATE:
2528	case KVM_CAP_SYNC_MMU:
2529	case KVM_CAP_USER_NMI:
2530	case KVM_CAP_REINJECT_CONTROL:
2531	case KVM_CAP_IRQ_INJECT_STATUS:
2532	case KVM_CAP_IOEVENTFD:
2533	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2534	case KVM_CAP_PIT2:
2535	case KVM_CAP_PIT_STATE2:
2536	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2537	case KVM_CAP_XEN_HVM:
2538	case KVM_CAP_ADJUST_CLOCK:
2539	case KVM_CAP_VCPU_EVENTS:
2540	case KVM_CAP_HYPERV:
2541	case KVM_CAP_HYPERV_VAPIC:
2542	case KVM_CAP_HYPERV_SPIN:
2543	case KVM_CAP_PCI_SEGMENT:
2544	case KVM_CAP_DEBUGREGS:
2545	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2546	case KVM_CAP_XSAVE:
2547	case KVM_CAP_ASYNC_PF:
2548	case KVM_CAP_GET_TSC_KHZ:
2549	case KVM_CAP_KVMCLOCK_CTRL:
2550	case KVM_CAP_READONLY_MEM:
2551	case KVM_CAP_HYPERV_TIME:
2552	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2553	case KVM_CAP_TSC_DEADLINE_TIMER:
2554	case KVM_CAP_ENABLE_CAP_VM:
2555	case KVM_CAP_DISABLE_QUIRKS:
2556	case KVM_CAP_SET_BOOT_CPU_ID:
2557 	case KVM_CAP_SPLIT_IRQCHIP:
2558#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2559	case KVM_CAP_ASSIGN_DEV_IRQ:
2560	case KVM_CAP_PCI_2_3:
2561#endif
2562		r = 1;
2563		break;
2564	case KVM_CAP_X86_SMM:
2565		/* SMBASE is usually relocated above 1M on modern chipsets,
2566		 * and SMM handlers might indeed rely on 4G segment limits,
2567		 * so do not report SMM to be available if real mode is
2568		 * emulated via vm86 mode.  Still, do not go to great lengths
2569		 * to avoid userspace's usage of the feature, because it is a
2570		 * fringe case that is not enabled except via specific settings
2571		 * of the module parameters.
2572		 */
2573		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2574		break;
2575	case KVM_CAP_COALESCED_MMIO:
2576		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2577		break;
2578	case KVM_CAP_VAPIC:
2579		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2580		break;
2581	case KVM_CAP_NR_VCPUS:
2582		r = KVM_SOFT_MAX_VCPUS;
2583		break;
2584	case KVM_CAP_MAX_VCPUS:
2585		r = KVM_MAX_VCPUS;
2586		break;
2587	case KVM_CAP_NR_MEMSLOTS:
2588		r = KVM_USER_MEM_SLOTS;
2589		break;
2590	case KVM_CAP_PV_MMU:	/* obsolete */
2591		r = 0;
2592		break;
2593#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2594	case KVM_CAP_IOMMU:
2595		r = iommu_present(&pci_bus_type);
2596		break;
2597#endif
2598	case KVM_CAP_MCE:
2599		r = KVM_MAX_MCE_BANKS;
2600		break;
2601	case KVM_CAP_XCRS:
2602		r = cpu_has_xsave;
2603		break;
2604	case KVM_CAP_TSC_CONTROL:
2605		r = kvm_has_tsc_control;
2606		break;
2607	default:
2608		r = 0;
2609		break;
2610	}
2611	return r;
2612
2613}
2614
2615long kvm_arch_dev_ioctl(struct file *filp,
2616			unsigned int ioctl, unsigned long arg)
2617{
2618	void __user *argp = (void __user *)arg;
2619	long r;
2620
2621	switch (ioctl) {
2622	case KVM_GET_MSR_INDEX_LIST: {
2623		struct kvm_msr_list __user *user_msr_list = argp;
2624		struct kvm_msr_list msr_list;
2625		unsigned n;
2626
2627		r = -EFAULT;
2628		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2629			goto out;
2630		n = msr_list.nmsrs;
2631		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2632		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2633			goto out;
2634		r = -E2BIG;
2635		if (n < msr_list.nmsrs)
2636			goto out;
2637		r = -EFAULT;
2638		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2639				 num_msrs_to_save * sizeof(u32)))
2640			goto out;
2641		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2642				 &emulated_msrs,
2643				 num_emulated_msrs * sizeof(u32)))
2644			goto out;
2645		r = 0;
2646		break;
2647	}
2648	case KVM_GET_SUPPORTED_CPUID:
2649	case KVM_GET_EMULATED_CPUID: {
2650		struct kvm_cpuid2 __user *cpuid_arg = argp;
2651		struct kvm_cpuid2 cpuid;
2652
2653		r = -EFAULT;
2654		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2655			goto out;
2656
2657		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2658					    ioctl);
2659		if (r)
2660			goto out;
2661
2662		r = -EFAULT;
2663		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2664			goto out;
2665		r = 0;
2666		break;
2667	}
2668	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2669		u64 mce_cap;
2670
2671		mce_cap = KVM_MCE_CAP_SUPPORTED;
2672		r = -EFAULT;
2673		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2674			goto out;
2675		r = 0;
2676		break;
2677	}
2678	default:
2679		r = -EINVAL;
2680	}
2681out:
2682	return r;
2683}
2684
2685static void wbinvd_ipi(void *garbage)
2686{
2687	wbinvd();
2688}
2689
2690static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2691{
2692	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2693}
2694
2695void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2696{
2697	/* Address WBINVD may be executed by guest */
2698	if (need_emulate_wbinvd(vcpu)) {
2699		if (kvm_x86_ops->has_wbinvd_exit())
2700			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2701		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2702			smp_call_function_single(vcpu->cpu,
2703					wbinvd_ipi, NULL, 1);
2704	}
2705
2706	kvm_x86_ops->vcpu_load(vcpu, cpu);
2707
2708	/* Apply any externally detected TSC adjustments (due to suspend) */
2709	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2710		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2711		vcpu->arch.tsc_offset_adjustment = 0;
2712		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2713	}
2714
2715	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2716		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2717				rdtsc() - vcpu->arch.last_host_tsc;
2718		if (tsc_delta < 0)
2719			mark_tsc_unstable("KVM discovered backwards TSC");
2720		if (check_tsc_unstable()) {
2721			u64 offset = kvm_compute_tsc_offset(vcpu,
2722						vcpu->arch.last_guest_tsc);
2723			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2724			vcpu->arch.tsc_catchup = 1;
2725		}
2726		/*
2727		 * On a host with synchronized TSC, there is no need to update
2728		 * kvmclock on vcpu->cpu migration
2729		 */
2730		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2731			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2732		if (vcpu->cpu != cpu)
2733			kvm_migrate_timers(vcpu);
2734		vcpu->cpu = cpu;
2735	}
2736
2737	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2738	vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
2739}
2740
2741void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2742{
2743	kvm_x86_ops->vcpu_put(vcpu);
2744	kvm_put_guest_fpu(vcpu);
2745	vcpu->arch.last_host_tsc = rdtsc();
2746}
2747
2748static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2749				    struct kvm_lapic_state *s)
2750{
2751	kvm_x86_ops->sync_pir_to_irr(vcpu);
2752	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2753
2754	return 0;
2755}
2756
2757static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2758				    struct kvm_lapic_state *s)
2759{
2760	kvm_apic_post_state_restore(vcpu, s);
2761	update_cr8_intercept(vcpu);
2762
2763	return 0;
2764}
2765
2766static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2767{
2768	return (!lapic_in_kernel(vcpu) ||
2769		kvm_apic_accept_pic_intr(vcpu));
2770}
2771
2772/*
2773 * if userspace requested an interrupt window, check that the
2774 * interrupt window is open.
2775 *
2776 * No need to exit to userspace if we already have an interrupt queued.
2777 */
2778static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2779{
2780	return kvm_arch_interrupt_allowed(vcpu) &&
2781		!kvm_cpu_has_interrupt(vcpu) &&
2782		!kvm_event_needs_reinjection(vcpu) &&
2783		kvm_cpu_accept_dm_intr(vcpu);
2784}
2785
2786static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2787				    struct kvm_interrupt *irq)
2788{
2789	if (irq->irq >= KVM_NR_INTERRUPTS)
2790		return -EINVAL;
2791
2792	if (!irqchip_in_kernel(vcpu->kvm)) {
2793		kvm_queue_interrupt(vcpu, irq->irq, false);
2794		kvm_make_request(KVM_REQ_EVENT, vcpu);
2795		return 0;
2796	}
2797
2798	/*
2799	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2800	 * fail for in-kernel 8259.
2801	 */
2802	if (pic_in_kernel(vcpu->kvm))
2803		return -ENXIO;
2804
2805	if (vcpu->arch.pending_external_vector != -1)
2806		return -EEXIST;
2807
2808	vcpu->arch.pending_external_vector = irq->irq;
2809	kvm_make_request(KVM_REQ_EVENT, vcpu);
2810	return 0;
2811}
2812
2813static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2814{
2815	kvm_inject_nmi(vcpu);
2816
2817	return 0;
2818}
2819
2820static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2821{
2822	kvm_make_request(KVM_REQ_SMI, vcpu);
2823
2824	return 0;
2825}
2826
2827static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2828					   struct kvm_tpr_access_ctl *tac)
2829{
2830	if (tac->flags)
2831		return -EINVAL;
2832	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2833	return 0;
2834}
2835
2836static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2837					u64 mcg_cap)
2838{
2839	int r;
2840	unsigned bank_num = mcg_cap & 0xff, bank;
2841
2842	r = -EINVAL;
2843	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2844		goto out;
2845	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2846		goto out;
2847	r = 0;
2848	vcpu->arch.mcg_cap = mcg_cap;
2849	/* Init IA32_MCG_CTL to all 1s */
2850	if (mcg_cap & MCG_CTL_P)
2851		vcpu->arch.mcg_ctl = ~(u64)0;
2852	/* Init IA32_MCi_CTL to all 1s */
2853	for (bank = 0; bank < bank_num; bank++)
2854		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2855out:
2856	return r;
2857}
2858
2859static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2860				      struct kvm_x86_mce *mce)
2861{
2862	u64 mcg_cap = vcpu->arch.mcg_cap;
2863	unsigned bank_num = mcg_cap & 0xff;
2864	u64 *banks = vcpu->arch.mce_banks;
2865
2866	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2867		return -EINVAL;
2868	/*
2869	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2870	 * reporting is disabled
2871	 */
2872	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2873	    vcpu->arch.mcg_ctl != ~(u64)0)
2874		return 0;
2875	banks += 4 * mce->bank;
2876	/*
2877	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2878	 * reporting is disabled for the bank
2879	 */
2880	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2881		return 0;
2882	if (mce->status & MCI_STATUS_UC) {
2883		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2884		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2885			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2886			return 0;
2887		}
2888		if (banks[1] & MCI_STATUS_VAL)
2889			mce->status |= MCI_STATUS_OVER;
2890		banks[2] = mce->addr;
2891		banks[3] = mce->misc;
2892		vcpu->arch.mcg_status = mce->mcg_status;
2893		banks[1] = mce->status;
2894		kvm_queue_exception(vcpu, MC_VECTOR);
2895	} else if (!(banks[1] & MCI_STATUS_VAL)
2896		   || !(banks[1] & MCI_STATUS_UC)) {
2897		if (banks[1] & MCI_STATUS_VAL)
2898			mce->status |= MCI_STATUS_OVER;
2899		banks[2] = mce->addr;
2900		banks[3] = mce->misc;
2901		banks[1] = mce->status;
2902	} else
2903		banks[1] |= MCI_STATUS_OVER;
2904	return 0;
2905}
2906
2907static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2908					       struct kvm_vcpu_events *events)
2909{
2910	process_nmi(vcpu);
2911	events->exception.injected =
2912		vcpu->arch.exception.pending &&
2913		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2914	events->exception.nr = vcpu->arch.exception.nr;
2915	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2916	events->exception.pad = 0;
2917	events->exception.error_code = vcpu->arch.exception.error_code;
2918
2919	events->interrupt.injected =
2920		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2921	events->interrupt.nr = vcpu->arch.interrupt.nr;
2922	events->interrupt.soft = 0;
2923	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2924
2925	events->nmi.injected = vcpu->arch.nmi_injected;
2926	events->nmi.pending = vcpu->arch.nmi_pending != 0;
2927	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2928	events->nmi.pad = 0;
2929
2930	events->sipi_vector = 0; /* never valid when reporting to user space */
2931
2932	events->smi.smm = is_smm(vcpu);
2933	events->smi.pending = vcpu->arch.smi_pending;
2934	events->smi.smm_inside_nmi =
2935		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2936	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2937
2938	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2939			 | KVM_VCPUEVENT_VALID_SHADOW
2940			 | KVM_VCPUEVENT_VALID_SMM);
2941	memset(&events->reserved, 0, sizeof(events->reserved));
2942}
2943
2944static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2945					      struct kvm_vcpu_events *events)
2946{
2947	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2948			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2949			      | KVM_VCPUEVENT_VALID_SHADOW
2950			      | KVM_VCPUEVENT_VALID_SMM))
2951		return -EINVAL;
2952
2953	process_nmi(vcpu);
2954	vcpu->arch.exception.pending = events->exception.injected;
2955	vcpu->arch.exception.nr = events->exception.nr;
2956	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2957	vcpu->arch.exception.error_code = events->exception.error_code;
2958
2959	vcpu->arch.interrupt.pending = events->interrupt.injected;
2960	vcpu->arch.interrupt.nr = events->interrupt.nr;
2961	vcpu->arch.interrupt.soft = events->interrupt.soft;
2962	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2963		kvm_x86_ops->set_interrupt_shadow(vcpu,
2964						  events->interrupt.shadow);
2965
2966	vcpu->arch.nmi_injected = events->nmi.injected;
2967	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2968		vcpu->arch.nmi_pending = events->nmi.pending;
2969	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2970
2971	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2972	    kvm_vcpu_has_lapic(vcpu))
2973		vcpu->arch.apic->sipi_vector = events->sipi_vector;
2974
2975	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2976		if (events->smi.smm)
2977			vcpu->arch.hflags |= HF_SMM_MASK;
2978		else
2979			vcpu->arch.hflags &= ~HF_SMM_MASK;
2980		vcpu->arch.smi_pending = events->smi.pending;
2981		if (events->smi.smm_inside_nmi)
2982			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2983		else
2984			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2985		if (kvm_vcpu_has_lapic(vcpu)) {
2986			if (events->smi.latched_init)
2987				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2988			else
2989				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2990		}
2991	}
2992
2993	kvm_make_request(KVM_REQ_EVENT, vcpu);
2994
2995	return 0;
2996}
2997
2998static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2999					     struct kvm_debugregs *dbgregs)
3000{
3001	unsigned long val;
3002
3003	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3004	kvm_get_dr(vcpu, 6, &val);
3005	dbgregs->dr6 = val;
3006	dbgregs->dr7 = vcpu->arch.dr7;
3007	dbgregs->flags = 0;
3008	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3009}
3010
3011static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3012					    struct kvm_debugregs *dbgregs)
3013{
3014	if (dbgregs->flags)
3015		return -EINVAL;
3016
3017	if (dbgregs->dr6 & ~0xffffffffull)
3018		return -EINVAL;
3019	if (dbgregs->dr7 & ~0xffffffffull)
3020		return -EINVAL;
3021
3022	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3023	kvm_update_dr0123(vcpu);
3024	vcpu->arch.dr6 = dbgregs->dr6;
3025	kvm_update_dr6(vcpu);
3026	vcpu->arch.dr7 = dbgregs->dr7;
3027	kvm_update_dr7(vcpu);
3028
3029	return 0;
3030}
3031
3032#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3033
3034static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3035{
3036	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3037	u64 xstate_bv = xsave->header.xfeatures;
3038	u64 valid;
3039
3040	/*
3041	 * Copy legacy XSAVE area, to avoid complications with CPUID
3042	 * leaves 0 and 1 in the loop below.
3043	 */
3044	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3045
3046	/* Set XSTATE_BV */
3047	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3048
3049	/*
3050	 * Copy each region from the possibly compacted offset to the
3051	 * non-compacted offset.
3052	 */
3053	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3054	while (valid) {
3055		u64 feature = valid & -valid;
3056		int index = fls64(feature) - 1;
3057		void *src = get_xsave_addr(xsave, feature);
3058
3059		if (src) {
3060			u32 size, offset, ecx, edx;
3061			cpuid_count(XSTATE_CPUID, index,
3062				    &size, &offset, &ecx, &edx);
3063			memcpy(dest + offset, src, size);
3064		}
3065
3066		valid -= feature;
3067	}
3068}
3069
3070static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3071{
3072	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3073	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3074	u64 valid;
3075
3076	/*
3077	 * Copy legacy XSAVE area, to avoid complications with CPUID
3078	 * leaves 0 and 1 in the loop below.
3079	 */
3080	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3081
3082	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3083	xsave->header.xfeatures = xstate_bv;
3084	if (cpu_has_xsaves)
3085		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3086
3087	/*
3088	 * Copy each region from the non-compacted offset to the
3089	 * possibly compacted offset.
3090	 */
3091	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3092	while (valid) {
3093		u64 feature = valid & -valid;
3094		int index = fls64(feature) - 1;
3095		void *dest = get_xsave_addr(xsave, feature);
3096
3097		if (dest) {
3098			u32 size, offset, ecx, edx;
3099			cpuid_count(XSTATE_CPUID, index,
3100				    &size, &offset, &ecx, &edx);
3101			memcpy(dest, src + offset, size);
3102		}
3103
3104		valid -= feature;
3105	}
3106}
3107
3108static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3109					 struct kvm_xsave *guest_xsave)
3110{
3111	if (cpu_has_xsave) {
3112		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3113		fill_xsave((u8 *) guest_xsave->region, vcpu);
3114	} else {
3115		memcpy(guest_xsave->region,
3116			&vcpu->arch.guest_fpu.state.fxsave,
3117			sizeof(struct fxregs_state));
3118		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3119			XFEATURE_MASK_FPSSE;
3120	}
3121}
3122
3123static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3124					struct kvm_xsave *guest_xsave)
3125{
3126	u64 xstate_bv =
3127		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3128
3129	if (cpu_has_xsave) {
3130		/*
3131		 * Here we allow setting states that are not present in
3132		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3133		 * with old userspace.
3134		 */
3135		if (xstate_bv & ~kvm_supported_xcr0())
3136			return -EINVAL;
3137		load_xsave(vcpu, (u8 *)guest_xsave->region);
3138	} else {
3139		if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3140			return -EINVAL;
3141		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3142			guest_xsave->region, sizeof(struct fxregs_state));
3143	}
3144	return 0;
3145}
3146
3147static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3148					struct kvm_xcrs *guest_xcrs)
3149{
3150	if (!cpu_has_xsave) {
3151		guest_xcrs->nr_xcrs = 0;
3152		return;
3153	}
3154
3155	guest_xcrs->nr_xcrs = 1;
3156	guest_xcrs->flags = 0;
3157	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3158	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3159}
3160
3161static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3162				       struct kvm_xcrs *guest_xcrs)
3163{
3164	int i, r = 0;
3165
3166	if (!cpu_has_xsave)
3167		return -EINVAL;
3168
3169	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3170		return -EINVAL;
3171
3172	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3173		/* Only support XCR0 currently */
3174		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3175			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3176				guest_xcrs->xcrs[i].value);
3177			break;
3178		}
3179	if (r)
3180		r = -EINVAL;
3181	return r;
3182}
3183
3184/*
3185 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3186 * stopped by the hypervisor.  This function will be called from the host only.
3187 * EINVAL is returned when the host attempts to set the flag for a guest that
3188 * does not support pv clocks.
3189 */
3190static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3191{
3192	if (!vcpu->arch.pv_time_enabled)
3193		return -EINVAL;
3194	vcpu->arch.pvclock_set_guest_stopped_request = true;
3195	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3196	return 0;
3197}
3198
3199long kvm_arch_vcpu_ioctl(struct file *filp,
3200			 unsigned int ioctl, unsigned long arg)
3201{
3202	struct kvm_vcpu *vcpu = filp->private_data;
3203	void __user *argp = (void __user *)arg;
3204	int r;
3205	union {
3206		struct kvm_lapic_state *lapic;
3207		struct kvm_xsave *xsave;
3208		struct kvm_xcrs *xcrs;
3209		void *buffer;
3210	} u;
3211
3212	u.buffer = NULL;
3213	switch (ioctl) {
3214	case KVM_GET_LAPIC: {
3215		r = -EINVAL;
3216		if (!vcpu->arch.apic)
3217			goto out;
3218		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3219
3220		r = -ENOMEM;
3221		if (!u.lapic)
3222			goto out;
3223		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3224		if (r)
3225			goto out;
3226		r = -EFAULT;
3227		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3228			goto out;
3229		r = 0;
3230		break;
3231	}
3232	case KVM_SET_LAPIC: {
3233		r = -EINVAL;
3234		if (!vcpu->arch.apic)
3235			goto out;
3236		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3237		if (IS_ERR(u.lapic))
3238			return PTR_ERR(u.lapic);
3239
3240		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3241		break;
3242	}
3243	case KVM_INTERRUPT: {
3244		struct kvm_interrupt irq;
3245
3246		r = -EFAULT;
3247		if (copy_from_user(&irq, argp, sizeof irq))
3248			goto out;
3249		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3250		break;
3251	}
3252	case KVM_NMI: {
3253		r = kvm_vcpu_ioctl_nmi(vcpu);
3254		break;
3255	}
3256	case KVM_SMI: {
3257		r = kvm_vcpu_ioctl_smi(vcpu);
3258		break;
3259	}
3260	case KVM_SET_CPUID: {
3261		struct kvm_cpuid __user *cpuid_arg = argp;
3262		struct kvm_cpuid cpuid;
3263
3264		r = -EFAULT;
3265		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3266			goto out;
3267		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3268		break;
3269	}
3270	case KVM_SET_CPUID2: {
3271		struct kvm_cpuid2 __user *cpuid_arg = argp;
3272		struct kvm_cpuid2 cpuid;
3273
3274		r = -EFAULT;
3275		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3276			goto out;
3277		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3278					      cpuid_arg->entries);
3279		break;
3280	}
3281	case KVM_GET_CPUID2: {
3282		struct kvm_cpuid2 __user *cpuid_arg = argp;
3283		struct kvm_cpuid2 cpuid;
3284
3285		r = -EFAULT;
3286		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3287			goto out;
3288		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3289					      cpuid_arg->entries);
3290		if (r)
3291			goto out;
3292		r = -EFAULT;
3293		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3294			goto out;
3295		r = 0;
3296		break;
3297	}
3298	case KVM_GET_MSRS:
3299		r = msr_io(vcpu, argp, do_get_msr, 1);
3300		break;
3301	case KVM_SET_MSRS:
3302		r = msr_io(vcpu, argp, do_set_msr, 0);
3303		break;
3304	case KVM_TPR_ACCESS_REPORTING: {
3305		struct kvm_tpr_access_ctl tac;
3306
3307		r = -EFAULT;
3308		if (copy_from_user(&tac, argp, sizeof tac))
3309			goto out;
3310		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3311		if (r)
3312			goto out;
3313		r = -EFAULT;
3314		if (copy_to_user(argp, &tac, sizeof tac))
3315			goto out;
3316		r = 0;
3317		break;
3318	};
3319	case KVM_SET_VAPIC_ADDR: {
3320		struct kvm_vapic_addr va;
3321
3322		r = -EINVAL;
3323		if (!lapic_in_kernel(vcpu))
3324			goto out;
3325		r = -EFAULT;
3326		if (copy_from_user(&va, argp, sizeof va))
3327			goto out;
3328		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3329		break;
3330	}
3331	case KVM_X86_SETUP_MCE: {
3332		u64 mcg_cap;
3333
3334		r = -EFAULT;
3335		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3336			goto out;
3337		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3338		break;
3339	}
3340	case KVM_X86_SET_MCE: {
3341		struct kvm_x86_mce mce;
3342
3343		r = -EFAULT;
3344		if (copy_from_user(&mce, argp, sizeof mce))
3345			goto out;
3346		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3347		break;
3348	}
3349	case KVM_GET_VCPU_EVENTS: {
3350		struct kvm_vcpu_events events;
3351
3352		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3353
3354		r = -EFAULT;
3355		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3356			break;
3357		r = 0;
3358		break;
3359	}
3360	case KVM_SET_VCPU_EVENTS: {
3361		struct kvm_vcpu_events events;
3362
3363		r = -EFAULT;
3364		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3365			break;
3366
3367		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3368		break;
3369	}
3370	case KVM_GET_DEBUGREGS: {
3371		struct kvm_debugregs dbgregs;
3372
3373		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3374
3375		r = -EFAULT;
3376		if (copy_to_user(argp, &dbgregs,
3377				 sizeof(struct kvm_debugregs)))
3378			break;
3379		r = 0;
3380		break;
3381	}
3382	case KVM_SET_DEBUGREGS: {
3383		struct kvm_debugregs dbgregs;
3384
3385		r = -EFAULT;
3386		if (copy_from_user(&dbgregs, argp,
3387				   sizeof(struct kvm_debugregs)))
3388			break;
3389
3390		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3391		break;
3392	}
3393	case KVM_GET_XSAVE: {
3394		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3395		r = -ENOMEM;
3396		if (!u.xsave)
3397			break;
3398
3399		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3400
3401		r = -EFAULT;
3402		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3403			break;
3404		r = 0;
3405		break;
3406	}
3407	case KVM_SET_XSAVE: {
3408		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3409		if (IS_ERR(u.xsave))
3410			return PTR_ERR(u.xsave);
3411
3412		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3413		break;
3414	}
3415	case KVM_GET_XCRS: {
3416		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3417		r = -ENOMEM;
3418		if (!u.xcrs)
3419			break;
3420
3421		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3422
3423		r = -EFAULT;
3424		if (copy_to_user(argp, u.xcrs,
3425				 sizeof(struct kvm_xcrs)))
3426			break;
3427		r = 0;
3428		break;
3429	}
3430	case KVM_SET_XCRS: {
3431		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3432		if (IS_ERR(u.xcrs))
3433			return PTR_ERR(u.xcrs);
3434
3435		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3436		break;
3437	}
3438	case KVM_SET_TSC_KHZ: {
3439		u32 user_tsc_khz;
3440
3441		r = -EINVAL;
3442		user_tsc_khz = (u32)arg;
3443
3444		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3445			goto out;
3446
3447		if (user_tsc_khz == 0)
3448			user_tsc_khz = tsc_khz;
3449
3450		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3451			r = 0;
3452
3453		goto out;
3454	}
3455	case KVM_GET_TSC_KHZ: {
3456		r = vcpu->arch.virtual_tsc_khz;
3457		goto out;
3458	}
3459	case KVM_KVMCLOCK_CTRL: {
3460		r = kvm_set_guest_paused(vcpu);
3461		goto out;
3462	}
3463	default:
3464		r = -EINVAL;
3465	}
3466out:
3467	kfree(u.buffer);
3468	return r;
3469}
3470
3471int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3472{
3473	return VM_FAULT_SIGBUS;
3474}
3475
3476static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3477{
3478	int ret;
3479
3480	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3481		return -EINVAL;
3482	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3483	return ret;
3484}
3485
3486static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3487					      u64 ident_addr)
3488{
3489	kvm->arch.ept_identity_map_addr = ident_addr;
3490	return 0;
3491}
3492
3493static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3494					  u32 kvm_nr_mmu_pages)
3495{
3496	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3497		return -EINVAL;
3498
3499	mutex_lock(&kvm->slots_lock);
3500
3501	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3502	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3503
3504	mutex_unlock(&kvm->slots_lock);
3505	return 0;
3506}
3507
3508static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3509{
3510	return kvm->arch.n_max_mmu_pages;
3511}
3512
3513static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3514{
3515	int r;
3516
3517	r = 0;
3518	switch (chip->chip_id) {
3519	case KVM_IRQCHIP_PIC_MASTER:
3520		memcpy(&chip->chip.pic,
3521			&pic_irqchip(kvm)->pics[0],
3522			sizeof(struct kvm_pic_state));
3523		break;
3524	case KVM_IRQCHIP_PIC_SLAVE:
3525		memcpy(&chip->chip.pic,
3526			&pic_irqchip(kvm)->pics[1],
3527			sizeof(struct kvm_pic_state));
3528		break;
3529	case KVM_IRQCHIP_IOAPIC:
3530		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3531		break;
3532	default:
3533		r = -EINVAL;
3534		break;
3535	}
3536	return r;
3537}
3538
3539static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3540{
3541	int r;
3542
3543	r = 0;
3544	switch (chip->chip_id) {
3545	case KVM_IRQCHIP_PIC_MASTER:
3546		spin_lock(&pic_irqchip(kvm)->lock);
3547		memcpy(&pic_irqchip(kvm)->pics[0],
3548			&chip->chip.pic,
3549			sizeof(struct kvm_pic_state));
3550		spin_unlock(&pic_irqchip(kvm)->lock);
3551		break;
3552	case KVM_IRQCHIP_PIC_SLAVE:
3553		spin_lock(&pic_irqchip(kvm)->lock);
3554		memcpy(&pic_irqchip(kvm)->pics[1],
3555			&chip->chip.pic,
3556			sizeof(struct kvm_pic_state));
3557		spin_unlock(&pic_irqchip(kvm)->lock);
3558		break;
3559	case KVM_IRQCHIP_IOAPIC:
3560		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3561		break;
3562	default:
3563		r = -EINVAL;
3564		break;
3565	}
3566	kvm_pic_update_irq(pic_irqchip(kvm));
3567	return r;
3568}
3569
3570static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3571{
3572	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3573	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3574	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3575	return 0;
3576}
3577
3578static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3579{
3580	int i;
3581	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3582	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3583	for (i = 0; i < 3; i++)
3584		kvm_pit_load_count(kvm, i, ps->channels[i].count, 0);
3585	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3586	return 0;
3587}
3588
3589static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3590{
3591	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3592	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3593		sizeof(ps->channels));
3594	ps->flags = kvm->arch.vpit->pit_state.flags;
3595	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3596	memset(&ps->reserved, 0, sizeof(ps->reserved));
3597	return 0;
3598}
3599
3600static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3601{
3602	int start = 0;
3603	int i;
3604	u32 prev_legacy, cur_legacy;
3605	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3606	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3607	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3608	if (!prev_legacy && cur_legacy)
3609		start = 1;
3610	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3611	       sizeof(kvm->arch.vpit->pit_state.channels));
3612	kvm->arch.vpit->pit_state.flags = ps->flags;
3613	for (i = 0; i < 3; i++)
3614		kvm_pit_load_count(kvm, i, kvm->arch.vpit->pit_state.channels[i].count,
3615				   start && i == 0);
3616	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3617	return 0;
3618}
3619
3620static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3621				 struct kvm_reinject_control *control)
3622{
3623	if (!kvm->arch.vpit)
3624		return -ENXIO;
3625	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3626	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3627	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3628	return 0;
3629}
3630
3631/**
3632 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3633 * @kvm: kvm instance
3634 * @log: slot id and address to which we copy the log
3635 *
3636 * Steps 1-4 below provide general overview of dirty page logging. See
3637 * kvm_get_dirty_log_protect() function description for additional details.
3638 *
3639 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3640 * always flush the TLB (step 4) even if previous step failed  and the dirty
3641 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3642 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3643 * writes will be marked dirty for next log read.
3644 *
3645 *   1. Take a snapshot of the bit and clear it if needed.
3646 *   2. Write protect the corresponding page.
3647 *   3. Copy the snapshot to the userspace.
3648 *   4. Flush TLB's if needed.
3649 */
3650int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3651{
3652	bool is_dirty = false;
3653	int r;
3654
3655	mutex_lock(&kvm->slots_lock);
3656
3657	/*
3658	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3659	 */
3660	if (kvm_x86_ops->flush_log_dirty)
3661		kvm_x86_ops->flush_log_dirty(kvm);
3662
3663	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3664
3665	/*
3666	 * All the TLBs can be flushed out of mmu lock, see the comments in
3667	 * kvm_mmu_slot_remove_write_access().
3668	 */
3669	lockdep_assert_held(&kvm->slots_lock);
3670	if (is_dirty)
3671		kvm_flush_remote_tlbs(kvm);
3672
3673	mutex_unlock(&kvm->slots_lock);
3674	return r;
3675}
3676
3677int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3678			bool line_status)
3679{
3680	if (!irqchip_in_kernel(kvm))
3681		return -ENXIO;
3682
3683	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3684					irq_event->irq, irq_event->level,
3685					line_status);
3686	return 0;
3687}
3688
3689static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3690				   struct kvm_enable_cap *cap)
3691{
3692	int r;
3693
3694	if (cap->flags)
3695		return -EINVAL;
3696
3697	switch (cap->cap) {
3698	case KVM_CAP_DISABLE_QUIRKS:
3699		kvm->arch.disabled_quirks = cap->args[0];
3700		r = 0;
3701		break;
3702	case KVM_CAP_SPLIT_IRQCHIP: {
3703		mutex_lock(&kvm->lock);
3704		r = -EINVAL;
3705		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3706			goto split_irqchip_unlock;
3707		r = -EEXIST;
3708		if (irqchip_in_kernel(kvm))
3709			goto split_irqchip_unlock;
3710		if (atomic_read(&kvm->online_vcpus))
3711			goto split_irqchip_unlock;
3712		r = kvm_setup_empty_irq_routing(kvm);
3713		if (r)
3714			goto split_irqchip_unlock;
3715		/* Pairs with irqchip_in_kernel. */
3716		smp_wmb();
3717		kvm->arch.irqchip_split = true;
3718		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3719		r = 0;
3720split_irqchip_unlock:
3721		mutex_unlock(&kvm->lock);
3722		break;
3723	}
3724	default:
3725		r = -EINVAL;
3726		break;
3727	}
3728	return r;
3729}
3730
3731long kvm_arch_vm_ioctl(struct file *filp,
3732		       unsigned int ioctl, unsigned long arg)
3733{
3734	struct kvm *kvm = filp->private_data;
3735	void __user *argp = (void __user *)arg;
3736	int r = -ENOTTY;
3737	/*
3738	 * This union makes it completely explicit to gcc-3.x
3739	 * that these two variables' stack usage should be
3740	 * combined, not added together.
3741	 */
3742	union {
3743		struct kvm_pit_state ps;
3744		struct kvm_pit_state2 ps2;
3745		struct kvm_pit_config pit_config;
3746	} u;
3747
3748	switch (ioctl) {
3749	case KVM_SET_TSS_ADDR:
3750		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3751		break;
3752	case KVM_SET_IDENTITY_MAP_ADDR: {
3753		u64 ident_addr;
3754
3755		r = -EFAULT;
3756		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3757			goto out;
3758		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3759		break;
3760	}
3761	case KVM_SET_NR_MMU_PAGES:
3762		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3763		break;
3764	case KVM_GET_NR_MMU_PAGES:
3765		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3766		break;
3767	case KVM_CREATE_IRQCHIP: {
3768		struct kvm_pic *vpic;
3769
3770		mutex_lock(&kvm->lock);
3771		r = -EEXIST;
3772		if (kvm->arch.vpic)
3773			goto create_irqchip_unlock;
3774		r = -EINVAL;
3775		if (atomic_read(&kvm->online_vcpus))
3776			goto create_irqchip_unlock;
3777		r = -ENOMEM;
3778		vpic = kvm_create_pic(kvm);
3779		if (vpic) {
3780			r = kvm_ioapic_init(kvm);
3781			if (r) {
3782				mutex_lock(&kvm->slots_lock);
3783				kvm_destroy_pic(vpic);
3784				mutex_unlock(&kvm->slots_lock);
3785				goto create_irqchip_unlock;
3786			}
3787		} else
3788			goto create_irqchip_unlock;
3789		r = kvm_setup_default_irq_routing(kvm);
3790		if (r) {
3791			mutex_lock(&kvm->slots_lock);
3792			mutex_lock(&kvm->irq_lock);
3793			kvm_ioapic_destroy(kvm);
3794			kvm_destroy_pic(vpic);
3795			mutex_unlock(&kvm->irq_lock);
3796			mutex_unlock(&kvm->slots_lock);
3797			goto create_irqchip_unlock;
3798		}
3799		/* Write kvm->irq_routing before kvm->arch.vpic.  */
3800		smp_wmb();
3801		kvm->arch.vpic = vpic;
3802	create_irqchip_unlock:
3803		mutex_unlock(&kvm->lock);
3804		break;
3805	}
3806	case KVM_CREATE_PIT:
3807		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3808		goto create_pit;
3809	case KVM_CREATE_PIT2:
3810		r = -EFAULT;
3811		if (copy_from_user(&u.pit_config, argp,
3812				   sizeof(struct kvm_pit_config)))
3813			goto out;
3814	create_pit:
3815		mutex_lock(&kvm->slots_lock);
3816		r = -EEXIST;
3817		if (kvm->arch.vpit)
3818			goto create_pit_unlock;
3819		r = -ENOMEM;
3820		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3821		if (kvm->arch.vpit)
3822			r = 0;
3823	create_pit_unlock:
3824		mutex_unlock(&kvm->slots_lock);
3825		break;
3826	case KVM_GET_IRQCHIP: {
3827		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3828		struct kvm_irqchip *chip;
3829
3830		chip = memdup_user(argp, sizeof(*chip));
3831		if (IS_ERR(chip)) {
3832			r = PTR_ERR(chip);
3833			goto out;
3834		}
3835
3836		r = -ENXIO;
3837		if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3838			goto get_irqchip_out;
3839		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3840		if (r)
3841			goto get_irqchip_out;
3842		r = -EFAULT;
3843		if (copy_to_user(argp, chip, sizeof *chip))
3844			goto get_irqchip_out;
3845		r = 0;
3846	get_irqchip_out:
3847		kfree(chip);
3848		break;
3849	}
3850	case KVM_SET_IRQCHIP: {
3851		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3852		struct kvm_irqchip *chip;
3853
3854		chip = memdup_user(argp, sizeof(*chip));
3855		if (IS_ERR(chip)) {
3856			r = PTR_ERR(chip);
3857			goto out;
3858		}
3859
3860		r = -ENXIO;
3861		if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3862			goto set_irqchip_out;
3863		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3864		if (r)
3865			goto set_irqchip_out;
3866		r = 0;
3867	set_irqchip_out:
3868		kfree(chip);
3869		break;
3870	}
3871	case KVM_GET_PIT: {
3872		r = -EFAULT;
3873		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3874			goto out;
3875		r = -ENXIO;
3876		if (!kvm->arch.vpit)
3877			goto out;
3878		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3879		if (r)
3880			goto out;
3881		r = -EFAULT;
3882		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3883			goto out;
3884		r = 0;
3885		break;
3886	}
3887	case KVM_SET_PIT: {
3888		r = -EFAULT;
3889		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3890			goto out;
3891		r = -ENXIO;
3892		if (!kvm->arch.vpit)
3893			goto out;
3894		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3895		break;
3896	}
3897	case KVM_GET_PIT2: {
3898		r = -ENXIO;
3899		if (!kvm->arch.vpit)
3900			goto out;
3901		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3902		if (r)
3903			goto out;
3904		r = -EFAULT;
3905		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3906			goto out;
3907		r = 0;
3908		break;
3909	}
3910	case KVM_SET_PIT2: {
3911		r = -EFAULT;
3912		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3913			goto out;
3914		r = -ENXIO;
3915		if (!kvm->arch.vpit)
3916			goto out;
3917		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3918		break;
3919	}
3920	case KVM_REINJECT_CONTROL: {
3921		struct kvm_reinject_control control;
3922		r =  -EFAULT;
3923		if (copy_from_user(&control, argp, sizeof(control)))
3924			goto out;
3925		r = kvm_vm_ioctl_reinject(kvm, &control);
3926		break;
3927	}
3928	case KVM_SET_BOOT_CPU_ID:
3929		r = 0;
3930		mutex_lock(&kvm->lock);
3931		if (atomic_read(&kvm->online_vcpus) != 0)
3932			r = -EBUSY;
3933		else
3934			kvm->arch.bsp_vcpu_id = arg;
3935		mutex_unlock(&kvm->lock);
3936		break;
3937	case KVM_XEN_HVM_CONFIG: {
3938		r = -EFAULT;
3939		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3940				   sizeof(struct kvm_xen_hvm_config)))
3941			goto out;
3942		r = -EINVAL;
3943		if (kvm->arch.xen_hvm_config.flags)
3944			goto out;
3945		r = 0;
3946		break;
3947	}
3948	case KVM_SET_CLOCK: {
3949		struct kvm_clock_data user_ns;
3950		u64 now_ns;
3951		s64 delta;
3952
3953		r = -EFAULT;
3954		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3955			goto out;
3956
3957		r = -EINVAL;
3958		if (user_ns.flags)
3959			goto out;
3960
3961		r = 0;
3962		local_irq_disable();
3963		now_ns = get_kernel_ns();
3964		delta = user_ns.clock - now_ns;
3965		local_irq_enable();
3966		kvm->arch.kvmclock_offset = delta;
3967		kvm_gen_update_masterclock(kvm);
3968		break;
3969	}
3970	case KVM_GET_CLOCK: {
3971		struct kvm_clock_data user_ns;
3972		u64 now_ns;
3973
3974		local_irq_disable();
3975		now_ns = get_kernel_ns();
3976		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3977		local_irq_enable();
3978		user_ns.flags = 0;
3979		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3980
3981		r = -EFAULT;
3982		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3983			goto out;
3984		r = 0;
3985		break;
3986	}
3987	case KVM_ENABLE_CAP: {
3988		struct kvm_enable_cap cap;
3989
3990		r = -EFAULT;
3991		if (copy_from_user(&cap, argp, sizeof(cap)))
3992			goto out;
3993		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3994		break;
3995	}
3996	default:
3997		r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
3998	}
3999out:
4000	return r;
4001}
4002
4003static void kvm_init_msr_list(void)
4004{
4005	u32 dummy[2];
4006	unsigned i, j;
4007
4008	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4009		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4010			continue;
4011
4012		/*
4013		 * Even MSRs that are valid in the host may not be exposed
4014		 * to the guests in some cases.
4015		 */
4016		switch (msrs_to_save[i]) {
4017		case MSR_IA32_BNDCFGS:
4018			if (!kvm_x86_ops->mpx_supported())
4019				continue;
4020			break;
4021		case MSR_TSC_AUX:
4022			if (!kvm_x86_ops->rdtscp_supported())
4023				continue;
4024			break;
4025		default:
4026			break;
4027		}
4028
4029		if (j < i)
4030			msrs_to_save[j] = msrs_to_save[i];
4031		j++;
4032	}
4033	num_msrs_to_save = j;
4034
4035	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4036		switch (emulated_msrs[i]) {
4037		case MSR_IA32_SMBASE:
4038			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4039				continue;
4040			break;
4041		default:
4042			break;
4043		}
4044
4045		if (j < i)
4046			emulated_msrs[j] = emulated_msrs[i];
4047		j++;
4048	}
4049	num_emulated_msrs = j;
4050}
4051
4052static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4053			   const void *v)
4054{
4055	int handled = 0;
4056	int n;
4057
4058	do {
4059		n = min(len, 8);
4060		if (!(vcpu->arch.apic &&
4061		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4062		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4063			break;
4064		handled += n;
4065		addr += n;
4066		len -= n;
4067		v += n;
4068	} while (len);
4069
4070	return handled;
4071}
4072
4073static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4074{
4075	int handled = 0;
4076	int n;
4077
4078	do {
4079		n = min(len, 8);
4080		if (!(vcpu->arch.apic &&
4081		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4082					 addr, n, v))
4083		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4084			break;
4085		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4086		handled += n;
4087		addr += n;
4088		len -= n;
4089		v += n;
4090	} while (len);
4091
4092	return handled;
4093}
4094
4095static void kvm_set_segment(struct kvm_vcpu *vcpu,
4096			struct kvm_segment *var, int seg)
4097{
4098	kvm_x86_ops->set_segment(vcpu, var, seg);
4099}
4100
4101void kvm_get_segment(struct kvm_vcpu *vcpu,
4102		     struct kvm_segment *var, int seg)
4103{
4104	kvm_x86_ops->get_segment(vcpu, var, seg);
4105}
4106
4107gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4108			   struct x86_exception *exception)
4109{
4110	gpa_t t_gpa;
4111
4112	BUG_ON(!mmu_is_nested(vcpu));
4113
4114	/* NPT walks are always user-walks */
4115	access |= PFERR_USER_MASK;
4116	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4117
4118	return t_gpa;
4119}
4120
4121gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4122			      struct x86_exception *exception)
4123{
4124	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4125	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4126}
4127
4128 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4129				struct x86_exception *exception)
4130{
4131	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4132	access |= PFERR_FETCH_MASK;
4133	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4134}
4135
4136gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4137			       struct x86_exception *exception)
4138{
4139	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4140	access |= PFERR_WRITE_MASK;
4141	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4142}
4143
4144/* uses this to access any guest's mapped memory without checking CPL */
4145gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4146				struct x86_exception *exception)
4147{
4148	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4149}
4150
4151static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4152				      struct kvm_vcpu *vcpu, u32 access,
4153				      struct x86_exception *exception)
4154{
4155	void *data = val;
4156	int r = X86EMUL_CONTINUE;
4157
4158	while (bytes) {
4159		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4160							    exception);
4161		unsigned offset = addr & (PAGE_SIZE-1);
4162		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4163		int ret;
4164
4165		if (gpa == UNMAPPED_GVA)
4166			return X86EMUL_PROPAGATE_FAULT;
4167		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4168					       offset, toread);
4169		if (ret < 0) {
4170			r = X86EMUL_IO_NEEDED;
4171			goto out;
4172		}
4173
4174		bytes -= toread;
4175		data += toread;
4176		addr += toread;
4177	}
4178out:
4179	return r;
4180}
4181
4182/* used for instruction fetching */
4183static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4184				gva_t addr, void *val, unsigned int bytes,
4185				struct x86_exception *exception)
4186{
4187	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4188	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4189	unsigned offset;
4190	int ret;
4191
4192	/* Inline kvm_read_guest_virt_helper for speed.  */
4193	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4194						    exception);
4195	if (unlikely(gpa == UNMAPPED_GVA))
4196		return X86EMUL_PROPAGATE_FAULT;
4197
4198	offset = addr & (PAGE_SIZE-1);
4199	if (WARN_ON(offset + bytes > PAGE_SIZE))
4200		bytes = (unsigned)PAGE_SIZE - offset;
4201	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4202				       offset, bytes);
4203	if (unlikely(ret < 0))
4204		return X86EMUL_IO_NEEDED;
4205
4206	return X86EMUL_CONTINUE;
4207}
4208
4209int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4210			       gva_t addr, void *val, unsigned int bytes,
4211			       struct x86_exception *exception)
4212{
4213	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4214	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4215
4216	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4217					  exception);
4218}
4219EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4220
4221static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4222				      gva_t addr, void *val, unsigned int bytes,
4223				      struct x86_exception *exception)
4224{
4225	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4226	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4227}
4228
4229static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4230		unsigned long addr, void *val, unsigned int bytes)
4231{
4232	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4233	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4234
4235	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4236}
4237
4238int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4239				       gva_t addr, void *val,
4240				       unsigned int bytes,
4241				       struct x86_exception *exception)
4242{
4243	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4244	void *data = val;
4245	int r = X86EMUL_CONTINUE;
4246
4247	while (bytes) {
4248		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4249							     PFERR_WRITE_MASK,
4250							     exception);
4251		unsigned offset = addr & (PAGE_SIZE-1);
4252		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4253		int ret;
4254
4255		if (gpa == UNMAPPED_GVA)
4256			return X86EMUL_PROPAGATE_FAULT;
4257		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4258		if (ret < 0) {
4259			r = X86EMUL_IO_NEEDED;
4260			goto out;
4261		}
4262
4263		bytes -= towrite;
4264		data += towrite;
4265		addr += towrite;
4266	}
4267out:
4268	return r;
4269}
4270EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4271
4272static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4273				gpa_t *gpa, struct x86_exception *exception,
4274				bool write)
4275{
4276	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4277		| (write ? PFERR_WRITE_MASK : 0);
4278
4279	if (vcpu_match_mmio_gva(vcpu, gva)
4280	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4281				 vcpu->arch.access, access)) {
4282		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4283					(gva & (PAGE_SIZE - 1));
4284		trace_vcpu_match_mmio(gva, *gpa, write, false);
4285		return 1;
4286	}
4287
4288	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4289
4290	if (*gpa == UNMAPPED_GVA)
4291		return -1;
4292
4293	/* For APIC access vmexit */
4294	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4295		return 1;
4296
4297	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4298		trace_vcpu_match_mmio(gva, *gpa, write, true);
4299		return 1;
4300	}
4301
4302	return 0;
4303}
4304
4305int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4306			const void *val, int bytes)
4307{
4308	int ret;
4309
4310	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4311	if (ret < 0)
4312		return 0;
4313	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4314	return 1;
4315}
4316
4317struct read_write_emulator_ops {
4318	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4319				  int bytes);
4320	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4321				  void *val, int bytes);
4322	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4323			       int bytes, void *val);
4324	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4325				    void *val, int bytes);
4326	bool write;
4327};
4328
4329static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4330{
4331	if (vcpu->mmio_read_completed) {
4332		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4333			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4334		vcpu->mmio_read_completed = 0;
4335		return 1;
4336	}
4337
4338	return 0;
4339}
4340
4341static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4342			void *val, int bytes)
4343{
4344	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4345}
4346
4347static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4348			 void *val, int bytes)
4349{
4350	return emulator_write_phys(vcpu, gpa, val, bytes);
4351}
4352
4353static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4354{
4355	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4356	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4357}
4358
4359static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4360			  void *val, int bytes)
4361{
4362	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4363	return X86EMUL_IO_NEEDED;
4364}
4365
4366static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4367			   void *val, int bytes)
4368{
4369	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4370
4371	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4372	return X86EMUL_CONTINUE;
4373}
4374
4375static const struct read_write_emulator_ops read_emultor = {
4376	.read_write_prepare = read_prepare,
4377	.read_write_emulate = read_emulate,
4378	.read_write_mmio = vcpu_mmio_read,
4379	.read_write_exit_mmio = read_exit_mmio,
4380};
4381
4382static const struct read_write_emulator_ops write_emultor = {
4383	.read_write_emulate = write_emulate,
4384	.read_write_mmio = write_mmio,
4385	.read_write_exit_mmio = write_exit_mmio,
4386	.write = true,
4387};
4388
4389static int emulator_read_write_onepage(unsigned long addr, void *val,
4390				       unsigned int bytes,
4391				       struct x86_exception *exception,
4392				       struct kvm_vcpu *vcpu,
4393				       const struct read_write_emulator_ops *ops)
4394{
4395	gpa_t gpa;
4396	int handled, ret;
4397	bool write = ops->write;
4398	struct kvm_mmio_fragment *frag;
4399
4400	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4401
4402	if (ret < 0)
4403		return X86EMUL_PROPAGATE_FAULT;
4404
4405	/* For APIC access vmexit */
4406	if (ret)
4407		goto mmio;
4408
4409	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4410		return X86EMUL_CONTINUE;
4411
4412mmio:
4413	/*
4414	 * Is this MMIO handled locally?
4415	 */
4416	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4417	if (handled == bytes)
4418		return X86EMUL_CONTINUE;
4419
4420	gpa += handled;
4421	bytes -= handled;
4422	val += handled;
4423
4424	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4425	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4426	frag->gpa = gpa;
4427	frag->data = val;
4428	frag->len = bytes;
4429	return X86EMUL_CONTINUE;
4430}
4431
4432static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4433			unsigned long addr,
4434			void *val, unsigned int bytes,
4435			struct x86_exception *exception,
4436			const struct read_write_emulator_ops *ops)
4437{
4438	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4439	gpa_t gpa;
4440	int rc;
4441
4442	if (ops->read_write_prepare &&
4443		  ops->read_write_prepare(vcpu, val, bytes))
4444		return X86EMUL_CONTINUE;
4445
4446	vcpu->mmio_nr_fragments = 0;
4447
4448	/* Crossing a page boundary? */
4449	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4450		int now;
4451
4452		now = -addr & ~PAGE_MASK;
4453		rc = emulator_read_write_onepage(addr, val, now, exception,
4454						 vcpu, ops);
4455
4456		if (rc != X86EMUL_CONTINUE)
4457			return rc;
4458		addr += now;
4459		if (ctxt->mode != X86EMUL_MODE_PROT64)
4460			addr = (u32)addr;
4461		val += now;
4462		bytes -= now;
4463	}
4464
4465	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4466					 vcpu, ops);
4467	if (rc != X86EMUL_CONTINUE)
4468		return rc;
4469
4470	if (!vcpu->mmio_nr_fragments)
4471		return rc;
4472
4473	gpa = vcpu->mmio_fragments[0].gpa;
4474
4475	vcpu->mmio_needed = 1;
4476	vcpu->mmio_cur_fragment = 0;
4477
4478	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4479	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4480	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4481	vcpu->run->mmio.phys_addr = gpa;
4482
4483	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4484}
4485
4486static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4487				  unsigned long addr,
4488				  void *val,
4489				  unsigned int bytes,
4490				  struct x86_exception *exception)
4491{
4492	return emulator_read_write(ctxt, addr, val, bytes,
4493				   exception, &read_emultor);
4494}
4495
4496static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4497			    unsigned long addr,
4498			    const void *val,
4499			    unsigned int bytes,
4500			    struct x86_exception *exception)
4501{
4502	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4503				   exception, &write_emultor);
4504}
4505
4506#define CMPXCHG_TYPE(t, ptr, old, new) \
4507	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4508
4509#ifdef CONFIG_X86_64
4510#  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4511#else
4512#  define CMPXCHG64(ptr, old, new) \
4513	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4514#endif
4515
4516static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4517				     unsigned long addr,
4518				     const void *old,
4519				     const void *new,
4520				     unsigned int bytes,
4521				     struct x86_exception *exception)
4522{
4523	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4524	gpa_t gpa;
4525	struct page *page;
4526	char *kaddr;
4527	bool exchanged;
4528
4529	/* guests cmpxchg8b have to be emulated atomically */
4530	if (bytes > 8 || (bytes & (bytes - 1)))
4531		goto emul_write;
4532
4533	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4534
4535	if (gpa == UNMAPPED_GVA ||
4536	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4537		goto emul_write;
4538
4539	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4540		goto emul_write;
4541
4542	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4543	if (is_error_page(page))
4544		goto emul_write;
4545
4546	kaddr = kmap_atomic(page);
4547	kaddr += offset_in_page(gpa);
4548	switch (bytes) {
4549	case 1:
4550		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4551		break;
4552	case 2:
4553		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4554		break;
4555	case 4:
4556		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4557		break;
4558	case 8:
4559		exchanged = CMPXCHG64(kaddr, old, new);
4560		break;
4561	default:
4562		BUG();
4563	}
4564	kunmap_atomic(kaddr);
4565	kvm_release_page_dirty(page);
4566
4567	if (!exchanged)
4568		return X86EMUL_CMPXCHG_FAILED;
4569
4570	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4571	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4572
4573	return X86EMUL_CONTINUE;
4574
4575emul_write:
4576	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4577
4578	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4579}
4580
4581static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4582{
4583	/* TODO: String I/O for in kernel device */
4584	int r;
4585
4586	if (vcpu->arch.pio.in)
4587		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4588				    vcpu->arch.pio.size, pd);
4589	else
4590		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4591				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4592				     pd);
4593	return r;
4594}
4595
4596static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4597			       unsigned short port, void *val,
4598			       unsigned int count, bool in)
4599{
4600	vcpu->arch.pio.port = port;
4601	vcpu->arch.pio.in = in;
4602	vcpu->arch.pio.count  = count;
4603	vcpu->arch.pio.size = size;
4604
4605	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4606		vcpu->arch.pio.count = 0;
4607		return 1;
4608	}
4609
4610	vcpu->run->exit_reason = KVM_EXIT_IO;
4611	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4612	vcpu->run->io.size = size;
4613	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4614	vcpu->run->io.count = count;
4615	vcpu->run->io.port = port;
4616
4617	return 0;
4618}
4619
4620static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4621				    int size, unsigned short port, void *val,
4622				    unsigned int count)
4623{
4624	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4625	int ret;
4626
4627	if (vcpu->arch.pio.count)
4628		goto data_avail;
4629
4630	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4631	if (ret) {
4632data_avail:
4633		memcpy(val, vcpu->arch.pio_data, size * count);
4634		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4635		vcpu->arch.pio.count = 0;
4636		return 1;
4637	}
4638
4639	return 0;
4640}
4641
4642static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4643				     int size, unsigned short port,
4644				     const void *val, unsigned int count)
4645{
4646	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4647
4648	memcpy(vcpu->arch.pio_data, val, size * count);
4649	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4650	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4651}
4652
4653static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4654{
4655	return kvm_x86_ops->get_segment_base(vcpu, seg);
4656}
4657
4658static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4659{
4660	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4661}
4662
4663int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4664{
4665	if (!need_emulate_wbinvd(vcpu))
4666		return X86EMUL_CONTINUE;
4667
4668	if (kvm_x86_ops->has_wbinvd_exit()) {
4669		int cpu = get_cpu();
4670
4671		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4672		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4673				wbinvd_ipi, NULL, 1);
4674		put_cpu();
4675		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4676	} else
4677		wbinvd();
4678	return X86EMUL_CONTINUE;
4679}
4680
4681int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4682{
4683	kvm_x86_ops->skip_emulated_instruction(vcpu);
4684	return kvm_emulate_wbinvd_noskip(vcpu);
4685}
4686EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4687
4688
4689
4690static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4691{
4692	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4693}
4694
4695static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4696			   unsigned long *dest)
4697{
4698	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4699}
4700
4701static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4702			   unsigned long value)
4703{
4704
4705	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4706}
4707
4708static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4709{
4710	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4711}
4712
4713static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4714{
4715	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4716	unsigned long value;
4717
4718	switch (cr) {
4719	case 0:
4720		value = kvm_read_cr0(vcpu);
4721		break;
4722	case 2:
4723		value = vcpu->arch.cr2;
4724		break;
4725	case 3:
4726		value = kvm_read_cr3(vcpu);
4727		break;
4728	case 4:
4729		value = kvm_read_cr4(vcpu);
4730		break;
4731	case 8:
4732		value = kvm_get_cr8(vcpu);
4733		break;
4734	default:
4735		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4736		return 0;
4737	}
4738
4739	return value;
4740}
4741
4742static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4743{
4744	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4745	int res = 0;
4746
4747	switch (cr) {
4748	case 0:
4749		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4750		break;
4751	case 2:
4752		vcpu->arch.cr2 = val;
4753		break;
4754	case 3:
4755		res = kvm_set_cr3(vcpu, val);
4756		break;
4757	case 4:
4758		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4759		break;
4760	case 8:
4761		res = kvm_set_cr8(vcpu, val);
4762		break;
4763	default:
4764		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4765		res = -1;
4766	}
4767
4768	return res;
4769}
4770
4771static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4772{
4773	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4774}
4775
4776static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4777{
4778	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4779}
4780
4781static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4782{
4783	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4784}
4785
4786static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4787{
4788	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4789}
4790
4791static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4792{
4793	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4794}
4795
4796static unsigned long emulator_get_cached_segment_base(
4797	struct x86_emulate_ctxt *ctxt, int seg)
4798{
4799	return get_segment_base(emul_to_vcpu(ctxt), seg);
4800}
4801
4802static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4803				 struct desc_struct *desc, u32 *base3,
4804				 int seg)
4805{
4806	struct kvm_segment var;
4807
4808	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4809	*selector = var.selector;
4810
4811	if (var.unusable) {
4812		memset(desc, 0, sizeof(*desc));
4813		return false;
4814	}
4815
4816	if (var.g)
4817		var.limit >>= 12;
4818	set_desc_limit(desc, var.limit);
4819	set_desc_base(desc, (unsigned long)var.base);
4820#ifdef CONFIG_X86_64
4821	if (base3)
4822		*base3 = var.base >> 32;
4823#endif
4824	desc->type = var.type;
4825	desc->s = var.s;
4826	desc->dpl = var.dpl;
4827	desc->p = var.present;
4828	desc->avl = var.avl;
4829	desc->l = var.l;
4830	desc->d = var.db;
4831	desc->g = var.g;
4832
4833	return true;
4834}
4835
4836static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4837				 struct desc_struct *desc, u32 base3,
4838				 int seg)
4839{
4840	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4841	struct kvm_segment var;
4842
4843	var.selector = selector;
4844	var.base = get_desc_base(desc);
4845#ifdef CONFIG_X86_64
4846	var.base |= ((u64)base3) << 32;
4847#endif
4848	var.limit = get_desc_limit(desc);
4849	if (desc->g)
4850		var.limit = (var.limit << 12) | 0xfff;
4851	var.type = desc->type;
4852	var.dpl = desc->dpl;
4853	var.db = desc->d;
4854	var.s = desc->s;
4855	var.l = desc->l;
4856	var.g = desc->g;
4857	var.avl = desc->avl;
4858	var.present = desc->p;
4859	var.unusable = !var.present;
4860	var.padding = 0;
4861
4862	kvm_set_segment(vcpu, &var, seg);
4863	return;
4864}
4865
4866static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4867			    u32 msr_index, u64 *pdata)
4868{
4869	struct msr_data msr;
4870	int r;
4871
4872	msr.index = msr_index;
4873	msr.host_initiated = false;
4874	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4875	if (r)
4876		return r;
4877
4878	*pdata = msr.data;
4879	return 0;
4880}
4881
4882static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4883			    u32 msr_index, u64 data)
4884{
4885	struct msr_data msr;
4886
4887	msr.data = data;
4888	msr.index = msr_index;
4889	msr.host_initiated = false;
4890	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4891}
4892
4893static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4894{
4895	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4896
4897	return vcpu->arch.smbase;
4898}
4899
4900static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4901{
4902	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4903
4904	vcpu->arch.smbase = smbase;
4905}
4906
4907static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4908			      u32 pmc)
4909{
4910	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4911}
4912
4913static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4914			     u32 pmc, u64 *pdata)
4915{
4916	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4917}
4918
4919static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4920{
4921	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4922}
4923
4924static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4925{
4926	preempt_disable();
4927	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4928	/*
4929	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4930	 * so it may be clear at this point.
4931	 */
4932	clts();
4933}
4934
4935static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4936{
4937	preempt_enable();
4938}
4939
4940static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4941			      struct x86_instruction_info *info,
4942			      enum x86_intercept_stage stage)
4943{
4944	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4945}
4946
4947static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4948			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4949{
4950	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4951}
4952
4953static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4954{
4955	return kvm_register_read(emul_to_vcpu(ctxt), reg);
4956}
4957
4958static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4959{
4960	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4961}
4962
4963static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4964{
4965	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4966}
4967
4968static const struct x86_emulate_ops emulate_ops = {
4969	.read_gpr            = emulator_read_gpr,
4970	.write_gpr           = emulator_write_gpr,
4971	.read_std            = kvm_read_guest_virt_system,
4972	.write_std           = kvm_write_guest_virt_system,
4973	.read_phys           = kvm_read_guest_phys_system,
4974	.fetch               = kvm_fetch_guest_virt,
4975	.read_emulated       = emulator_read_emulated,
4976	.write_emulated      = emulator_write_emulated,
4977	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
4978	.invlpg              = emulator_invlpg,
4979	.pio_in_emulated     = emulator_pio_in_emulated,
4980	.pio_out_emulated    = emulator_pio_out_emulated,
4981	.get_segment         = emulator_get_segment,
4982	.set_segment         = emulator_set_segment,
4983	.get_cached_segment_base = emulator_get_cached_segment_base,
4984	.get_gdt             = emulator_get_gdt,
4985	.get_idt	     = emulator_get_idt,
4986	.set_gdt             = emulator_set_gdt,
4987	.set_idt	     = emulator_set_idt,
4988	.get_cr              = emulator_get_cr,
4989	.set_cr              = emulator_set_cr,
4990	.cpl                 = emulator_get_cpl,
4991	.get_dr              = emulator_get_dr,
4992	.set_dr              = emulator_set_dr,
4993	.get_smbase          = emulator_get_smbase,
4994	.set_smbase          = emulator_set_smbase,
4995	.set_msr             = emulator_set_msr,
4996	.get_msr             = emulator_get_msr,
4997	.check_pmc	     = emulator_check_pmc,
4998	.read_pmc            = emulator_read_pmc,
4999	.halt                = emulator_halt,
5000	.wbinvd              = emulator_wbinvd,
5001	.fix_hypercall       = emulator_fix_hypercall,
5002	.get_fpu             = emulator_get_fpu,
5003	.put_fpu             = emulator_put_fpu,
5004	.intercept           = emulator_intercept,
5005	.get_cpuid           = emulator_get_cpuid,
5006	.set_nmi_mask        = emulator_set_nmi_mask,
5007};
5008
5009static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5010{
5011	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5012	/*
5013	 * an sti; sti; sequence only disable interrupts for the first
5014	 * instruction. So, if the last instruction, be it emulated or
5015	 * not, left the system with the INT_STI flag enabled, it
5016	 * means that the last instruction is an sti. We should not
5017	 * leave the flag on in this case. The same goes for mov ss
5018	 */
5019	if (int_shadow & mask)
5020		mask = 0;
5021	if (unlikely(int_shadow || mask)) {
5022		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5023		if (!mask)
5024			kvm_make_request(KVM_REQ_EVENT, vcpu);
5025	}
5026}
5027
5028static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5029{
5030	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5031	if (ctxt->exception.vector == PF_VECTOR)
5032		return kvm_propagate_fault(vcpu, &ctxt->exception);
5033
5034	if (ctxt->exception.error_code_valid)
5035		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5036				      ctxt->exception.error_code);
5037	else
5038		kvm_queue_exception(vcpu, ctxt->exception.vector);
5039	return false;
5040}
5041
5042static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5043{
5044	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5045	int cs_db, cs_l;
5046
5047	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5048
5049	ctxt->eflags = kvm_get_rflags(vcpu);
5050	ctxt->eip = kvm_rip_read(vcpu);
5051	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5052		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5053		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5054		     cs_db				? X86EMUL_MODE_PROT32 :
5055							  X86EMUL_MODE_PROT16;
5056	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5057	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5058	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5059	ctxt->emul_flags = vcpu->arch.hflags;
5060
5061	init_decode_cache(ctxt);
5062	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5063}
5064
5065int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5066{
5067	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5068	int ret;
5069
5070	init_emulate_ctxt(vcpu);
5071
5072	ctxt->op_bytes = 2;
5073	ctxt->ad_bytes = 2;
5074	ctxt->_eip = ctxt->eip + inc_eip;
5075	ret = emulate_int_real(ctxt, irq);
5076
5077	if (ret != X86EMUL_CONTINUE)
5078		return EMULATE_FAIL;
5079
5080	ctxt->eip = ctxt->_eip;
5081	kvm_rip_write(vcpu, ctxt->eip);
5082	kvm_set_rflags(vcpu, ctxt->eflags);
5083
5084	if (irq == NMI_VECTOR)
5085		vcpu->arch.nmi_pending = 0;
5086	else
5087		vcpu->arch.interrupt.pending = false;
5088
5089	return EMULATE_DONE;
5090}
5091EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5092
5093static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5094{
5095	int r = EMULATE_DONE;
5096
5097	++vcpu->stat.insn_emulation_fail;
5098	trace_kvm_emulate_insn_failed(vcpu);
5099	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5100		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5101		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5102		vcpu->run->internal.ndata = 0;
5103		r = EMULATE_FAIL;
5104	}
5105	kvm_queue_exception(vcpu, UD_VECTOR);
5106
5107	return r;
5108}
5109
5110static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5111				  bool write_fault_to_shadow_pgtable,
5112				  int emulation_type)
5113{
5114	gpa_t gpa = cr2;
5115	pfn_t pfn;
5116
5117	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5118		return false;
5119
5120	if (!vcpu->arch.mmu.direct_map) {
5121		/*
5122		 * Write permission should be allowed since only
5123		 * write access need to be emulated.
5124		 */
5125		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5126
5127		/*
5128		 * If the mapping is invalid in guest, let cpu retry
5129		 * it to generate fault.
5130		 */
5131		if (gpa == UNMAPPED_GVA)
5132			return true;
5133	}
5134
5135	/*
5136	 * Do not retry the unhandleable instruction if it faults on the
5137	 * readonly host memory, otherwise it will goto a infinite loop:
5138	 * retry instruction -> write #PF -> emulation fail -> retry
5139	 * instruction -> ...
5140	 */
5141	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5142
5143	/*
5144	 * If the instruction failed on the error pfn, it can not be fixed,
5145	 * report the error to userspace.
5146	 */
5147	if (is_error_noslot_pfn(pfn))
5148		return false;
5149
5150	kvm_release_pfn_clean(pfn);
5151
5152	/* The instructions are well-emulated on direct mmu. */
5153	if (vcpu->arch.mmu.direct_map) {
5154		unsigned int indirect_shadow_pages;
5155
5156		spin_lock(&vcpu->kvm->mmu_lock);
5157		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5158		spin_unlock(&vcpu->kvm->mmu_lock);
5159
5160		if (indirect_shadow_pages)
5161			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5162
5163		return true;
5164	}
5165
5166	/*
5167	 * if emulation was due to access to shadowed page table
5168	 * and it failed try to unshadow page and re-enter the
5169	 * guest to let CPU execute the instruction.
5170	 */
5171	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5172
5173	/*
5174	 * If the access faults on its page table, it can not
5175	 * be fixed by unprotecting shadow page and it should
5176	 * be reported to userspace.
5177	 */
5178	return !write_fault_to_shadow_pgtable;
5179}
5180
5181static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5182			      unsigned long cr2,  int emulation_type)
5183{
5184	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5185	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5186
5187	last_retry_eip = vcpu->arch.last_retry_eip;
5188	last_retry_addr = vcpu->arch.last_retry_addr;
5189
5190	/*
5191	 * If the emulation is caused by #PF and it is non-page_table
5192	 * writing instruction, it means the VM-EXIT is caused by shadow
5193	 * page protected, we can zap the shadow page and retry this
5194	 * instruction directly.
5195	 *
5196	 * Note: if the guest uses a non-page-table modifying instruction
5197	 * on the PDE that points to the instruction, then we will unmap
5198	 * the instruction and go to an infinite loop. So, we cache the
5199	 * last retried eip and the last fault address, if we meet the eip
5200	 * and the address again, we can break out of the potential infinite
5201	 * loop.
5202	 */
5203	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5204
5205	if (!(emulation_type & EMULTYPE_RETRY))
5206		return false;
5207
5208	if (x86_page_table_writing_insn(ctxt))
5209		return false;
5210
5211	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5212		return false;
5213
5214	vcpu->arch.last_retry_eip = ctxt->eip;
5215	vcpu->arch.last_retry_addr = cr2;
5216
5217	if (!vcpu->arch.mmu.direct_map)
5218		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5219
5220	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5221
5222	return true;
5223}
5224
5225static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5226static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5227
5228static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5229{
5230	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5231		/* This is a good place to trace that we are exiting SMM.  */
5232		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5233
5234		if (unlikely(vcpu->arch.smi_pending)) {
5235			kvm_make_request(KVM_REQ_SMI, vcpu);
5236			vcpu->arch.smi_pending = 0;
5237		} else {
5238			/* Process a latched INIT, if any.  */
5239			kvm_make_request(KVM_REQ_EVENT, vcpu);
5240		}
5241	}
5242
5243	kvm_mmu_reset_context(vcpu);
5244}
5245
5246static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5247{
5248	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5249
5250	vcpu->arch.hflags = emul_flags;
5251
5252	if (changed & HF_SMM_MASK)
5253		kvm_smm_changed(vcpu);
5254}
5255
5256static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5257				unsigned long *db)
5258{
5259	u32 dr6 = 0;
5260	int i;
5261	u32 enable, rwlen;
5262
5263	enable = dr7;
5264	rwlen = dr7 >> 16;
5265	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5266		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5267			dr6 |= (1 << i);
5268	return dr6;
5269}
5270
5271static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5272{
5273	struct kvm_run *kvm_run = vcpu->run;
5274
5275	/*
5276	 * rflags is the old, "raw" value of the flags.  The new value has
5277	 * not been saved yet.
5278	 *
5279	 * This is correct even for TF set by the guest, because "the
5280	 * processor will not generate this exception after the instruction
5281	 * that sets the TF flag".
5282	 */
5283	if (unlikely(rflags & X86_EFLAGS_TF)) {
5284		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5285			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5286						  DR6_RTM;
5287			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5288			kvm_run->debug.arch.exception = DB_VECTOR;
5289			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5290			*r = EMULATE_USER_EXIT;
5291		} else {
5292			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5293			/*
5294			 * "Certain debug exceptions may clear bit 0-3.  The
5295			 * remaining contents of the DR6 register are never
5296			 * cleared by the processor".
5297			 */
5298			vcpu->arch.dr6 &= ~15;
5299			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5300			kvm_queue_exception(vcpu, DB_VECTOR);
5301		}
5302	}
5303}
5304
5305static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5306{
5307	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5308	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5309		struct kvm_run *kvm_run = vcpu->run;
5310		unsigned long eip = kvm_get_linear_rip(vcpu);
5311		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5312					   vcpu->arch.guest_debug_dr7,
5313					   vcpu->arch.eff_db);
5314
5315		if (dr6 != 0) {
5316			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5317			kvm_run->debug.arch.pc = eip;
5318			kvm_run->debug.arch.exception = DB_VECTOR;
5319			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5320			*r = EMULATE_USER_EXIT;
5321			return true;
5322		}
5323	}
5324
5325	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5326	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5327		unsigned long eip = kvm_get_linear_rip(vcpu);
5328		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5329					   vcpu->arch.dr7,
5330					   vcpu->arch.db);
5331
5332		if (dr6 != 0) {
5333			vcpu->arch.dr6 &= ~15;
5334			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5335			kvm_queue_exception(vcpu, DB_VECTOR);
5336			*r = EMULATE_DONE;
5337			return true;
5338		}
5339	}
5340
5341	return false;
5342}
5343
5344int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5345			    unsigned long cr2,
5346			    int emulation_type,
5347			    void *insn,
5348			    int insn_len)
5349{
5350	int r;
5351	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5352	bool writeback = true;
5353	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5354
5355	/*
5356	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5357	 * never reused.
5358	 */
5359	vcpu->arch.write_fault_to_shadow_pgtable = false;
5360	kvm_clear_exception_queue(vcpu);
5361
5362	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5363		init_emulate_ctxt(vcpu);
5364
5365		/*
5366		 * We will reenter on the same instruction since
5367		 * we do not set complete_userspace_io.  This does not
5368		 * handle watchpoints yet, those would be handled in
5369		 * the emulate_ops.
5370		 */
5371		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5372			return r;
5373
5374		ctxt->interruptibility = 0;
5375		ctxt->have_exception = false;
5376		ctxt->exception.vector = -1;
5377		ctxt->perm_ok = false;
5378
5379		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5380
5381		r = x86_decode_insn(ctxt, insn, insn_len);
5382
5383		trace_kvm_emulate_insn_start(vcpu);
5384		++vcpu->stat.insn_emulation;
5385		if (r != EMULATION_OK)  {
5386			if (emulation_type & EMULTYPE_TRAP_UD)
5387				return EMULATE_FAIL;
5388			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5389						emulation_type))
5390				return EMULATE_DONE;
5391			if (emulation_type & EMULTYPE_SKIP)
5392				return EMULATE_FAIL;
5393			return handle_emulation_failure(vcpu);
5394		}
5395	}
5396
5397	if (emulation_type & EMULTYPE_SKIP) {
5398		kvm_rip_write(vcpu, ctxt->_eip);
5399		if (ctxt->eflags & X86_EFLAGS_RF)
5400			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5401		return EMULATE_DONE;
5402	}
5403
5404	if (retry_instruction(ctxt, cr2, emulation_type))
5405		return EMULATE_DONE;
5406
5407	/* this is needed for vmware backdoor interface to work since it
5408	   changes registers values  during IO operation */
5409	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5410		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5411		emulator_invalidate_register_cache(ctxt);
5412	}
5413
5414restart:
5415	r = x86_emulate_insn(ctxt);
5416
5417	if (r == EMULATION_INTERCEPTED)
5418		return EMULATE_DONE;
5419
5420	if (r == EMULATION_FAILED) {
5421		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5422					emulation_type))
5423			return EMULATE_DONE;
5424
5425		return handle_emulation_failure(vcpu);
5426	}
5427
5428	if (ctxt->have_exception) {
5429		r = EMULATE_DONE;
5430		if (inject_emulated_exception(vcpu))
5431			return r;
5432	} else if (vcpu->arch.pio.count) {
5433		if (!vcpu->arch.pio.in) {
5434			/* FIXME: return into emulator if single-stepping.  */
5435			vcpu->arch.pio.count = 0;
5436		} else {
5437			writeback = false;
5438			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5439		}
5440		r = EMULATE_USER_EXIT;
5441	} else if (vcpu->mmio_needed) {
5442		if (!vcpu->mmio_is_write)
5443			writeback = false;
5444		r = EMULATE_USER_EXIT;
5445		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5446	} else if (r == EMULATION_RESTART)
5447		goto restart;
5448	else
5449		r = EMULATE_DONE;
5450
5451	if (writeback) {
5452		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5453		toggle_interruptibility(vcpu, ctxt->interruptibility);
5454		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5455		if (vcpu->arch.hflags != ctxt->emul_flags)
5456			kvm_set_hflags(vcpu, ctxt->emul_flags);
5457		kvm_rip_write(vcpu, ctxt->eip);
5458		if (r == EMULATE_DONE)
5459			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5460		if (!ctxt->have_exception ||
5461		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5462			__kvm_set_rflags(vcpu, ctxt->eflags);
5463
5464		/*
5465		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5466		 * do nothing, and it will be requested again as soon as
5467		 * the shadow expires.  But we still need to check here,
5468		 * because POPF has no interrupt shadow.
5469		 */
5470		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5471			kvm_make_request(KVM_REQ_EVENT, vcpu);
5472	} else
5473		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5474
5475	return r;
5476}
5477EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5478
5479int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5480{
5481	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5482	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5483					    size, port, &val, 1);
5484	/* do not return to emulator after return from userspace */
5485	vcpu->arch.pio.count = 0;
5486	return ret;
5487}
5488EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5489
5490static void tsc_bad(void *info)
5491{
5492	__this_cpu_write(cpu_tsc_khz, 0);
5493}
5494
5495static void tsc_khz_changed(void *data)
5496{
5497	struct cpufreq_freqs *freq = data;
5498	unsigned long khz = 0;
5499
5500	if (data)
5501		khz = freq->new;
5502	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5503		khz = cpufreq_quick_get(raw_smp_processor_id());
5504	if (!khz)
5505		khz = tsc_khz;
5506	__this_cpu_write(cpu_tsc_khz, khz);
5507}
5508
5509static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5510				     void *data)
5511{
5512	struct cpufreq_freqs *freq = data;
5513	struct kvm *kvm;
5514	struct kvm_vcpu *vcpu;
5515	int i, send_ipi = 0;
5516
5517	/*
5518	 * We allow guests to temporarily run on slowing clocks,
5519	 * provided we notify them after, or to run on accelerating
5520	 * clocks, provided we notify them before.  Thus time never
5521	 * goes backwards.
5522	 *
5523	 * However, we have a problem.  We can't atomically update
5524	 * the frequency of a given CPU from this function; it is
5525	 * merely a notifier, which can be called from any CPU.
5526	 * Changing the TSC frequency at arbitrary points in time
5527	 * requires a recomputation of local variables related to
5528	 * the TSC for each VCPU.  We must flag these local variables
5529	 * to be updated and be sure the update takes place with the
5530	 * new frequency before any guests proceed.
5531	 *
5532	 * Unfortunately, the combination of hotplug CPU and frequency
5533	 * change creates an intractable locking scenario; the order
5534	 * of when these callouts happen is undefined with respect to
5535	 * CPU hotplug, and they can race with each other.  As such,
5536	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5537	 * undefined; you can actually have a CPU frequency change take
5538	 * place in between the computation of X and the setting of the
5539	 * variable.  To protect against this problem, all updates of
5540	 * the per_cpu tsc_khz variable are done in an interrupt
5541	 * protected IPI, and all callers wishing to update the value
5542	 * must wait for a synchronous IPI to complete (which is trivial
5543	 * if the caller is on the CPU already).  This establishes the
5544	 * necessary total order on variable updates.
5545	 *
5546	 * Note that because a guest time update may take place
5547	 * anytime after the setting of the VCPU's request bit, the
5548	 * correct TSC value must be set before the request.  However,
5549	 * to ensure the update actually makes it to any guest which
5550	 * starts running in hardware virtualization between the set
5551	 * and the acquisition of the spinlock, we must also ping the
5552	 * CPU after setting the request bit.
5553	 *
5554	 */
5555
5556	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5557		return 0;
5558	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5559		return 0;
5560
5561	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5562
5563	spin_lock(&kvm_lock);
5564	list_for_each_entry(kvm, &vm_list, vm_list) {
5565		kvm_for_each_vcpu(i, vcpu, kvm) {
5566			if (vcpu->cpu != freq->cpu)
5567				continue;
5568			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5569			if (vcpu->cpu != smp_processor_id())
5570				send_ipi = 1;
5571		}
5572	}
5573	spin_unlock(&kvm_lock);
5574
5575	if (freq->old < freq->new && send_ipi) {
5576		/*
5577		 * We upscale the frequency.  Must make the guest
5578		 * doesn't see old kvmclock values while running with
5579		 * the new frequency, otherwise we risk the guest sees
5580		 * time go backwards.
5581		 *
5582		 * In case we update the frequency for another cpu
5583		 * (which might be in guest context) send an interrupt
5584		 * to kick the cpu out of guest context.  Next time
5585		 * guest context is entered kvmclock will be updated,
5586		 * so the guest will not see stale values.
5587		 */
5588		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5589	}
5590	return 0;
5591}
5592
5593static struct notifier_block kvmclock_cpufreq_notifier_block = {
5594	.notifier_call  = kvmclock_cpufreq_notifier
5595};
5596
5597static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5598					unsigned long action, void *hcpu)
5599{
5600	unsigned int cpu = (unsigned long)hcpu;
5601
5602	switch (action) {
5603		case CPU_ONLINE:
5604		case CPU_DOWN_FAILED:
5605			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5606			break;
5607		case CPU_DOWN_PREPARE:
5608			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5609			break;
5610	}
5611	return NOTIFY_OK;
5612}
5613
5614static struct notifier_block kvmclock_cpu_notifier_block = {
5615	.notifier_call  = kvmclock_cpu_notifier,
5616	.priority = -INT_MAX
5617};
5618
5619static void kvm_timer_init(void)
5620{
5621	int cpu;
5622
5623	max_tsc_khz = tsc_khz;
5624
5625	cpu_notifier_register_begin();
5626	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5627#ifdef CONFIG_CPU_FREQ
5628		struct cpufreq_policy policy;
5629		memset(&policy, 0, sizeof(policy));
5630		cpu = get_cpu();
5631		cpufreq_get_policy(&policy, cpu);
5632		if (policy.cpuinfo.max_freq)
5633			max_tsc_khz = policy.cpuinfo.max_freq;
5634		put_cpu();
5635#endif
5636		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5637					  CPUFREQ_TRANSITION_NOTIFIER);
5638	}
5639	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5640	for_each_online_cpu(cpu)
5641		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5642
5643	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5644	cpu_notifier_register_done();
5645
5646}
5647
5648static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5649
5650int kvm_is_in_guest(void)
5651{
5652	return __this_cpu_read(current_vcpu) != NULL;
5653}
5654
5655static int kvm_is_user_mode(void)
5656{
5657	int user_mode = 3;
5658
5659	if (__this_cpu_read(current_vcpu))
5660		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5661
5662	return user_mode != 0;
5663}
5664
5665static unsigned long kvm_get_guest_ip(void)
5666{
5667	unsigned long ip = 0;
5668
5669	if (__this_cpu_read(current_vcpu))
5670		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5671
5672	return ip;
5673}
5674
5675static struct perf_guest_info_callbacks kvm_guest_cbs = {
5676	.is_in_guest		= kvm_is_in_guest,
5677	.is_user_mode		= kvm_is_user_mode,
5678	.get_guest_ip		= kvm_get_guest_ip,
5679};
5680
5681void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5682{
5683	__this_cpu_write(current_vcpu, vcpu);
5684}
5685EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5686
5687void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5688{
5689	__this_cpu_write(current_vcpu, NULL);
5690}
5691EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5692
5693static void kvm_set_mmio_spte_mask(void)
5694{
5695	u64 mask;
5696	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5697
5698	/*
5699	 * Set the reserved bits and the present bit of an paging-structure
5700	 * entry to generate page fault with PFER.RSV = 1.
5701	 */
5702	 /* Mask the reserved physical address bits. */
5703	mask = rsvd_bits(maxphyaddr, 51);
5704
5705	/* Bit 62 is always reserved for 32bit host. */
5706	mask |= 0x3ull << 62;
5707
5708	/* Set the present bit. */
5709	mask |= 1ull;
5710
5711#ifdef CONFIG_X86_64
5712	/*
5713	 * If reserved bit is not supported, clear the present bit to disable
5714	 * mmio page fault.
5715	 */
5716	if (maxphyaddr == 52)
5717		mask &= ~1ull;
5718#endif
5719
5720	kvm_mmu_set_mmio_spte_mask(mask);
5721}
5722
5723#ifdef CONFIG_X86_64
5724static void pvclock_gtod_update_fn(struct work_struct *work)
5725{
5726	struct kvm *kvm;
5727
5728	struct kvm_vcpu *vcpu;
5729	int i;
5730
5731	spin_lock(&kvm_lock);
5732	list_for_each_entry(kvm, &vm_list, vm_list)
5733		kvm_for_each_vcpu(i, vcpu, kvm)
5734			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5735	atomic_set(&kvm_guest_has_master_clock, 0);
5736	spin_unlock(&kvm_lock);
5737}
5738
5739static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5740
5741/*
5742 * Notification about pvclock gtod data update.
5743 */
5744static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5745			       void *priv)
5746{
5747	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5748	struct timekeeper *tk = priv;
5749
5750	update_pvclock_gtod(tk);
5751
5752	/* disable master clock if host does not trust, or does not
5753	 * use, TSC clocksource
5754	 */
5755	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5756	    atomic_read(&kvm_guest_has_master_clock) != 0)
5757		queue_work(system_long_wq, &pvclock_gtod_work);
5758
5759	return 0;
5760}
5761
5762static struct notifier_block pvclock_gtod_notifier = {
5763	.notifier_call = pvclock_gtod_notify,
5764};
5765#endif
5766
5767int kvm_arch_init(void *opaque)
5768{
5769	int r;
5770	struct kvm_x86_ops *ops = opaque;
5771
5772	if (kvm_x86_ops) {
5773		printk(KERN_ERR "kvm: already loaded the other module\n");
5774		r = -EEXIST;
5775		goto out;
5776	}
5777
5778	if (!ops->cpu_has_kvm_support()) {
5779		printk(KERN_ERR "kvm: no hardware support\n");
5780		r = -EOPNOTSUPP;
5781		goto out;
5782	}
5783	if (ops->disabled_by_bios()) {
5784		printk(KERN_ERR "kvm: disabled by bios\n");
5785		r = -EOPNOTSUPP;
5786		goto out;
5787	}
5788
5789	r = -ENOMEM;
5790	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5791	if (!shared_msrs) {
5792		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5793		goto out;
5794	}
5795
5796	r = kvm_mmu_module_init();
5797	if (r)
5798		goto out_free_percpu;
5799
5800	kvm_set_mmio_spte_mask();
5801
5802	kvm_x86_ops = ops;
5803
5804	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5805			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5806
5807	kvm_timer_init();
5808
5809	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5810
5811	if (cpu_has_xsave)
5812		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5813
5814	kvm_lapic_init();
5815#ifdef CONFIG_X86_64
5816	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5817#endif
5818
5819	return 0;
5820
5821out_free_percpu:
5822	free_percpu(shared_msrs);
5823out:
5824	return r;
5825}
5826
5827void kvm_arch_exit(void)
5828{
5829	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5830
5831	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5832		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5833					    CPUFREQ_TRANSITION_NOTIFIER);
5834	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5835#ifdef CONFIG_X86_64
5836	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5837#endif
5838	kvm_x86_ops = NULL;
5839	kvm_mmu_module_exit();
5840	free_percpu(shared_msrs);
5841}
5842
5843int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5844{
5845	++vcpu->stat.halt_exits;
5846	if (lapic_in_kernel(vcpu)) {
5847		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5848		return 1;
5849	} else {
5850		vcpu->run->exit_reason = KVM_EXIT_HLT;
5851		return 0;
5852	}
5853}
5854EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5855
5856int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5857{
5858	kvm_x86_ops->skip_emulated_instruction(vcpu);
5859	return kvm_vcpu_halt(vcpu);
5860}
5861EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5862
5863/*
5864 * kvm_pv_kick_cpu_op:  Kick a vcpu.
5865 *
5866 * @apicid - apicid of vcpu to be kicked.
5867 */
5868static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5869{
5870	struct kvm_lapic_irq lapic_irq;
5871
5872	lapic_irq.shorthand = 0;
5873	lapic_irq.dest_mode = 0;
5874	lapic_irq.dest_id = apicid;
5875	lapic_irq.msi_redir_hint = false;
5876
5877	lapic_irq.delivery_mode = APIC_DM_REMRD;
5878	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5879}
5880
5881int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5882{
5883	unsigned long nr, a0, a1, a2, a3, ret;
5884	int op_64_bit, r = 1;
5885
5886	kvm_x86_ops->skip_emulated_instruction(vcpu);
5887
5888	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5889		return kvm_hv_hypercall(vcpu);
5890
5891	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5892	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5893	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5894	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5895	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5896
5897	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5898
5899	op_64_bit = is_64_bit_mode(vcpu);
5900	if (!op_64_bit) {
5901		nr &= 0xFFFFFFFF;
5902		a0 &= 0xFFFFFFFF;
5903		a1 &= 0xFFFFFFFF;
5904		a2 &= 0xFFFFFFFF;
5905		a3 &= 0xFFFFFFFF;
5906	}
5907
5908	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5909		ret = -KVM_EPERM;
5910		goto out;
5911	}
5912
5913	switch (nr) {
5914	case KVM_HC_VAPIC_POLL_IRQ:
5915		ret = 0;
5916		break;
5917	case KVM_HC_KICK_CPU:
5918		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5919		ret = 0;
5920		break;
5921	default:
5922		ret = -KVM_ENOSYS;
5923		break;
5924	}
5925out:
5926	if (!op_64_bit)
5927		ret = (u32)ret;
5928	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5929	++vcpu->stat.hypercalls;
5930	return r;
5931}
5932EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5933
5934static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5935{
5936	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5937	char instruction[3];
5938	unsigned long rip = kvm_rip_read(vcpu);
5939
5940	kvm_x86_ops->patch_hypercall(vcpu, instruction);
5941
5942	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5943}
5944
5945static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5946{
5947	return vcpu->run->request_interrupt_window &&
5948		likely(!pic_in_kernel(vcpu->kvm));
5949}
5950
5951static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5952{
5953	struct kvm_run *kvm_run = vcpu->run;
5954
5955	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5956	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5957	kvm_run->cr8 = kvm_get_cr8(vcpu);
5958	kvm_run->apic_base = kvm_get_apic_base(vcpu);
5959	kvm_run->ready_for_interrupt_injection =
5960		pic_in_kernel(vcpu->kvm) ||
5961		kvm_vcpu_ready_for_interrupt_injection(vcpu);
5962}
5963
5964static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5965{
5966	int max_irr, tpr;
5967
5968	if (!kvm_x86_ops->update_cr8_intercept)
5969		return;
5970
5971	if (!vcpu->arch.apic)
5972		return;
5973
5974	if (!vcpu->arch.apic->vapic_addr)
5975		max_irr = kvm_lapic_find_highest_irr(vcpu);
5976	else
5977		max_irr = -1;
5978
5979	if (max_irr != -1)
5980		max_irr >>= 4;
5981
5982	tpr = kvm_lapic_get_cr8(vcpu);
5983
5984	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5985}
5986
5987static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5988{
5989	int r;
5990
5991	/* try to reinject previous events if any */
5992	if (vcpu->arch.exception.pending) {
5993		trace_kvm_inj_exception(vcpu->arch.exception.nr,
5994					vcpu->arch.exception.has_error_code,
5995					vcpu->arch.exception.error_code);
5996
5997		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5998			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5999					     X86_EFLAGS_RF);
6000
6001		if (vcpu->arch.exception.nr == DB_VECTOR &&
6002		    (vcpu->arch.dr7 & DR7_GD)) {
6003			vcpu->arch.dr7 &= ~DR7_GD;
6004			kvm_update_dr7(vcpu);
6005		}
6006
6007		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6008					  vcpu->arch.exception.has_error_code,
6009					  vcpu->arch.exception.error_code,
6010					  vcpu->arch.exception.reinject);
6011		return 0;
6012	}
6013
6014	if (vcpu->arch.nmi_injected) {
6015		kvm_x86_ops->set_nmi(vcpu);
6016		return 0;
6017	}
6018
6019	if (vcpu->arch.interrupt.pending) {
6020		kvm_x86_ops->set_irq(vcpu);
6021		return 0;
6022	}
6023
6024	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6025		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6026		if (r != 0)
6027			return r;
6028	}
6029
6030	/* try to inject new event if pending */
6031	if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6032		--vcpu->arch.nmi_pending;
6033		vcpu->arch.nmi_injected = true;
6034		kvm_x86_ops->set_nmi(vcpu);
6035	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6036		/*
6037		 * Because interrupts can be injected asynchronously, we are
6038		 * calling check_nested_events again here to avoid a race condition.
6039		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6040		 * proposal and current concerns.  Perhaps we should be setting
6041		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6042		 */
6043		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6044			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6045			if (r != 0)
6046				return r;
6047		}
6048		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6049			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6050					    false);
6051			kvm_x86_ops->set_irq(vcpu);
6052		}
6053	}
6054	return 0;
6055}
6056
6057static void process_nmi(struct kvm_vcpu *vcpu)
6058{
6059	unsigned limit = 2;
6060
6061	/*
6062	 * x86 is limited to one NMI running, and one NMI pending after it.
6063	 * If an NMI is already in progress, limit further NMIs to just one.
6064	 * Otherwise, allow two (and we'll inject the first one immediately).
6065	 */
6066	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6067		limit = 1;
6068
6069	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6070	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6071	kvm_make_request(KVM_REQ_EVENT, vcpu);
6072}
6073
6074#define put_smstate(type, buf, offset, val)			  \
6075	*(type *)((buf) + (offset) - 0x7e00) = val
6076
6077static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6078{
6079	u32 flags = 0;
6080	flags |= seg->g       << 23;
6081	flags |= seg->db      << 22;
6082	flags |= seg->l       << 21;
6083	flags |= seg->avl     << 20;
6084	flags |= seg->present << 15;
6085	flags |= seg->dpl     << 13;
6086	flags |= seg->s       << 12;
6087	flags |= seg->type    << 8;
6088	return flags;
6089}
6090
6091static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6092{
6093	struct kvm_segment seg;
6094	int offset;
6095
6096	kvm_get_segment(vcpu, &seg, n);
6097	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6098
6099	if (n < 3)
6100		offset = 0x7f84 + n * 12;
6101	else
6102		offset = 0x7f2c + (n - 3) * 12;
6103
6104	put_smstate(u32, buf, offset + 8, seg.base);
6105	put_smstate(u32, buf, offset + 4, seg.limit);
6106	put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6107}
6108
6109#ifdef CONFIG_X86_64
6110static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6111{
6112	struct kvm_segment seg;
6113	int offset;
6114	u16 flags;
6115
6116	kvm_get_segment(vcpu, &seg, n);
6117	offset = 0x7e00 + n * 16;
6118
6119	flags = process_smi_get_segment_flags(&seg) >> 8;
6120	put_smstate(u16, buf, offset, seg.selector);
6121	put_smstate(u16, buf, offset + 2, flags);
6122	put_smstate(u32, buf, offset + 4, seg.limit);
6123	put_smstate(u64, buf, offset + 8, seg.base);
6124}
6125#endif
6126
6127static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6128{
6129	struct desc_ptr dt;
6130	struct kvm_segment seg;
6131	unsigned long val;
6132	int i;
6133
6134	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6135	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6136	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6137	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6138
6139	for (i = 0; i < 8; i++)
6140		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6141
6142	kvm_get_dr(vcpu, 6, &val);
6143	put_smstate(u32, buf, 0x7fcc, (u32)val);
6144	kvm_get_dr(vcpu, 7, &val);
6145	put_smstate(u32, buf, 0x7fc8, (u32)val);
6146
6147	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6148	put_smstate(u32, buf, 0x7fc4, seg.selector);
6149	put_smstate(u32, buf, 0x7f64, seg.base);
6150	put_smstate(u32, buf, 0x7f60, seg.limit);
6151	put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6152
6153	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6154	put_smstate(u32, buf, 0x7fc0, seg.selector);
6155	put_smstate(u32, buf, 0x7f80, seg.base);
6156	put_smstate(u32, buf, 0x7f7c, seg.limit);
6157	put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6158
6159	kvm_x86_ops->get_gdt(vcpu, &dt);
6160	put_smstate(u32, buf, 0x7f74, dt.address);
6161	put_smstate(u32, buf, 0x7f70, dt.size);
6162
6163	kvm_x86_ops->get_idt(vcpu, &dt);
6164	put_smstate(u32, buf, 0x7f58, dt.address);
6165	put_smstate(u32, buf, 0x7f54, dt.size);
6166
6167	for (i = 0; i < 6; i++)
6168		process_smi_save_seg_32(vcpu, buf, i);
6169
6170	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6171
6172	/* revision id */
6173	put_smstate(u32, buf, 0x7efc, 0x00020000);
6174	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6175}
6176
6177static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6178{
6179#ifdef CONFIG_X86_64
6180	struct desc_ptr dt;
6181	struct kvm_segment seg;
6182	unsigned long val;
6183	int i;
6184
6185	for (i = 0; i < 16; i++)
6186		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6187
6188	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6189	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6190
6191	kvm_get_dr(vcpu, 6, &val);
6192	put_smstate(u64, buf, 0x7f68, val);
6193	kvm_get_dr(vcpu, 7, &val);
6194	put_smstate(u64, buf, 0x7f60, val);
6195
6196	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6197	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6198	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6199
6200	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6201
6202	/* revision id */
6203	put_smstate(u32, buf, 0x7efc, 0x00020064);
6204
6205	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6206
6207	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6208	put_smstate(u16, buf, 0x7e90, seg.selector);
6209	put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6210	put_smstate(u32, buf, 0x7e94, seg.limit);
6211	put_smstate(u64, buf, 0x7e98, seg.base);
6212
6213	kvm_x86_ops->get_idt(vcpu, &dt);
6214	put_smstate(u32, buf, 0x7e84, dt.size);
6215	put_smstate(u64, buf, 0x7e88, dt.address);
6216
6217	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6218	put_smstate(u16, buf, 0x7e70, seg.selector);
6219	put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6220	put_smstate(u32, buf, 0x7e74, seg.limit);
6221	put_smstate(u64, buf, 0x7e78, seg.base);
6222
6223	kvm_x86_ops->get_gdt(vcpu, &dt);
6224	put_smstate(u32, buf, 0x7e64, dt.size);
6225	put_smstate(u64, buf, 0x7e68, dt.address);
6226
6227	for (i = 0; i < 6; i++)
6228		process_smi_save_seg_64(vcpu, buf, i);
6229#else
6230	WARN_ON_ONCE(1);
6231#endif
6232}
6233
6234static void process_smi(struct kvm_vcpu *vcpu)
6235{
6236	struct kvm_segment cs, ds;
6237	struct desc_ptr dt;
6238	char buf[512];
6239	u32 cr0;
6240
6241	if (is_smm(vcpu)) {
6242		vcpu->arch.smi_pending = true;
6243		return;
6244	}
6245
6246	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6247	vcpu->arch.hflags |= HF_SMM_MASK;
6248	memset(buf, 0, 512);
6249	if (guest_cpuid_has_longmode(vcpu))
6250		process_smi_save_state_64(vcpu, buf);
6251	else
6252		process_smi_save_state_32(vcpu, buf);
6253
6254	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6255
6256	if (kvm_x86_ops->get_nmi_mask(vcpu))
6257		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6258	else
6259		kvm_x86_ops->set_nmi_mask(vcpu, true);
6260
6261	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6262	kvm_rip_write(vcpu, 0x8000);
6263
6264	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6265	kvm_x86_ops->set_cr0(vcpu, cr0);
6266	vcpu->arch.cr0 = cr0;
6267
6268	kvm_x86_ops->set_cr4(vcpu, 0);
6269
6270	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6271	dt.address = dt.size = 0;
6272	kvm_x86_ops->set_idt(vcpu, &dt);
6273
6274	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6275
6276	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6277	cs.base = vcpu->arch.smbase;
6278
6279	ds.selector = 0;
6280	ds.base = 0;
6281
6282	cs.limit    = ds.limit = 0xffffffff;
6283	cs.type     = ds.type = 0x3;
6284	cs.dpl      = ds.dpl = 0;
6285	cs.db       = ds.db = 0;
6286	cs.s        = ds.s = 1;
6287	cs.l        = ds.l = 0;
6288	cs.g        = ds.g = 1;
6289	cs.avl      = ds.avl = 0;
6290	cs.present  = ds.present = 1;
6291	cs.unusable = ds.unusable = 0;
6292	cs.padding  = ds.padding = 0;
6293
6294	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6295	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6296	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6297	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6298	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6299	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6300
6301	if (guest_cpuid_has_longmode(vcpu))
6302		kvm_x86_ops->set_efer(vcpu, 0);
6303
6304	kvm_update_cpuid(vcpu);
6305	kvm_mmu_reset_context(vcpu);
6306}
6307
6308static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6309{
6310	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6311		return;
6312
6313	memset(vcpu->arch.eoi_exit_bitmap, 0, 256 / 8);
6314
6315	if (irqchip_split(vcpu->kvm))
6316		kvm_scan_ioapic_routes(vcpu, vcpu->arch.eoi_exit_bitmap);
6317	else {
6318		kvm_x86_ops->sync_pir_to_irr(vcpu);
6319		kvm_ioapic_scan_entry(vcpu, vcpu->arch.eoi_exit_bitmap);
6320	}
6321	kvm_x86_ops->load_eoi_exitmap(vcpu);
6322}
6323
6324static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6325{
6326	++vcpu->stat.tlb_flush;
6327	kvm_x86_ops->tlb_flush(vcpu);
6328}
6329
6330void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6331{
6332	struct page *page = NULL;
6333
6334	if (!lapic_in_kernel(vcpu))
6335		return;
6336
6337	if (!kvm_x86_ops->set_apic_access_page_addr)
6338		return;
6339
6340	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6341	if (is_error_page(page))
6342		return;
6343	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6344
6345	/*
6346	 * Do not pin apic access page in memory, the MMU notifier
6347	 * will call us again if it is migrated or swapped out.
6348	 */
6349	put_page(page);
6350}
6351EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6352
6353void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6354					   unsigned long address)
6355{
6356	/*
6357	 * The physical address of apic access page is stored in the VMCS.
6358	 * Update it when it becomes invalid.
6359	 */
6360	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6361		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6362}
6363
6364/*
6365 * Returns 1 to let vcpu_run() continue the guest execution loop without
6366 * exiting to the userspace.  Otherwise, the value will be returned to the
6367 * userspace.
6368 */
6369static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6370{
6371	int r;
6372	bool req_int_win =
6373		dm_request_for_irq_injection(vcpu) &&
6374		kvm_cpu_accept_dm_intr(vcpu);
6375
6376	bool req_immediate_exit = false;
6377
6378	if (vcpu->requests) {
6379		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6380			kvm_mmu_unload(vcpu);
6381		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6382			__kvm_migrate_timers(vcpu);
6383		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6384			kvm_gen_update_masterclock(vcpu->kvm);
6385		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6386			kvm_gen_kvmclock_update(vcpu);
6387		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6388			r = kvm_guest_time_update(vcpu);
6389			if (unlikely(r))
6390				goto out;
6391		}
6392		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6393			kvm_mmu_sync_roots(vcpu);
6394		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6395			kvm_vcpu_flush_tlb(vcpu);
6396		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6397			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6398			r = 0;
6399			goto out;
6400		}
6401		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6402			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6403			r = 0;
6404			goto out;
6405		}
6406		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6407			vcpu->fpu_active = 0;
6408			kvm_x86_ops->fpu_deactivate(vcpu);
6409		}
6410		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6411			/* Page is swapped out. Do synthetic halt */
6412			vcpu->arch.apf.halted = true;
6413			r = 1;
6414			goto out;
6415		}
6416		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6417			record_steal_time(vcpu);
6418		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6419			process_smi(vcpu);
6420		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6421			process_nmi(vcpu);
6422		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6423			kvm_pmu_handle_event(vcpu);
6424		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6425			kvm_pmu_deliver_pmi(vcpu);
6426		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6427			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6428			if (test_bit(vcpu->arch.pending_ioapic_eoi,
6429				     (void *) vcpu->arch.eoi_exit_bitmap)) {
6430				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6431				vcpu->run->eoi.vector =
6432						vcpu->arch.pending_ioapic_eoi;
6433				r = 0;
6434				goto out;
6435			}
6436		}
6437		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6438			vcpu_scan_ioapic(vcpu);
6439		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6440			kvm_vcpu_reload_apic_access_page(vcpu);
6441		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6442			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6443			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6444			r = 0;
6445			goto out;
6446		}
6447		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6448			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6449			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6450			r = 0;
6451			goto out;
6452		}
6453	}
6454
6455	/*
6456	 * KVM_REQ_EVENT is not set when posted interrupts are set by
6457	 * VT-d hardware, so we have to update RVI unconditionally.
6458	 */
6459	if (kvm_lapic_enabled(vcpu)) {
6460		/*
6461		 * Update architecture specific hints for APIC
6462		 * virtual interrupt delivery.
6463		 */
6464		if (kvm_x86_ops->hwapic_irr_update)
6465			kvm_x86_ops->hwapic_irr_update(vcpu,
6466				kvm_lapic_find_highest_irr(vcpu));
6467	}
6468
6469	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6470		kvm_apic_accept_events(vcpu);
6471		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6472			r = 1;
6473			goto out;
6474		}
6475
6476		if (inject_pending_event(vcpu, req_int_win) != 0)
6477			req_immediate_exit = true;
6478		/* enable NMI/IRQ window open exits if needed */
6479		else {
6480			if (vcpu->arch.nmi_pending)
6481				kvm_x86_ops->enable_nmi_window(vcpu);
6482			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6483				kvm_x86_ops->enable_irq_window(vcpu);
6484		}
6485
6486		if (kvm_lapic_enabled(vcpu)) {
6487			update_cr8_intercept(vcpu);
6488			kvm_lapic_sync_to_vapic(vcpu);
6489		}
6490	}
6491
6492	r = kvm_mmu_reload(vcpu);
6493	if (unlikely(r)) {
6494		goto cancel_injection;
6495	}
6496
6497	preempt_disable();
6498
6499	kvm_x86_ops->prepare_guest_switch(vcpu);
6500	if (vcpu->fpu_active)
6501		kvm_load_guest_fpu(vcpu);
6502	vcpu->mode = IN_GUEST_MODE;
6503
6504	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6505
6506	/* We should set ->mode before check ->requests,
6507	 * see the comment in make_all_cpus_request.
6508	 */
6509	smp_mb__after_srcu_read_unlock();
6510
6511	local_irq_disable();
6512
6513	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6514	    || need_resched() || signal_pending(current)) {
6515		vcpu->mode = OUTSIDE_GUEST_MODE;
6516		smp_wmb();
6517		local_irq_enable();
6518		preempt_enable();
6519		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6520		r = 1;
6521		goto cancel_injection;
6522	}
6523
6524	kvm_load_guest_xcr0(vcpu);
6525
6526	if (req_immediate_exit)
6527		smp_send_reschedule(vcpu->cpu);
6528
6529	trace_kvm_entry(vcpu->vcpu_id);
6530	wait_lapic_expire(vcpu);
6531	__kvm_guest_enter();
6532
6533	if (unlikely(vcpu->arch.switch_db_regs)) {
6534		set_debugreg(0, 7);
6535		set_debugreg(vcpu->arch.eff_db[0], 0);
6536		set_debugreg(vcpu->arch.eff_db[1], 1);
6537		set_debugreg(vcpu->arch.eff_db[2], 2);
6538		set_debugreg(vcpu->arch.eff_db[3], 3);
6539		set_debugreg(vcpu->arch.dr6, 6);
6540		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6541	}
6542
6543	kvm_x86_ops->run(vcpu);
6544
6545	/*
6546	 * Do this here before restoring debug registers on the host.  And
6547	 * since we do this before handling the vmexit, a DR access vmexit
6548	 * can (a) read the correct value of the debug registers, (b) set
6549	 * KVM_DEBUGREG_WONT_EXIT again.
6550	 */
6551	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6552		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6553		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6554		kvm_update_dr0123(vcpu);
6555		kvm_update_dr6(vcpu);
6556		kvm_update_dr7(vcpu);
6557		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6558	}
6559
6560	/*
6561	 * If the guest has used debug registers, at least dr7
6562	 * will be disabled while returning to the host.
6563	 * If we don't have active breakpoints in the host, we don't
6564	 * care about the messed up debug address registers. But if
6565	 * we have some of them active, restore the old state.
6566	 */
6567	if (hw_breakpoint_active())
6568		hw_breakpoint_restore();
6569
6570	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6571
6572	vcpu->mode = OUTSIDE_GUEST_MODE;
6573	smp_wmb();
6574
6575	kvm_put_guest_xcr0(vcpu);
6576
6577	/* Interrupt is enabled by handle_external_intr() */
6578	kvm_x86_ops->handle_external_intr(vcpu);
6579
6580	++vcpu->stat.exits;
6581
6582	/*
6583	 * We must have an instruction between local_irq_enable() and
6584	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6585	 * the interrupt shadow.  The stat.exits increment will do nicely.
6586	 * But we need to prevent reordering, hence this barrier():
6587	 */
6588	barrier();
6589
6590	kvm_guest_exit();
6591
6592	preempt_enable();
6593
6594	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6595
6596	/*
6597	 * Profile KVM exit RIPs:
6598	 */
6599	if (unlikely(prof_on == KVM_PROFILING)) {
6600		unsigned long rip = kvm_rip_read(vcpu);
6601		profile_hit(KVM_PROFILING, (void *)rip);
6602	}
6603
6604	if (unlikely(vcpu->arch.tsc_always_catchup))
6605		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6606
6607	if (vcpu->arch.apic_attention)
6608		kvm_lapic_sync_from_vapic(vcpu);
6609
6610	r = kvm_x86_ops->handle_exit(vcpu);
6611	return r;
6612
6613cancel_injection:
6614	kvm_x86_ops->cancel_injection(vcpu);
6615	if (unlikely(vcpu->arch.apic_attention))
6616		kvm_lapic_sync_from_vapic(vcpu);
6617out:
6618	return r;
6619}
6620
6621static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6622{
6623	if (!kvm_arch_vcpu_runnable(vcpu) &&
6624	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6625		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6626		kvm_vcpu_block(vcpu);
6627		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6628
6629		if (kvm_x86_ops->post_block)
6630			kvm_x86_ops->post_block(vcpu);
6631
6632		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6633			return 1;
6634	}
6635
6636	kvm_apic_accept_events(vcpu);
6637	switch(vcpu->arch.mp_state) {
6638	case KVM_MP_STATE_HALTED:
6639		vcpu->arch.pv.pv_unhalted = false;
6640		vcpu->arch.mp_state =
6641			KVM_MP_STATE_RUNNABLE;
6642	case KVM_MP_STATE_RUNNABLE:
6643		vcpu->arch.apf.halted = false;
6644		break;
6645	case KVM_MP_STATE_INIT_RECEIVED:
6646		break;
6647	default:
6648		return -EINTR;
6649		break;
6650	}
6651	return 1;
6652}
6653
6654static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6655{
6656	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6657		!vcpu->arch.apf.halted);
6658}
6659
6660static int vcpu_run(struct kvm_vcpu *vcpu)
6661{
6662	int r;
6663	struct kvm *kvm = vcpu->kvm;
6664
6665	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6666
6667	for (;;) {
6668		if (kvm_vcpu_running(vcpu)) {
6669			r = vcpu_enter_guest(vcpu);
6670		} else {
6671			r = vcpu_block(kvm, vcpu);
6672		}
6673
6674		if (r <= 0)
6675			break;
6676
6677		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6678		if (kvm_cpu_has_pending_timer(vcpu))
6679			kvm_inject_pending_timer_irqs(vcpu);
6680
6681		if (dm_request_for_irq_injection(vcpu) &&
6682			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6683			r = 0;
6684			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6685			++vcpu->stat.request_irq_exits;
6686			break;
6687		}
6688
6689		kvm_check_async_pf_completion(vcpu);
6690
6691		if (signal_pending(current)) {
6692			r = -EINTR;
6693			vcpu->run->exit_reason = KVM_EXIT_INTR;
6694			++vcpu->stat.signal_exits;
6695			break;
6696		}
6697		if (need_resched()) {
6698			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6699			cond_resched();
6700			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6701		}
6702	}
6703
6704	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6705
6706	return r;
6707}
6708
6709static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6710{
6711	int r;
6712	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6713	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6714	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6715	if (r != EMULATE_DONE)
6716		return 0;
6717	return 1;
6718}
6719
6720static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6721{
6722	BUG_ON(!vcpu->arch.pio.count);
6723
6724	return complete_emulated_io(vcpu);
6725}
6726
6727/*
6728 * Implements the following, as a state machine:
6729 *
6730 * read:
6731 *   for each fragment
6732 *     for each mmio piece in the fragment
6733 *       write gpa, len
6734 *       exit
6735 *       copy data
6736 *   execute insn
6737 *
6738 * write:
6739 *   for each fragment
6740 *     for each mmio piece in the fragment
6741 *       write gpa, len
6742 *       copy data
6743 *       exit
6744 */
6745static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6746{
6747	struct kvm_run *run = vcpu->run;
6748	struct kvm_mmio_fragment *frag;
6749	unsigned len;
6750
6751	BUG_ON(!vcpu->mmio_needed);
6752
6753	/* Complete previous fragment */
6754	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6755	len = min(8u, frag->len);
6756	if (!vcpu->mmio_is_write)
6757		memcpy(frag->data, run->mmio.data, len);
6758
6759	if (frag->len <= 8) {
6760		/* Switch to the next fragment. */
6761		frag++;
6762		vcpu->mmio_cur_fragment++;
6763	} else {
6764		/* Go forward to the next mmio piece. */
6765		frag->data += len;
6766		frag->gpa += len;
6767		frag->len -= len;
6768	}
6769
6770	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6771		vcpu->mmio_needed = 0;
6772
6773		/* FIXME: return into emulator if single-stepping.  */
6774		if (vcpu->mmio_is_write)
6775			return 1;
6776		vcpu->mmio_read_completed = 1;
6777		return complete_emulated_io(vcpu);
6778	}
6779
6780	run->exit_reason = KVM_EXIT_MMIO;
6781	run->mmio.phys_addr = frag->gpa;
6782	if (vcpu->mmio_is_write)
6783		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6784	run->mmio.len = min(8u, frag->len);
6785	run->mmio.is_write = vcpu->mmio_is_write;
6786	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6787	return 0;
6788}
6789
6790
6791int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6792{
6793	struct fpu *fpu = &current->thread.fpu;
6794	int r;
6795	sigset_t sigsaved;
6796
6797	fpu__activate_curr(fpu);
6798
6799	if (vcpu->sigset_active)
6800		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6801
6802	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6803		kvm_vcpu_block(vcpu);
6804		kvm_apic_accept_events(vcpu);
6805		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6806		r = -EAGAIN;
6807		goto out;
6808	}
6809
6810	/* re-sync apic's tpr */
6811	if (!lapic_in_kernel(vcpu)) {
6812		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6813			r = -EINVAL;
6814			goto out;
6815		}
6816	}
6817
6818	if (unlikely(vcpu->arch.complete_userspace_io)) {
6819		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6820		vcpu->arch.complete_userspace_io = NULL;
6821		r = cui(vcpu);
6822		if (r <= 0)
6823			goto out;
6824	} else
6825		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6826
6827	r = vcpu_run(vcpu);
6828
6829out:
6830	post_kvm_run_save(vcpu);
6831	if (vcpu->sigset_active)
6832		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6833
6834	return r;
6835}
6836
6837int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6838{
6839	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6840		/*
6841		 * We are here if userspace calls get_regs() in the middle of
6842		 * instruction emulation. Registers state needs to be copied
6843		 * back from emulation context to vcpu. Userspace shouldn't do
6844		 * that usually, but some bad designed PV devices (vmware
6845		 * backdoor interface) need this to work
6846		 */
6847		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6848		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6849	}
6850	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6851	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6852	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6853	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6854	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6855	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6856	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6857	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6858#ifdef CONFIG_X86_64
6859	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6860	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6861	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6862	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6863	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6864	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6865	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6866	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6867#endif
6868
6869	regs->rip = kvm_rip_read(vcpu);
6870	regs->rflags = kvm_get_rflags(vcpu);
6871
6872	return 0;
6873}
6874
6875int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6876{
6877	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6878	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6879
6880	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6881	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6882	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6883	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6884	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6885	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6886	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6887	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6888#ifdef CONFIG_X86_64
6889	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6890	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6891	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6892	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6893	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6894	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6895	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6896	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6897#endif
6898
6899	kvm_rip_write(vcpu, regs->rip);
6900	kvm_set_rflags(vcpu, regs->rflags);
6901
6902	vcpu->arch.exception.pending = false;
6903
6904	kvm_make_request(KVM_REQ_EVENT, vcpu);
6905
6906	return 0;
6907}
6908
6909void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6910{
6911	struct kvm_segment cs;
6912
6913	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6914	*db = cs.db;
6915	*l = cs.l;
6916}
6917EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6918
6919int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6920				  struct kvm_sregs *sregs)
6921{
6922	struct desc_ptr dt;
6923
6924	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6925	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6926	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6927	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6928	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6929	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6930
6931	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6932	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6933
6934	kvm_x86_ops->get_idt(vcpu, &dt);
6935	sregs->idt.limit = dt.size;
6936	sregs->idt.base = dt.address;
6937	kvm_x86_ops->get_gdt(vcpu, &dt);
6938	sregs->gdt.limit = dt.size;
6939	sregs->gdt.base = dt.address;
6940
6941	sregs->cr0 = kvm_read_cr0(vcpu);
6942	sregs->cr2 = vcpu->arch.cr2;
6943	sregs->cr3 = kvm_read_cr3(vcpu);
6944	sregs->cr4 = kvm_read_cr4(vcpu);
6945	sregs->cr8 = kvm_get_cr8(vcpu);
6946	sregs->efer = vcpu->arch.efer;
6947	sregs->apic_base = kvm_get_apic_base(vcpu);
6948
6949	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6950
6951	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6952		set_bit(vcpu->arch.interrupt.nr,
6953			(unsigned long *)sregs->interrupt_bitmap);
6954
6955	return 0;
6956}
6957
6958int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6959				    struct kvm_mp_state *mp_state)
6960{
6961	kvm_apic_accept_events(vcpu);
6962	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6963					vcpu->arch.pv.pv_unhalted)
6964		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6965	else
6966		mp_state->mp_state = vcpu->arch.mp_state;
6967
6968	return 0;
6969}
6970
6971int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6972				    struct kvm_mp_state *mp_state)
6973{
6974	if (!kvm_vcpu_has_lapic(vcpu) &&
6975	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6976		return -EINVAL;
6977
6978	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6979		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6980		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6981	} else
6982		vcpu->arch.mp_state = mp_state->mp_state;
6983	kvm_make_request(KVM_REQ_EVENT, vcpu);
6984	return 0;
6985}
6986
6987int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6988		    int reason, bool has_error_code, u32 error_code)
6989{
6990	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6991	int ret;
6992
6993	init_emulate_ctxt(vcpu);
6994
6995	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6996				   has_error_code, error_code);
6997
6998	if (ret)
6999		return EMULATE_FAIL;
7000
7001	kvm_rip_write(vcpu, ctxt->eip);
7002	kvm_set_rflags(vcpu, ctxt->eflags);
7003	kvm_make_request(KVM_REQ_EVENT, vcpu);
7004	return EMULATE_DONE;
7005}
7006EXPORT_SYMBOL_GPL(kvm_task_switch);
7007
7008int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7009				  struct kvm_sregs *sregs)
7010{
7011	struct msr_data apic_base_msr;
7012	int mmu_reset_needed = 0;
7013	int pending_vec, max_bits, idx;
7014	struct desc_ptr dt;
7015
7016	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7017		return -EINVAL;
7018
7019	dt.size = sregs->idt.limit;
7020	dt.address = sregs->idt.base;
7021	kvm_x86_ops->set_idt(vcpu, &dt);
7022	dt.size = sregs->gdt.limit;
7023	dt.address = sregs->gdt.base;
7024	kvm_x86_ops->set_gdt(vcpu, &dt);
7025
7026	vcpu->arch.cr2 = sregs->cr2;
7027	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7028	vcpu->arch.cr3 = sregs->cr3;
7029	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7030
7031	kvm_set_cr8(vcpu, sregs->cr8);
7032
7033	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7034	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7035	apic_base_msr.data = sregs->apic_base;
7036	apic_base_msr.host_initiated = true;
7037	kvm_set_apic_base(vcpu, &apic_base_msr);
7038
7039	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7040	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7041	vcpu->arch.cr0 = sregs->cr0;
7042
7043	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7044	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7045	if (sregs->cr4 & X86_CR4_OSXSAVE)
7046		kvm_update_cpuid(vcpu);
7047
7048	idx = srcu_read_lock(&vcpu->kvm->srcu);
7049	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7050		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7051		mmu_reset_needed = 1;
7052	}
7053	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7054
7055	if (mmu_reset_needed)
7056		kvm_mmu_reset_context(vcpu);
7057
7058	max_bits = KVM_NR_INTERRUPTS;
7059	pending_vec = find_first_bit(
7060		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7061	if (pending_vec < max_bits) {
7062		kvm_queue_interrupt(vcpu, pending_vec, false);
7063		pr_debug("Set back pending irq %d\n", pending_vec);
7064	}
7065
7066	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7067	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7068	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7069	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7070	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7071	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7072
7073	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7074	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7075
7076	update_cr8_intercept(vcpu);
7077
7078	/* Older userspace won't unhalt the vcpu on reset. */
7079	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7080	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7081	    !is_protmode(vcpu))
7082		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7083
7084	kvm_make_request(KVM_REQ_EVENT, vcpu);
7085
7086	return 0;
7087}
7088
7089int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7090					struct kvm_guest_debug *dbg)
7091{
7092	unsigned long rflags;
7093	int i, r;
7094
7095	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7096		r = -EBUSY;
7097		if (vcpu->arch.exception.pending)
7098			goto out;
7099		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7100			kvm_queue_exception(vcpu, DB_VECTOR);
7101		else
7102			kvm_queue_exception(vcpu, BP_VECTOR);
7103	}
7104
7105	/*
7106	 * Read rflags as long as potentially injected trace flags are still
7107	 * filtered out.
7108	 */
7109	rflags = kvm_get_rflags(vcpu);
7110
7111	vcpu->guest_debug = dbg->control;
7112	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7113		vcpu->guest_debug = 0;
7114
7115	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7116		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7117			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7118		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7119	} else {
7120		for (i = 0; i < KVM_NR_DB_REGS; i++)
7121			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7122	}
7123	kvm_update_dr7(vcpu);
7124
7125	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7126		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7127			get_segment_base(vcpu, VCPU_SREG_CS);
7128
7129	/*
7130	 * Trigger an rflags update that will inject or remove the trace
7131	 * flags.
7132	 */
7133	kvm_set_rflags(vcpu, rflags);
7134
7135	kvm_x86_ops->update_bp_intercept(vcpu);
7136
7137	r = 0;
7138
7139out:
7140
7141	return r;
7142}
7143
7144/*
7145 * Translate a guest virtual address to a guest physical address.
7146 */
7147int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7148				    struct kvm_translation *tr)
7149{
7150	unsigned long vaddr = tr->linear_address;
7151	gpa_t gpa;
7152	int idx;
7153
7154	idx = srcu_read_lock(&vcpu->kvm->srcu);
7155	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7156	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7157	tr->physical_address = gpa;
7158	tr->valid = gpa != UNMAPPED_GVA;
7159	tr->writeable = 1;
7160	tr->usermode = 0;
7161
7162	return 0;
7163}
7164
7165int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7166{
7167	struct fxregs_state *fxsave =
7168			&vcpu->arch.guest_fpu.state.fxsave;
7169
7170	memcpy(fpu->fpr, fxsave->st_space, 128);
7171	fpu->fcw = fxsave->cwd;
7172	fpu->fsw = fxsave->swd;
7173	fpu->ftwx = fxsave->twd;
7174	fpu->last_opcode = fxsave->fop;
7175	fpu->last_ip = fxsave->rip;
7176	fpu->last_dp = fxsave->rdp;
7177	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7178
7179	return 0;
7180}
7181
7182int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7183{
7184	struct fxregs_state *fxsave =
7185			&vcpu->arch.guest_fpu.state.fxsave;
7186
7187	memcpy(fxsave->st_space, fpu->fpr, 128);
7188	fxsave->cwd = fpu->fcw;
7189	fxsave->swd = fpu->fsw;
7190	fxsave->twd = fpu->ftwx;
7191	fxsave->fop = fpu->last_opcode;
7192	fxsave->rip = fpu->last_ip;
7193	fxsave->rdp = fpu->last_dp;
7194	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7195
7196	return 0;
7197}
7198
7199static void fx_init(struct kvm_vcpu *vcpu)
7200{
7201	fpstate_init(&vcpu->arch.guest_fpu.state);
7202	if (cpu_has_xsaves)
7203		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7204			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7205
7206	/*
7207	 * Ensure guest xcr0 is valid for loading
7208	 */
7209	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7210
7211	vcpu->arch.cr0 |= X86_CR0_ET;
7212}
7213
7214void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7215{
7216	if (vcpu->guest_fpu_loaded)
7217		return;
7218
7219	/*
7220	 * Restore all possible states in the guest,
7221	 * and assume host would use all available bits.
7222	 * Guest xcr0 would be loaded later.
7223	 */
7224	vcpu->guest_fpu_loaded = 1;
7225	__kernel_fpu_begin();
7226	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7227	trace_kvm_fpu(1);
7228}
7229
7230void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7231{
7232	if (!vcpu->guest_fpu_loaded) {
7233		vcpu->fpu_counter = 0;
7234		return;
7235	}
7236
7237	vcpu->guest_fpu_loaded = 0;
7238	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7239	__kernel_fpu_end();
7240	++vcpu->stat.fpu_reload;
7241	/*
7242	 * If using eager FPU mode, or if the guest is a frequent user
7243	 * of the FPU, just leave the FPU active for next time.
7244	 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7245	 * the FPU in bursts will revert to loading it on demand.
7246	 */
7247	if (!vcpu->arch.eager_fpu) {
7248		if (++vcpu->fpu_counter < 5)
7249			kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7250	}
7251	trace_kvm_fpu(0);
7252}
7253
7254void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7255{
7256	kvmclock_reset(vcpu);
7257
7258	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7259	kvm_x86_ops->vcpu_free(vcpu);
7260}
7261
7262struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7263						unsigned int id)
7264{
7265	struct kvm_vcpu *vcpu;
7266
7267	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7268		printk_once(KERN_WARNING
7269		"kvm: SMP vm created on host with unstable TSC; "
7270		"guest TSC will not be reliable\n");
7271
7272	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7273
7274	return vcpu;
7275}
7276
7277int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7278{
7279	int r;
7280
7281	kvm_vcpu_mtrr_init(vcpu);
7282	r = vcpu_load(vcpu);
7283	if (r)
7284		return r;
7285	kvm_vcpu_reset(vcpu, false);
7286	kvm_mmu_setup(vcpu);
7287	vcpu_put(vcpu);
7288	return r;
7289}
7290
7291void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7292{
7293	struct msr_data msr;
7294	struct kvm *kvm = vcpu->kvm;
7295
7296	if (vcpu_load(vcpu))
7297		return;
7298	msr.data = 0x0;
7299	msr.index = MSR_IA32_TSC;
7300	msr.host_initiated = true;
7301	kvm_write_tsc(vcpu, &msr);
7302	vcpu_put(vcpu);
7303
7304	if (!kvmclock_periodic_sync)
7305		return;
7306
7307	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7308					KVMCLOCK_SYNC_PERIOD);
7309}
7310
7311void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7312{
7313	int r;
7314	vcpu->arch.apf.msr_val = 0;
7315
7316	r = vcpu_load(vcpu);
7317	BUG_ON(r);
7318	kvm_mmu_unload(vcpu);
7319	vcpu_put(vcpu);
7320
7321	kvm_x86_ops->vcpu_free(vcpu);
7322}
7323
7324void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7325{
7326	vcpu->arch.hflags = 0;
7327
7328	atomic_set(&vcpu->arch.nmi_queued, 0);
7329	vcpu->arch.nmi_pending = 0;
7330	vcpu->arch.nmi_injected = false;
7331	kvm_clear_interrupt_queue(vcpu);
7332	kvm_clear_exception_queue(vcpu);
7333
7334	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7335	kvm_update_dr0123(vcpu);
7336	vcpu->arch.dr6 = DR6_INIT;
7337	kvm_update_dr6(vcpu);
7338	vcpu->arch.dr7 = DR7_FIXED_1;
7339	kvm_update_dr7(vcpu);
7340
7341	vcpu->arch.cr2 = 0;
7342
7343	kvm_make_request(KVM_REQ_EVENT, vcpu);
7344	vcpu->arch.apf.msr_val = 0;
7345	vcpu->arch.st.msr_val = 0;
7346
7347	kvmclock_reset(vcpu);
7348
7349	kvm_clear_async_pf_completion_queue(vcpu);
7350	kvm_async_pf_hash_reset(vcpu);
7351	vcpu->arch.apf.halted = false;
7352
7353	if (!init_event) {
7354		kvm_pmu_reset(vcpu);
7355		vcpu->arch.smbase = 0x30000;
7356	}
7357
7358	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7359	vcpu->arch.regs_avail = ~0;
7360	vcpu->arch.regs_dirty = ~0;
7361
7362	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7363}
7364
7365void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7366{
7367	struct kvm_segment cs;
7368
7369	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7370	cs.selector = vector << 8;
7371	cs.base = vector << 12;
7372	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7373	kvm_rip_write(vcpu, 0);
7374}
7375
7376int kvm_arch_hardware_enable(void)
7377{
7378	struct kvm *kvm;
7379	struct kvm_vcpu *vcpu;
7380	int i;
7381	int ret;
7382	u64 local_tsc;
7383	u64 max_tsc = 0;
7384	bool stable, backwards_tsc = false;
7385
7386	kvm_shared_msr_cpu_online();
7387	ret = kvm_x86_ops->hardware_enable();
7388	if (ret != 0)
7389		return ret;
7390
7391	local_tsc = rdtsc();
7392	stable = !check_tsc_unstable();
7393	list_for_each_entry(kvm, &vm_list, vm_list) {
7394		kvm_for_each_vcpu(i, vcpu, kvm) {
7395			if (!stable && vcpu->cpu == smp_processor_id())
7396				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7397			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7398				backwards_tsc = true;
7399				if (vcpu->arch.last_host_tsc > max_tsc)
7400					max_tsc = vcpu->arch.last_host_tsc;
7401			}
7402		}
7403	}
7404
7405	/*
7406	 * Sometimes, even reliable TSCs go backwards.  This happens on
7407	 * platforms that reset TSC during suspend or hibernate actions, but
7408	 * maintain synchronization.  We must compensate.  Fortunately, we can
7409	 * detect that condition here, which happens early in CPU bringup,
7410	 * before any KVM threads can be running.  Unfortunately, we can't
7411	 * bring the TSCs fully up to date with real time, as we aren't yet far
7412	 * enough into CPU bringup that we know how much real time has actually
7413	 * elapsed; our helper function, get_kernel_ns() will be using boot
7414	 * variables that haven't been updated yet.
7415	 *
7416	 * So we simply find the maximum observed TSC above, then record the
7417	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7418	 * the adjustment will be applied.  Note that we accumulate
7419	 * adjustments, in case multiple suspend cycles happen before some VCPU
7420	 * gets a chance to run again.  In the event that no KVM threads get a
7421	 * chance to run, we will miss the entire elapsed period, as we'll have
7422	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7423	 * loose cycle time.  This isn't too big a deal, since the loss will be
7424	 * uniform across all VCPUs (not to mention the scenario is extremely
7425	 * unlikely). It is possible that a second hibernate recovery happens
7426	 * much faster than a first, causing the observed TSC here to be
7427	 * smaller; this would require additional padding adjustment, which is
7428	 * why we set last_host_tsc to the local tsc observed here.
7429	 *
7430	 * N.B. - this code below runs only on platforms with reliable TSC,
7431	 * as that is the only way backwards_tsc is set above.  Also note
7432	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7433	 * have the same delta_cyc adjustment applied if backwards_tsc
7434	 * is detected.  Note further, this adjustment is only done once,
7435	 * as we reset last_host_tsc on all VCPUs to stop this from being
7436	 * called multiple times (one for each physical CPU bringup).
7437	 *
7438	 * Platforms with unreliable TSCs don't have to deal with this, they
7439	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7440	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7441	 * guarantee that they stay in perfect synchronization.
7442	 */
7443	if (backwards_tsc) {
7444		u64 delta_cyc = max_tsc - local_tsc;
7445		backwards_tsc_observed = true;
7446		list_for_each_entry(kvm, &vm_list, vm_list) {
7447			kvm_for_each_vcpu(i, vcpu, kvm) {
7448				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7449				vcpu->arch.last_host_tsc = local_tsc;
7450				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7451			}
7452
7453			/*
7454			 * We have to disable TSC offset matching.. if you were
7455			 * booting a VM while issuing an S4 host suspend....
7456			 * you may have some problem.  Solving this issue is
7457			 * left as an exercise to the reader.
7458			 */
7459			kvm->arch.last_tsc_nsec = 0;
7460			kvm->arch.last_tsc_write = 0;
7461		}
7462
7463	}
7464	return 0;
7465}
7466
7467void kvm_arch_hardware_disable(void)
7468{
7469	kvm_x86_ops->hardware_disable();
7470	drop_user_return_notifiers();
7471}
7472
7473int kvm_arch_hardware_setup(void)
7474{
7475	int r;
7476
7477	r = kvm_x86_ops->hardware_setup();
7478	if (r != 0)
7479		return r;
7480
7481	if (kvm_has_tsc_control) {
7482		/*
7483		 * Make sure the user can only configure tsc_khz values that
7484		 * fit into a signed integer.
7485		 * A min value is not calculated needed because it will always
7486		 * be 1 on all machines.
7487		 */
7488		u64 max = min(0x7fffffffULL,
7489			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7490		kvm_max_guest_tsc_khz = max;
7491
7492		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7493	}
7494
7495	kvm_init_msr_list();
7496	return 0;
7497}
7498
7499void kvm_arch_hardware_unsetup(void)
7500{
7501	kvm_x86_ops->hardware_unsetup();
7502}
7503
7504void kvm_arch_check_processor_compat(void *rtn)
7505{
7506	kvm_x86_ops->check_processor_compatibility(rtn);
7507}
7508
7509bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7510{
7511	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7512}
7513EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7514
7515bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7516{
7517	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7518}
7519
7520bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7521{
7522	return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7523}
7524
7525struct static_key kvm_no_apic_vcpu __read_mostly;
7526
7527int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7528{
7529	struct page *page;
7530	struct kvm *kvm;
7531	int r;
7532
7533	BUG_ON(vcpu->kvm == NULL);
7534	kvm = vcpu->kvm;
7535
7536	vcpu->arch.pv.pv_unhalted = false;
7537	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7538	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7539		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7540	else
7541		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7542
7543	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7544	if (!page) {
7545		r = -ENOMEM;
7546		goto fail;
7547	}
7548	vcpu->arch.pio_data = page_address(page);
7549
7550	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7551
7552	r = kvm_mmu_create(vcpu);
7553	if (r < 0)
7554		goto fail_free_pio_data;
7555
7556	if (irqchip_in_kernel(kvm)) {
7557		r = kvm_create_lapic(vcpu);
7558		if (r < 0)
7559			goto fail_mmu_destroy;
7560	} else
7561		static_key_slow_inc(&kvm_no_apic_vcpu);
7562
7563	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7564				       GFP_KERNEL);
7565	if (!vcpu->arch.mce_banks) {
7566		r = -ENOMEM;
7567		goto fail_free_lapic;
7568	}
7569	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7570
7571	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7572		r = -ENOMEM;
7573		goto fail_free_mce_banks;
7574	}
7575
7576	fx_init(vcpu);
7577
7578	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7579	vcpu->arch.pv_time_enabled = false;
7580
7581	vcpu->arch.guest_supported_xcr0 = 0;
7582	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7583
7584	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7585
7586	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7587
7588	kvm_async_pf_hash_reset(vcpu);
7589	kvm_pmu_init(vcpu);
7590
7591	vcpu->arch.pending_external_vector = -1;
7592
7593	return 0;
7594
7595fail_free_mce_banks:
7596	kfree(vcpu->arch.mce_banks);
7597fail_free_lapic:
7598	kvm_free_lapic(vcpu);
7599fail_mmu_destroy:
7600	kvm_mmu_destroy(vcpu);
7601fail_free_pio_data:
7602	free_page((unsigned long)vcpu->arch.pio_data);
7603fail:
7604	return r;
7605}
7606
7607void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7608{
7609	int idx;
7610
7611	kvm_pmu_destroy(vcpu);
7612	kfree(vcpu->arch.mce_banks);
7613	kvm_free_lapic(vcpu);
7614	idx = srcu_read_lock(&vcpu->kvm->srcu);
7615	kvm_mmu_destroy(vcpu);
7616	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7617	free_page((unsigned long)vcpu->arch.pio_data);
7618	if (!lapic_in_kernel(vcpu))
7619		static_key_slow_dec(&kvm_no_apic_vcpu);
7620}
7621
7622void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7623{
7624	kvm_x86_ops->sched_in(vcpu, cpu);
7625}
7626
7627int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7628{
7629	if (type)
7630		return -EINVAL;
7631
7632	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7633	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7634	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7635	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7636	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7637
7638	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7639	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7640	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7641	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7642		&kvm->arch.irq_sources_bitmap);
7643
7644	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7645	mutex_init(&kvm->arch.apic_map_lock);
7646	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7647
7648	pvclock_update_vm_gtod_copy(kvm);
7649
7650	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7651	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7652
7653	return 0;
7654}
7655
7656static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7657{
7658	int r;
7659	r = vcpu_load(vcpu);
7660	BUG_ON(r);
7661	kvm_mmu_unload(vcpu);
7662	vcpu_put(vcpu);
7663}
7664
7665static void kvm_free_vcpus(struct kvm *kvm)
7666{
7667	unsigned int i;
7668	struct kvm_vcpu *vcpu;
7669
7670	/*
7671	 * Unpin any mmu pages first.
7672	 */
7673	kvm_for_each_vcpu(i, vcpu, kvm) {
7674		kvm_clear_async_pf_completion_queue(vcpu);
7675		kvm_unload_vcpu_mmu(vcpu);
7676	}
7677	kvm_for_each_vcpu(i, vcpu, kvm)
7678		kvm_arch_vcpu_free(vcpu);
7679
7680	mutex_lock(&kvm->lock);
7681	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7682		kvm->vcpus[i] = NULL;
7683
7684	atomic_set(&kvm->online_vcpus, 0);
7685	mutex_unlock(&kvm->lock);
7686}
7687
7688void kvm_arch_sync_events(struct kvm *kvm)
7689{
7690	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7691	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7692	kvm_free_all_assigned_devices(kvm);
7693	kvm_free_pit(kvm);
7694}
7695
7696int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7697{
7698	int i, r;
7699	unsigned long hva;
7700	struct kvm_memslots *slots = kvm_memslots(kvm);
7701	struct kvm_memory_slot *slot, old;
7702
7703	/* Called with kvm->slots_lock held.  */
7704	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7705		return -EINVAL;
7706
7707	slot = id_to_memslot(slots, id);
7708	if (size) {
7709		if (WARN_ON(slot->npages))
7710			return -EEXIST;
7711
7712		/*
7713		 * MAP_SHARED to prevent internal slot pages from being moved
7714		 * by fork()/COW.
7715		 */
7716		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7717			      MAP_SHARED | MAP_ANONYMOUS, 0);
7718		if (IS_ERR((void *)hva))
7719			return PTR_ERR((void *)hva);
7720	} else {
7721		if (!slot->npages)
7722			return 0;
7723
7724		hva = 0;
7725	}
7726
7727	old = *slot;
7728	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7729		struct kvm_userspace_memory_region m;
7730
7731		m.slot = id | (i << 16);
7732		m.flags = 0;
7733		m.guest_phys_addr = gpa;
7734		m.userspace_addr = hva;
7735		m.memory_size = size;
7736		r = __kvm_set_memory_region(kvm, &m);
7737		if (r < 0)
7738			return r;
7739	}
7740
7741	if (!size) {
7742		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7743		WARN_ON(r < 0);
7744	}
7745
7746	return 0;
7747}
7748EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7749
7750int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7751{
7752	int r;
7753
7754	mutex_lock(&kvm->slots_lock);
7755	r = __x86_set_memory_region(kvm, id, gpa, size);
7756	mutex_unlock(&kvm->slots_lock);
7757
7758	return r;
7759}
7760EXPORT_SYMBOL_GPL(x86_set_memory_region);
7761
7762void kvm_arch_destroy_vm(struct kvm *kvm)
7763{
7764	if (current->mm == kvm->mm) {
7765		/*
7766		 * Free memory regions allocated on behalf of userspace,
7767		 * unless the the memory map has changed due to process exit
7768		 * or fd copying.
7769		 */
7770		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7771		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7772		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7773	}
7774	kvm_iommu_unmap_guest(kvm);
7775	kfree(kvm->arch.vpic);
7776	kfree(kvm->arch.vioapic);
7777	kvm_free_vcpus(kvm);
7778	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7779}
7780
7781void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7782			   struct kvm_memory_slot *dont)
7783{
7784	int i;
7785
7786	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7787		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7788			kvfree(free->arch.rmap[i]);
7789			free->arch.rmap[i] = NULL;
7790		}
7791		if (i == 0)
7792			continue;
7793
7794		if (!dont || free->arch.lpage_info[i - 1] !=
7795			     dont->arch.lpage_info[i - 1]) {
7796			kvfree(free->arch.lpage_info[i - 1]);
7797			free->arch.lpage_info[i - 1] = NULL;
7798		}
7799	}
7800}
7801
7802int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7803			    unsigned long npages)
7804{
7805	int i;
7806
7807	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7808		unsigned long ugfn;
7809		int lpages;
7810		int level = i + 1;
7811
7812		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7813				      slot->base_gfn, level) + 1;
7814
7815		slot->arch.rmap[i] =
7816			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7817		if (!slot->arch.rmap[i])
7818			goto out_free;
7819		if (i == 0)
7820			continue;
7821
7822		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7823					sizeof(*slot->arch.lpage_info[i - 1]));
7824		if (!slot->arch.lpage_info[i - 1])
7825			goto out_free;
7826
7827		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7828			slot->arch.lpage_info[i - 1][0].write_count = 1;
7829		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7830			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7831		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7832		/*
7833		 * If the gfn and userspace address are not aligned wrt each
7834		 * other, or if explicitly asked to, disable large page
7835		 * support for this slot
7836		 */
7837		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7838		    !kvm_largepages_enabled()) {
7839			unsigned long j;
7840
7841			for (j = 0; j < lpages; ++j)
7842				slot->arch.lpage_info[i - 1][j].write_count = 1;
7843		}
7844	}
7845
7846	return 0;
7847
7848out_free:
7849	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7850		kvfree(slot->arch.rmap[i]);
7851		slot->arch.rmap[i] = NULL;
7852		if (i == 0)
7853			continue;
7854
7855		kvfree(slot->arch.lpage_info[i - 1]);
7856		slot->arch.lpage_info[i - 1] = NULL;
7857	}
7858	return -ENOMEM;
7859}
7860
7861void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7862{
7863	/*
7864	 * memslots->generation has been incremented.
7865	 * mmio generation may have reached its maximum value.
7866	 */
7867	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7868}
7869
7870int kvm_arch_prepare_memory_region(struct kvm *kvm,
7871				struct kvm_memory_slot *memslot,
7872				const struct kvm_userspace_memory_region *mem,
7873				enum kvm_mr_change change)
7874{
7875	return 0;
7876}
7877
7878static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7879				     struct kvm_memory_slot *new)
7880{
7881	/* Still write protect RO slot */
7882	if (new->flags & KVM_MEM_READONLY) {
7883		kvm_mmu_slot_remove_write_access(kvm, new);
7884		return;
7885	}
7886
7887	/*
7888	 * Call kvm_x86_ops dirty logging hooks when they are valid.
7889	 *
7890	 * kvm_x86_ops->slot_disable_log_dirty is called when:
7891	 *
7892	 *  - KVM_MR_CREATE with dirty logging is disabled
7893	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7894	 *
7895	 * The reason is, in case of PML, we need to set D-bit for any slots
7896	 * with dirty logging disabled in order to eliminate unnecessary GPA
7897	 * logging in PML buffer (and potential PML buffer full VMEXT). This
7898	 * guarantees leaving PML enabled during guest's lifetime won't have
7899	 * any additonal overhead from PML when guest is running with dirty
7900	 * logging disabled for memory slots.
7901	 *
7902	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7903	 * to dirty logging mode.
7904	 *
7905	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7906	 *
7907	 * In case of write protect:
7908	 *
7909	 * Write protect all pages for dirty logging.
7910	 *
7911	 * All the sptes including the large sptes which point to this
7912	 * slot are set to readonly. We can not create any new large
7913	 * spte on this slot until the end of the logging.
7914	 *
7915	 * See the comments in fast_page_fault().
7916	 */
7917	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7918		if (kvm_x86_ops->slot_enable_log_dirty)
7919			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7920		else
7921			kvm_mmu_slot_remove_write_access(kvm, new);
7922	} else {
7923		if (kvm_x86_ops->slot_disable_log_dirty)
7924			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7925	}
7926}
7927
7928void kvm_arch_commit_memory_region(struct kvm *kvm,
7929				const struct kvm_userspace_memory_region *mem,
7930				const struct kvm_memory_slot *old,
7931				const struct kvm_memory_slot *new,
7932				enum kvm_mr_change change)
7933{
7934	int nr_mmu_pages = 0;
7935
7936	if (!kvm->arch.n_requested_mmu_pages)
7937		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7938
7939	if (nr_mmu_pages)
7940		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7941
7942	/*
7943	 * Dirty logging tracks sptes in 4k granularity, meaning that large
7944	 * sptes have to be split.  If live migration is successful, the guest
7945	 * in the source machine will be destroyed and large sptes will be
7946	 * created in the destination. However, if the guest continues to run
7947	 * in the source machine (for example if live migration fails), small
7948	 * sptes will remain around and cause bad performance.
7949	 *
7950	 * Scan sptes if dirty logging has been stopped, dropping those
7951	 * which can be collapsed into a single large-page spte.  Later
7952	 * page faults will create the large-page sptes.
7953	 */
7954	if ((change != KVM_MR_DELETE) &&
7955		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7956		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7957		kvm_mmu_zap_collapsible_sptes(kvm, new);
7958
7959	/*
7960	 * Set up write protection and/or dirty logging for the new slot.
7961	 *
7962	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7963	 * been zapped so no dirty logging staff is needed for old slot. For
7964	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7965	 * new and it's also covered when dealing with the new slot.
7966	 *
7967	 * FIXME: const-ify all uses of struct kvm_memory_slot.
7968	 */
7969	if (change != KVM_MR_DELETE)
7970		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7971}
7972
7973void kvm_arch_flush_shadow_all(struct kvm *kvm)
7974{
7975	kvm_mmu_invalidate_zap_all_pages(kvm);
7976}
7977
7978void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7979				   struct kvm_memory_slot *slot)
7980{
7981	kvm_mmu_invalidate_zap_all_pages(kvm);
7982}
7983
7984static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
7985{
7986	if (!list_empty_careful(&vcpu->async_pf.done))
7987		return true;
7988
7989	if (kvm_apic_has_events(vcpu))
7990		return true;
7991
7992	if (vcpu->arch.pv.pv_unhalted)
7993		return true;
7994
7995	if (atomic_read(&vcpu->arch.nmi_queued))
7996		return true;
7997
7998	if (test_bit(KVM_REQ_SMI, &vcpu->requests))
7999		return true;
8000
8001	if (kvm_arch_interrupt_allowed(vcpu) &&
8002	    kvm_cpu_has_interrupt(vcpu))
8003		return true;
8004
8005	return false;
8006}
8007
8008int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8009{
8010	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8011		kvm_x86_ops->check_nested_events(vcpu, false);
8012
8013	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8014}
8015
8016int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8017{
8018	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8019}
8020
8021int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8022{
8023	return kvm_x86_ops->interrupt_allowed(vcpu);
8024}
8025
8026unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8027{
8028	if (is_64_bit_mode(vcpu))
8029		return kvm_rip_read(vcpu);
8030	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8031		     kvm_rip_read(vcpu));
8032}
8033EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8034
8035bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8036{
8037	return kvm_get_linear_rip(vcpu) == linear_rip;
8038}
8039EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8040
8041unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8042{
8043	unsigned long rflags;
8044
8045	rflags = kvm_x86_ops->get_rflags(vcpu);
8046	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8047		rflags &= ~X86_EFLAGS_TF;
8048	return rflags;
8049}
8050EXPORT_SYMBOL_GPL(kvm_get_rflags);
8051
8052static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8053{
8054	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8055	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8056		rflags |= X86_EFLAGS_TF;
8057	kvm_x86_ops->set_rflags(vcpu, rflags);
8058}
8059
8060void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8061{
8062	__kvm_set_rflags(vcpu, rflags);
8063	kvm_make_request(KVM_REQ_EVENT, vcpu);
8064}
8065EXPORT_SYMBOL_GPL(kvm_set_rflags);
8066
8067void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8068{
8069	int r;
8070
8071	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8072	      work->wakeup_all)
8073		return;
8074
8075	r = kvm_mmu_reload(vcpu);
8076	if (unlikely(r))
8077		return;
8078
8079	if (!vcpu->arch.mmu.direct_map &&
8080	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8081		return;
8082
8083	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8084}
8085
8086static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8087{
8088	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8089}
8090
8091static inline u32 kvm_async_pf_next_probe(u32 key)
8092{
8093	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8094}
8095
8096static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8097{
8098	u32 key = kvm_async_pf_hash_fn(gfn);
8099
8100	while (vcpu->arch.apf.gfns[key] != ~0)
8101		key = kvm_async_pf_next_probe(key);
8102
8103	vcpu->arch.apf.gfns[key] = gfn;
8104}
8105
8106static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8107{
8108	int i;
8109	u32 key = kvm_async_pf_hash_fn(gfn);
8110
8111	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8112		     (vcpu->arch.apf.gfns[key] != gfn &&
8113		      vcpu->arch.apf.gfns[key] != ~0); i++)
8114		key = kvm_async_pf_next_probe(key);
8115
8116	return key;
8117}
8118
8119bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8120{
8121	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8122}
8123
8124static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8125{
8126	u32 i, j, k;
8127
8128	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8129	while (true) {
8130		vcpu->arch.apf.gfns[i] = ~0;
8131		do {
8132			j = kvm_async_pf_next_probe(j);
8133			if (vcpu->arch.apf.gfns[j] == ~0)
8134				return;
8135			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8136			/*
8137			 * k lies cyclically in ]i,j]
8138			 * |    i.k.j |
8139			 * |....j i.k.| or  |.k..j i...|
8140			 */
8141		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8142		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8143		i = j;
8144	}
8145}
8146
8147static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8148{
8149
8150	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8151				      sizeof(val));
8152}
8153
8154void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8155				     struct kvm_async_pf *work)
8156{
8157	struct x86_exception fault;
8158
8159	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8160	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8161
8162	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8163	    (vcpu->arch.apf.send_user_only &&
8164	     kvm_x86_ops->get_cpl(vcpu) == 0))
8165		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8166	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8167		fault.vector = PF_VECTOR;
8168		fault.error_code_valid = true;
8169		fault.error_code = 0;
8170		fault.nested_page_fault = false;
8171		fault.address = work->arch.token;
8172		kvm_inject_page_fault(vcpu, &fault);
8173	}
8174}
8175
8176void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8177				 struct kvm_async_pf *work)
8178{
8179	struct x86_exception fault;
8180
8181	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8182	if (work->wakeup_all)
8183		work->arch.token = ~0; /* broadcast wakeup */
8184	else
8185		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8186
8187	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8188	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8189		fault.vector = PF_VECTOR;
8190		fault.error_code_valid = true;
8191		fault.error_code = 0;
8192		fault.nested_page_fault = false;
8193		fault.address = work->arch.token;
8194		kvm_inject_page_fault(vcpu, &fault);
8195	}
8196	vcpu->arch.apf.halted = false;
8197	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8198}
8199
8200bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8201{
8202	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8203		return true;
8204	else
8205		return !kvm_event_needs_reinjection(vcpu) &&
8206			kvm_x86_ops->interrupt_allowed(vcpu);
8207}
8208
8209void kvm_arch_start_assignment(struct kvm *kvm)
8210{
8211	atomic_inc(&kvm->arch.assigned_device_count);
8212}
8213EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8214
8215void kvm_arch_end_assignment(struct kvm *kvm)
8216{
8217	atomic_dec(&kvm->arch.assigned_device_count);
8218}
8219EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8220
8221bool kvm_arch_has_assigned_device(struct kvm *kvm)
8222{
8223	return atomic_read(&kvm->arch.assigned_device_count);
8224}
8225EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8226
8227void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8228{
8229	atomic_inc(&kvm->arch.noncoherent_dma_count);
8230}
8231EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8232
8233void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8234{
8235	atomic_dec(&kvm->arch.noncoherent_dma_count);
8236}
8237EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8238
8239bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8240{
8241	return atomic_read(&kvm->arch.noncoherent_dma_count);
8242}
8243EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8244
8245int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8246				      struct irq_bypass_producer *prod)
8247{
8248	struct kvm_kernel_irqfd *irqfd =
8249		container_of(cons, struct kvm_kernel_irqfd, consumer);
8250
8251	if (kvm_x86_ops->update_pi_irte) {
8252		irqfd->producer = prod;
8253		return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8254				prod->irq, irqfd->gsi, 1);
8255	}
8256
8257	return -EINVAL;
8258}
8259
8260void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8261				      struct irq_bypass_producer *prod)
8262{
8263	int ret;
8264	struct kvm_kernel_irqfd *irqfd =
8265		container_of(cons, struct kvm_kernel_irqfd, consumer);
8266
8267	if (!kvm_x86_ops->update_pi_irte) {
8268		WARN_ON(irqfd->producer != NULL);
8269		return;
8270	}
8271
8272	WARN_ON(irqfd->producer != prod);
8273	irqfd->producer = NULL;
8274
8275	/*
8276	 * When producer of consumer is unregistered, we change back to
8277	 * remapped mode, so we can re-use the current implementation
8278	 * when the irq is masked/disabed or the consumer side (KVM
8279	 * int this case doesn't want to receive the interrupts.
8280	*/
8281	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8282	if (ret)
8283		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8284		       " fails: %d\n", irqfd->consumer.token, ret);
8285}
8286
8287int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8288				   uint32_t guest_irq, bool set)
8289{
8290	if (!kvm_x86_ops->update_pi_irte)
8291		return -EINVAL;
8292
8293	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8294}
8295
8296EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8297EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8298EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8299EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8300EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8301EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8302EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8303EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8304EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8305EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8306EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8307EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8308EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8309EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8310EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8311EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8312EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8313