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
32#include <linux/clocksource.h>
33#include <linux/interrupt.h>
34#include <linux/kvm.h>
35#include <linux/fs.h>
36#include <linux/vmalloc.h>
37#include <linux/module.h>
38#include <linux/mman.h>
39#include <linux/highmem.h>
40#include <linux/iommu.h>
41#include <linux/intel-iommu.h>
42#include <linux/cpufreq.h>
43#include <linux/user-return-notifier.h>
44#include <linux/srcu.h>
45#include <linux/slab.h>
46#include <linux/perf_event.h>
47#include <linux/uaccess.h>
48#include <linux/hash.h>
49#include <linux/pci.h>
50#include <linux/timekeeper_internal.h>
51#include <linux/pvclock_gtod.h>
52#include <trace/events/kvm.h>
53
54#define CREATE_TRACE_POINTS
55#include "trace.h"
56
57#include <asm/debugreg.h>
58#include <asm/msr.h>
59#include <asm/desc.h>
60#include <asm/mtrr.h>
61#include <asm/mce.h>
62#include <asm/i387.h>
63#include <asm/fpu-internal.h> /* Ugh! */
64#include <asm/xcr.h>
65#include <asm/pvclock.h>
66#include <asm/div64.h>
67
68#define MAX_IO_MSRS 256
69#define KVM_MAX_MCE_BANKS 32
70#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71
72#define emul_to_vcpu(ctxt) \
73	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
74
75/* EFER defaults:
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
78 */
79#ifdef CONFIG_X86_64
80static
81u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82#else
83static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
84#endif
85
86#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88
89static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90static void process_nmi(struct kvm_vcpu *vcpu);
91static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
92
93struct kvm_x86_ops *kvm_x86_ops;
94EXPORT_SYMBOL_GPL(kvm_x86_ops);
95
96static bool ignore_msrs = 0;
97module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
98
99unsigned int min_timer_period_us = 500;
100module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
101
102bool kvm_has_tsc_control;
103EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
104u32  kvm_max_guest_tsc_khz;
105EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
106
107/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
108static u32 tsc_tolerance_ppm = 250;
109module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
110
111/* lapic timer advance (tscdeadline mode only) in nanoseconds */
112unsigned int lapic_timer_advance_ns = 0;
113module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
114
115static bool backwards_tsc_observed = false;
116
117#define KVM_NR_SHARED_MSRS 16
118
119struct kvm_shared_msrs_global {
120	int nr;
121	u32 msrs[KVM_NR_SHARED_MSRS];
122};
123
124struct kvm_shared_msrs {
125	struct user_return_notifier urn;
126	bool registered;
127	struct kvm_shared_msr_values {
128		u64 host;
129		u64 curr;
130	} values[KVM_NR_SHARED_MSRS];
131};
132
133static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
134static struct kvm_shared_msrs __percpu *shared_msrs;
135
136struct kvm_stats_debugfs_item debugfs_entries[] = {
137	{ "pf_fixed", VCPU_STAT(pf_fixed) },
138	{ "pf_guest", VCPU_STAT(pf_guest) },
139	{ "tlb_flush", VCPU_STAT(tlb_flush) },
140	{ "invlpg", VCPU_STAT(invlpg) },
141	{ "exits", VCPU_STAT(exits) },
142	{ "io_exits", VCPU_STAT(io_exits) },
143	{ "mmio_exits", VCPU_STAT(mmio_exits) },
144	{ "signal_exits", VCPU_STAT(signal_exits) },
145	{ "irq_window", VCPU_STAT(irq_window_exits) },
146	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
147	{ "halt_exits", VCPU_STAT(halt_exits) },
148	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
149	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
150	{ "hypercalls", VCPU_STAT(hypercalls) },
151	{ "request_irq", VCPU_STAT(request_irq_exits) },
152	{ "irq_exits", VCPU_STAT(irq_exits) },
153	{ "host_state_reload", VCPU_STAT(host_state_reload) },
154	{ "efer_reload", VCPU_STAT(efer_reload) },
155	{ "fpu_reload", VCPU_STAT(fpu_reload) },
156	{ "insn_emulation", VCPU_STAT(insn_emulation) },
157	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
158	{ "irq_injections", VCPU_STAT(irq_injections) },
159	{ "nmi_injections", VCPU_STAT(nmi_injections) },
160	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
161	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
162	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
163	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
164	{ "mmu_flooded", VM_STAT(mmu_flooded) },
165	{ "mmu_recycled", VM_STAT(mmu_recycled) },
166	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
167	{ "mmu_unsync", VM_STAT(mmu_unsync) },
168	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
169	{ "largepages", VM_STAT(lpages) },
170	{ NULL }
171};
172
173u64 __read_mostly host_xcr0;
174
175static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
176
177static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
178{
179	int i;
180	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
181		vcpu->arch.apf.gfns[i] = ~0;
182}
183
184static void kvm_on_user_return(struct user_return_notifier *urn)
185{
186	unsigned slot;
187	struct kvm_shared_msrs *locals
188		= container_of(urn, struct kvm_shared_msrs, urn);
189	struct kvm_shared_msr_values *values;
190
191	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
192		values = &locals->values[slot];
193		if (values->host != values->curr) {
194			wrmsrl(shared_msrs_global.msrs[slot], values->host);
195			values->curr = values->host;
196		}
197	}
198	locals->registered = false;
199	user_return_notifier_unregister(urn);
200}
201
202static void shared_msr_update(unsigned slot, u32 msr)
203{
204	u64 value;
205	unsigned int cpu = smp_processor_id();
206	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
207
208	/* only read, and nobody should modify it at this time,
209	 * so don't need lock */
210	if (slot >= shared_msrs_global.nr) {
211		printk(KERN_ERR "kvm: invalid MSR slot!");
212		return;
213	}
214	rdmsrl_safe(msr, &value);
215	smsr->values[slot].host = value;
216	smsr->values[slot].curr = value;
217}
218
219void kvm_define_shared_msr(unsigned slot, u32 msr)
220{
221	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
222	if (slot >= shared_msrs_global.nr)
223		shared_msrs_global.nr = slot + 1;
224	shared_msrs_global.msrs[slot] = msr;
225	/* we need ensured the shared_msr_global have been updated */
226	smp_wmb();
227}
228EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
229
230static void kvm_shared_msr_cpu_online(void)
231{
232	unsigned i;
233
234	for (i = 0; i < shared_msrs_global.nr; ++i)
235		shared_msr_update(i, shared_msrs_global.msrs[i]);
236}
237
238int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
239{
240	unsigned int cpu = smp_processor_id();
241	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
242	int err;
243
244	if (((value ^ smsr->values[slot].curr) & mask) == 0)
245		return 0;
246	smsr->values[slot].curr = value;
247	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
248	if (err)
249		return 1;
250
251	if (!smsr->registered) {
252		smsr->urn.on_user_return = kvm_on_user_return;
253		user_return_notifier_register(&smsr->urn);
254		smsr->registered = true;
255	}
256	return 0;
257}
258EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
259
260static void drop_user_return_notifiers(void)
261{
262	unsigned int cpu = smp_processor_id();
263	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
264
265	if (smsr->registered)
266		kvm_on_user_return(&smsr->urn);
267}
268
269u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
270{
271	return vcpu->arch.apic_base;
272}
273EXPORT_SYMBOL_GPL(kvm_get_apic_base);
274
275int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
276{
277	u64 old_state = vcpu->arch.apic_base &
278		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
279	u64 new_state = msr_info->data &
280		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
282		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
283
284	if (!msr_info->host_initiated &&
285	    ((msr_info->data & reserved_bits) != 0 ||
286	     new_state == X2APIC_ENABLE ||
287	     (new_state == MSR_IA32_APICBASE_ENABLE &&
288	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
289	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
290	      old_state == 0)))
291		return 1;
292
293	kvm_lapic_set_base(vcpu, msr_info->data);
294	return 0;
295}
296EXPORT_SYMBOL_GPL(kvm_set_apic_base);
297
298asmlinkage __visible void kvm_spurious_fault(void)
299{
300	/* Fault while not rebooting.  We want the trace. */
301	BUG();
302}
303EXPORT_SYMBOL_GPL(kvm_spurious_fault);
304
305#define EXCPT_BENIGN		0
306#define EXCPT_CONTRIBUTORY	1
307#define EXCPT_PF		2
308
309static int exception_class(int vector)
310{
311	switch (vector) {
312	case PF_VECTOR:
313		return EXCPT_PF;
314	case DE_VECTOR:
315	case TS_VECTOR:
316	case NP_VECTOR:
317	case SS_VECTOR:
318	case GP_VECTOR:
319		return EXCPT_CONTRIBUTORY;
320	default:
321		break;
322	}
323	return EXCPT_BENIGN;
324}
325
326#define EXCPT_FAULT		0
327#define EXCPT_TRAP		1
328#define EXCPT_ABORT		2
329#define EXCPT_INTERRUPT		3
330
331static int exception_type(int vector)
332{
333	unsigned int mask;
334
335	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
336		return EXCPT_INTERRUPT;
337
338	mask = 1 << vector;
339
340	/* #DB is trap, as instruction watchpoints are handled elsewhere */
341	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
342		return EXCPT_TRAP;
343
344	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
345		return EXCPT_ABORT;
346
347	/* Reserved exceptions will result in fault */
348	return EXCPT_FAULT;
349}
350
351static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
352		unsigned nr, bool has_error, u32 error_code,
353		bool reinject)
354{
355	u32 prev_nr;
356	int class1, class2;
357
358	kvm_make_request(KVM_REQ_EVENT, vcpu);
359
360	if (!vcpu->arch.exception.pending) {
361	queue:
362		if (has_error && !is_protmode(vcpu))
363			has_error = false;
364		vcpu->arch.exception.pending = true;
365		vcpu->arch.exception.has_error_code = has_error;
366		vcpu->arch.exception.nr = nr;
367		vcpu->arch.exception.error_code = error_code;
368		vcpu->arch.exception.reinject = reinject;
369		return;
370	}
371
372	/* to check exception */
373	prev_nr = vcpu->arch.exception.nr;
374	if (prev_nr == DF_VECTOR) {
375		/* triple fault -> shutdown */
376		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
377		return;
378	}
379	class1 = exception_class(prev_nr);
380	class2 = exception_class(nr);
381	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
382		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
383		/* generate double fault per SDM Table 5-5 */
384		vcpu->arch.exception.pending = true;
385		vcpu->arch.exception.has_error_code = true;
386		vcpu->arch.exception.nr = DF_VECTOR;
387		vcpu->arch.exception.error_code = 0;
388	} else
389		/* replace previous exception with a new one in a hope
390		   that instruction re-execution will regenerate lost
391		   exception */
392		goto queue;
393}
394
395void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
396{
397	kvm_multiple_exception(vcpu, nr, false, 0, false);
398}
399EXPORT_SYMBOL_GPL(kvm_queue_exception);
400
401void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
402{
403	kvm_multiple_exception(vcpu, nr, false, 0, true);
404}
405EXPORT_SYMBOL_GPL(kvm_requeue_exception);
406
407void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
408{
409	if (err)
410		kvm_inject_gp(vcpu, 0);
411	else
412		kvm_x86_ops->skip_emulated_instruction(vcpu);
413}
414EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
415
416void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
417{
418	++vcpu->stat.pf_guest;
419	vcpu->arch.cr2 = fault->address;
420	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
421}
422EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
423
424static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
425{
426	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
427		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
428	else
429		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
430
431	return fault->nested_page_fault;
432}
433
434void kvm_inject_nmi(struct kvm_vcpu *vcpu)
435{
436	atomic_inc(&vcpu->arch.nmi_queued);
437	kvm_make_request(KVM_REQ_NMI, vcpu);
438}
439EXPORT_SYMBOL_GPL(kvm_inject_nmi);
440
441void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
442{
443	kvm_multiple_exception(vcpu, nr, true, error_code, false);
444}
445EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
446
447void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
448{
449	kvm_multiple_exception(vcpu, nr, true, error_code, true);
450}
451EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
452
453/*
454 * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
455 * a #GP and return false.
456 */
457bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
458{
459	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
460		return true;
461	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
462	return false;
463}
464EXPORT_SYMBOL_GPL(kvm_require_cpl);
465
466bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
467{
468	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
469		return true;
470
471	kvm_queue_exception(vcpu, UD_VECTOR);
472	return false;
473}
474EXPORT_SYMBOL_GPL(kvm_require_dr);
475
476/*
477 * This function will be used to read from the physical memory of the currently
478 * running guest. The difference to kvm_read_guest_page is that this function
479 * can read from guest physical or from the guest's guest physical memory.
480 */
481int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
482			    gfn_t ngfn, void *data, int offset, int len,
483			    u32 access)
484{
485	struct x86_exception exception;
486	gfn_t real_gfn;
487	gpa_t ngpa;
488
489	ngpa     = gfn_to_gpa(ngfn);
490	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
491	if (real_gfn == UNMAPPED_GVA)
492		return -EFAULT;
493
494	real_gfn = gpa_to_gfn(real_gfn);
495
496	return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
497}
498EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
499
500static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
501			       void *data, int offset, int len, u32 access)
502{
503	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
504				       data, offset, len, access);
505}
506
507/*
508 * Load the pae pdptrs.  Return true is they are all valid.
509 */
510int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
511{
512	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
513	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
514	int i;
515	int ret;
516	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
517
518	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
519				      offset * sizeof(u64), sizeof(pdpte),
520				      PFERR_USER_MASK|PFERR_WRITE_MASK);
521	if (ret < 0) {
522		ret = 0;
523		goto out;
524	}
525	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
526		if (is_present_gpte(pdpte[i]) &&
527		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
528			ret = 0;
529			goto out;
530		}
531	}
532	ret = 1;
533
534	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
535	__set_bit(VCPU_EXREG_PDPTR,
536		  (unsigned long *)&vcpu->arch.regs_avail);
537	__set_bit(VCPU_EXREG_PDPTR,
538		  (unsigned long *)&vcpu->arch.regs_dirty);
539out:
540
541	return ret;
542}
543EXPORT_SYMBOL_GPL(load_pdptrs);
544
545static bool pdptrs_changed(struct kvm_vcpu *vcpu)
546{
547	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
548	bool changed = true;
549	int offset;
550	gfn_t gfn;
551	int r;
552
553	if (is_long_mode(vcpu) || !is_pae(vcpu))
554		return false;
555
556	if (!test_bit(VCPU_EXREG_PDPTR,
557		      (unsigned long *)&vcpu->arch.regs_avail))
558		return true;
559
560	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
561	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
562	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
563				       PFERR_USER_MASK | PFERR_WRITE_MASK);
564	if (r < 0)
565		goto out;
566	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
567out:
568
569	return changed;
570}
571
572int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
573{
574	unsigned long old_cr0 = kvm_read_cr0(vcpu);
575	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
576				    X86_CR0_CD | X86_CR0_NW;
577
578	cr0 |= X86_CR0_ET;
579
580#ifdef CONFIG_X86_64
581	if (cr0 & 0xffffffff00000000UL)
582		return 1;
583#endif
584
585	cr0 &= ~CR0_RESERVED_BITS;
586
587	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
588		return 1;
589
590	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
591		return 1;
592
593	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
594#ifdef CONFIG_X86_64
595		if ((vcpu->arch.efer & EFER_LME)) {
596			int cs_db, cs_l;
597
598			if (!is_pae(vcpu))
599				return 1;
600			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
601			if (cs_l)
602				return 1;
603		} else
604#endif
605		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
606						 kvm_read_cr3(vcpu)))
607			return 1;
608	}
609
610	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
611		return 1;
612
613	kvm_x86_ops->set_cr0(vcpu, cr0);
614
615	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
616		kvm_clear_async_pf_completion_queue(vcpu);
617		kvm_async_pf_hash_reset(vcpu);
618	}
619
620	if ((cr0 ^ old_cr0) & update_bits)
621		kvm_mmu_reset_context(vcpu);
622	return 0;
623}
624EXPORT_SYMBOL_GPL(kvm_set_cr0);
625
626void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
627{
628	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
629}
630EXPORT_SYMBOL_GPL(kvm_lmsw);
631
632static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
633{
634	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
635			!vcpu->guest_xcr0_loaded) {
636		/* kvm_set_xcr() also depends on this */
637		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
638		vcpu->guest_xcr0_loaded = 1;
639	}
640}
641
642static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
643{
644	if (vcpu->guest_xcr0_loaded) {
645		if (vcpu->arch.xcr0 != host_xcr0)
646			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
647		vcpu->guest_xcr0_loaded = 0;
648	}
649}
650
651static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
652{
653	u64 xcr0 = xcr;
654	u64 old_xcr0 = vcpu->arch.xcr0;
655	u64 valid_bits;
656
657	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
658	if (index != XCR_XFEATURE_ENABLED_MASK)
659		return 1;
660	if (!(xcr0 & XSTATE_FP))
661		return 1;
662	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
663		return 1;
664
665	/*
666	 * Do not allow the guest to set bits that we do not support
667	 * saving.  However, xcr0 bit 0 is always set, even if the
668	 * emulated CPU does not support XSAVE (see fx_init).
669	 */
670	valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
671	if (xcr0 & ~valid_bits)
672		return 1;
673
674	if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
675		return 1;
676
677	if (xcr0 & XSTATE_AVX512) {
678		if (!(xcr0 & XSTATE_YMM))
679			return 1;
680		if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
681			return 1;
682	}
683	vcpu->arch.xcr0 = xcr0;
684
685	if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
686		kvm_update_cpuid(vcpu);
687	return 0;
688}
689
690int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
691{
692	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
693	    __kvm_set_xcr(vcpu, index, xcr)) {
694		kvm_inject_gp(vcpu, 0);
695		return 1;
696	}
697	return 0;
698}
699EXPORT_SYMBOL_GPL(kvm_set_xcr);
700
701int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
702{
703	unsigned long old_cr4 = kvm_read_cr4(vcpu);
704	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
705				   X86_CR4_SMEP | X86_CR4_SMAP;
706
707	if (cr4 & CR4_RESERVED_BITS)
708		return 1;
709
710	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
711		return 1;
712
713	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
714		return 1;
715
716	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
717		return 1;
718
719	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
720		return 1;
721
722	if (is_long_mode(vcpu)) {
723		if (!(cr4 & X86_CR4_PAE))
724			return 1;
725	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
726		   && ((cr4 ^ old_cr4) & pdptr_bits)
727		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
728				   kvm_read_cr3(vcpu)))
729		return 1;
730
731	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
732		if (!guest_cpuid_has_pcid(vcpu))
733			return 1;
734
735		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
736		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
737			return 1;
738	}
739
740	if (kvm_x86_ops->set_cr4(vcpu, cr4))
741		return 1;
742
743	if (((cr4 ^ old_cr4) & pdptr_bits) ||
744	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
745		kvm_mmu_reset_context(vcpu);
746
747	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
748		kvm_update_cpuid(vcpu);
749
750	return 0;
751}
752EXPORT_SYMBOL_GPL(kvm_set_cr4);
753
754int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
755{
756#ifdef CONFIG_X86_64
757	cr3 &= ~CR3_PCID_INVD;
758#endif
759
760	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
761		kvm_mmu_sync_roots(vcpu);
762		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
763		return 0;
764	}
765
766	if (is_long_mode(vcpu)) {
767		if (cr3 & CR3_L_MODE_RESERVED_BITS)
768			return 1;
769	} else if (is_pae(vcpu) && is_paging(vcpu) &&
770		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
771		return 1;
772
773	vcpu->arch.cr3 = cr3;
774	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
775	kvm_mmu_new_cr3(vcpu);
776	return 0;
777}
778EXPORT_SYMBOL_GPL(kvm_set_cr3);
779
780int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
781{
782	if (cr8 & CR8_RESERVED_BITS)
783		return 1;
784	if (irqchip_in_kernel(vcpu->kvm))
785		kvm_lapic_set_tpr(vcpu, cr8);
786	else
787		vcpu->arch.cr8 = cr8;
788	return 0;
789}
790EXPORT_SYMBOL_GPL(kvm_set_cr8);
791
792unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
793{
794	if (irqchip_in_kernel(vcpu->kvm))
795		return kvm_lapic_get_cr8(vcpu);
796	else
797		return vcpu->arch.cr8;
798}
799EXPORT_SYMBOL_GPL(kvm_get_cr8);
800
801static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
802{
803	int i;
804
805	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
806		for (i = 0; i < KVM_NR_DB_REGS; i++)
807			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
808		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
809	}
810}
811
812static void kvm_update_dr6(struct kvm_vcpu *vcpu)
813{
814	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
815		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
816}
817
818static void kvm_update_dr7(struct kvm_vcpu *vcpu)
819{
820	unsigned long dr7;
821
822	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
823		dr7 = vcpu->arch.guest_debug_dr7;
824	else
825		dr7 = vcpu->arch.dr7;
826	kvm_x86_ops->set_dr7(vcpu, dr7);
827	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
828	if (dr7 & DR7_BP_EN_MASK)
829		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
830}
831
832static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
833{
834	u64 fixed = DR6_FIXED_1;
835
836	if (!guest_cpuid_has_rtm(vcpu))
837		fixed |= DR6_RTM;
838	return fixed;
839}
840
841static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
842{
843	switch (dr) {
844	case 0 ... 3:
845		vcpu->arch.db[dr] = val;
846		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
847			vcpu->arch.eff_db[dr] = val;
848		break;
849	case 4:
850		/* fall through */
851	case 6:
852		if (val & 0xffffffff00000000ULL)
853			return -1; /* #GP */
854		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
855		kvm_update_dr6(vcpu);
856		break;
857	case 5:
858		/* fall through */
859	default: /* 7 */
860		if (val & 0xffffffff00000000ULL)
861			return -1; /* #GP */
862		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
863		kvm_update_dr7(vcpu);
864		break;
865	}
866
867	return 0;
868}
869
870int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
871{
872	if (__kvm_set_dr(vcpu, dr, val)) {
873		kvm_inject_gp(vcpu, 0);
874		return 1;
875	}
876	return 0;
877}
878EXPORT_SYMBOL_GPL(kvm_set_dr);
879
880int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
881{
882	switch (dr) {
883	case 0 ... 3:
884		*val = vcpu->arch.db[dr];
885		break;
886	case 4:
887		/* fall through */
888	case 6:
889		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
890			*val = vcpu->arch.dr6;
891		else
892			*val = kvm_x86_ops->get_dr6(vcpu);
893		break;
894	case 5:
895		/* fall through */
896	default: /* 7 */
897		*val = vcpu->arch.dr7;
898		break;
899	}
900	return 0;
901}
902EXPORT_SYMBOL_GPL(kvm_get_dr);
903
904bool kvm_rdpmc(struct kvm_vcpu *vcpu)
905{
906	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
907	u64 data;
908	int err;
909
910	err = kvm_pmu_read_pmc(vcpu, ecx, &data);
911	if (err)
912		return err;
913	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
914	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
915	return err;
916}
917EXPORT_SYMBOL_GPL(kvm_rdpmc);
918
919/*
920 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
921 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
922 *
923 * This list is modified at module load time to reflect the
924 * capabilities of the host cpu. This capabilities test skips MSRs that are
925 * kvm-specific. Those are put in the beginning of the list.
926 */
927
928#define KVM_SAVE_MSRS_BEGIN	12
929static u32 msrs_to_save[] = {
930	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
931	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
932	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
933	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
934	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
935	MSR_KVM_PV_EOI_EN,
936	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
937	MSR_STAR,
938#ifdef CONFIG_X86_64
939	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
940#endif
941	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
942	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
943};
944
945static unsigned num_msrs_to_save;
946
947static const u32 emulated_msrs[] = {
948	MSR_IA32_TSC_ADJUST,
949	MSR_IA32_TSCDEADLINE,
950	MSR_IA32_MISC_ENABLE,
951	MSR_IA32_MCG_STATUS,
952	MSR_IA32_MCG_CTL,
953};
954
955bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
956{
957	if (efer & efer_reserved_bits)
958		return false;
959
960	if (efer & EFER_FFXSR) {
961		struct kvm_cpuid_entry2 *feat;
962
963		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
964		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
965			return false;
966	}
967
968	if (efer & EFER_SVME) {
969		struct kvm_cpuid_entry2 *feat;
970
971		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
972		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
973			return false;
974	}
975
976	return true;
977}
978EXPORT_SYMBOL_GPL(kvm_valid_efer);
979
980static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
981{
982	u64 old_efer = vcpu->arch.efer;
983
984	if (!kvm_valid_efer(vcpu, efer))
985		return 1;
986
987	if (is_paging(vcpu)
988	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
989		return 1;
990
991	efer &= ~EFER_LMA;
992	efer |= vcpu->arch.efer & EFER_LMA;
993
994	kvm_x86_ops->set_efer(vcpu, efer);
995
996	/* Update reserved bits */
997	if ((efer ^ old_efer) & EFER_NX)
998		kvm_mmu_reset_context(vcpu);
999
1000	return 0;
1001}
1002
1003void kvm_enable_efer_bits(u64 mask)
1004{
1005       efer_reserved_bits &= ~mask;
1006}
1007EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1008
1009/*
1010 * Writes msr value into into the appropriate "register".
1011 * Returns 0 on success, non-0 otherwise.
1012 * Assumes vcpu_load() was already called.
1013 */
1014int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1015{
1016	switch (msr->index) {
1017	case MSR_FS_BASE:
1018	case MSR_GS_BASE:
1019	case MSR_KERNEL_GS_BASE:
1020	case MSR_CSTAR:
1021	case MSR_LSTAR:
1022		if (is_noncanonical_address(msr->data))
1023			return 1;
1024		break;
1025	case MSR_IA32_SYSENTER_EIP:
1026	case MSR_IA32_SYSENTER_ESP:
1027		/*
1028		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1029		 * non-canonical address is written on Intel but not on
1030		 * AMD (which ignores the top 32-bits, because it does
1031		 * not implement 64-bit SYSENTER).
1032		 *
1033		 * 64-bit code should hence be able to write a non-canonical
1034		 * value on AMD.  Making the address canonical ensures that
1035		 * vmentry does not fail on Intel after writing a non-canonical
1036		 * value, and that something deterministic happens if the guest
1037		 * invokes 64-bit SYSENTER.
1038		 */
1039		msr->data = get_canonical(msr->data);
1040	}
1041	return kvm_x86_ops->set_msr(vcpu, msr);
1042}
1043EXPORT_SYMBOL_GPL(kvm_set_msr);
1044
1045/*
1046 * Adapt set_msr() to msr_io()'s calling convention
1047 */
1048static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1049{
1050	struct msr_data msr;
1051
1052	msr.data = *data;
1053	msr.index = index;
1054	msr.host_initiated = true;
1055	return kvm_set_msr(vcpu, &msr);
1056}
1057
1058#ifdef CONFIG_X86_64
1059struct pvclock_gtod_data {
1060	seqcount_t	seq;
1061
1062	struct { /* extract of a clocksource struct */
1063		int vclock_mode;
1064		cycle_t	cycle_last;
1065		cycle_t	mask;
1066		u32	mult;
1067		u32	shift;
1068	} clock;
1069
1070	u64		boot_ns;
1071	u64		nsec_base;
1072};
1073
1074static struct pvclock_gtod_data pvclock_gtod_data;
1075
1076static void update_pvclock_gtod(struct timekeeper *tk)
1077{
1078	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1079	u64 boot_ns;
1080
1081	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1082
1083	write_seqcount_begin(&vdata->seq);
1084
1085	/* copy pvclock gtod data */
1086	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1087	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1088	vdata->clock.mask		= tk->tkr_mono.mask;
1089	vdata->clock.mult		= tk->tkr_mono.mult;
1090	vdata->clock.shift		= tk->tkr_mono.shift;
1091
1092	vdata->boot_ns			= boot_ns;
1093	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1094
1095	write_seqcount_end(&vdata->seq);
1096}
1097#endif
1098
1099void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1100{
1101	/*
1102	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1103	 * vcpu_enter_guest.  This function is only called from
1104	 * the physical CPU that is running vcpu.
1105	 */
1106	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1107}
1108
1109static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1110{
1111	int version;
1112	int r;
1113	struct pvclock_wall_clock wc;
1114	struct timespec boot;
1115
1116	if (!wall_clock)
1117		return;
1118
1119	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1120	if (r)
1121		return;
1122
1123	if (version & 1)
1124		++version;  /* first time write, random junk */
1125
1126	++version;
1127
1128	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1129
1130	/*
1131	 * The guest calculates current wall clock time by adding
1132	 * system time (updated by kvm_guest_time_update below) to the
1133	 * wall clock specified here.  guest system time equals host
1134	 * system time for us, thus we must fill in host boot time here.
1135	 */
1136	getboottime(&boot);
1137
1138	if (kvm->arch.kvmclock_offset) {
1139		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1140		boot = timespec_sub(boot, ts);
1141	}
1142	wc.sec = boot.tv_sec;
1143	wc.nsec = boot.tv_nsec;
1144	wc.version = version;
1145
1146	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1147
1148	version++;
1149	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1150}
1151
1152static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1153{
1154	uint32_t quotient, remainder;
1155
1156	/* Don't try to replace with do_div(), this one calculates
1157	 * "(dividend << 32) / divisor" */
1158	__asm__ ( "divl %4"
1159		  : "=a" (quotient), "=d" (remainder)
1160		  : "0" (0), "1" (dividend), "r" (divisor) );
1161	return quotient;
1162}
1163
1164static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1165			       s8 *pshift, u32 *pmultiplier)
1166{
1167	uint64_t scaled64;
1168	int32_t  shift = 0;
1169	uint64_t tps64;
1170	uint32_t tps32;
1171
1172	tps64 = base_khz * 1000LL;
1173	scaled64 = scaled_khz * 1000LL;
1174	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1175		tps64 >>= 1;
1176		shift--;
1177	}
1178
1179	tps32 = (uint32_t)tps64;
1180	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1181		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1182			scaled64 >>= 1;
1183		else
1184			tps32 <<= 1;
1185		shift++;
1186	}
1187
1188	*pshift = shift;
1189	*pmultiplier = div_frac(scaled64, tps32);
1190
1191	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1192		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1193}
1194
1195static inline u64 get_kernel_ns(void)
1196{
1197	return ktime_get_boot_ns();
1198}
1199
1200#ifdef CONFIG_X86_64
1201static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1202#endif
1203
1204static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1205static unsigned long max_tsc_khz;
1206
1207static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1208{
1209	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1210				   vcpu->arch.virtual_tsc_shift);
1211}
1212
1213static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1214{
1215	u64 v = (u64)khz * (1000000 + ppm);
1216	do_div(v, 1000000);
1217	return v;
1218}
1219
1220static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1221{
1222	u32 thresh_lo, thresh_hi;
1223	int use_scaling = 0;
1224
1225	/* tsc_khz can be zero if TSC calibration fails */
1226	if (this_tsc_khz == 0)
1227		return;
1228
1229	/* Compute a scale to convert nanoseconds in TSC cycles */
1230	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1231			   &vcpu->arch.virtual_tsc_shift,
1232			   &vcpu->arch.virtual_tsc_mult);
1233	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1234
1235	/*
1236	 * Compute the variation in TSC rate which is acceptable
1237	 * within the range of tolerance and decide if the
1238	 * rate being applied is within that bounds of the hardware
1239	 * rate.  If so, no scaling or compensation need be done.
1240	 */
1241	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1242	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1243	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1244		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1245		use_scaling = 1;
1246	}
1247	kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1248}
1249
1250static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1251{
1252	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1253				      vcpu->arch.virtual_tsc_mult,
1254				      vcpu->arch.virtual_tsc_shift);
1255	tsc += vcpu->arch.this_tsc_write;
1256	return tsc;
1257}
1258
1259static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1260{
1261#ifdef CONFIG_X86_64
1262	bool vcpus_matched;
1263	struct kvm_arch *ka = &vcpu->kvm->arch;
1264	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1265
1266	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1267			 atomic_read(&vcpu->kvm->online_vcpus));
1268
1269	/*
1270	 * Once the masterclock is enabled, always perform request in
1271	 * order to update it.
1272	 *
1273	 * In order to enable masterclock, the host clocksource must be TSC
1274	 * and the vcpus need to have matched TSCs.  When that happens,
1275	 * perform request to enable masterclock.
1276	 */
1277	if (ka->use_master_clock ||
1278	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1279		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1280
1281	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1282			    atomic_read(&vcpu->kvm->online_vcpus),
1283		            ka->use_master_clock, gtod->clock.vclock_mode);
1284#endif
1285}
1286
1287static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1288{
1289	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1290	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1291}
1292
1293void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1294{
1295	struct kvm *kvm = vcpu->kvm;
1296	u64 offset, ns, elapsed;
1297	unsigned long flags;
1298	s64 usdiff;
1299	bool matched;
1300	bool already_matched;
1301	u64 data = msr->data;
1302
1303	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1304	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1305	ns = get_kernel_ns();
1306	elapsed = ns - kvm->arch.last_tsc_nsec;
1307
1308	if (vcpu->arch.virtual_tsc_khz) {
1309		int faulted = 0;
1310
1311		/* n.b - signed multiplication and division required */
1312		usdiff = data - kvm->arch.last_tsc_write;
1313#ifdef CONFIG_X86_64
1314		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1315#else
1316		/* do_div() only does unsigned */
1317		asm("1: idivl %[divisor]\n"
1318		    "2: xor %%edx, %%edx\n"
1319		    "   movl $0, %[faulted]\n"
1320		    "3:\n"
1321		    ".section .fixup,\"ax\"\n"
1322		    "4: movl $1, %[faulted]\n"
1323		    "   jmp  3b\n"
1324		    ".previous\n"
1325
1326		_ASM_EXTABLE(1b, 4b)
1327
1328		: "=A"(usdiff), [faulted] "=r" (faulted)
1329		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1330
1331#endif
1332		do_div(elapsed, 1000);
1333		usdiff -= elapsed;
1334		if (usdiff < 0)
1335			usdiff = -usdiff;
1336
1337		/* idivl overflow => difference is larger than USEC_PER_SEC */
1338		if (faulted)
1339			usdiff = USEC_PER_SEC;
1340	} else
1341		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1342
1343	/*
1344	 * Special case: TSC write with a small delta (1 second) of virtual
1345	 * cycle time against real time is interpreted as an attempt to
1346	 * synchronize the CPU.
1347         *
1348	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1349	 * TSC, we add elapsed time in this computation.  We could let the
1350	 * compensation code attempt to catch up if we fall behind, but
1351	 * it's better to try to match offsets from the beginning.
1352         */
1353	if (usdiff < USEC_PER_SEC &&
1354	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1355		if (!check_tsc_unstable()) {
1356			offset = kvm->arch.cur_tsc_offset;
1357			pr_debug("kvm: matched tsc offset for %llu\n", data);
1358		} else {
1359			u64 delta = nsec_to_cycles(vcpu, elapsed);
1360			data += delta;
1361			offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1362			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1363		}
1364		matched = true;
1365		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1366	} else {
1367		/*
1368		 * We split periods of matched TSC writes into generations.
1369		 * For each generation, we track the original measured
1370		 * nanosecond time, offset, and write, so if TSCs are in
1371		 * sync, we can match exact offset, and if not, we can match
1372		 * exact software computation in compute_guest_tsc()
1373		 *
1374		 * These values are tracked in kvm->arch.cur_xxx variables.
1375		 */
1376		kvm->arch.cur_tsc_generation++;
1377		kvm->arch.cur_tsc_nsec = ns;
1378		kvm->arch.cur_tsc_write = data;
1379		kvm->arch.cur_tsc_offset = offset;
1380		matched = false;
1381		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1382			 kvm->arch.cur_tsc_generation, data);
1383	}
1384
1385	/*
1386	 * We also track th most recent recorded KHZ, write and time to
1387	 * allow the matching interval to be extended at each write.
1388	 */
1389	kvm->arch.last_tsc_nsec = ns;
1390	kvm->arch.last_tsc_write = data;
1391	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1392
1393	vcpu->arch.last_guest_tsc = data;
1394
1395	/* Keep track of which generation this VCPU has synchronized to */
1396	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1397	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1398	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1399
1400	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1401		update_ia32_tsc_adjust_msr(vcpu, offset);
1402	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1403	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1404
1405	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1406	if (!matched) {
1407		kvm->arch.nr_vcpus_matched_tsc = 0;
1408	} else if (!already_matched) {
1409		kvm->arch.nr_vcpus_matched_tsc++;
1410	}
1411
1412	kvm_track_tsc_matching(vcpu);
1413	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1414}
1415
1416EXPORT_SYMBOL_GPL(kvm_write_tsc);
1417
1418#ifdef CONFIG_X86_64
1419
1420static cycle_t read_tsc(void)
1421{
1422	cycle_t ret;
1423	u64 last;
1424
1425	/*
1426	 * Empirically, a fence (of type that depends on the CPU)
1427	 * before rdtsc is enough to ensure that rdtsc is ordered
1428	 * with respect to loads.  The various CPU manuals are unclear
1429	 * as to whether rdtsc can be reordered with later loads,
1430	 * but no one has ever seen it happen.
1431	 */
1432	rdtsc_barrier();
1433	ret = (cycle_t)vget_cycles();
1434
1435	last = pvclock_gtod_data.clock.cycle_last;
1436
1437	if (likely(ret >= last))
1438		return ret;
1439
1440	/*
1441	 * GCC likes to generate cmov here, but this branch is extremely
1442	 * predictable (it's just a funciton of time and the likely is
1443	 * very likely) and there's a data dependence, so force GCC
1444	 * to generate a branch instead.  I don't barrier() because
1445	 * we don't actually need a barrier, and if this function
1446	 * ever gets inlined it will generate worse code.
1447	 */
1448	asm volatile ("");
1449	return last;
1450}
1451
1452static inline u64 vgettsc(cycle_t *cycle_now)
1453{
1454	long v;
1455	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1456
1457	*cycle_now = read_tsc();
1458
1459	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1460	return v * gtod->clock.mult;
1461}
1462
1463static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1464{
1465	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1466	unsigned long seq;
1467	int mode;
1468	u64 ns;
1469
1470	do {
1471		seq = read_seqcount_begin(&gtod->seq);
1472		mode = gtod->clock.vclock_mode;
1473		ns = gtod->nsec_base;
1474		ns += vgettsc(cycle_now);
1475		ns >>= gtod->clock.shift;
1476		ns += gtod->boot_ns;
1477	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1478	*t = ns;
1479
1480	return mode;
1481}
1482
1483/* returns true if host is using tsc clocksource */
1484static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1485{
1486	/* checked again under seqlock below */
1487	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1488		return false;
1489
1490	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1491}
1492#endif
1493
1494/*
1495 *
1496 * Assuming a stable TSC across physical CPUS, and a stable TSC
1497 * across virtual CPUs, the following condition is possible.
1498 * Each numbered line represents an event visible to both
1499 * CPUs at the next numbered event.
1500 *
1501 * "timespecX" represents host monotonic time. "tscX" represents
1502 * RDTSC value.
1503 *
1504 * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1505 *
1506 * 1.  read timespec0,tsc0
1507 * 2.					| timespec1 = timespec0 + N
1508 * 					| tsc1 = tsc0 + M
1509 * 3. transition to guest		| transition to guest
1510 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1511 * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1512 * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1513 *
1514 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1515 *
1516 * 	- ret0 < ret1
1517 *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1518 *		...
1519 *	- 0 < N - M => M < N
1520 *
1521 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1522 * always the case (the difference between two distinct xtime instances
1523 * might be smaller then the difference between corresponding TSC reads,
1524 * when updating guest vcpus pvclock areas).
1525 *
1526 * To avoid that problem, do not allow visibility of distinct
1527 * system_timestamp/tsc_timestamp values simultaneously: use a master
1528 * copy of host monotonic time values. Update that master copy
1529 * in lockstep.
1530 *
1531 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1532 *
1533 */
1534
1535static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1536{
1537#ifdef CONFIG_X86_64
1538	struct kvm_arch *ka = &kvm->arch;
1539	int vclock_mode;
1540	bool host_tsc_clocksource, vcpus_matched;
1541
1542	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1543			atomic_read(&kvm->online_vcpus));
1544
1545	/*
1546	 * If the host uses TSC clock, then passthrough TSC as stable
1547	 * to the guest.
1548	 */
1549	host_tsc_clocksource = kvm_get_time_and_clockread(
1550					&ka->master_kernel_ns,
1551					&ka->master_cycle_now);
1552
1553	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1554				&& !backwards_tsc_observed
1555				&& !ka->boot_vcpu_runs_old_kvmclock;
1556
1557	if (ka->use_master_clock)
1558		atomic_set(&kvm_guest_has_master_clock, 1);
1559
1560	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1561	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1562					vcpus_matched);
1563#endif
1564}
1565
1566static void kvm_gen_update_masterclock(struct kvm *kvm)
1567{
1568#ifdef CONFIG_X86_64
1569	int i;
1570	struct kvm_vcpu *vcpu;
1571	struct kvm_arch *ka = &kvm->arch;
1572
1573	spin_lock(&ka->pvclock_gtod_sync_lock);
1574	kvm_make_mclock_inprogress_request(kvm);
1575	/* no guest entries from this point */
1576	pvclock_update_vm_gtod_copy(kvm);
1577
1578	kvm_for_each_vcpu(i, vcpu, kvm)
1579		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1580
1581	/* guest entries allowed */
1582	kvm_for_each_vcpu(i, vcpu, kvm)
1583		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1584
1585	spin_unlock(&ka->pvclock_gtod_sync_lock);
1586#endif
1587}
1588
1589static int kvm_guest_time_update(struct kvm_vcpu *v)
1590{
1591	unsigned long flags, this_tsc_khz;
1592	struct kvm_vcpu_arch *vcpu = &v->arch;
1593	struct kvm_arch *ka = &v->kvm->arch;
1594	s64 kernel_ns;
1595	u64 tsc_timestamp, host_tsc;
1596	struct pvclock_vcpu_time_info guest_hv_clock;
1597	u8 pvclock_flags;
1598	bool use_master_clock;
1599
1600	kernel_ns = 0;
1601	host_tsc = 0;
1602
1603	/*
1604	 * If the host uses TSC clock, then passthrough TSC as stable
1605	 * to the guest.
1606	 */
1607	spin_lock(&ka->pvclock_gtod_sync_lock);
1608	use_master_clock = ka->use_master_clock;
1609	if (use_master_clock) {
1610		host_tsc = ka->master_cycle_now;
1611		kernel_ns = ka->master_kernel_ns;
1612	}
1613	spin_unlock(&ka->pvclock_gtod_sync_lock);
1614
1615	/* Keep irq disabled to prevent changes to the clock */
1616	local_irq_save(flags);
1617	this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1618	if (unlikely(this_tsc_khz == 0)) {
1619		local_irq_restore(flags);
1620		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1621		return 1;
1622	}
1623	if (!use_master_clock) {
1624		host_tsc = native_read_tsc();
1625		kernel_ns = get_kernel_ns();
1626	}
1627
1628	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1629
1630	/*
1631	 * We may have to catch up the TSC to match elapsed wall clock
1632	 * time for two reasons, even if kvmclock is used.
1633	 *   1) CPU could have been running below the maximum TSC rate
1634	 *   2) Broken TSC compensation resets the base at each VCPU
1635	 *      entry to avoid unknown leaps of TSC even when running
1636	 *      again on the same CPU.  This may cause apparent elapsed
1637	 *      time to disappear, and the guest to stand still or run
1638	 *	very slowly.
1639	 */
1640	if (vcpu->tsc_catchup) {
1641		u64 tsc = compute_guest_tsc(v, kernel_ns);
1642		if (tsc > tsc_timestamp) {
1643			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1644			tsc_timestamp = tsc;
1645		}
1646	}
1647
1648	local_irq_restore(flags);
1649
1650	if (!vcpu->pv_time_enabled)
1651		return 0;
1652
1653	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1654		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1655				   &vcpu->hv_clock.tsc_shift,
1656				   &vcpu->hv_clock.tsc_to_system_mul);
1657		vcpu->hw_tsc_khz = this_tsc_khz;
1658	}
1659
1660	/* With all the info we got, fill in the values */
1661	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1662	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1663	vcpu->last_guest_tsc = tsc_timestamp;
1664
1665	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1666		&guest_hv_clock, sizeof(guest_hv_clock))))
1667		return 0;
1668
1669	/* This VCPU is paused, but it's legal for a guest to read another
1670	 * VCPU's kvmclock, so we really have to follow the specification where
1671	 * it says that version is odd if data is being modified, and even after
1672	 * it is consistent.
1673	 *
1674	 * Version field updates must be kept separate.  This is because
1675	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1676	 * writes within a string instruction are weakly ordered.  So there
1677	 * are three writes overall.
1678	 *
1679	 * As a small optimization, only write the version field in the first
1680	 * and third write.  The vcpu->pv_time cache is still valid, because the
1681	 * version field is the first in the struct.
1682	 */
1683	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1684
1685	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1686	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1687				&vcpu->hv_clock,
1688				sizeof(vcpu->hv_clock.version));
1689
1690	smp_wmb();
1691
1692	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1693	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1694
1695	if (vcpu->pvclock_set_guest_stopped_request) {
1696		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1697		vcpu->pvclock_set_guest_stopped_request = false;
1698	}
1699
1700	/* If the host uses TSC clocksource, then it is stable */
1701	if (use_master_clock)
1702		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1703
1704	vcpu->hv_clock.flags = pvclock_flags;
1705
1706	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1707
1708	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1709				&vcpu->hv_clock,
1710				sizeof(vcpu->hv_clock));
1711
1712	smp_wmb();
1713
1714	vcpu->hv_clock.version++;
1715	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1716				&vcpu->hv_clock,
1717				sizeof(vcpu->hv_clock.version));
1718	return 0;
1719}
1720
1721/*
1722 * kvmclock updates which are isolated to a given vcpu, such as
1723 * vcpu->cpu migration, should not allow system_timestamp from
1724 * the rest of the vcpus to remain static. Otherwise ntp frequency
1725 * correction applies to one vcpu's system_timestamp but not
1726 * the others.
1727 *
1728 * So in those cases, request a kvmclock update for all vcpus.
1729 * We need to rate-limit these requests though, as they can
1730 * considerably slow guests that have a large number of vcpus.
1731 * The time for a remote vcpu to update its kvmclock is bound
1732 * by the delay we use to rate-limit the updates.
1733 */
1734
1735#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1736
1737static void kvmclock_update_fn(struct work_struct *work)
1738{
1739	int i;
1740	struct delayed_work *dwork = to_delayed_work(work);
1741	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1742					   kvmclock_update_work);
1743	struct kvm *kvm = container_of(ka, struct kvm, arch);
1744	struct kvm_vcpu *vcpu;
1745
1746	kvm_for_each_vcpu(i, vcpu, kvm) {
1747		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1748		kvm_vcpu_kick(vcpu);
1749	}
1750}
1751
1752static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1753{
1754	struct kvm *kvm = v->kvm;
1755
1756	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1757	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1758					KVMCLOCK_UPDATE_DELAY);
1759}
1760
1761#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1762
1763static void kvmclock_sync_fn(struct work_struct *work)
1764{
1765	struct delayed_work *dwork = to_delayed_work(work);
1766	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1767					   kvmclock_sync_work);
1768	struct kvm *kvm = container_of(ka, struct kvm, arch);
1769
1770	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1771	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1772					KVMCLOCK_SYNC_PERIOD);
1773}
1774
1775static bool msr_mtrr_valid(unsigned msr)
1776{
1777	switch (msr) {
1778	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1779	case MSR_MTRRfix64K_00000:
1780	case MSR_MTRRfix16K_80000:
1781	case MSR_MTRRfix16K_A0000:
1782	case MSR_MTRRfix4K_C0000:
1783	case MSR_MTRRfix4K_C8000:
1784	case MSR_MTRRfix4K_D0000:
1785	case MSR_MTRRfix4K_D8000:
1786	case MSR_MTRRfix4K_E0000:
1787	case MSR_MTRRfix4K_E8000:
1788	case MSR_MTRRfix4K_F0000:
1789	case MSR_MTRRfix4K_F8000:
1790	case MSR_MTRRdefType:
1791	case MSR_IA32_CR_PAT:
1792		return true;
1793	case 0x2f8:
1794		return true;
1795	}
1796	return false;
1797}
1798
1799static bool valid_pat_type(unsigned t)
1800{
1801	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1802}
1803
1804static bool valid_mtrr_type(unsigned t)
1805{
1806	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1807}
1808
1809bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1810{
1811	int i;
1812	u64 mask;
1813
1814	if (!msr_mtrr_valid(msr))
1815		return false;
1816
1817	if (msr == MSR_IA32_CR_PAT) {
1818		for (i = 0; i < 8; i++)
1819			if (!valid_pat_type((data >> (i * 8)) & 0xff))
1820				return false;
1821		return true;
1822	} else if (msr == MSR_MTRRdefType) {
1823		if (data & ~0xcff)
1824			return false;
1825		return valid_mtrr_type(data & 0xff);
1826	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1827		for (i = 0; i < 8 ; i++)
1828			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1829				return false;
1830		return true;
1831	}
1832
1833	/* variable MTRRs */
1834	WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1835
1836	mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1837	if ((msr & 1) == 0) {
1838		/* MTRR base */
1839		if (!valid_mtrr_type(data & 0xff))
1840			return false;
1841		mask |= 0xf00;
1842	} else
1843		/* MTRR mask */
1844		mask |= 0x7ff;
1845	if (data & mask) {
1846		kvm_inject_gp(vcpu, 0);
1847		return false;
1848	}
1849
1850	return true;
1851}
1852EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1853
1854static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1855{
1856	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1857
1858	if (!kvm_mtrr_valid(vcpu, msr, data))
1859		return 1;
1860
1861	if (msr == MSR_MTRRdefType) {
1862		vcpu->arch.mtrr_state.def_type = data;
1863		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1864	} else if (msr == MSR_MTRRfix64K_00000)
1865		p[0] = data;
1866	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1867		p[1 + msr - MSR_MTRRfix16K_80000] = data;
1868	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1869		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1870	else if (msr == MSR_IA32_CR_PAT)
1871		vcpu->arch.pat = data;
1872	else {	/* Variable MTRRs */
1873		int idx, is_mtrr_mask;
1874		u64 *pt;
1875
1876		idx = (msr - 0x200) / 2;
1877		is_mtrr_mask = msr - 0x200 - 2 * idx;
1878		if (!is_mtrr_mask)
1879			pt =
1880			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1881		else
1882			pt =
1883			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1884		*pt = data;
1885	}
1886
1887	kvm_mmu_reset_context(vcpu);
1888	return 0;
1889}
1890
1891static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1892{
1893	u64 mcg_cap = vcpu->arch.mcg_cap;
1894	unsigned bank_num = mcg_cap & 0xff;
1895
1896	switch (msr) {
1897	case MSR_IA32_MCG_STATUS:
1898		vcpu->arch.mcg_status = data;
1899		break;
1900	case MSR_IA32_MCG_CTL:
1901		if (!(mcg_cap & MCG_CTL_P))
1902			return 1;
1903		if (data != 0 && data != ~(u64)0)
1904			return -1;
1905		vcpu->arch.mcg_ctl = data;
1906		break;
1907	default:
1908		if (msr >= MSR_IA32_MC0_CTL &&
1909		    msr < MSR_IA32_MCx_CTL(bank_num)) {
1910			u32 offset = msr - MSR_IA32_MC0_CTL;
1911			/* only 0 or all 1s can be written to IA32_MCi_CTL
1912			 * some Linux kernels though clear bit 10 in bank 4 to
1913			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1914			 * this to avoid an uncatched #GP in the guest
1915			 */
1916			if ((offset & 0x3) == 0 &&
1917			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1918				return -1;
1919			vcpu->arch.mce_banks[offset] = data;
1920			break;
1921		}
1922		return 1;
1923	}
1924	return 0;
1925}
1926
1927static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1928{
1929	struct kvm *kvm = vcpu->kvm;
1930	int lm = is_long_mode(vcpu);
1931	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1932		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1933	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1934		: kvm->arch.xen_hvm_config.blob_size_32;
1935	u32 page_num = data & ~PAGE_MASK;
1936	u64 page_addr = data & PAGE_MASK;
1937	u8 *page;
1938	int r;
1939
1940	r = -E2BIG;
1941	if (page_num >= blob_size)
1942		goto out;
1943	r = -ENOMEM;
1944	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1945	if (IS_ERR(page)) {
1946		r = PTR_ERR(page);
1947		goto out;
1948	}
1949	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1950		goto out_free;
1951	r = 0;
1952out_free:
1953	kfree(page);
1954out:
1955	return r;
1956}
1957
1958static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1959{
1960	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1961}
1962
1963static bool kvm_hv_msr_partition_wide(u32 msr)
1964{
1965	bool r = false;
1966	switch (msr) {
1967	case HV_X64_MSR_GUEST_OS_ID:
1968	case HV_X64_MSR_HYPERCALL:
1969	case HV_X64_MSR_REFERENCE_TSC:
1970	case HV_X64_MSR_TIME_REF_COUNT:
1971		r = true;
1972		break;
1973	}
1974
1975	return r;
1976}
1977
1978static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1979{
1980	struct kvm *kvm = vcpu->kvm;
1981
1982	switch (msr) {
1983	case HV_X64_MSR_GUEST_OS_ID:
1984		kvm->arch.hv_guest_os_id = data;
1985		/* setting guest os id to zero disables hypercall page */
1986		if (!kvm->arch.hv_guest_os_id)
1987			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1988		break;
1989	case HV_X64_MSR_HYPERCALL: {
1990		u64 gfn;
1991		unsigned long addr;
1992		u8 instructions[4];
1993
1994		/* if guest os id is not set hypercall should remain disabled */
1995		if (!kvm->arch.hv_guest_os_id)
1996			break;
1997		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1998			kvm->arch.hv_hypercall = data;
1999			break;
2000		}
2001		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
2002		addr = gfn_to_hva(kvm, gfn);
2003		if (kvm_is_error_hva(addr))
2004			return 1;
2005		kvm_x86_ops->patch_hypercall(vcpu, instructions);
2006		((unsigned char *)instructions)[3] = 0xc3; /* ret */
2007		if (__copy_to_user((void __user *)addr, instructions, 4))
2008			return 1;
2009		kvm->arch.hv_hypercall = data;
2010		mark_page_dirty(kvm, gfn);
2011		break;
2012	}
2013	case HV_X64_MSR_REFERENCE_TSC: {
2014		u64 gfn;
2015		HV_REFERENCE_TSC_PAGE tsc_ref;
2016		memset(&tsc_ref, 0, sizeof(tsc_ref));
2017		kvm->arch.hv_tsc_page = data;
2018		if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
2019			break;
2020		gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
2021		if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
2022			&tsc_ref, sizeof(tsc_ref)))
2023			return 1;
2024		mark_page_dirty(kvm, gfn);
2025		break;
2026	}
2027	default:
2028		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2029			    "data 0x%llx\n", msr, data);
2030		return 1;
2031	}
2032	return 0;
2033}
2034
2035static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2036{
2037	switch (msr) {
2038	case HV_X64_MSR_APIC_ASSIST_PAGE: {
2039		u64 gfn;
2040		unsigned long addr;
2041
2042		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2043			vcpu->arch.hv_vapic = data;
2044			if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2045				return 1;
2046			break;
2047		}
2048		gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2049		addr = gfn_to_hva(vcpu->kvm, gfn);
2050		if (kvm_is_error_hva(addr))
2051			return 1;
2052		if (__clear_user((void __user *)addr, PAGE_SIZE))
2053			return 1;
2054		vcpu->arch.hv_vapic = data;
2055		mark_page_dirty(vcpu->kvm, gfn);
2056		if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2057			return 1;
2058		break;
2059	}
2060	case HV_X64_MSR_EOI:
2061		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2062	case HV_X64_MSR_ICR:
2063		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2064	case HV_X64_MSR_TPR:
2065		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2066	default:
2067		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2068			    "data 0x%llx\n", msr, data);
2069		return 1;
2070	}
2071
2072	return 0;
2073}
2074
2075static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2076{
2077	gpa_t gpa = data & ~0x3f;
2078
2079	/* Bits 2:5 are reserved, Should be zero */
2080	if (data & 0x3c)
2081		return 1;
2082
2083	vcpu->arch.apf.msr_val = data;
2084
2085	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2086		kvm_clear_async_pf_completion_queue(vcpu);
2087		kvm_async_pf_hash_reset(vcpu);
2088		return 0;
2089	}
2090
2091	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2092					sizeof(u32)))
2093		return 1;
2094
2095	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2096	kvm_async_pf_wakeup_all(vcpu);
2097	return 0;
2098}
2099
2100static void kvmclock_reset(struct kvm_vcpu *vcpu)
2101{
2102	vcpu->arch.pv_time_enabled = false;
2103}
2104
2105static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2106{
2107	u64 delta;
2108
2109	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2110		return;
2111
2112	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2113	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2114	vcpu->arch.st.accum_steal = delta;
2115}
2116
2117static void record_steal_time(struct kvm_vcpu *vcpu)
2118{
2119	accumulate_steal_time(vcpu);
2120
2121	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2122		return;
2123
2124	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2125		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2126		return;
2127
2128	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2129	vcpu->arch.st.steal.version += 2;
2130	vcpu->arch.st.accum_steal = 0;
2131
2132	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2133		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2134}
2135
2136int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2137{
2138	bool pr = false;
2139	u32 msr = msr_info->index;
2140	u64 data = msr_info->data;
2141
2142	switch (msr) {
2143	case MSR_AMD64_NB_CFG:
2144	case MSR_IA32_UCODE_REV:
2145	case MSR_IA32_UCODE_WRITE:
2146	case MSR_VM_HSAVE_PA:
2147	case MSR_AMD64_PATCH_LOADER:
2148	case MSR_AMD64_BU_CFG2:
2149		break;
2150
2151	case MSR_EFER:
2152		return set_efer(vcpu, data);
2153	case MSR_K7_HWCR:
2154		data &= ~(u64)0x40;	/* ignore flush filter disable */
2155		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2156		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2157		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2158		if (data != 0) {
2159			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2160				    data);
2161			return 1;
2162		}
2163		break;
2164	case MSR_FAM10H_MMIO_CONF_BASE:
2165		if (data != 0) {
2166			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2167				    "0x%llx\n", data);
2168			return 1;
2169		}
2170		break;
2171	case MSR_IA32_DEBUGCTLMSR:
2172		if (!data) {
2173			/* We support the non-activated case already */
2174			break;
2175		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2176			/* Values other than LBR and BTF are vendor-specific,
2177			   thus reserved and should throw a #GP */
2178			return 1;
2179		}
2180		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2181			    __func__, data);
2182		break;
2183	case 0x200 ... 0x2ff:
2184		return set_msr_mtrr(vcpu, msr, data);
2185	case MSR_IA32_APICBASE:
2186		return kvm_set_apic_base(vcpu, msr_info);
2187	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2188		return kvm_x2apic_msr_write(vcpu, msr, data);
2189	case MSR_IA32_TSCDEADLINE:
2190		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2191		break;
2192	case MSR_IA32_TSC_ADJUST:
2193		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2194			if (!msr_info->host_initiated) {
2195				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2196				adjust_tsc_offset_guest(vcpu, adj);
2197			}
2198			vcpu->arch.ia32_tsc_adjust_msr = data;
2199		}
2200		break;
2201	case MSR_IA32_MISC_ENABLE:
2202		vcpu->arch.ia32_misc_enable_msr = data;
2203		break;
2204	case MSR_KVM_WALL_CLOCK_NEW:
2205	case MSR_KVM_WALL_CLOCK:
2206		vcpu->kvm->arch.wall_clock = data;
2207		kvm_write_wall_clock(vcpu->kvm, data);
2208		break;
2209	case MSR_KVM_SYSTEM_TIME_NEW:
2210	case MSR_KVM_SYSTEM_TIME: {
2211		u64 gpa_offset;
2212		struct kvm_arch *ka = &vcpu->kvm->arch;
2213
2214		kvmclock_reset(vcpu);
2215
2216		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2217			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2218
2219			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2220				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2221					&vcpu->requests);
2222
2223			ka->boot_vcpu_runs_old_kvmclock = tmp;
2224		}
2225
2226		vcpu->arch.time = data;
2227		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2228
2229		/* we verify if the enable bit is set... */
2230		if (!(data & 1))
2231			break;
2232
2233		gpa_offset = data & ~(PAGE_MASK | 1);
2234
2235		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2236		     &vcpu->arch.pv_time, data & ~1ULL,
2237		     sizeof(struct pvclock_vcpu_time_info)))
2238			vcpu->arch.pv_time_enabled = false;
2239		else
2240			vcpu->arch.pv_time_enabled = true;
2241
2242		break;
2243	}
2244	case MSR_KVM_ASYNC_PF_EN:
2245		if (kvm_pv_enable_async_pf(vcpu, data))
2246			return 1;
2247		break;
2248	case MSR_KVM_STEAL_TIME:
2249
2250		if (unlikely(!sched_info_on()))
2251			return 1;
2252
2253		if (data & KVM_STEAL_RESERVED_MASK)
2254			return 1;
2255
2256		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2257						data & KVM_STEAL_VALID_BITS,
2258						sizeof(struct kvm_steal_time)))
2259			return 1;
2260
2261		vcpu->arch.st.msr_val = data;
2262
2263		if (!(data & KVM_MSR_ENABLED))
2264			break;
2265
2266		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2267
2268		break;
2269	case MSR_KVM_PV_EOI_EN:
2270		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2271			return 1;
2272		break;
2273
2274	case MSR_IA32_MCG_CTL:
2275	case MSR_IA32_MCG_STATUS:
2276	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2277		return set_msr_mce(vcpu, msr, data);
2278
2279	/* Performance counters are not protected by a CPUID bit,
2280	 * so we should check all of them in the generic path for the sake of
2281	 * cross vendor migration.
2282	 * Writing a zero into the event select MSRs disables them,
2283	 * which we perfectly emulate ;-). Any other value should be at least
2284	 * reported, some guests depend on them.
2285	 */
2286	case MSR_K7_EVNTSEL0:
2287	case MSR_K7_EVNTSEL1:
2288	case MSR_K7_EVNTSEL2:
2289	case MSR_K7_EVNTSEL3:
2290		if (data != 0)
2291			vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2292				    "0x%x data 0x%llx\n", msr, data);
2293		break;
2294	/* at least RHEL 4 unconditionally writes to the perfctr registers,
2295	 * so we ignore writes to make it happy.
2296	 */
2297	case MSR_K7_PERFCTR0:
2298	case MSR_K7_PERFCTR1:
2299	case MSR_K7_PERFCTR2:
2300	case MSR_K7_PERFCTR3:
2301		vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2302			    "0x%x data 0x%llx\n", msr, data);
2303		break;
2304	case MSR_P6_PERFCTR0:
2305	case MSR_P6_PERFCTR1:
2306		pr = true;
2307	case MSR_P6_EVNTSEL0:
2308	case MSR_P6_EVNTSEL1:
2309		if (kvm_pmu_msr(vcpu, msr))
2310			return kvm_pmu_set_msr(vcpu, msr_info);
2311
2312		if (pr || data != 0)
2313			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2314				    "0x%x data 0x%llx\n", msr, data);
2315		break;
2316	case MSR_K7_CLK_CTL:
2317		/*
2318		 * Ignore all writes to this no longer documented MSR.
2319		 * Writes are only relevant for old K7 processors,
2320		 * all pre-dating SVM, but a recommended workaround from
2321		 * AMD for these chips. It is possible to specify the
2322		 * affected processor models on the command line, hence
2323		 * the need to ignore the workaround.
2324		 */
2325		break;
2326	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2327		if (kvm_hv_msr_partition_wide(msr)) {
2328			int r;
2329			mutex_lock(&vcpu->kvm->lock);
2330			r = set_msr_hyperv_pw(vcpu, msr, data);
2331			mutex_unlock(&vcpu->kvm->lock);
2332			return r;
2333		} else
2334			return set_msr_hyperv(vcpu, msr, data);
2335		break;
2336	case MSR_IA32_BBL_CR_CTL3:
2337		/* Drop writes to this legacy MSR -- see rdmsr
2338		 * counterpart for further detail.
2339		 */
2340		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2341		break;
2342	case MSR_AMD64_OSVW_ID_LENGTH:
2343		if (!guest_cpuid_has_osvw(vcpu))
2344			return 1;
2345		vcpu->arch.osvw.length = data;
2346		break;
2347	case MSR_AMD64_OSVW_STATUS:
2348		if (!guest_cpuid_has_osvw(vcpu))
2349			return 1;
2350		vcpu->arch.osvw.status = data;
2351		break;
2352	default:
2353		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2354			return xen_hvm_config(vcpu, data);
2355		if (kvm_pmu_msr(vcpu, msr))
2356			return kvm_pmu_set_msr(vcpu, msr_info);
2357		if (!ignore_msrs) {
2358			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2359				    msr, data);
2360			return 1;
2361		} else {
2362			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2363				    msr, data);
2364			break;
2365		}
2366	}
2367	return 0;
2368}
2369EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2370
2371
2372/*
2373 * Reads an msr value (of 'msr_index') into 'pdata'.
2374 * Returns 0 on success, non-0 otherwise.
2375 * Assumes vcpu_load() was already called.
2376 */
2377int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2378{
2379	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2380}
2381EXPORT_SYMBOL_GPL(kvm_get_msr);
2382
2383static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2384{
2385	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2386
2387	if (!msr_mtrr_valid(msr))
2388		return 1;
2389
2390	if (msr == MSR_MTRRdefType)
2391		*pdata = vcpu->arch.mtrr_state.def_type +
2392			 (vcpu->arch.mtrr_state.enabled << 10);
2393	else if (msr == MSR_MTRRfix64K_00000)
2394		*pdata = p[0];
2395	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2396		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2397	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2398		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2399	else if (msr == MSR_IA32_CR_PAT)
2400		*pdata = vcpu->arch.pat;
2401	else {	/* Variable MTRRs */
2402		int idx, is_mtrr_mask;
2403		u64 *pt;
2404
2405		idx = (msr - 0x200) / 2;
2406		is_mtrr_mask = msr - 0x200 - 2 * idx;
2407		if (!is_mtrr_mask)
2408			pt =
2409			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2410		else
2411			pt =
2412			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2413		*pdata = *pt;
2414	}
2415
2416	return 0;
2417}
2418
2419static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2420{
2421	u64 data;
2422	u64 mcg_cap = vcpu->arch.mcg_cap;
2423	unsigned bank_num = mcg_cap & 0xff;
2424
2425	switch (msr) {
2426	case MSR_IA32_P5_MC_ADDR:
2427	case MSR_IA32_P5_MC_TYPE:
2428		data = 0;
2429		break;
2430	case MSR_IA32_MCG_CAP:
2431		data = vcpu->arch.mcg_cap;
2432		break;
2433	case MSR_IA32_MCG_CTL:
2434		if (!(mcg_cap & MCG_CTL_P))
2435			return 1;
2436		data = vcpu->arch.mcg_ctl;
2437		break;
2438	case MSR_IA32_MCG_STATUS:
2439		data = vcpu->arch.mcg_status;
2440		break;
2441	default:
2442		if (msr >= MSR_IA32_MC0_CTL &&
2443		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2444			u32 offset = msr - MSR_IA32_MC0_CTL;
2445			data = vcpu->arch.mce_banks[offset];
2446			break;
2447		}
2448		return 1;
2449	}
2450	*pdata = data;
2451	return 0;
2452}
2453
2454static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2455{
2456	u64 data = 0;
2457	struct kvm *kvm = vcpu->kvm;
2458
2459	switch (msr) {
2460	case HV_X64_MSR_GUEST_OS_ID:
2461		data = kvm->arch.hv_guest_os_id;
2462		break;
2463	case HV_X64_MSR_HYPERCALL:
2464		data = kvm->arch.hv_hypercall;
2465		break;
2466	case HV_X64_MSR_TIME_REF_COUNT: {
2467		data =
2468		     div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2469		break;
2470	}
2471	case HV_X64_MSR_REFERENCE_TSC:
2472		data = kvm->arch.hv_tsc_page;
2473		break;
2474	default:
2475		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2476		return 1;
2477	}
2478
2479	*pdata = data;
2480	return 0;
2481}
2482
2483static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2484{
2485	u64 data = 0;
2486
2487	switch (msr) {
2488	case HV_X64_MSR_VP_INDEX: {
2489		int r;
2490		struct kvm_vcpu *v;
2491		kvm_for_each_vcpu(r, v, vcpu->kvm) {
2492			if (v == vcpu) {
2493				data = r;
2494				break;
2495			}
2496		}
2497		break;
2498	}
2499	case HV_X64_MSR_EOI:
2500		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2501	case HV_X64_MSR_ICR:
2502		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2503	case HV_X64_MSR_TPR:
2504		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2505	case HV_X64_MSR_APIC_ASSIST_PAGE:
2506		data = vcpu->arch.hv_vapic;
2507		break;
2508	default:
2509		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2510		return 1;
2511	}
2512	*pdata = data;
2513	return 0;
2514}
2515
2516int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2517{
2518	u64 data;
2519
2520	switch (msr) {
2521	case MSR_IA32_PLATFORM_ID:
2522	case MSR_IA32_EBL_CR_POWERON:
2523	case MSR_IA32_DEBUGCTLMSR:
2524	case MSR_IA32_LASTBRANCHFROMIP:
2525	case MSR_IA32_LASTBRANCHTOIP:
2526	case MSR_IA32_LASTINTFROMIP:
2527	case MSR_IA32_LASTINTTOIP:
2528	case MSR_K8_SYSCFG:
2529	case MSR_K7_HWCR:
2530	case MSR_VM_HSAVE_PA:
2531	case MSR_K7_EVNTSEL0:
2532	case MSR_K7_EVNTSEL1:
2533	case MSR_K7_EVNTSEL2:
2534	case MSR_K7_EVNTSEL3:
2535	case MSR_K7_PERFCTR0:
2536	case MSR_K7_PERFCTR1:
2537	case MSR_K7_PERFCTR2:
2538	case MSR_K7_PERFCTR3:
2539	case MSR_K8_INT_PENDING_MSG:
2540	case MSR_AMD64_NB_CFG:
2541	case MSR_FAM10H_MMIO_CONF_BASE:
2542	case MSR_AMD64_BU_CFG2:
2543		data = 0;
2544		break;
2545	case MSR_P6_PERFCTR0:
2546	case MSR_P6_PERFCTR1:
2547	case MSR_P6_EVNTSEL0:
2548	case MSR_P6_EVNTSEL1:
2549		if (kvm_pmu_msr(vcpu, msr))
2550			return kvm_pmu_get_msr(vcpu, msr, pdata);
2551		data = 0;
2552		break;
2553	case MSR_IA32_UCODE_REV:
2554		data = 0x100000000ULL;
2555		break;
2556	case MSR_MTRRcap:
2557		data = 0x500 | KVM_NR_VAR_MTRR;
2558		break;
2559	case 0x200 ... 0x2ff:
2560		return get_msr_mtrr(vcpu, msr, pdata);
2561	case 0xcd: /* fsb frequency */
2562		data = 3;
2563		break;
2564		/*
2565		 * MSR_EBC_FREQUENCY_ID
2566		 * Conservative value valid for even the basic CPU models.
2567		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2568		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2569		 * and 266MHz for model 3, or 4. Set Core Clock
2570		 * Frequency to System Bus Frequency Ratio to 1 (bits
2571		 * 31:24) even though these are only valid for CPU
2572		 * models > 2, however guests may end up dividing or
2573		 * multiplying by zero otherwise.
2574		 */
2575	case MSR_EBC_FREQUENCY_ID:
2576		data = 1 << 24;
2577		break;
2578	case MSR_IA32_APICBASE:
2579		data = kvm_get_apic_base(vcpu);
2580		break;
2581	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2582		return kvm_x2apic_msr_read(vcpu, msr, pdata);
2583		break;
2584	case MSR_IA32_TSCDEADLINE:
2585		data = kvm_get_lapic_tscdeadline_msr(vcpu);
2586		break;
2587	case MSR_IA32_TSC_ADJUST:
2588		data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2589		break;
2590	case MSR_IA32_MISC_ENABLE:
2591		data = vcpu->arch.ia32_misc_enable_msr;
2592		break;
2593	case MSR_IA32_PERF_STATUS:
2594		/* TSC increment by tick */
2595		data = 1000ULL;
2596		/* CPU multiplier */
2597		data |= (((uint64_t)4ULL) << 40);
2598		break;
2599	case MSR_EFER:
2600		data = vcpu->arch.efer;
2601		break;
2602	case MSR_KVM_WALL_CLOCK:
2603	case MSR_KVM_WALL_CLOCK_NEW:
2604		data = vcpu->kvm->arch.wall_clock;
2605		break;
2606	case MSR_KVM_SYSTEM_TIME:
2607	case MSR_KVM_SYSTEM_TIME_NEW:
2608		data = vcpu->arch.time;
2609		break;
2610	case MSR_KVM_ASYNC_PF_EN:
2611		data = vcpu->arch.apf.msr_val;
2612		break;
2613	case MSR_KVM_STEAL_TIME:
2614		data = vcpu->arch.st.msr_val;
2615		break;
2616	case MSR_KVM_PV_EOI_EN:
2617		data = vcpu->arch.pv_eoi.msr_val;
2618		break;
2619	case MSR_IA32_P5_MC_ADDR:
2620	case MSR_IA32_P5_MC_TYPE:
2621	case MSR_IA32_MCG_CAP:
2622	case MSR_IA32_MCG_CTL:
2623	case MSR_IA32_MCG_STATUS:
2624	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2625		return get_msr_mce(vcpu, msr, pdata);
2626	case MSR_K7_CLK_CTL:
2627		/*
2628		 * Provide expected ramp-up count for K7. All other
2629		 * are set to zero, indicating minimum divisors for
2630		 * every field.
2631		 *
2632		 * This prevents guest kernels on AMD host with CPU
2633		 * type 6, model 8 and higher from exploding due to
2634		 * the rdmsr failing.
2635		 */
2636		data = 0x20000000;
2637		break;
2638	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2639		if (kvm_hv_msr_partition_wide(msr)) {
2640			int r;
2641			mutex_lock(&vcpu->kvm->lock);
2642			r = get_msr_hyperv_pw(vcpu, msr, pdata);
2643			mutex_unlock(&vcpu->kvm->lock);
2644			return r;
2645		} else
2646			return get_msr_hyperv(vcpu, msr, pdata);
2647		break;
2648	case MSR_IA32_BBL_CR_CTL3:
2649		/* This legacy MSR exists but isn't fully documented in current
2650		 * silicon.  It is however accessed by winxp in very narrow
2651		 * scenarios where it sets bit #19, itself documented as
2652		 * a "reserved" bit.  Best effort attempt to source coherent
2653		 * read data here should the balance of the register be
2654		 * interpreted by the guest:
2655		 *
2656		 * L2 cache control register 3: 64GB range, 256KB size,
2657		 * enabled, latency 0x1, configured
2658		 */
2659		data = 0xbe702111;
2660		break;
2661	case MSR_AMD64_OSVW_ID_LENGTH:
2662		if (!guest_cpuid_has_osvw(vcpu))
2663			return 1;
2664		data = vcpu->arch.osvw.length;
2665		break;
2666	case MSR_AMD64_OSVW_STATUS:
2667		if (!guest_cpuid_has_osvw(vcpu))
2668			return 1;
2669		data = vcpu->arch.osvw.status;
2670		break;
2671	default:
2672		if (kvm_pmu_msr(vcpu, msr))
2673			return kvm_pmu_get_msr(vcpu, msr, pdata);
2674		if (!ignore_msrs) {
2675			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2676			return 1;
2677		} else {
2678			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2679			data = 0;
2680		}
2681		break;
2682	}
2683	*pdata = data;
2684	return 0;
2685}
2686EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2687
2688/*
2689 * Read or write a bunch of msrs. All parameters are kernel addresses.
2690 *
2691 * @return number of msrs set successfully.
2692 */
2693static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2694		    struct kvm_msr_entry *entries,
2695		    int (*do_msr)(struct kvm_vcpu *vcpu,
2696				  unsigned index, u64 *data))
2697{
2698	int i, idx;
2699
2700	idx = srcu_read_lock(&vcpu->kvm->srcu);
2701	for (i = 0; i < msrs->nmsrs; ++i)
2702		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2703			break;
2704	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2705
2706	return i;
2707}
2708
2709/*
2710 * Read or write a bunch of msrs. Parameters are user addresses.
2711 *
2712 * @return number of msrs set successfully.
2713 */
2714static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2715		  int (*do_msr)(struct kvm_vcpu *vcpu,
2716				unsigned index, u64 *data),
2717		  int writeback)
2718{
2719	struct kvm_msrs msrs;
2720	struct kvm_msr_entry *entries;
2721	int r, n;
2722	unsigned size;
2723
2724	r = -EFAULT;
2725	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2726		goto out;
2727
2728	r = -E2BIG;
2729	if (msrs.nmsrs >= MAX_IO_MSRS)
2730		goto out;
2731
2732	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2733	entries = memdup_user(user_msrs->entries, size);
2734	if (IS_ERR(entries)) {
2735		r = PTR_ERR(entries);
2736		goto out;
2737	}
2738
2739	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2740	if (r < 0)
2741		goto out_free;
2742
2743	r = -EFAULT;
2744	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2745		goto out_free;
2746
2747	r = n;
2748
2749out_free:
2750	kfree(entries);
2751out:
2752	return r;
2753}
2754
2755int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2756{
2757	int r;
2758
2759	switch (ext) {
2760	case KVM_CAP_IRQCHIP:
2761	case KVM_CAP_HLT:
2762	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2763	case KVM_CAP_SET_TSS_ADDR:
2764	case KVM_CAP_EXT_CPUID:
2765	case KVM_CAP_EXT_EMUL_CPUID:
2766	case KVM_CAP_CLOCKSOURCE:
2767	case KVM_CAP_PIT:
2768	case KVM_CAP_NOP_IO_DELAY:
2769	case KVM_CAP_MP_STATE:
2770	case KVM_CAP_SYNC_MMU:
2771	case KVM_CAP_USER_NMI:
2772	case KVM_CAP_REINJECT_CONTROL:
2773	case KVM_CAP_IRQ_INJECT_STATUS:
2774	case KVM_CAP_IOEVENTFD:
2775	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2776	case KVM_CAP_PIT2:
2777	case KVM_CAP_PIT_STATE2:
2778	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2779	case KVM_CAP_XEN_HVM:
2780	case KVM_CAP_ADJUST_CLOCK:
2781	case KVM_CAP_VCPU_EVENTS:
2782	case KVM_CAP_HYPERV:
2783	case KVM_CAP_HYPERV_VAPIC:
2784	case KVM_CAP_HYPERV_SPIN:
2785	case KVM_CAP_PCI_SEGMENT:
2786	case KVM_CAP_DEBUGREGS:
2787	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2788	case KVM_CAP_XSAVE:
2789	case KVM_CAP_ASYNC_PF:
2790	case KVM_CAP_GET_TSC_KHZ:
2791	case KVM_CAP_KVMCLOCK_CTRL:
2792	case KVM_CAP_READONLY_MEM:
2793	case KVM_CAP_HYPERV_TIME:
2794	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2795	case KVM_CAP_TSC_DEADLINE_TIMER:
2796#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2797	case KVM_CAP_ASSIGN_DEV_IRQ:
2798	case KVM_CAP_PCI_2_3:
2799#endif
2800		r = 1;
2801		break;
2802	case KVM_CAP_COALESCED_MMIO:
2803		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2804		break;
2805	case KVM_CAP_VAPIC:
2806		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2807		break;
2808	case KVM_CAP_NR_VCPUS:
2809		r = KVM_SOFT_MAX_VCPUS;
2810		break;
2811	case KVM_CAP_MAX_VCPUS:
2812		r = KVM_MAX_VCPUS;
2813		break;
2814	case KVM_CAP_NR_MEMSLOTS:
2815		r = KVM_USER_MEM_SLOTS;
2816		break;
2817	case KVM_CAP_PV_MMU:	/* obsolete */
2818		r = 0;
2819		break;
2820#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2821	case KVM_CAP_IOMMU:
2822		r = iommu_present(&pci_bus_type);
2823		break;
2824#endif
2825	case KVM_CAP_MCE:
2826		r = KVM_MAX_MCE_BANKS;
2827		break;
2828	case KVM_CAP_XCRS:
2829		r = cpu_has_xsave;
2830		break;
2831	case KVM_CAP_TSC_CONTROL:
2832		r = kvm_has_tsc_control;
2833		break;
2834	default:
2835		r = 0;
2836		break;
2837	}
2838	return r;
2839
2840}
2841
2842long kvm_arch_dev_ioctl(struct file *filp,
2843			unsigned int ioctl, unsigned long arg)
2844{
2845	void __user *argp = (void __user *)arg;
2846	long r;
2847
2848	switch (ioctl) {
2849	case KVM_GET_MSR_INDEX_LIST: {
2850		struct kvm_msr_list __user *user_msr_list = argp;
2851		struct kvm_msr_list msr_list;
2852		unsigned n;
2853
2854		r = -EFAULT;
2855		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2856			goto out;
2857		n = msr_list.nmsrs;
2858		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2859		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2860			goto out;
2861		r = -E2BIG;
2862		if (n < msr_list.nmsrs)
2863			goto out;
2864		r = -EFAULT;
2865		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2866				 num_msrs_to_save * sizeof(u32)))
2867			goto out;
2868		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2869				 &emulated_msrs,
2870				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2871			goto out;
2872		r = 0;
2873		break;
2874	}
2875	case KVM_GET_SUPPORTED_CPUID:
2876	case KVM_GET_EMULATED_CPUID: {
2877		struct kvm_cpuid2 __user *cpuid_arg = argp;
2878		struct kvm_cpuid2 cpuid;
2879
2880		r = -EFAULT;
2881		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2882			goto out;
2883
2884		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2885					    ioctl);
2886		if (r)
2887			goto out;
2888
2889		r = -EFAULT;
2890		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2891			goto out;
2892		r = 0;
2893		break;
2894	}
2895	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2896		u64 mce_cap;
2897
2898		mce_cap = KVM_MCE_CAP_SUPPORTED;
2899		r = -EFAULT;
2900		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2901			goto out;
2902		r = 0;
2903		break;
2904	}
2905	default:
2906		r = -EINVAL;
2907	}
2908out:
2909	return r;
2910}
2911
2912static void wbinvd_ipi(void *garbage)
2913{
2914	wbinvd();
2915}
2916
2917static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2918{
2919	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2920}
2921
2922void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2923{
2924	/* Address WBINVD may be executed by guest */
2925	if (need_emulate_wbinvd(vcpu)) {
2926		if (kvm_x86_ops->has_wbinvd_exit())
2927			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2928		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2929			smp_call_function_single(vcpu->cpu,
2930					wbinvd_ipi, NULL, 1);
2931	}
2932
2933	kvm_x86_ops->vcpu_load(vcpu, cpu);
2934
2935	/* Apply any externally detected TSC adjustments (due to suspend) */
2936	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2937		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2938		vcpu->arch.tsc_offset_adjustment = 0;
2939		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2940	}
2941
2942	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2943		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2944				native_read_tsc() - vcpu->arch.last_host_tsc;
2945		if (tsc_delta < 0)
2946			mark_tsc_unstable("KVM discovered backwards TSC");
2947		if (check_tsc_unstable()) {
2948			u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2949						vcpu->arch.last_guest_tsc);
2950			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2951			vcpu->arch.tsc_catchup = 1;
2952		}
2953		/*
2954		 * On a host with synchronized TSC, there is no need to update
2955		 * kvmclock on vcpu->cpu migration
2956		 */
2957		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2958			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2959		if (vcpu->cpu != cpu)
2960			kvm_migrate_timers(vcpu);
2961		vcpu->cpu = cpu;
2962	}
2963
2964	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2965}
2966
2967void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2968{
2969	kvm_x86_ops->vcpu_put(vcpu);
2970	kvm_put_guest_fpu(vcpu);
2971	vcpu->arch.last_host_tsc = native_read_tsc();
2972}
2973
2974static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2975				    struct kvm_lapic_state *s)
2976{
2977	kvm_x86_ops->sync_pir_to_irr(vcpu);
2978	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2979
2980	return 0;
2981}
2982
2983static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2984				    struct kvm_lapic_state *s)
2985{
2986	kvm_apic_post_state_restore(vcpu, s);
2987	update_cr8_intercept(vcpu);
2988
2989	return 0;
2990}
2991
2992static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2993				    struct kvm_interrupt *irq)
2994{
2995	if (irq->irq >= KVM_NR_INTERRUPTS)
2996		return -EINVAL;
2997	if (irqchip_in_kernel(vcpu->kvm))
2998		return -ENXIO;
2999
3000	kvm_queue_interrupt(vcpu, irq->irq, false);
3001	kvm_make_request(KVM_REQ_EVENT, vcpu);
3002
3003	return 0;
3004}
3005
3006static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3007{
3008	kvm_inject_nmi(vcpu);
3009
3010	return 0;
3011}
3012
3013static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3014					   struct kvm_tpr_access_ctl *tac)
3015{
3016	if (tac->flags)
3017		return -EINVAL;
3018	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3019	return 0;
3020}
3021
3022static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3023					u64 mcg_cap)
3024{
3025	int r;
3026	unsigned bank_num = mcg_cap & 0xff, bank;
3027
3028	r = -EINVAL;
3029	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3030		goto out;
3031	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
3032		goto out;
3033	r = 0;
3034	vcpu->arch.mcg_cap = mcg_cap;
3035	/* Init IA32_MCG_CTL to all 1s */
3036	if (mcg_cap & MCG_CTL_P)
3037		vcpu->arch.mcg_ctl = ~(u64)0;
3038	/* Init IA32_MCi_CTL to all 1s */
3039	for (bank = 0; bank < bank_num; bank++)
3040		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3041out:
3042	return r;
3043}
3044
3045static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3046				      struct kvm_x86_mce *mce)
3047{
3048	u64 mcg_cap = vcpu->arch.mcg_cap;
3049	unsigned bank_num = mcg_cap & 0xff;
3050	u64 *banks = vcpu->arch.mce_banks;
3051
3052	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3053		return -EINVAL;
3054	/*
3055	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3056	 * reporting is disabled
3057	 */
3058	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3059	    vcpu->arch.mcg_ctl != ~(u64)0)
3060		return 0;
3061	banks += 4 * mce->bank;
3062	/*
3063	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3064	 * reporting is disabled for the bank
3065	 */
3066	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3067		return 0;
3068	if (mce->status & MCI_STATUS_UC) {
3069		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3070		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3071			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3072			return 0;
3073		}
3074		if (banks[1] & MCI_STATUS_VAL)
3075			mce->status |= MCI_STATUS_OVER;
3076		banks[2] = mce->addr;
3077		banks[3] = mce->misc;
3078		vcpu->arch.mcg_status = mce->mcg_status;
3079		banks[1] = mce->status;
3080		kvm_queue_exception(vcpu, MC_VECTOR);
3081	} else if (!(banks[1] & MCI_STATUS_VAL)
3082		   || !(banks[1] & MCI_STATUS_UC)) {
3083		if (banks[1] & MCI_STATUS_VAL)
3084			mce->status |= MCI_STATUS_OVER;
3085		banks[2] = mce->addr;
3086		banks[3] = mce->misc;
3087		banks[1] = mce->status;
3088	} else
3089		banks[1] |= MCI_STATUS_OVER;
3090	return 0;
3091}
3092
3093static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3094					       struct kvm_vcpu_events *events)
3095{
3096	process_nmi(vcpu);
3097	events->exception.injected =
3098		vcpu->arch.exception.pending &&
3099		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3100	events->exception.nr = vcpu->arch.exception.nr;
3101	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3102	events->exception.pad = 0;
3103	events->exception.error_code = vcpu->arch.exception.error_code;
3104
3105	events->interrupt.injected =
3106		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3107	events->interrupt.nr = vcpu->arch.interrupt.nr;
3108	events->interrupt.soft = 0;
3109	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3110
3111	events->nmi.injected = vcpu->arch.nmi_injected;
3112	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3113	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3114	events->nmi.pad = 0;
3115
3116	events->sipi_vector = 0; /* never valid when reporting to user space */
3117
3118	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3119			 | KVM_VCPUEVENT_VALID_SHADOW);
3120	memset(&events->reserved, 0, sizeof(events->reserved));
3121}
3122
3123static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3124					      struct kvm_vcpu_events *events)
3125{
3126	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3127			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3128			      | KVM_VCPUEVENT_VALID_SHADOW))
3129		return -EINVAL;
3130
3131	process_nmi(vcpu);
3132	vcpu->arch.exception.pending = events->exception.injected;
3133	vcpu->arch.exception.nr = events->exception.nr;
3134	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3135	vcpu->arch.exception.error_code = events->exception.error_code;
3136
3137	vcpu->arch.interrupt.pending = events->interrupt.injected;
3138	vcpu->arch.interrupt.nr = events->interrupt.nr;
3139	vcpu->arch.interrupt.soft = events->interrupt.soft;
3140	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3141		kvm_x86_ops->set_interrupt_shadow(vcpu,
3142						  events->interrupt.shadow);
3143
3144	vcpu->arch.nmi_injected = events->nmi.injected;
3145	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3146		vcpu->arch.nmi_pending = events->nmi.pending;
3147	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3148
3149	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3150	    kvm_vcpu_has_lapic(vcpu))
3151		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3152
3153	kvm_make_request(KVM_REQ_EVENT, vcpu);
3154
3155	return 0;
3156}
3157
3158static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3159					     struct kvm_debugregs *dbgregs)
3160{
3161	unsigned long val;
3162
3163	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3164	kvm_get_dr(vcpu, 6, &val);
3165	dbgregs->dr6 = val;
3166	dbgregs->dr7 = vcpu->arch.dr7;
3167	dbgregs->flags = 0;
3168	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3169}
3170
3171static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3172					    struct kvm_debugregs *dbgregs)
3173{
3174	if (dbgregs->flags)
3175		return -EINVAL;
3176
3177	if (dbgregs->dr6 & ~0xffffffffull)
3178		return -EINVAL;
3179	if (dbgregs->dr7 & ~0xffffffffull)
3180		return -EINVAL;
3181
3182	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3183	kvm_update_dr0123(vcpu);
3184	vcpu->arch.dr6 = dbgregs->dr6;
3185	kvm_update_dr6(vcpu);
3186	vcpu->arch.dr7 = dbgregs->dr7;
3187	kvm_update_dr7(vcpu);
3188
3189	return 0;
3190}
3191
3192#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3193
3194static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3195{
3196	struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3197	u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3198	u64 valid;
3199
3200	/*
3201	 * Copy legacy XSAVE area, to avoid complications with CPUID
3202	 * leaves 0 and 1 in the loop below.
3203	 */
3204	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3205
3206	/* Set XSTATE_BV */
3207	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3208
3209	/*
3210	 * Copy each region from the possibly compacted offset to the
3211	 * non-compacted offset.
3212	 */
3213	valid = xstate_bv & ~XSTATE_FPSSE;
3214	while (valid) {
3215		u64 feature = valid & -valid;
3216		int index = fls64(feature) - 1;
3217		void *src = get_xsave_addr(xsave, feature);
3218
3219		if (src) {
3220			u32 size, offset, ecx, edx;
3221			cpuid_count(XSTATE_CPUID, index,
3222				    &size, &offset, &ecx, &edx);
3223			memcpy(dest + offset, src, size);
3224		}
3225
3226		valid -= feature;
3227	}
3228}
3229
3230static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3231{
3232	struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3233	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3234	u64 valid;
3235
3236	/*
3237	 * Copy legacy XSAVE area, to avoid complications with CPUID
3238	 * leaves 0 and 1 in the loop below.
3239	 */
3240	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3241
3242	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3243	xsave->xsave_hdr.xstate_bv = xstate_bv;
3244	if (cpu_has_xsaves)
3245		xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3246
3247	/*
3248	 * Copy each region from the non-compacted offset to the
3249	 * possibly compacted offset.
3250	 */
3251	valid = xstate_bv & ~XSTATE_FPSSE;
3252	while (valid) {
3253		u64 feature = valid & -valid;
3254		int index = fls64(feature) - 1;
3255		void *dest = get_xsave_addr(xsave, feature);
3256
3257		if (dest) {
3258			u32 size, offset, ecx, edx;
3259			cpuid_count(XSTATE_CPUID, index,
3260				    &size, &offset, &ecx, &edx);
3261			memcpy(dest, src + offset, size);
3262		} else
3263			WARN_ON_ONCE(1);
3264
3265		valid -= feature;
3266	}
3267}
3268
3269static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3270					 struct kvm_xsave *guest_xsave)
3271{
3272	if (cpu_has_xsave) {
3273		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3274		fill_xsave((u8 *) guest_xsave->region, vcpu);
3275	} else {
3276		memcpy(guest_xsave->region,
3277			&vcpu->arch.guest_fpu.state->fxsave,
3278			sizeof(struct i387_fxsave_struct));
3279		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3280			XSTATE_FPSSE;
3281	}
3282}
3283
3284static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3285					struct kvm_xsave *guest_xsave)
3286{
3287	u64 xstate_bv =
3288		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3289
3290	if (cpu_has_xsave) {
3291		/*
3292		 * Here we allow setting states that are not present in
3293		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3294		 * with old userspace.
3295		 */
3296		if (xstate_bv & ~kvm_supported_xcr0())
3297			return -EINVAL;
3298		load_xsave(vcpu, (u8 *)guest_xsave->region);
3299	} else {
3300		if (xstate_bv & ~XSTATE_FPSSE)
3301			return -EINVAL;
3302		memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3303			guest_xsave->region, sizeof(struct i387_fxsave_struct));
3304	}
3305	return 0;
3306}
3307
3308static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3309					struct kvm_xcrs *guest_xcrs)
3310{
3311	if (!cpu_has_xsave) {
3312		guest_xcrs->nr_xcrs = 0;
3313		return;
3314	}
3315
3316	guest_xcrs->nr_xcrs = 1;
3317	guest_xcrs->flags = 0;
3318	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3319	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3320}
3321
3322static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3323				       struct kvm_xcrs *guest_xcrs)
3324{
3325	int i, r = 0;
3326
3327	if (!cpu_has_xsave)
3328		return -EINVAL;
3329
3330	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3331		return -EINVAL;
3332
3333	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3334		/* Only support XCR0 currently */
3335		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3336			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3337				guest_xcrs->xcrs[i].value);
3338			break;
3339		}
3340	if (r)
3341		r = -EINVAL;
3342	return r;
3343}
3344
3345/*
3346 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3347 * stopped by the hypervisor.  This function will be called from the host only.
3348 * EINVAL is returned when the host attempts to set the flag for a guest that
3349 * does not support pv clocks.
3350 */
3351static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3352{
3353	if (!vcpu->arch.pv_time_enabled)
3354		return -EINVAL;
3355	vcpu->arch.pvclock_set_guest_stopped_request = true;
3356	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3357	return 0;
3358}
3359
3360long kvm_arch_vcpu_ioctl(struct file *filp,
3361			 unsigned int ioctl, unsigned long arg)
3362{
3363	struct kvm_vcpu *vcpu = filp->private_data;
3364	void __user *argp = (void __user *)arg;
3365	int r;
3366	union {
3367		struct kvm_lapic_state *lapic;
3368		struct kvm_xsave *xsave;
3369		struct kvm_xcrs *xcrs;
3370		void *buffer;
3371	} u;
3372
3373	u.buffer = NULL;
3374	switch (ioctl) {
3375	case KVM_GET_LAPIC: {
3376		r = -EINVAL;
3377		if (!vcpu->arch.apic)
3378			goto out;
3379		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3380
3381		r = -ENOMEM;
3382		if (!u.lapic)
3383			goto out;
3384		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3385		if (r)
3386			goto out;
3387		r = -EFAULT;
3388		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3389			goto out;
3390		r = 0;
3391		break;
3392	}
3393	case KVM_SET_LAPIC: {
3394		r = -EINVAL;
3395		if (!vcpu->arch.apic)
3396			goto out;
3397		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3398		if (IS_ERR(u.lapic))
3399			return PTR_ERR(u.lapic);
3400
3401		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3402		break;
3403	}
3404	case KVM_INTERRUPT: {
3405		struct kvm_interrupt irq;
3406
3407		r = -EFAULT;
3408		if (copy_from_user(&irq, argp, sizeof irq))
3409			goto out;
3410		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3411		break;
3412	}
3413	case KVM_NMI: {
3414		r = kvm_vcpu_ioctl_nmi(vcpu);
3415		break;
3416	}
3417	case KVM_SET_CPUID: {
3418		struct kvm_cpuid __user *cpuid_arg = argp;
3419		struct kvm_cpuid cpuid;
3420
3421		r = -EFAULT;
3422		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3423			goto out;
3424		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3425		break;
3426	}
3427	case KVM_SET_CPUID2: {
3428		struct kvm_cpuid2 __user *cpuid_arg = argp;
3429		struct kvm_cpuid2 cpuid;
3430
3431		r = -EFAULT;
3432		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3433			goto out;
3434		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3435					      cpuid_arg->entries);
3436		break;
3437	}
3438	case KVM_GET_CPUID2: {
3439		struct kvm_cpuid2 __user *cpuid_arg = argp;
3440		struct kvm_cpuid2 cpuid;
3441
3442		r = -EFAULT;
3443		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3444			goto out;
3445		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3446					      cpuid_arg->entries);
3447		if (r)
3448			goto out;
3449		r = -EFAULT;
3450		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3451			goto out;
3452		r = 0;
3453		break;
3454	}
3455	case KVM_GET_MSRS:
3456		r = msr_io(vcpu, argp, kvm_get_msr, 1);
3457		break;
3458	case KVM_SET_MSRS:
3459		r = msr_io(vcpu, argp, do_set_msr, 0);
3460		break;
3461	case KVM_TPR_ACCESS_REPORTING: {
3462		struct kvm_tpr_access_ctl tac;
3463
3464		r = -EFAULT;
3465		if (copy_from_user(&tac, argp, sizeof tac))
3466			goto out;
3467		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3468		if (r)
3469			goto out;
3470		r = -EFAULT;
3471		if (copy_to_user(argp, &tac, sizeof tac))
3472			goto out;
3473		r = 0;
3474		break;
3475	};
3476	case KVM_SET_VAPIC_ADDR: {
3477		struct kvm_vapic_addr va;
3478
3479		r = -EINVAL;
3480		if (!irqchip_in_kernel(vcpu->kvm))
3481			goto out;
3482		r = -EFAULT;
3483		if (copy_from_user(&va, argp, sizeof va))
3484			goto out;
3485		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3486		break;
3487	}
3488	case KVM_X86_SETUP_MCE: {
3489		u64 mcg_cap;
3490
3491		r = -EFAULT;
3492		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3493			goto out;
3494		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3495		break;
3496	}
3497	case KVM_X86_SET_MCE: {
3498		struct kvm_x86_mce mce;
3499
3500		r = -EFAULT;
3501		if (copy_from_user(&mce, argp, sizeof mce))
3502			goto out;
3503		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3504		break;
3505	}
3506	case KVM_GET_VCPU_EVENTS: {
3507		struct kvm_vcpu_events events;
3508
3509		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3510
3511		r = -EFAULT;
3512		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3513			break;
3514		r = 0;
3515		break;
3516	}
3517	case KVM_SET_VCPU_EVENTS: {
3518		struct kvm_vcpu_events events;
3519
3520		r = -EFAULT;
3521		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3522			break;
3523
3524		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3525		break;
3526	}
3527	case KVM_GET_DEBUGREGS: {
3528		struct kvm_debugregs dbgregs;
3529
3530		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3531
3532		r = -EFAULT;
3533		if (copy_to_user(argp, &dbgregs,
3534				 sizeof(struct kvm_debugregs)))
3535			break;
3536		r = 0;
3537		break;
3538	}
3539	case KVM_SET_DEBUGREGS: {
3540		struct kvm_debugregs dbgregs;
3541
3542		r = -EFAULT;
3543		if (copy_from_user(&dbgregs, argp,
3544				   sizeof(struct kvm_debugregs)))
3545			break;
3546
3547		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3548		break;
3549	}
3550	case KVM_GET_XSAVE: {
3551		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3552		r = -ENOMEM;
3553		if (!u.xsave)
3554			break;
3555
3556		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3557
3558		r = -EFAULT;
3559		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3560			break;
3561		r = 0;
3562		break;
3563	}
3564	case KVM_SET_XSAVE: {
3565		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3566		if (IS_ERR(u.xsave))
3567			return PTR_ERR(u.xsave);
3568
3569		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3570		break;
3571	}
3572	case KVM_GET_XCRS: {
3573		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3574		r = -ENOMEM;
3575		if (!u.xcrs)
3576			break;
3577
3578		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3579
3580		r = -EFAULT;
3581		if (copy_to_user(argp, u.xcrs,
3582				 sizeof(struct kvm_xcrs)))
3583			break;
3584		r = 0;
3585		break;
3586	}
3587	case KVM_SET_XCRS: {
3588		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3589		if (IS_ERR(u.xcrs))
3590			return PTR_ERR(u.xcrs);
3591
3592		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3593		break;
3594	}
3595	case KVM_SET_TSC_KHZ: {
3596		u32 user_tsc_khz;
3597
3598		r = -EINVAL;
3599		user_tsc_khz = (u32)arg;
3600
3601		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3602			goto out;
3603
3604		if (user_tsc_khz == 0)
3605			user_tsc_khz = tsc_khz;
3606
3607		kvm_set_tsc_khz(vcpu, user_tsc_khz);
3608
3609		r = 0;
3610		goto out;
3611	}
3612	case KVM_GET_TSC_KHZ: {
3613		r = vcpu->arch.virtual_tsc_khz;
3614		goto out;
3615	}
3616	case KVM_KVMCLOCK_CTRL: {
3617		r = kvm_set_guest_paused(vcpu);
3618		goto out;
3619	}
3620	default:
3621		r = -EINVAL;
3622	}
3623out:
3624	kfree(u.buffer);
3625	return r;
3626}
3627
3628int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3629{
3630	return VM_FAULT_SIGBUS;
3631}
3632
3633static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3634{
3635	int ret;
3636
3637	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3638		return -EINVAL;
3639	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3640	return ret;
3641}
3642
3643static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3644					      u64 ident_addr)
3645{
3646	kvm->arch.ept_identity_map_addr = ident_addr;
3647	return 0;
3648}
3649
3650static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3651					  u32 kvm_nr_mmu_pages)
3652{
3653	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3654		return -EINVAL;
3655
3656	mutex_lock(&kvm->slots_lock);
3657
3658	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3659	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3660
3661	mutex_unlock(&kvm->slots_lock);
3662	return 0;
3663}
3664
3665static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3666{
3667	return kvm->arch.n_max_mmu_pages;
3668}
3669
3670static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3671{
3672	int r;
3673
3674	r = 0;
3675	switch (chip->chip_id) {
3676	case KVM_IRQCHIP_PIC_MASTER:
3677		memcpy(&chip->chip.pic,
3678			&pic_irqchip(kvm)->pics[0],
3679			sizeof(struct kvm_pic_state));
3680		break;
3681	case KVM_IRQCHIP_PIC_SLAVE:
3682		memcpy(&chip->chip.pic,
3683			&pic_irqchip(kvm)->pics[1],
3684			sizeof(struct kvm_pic_state));
3685		break;
3686	case KVM_IRQCHIP_IOAPIC:
3687		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3688		break;
3689	default:
3690		r = -EINVAL;
3691		break;
3692	}
3693	return r;
3694}
3695
3696static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3697{
3698	int r;
3699
3700	r = 0;
3701	switch (chip->chip_id) {
3702	case KVM_IRQCHIP_PIC_MASTER:
3703		spin_lock(&pic_irqchip(kvm)->lock);
3704		memcpy(&pic_irqchip(kvm)->pics[0],
3705			&chip->chip.pic,
3706			sizeof(struct kvm_pic_state));
3707		spin_unlock(&pic_irqchip(kvm)->lock);
3708		break;
3709	case KVM_IRQCHIP_PIC_SLAVE:
3710		spin_lock(&pic_irqchip(kvm)->lock);
3711		memcpy(&pic_irqchip(kvm)->pics[1],
3712			&chip->chip.pic,
3713			sizeof(struct kvm_pic_state));
3714		spin_unlock(&pic_irqchip(kvm)->lock);
3715		break;
3716	case KVM_IRQCHIP_IOAPIC:
3717		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3718		break;
3719	default:
3720		r = -EINVAL;
3721		break;
3722	}
3723	kvm_pic_update_irq(pic_irqchip(kvm));
3724	return r;
3725}
3726
3727static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3728{
3729	int r = 0;
3730
3731	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3732	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3733	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3734	return r;
3735}
3736
3737static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3738{
3739	int i;
3740	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3741	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3742	for (i = 0; i < 3; i++)
3743		kvm_pit_load_count(kvm, i, ps->channels[i].count, 0);
3744	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3745	return 0;
3746}
3747
3748static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3749{
3750	int r = 0;
3751
3752	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3753	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3754		sizeof(ps->channels));
3755	ps->flags = kvm->arch.vpit->pit_state.flags;
3756	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3757	memset(&ps->reserved, 0, sizeof(ps->reserved));
3758	return r;
3759}
3760
3761static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3762{
3763	int r = 0, start = 0;
3764	int i;
3765	u32 prev_legacy, cur_legacy;
3766	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3767	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3768	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3769	if (!prev_legacy && cur_legacy)
3770		start = 1;
3771	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3772	       sizeof(kvm->arch.vpit->pit_state.channels));
3773	kvm->arch.vpit->pit_state.flags = ps->flags;
3774	for (i = 0; i < 3; i++)
3775		kvm_pit_load_count(kvm, i, kvm->arch.vpit->pit_state.channels[i].count, start);
3776	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3777	return r;
3778}
3779
3780static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3781				 struct kvm_reinject_control *control)
3782{
3783	if (!kvm->arch.vpit)
3784		return -ENXIO;
3785	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3786	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3787	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3788	return 0;
3789}
3790
3791/**
3792 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3793 * @kvm: kvm instance
3794 * @log: slot id and address to which we copy the log
3795 *
3796 * Steps 1-4 below provide general overview of dirty page logging. See
3797 * kvm_get_dirty_log_protect() function description for additional details.
3798 *
3799 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3800 * always flush the TLB (step 4) even if previous step failed  and the dirty
3801 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3802 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3803 * writes will be marked dirty for next log read.
3804 *
3805 *   1. Take a snapshot of the bit and clear it if needed.
3806 *   2. Write protect the corresponding page.
3807 *   3. Copy the snapshot to the userspace.
3808 *   4. Flush TLB's if needed.
3809 */
3810int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3811{
3812	bool is_dirty = false;
3813	int r;
3814
3815	mutex_lock(&kvm->slots_lock);
3816
3817	/*
3818	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3819	 */
3820	if (kvm_x86_ops->flush_log_dirty)
3821		kvm_x86_ops->flush_log_dirty(kvm);
3822
3823	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3824
3825	/*
3826	 * All the TLBs can be flushed out of mmu lock, see the comments in
3827	 * kvm_mmu_slot_remove_write_access().
3828	 */
3829	lockdep_assert_held(&kvm->slots_lock);
3830	if (is_dirty)
3831		kvm_flush_remote_tlbs(kvm);
3832
3833	mutex_unlock(&kvm->slots_lock);
3834	return r;
3835}
3836
3837int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3838			bool line_status)
3839{
3840	if (!irqchip_in_kernel(kvm))
3841		return -ENXIO;
3842
3843	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3844					irq_event->irq, irq_event->level,
3845					line_status);
3846	return 0;
3847}
3848
3849long kvm_arch_vm_ioctl(struct file *filp,
3850		       unsigned int ioctl, unsigned long arg)
3851{
3852	struct kvm *kvm = filp->private_data;
3853	void __user *argp = (void __user *)arg;
3854	int r = -ENOTTY;
3855	/*
3856	 * This union makes it completely explicit to gcc-3.x
3857	 * that these two variables' stack usage should be
3858	 * combined, not added together.
3859	 */
3860	union {
3861		struct kvm_pit_state ps;
3862		struct kvm_pit_state2 ps2;
3863		struct kvm_pit_config pit_config;
3864	} u;
3865
3866	switch (ioctl) {
3867	case KVM_SET_TSS_ADDR:
3868		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3869		break;
3870	case KVM_SET_IDENTITY_MAP_ADDR: {
3871		u64 ident_addr;
3872
3873		r = -EFAULT;
3874		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3875			goto out;
3876		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3877		break;
3878	}
3879	case KVM_SET_NR_MMU_PAGES:
3880		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3881		break;
3882	case KVM_GET_NR_MMU_PAGES:
3883		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3884		break;
3885	case KVM_CREATE_IRQCHIP: {
3886		struct kvm_pic *vpic;
3887
3888		mutex_lock(&kvm->lock);
3889		r = -EEXIST;
3890		if (kvm->arch.vpic)
3891			goto create_irqchip_unlock;
3892		r = -EINVAL;
3893		if (atomic_read(&kvm->online_vcpus))
3894			goto create_irqchip_unlock;
3895		r = -ENOMEM;
3896		vpic = kvm_create_pic(kvm);
3897		if (vpic) {
3898			r = kvm_ioapic_init(kvm);
3899			if (r) {
3900				mutex_lock(&kvm->slots_lock);
3901				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3902							  &vpic->dev_master);
3903				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3904							  &vpic->dev_slave);
3905				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3906							  &vpic->dev_eclr);
3907				mutex_unlock(&kvm->slots_lock);
3908				kfree(vpic);
3909				goto create_irqchip_unlock;
3910			}
3911		} else
3912			goto create_irqchip_unlock;
3913		smp_wmb();
3914		kvm->arch.vpic = vpic;
3915		smp_wmb();
3916		r = kvm_setup_default_irq_routing(kvm);
3917		if (r) {
3918			mutex_lock(&kvm->slots_lock);
3919			mutex_lock(&kvm->irq_lock);
3920			kvm_ioapic_destroy(kvm);
3921			kvm_destroy_pic(kvm);
3922			mutex_unlock(&kvm->irq_lock);
3923			mutex_unlock(&kvm->slots_lock);
3924		}
3925	create_irqchip_unlock:
3926		mutex_unlock(&kvm->lock);
3927		break;
3928	}
3929	case KVM_CREATE_PIT:
3930		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3931		goto create_pit;
3932	case KVM_CREATE_PIT2:
3933		r = -EFAULT;
3934		if (copy_from_user(&u.pit_config, argp,
3935				   sizeof(struct kvm_pit_config)))
3936			goto out;
3937	create_pit:
3938		mutex_lock(&kvm->slots_lock);
3939		r = -EEXIST;
3940		if (kvm->arch.vpit)
3941			goto create_pit_unlock;
3942		r = -ENOMEM;
3943		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3944		if (kvm->arch.vpit)
3945			r = 0;
3946	create_pit_unlock:
3947		mutex_unlock(&kvm->slots_lock);
3948		break;
3949	case KVM_GET_IRQCHIP: {
3950		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3951		struct kvm_irqchip *chip;
3952
3953		chip = memdup_user(argp, sizeof(*chip));
3954		if (IS_ERR(chip)) {
3955			r = PTR_ERR(chip);
3956			goto out;
3957		}
3958
3959		r = -ENXIO;
3960		if (!irqchip_in_kernel(kvm))
3961			goto get_irqchip_out;
3962		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3963		if (r)
3964			goto get_irqchip_out;
3965		r = -EFAULT;
3966		if (copy_to_user(argp, chip, sizeof *chip))
3967			goto get_irqchip_out;
3968		r = 0;
3969	get_irqchip_out:
3970		kfree(chip);
3971		break;
3972	}
3973	case KVM_SET_IRQCHIP: {
3974		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3975		struct kvm_irqchip *chip;
3976
3977		chip = memdup_user(argp, sizeof(*chip));
3978		if (IS_ERR(chip)) {
3979			r = PTR_ERR(chip);
3980			goto out;
3981		}
3982
3983		r = -ENXIO;
3984		if (!irqchip_in_kernel(kvm))
3985			goto set_irqchip_out;
3986		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3987		if (r)
3988			goto set_irqchip_out;
3989		r = 0;
3990	set_irqchip_out:
3991		kfree(chip);
3992		break;
3993	}
3994	case KVM_GET_PIT: {
3995		r = -EFAULT;
3996		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3997			goto out;
3998		r = -ENXIO;
3999		if (!kvm->arch.vpit)
4000			goto out;
4001		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4002		if (r)
4003			goto out;
4004		r = -EFAULT;
4005		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4006			goto out;
4007		r = 0;
4008		break;
4009	}
4010	case KVM_SET_PIT: {
4011		r = -EFAULT;
4012		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4013			goto out;
4014		r = -ENXIO;
4015		if (!kvm->arch.vpit)
4016			goto out;
4017		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4018		break;
4019	}
4020	case KVM_GET_PIT2: {
4021		r = -ENXIO;
4022		if (!kvm->arch.vpit)
4023			goto out;
4024		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4025		if (r)
4026			goto out;
4027		r = -EFAULT;
4028		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4029			goto out;
4030		r = 0;
4031		break;
4032	}
4033	case KVM_SET_PIT2: {
4034		r = -EFAULT;
4035		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4036			goto out;
4037		r = -ENXIO;
4038		if (!kvm->arch.vpit)
4039			goto out;
4040		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4041		break;
4042	}
4043	case KVM_REINJECT_CONTROL: {
4044		struct kvm_reinject_control control;
4045		r =  -EFAULT;
4046		if (copy_from_user(&control, argp, sizeof(control)))
4047			goto out;
4048		r = kvm_vm_ioctl_reinject(kvm, &control);
4049		break;
4050	}
4051	case KVM_XEN_HVM_CONFIG: {
4052		r = -EFAULT;
4053		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4054				   sizeof(struct kvm_xen_hvm_config)))
4055			goto out;
4056		r = -EINVAL;
4057		if (kvm->arch.xen_hvm_config.flags)
4058			goto out;
4059		r = 0;
4060		break;
4061	}
4062	case KVM_SET_CLOCK: {
4063		struct kvm_clock_data user_ns;
4064		u64 now_ns;
4065		s64 delta;
4066
4067		r = -EFAULT;
4068		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4069			goto out;
4070
4071		r = -EINVAL;
4072		if (user_ns.flags)
4073			goto out;
4074
4075		r = 0;
4076		local_irq_disable();
4077		now_ns = get_kernel_ns();
4078		delta = user_ns.clock - now_ns;
4079		local_irq_enable();
4080		kvm->arch.kvmclock_offset = delta;
4081		kvm_gen_update_masterclock(kvm);
4082		break;
4083	}
4084	case KVM_GET_CLOCK: {
4085		struct kvm_clock_data user_ns;
4086		u64 now_ns;
4087
4088		local_irq_disable();
4089		now_ns = get_kernel_ns();
4090		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4091		local_irq_enable();
4092		user_ns.flags = 0;
4093		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4094
4095		r = -EFAULT;
4096		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4097			goto out;
4098		r = 0;
4099		break;
4100	}
4101
4102	default:
4103		r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4104	}
4105out:
4106	return r;
4107}
4108
4109static void kvm_init_msr_list(void)
4110{
4111	u32 dummy[2];
4112	unsigned i, j;
4113
4114	/* skip the first msrs in the list. KVM-specific */
4115	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
4116		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4117			continue;
4118
4119		/*
4120		 * Even MSRs that are valid in the host may not be exposed
4121		 * to the guests in some cases.
4122		 */
4123		switch (msrs_to_save[i]) {
4124		case MSR_IA32_BNDCFGS:
4125			if (!kvm_x86_ops->mpx_supported())
4126				continue;
4127			break;
4128		case MSR_TSC_AUX:
4129			if (!kvm_x86_ops->rdtscp_supported())
4130				continue;
4131			break;
4132		default:
4133			break;
4134		}
4135
4136		if (j < i)
4137			msrs_to_save[j] = msrs_to_save[i];
4138		j++;
4139	}
4140	num_msrs_to_save = j;
4141}
4142
4143static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4144			   const void *v)
4145{
4146	int handled = 0;
4147	int n;
4148
4149	do {
4150		n = min(len, 8);
4151		if (!(vcpu->arch.apic &&
4152		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4153		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4154			break;
4155		handled += n;
4156		addr += n;
4157		len -= n;
4158		v += n;
4159	} while (len);
4160
4161	return handled;
4162}
4163
4164static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4165{
4166	int handled = 0;
4167	int n;
4168
4169	do {
4170		n = min(len, 8);
4171		if (!(vcpu->arch.apic &&
4172		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4173					 addr, n, v))
4174		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4175			break;
4176		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4177		handled += n;
4178		addr += n;
4179		len -= n;
4180		v += n;
4181	} while (len);
4182
4183	return handled;
4184}
4185
4186static void kvm_set_segment(struct kvm_vcpu *vcpu,
4187			struct kvm_segment *var, int seg)
4188{
4189	kvm_x86_ops->set_segment(vcpu, var, seg);
4190}
4191
4192void kvm_get_segment(struct kvm_vcpu *vcpu,
4193		     struct kvm_segment *var, int seg)
4194{
4195	kvm_x86_ops->get_segment(vcpu, var, seg);
4196}
4197
4198gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4199			   struct x86_exception *exception)
4200{
4201	gpa_t t_gpa;
4202
4203	BUG_ON(!mmu_is_nested(vcpu));
4204
4205	/* NPT walks are always user-walks */
4206	access |= PFERR_USER_MASK;
4207	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4208
4209	return t_gpa;
4210}
4211
4212gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4213			      struct x86_exception *exception)
4214{
4215	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4216	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4217}
4218
4219 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4220				struct x86_exception *exception)
4221{
4222	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4223	access |= PFERR_FETCH_MASK;
4224	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4225}
4226
4227gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4228			       struct x86_exception *exception)
4229{
4230	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4231	access |= PFERR_WRITE_MASK;
4232	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4233}
4234
4235/* uses this to access any guest's mapped memory without checking CPL */
4236gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4237				struct x86_exception *exception)
4238{
4239	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4240}
4241
4242static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4243				      struct kvm_vcpu *vcpu, u32 access,
4244				      struct x86_exception *exception)
4245{
4246	void *data = val;
4247	int r = X86EMUL_CONTINUE;
4248
4249	while (bytes) {
4250		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4251							    exception);
4252		unsigned offset = addr & (PAGE_SIZE-1);
4253		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4254		int ret;
4255
4256		if (gpa == UNMAPPED_GVA)
4257			return X86EMUL_PROPAGATE_FAULT;
4258		ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4259					  offset, toread);
4260		if (ret < 0) {
4261			r = X86EMUL_IO_NEEDED;
4262			goto out;
4263		}
4264
4265		bytes -= toread;
4266		data += toread;
4267		addr += toread;
4268	}
4269out:
4270	return r;
4271}
4272
4273/* used for instruction fetching */
4274static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4275				gva_t addr, void *val, unsigned int bytes,
4276				struct x86_exception *exception)
4277{
4278	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4279	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4280	unsigned offset;
4281	int ret;
4282
4283	/* Inline kvm_read_guest_virt_helper for speed.  */
4284	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4285						    exception);
4286	if (unlikely(gpa == UNMAPPED_GVA))
4287		return X86EMUL_PROPAGATE_FAULT;
4288
4289	offset = addr & (PAGE_SIZE-1);
4290	if (WARN_ON(offset + bytes > PAGE_SIZE))
4291		bytes = (unsigned)PAGE_SIZE - offset;
4292	ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4293				  offset, bytes);
4294	if (unlikely(ret < 0))
4295		return X86EMUL_IO_NEEDED;
4296
4297	return X86EMUL_CONTINUE;
4298}
4299
4300int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4301			       gva_t addr, void *val, unsigned int bytes,
4302			       struct x86_exception *exception)
4303{
4304	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4305	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4306
4307	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4308					  exception);
4309}
4310EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4311
4312static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4313				      gva_t addr, void *val, unsigned int bytes,
4314				      struct x86_exception *exception)
4315{
4316	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4317	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4318}
4319
4320int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4321				       gva_t addr, void *val,
4322				       unsigned int bytes,
4323				       struct x86_exception *exception)
4324{
4325	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4326	void *data = val;
4327	int r = X86EMUL_CONTINUE;
4328
4329	while (bytes) {
4330		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4331							     PFERR_WRITE_MASK,
4332							     exception);
4333		unsigned offset = addr & (PAGE_SIZE-1);
4334		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4335		int ret;
4336
4337		if (gpa == UNMAPPED_GVA)
4338			return X86EMUL_PROPAGATE_FAULT;
4339		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4340		if (ret < 0) {
4341			r = X86EMUL_IO_NEEDED;
4342			goto out;
4343		}
4344
4345		bytes -= towrite;
4346		data += towrite;
4347		addr += towrite;
4348	}
4349out:
4350	return r;
4351}
4352EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4353
4354static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4355				gpa_t *gpa, struct x86_exception *exception,
4356				bool write)
4357{
4358	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4359		| (write ? PFERR_WRITE_MASK : 0);
4360
4361	if (vcpu_match_mmio_gva(vcpu, gva)
4362	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4363				 vcpu->arch.access, access)) {
4364		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4365					(gva & (PAGE_SIZE - 1));
4366		trace_vcpu_match_mmio(gva, *gpa, write, false);
4367		return 1;
4368	}
4369
4370	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4371
4372	if (*gpa == UNMAPPED_GVA)
4373		return -1;
4374
4375	/* For APIC access vmexit */
4376	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4377		return 1;
4378
4379	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4380		trace_vcpu_match_mmio(gva, *gpa, write, true);
4381		return 1;
4382	}
4383
4384	return 0;
4385}
4386
4387int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4388			const void *val, int bytes)
4389{
4390	int ret;
4391
4392	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4393	if (ret < 0)
4394		return 0;
4395	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4396	return 1;
4397}
4398
4399struct read_write_emulator_ops {
4400	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4401				  int bytes);
4402	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4403				  void *val, int bytes);
4404	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4405			       int bytes, void *val);
4406	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4407				    void *val, int bytes);
4408	bool write;
4409};
4410
4411static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4412{
4413	if (vcpu->mmio_read_completed) {
4414		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4415			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4416		vcpu->mmio_read_completed = 0;
4417		return 1;
4418	}
4419
4420	return 0;
4421}
4422
4423static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4424			void *val, int bytes)
4425{
4426	return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4427}
4428
4429static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4430			 void *val, int bytes)
4431{
4432	return emulator_write_phys(vcpu, gpa, val, bytes);
4433}
4434
4435static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4436{
4437	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4438	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4439}
4440
4441static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4442			  void *val, int bytes)
4443{
4444	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4445	return X86EMUL_IO_NEEDED;
4446}
4447
4448static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4449			   void *val, int bytes)
4450{
4451	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4452
4453	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4454	return X86EMUL_CONTINUE;
4455}
4456
4457static const struct read_write_emulator_ops read_emultor = {
4458	.read_write_prepare = read_prepare,
4459	.read_write_emulate = read_emulate,
4460	.read_write_mmio = vcpu_mmio_read,
4461	.read_write_exit_mmio = read_exit_mmio,
4462};
4463
4464static const struct read_write_emulator_ops write_emultor = {
4465	.read_write_emulate = write_emulate,
4466	.read_write_mmio = write_mmio,
4467	.read_write_exit_mmio = write_exit_mmio,
4468	.write = true,
4469};
4470
4471static int emulator_read_write_onepage(unsigned long addr, void *val,
4472				       unsigned int bytes,
4473				       struct x86_exception *exception,
4474				       struct kvm_vcpu *vcpu,
4475				       const struct read_write_emulator_ops *ops)
4476{
4477	gpa_t gpa;
4478	int handled, ret;
4479	bool write = ops->write;
4480	struct kvm_mmio_fragment *frag;
4481
4482	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4483
4484	if (ret < 0)
4485		return X86EMUL_PROPAGATE_FAULT;
4486
4487	/* For APIC access vmexit */
4488	if (ret)
4489		goto mmio;
4490
4491	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4492		return X86EMUL_CONTINUE;
4493
4494mmio:
4495	/*
4496	 * Is this MMIO handled locally?
4497	 */
4498	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4499	if (handled == bytes)
4500		return X86EMUL_CONTINUE;
4501
4502	gpa += handled;
4503	bytes -= handled;
4504	val += handled;
4505
4506	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4507	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4508	frag->gpa = gpa;
4509	frag->data = val;
4510	frag->len = bytes;
4511	return X86EMUL_CONTINUE;
4512}
4513
4514static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4515			unsigned long addr,
4516			void *val, unsigned int bytes,
4517			struct x86_exception *exception,
4518			const struct read_write_emulator_ops *ops)
4519{
4520	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4521	gpa_t gpa;
4522	int rc;
4523
4524	if (ops->read_write_prepare &&
4525		  ops->read_write_prepare(vcpu, val, bytes))
4526		return X86EMUL_CONTINUE;
4527
4528	vcpu->mmio_nr_fragments = 0;
4529
4530	/* Crossing a page boundary? */
4531	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4532		int now;
4533
4534		now = -addr & ~PAGE_MASK;
4535		rc = emulator_read_write_onepage(addr, val, now, exception,
4536						 vcpu, ops);
4537
4538		if (rc != X86EMUL_CONTINUE)
4539			return rc;
4540		addr += now;
4541		if (ctxt->mode != X86EMUL_MODE_PROT64)
4542			addr = (u32)addr;
4543		val += now;
4544		bytes -= now;
4545	}
4546
4547	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4548					 vcpu, ops);
4549	if (rc != X86EMUL_CONTINUE)
4550		return rc;
4551
4552	if (!vcpu->mmio_nr_fragments)
4553		return rc;
4554
4555	gpa = vcpu->mmio_fragments[0].gpa;
4556
4557	vcpu->mmio_needed = 1;
4558	vcpu->mmio_cur_fragment = 0;
4559
4560	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4561	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4562	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4563	vcpu->run->mmio.phys_addr = gpa;
4564
4565	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4566}
4567
4568static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4569				  unsigned long addr,
4570				  void *val,
4571				  unsigned int bytes,
4572				  struct x86_exception *exception)
4573{
4574	return emulator_read_write(ctxt, addr, val, bytes,
4575				   exception, &read_emultor);
4576}
4577
4578static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4579			    unsigned long addr,
4580			    const void *val,
4581			    unsigned int bytes,
4582			    struct x86_exception *exception)
4583{
4584	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4585				   exception, &write_emultor);
4586}
4587
4588#define CMPXCHG_TYPE(t, ptr, old, new) \
4589	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4590
4591#ifdef CONFIG_X86_64
4592#  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4593#else
4594#  define CMPXCHG64(ptr, old, new) \
4595	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4596#endif
4597
4598static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4599				     unsigned long addr,
4600				     const void *old,
4601				     const void *new,
4602				     unsigned int bytes,
4603				     struct x86_exception *exception)
4604{
4605	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4606	gpa_t gpa;
4607	struct page *page;
4608	char *kaddr;
4609	bool exchanged;
4610
4611	/* guests cmpxchg8b have to be emulated atomically */
4612	if (bytes > 8 || (bytes & (bytes - 1)))
4613		goto emul_write;
4614
4615	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4616
4617	if (gpa == UNMAPPED_GVA ||
4618	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4619		goto emul_write;
4620
4621	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4622		goto emul_write;
4623
4624	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4625	if (is_error_page(page))
4626		goto emul_write;
4627
4628	kaddr = kmap_atomic(page);
4629	kaddr += offset_in_page(gpa);
4630	switch (bytes) {
4631	case 1:
4632		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4633		break;
4634	case 2:
4635		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4636		break;
4637	case 4:
4638		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4639		break;
4640	case 8:
4641		exchanged = CMPXCHG64(kaddr, old, new);
4642		break;
4643	default:
4644		BUG();
4645	}
4646	kunmap_atomic(kaddr);
4647	kvm_release_page_dirty(page);
4648
4649	if (!exchanged)
4650		return X86EMUL_CMPXCHG_FAILED;
4651
4652	mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4653	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4654
4655	return X86EMUL_CONTINUE;
4656
4657emul_write:
4658	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4659
4660	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4661}
4662
4663static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4664{
4665	/* TODO: String I/O for in kernel device */
4666	int r;
4667
4668	if (vcpu->arch.pio.in)
4669		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4670				    vcpu->arch.pio.size, pd);
4671	else
4672		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4673				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4674				     pd);
4675	return r;
4676}
4677
4678static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4679			       unsigned short port, void *val,
4680			       unsigned int count, bool in)
4681{
4682	vcpu->arch.pio.port = port;
4683	vcpu->arch.pio.in = in;
4684	vcpu->arch.pio.count  = count;
4685	vcpu->arch.pio.size = size;
4686
4687	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4688		vcpu->arch.pio.count = 0;
4689		return 1;
4690	}
4691
4692	vcpu->run->exit_reason = KVM_EXIT_IO;
4693	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4694	vcpu->run->io.size = size;
4695	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4696	vcpu->run->io.count = count;
4697	vcpu->run->io.port = port;
4698
4699	return 0;
4700}
4701
4702static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4703				    int size, unsigned short port, void *val,
4704				    unsigned int count)
4705{
4706	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4707	int ret;
4708
4709	if (vcpu->arch.pio.count)
4710		goto data_avail;
4711
4712	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4713	if (ret) {
4714data_avail:
4715		memcpy(val, vcpu->arch.pio_data, size * count);
4716		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4717		vcpu->arch.pio.count = 0;
4718		return 1;
4719	}
4720
4721	return 0;
4722}
4723
4724static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4725				     int size, unsigned short port,
4726				     const void *val, unsigned int count)
4727{
4728	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4729
4730	memcpy(vcpu->arch.pio_data, val, size * count);
4731	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4732	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4733}
4734
4735static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4736{
4737	return kvm_x86_ops->get_segment_base(vcpu, seg);
4738}
4739
4740static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4741{
4742	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4743}
4744
4745int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4746{
4747	if (!need_emulate_wbinvd(vcpu))
4748		return X86EMUL_CONTINUE;
4749
4750	if (kvm_x86_ops->has_wbinvd_exit()) {
4751		int cpu = get_cpu();
4752
4753		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4754		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4755				wbinvd_ipi, NULL, 1);
4756		put_cpu();
4757		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4758	} else
4759		wbinvd();
4760	return X86EMUL_CONTINUE;
4761}
4762
4763int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4764{
4765	kvm_x86_ops->skip_emulated_instruction(vcpu);
4766	return kvm_emulate_wbinvd_noskip(vcpu);
4767}
4768EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4769
4770
4771
4772static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4773{
4774	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4775}
4776
4777static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4778			   unsigned long *dest)
4779{
4780	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4781}
4782
4783static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4784			   unsigned long value)
4785{
4786
4787	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4788}
4789
4790static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4791{
4792	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4793}
4794
4795static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4796{
4797	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4798	unsigned long value;
4799
4800	switch (cr) {
4801	case 0:
4802		value = kvm_read_cr0(vcpu);
4803		break;
4804	case 2:
4805		value = vcpu->arch.cr2;
4806		break;
4807	case 3:
4808		value = kvm_read_cr3(vcpu);
4809		break;
4810	case 4:
4811		value = kvm_read_cr4(vcpu);
4812		break;
4813	case 8:
4814		value = kvm_get_cr8(vcpu);
4815		break;
4816	default:
4817		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4818		return 0;
4819	}
4820
4821	return value;
4822}
4823
4824static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4825{
4826	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4827	int res = 0;
4828
4829	switch (cr) {
4830	case 0:
4831		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4832		break;
4833	case 2:
4834		vcpu->arch.cr2 = val;
4835		break;
4836	case 3:
4837		res = kvm_set_cr3(vcpu, val);
4838		break;
4839	case 4:
4840		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4841		break;
4842	case 8:
4843		res = kvm_set_cr8(vcpu, val);
4844		break;
4845	default:
4846		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4847		res = -1;
4848	}
4849
4850	return res;
4851}
4852
4853static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4854{
4855	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4856}
4857
4858static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4859{
4860	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4861}
4862
4863static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4864{
4865	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4866}
4867
4868static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4869{
4870	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4871}
4872
4873static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4874{
4875	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4876}
4877
4878static unsigned long emulator_get_cached_segment_base(
4879	struct x86_emulate_ctxt *ctxt, int seg)
4880{
4881	return get_segment_base(emul_to_vcpu(ctxt), seg);
4882}
4883
4884static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4885				 struct desc_struct *desc, u32 *base3,
4886				 int seg)
4887{
4888	struct kvm_segment var;
4889
4890	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4891	*selector = var.selector;
4892
4893	if (var.unusable) {
4894		memset(desc, 0, sizeof(*desc));
4895		return false;
4896	}
4897
4898	if (var.g)
4899		var.limit >>= 12;
4900	set_desc_limit(desc, var.limit);
4901	set_desc_base(desc, (unsigned long)var.base);
4902#ifdef CONFIG_X86_64
4903	if (base3)
4904		*base3 = var.base >> 32;
4905#endif
4906	desc->type = var.type;
4907	desc->s = var.s;
4908	desc->dpl = var.dpl;
4909	desc->p = var.present;
4910	desc->avl = var.avl;
4911	desc->l = var.l;
4912	desc->d = var.db;
4913	desc->g = var.g;
4914
4915	return true;
4916}
4917
4918static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4919				 struct desc_struct *desc, u32 base3,
4920				 int seg)
4921{
4922	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4923	struct kvm_segment var;
4924
4925	var.selector = selector;
4926	var.base = get_desc_base(desc);
4927#ifdef CONFIG_X86_64
4928	var.base |= ((u64)base3) << 32;
4929#endif
4930	var.limit = get_desc_limit(desc);
4931	if (desc->g)
4932		var.limit = (var.limit << 12) | 0xfff;
4933	var.type = desc->type;
4934	var.dpl = desc->dpl;
4935	var.db = desc->d;
4936	var.s = desc->s;
4937	var.l = desc->l;
4938	var.g = desc->g;
4939	var.avl = desc->avl;
4940	var.present = desc->p;
4941	var.unusable = !var.present;
4942	var.padding = 0;
4943
4944	kvm_set_segment(vcpu, &var, seg);
4945	return;
4946}
4947
4948static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4949			    u32 msr_index, u64 *pdata)
4950{
4951	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4952}
4953
4954static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4955			    u32 msr_index, u64 data)
4956{
4957	struct msr_data msr;
4958
4959	msr.data = data;
4960	msr.index = msr_index;
4961	msr.host_initiated = false;
4962	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4963}
4964
4965static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4966			      u32 pmc)
4967{
4968	return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4969}
4970
4971static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4972			     u32 pmc, u64 *pdata)
4973{
4974	return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4975}
4976
4977static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4978{
4979	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4980}
4981
4982static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4983{
4984	preempt_disable();
4985	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4986	/*
4987	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4988	 * so it may be clear at this point.
4989	 */
4990	clts();
4991}
4992
4993static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4994{
4995	preempt_enable();
4996}
4997
4998static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4999			      struct x86_instruction_info *info,
5000			      enum x86_intercept_stage stage)
5001{
5002	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5003}
5004
5005static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5006			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5007{
5008	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5009}
5010
5011static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5012{
5013	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5014}
5015
5016static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5017{
5018	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5019}
5020
5021static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5022{
5023	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5024}
5025
5026static const struct x86_emulate_ops emulate_ops = {
5027	.read_gpr            = emulator_read_gpr,
5028	.write_gpr           = emulator_write_gpr,
5029	.read_std            = kvm_read_guest_virt_system,
5030	.write_std           = kvm_write_guest_virt_system,
5031	.fetch               = kvm_fetch_guest_virt,
5032	.read_emulated       = emulator_read_emulated,
5033	.write_emulated      = emulator_write_emulated,
5034	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5035	.invlpg              = emulator_invlpg,
5036	.pio_in_emulated     = emulator_pio_in_emulated,
5037	.pio_out_emulated    = emulator_pio_out_emulated,
5038	.get_segment         = emulator_get_segment,
5039	.set_segment         = emulator_set_segment,
5040	.get_cached_segment_base = emulator_get_cached_segment_base,
5041	.get_gdt             = emulator_get_gdt,
5042	.get_idt	     = emulator_get_idt,
5043	.set_gdt             = emulator_set_gdt,
5044	.set_idt	     = emulator_set_idt,
5045	.get_cr              = emulator_get_cr,
5046	.set_cr              = emulator_set_cr,
5047	.cpl                 = emulator_get_cpl,
5048	.get_dr              = emulator_get_dr,
5049	.set_dr              = emulator_set_dr,
5050	.set_msr             = emulator_set_msr,
5051	.get_msr             = emulator_get_msr,
5052	.check_pmc	     = emulator_check_pmc,
5053	.read_pmc            = emulator_read_pmc,
5054	.halt                = emulator_halt,
5055	.wbinvd              = emulator_wbinvd,
5056	.fix_hypercall       = emulator_fix_hypercall,
5057	.get_fpu             = emulator_get_fpu,
5058	.put_fpu             = emulator_put_fpu,
5059	.intercept           = emulator_intercept,
5060	.get_cpuid           = emulator_get_cpuid,
5061	.set_nmi_mask        = emulator_set_nmi_mask,
5062};
5063
5064static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5065{
5066	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5067	/*
5068	 * an sti; sti; sequence only disable interrupts for the first
5069	 * instruction. So, if the last instruction, be it emulated or
5070	 * not, left the system with the INT_STI flag enabled, it
5071	 * means that the last instruction is an sti. We should not
5072	 * leave the flag on in this case. The same goes for mov ss
5073	 */
5074	if (int_shadow & mask)
5075		mask = 0;
5076	if (unlikely(int_shadow || mask)) {
5077		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5078		if (!mask)
5079			kvm_make_request(KVM_REQ_EVENT, vcpu);
5080	}
5081}
5082
5083static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5084{
5085	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5086	if (ctxt->exception.vector == PF_VECTOR)
5087		return kvm_propagate_fault(vcpu, &ctxt->exception);
5088
5089	if (ctxt->exception.error_code_valid)
5090		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5091				      ctxt->exception.error_code);
5092	else
5093		kvm_queue_exception(vcpu, ctxt->exception.vector);
5094	return false;
5095}
5096
5097static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5098{
5099	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5100	int cs_db, cs_l;
5101
5102	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5103
5104	ctxt->eflags = kvm_get_rflags(vcpu);
5105	ctxt->eip = kvm_rip_read(vcpu);
5106	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5107		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5108		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5109		     cs_db				? X86EMUL_MODE_PROT32 :
5110							  X86EMUL_MODE_PROT16;
5111	ctxt->guest_mode = is_guest_mode(vcpu);
5112
5113	init_decode_cache(ctxt);
5114	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5115}
5116
5117int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5118{
5119	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5120	int ret;
5121
5122	init_emulate_ctxt(vcpu);
5123
5124	ctxt->op_bytes = 2;
5125	ctxt->ad_bytes = 2;
5126	ctxt->_eip = ctxt->eip + inc_eip;
5127	ret = emulate_int_real(ctxt, irq);
5128
5129	if (ret != X86EMUL_CONTINUE)
5130		return EMULATE_FAIL;
5131
5132	ctxt->eip = ctxt->_eip;
5133	kvm_rip_write(vcpu, ctxt->eip);
5134	kvm_set_rflags(vcpu, ctxt->eflags);
5135
5136	if (irq == NMI_VECTOR)
5137		vcpu->arch.nmi_pending = 0;
5138	else
5139		vcpu->arch.interrupt.pending = false;
5140
5141	return EMULATE_DONE;
5142}
5143EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5144
5145static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5146{
5147	int r = EMULATE_DONE;
5148
5149	++vcpu->stat.insn_emulation_fail;
5150	trace_kvm_emulate_insn_failed(vcpu);
5151	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5152		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5153		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5154		vcpu->run->internal.ndata = 0;
5155		r = EMULATE_FAIL;
5156	}
5157	kvm_queue_exception(vcpu, UD_VECTOR);
5158
5159	return r;
5160}
5161
5162static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5163				  bool write_fault_to_shadow_pgtable,
5164				  int emulation_type)
5165{
5166	gpa_t gpa = cr2;
5167	pfn_t pfn;
5168
5169	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5170		return false;
5171
5172	if (!vcpu->arch.mmu.direct_map) {
5173		/*
5174		 * Write permission should be allowed since only
5175		 * write access need to be emulated.
5176		 */
5177		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5178
5179		/*
5180		 * If the mapping is invalid in guest, let cpu retry
5181		 * it to generate fault.
5182		 */
5183		if (gpa == UNMAPPED_GVA)
5184			return true;
5185	}
5186
5187	/*
5188	 * Do not retry the unhandleable instruction if it faults on the
5189	 * readonly host memory, otherwise it will goto a infinite loop:
5190	 * retry instruction -> write #PF -> emulation fail -> retry
5191	 * instruction -> ...
5192	 */
5193	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5194
5195	/*
5196	 * If the instruction failed on the error pfn, it can not be fixed,
5197	 * report the error to userspace.
5198	 */
5199	if (is_error_noslot_pfn(pfn))
5200		return false;
5201
5202	kvm_release_pfn_clean(pfn);
5203
5204	/* The instructions are well-emulated on direct mmu. */
5205	if (vcpu->arch.mmu.direct_map) {
5206		unsigned int indirect_shadow_pages;
5207
5208		spin_lock(&vcpu->kvm->mmu_lock);
5209		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5210		spin_unlock(&vcpu->kvm->mmu_lock);
5211
5212		if (indirect_shadow_pages)
5213			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5214
5215		return true;
5216	}
5217
5218	/*
5219	 * if emulation was due to access to shadowed page table
5220	 * and it failed try to unshadow page and re-enter the
5221	 * guest to let CPU execute the instruction.
5222	 */
5223	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5224
5225	/*
5226	 * If the access faults on its page table, it can not
5227	 * be fixed by unprotecting shadow page and it should
5228	 * be reported to userspace.
5229	 */
5230	return !write_fault_to_shadow_pgtable;
5231}
5232
5233static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5234			      unsigned long cr2,  int emulation_type)
5235{
5236	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5237	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5238
5239	last_retry_eip = vcpu->arch.last_retry_eip;
5240	last_retry_addr = vcpu->arch.last_retry_addr;
5241
5242	/*
5243	 * If the emulation is caused by #PF and it is non-page_table
5244	 * writing instruction, it means the VM-EXIT is caused by shadow
5245	 * page protected, we can zap the shadow page and retry this
5246	 * instruction directly.
5247	 *
5248	 * Note: if the guest uses a non-page-table modifying instruction
5249	 * on the PDE that points to the instruction, then we will unmap
5250	 * the instruction and go to an infinite loop. So, we cache the
5251	 * last retried eip and the last fault address, if we meet the eip
5252	 * and the address again, we can break out of the potential infinite
5253	 * loop.
5254	 */
5255	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5256
5257	if (!(emulation_type & EMULTYPE_RETRY))
5258		return false;
5259
5260	if (x86_page_table_writing_insn(ctxt))
5261		return false;
5262
5263	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5264		return false;
5265
5266	vcpu->arch.last_retry_eip = ctxt->eip;
5267	vcpu->arch.last_retry_addr = cr2;
5268
5269	if (!vcpu->arch.mmu.direct_map)
5270		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5271
5272	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5273
5274	return true;
5275}
5276
5277static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5278static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5279
5280static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5281				unsigned long *db)
5282{
5283	u32 dr6 = 0;
5284	int i;
5285	u32 enable, rwlen;
5286
5287	enable = dr7;
5288	rwlen = dr7 >> 16;
5289	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5290		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5291			dr6 |= (1 << i);
5292	return dr6;
5293}
5294
5295static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5296{
5297	struct kvm_run *kvm_run = vcpu->run;
5298
5299	/*
5300	 * rflags is the old, "raw" value of the flags.  The new value has
5301	 * not been saved yet.
5302	 *
5303	 * This is correct even for TF set by the guest, because "the
5304	 * processor will not generate this exception after the instruction
5305	 * that sets the TF flag".
5306	 */
5307	if (unlikely(rflags & X86_EFLAGS_TF)) {
5308		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5309			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5310						  DR6_RTM;
5311			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5312			kvm_run->debug.arch.exception = DB_VECTOR;
5313			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5314			*r = EMULATE_USER_EXIT;
5315		} else {
5316			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5317			/*
5318			 * "Certain debug exceptions may clear bit 0-3.  The
5319			 * remaining contents of the DR6 register are never
5320			 * cleared by the processor".
5321			 */
5322			vcpu->arch.dr6 &= ~15;
5323			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5324			kvm_queue_exception(vcpu, DB_VECTOR);
5325		}
5326	}
5327}
5328
5329static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5330{
5331	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5332	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5333		struct kvm_run *kvm_run = vcpu->run;
5334		unsigned long eip = kvm_get_linear_rip(vcpu);
5335		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5336					   vcpu->arch.guest_debug_dr7,
5337					   vcpu->arch.eff_db);
5338
5339		if (dr6 != 0) {
5340			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5341			kvm_run->debug.arch.pc = eip;
5342			kvm_run->debug.arch.exception = DB_VECTOR;
5343			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5344			*r = EMULATE_USER_EXIT;
5345			return true;
5346		}
5347	}
5348
5349	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5350	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5351		unsigned long eip = kvm_get_linear_rip(vcpu);
5352		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5353					   vcpu->arch.dr7,
5354					   vcpu->arch.db);
5355
5356		if (dr6 != 0) {
5357			vcpu->arch.dr6 &= ~15;
5358			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5359			kvm_queue_exception(vcpu, DB_VECTOR);
5360			*r = EMULATE_DONE;
5361			return true;
5362		}
5363	}
5364
5365	return false;
5366}
5367
5368int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5369			    unsigned long cr2,
5370			    int emulation_type,
5371			    void *insn,
5372			    int insn_len)
5373{
5374	int r;
5375	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5376	bool writeback = true;
5377	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5378
5379	/*
5380	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5381	 * never reused.
5382	 */
5383	vcpu->arch.write_fault_to_shadow_pgtable = false;
5384	kvm_clear_exception_queue(vcpu);
5385
5386	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5387		init_emulate_ctxt(vcpu);
5388
5389		/*
5390		 * We will reenter on the same instruction since
5391		 * we do not set complete_userspace_io.  This does not
5392		 * handle watchpoints yet, those would be handled in
5393		 * the emulate_ops.
5394		 */
5395		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5396			return r;
5397
5398		ctxt->interruptibility = 0;
5399		ctxt->have_exception = false;
5400		ctxt->exception.vector = -1;
5401		ctxt->perm_ok = false;
5402
5403		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5404
5405		r = x86_decode_insn(ctxt, insn, insn_len);
5406
5407		trace_kvm_emulate_insn_start(vcpu);
5408		++vcpu->stat.insn_emulation;
5409		if (r != EMULATION_OK)  {
5410			if (emulation_type & EMULTYPE_TRAP_UD)
5411				return EMULATE_FAIL;
5412			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5413						emulation_type))
5414				return EMULATE_DONE;
5415			if (emulation_type & EMULTYPE_SKIP)
5416				return EMULATE_FAIL;
5417			return handle_emulation_failure(vcpu);
5418		}
5419	}
5420
5421	if (emulation_type & EMULTYPE_SKIP) {
5422		kvm_rip_write(vcpu, ctxt->_eip);
5423		if (ctxt->eflags & X86_EFLAGS_RF)
5424			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5425		return EMULATE_DONE;
5426	}
5427
5428	if (retry_instruction(ctxt, cr2, emulation_type))
5429		return EMULATE_DONE;
5430
5431	/* this is needed for vmware backdoor interface to work since it
5432	   changes registers values  during IO operation */
5433	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5434		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5435		emulator_invalidate_register_cache(ctxt);
5436	}
5437
5438restart:
5439	r = x86_emulate_insn(ctxt);
5440
5441	if (r == EMULATION_INTERCEPTED)
5442		return EMULATE_DONE;
5443
5444	if (r == EMULATION_FAILED) {
5445		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5446					emulation_type))
5447			return EMULATE_DONE;
5448
5449		return handle_emulation_failure(vcpu);
5450	}
5451
5452	if (ctxt->have_exception) {
5453		r = EMULATE_DONE;
5454		if (inject_emulated_exception(vcpu))
5455			return r;
5456	} else if (vcpu->arch.pio.count) {
5457		if (!vcpu->arch.pio.in) {
5458			/* FIXME: return into emulator if single-stepping.  */
5459			vcpu->arch.pio.count = 0;
5460		} else {
5461			writeback = false;
5462			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5463		}
5464		r = EMULATE_USER_EXIT;
5465	} else if (vcpu->mmio_needed) {
5466		if (!vcpu->mmio_is_write)
5467			writeback = false;
5468		r = EMULATE_USER_EXIT;
5469		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5470	} else if (r == EMULATION_RESTART)
5471		goto restart;
5472	else
5473		r = EMULATE_DONE;
5474
5475	if (writeback) {
5476		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5477		toggle_interruptibility(vcpu, ctxt->interruptibility);
5478		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5479		kvm_rip_write(vcpu, ctxt->eip);
5480		if (r == EMULATE_DONE)
5481			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5482		if (!ctxt->have_exception ||
5483		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5484			__kvm_set_rflags(vcpu, ctxt->eflags);
5485
5486		/*
5487		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5488		 * do nothing, and it will be requested again as soon as
5489		 * the shadow expires.  But we still need to check here,
5490		 * because POPF has no interrupt shadow.
5491		 */
5492		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5493			kvm_make_request(KVM_REQ_EVENT, vcpu);
5494	} else
5495		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5496
5497	return r;
5498}
5499EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5500
5501int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5502{
5503	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5504	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5505					    size, port, &val, 1);
5506	/* do not return to emulator after return from userspace */
5507	vcpu->arch.pio.count = 0;
5508	return ret;
5509}
5510EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5511
5512static void tsc_bad(void *info)
5513{
5514	__this_cpu_write(cpu_tsc_khz, 0);
5515}
5516
5517static void tsc_khz_changed(void *data)
5518{
5519	struct cpufreq_freqs *freq = data;
5520	unsigned long khz = 0;
5521
5522	if (data)
5523		khz = freq->new;
5524	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5525		khz = cpufreq_quick_get(raw_smp_processor_id());
5526	if (!khz)
5527		khz = tsc_khz;
5528	__this_cpu_write(cpu_tsc_khz, khz);
5529}
5530
5531static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5532				     void *data)
5533{
5534	struct cpufreq_freqs *freq = data;
5535	struct kvm *kvm;
5536	struct kvm_vcpu *vcpu;
5537	int i, send_ipi = 0;
5538
5539	/*
5540	 * We allow guests to temporarily run on slowing clocks,
5541	 * provided we notify them after, or to run on accelerating
5542	 * clocks, provided we notify them before.  Thus time never
5543	 * goes backwards.
5544	 *
5545	 * However, we have a problem.  We can't atomically update
5546	 * the frequency of a given CPU from this function; it is
5547	 * merely a notifier, which can be called from any CPU.
5548	 * Changing the TSC frequency at arbitrary points in time
5549	 * requires a recomputation of local variables related to
5550	 * the TSC for each VCPU.  We must flag these local variables
5551	 * to be updated and be sure the update takes place with the
5552	 * new frequency before any guests proceed.
5553	 *
5554	 * Unfortunately, the combination of hotplug CPU and frequency
5555	 * change creates an intractable locking scenario; the order
5556	 * of when these callouts happen is undefined with respect to
5557	 * CPU hotplug, and they can race with each other.  As such,
5558	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5559	 * undefined; you can actually have a CPU frequency change take
5560	 * place in between the computation of X and the setting of the
5561	 * variable.  To protect against this problem, all updates of
5562	 * the per_cpu tsc_khz variable are done in an interrupt
5563	 * protected IPI, and all callers wishing to update the value
5564	 * must wait for a synchronous IPI to complete (which is trivial
5565	 * if the caller is on the CPU already).  This establishes the
5566	 * necessary total order on variable updates.
5567	 *
5568	 * Note that because a guest time update may take place
5569	 * anytime after the setting of the VCPU's request bit, the
5570	 * correct TSC value must be set before the request.  However,
5571	 * to ensure the update actually makes it to any guest which
5572	 * starts running in hardware virtualization between the set
5573	 * and the acquisition of the spinlock, we must also ping the
5574	 * CPU after setting the request bit.
5575	 *
5576	 */
5577
5578	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5579		return 0;
5580	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5581		return 0;
5582
5583	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5584
5585	spin_lock(&kvm_lock);
5586	list_for_each_entry(kvm, &vm_list, vm_list) {
5587		kvm_for_each_vcpu(i, vcpu, kvm) {
5588			if (vcpu->cpu != freq->cpu)
5589				continue;
5590			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5591			if (vcpu->cpu != smp_processor_id())
5592				send_ipi = 1;
5593		}
5594	}
5595	spin_unlock(&kvm_lock);
5596
5597	if (freq->old < freq->new && send_ipi) {
5598		/*
5599		 * We upscale the frequency.  Must make the guest
5600		 * doesn't see old kvmclock values while running with
5601		 * the new frequency, otherwise we risk the guest sees
5602		 * time go backwards.
5603		 *
5604		 * In case we update the frequency for another cpu
5605		 * (which might be in guest context) send an interrupt
5606		 * to kick the cpu out of guest context.  Next time
5607		 * guest context is entered kvmclock will be updated,
5608		 * so the guest will not see stale values.
5609		 */
5610		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5611	}
5612	return 0;
5613}
5614
5615static struct notifier_block kvmclock_cpufreq_notifier_block = {
5616	.notifier_call  = kvmclock_cpufreq_notifier
5617};
5618
5619static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5620					unsigned long action, void *hcpu)
5621{
5622	unsigned int cpu = (unsigned long)hcpu;
5623
5624	switch (action) {
5625		case CPU_ONLINE:
5626		case CPU_DOWN_FAILED:
5627			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5628			break;
5629		case CPU_DOWN_PREPARE:
5630			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5631			break;
5632	}
5633	return NOTIFY_OK;
5634}
5635
5636static struct notifier_block kvmclock_cpu_notifier_block = {
5637	.notifier_call  = kvmclock_cpu_notifier,
5638	.priority = -INT_MAX
5639};
5640
5641static void kvm_timer_init(void)
5642{
5643	int cpu;
5644
5645	max_tsc_khz = tsc_khz;
5646
5647	cpu_notifier_register_begin();
5648	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5649#ifdef CONFIG_CPU_FREQ
5650		struct cpufreq_policy policy;
5651		memset(&policy, 0, sizeof(policy));
5652		cpu = get_cpu();
5653		cpufreq_get_policy(&policy, cpu);
5654		if (policy.cpuinfo.max_freq)
5655			max_tsc_khz = policy.cpuinfo.max_freq;
5656		put_cpu();
5657#endif
5658		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5659					  CPUFREQ_TRANSITION_NOTIFIER);
5660	}
5661	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5662	for_each_online_cpu(cpu)
5663		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5664
5665	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5666	cpu_notifier_register_done();
5667
5668}
5669
5670static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5671
5672int kvm_is_in_guest(void)
5673{
5674	return __this_cpu_read(current_vcpu) != NULL;
5675}
5676
5677static int kvm_is_user_mode(void)
5678{
5679	int user_mode = 3;
5680
5681	if (__this_cpu_read(current_vcpu))
5682		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5683
5684	return user_mode != 0;
5685}
5686
5687static unsigned long kvm_get_guest_ip(void)
5688{
5689	unsigned long ip = 0;
5690
5691	if (__this_cpu_read(current_vcpu))
5692		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5693
5694	return ip;
5695}
5696
5697static struct perf_guest_info_callbacks kvm_guest_cbs = {
5698	.is_in_guest		= kvm_is_in_guest,
5699	.is_user_mode		= kvm_is_user_mode,
5700	.get_guest_ip		= kvm_get_guest_ip,
5701};
5702
5703void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5704{
5705	__this_cpu_write(current_vcpu, vcpu);
5706}
5707EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5708
5709void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5710{
5711	__this_cpu_write(current_vcpu, NULL);
5712}
5713EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5714
5715static void kvm_set_mmio_spte_mask(void)
5716{
5717	u64 mask;
5718	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5719
5720	/*
5721	 * Set the reserved bits and the present bit of an paging-structure
5722	 * entry to generate page fault with PFER.RSV = 1.
5723	 */
5724	 /* Mask the reserved physical address bits. */
5725	mask = rsvd_bits(maxphyaddr, 51);
5726
5727	/* Bit 62 is always reserved for 32bit host. */
5728	mask |= 0x3ull << 62;
5729
5730	/* Set the present bit. */
5731	mask |= 1ull;
5732
5733#ifdef CONFIG_X86_64
5734	/*
5735	 * If reserved bit is not supported, clear the present bit to disable
5736	 * mmio page fault.
5737	 */
5738	if (maxphyaddr == 52)
5739		mask &= ~1ull;
5740#endif
5741
5742	kvm_mmu_set_mmio_spte_mask(mask);
5743}
5744
5745#ifdef CONFIG_X86_64
5746static void pvclock_gtod_update_fn(struct work_struct *work)
5747{
5748	struct kvm *kvm;
5749
5750	struct kvm_vcpu *vcpu;
5751	int i;
5752
5753	spin_lock(&kvm_lock);
5754	list_for_each_entry(kvm, &vm_list, vm_list)
5755		kvm_for_each_vcpu(i, vcpu, kvm)
5756			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5757	atomic_set(&kvm_guest_has_master_clock, 0);
5758	spin_unlock(&kvm_lock);
5759}
5760
5761static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5762
5763/*
5764 * Notification about pvclock gtod data update.
5765 */
5766static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5767			       void *priv)
5768{
5769	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5770	struct timekeeper *tk = priv;
5771
5772	update_pvclock_gtod(tk);
5773
5774	/* disable master clock if host does not trust, or does not
5775	 * use, TSC clocksource
5776	 */
5777	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5778	    atomic_read(&kvm_guest_has_master_clock) != 0)
5779		queue_work(system_long_wq, &pvclock_gtod_work);
5780
5781	return 0;
5782}
5783
5784static struct notifier_block pvclock_gtod_notifier = {
5785	.notifier_call = pvclock_gtod_notify,
5786};
5787#endif
5788
5789int kvm_arch_init(void *opaque)
5790{
5791	int r;
5792	struct kvm_x86_ops *ops = opaque;
5793
5794	if (kvm_x86_ops) {
5795		printk(KERN_ERR "kvm: already loaded the other module\n");
5796		r = -EEXIST;
5797		goto out;
5798	}
5799
5800	if (!ops->cpu_has_kvm_support()) {
5801		printk(KERN_ERR "kvm: no hardware support\n");
5802		r = -EOPNOTSUPP;
5803		goto out;
5804	}
5805	if (ops->disabled_by_bios()) {
5806		printk(KERN_ERR "kvm: disabled by bios\n");
5807		r = -EOPNOTSUPP;
5808		goto out;
5809	}
5810
5811	r = -ENOMEM;
5812	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5813	if (!shared_msrs) {
5814		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5815		goto out;
5816	}
5817
5818	r = kvm_mmu_module_init();
5819	if (r)
5820		goto out_free_percpu;
5821
5822	kvm_set_mmio_spte_mask();
5823
5824	kvm_x86_ops = ops;
5825
5826	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5827			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5828
5829	kvm_timer_init();
5830
5831	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5832
5833	if (cpu_has_xsave)
5834		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5835
5836	kvm_lapic_init();
5837#ifdef CONFIG_X86_64
5838	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5839#endif
5840
5841	return 0;
5842
5843out_free_percpu:
5844	free_percpu(shared_msrs);
5845out:
5846	return r;
5847}
5848
5849void kvm_arch_exit(void)
5850{
5851	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5852
5853	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5854		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5855					    CPUFREQ_TRANSITION_NOTIFIER);
5856	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5857#ifdef CONFIG_X86_64
5858	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5859#endif
5860	kvm_x86_ops = NULL;
5861	kvm_mmu_module_exit();
5862	free_percpu(shared_msrs);
5863}
5864
5865int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5866{
5867	++vcpu->stat.halt_exits;
5868	if (irqchip_in_kernel(vcpu->kvm)) {
5869		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5870		return 1;
5871	} else {
5872		vcpu->run->exit_reason = KVM_EXIT_HLT;
5873		return 0;
5874	}
5875}
5876EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5877
5878int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5879{
5880	kvm_x86_ops->skip_emulated_instruction(vcpu);
5881	return kvm_vcpu_halt(vcpu);
5882}
5883EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5884
5885int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5886{
5887	u64 param, ingpa, outgpa, ret;
5888	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5889	bool fast, longmode;
5890
5891	/*
5892	 * hypercall generates UD from non zero cpl and real mode
5893	 * per HYPER-V spec
5894	 */
5895	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5896		kvm_queue_exception(vcpu, UD_VECTOR);
5897		return 0;
5898	}
5899
5900	longmode = is_64_bit_mode(vcpu);
5901
5902	if (!longmode) {
5903		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5904			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5905		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5906			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5907		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5908			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5909	}
5910#ifdef CONFIG_X86_64
5911	else {
5912		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5913		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5914		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5915	}
5916#endif
5917
5918	code = param & 0xffff;
5919	fast = (param >> 16) & 0x1;
5920	rep_cnt = (param >> 32) & 0xfff;
5921	rep_idx = (param >> 48) & 0xfff;
5922
5923	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5924
5925	switch (code) {
5926	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5927		kvm_vcpu_on_spin(vcpu);
5928		break;
5929	default:
5930		res = HV_STATUS_INVALID_HYPERCALL_CODE;
5931		break;
5932	}
5933
5934	ret = res | (((u64)rep_done & 0xfff) << 32);
5935	if (longmode) {
5936		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5937	} else {
5938		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5939		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5940	}
5941
5942	return 1;
5943}
5944
5945/*
5946 * kvm_pv_kick_cpu_op:  Kick a vcpu.
5947 *
5948 * @apicid - apicid of vcpu to be kicked.
5949 */
5950static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5951{
5952	struct kvm_lapic_irq lapic_irq;
5953
5954	lapic_irq.shorthand = 0;
5955	lapic_irq.dest_mode = 0;
5956	lapic_irq.dest_id = apicid;
5957
5958	lapic_irq.delivery_mode = APIC_DM_REMRD;
5959	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5960}
5961
5962int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5963{
5964	unsigned long nr, a0, a1, a2, a3, ret;
5965	int op_64_bit, r = 1;
5966
5967	kvm_x86_ops->skip_emulated_instruction(vcpu);
5968
5969	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5970		return kvm_hv_hypercall(vcpu);
5971
5972	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5973	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5974	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5975	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5976	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5977
5978	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5979
5980	op_64_bit = is_64_bit_mode(vcpu);
5981	if (!op_64_bit) {
5982		nr &= 0xFFFFFFFF;
5983		a0 &= 0xFFFFFFFF;
5984		a1 &= 0xFFFFFFFF;
5985		a2 &= 0xFFFFFFFF;
5986		a3 &= 0xFFFFFFFF;
5987	}
5988
5989	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5990		ret = -KVM_EPERM;
5991		goto out;
5992	}
5993
5994	switch (nr) {
5995	case KVM_HC_VAPIC_POLL_IRQ:
5996		ret = 0;
5997		break;
5998	case KVM_HC_KICK_CPU:
5999		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6000		ret = 0;
6001		break;
6002	default:
6003		ret = -KVM_ENOSYS;
6004		break;
6005	}
6006out:
6007	if (!op_64_bit)
6008		ret = (u32)ret;
6009	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6010	++vcpu->stat.hypercalls;
6011	return r;
6012}
6013EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6014
6015static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6016{
6017	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6018	char instruction[3];
6019	unsigned long rip = kvm_rip_read(vcpu);
6020
6021	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6022
6023	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6024}
6025
6026/*
6027 * Check if userspace requested an interrupt window, and that the
6028 * interrupt window is open.
6029 *
6030 * No need to exit to userspace if we already have an interrupt queued.
6031 */
6032static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6033{
6034	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
6035		vcpu->run->request_interrupt_window &&
6036		kvm_arch_interrupt_allowed(vcpu));
6037}
6038
6039static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6040{
6041	struct kvm_run *kvm_run = vcpu->run;
6042
6043	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6044	kvm_run->cr8 = kvm_get_cr8(vcpu);
6045	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6046	if (irqchip_in_kernel(vcpu->kvm))
6047		kvm_run->ready_for_interrupt_injection = 1;
6048	else
6049		kvm_run->ready_for_interrupt_injection =
6050			kvm_arch_interrupt_allowed(vcpu) &&
6051			!kvm_cpu_has_interrupt(vcpu) &&
6052			!kvm_event_needs_reinjection(vcpu);
6053}
6054
6055static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6056{
6057	int max_irr, tpr;
6058
6059	if (!kvm_x86_ops->update_cr8_intercept)
6060		return;
6061
6062	if (!vcpu->arch.apic)
6063		return;
6064
6065	if (!vcpu->arch.apic->vapic_addr)
6066		max_irr = kvm_lapic_find_highest_irr(vcpu);
6067	else
6068		max_irr = -1;
6069
6070	if (max_irr != -1)
6071		max_irr >>= 4;
6072
6073	tpr = kvm_lapic_get_cr8(vcpu);
6074
6075	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6076}
6077
6078static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6079{
6080	int r;
6081
6082	/* try to reinject previous events if any */
6083	if (vcpu->arch.exception.pending) {
6084		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6085					vcpu->arch.exception.has_error_code,
6086					vcpu->arch.exception.error_code);
6087
6088		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6089			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6090					     X86_EFLAGS_RF);
6091
6092		if (vcpu->arch.exception.nr == DB_VECTOR &&
6093		    (vcpu->arch.dr7 & DR7_GD)) {
6094			vcpu->arch.dr7 &= ~DR7_GD;
6095			kvm_update_dr7(vcpu);
6096		}
6097
6098		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6099					  vcpu->arch.exception.has_error_code,
6100					  vcpu->arch.exception.error_code,
6101					  vcpu->arch.exception.reinject);
6102		return 0;
6103	}
6104
6105	if (vcpu->arch.nmi_injected) {
6106		kvm_x86_ops->set_nmi(vcpu);
6107		return 0;
6108	}
6109
6110	if (vcpu->arch.interrupt.pending) {
6111		kvm_x86_ops->set_irq(vcpu);
6112		return 0;
6113	}
6114
6115	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6116		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6117		if (r != 0)
6118			return r;
6119	}
6120
6121	/* try to inject new event if pending */
6122	if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6123		--vcpu->arch.nmi_pending;
6124		vcpu->arch.nmi_injected = true;
6125		kvm_x86_ops->set_nmi(vcpu);
6126	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6127		/*
6128		 * Because interrupts can be injected asynchronously, we are
6129		 * calling check_nested_events again here to avoid a race condition.
6130		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6131		 * proposal and current concerns.  Perhaps we should be setting
6132		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6133		 */
6134		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6135			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6136			if (r != 0)
6137				return r;
6138		}
6139		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6140			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6141					    false);
6142			kvm_x86_ops->set_irq(vcpu);
6143		}
6144	}
6145	return 0;
6146}
6147
6148static void process_nmi(struct kvm_vcpu *vcpu)
6149{
6150	unsigned limit = 2;
6151
6152	/*
6153	 * x86 is limited to one NMI running, and one NMI pending after it.
6154	 * If an NMI is already in progress, limit further NMIs to just one.
6155	 * Otherwise, allow two (and we'll inject the first one immediately).
6156	 */
6157	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6158		limit = 1;
6159
6160	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6161	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6162	kvm_make_request(KVM_REQ_EVENT, vcpu);
6163}
6164
6165static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6166{
6167	u64 eoi_exit_bitmap[4];
6168	u32 tmr[8];
6169
6170	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6171		return;
6172
6173	memset(eoi_exit_bitmap, 0, 32);
6174	memset(tmr, 0, 32);
6175
6176	kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6177	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6178	kvm_apic_update_tmr(vcpu, tmr);
6179}
6180
6181static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6182{
6183	++vcpu->stat.tlb_flush;
6184	kvm_x86_ops->tlb_flush(vcpu);
6185}
6186
6187void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6188{
6189	struct page *page = NULL;
6190
6191	if (!irqchip_in_kernel(vcpu->kvm))
6192		return;
6193
6194	if (!kvm_x86_ops->set_apic_access_page_addr)
6195		return;
6196
6197	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6198	if (is_error_page(page))
6199		return;
6200	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6201
6202	/*
6203	 * Do not pin apic access page in memory, the MMU notifier
6204	 * will call us again if it is migrated or swapped out.
6205	 */
6206	put_page(page);
6207}
6208EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6209
6210void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6211					   unsigned long address)
6212{
6213	/*
6214	 * The physical address of apic access page is stored in the VMCS.
6215	 * Update it when it becomes invalid.
6216	 */
6217	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6218		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6219}
6220
6221/*
6222 * Returns 1 to let vcpu_run() continue the guest execution loop without
6223 * exiting to the userspace.  Otherwise, the value will be returned to the
6224 * userspace.
6225 */
6226static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6227{
6228	int r;
6229	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6230		vcpu->run->request_interrupt_window;
6231	bool req_immediate_exit = false;
6232
6233	if (vcpu->requests) {
6234		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6235			kvm_mmu_unload(vcpu);
6236		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6237			__kvm_migrate_timers(vcpu);
6238		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6239			kvm_gen_update_masterclock(vcpu->kvm);
6240		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6241			kvm_gen_kvmclock_update(vcpu);
6242		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6243			r = kvm_guest_time_update(vcpu);
6244			if (unlikely(r))
6245				goto out;
6246		}
6247		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6248			kvm_mmu_sync_roots(vcpu);
6249		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6250			kvm_vcpu_flush_tlb(vcpu);
6251		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6252			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6253			r = 0;
6254			goto out;
6255		}
6256		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6257			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6258			r = 0;
6259			goto out;
6260		}
6261		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6262			vcpu->fpu_active = 0;
6263			kvm_x86_ops->fpu_deactivate(vcpu);
6264		}
6265		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6266			/* Page is swapped out. Do synthetic halt */
6267			vcpu->arch.apf.halted = true;
6268			r = 1;
6269			goto out;
6270		}
6271		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6272			record_steal_time(vcpu);
6273		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6274			process_nmi(vcpu);
6275		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6276			kvm_handle_pmu_event(vcpu);
6277		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6278			kvm_deliver_pmi(vcpu);
6279		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6280			vcpu_scan_ioapic(vcpu);
6281		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6282			kvm_vcpu_reload_apic_access_page(vcpu);
6283	}
6284
6285	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6286		kvm_apic_accept_events(vcpu);
6287		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6288			r = 1;
6289			goto out;
6290		}
6291
6292		if (inject_pending_event(vcpu, req_int_win) != 0)
6293			req_immediate_exit = true;
6294		/* enable NMI/IRQ window open exits if needed */
6295		else {
6296			if (vcpu->arch.nmi_pending)
6297				kvm_x86_ops->enable_nmi_window(vcpu);
6298			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6299				kvm_x86_ops->enable_irq_window(vcpu);
6300		}
6301
6302		if (kvm_lapic_enabled(vcpu)) {
6303			/*
6304			 * Update architecture specific hints for APIC
6305			 * virtual interrupt delivery.
6306			 */
6307			if (kvm_x86_ops->hwapic_irr_update)
6308				kvm_x86_ops->hwapic_irr_update(vcpu,
6309					kvm_lapic_find_highest_irr(vcpu));
6310			update_cr8_intercept(vcpu);
6311			kvm_lapic_sync_to_vapic(vcpu);
6312		}
6313	}
6314
6315	r = kvm_mmu_reload(vcpu);
6316	if (unlikely(r)) {
6317		goto cancel_injection;
6318	}
6319
6320	preempt_disable();
6321
6322	kvm_x86_ops->prepare_guest_switch(vcpu);
6323	if (vcpu->fpu_active)
6324		kvm_load_guest_fpu(vcpu);
6325	vcpu->mode = IN_GUEST_MODE;
6326
6327	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6328
6329	/* We should set ->mode before check ->requests,
6330	 * see the comment in make_all_cpus_request.
6331	 */
6332	smp_mb__after_srcu_read_unlock();
6333
6334	local_irq_disable();
6335
6336	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6337	    || need_resched() || signal_pending(current)) {
6338		vcpu->mode = OUTSIDE_GUEST_MODE;
6339		smp_wmb();
6340		local_irq_enable();
6341		preempt_enable();
6342		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6343		r = 1;
6344		goto cancel_injection;
6345	}
6346
6347	kvm_load_guest_xcr0(vcpu);
6348
6349	if (req_immediate_exit)
6350		smp_send_reschedule(vcpu->cpu);
6351
6352	kvm_guest_enter();
6353
6354	if (unlikely(vcpu->arch.switch_db_regs)) {
6355		set_debugreg(0, 7);
6356		set_debugreg(vcpu->arch.eff_db[0], 0);
6357		set_debugreg(vcpu->arch.eff_db[1], 1);
6358		set_debugreg(vcpu->arch.eff_db[2], 2);
6359		set_debugreg(vcpu->arch.eff_db[3], 3);
6360		set_debugreg(vcpu->arch.dr6, 6);
6361		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6362	}
6363
6364	trace_kvm_entry(vcpu->vcpu_id);
6365	wait_lapic_expire(vcpu);
6366	kvm_x86_ops->run(vcpu);
6367
6368	/*
6369	 * Do this here before restoring debug registers on the host.  And
6370	 * since we do this before handling the vmexit, a DR access vmexit
6371	 * can (a) read the correct value of the debug registers, (b) set
6372	 * KVM_DEBUGREG_WONT_EXIT again.
6373	 */
6374	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6375		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6376		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6377		kvm_update_dr0123(vcpu);
6378		kvm_update_dr6(vcpu);
6379		kvm_update_dr7(vcpu);
6380		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6381	}
6382
6383	/*
6384	 * If the guest has used debug registers, at least dr7
6385	 * will be disabled while returning to the host.
6386	 * If we don't have active breakpoints in the host, we don't
6387	 * care about the messed up debug address registers. But if
6388	 * we have some of them active, restore the old state.
6389	 */
6390	if (hw_breakpoint_active())
6391		hw_breakpoint_restore();
6392
6393	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6394							   native_read_tsc());
6395
6396	vcpu->mode = OUTSIDE_GUEST_MODE;
6397	smp_wmb();
6398
6399	kvm_put_guest_xcr0(vcpu);
6400
6401	/* Interrupt is enabled by handle_external_intr() */
6402	kvm_x86_ops->handle_external_intr(vcpu);
6403
6404	++vcpu->stat.exits;
6405
6406	/*
6407	 * We must have an instruction between local_irq_enable() and
6408	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6409	 * the interrupt shadow.  The stat.exits increment will do nicely.
6410	 * But we need to prevent reordering, hence this barrier():
6411	 */
6412	barrier();
6413
6414	kvm_guest_exit();
6415
6416	preempt_enable();
6417
6418	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6419
6420	/*
6421	 * Profile KVM exit RIPs:
6422	 */
6423	if (unlikely(prof_on == KVM_PROFILING)) {
6424		unsigned long rip = kvm_rip_read(vcpu);
6425		profile_hit(KVM_PROFILING, (void *)rip);
6426	}
6427
6428	if (unlikely(vcpu->arch.tsc_always_catchup))
6429		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6430
6431	if (vcpu->arch.apic_attention)
6432		kvm_lapic_sync_from_vapic(vcpu);
6433
6434	r = kvm_x86_ops->handle_exit(vcpu);
6435	return r;
6436
6437cancel_injection:
6438	kvm_x86_ops->cancel_injection(vcpu);
6439	if (unlikely(vcpu->arch.apic_attention))
6440		kvm_lapic_sync_from_vapic(vcpu);
6441out:
6442	return r;
6443}
6444
6445static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6446{
6447	if (!kvm_arch_vcpu_runnable(vcpu)) {
6448		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6449		kvm_vcpu_block(vcpu);
6450		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6451		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6452			return 1;
6453	}
6454
6455	kvm_apic_accept_events(vcpu);
6456	switch(vcpu->arch.mp_state) {
6457	case KVM_MP_STATE_HALTED:
6458		vcpu->arch.pv.pv_unhalted = false;
6459		vcpu->arch.mp_state =
6460			KVM_MP_STATE_RUNNABLE;
6461	case KVM_MP_STATE_RUNNABLE:
6462		vcpu->arch.apf.halted = false;
6463		break;
6464	case KVM_MP_STATE_INIT_RECEIVED:
6465		break;
6466	default:
6467		return -EINTR;
6468		break;
6469	}
6470	return 1;
6471}
6472
6473static int vcpu_run(struct kvm_vcpu *vcpu)
6474{
6475	int r;
6476	struct kvm *kvm = vcpu->kvm;
6477
6478	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6479
6480	for (;;) {
6481		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6482		    !vcpu->arch.apf.halted)
6483			r = vcpu_enter_guest(vcpu);
6484		else
6485			r = vcpu_block(kvm, vcpu);
6486		if (r <= 0)
6487			break;
6488
6489		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6490		if (kvm_cpu_has_pending_timer(vcpu))
6491			kvm_inject_pending_timer_irqs(vcpu);
6492
6493		if (dm_request_for_irq_injection(vcpu)) {
6494			r = -EINTR;
6495			vcpu->run->exit_reason = KVM_EXIT_INTR;
6496			++vcpu->stat.request_irq_exits;
6497			break;
6498		}
6499
6500		kvm_check_async_pf_completion(vcpu);
6501
6502		if (signal_pending(current)) {
6503			r = -EINTR;
6504			vcpu->run->exit_reason = KVM_EXIT_INTR;
6505			++vcpu->stat.signal_exits;
6506			break;
6507		}
6508		if (need_resched()) {
6509			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6510			cond_resched();
6511			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6512		}
6513	}
6514
6515	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6516
6517	return r;
6518}
6519
6520static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6521{
6522	int r;
6523	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6524	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6525	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6526	if (r != EMULATE_DONE)
6527		return 0;
6528	return 1;
6529}
6530
6531static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6532{
6533	BUG_ON(!vcpu->arch.pio.count);
6534
6535	return complete_emulated_io(vcpu);
6536}
6537
6538/*
6539 * Implements the following, as a state machine:
6540 *
6541 * read:
6542 *   for each fragment
6543 *     for each mmio piece in the fragment
6544 *       write gpa, len
6545 *       exit
6546 *       copy data
6547 *   execute insn
6548 *
6549 * write:
6550 *   for each fragment
6551 *     for each mmio piece in the fragment
6552 *       write gpa, len
6553 *       copy data
6554 *       exit
6555 */
6556static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6557{
6558	struct kvm_run *run = vcpu->run;
6559	struct kvm_mmio_fragment *frag;
6560	unsigned len;
6561
6562	BUG_ON(!vcpu->mmio_needed);
6563
6564	/* Complete previous fragment */
6565	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6566	len = min(8u, frag->len);
6567	if (!vcpu->mmio_is_write)
6568		memcpy(frag->data, run->mmio.data, len);
6569
6570	if (frag->len <= 8) {
6571		/* Switch to the next fragment. */
6572		frag++;
6573		vcpu->mmio_cur_fragment++;
6574	} else {
6575		/* Go forward to the next mmio piece. */
6576		frag->data += len;
6577		frag->gpa += len;
6578		frag->len -= len;
6579	}
6580
6581	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6582		vcpu->mmio_needed = 0;
6583
6584		/* FIXME: return into emulator if single-stepping.  */
6585		if (vcpu->mmio_is_write)
6586			return 1;
6587		vcpu->mmio_read_completed = 1;
6588		return complete_emulated_io(vcpu);
6589	}
6590
6591	run->exit_reason = KVM_EXIT_MMIO;
6592	run->mmio.phys_addr = frag->gpa;
6593	if (vcpu->mmio_is_write)
6594		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6595	run->mmio.len = min(8u, frag->len);
6596	run->mmio.is_write = vcpu->mmio_is_write;
6597	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6598	return 0;
6599}
6600
6601
6602int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6603{
6604	int r;
6605	sigset_t sigsaved;
6606
6607	if (!tsk_used_math(current) && init_fpu(current))
6608		return -ENOMEM;
6609
6610	if (vcpu->sigset_active)
6611		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6612
6613	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6614		kvm_vcpu_block(vcpu);
6615		kvm_apic_accept_events(vcpu);
6616		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6617		r = -EAGAIN;
6618		goto out;
6619	}
6620
6621	/* re-sync apic's tpr */
6622	if (!irqchip_in_kernel(vcpu->kvm)) {
6623		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6624			r = -EINVAL;
6625			goto out;
6626		}
6627	}
6628
6629	if (unlikely(vcpu->arch.complete_userspace_io)) {
6630		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6631		vcpu->arch.complete_userspace_io = NULL;
6632		r = cui(vcpu);
6633		if (r <= 0)
6634			goto out;
6635	} else
6636		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6637
6638	r = vcpu_run(vcpu);
6639
6640out:
6641	post_kvm_run_save(vcpu);
6642	if (vcpu->sigset_active)
6643		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6644
6645	return r;
6646}
6647
6648int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6649{
6650	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6651		/*
6652		 * We are here if userspace calls get_regs() in the middle of
6653		 * instruction emulation. Registers state needs to be copied
6654		 * back from emulation context to vcpu. Userspace shouldn't do
6655		 * that usually, but some bad designed PV devices (vmware
6656		 * backdoor interface) need this to work
6657		 */
6658		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6659		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6660	}
6661	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6662	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6663	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6664	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6665	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6666	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6667	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6668	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6669#ifdef CONFIG_X86_64
6670	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6671	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6672	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6673	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6674	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6675	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6676	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6677	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6678#endif
6679
6680	regs->rip = kvm_rip_read(vcpu);
6681	regs->rflags = kvm_get_rflags(vcpu);
6682
6683	return 0;
6684}
6685
6686int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6687{
6688	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6689	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6690
6691	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6692	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6693	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6694	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6695	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6696	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6697	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6698	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6699#ifdef CONFIG_X86_64
6700	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6701	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6702	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6703	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6704	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6705	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6706	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6707	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6708#endif
6709
6710	kvm_rip_write(vcpu, regs->rip);
6711	kvm_set_rflags(vcpu, regs->rflags);
6712
6713	vcpu->arch.exception.pending = false;
6714
6715	kvm_make_request(KVM_REQ_EVENT, vcpu);
6716
6717	return 0;
6718}
6719
6720void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6721{
6722	struct kvm_segment cs;
6723
6724	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6725	*db = cs.db;
6726	*l = cs.l;
6727}
6728EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6729
6730int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6731				  struct kvm_sregs *sregs)
6732{
6733	struct desc_ptr dt;
6734
6735	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6736	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6737	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6738	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6739	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6740	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6741
6742	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6743	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6744
6745	kvm_x86_ops->get_idt(vcpu, &dt);
6746	sregs->idt.limit = dt.size;
6747	sregs->idt.base = dt.address;
6748	kvm_x86_ops->get_gdt(vcpu, &dt);
6749	sregs->gdt.limit = dt.size;
6750	sregs->gdt.base = dt.address;
6751
6752	sregs->cr0 = kvm_read_cr0(vcpu);
6753	sregs->cr2 = vcpu->arch.cr2;
6754	sregs->cr3 = kvm_read_cr3(vcpu);
6755	sregs->cr4 = kvm_read_cr4(vcpu);
6756	sregs->cr8 = kvm_get_cr8(vcpu);
6757	sregs->efer = vcpu->arch.efer;
6758	sregs->apic_base = kvm_get_apic_base(vcpu);
6759
6760	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6761
6762	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6763		set_bit(vcpu->arch.interrupt.nr,
6764			(unsigned long *)sregs->interrupt_bitmap);
6765
6766	return 0;
6767}
6768
6769int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6770				    struct kvm_mp_state *mp_state)
6771{
6772	kvm_apic_accept_events(vcpu);
6773	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6774					vcpu->arch.pv.pv_unhalted)
6775		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6776	else
6777		mp_state->mp_state = vcpu->arch.mp_state;
6778
6779	return 0;
6780}
6781
6782int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6783				    struct kvm_mp_state *mp_state)
6784{
6785	if (!kvm_vcpu_has_lapic(vcpu) &&
6786	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6787		return -EINVAL;
6788
6789	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6790		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6791		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6792	} else
6793		vcpu->arch.mp_state = mp_state->mp_state;
6794	kvm_make_request(KVM_REQ_EVENT, vcpu);
6795	return 0;
6796}
6797
6798int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6799		    int reason, bool has_error_code, u32 error_code)
6800{
6801	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6802	int ret;
6803
6804	init_emulate_ctxt(vcpu);
6805
6806	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6807				   has_error_code, error_code);
6808
6809	if (ret)
6810		return EMULATE_FAIL;
6811
6812	kvm_rip_write(vcpu, ctxt->eip);
6813	kvm_set_rflags(vcpu, ctxt->eflags);
6814	kvm_make_request(KVM_REQ_EVENT, vcpu);
6815	return EMULATE_DONE;
6816}
6817EXPORT_SYMBOL_GPL(kvm_task_switch);
6818
6819int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6820				  struct kvm_sregs *sregs)
6821{
6822	struct msr_data apic_base_msr;
6823	int mmu_reset_needed = 0;
6824	int pending_vec, max_bits, idx;
6825	struct desc_ptr dt;
6826
6827	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6828		return -EINVAL;
6829
6830	dt.size = sregs->idt.limit;
6831	dt.address = sregs->idt.base;
6832	kvm_x86_ops->set_idt(vcpu, &dt);
6833	dt.size = sregs->gdt.limit;
6834	dt.address = sregs->gdt.base;
6835	kvm_x86_ops->set_gdt(vcpu, &dt);
6836
6837	vcpu->arch.cr2 = sregs->cr2;
6838	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6839	vcpu->arch.cr3 = sregs->cr3;
6840	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6841
6842	kvm_set_cr8(vcpu, sregs->cr8);
6843
6844	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6845	kvm_x86_ops->set_efer(vcpu, sregs->efer);
6846	apic_base_msr.data = sregs->apic_base;
6847	apic_base_msr.host_initiated = true;
6848	kvm_set_apic_base(vcpu, &apic_base_msr);
6849
6850	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6851	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6852	vcpu->arch.cr0 = sregs->cr0;
6853
6854	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6855	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6856	if (sregs->cr4 & X86_CR4_OSXSAVE)
6857		kvm_update_cpuid(vcpu);
6858
6859	idx = srcu_read_lock(&vcpu->kvm->srcu);
6860	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6861		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6862		mmu_reset_needed = 1;
6863	}
6864	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6865
6866	if (mmu_reset_needed)
6867		kvm_mmu_reset_context(vcpu);
6868
6869	max_bits = KVM_NR_INTERRUPTS;
6870	pending_vec = find_first_bit(
6871		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
6872	if (pending_vec < max_bits) {
6873		kvm_queue_interrupt(vcpu, pending_vec, false);
6874		pr_debug("Set back pending irq %d\n", pending_vec);
6875	}
6876
6877	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6878	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6879	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6880	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6881	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6882	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6883
6884	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6885	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6886
6887	update_cr8_intercept(vcpu);
6888
6889	/* Older userspace won't unhalt the vcpu on reset. */
6890	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6891	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6892	    !is_protmode(vcpu))
6893		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6894
6895	kvm_make_request(KVM_REQ_EVENT, vcpu);
6896
6897	return 0;
6898}
6899
6900int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6901					struct kvm_guest_debug *dbg)
6902{
6903	unsigned long rflags;
6904	int i, r;
6905
6906	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6907		r = -EBUSY;
6908		if (vcpu->arch.exception.pending)
6909			goto out;
6910		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6911			kvm_queue_exception(vcpu, DB_VECTOR);
6912		else
6913			kvm_queue_exception(vcpu, BP_VECTOR);
6914	}
6915
6916	/*
6917	 * Read rflags as long as potentially injected trace flags are still
6918	 * filtered out.
6919	 */
6920	rflags = kvm_get_rflags(vcpu);
6921
6922	vcpu->guest_debug = dbg->control;
6923	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6924		vcpu->guest_debug = 0;
6925
6926	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6927		for (i = 0; i < KVM_NR_DB_REGS; ++i)
6928			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6929		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6930	} else {
6931		for (i = 0; i < KVM_NR_DB_REGS; i++)
6932			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6933	}
6934	kvm_update_dr7(vcpu);
6935
6936	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6937		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6938			get_segment_base(vcpu, VCPU_SREG_CS);
6939
6940	/*
6941	 * Trigger an rflags update that will inject or remove the trace
6942	 * flags.
6943	 */
6944	kvm_set_rflags(vcpu, rflags);
6945
6946	kvm_x86_ops->update_db_bp_intercept(vcpu);
6947
6948	r = 0;
6949
6950out:
6951
6952	return r;
6953}
6954
6955/*
6956 * Translate a guest virtual address to a guest physical address.
6957 */
6958int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6959				    struct kvm_translation *tr)
6960{
6961	unsigned long vaddr = tr->linear_address;
6962	gpa_t gpa;
6963	int idx;
6964
6965	idx = srcu_read_lock(&vcpu->kvm->srcu);
6966	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6967	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6968	tr->physical_address = gpa;
6969	tr->valid = gpa != UNMAPPED_GVA;
6970	tr->writeable = 1;
6971	tr->usermode = 0;
6972
6973	return 0;
6974}
6975
6976int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6977{
6978	struct i387_fxsave_struct *fxsave =
6979			&vcpu->arch.guest_fpu.state->fxsave;
6980
6981	memcpy(fpu->fpr, fxsave->st_space, 128);
6982	fpu->fcw = fxsave->cwd;
6983	fpu->fsw = fxsave->swd;
6984	fpu->ftwx = fxsave->twd;
6985	fpu->last_opcode = fxsave->fop;
6986	fpu->last_ip = fxsave->rip;
6987	fpu->last_dp = fxsave->rdp;
6988	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6989
6990	return 0;
6991}
6992
6993int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6994{
6995	struct i387_fxsave_struct *fxsave =
6996			&vcpu->arch.guest_fpu.state->fxsave;
6997
6998	memcpy(fxsave->st_space, fpu->fpr, 128);
6999	fxsave->cwd = fpu->fcw;
7000	fxsave->swd = fpu->fsw;
7001	fxsave->twd = fpu->ftwx;
7002	fxsave->fop = fpu->last_opcode;
7003	fxsave->rip = fpu->last_ip;
7004	fxsave->rdp = fpu->last_dp;
7005	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7006
7007	return 0;
7008}
7009
7010int fx_init(struct kvm_vcpu *vcpu)
7011{
7012	int err;
7013
7014	err = fpu_alloc(&vcpu->arch.guest_fpu);
7015	if (err)
7016		return err;
7017
7018	fpu_finit(&vcpu->arch.guest_fpu);
7019	if (cpu_has_xsaves)
7020		vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
7021			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7022
7023	/*
7024	 * Ensure guest xcr0 is valid for loading
7025	 */
7026	vcpu->arch.xcr0 = XSTATE_FP;
7027
7028	vcpu->arch.cr0 |= X86_CR0_ET;
7029
7030	return 0;
7031}
7032EXPORT_SYMBOL_GPL(fx_init);
7033
7034static void fx_free(struct kvm_vcpu *vcpu)
7035{
7036	fpu_free(&vcpu->arch.guest_fpu);
7037}
7038
7039void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7040{
7041	if (vcpu->guest_fpu_loaded)
7042		return;
7043
7044	/*
7045	 * Restore all possible states in the guest,
7046	 * and assume host would use all available bits.
7047	 * Guest xcr0 would be loaded later.
7048	 */
7049	vcpu->guest_fpu_loaded = 1;
7050	__kernel_fpu_begin();
7051	fpu_restore_checking(&vcpu->arch.guest_fpu);
7052	trace_kvm_fpu(1);
7053}
7054
7055void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7056{
7057	if (!vcpu->guest_fpu_loaded)
7058		return;
7059
7060	vcpu->guest_fpu_loaded = 0;
7061	fpu_save_init(&vcpu->arch.guest_fpu);
7062	__kernel_fpu_end();
7063	++vcpu->stat.fpu_reload;
7064	if (!vcpu->arch.eager_fpu)
7065		kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7066
7067	trace_kvm_fpu(0);
7068}
7069
7070void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7071{
7072	kvmclock_reset(vcpu);
7073
7074	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7075	fx_free(vcpu);
7076	kvm_x86_ops->vcpu_free(vcpu);
7077}
7078
7079struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7080						unsigned int id)
7081{
7082	struct kvm_vcpu *vcpu;
7083
7084	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7085		printk_once(KERN_WARNING
7086		"kvm: SMP vm created on host with unstable TSC; "
7087		"guest TSC will not be reliable\n");
7088
7089	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7090
7091	/*
7092	 * Activate fpu unconditionally in case the guest needs eager FPU.  It will be
7093	 * deactivated soon if it doesn't.
7094	 */
7095	kvm_x86_ops->fpu_activate(vcpu);
7096	return vcpu;
7097}
7098
7099int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7100{
7101	int r;
7102
7103	vcpu->arch.mtrr_state.have_fixed = 1;
7104	r = vcpu_load(vcpu);
7105	if (r)
7106		return r;
7107	kvm_vcpu_reset(vcpu);
7108	kvm_mmu_setup(vcpu);
7109	vcpu_put(vcpu);
7110
7111	return r;
7112}
7113
7114void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7115{
7116	struct msr_data msr;
7117	struct kvm *kvm = vcpu->kvm;
7118
7119	if (vcpu_load(vcpu))
7120		return;
7121	msr.data = 0x0;
7122	msr.index = MSR_IA32_TSC;
7123	msr.host_initiated = true;
7124	kvm_write_tsc(vcpu, &msr);
7125	vcpu_put(vcpu);
7126
7127	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7128					KVMCLOCK_SYNC_PERIOD);
7129}
7130
7131void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7132{
7133	int r;
7134	vcpu->arch.apf.msr_val = 0;
7135
7136	r = vcpu_load(vcpu);
7137	BUG_ON(r);
7138	kvm_mmu_unload(vcpu);
7139	vcpu_put(vcpu);
7140
7141	fx_free(vcpu);
7142	kvm_x86_ops->vcpu_free(vcpu);
7143}
7144
7145void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
7146{
7147	atomic_set(&vcpu->arch.nmi_queued, 0);
7148	vcpu->arch.nmi_pending = 0;
7149	vcpu->arch.nmi_injected = false;
7150	kvm_clear_interrupt_queue(vcpu);
7151	kvm_clear_exception_queue(vcpu);
7152
7153	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7154	kvm_update_dr0123(vcpu);
7155	vcpu->arch.dr6 = DR6_INIT;
7156	kvm_update_dr6(vcpu);
7157	vcpu->arch.dr7 = DR7_FIXED_1;
7158	kvm_update_dr7(vcpu);
7159
7160	vcpu->arch.cr2 = 0;
7161
7162	kvm_make_request(KVM_REQ_EVENT, vcpu);
7163	vcpu->arch.apf.msr_val = 0;
7164	vcpu->arch.st.msr_val = 0;
7165
7166	kvmclock_reset(vcpu);
7167
7168	kvm_clear_async_pf_completion_queue(vcpu);
7169	kvm_async_pf_hash_reset(vcpu);
7170	vcpu->arch.apf.halted = false;
7171
7172	kvm_pmu_reset(vcpu);
7173
7174	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7175	vcpu->arch.regs_avail = ~0;
7176	vcpu->arch.regs_dirty = ~0;
7177
7178	kvm_x86_ops->vcpu_reset(vcpu);
7179}
7180
7181void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7182{
7183	struct kvm_segment cs;
7184
7185	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7186	cs.selector = vector << 8;
7187	cs.base = vector << 12;
7188	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7189	kvm_rip_write(vcpu, 0);
7190}
7191
7192int kvm_arch_hardware_enable(void)
7193{
7194	struct kvm *kvm;
7195	struct kvm_vcpu *vcpu;
7196	int i;
7197	int ret;
7198	u64 local_tsc;
7199	u64 max_tsc = 0;
7200	bool stable, backwards_tsc = false;
7201
7202	kvm_shared_msr_cpu_online();
7203	ret = kvm_x86_ops->hardware_enable();
7204	if (ret != 0)
7205		return ret;
7206
7207	local_tsc = native_read_tsc();
7208	stable = !check_tsc_unstable();
7209	list_for_each_entry(kvm, &vm_list, vm_list) {
7210		kvm_for_each_vcpu(i, vcpu, kvm) {
7211			if (!stable && vcpu->cpu == smp_processor_id())
7212				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7213			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7214				backwards_tsc = true;
7215				if (vcpu->arch.last_host_tsc > max_tsc)
7216					max_tsc = vcpu->arch.last_host_tsc;
7217			}
7218		}
7219	}
7220
7221	/*
7222	 * Sometimes, even reliable TSCs go backwards.  This happens on
7223	 * platforms that reset TSC during suspend or hibernate actions, but
7224	 * maintain synchronization.  We must compensate.  Fortunately, we can
7225	 * detect that condition here, which happens early in CPU bringup,
7226	 * before any KVM threads can be running.  Unfortunately, we can't
7227	 * bring the TSCs fully up to date with real time, as we aren't yet far
7228	 * enough into CPU bringup that we know how much real time has actually
7229	 * elapsed; our helper function, get_kernel_ns() will be using boot
7230	 * variables that haven't been updated yet.
7231	 *
7232	 * So we simply find the maximum observed TSC above, then record the
7233	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7234	 * the adjustment will be applied.  Note that we accumulate
7235	 * adjustments, in case multiple suspend cycles happen before some VCPU
7236	 * gets a chance to run again.  In the event that no KVM threads get a
7237	 * chance to run, we will miss the entire elapsed period, as we'll have
7238	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7239	 * loose cycle time.  This isn't too big a deal, since the loss will be
7240	 * uniform across all VCPUs (not to mention the scenario is extremely
7241	 * unlikely). It is possible that a second hibernate recovery happens
7242	 * much faster than a first, causing the observed TSC here to be
7243	 * smaller; this would require additional padding adjustment, which is
7244	 * why we set last_host_tsc to the local tsc observed here.
7245	 *
7246	 * N.B. - this code below runs only on platforms with reliable TSC,
7247	 * as that is the only way backwards_tsc is set above.  Also note
7248	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7249	 * have the same delta_cyc adjustment applied if backwards_tsc
7250	 * is detected.  Note further, this adjustment is only done once,
7251	 * as we reset last_host_tsc on all VCPUs to stop this from being
7252	 * called multiple times (one for each physical CPU bringup).
7253	 *
7254	 * Platforms with unreliable TSCs don't have to deal with this, they
7255	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7256	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7257	 * guarantee that they stay in perfect synchronization.
7258	 */
7259	if (backwards_tsc) {
7260		u64 delta_cyc = max_tsc - local_tsc;
7261		backwards_tsc_observed = true;
7262		list_for_each_entry(kvm, &vm_list, vm_list) {
7263			kvm_for_each_vcpu(i, vcpu, kvm) {
7264				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7265				vcpu->arch.last_host_tsc = local_tsc;
7266				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7267			}
7268
7269			/*
7270			 * We have to disable TSC offset matching.. if you were
7271			 * booting a VM while issuing an S4 host suspend....
7272			 * you may have some problem.  Solving this issue is
7273			 * left as an exercise to the reader.
7274			 */
7275			kvm->arch.last_tsc_nsec = 0;
7276			kvm->arch.last_tsc_write = 0;
7277		}
7278
7279	}
7280	return 0;
7281}
7282
7283void kvm_arch_hardware_disable(void)
7284{
7285	kvm_x86_ops->hardware_disable();
7286	drop_user_return_notifiers();
7287}
7288
7289int kvm_arch_hardware_setup(void)
7290{
7291	int r;
7292
7293	r = kvm_x86_ops->hardware_setup();
7294	if (r != 0)
7295		return r;
7296
7297	kvm_init_msr_list();
7298	return 0;
7299}
7300
7301void kvm_arch_hardware_unsetup(void)
7302{
7303	kvm_x86_ops->hardware_unsetup();
7304}
7305
7306void kvm_arch_check_processor_compat(void *rtn)
7307{
7308	kvm_x86_ops->check_processor_compatibility(rtn);
7309}
7310
7311bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7312{
7313	return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7314}
7315
7316struct static_key kvm_no_apic_vcpu __read_mostly;
7317
7318int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7319{
7320	struct page *page;
7321	struct kvm *kvm;
7322	int r;
7323
7324	BUG_ON(vcpu->kvm == NULL);
7325	kvm = vcpu->kvm;
7326
7327	vcpu->arch.pv.pv_unhalted = false;
7328	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7329	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7330		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7331	else
7332		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7333
7334	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7335	if (!page) {
7336		r = -ENOMEM;
7337		goto fail;
7338	}
7339	vcpu->arch.pio_data = page_address(page);
7340
7341	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7342
7343	r = kvm_mmu_create(vcpu);
7344	if (r < 0)
7345		goto fail_free_pio_data;
7346
7347	if (irqchip_in_kernel(kvm)) {
7348		r = kvm_create_lapic(vcpu);
7349		if (r < 0)
7350			goto fail_mmu_destroy;
7351	} else
7352		static_key_slow_inc(&kvm_no_apic_vcpu);
7353
7354	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7355				       GFP_KERNEL);
7356	if (!vcpu->arch.mce_banks) {
7357		r = -ENOMEM;
7358		goto fail_free_lapic;
7359	}
7360	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7361
7362	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7363		r = -ENOMEM;
7364		goto fail_free_mce_banks;
7365	}
7366
7367	r = fx_init(vcpu);
7368	if (r)
7369		goto fail_free_wbinvd_dirty_mask;
7370
7371	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7372	vcpu->arch.pv_time_enabled = false;
7373
7374	vcpu->arch.guest_supported_xcr0 = 0;
7375	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7376
7377	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7378
7379	kvm_async_pf_hash_reset(vcpu);
7380	kvm_pmu_init(vcpu);
7381
7382	return 0;
7383fail_free_wbinvd_dirty_mask:
7384	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7385fail_free_mce_banks:
7386	kfree(vcpu->arch.mce_banks);
7387fail_free_lapic:
7388	kvm_free_lapic(vcpu);
7389fail_mmu_destroy:
7390	kvm_mmu_destroy(vcpu);
7391fail_free_pio_data:
7392	free_page((unsigned long)vcpu->arch.pio_data);
7393fail:
7394	return r;
7395}
7396
7397void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7398{
7399	int idx;
7400
7401	kvm_pmu_destroy(vcpu);
7402	kfree(vcpu->arch.mce_banks);
7403	kvm_free_lapic(vcpu);
7404	idx = srcu_read_lock(&vcpu->kvm->srcu);
7405	kvm_mmu_destroy(vcpu);
7406	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7407	free_page((unsigned long)vcpu->arch.pio_data);
7408	if (!irqchip_in_kernel(vcpu->kvm))
7409		static_key_slow_dec(&kvm_no_apic_vcpu);
7410}
7411
7412void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7413{
7414	kvm_x86_ops->sched_in(vcpu, cpu);
7415}
7416
7417int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7418{
7419	if (type)
7420		return -EINVAL;
7421
7422	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7423	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7424	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7425	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7426	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7427
7428	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7429	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7430	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7431	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7432		&kvm->arch.irq_sources_bitmap);
7433
7434	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7435	mutex_init(&kvm->arch.apic_map_lock);
7436	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7437
7438	pvclock_update_vm_gtod_copy(kvm);
7439
7440	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7441	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7442
7443	return 0;
7444}
7445
7446static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7447{
7448	int r;
7449	r = vcpu_load(vcpu);
7450	BUG_ON(r);
7451	kvm_mmu_unload(vcpu);
7452	vcpu_put(vcpu);
7453}
7454
7455static void kvm_free_vcpus(struct kvm *kvm)
7456{
7457	unsigned int i;
7458	struct kvm_vcpu *vcpu;
7459
7460	/*
7461	 * Unpin any mmu pages first.
7462	 */
7463	kvm_for_each_vcpu(i, vcpu, kvm) {
7464		kvm_clear_async_pf_completion_queue(vcpu);
7465		kvm_unload_vcpu_mmu(vcpu);
7466	}
7467	kvm_for_each_vcpu(i, vcpu, kvm)
7468		kvm_arch_vcpu_free(vcpu);
7469
7470	mutex_lock(&kvm->lock);
7471	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7472		kvm->vcpus[i] = NULL;
7473
7474	atomic_set(&kvm->online_vcpus, 0);
7475	mutex_unlock(&kvm->lock);
7476}
7477
7478void kvm_arch_sync_events(struct kvm *kvm)
7479{
7480	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7481	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7482	kvm_free_all_assigned_devices(kvm);
7483	kvm_free_pit(kvm);
7484}
7485
7486void kvm_arch_destroy_vm(struct kvm *kvm)
7487{
7488	if (current->mm == kvm->mm) {
7489		/*
7490		 * Free memory regions allocated on behalf of userspace,
7491		 * unless the the memory map has changed due to process exit
7492		 * or fd copying.
7493		 */
7494		struct kvm_userspace_memory_region mem;
7495		memset(&mem, 0, sizeof(mem));
7496		mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7497		kvm_set_memory_region(kvm, &mem);
7498
7499		mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7500		kvm_set_memory_region(kvm, &mem);
7501
7502		mem.slot = TSS_PRIVATE_MEMSLOT;
7503		kvm_set_memory_region(kvm, &mem);
7504	}
7505	kvm_iommu_unmap_guest(kvm);
7506	kfree(kvm->arch.vpic);
7507	kfree(kvm->arch.vioapic);
7508	kvm_free_vcpus(kvm);
7509	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7510}
7511
7512void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7513			   struct kvm_memory_slot *dont)
7514{
7515	int i;
7516
7517	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7518		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7519			kvfree(free->arch.rmap[i]);
7520			free->arch.rmap[i] = NULL;
7521		}
7522		if (i == 0)
7523			continue;
7524
7525		if (!dont || free->arch.lpage_info[i - 1] !=
7526			     dont->arch.lpage_info[i - 1]) {
7527			kvfree(free->arch.lpage_info[i - 1]);
7528			free->arch.lpage_info[i - 1] = NULL;
7529		}
7530	}
7531}
7532
7533int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7534			    unsigned long npages)
7535{
7536	int i;
7537
7538	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7539		unsigned long ugfn;
7540		int lpages;
7541		int level = i + 1;
7542
7543		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7544				      slot->base_gfn, level) + 1;
7545
7546		slot->arch.rmap[i] =
7547			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7548		if (!slot->arch.rmap[i])
7549			goto out_free;
7550		if (i == 0)
7551			continue;
7552
7553		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7554					sizeof(*slot->arch.lpage_info[i - 1]));
7555		if (!slot->arch.lpage_info[i - 1])
7556			goto out_free;
7557
7558		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7559			slot->arch.lpage_info[i - 1][0].write_count = 1;
7560		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7561			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7562		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7563		/*
7564		 * If the gfn and userspace address are not aligned wrt each
7565		 * other, or if explicitly asked to, disable large page
7566		 * support for this slot
7567		 */
7568		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7569		    !kvm_largepages_enabled()) {
7570			unsigned long j;
7571
7572			for (j = 0; j < lpages; ++j)
7573				slot->arch.lpage_info[i - 1][j].write_count = 1;
7574		}
7575	}
7576
7577	return 0;
7578
7579out_free:
7580	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7581		kvfree(slot->arch.rmap[i]);
7582		slot->arch.rmap[i] = NULL;
7583		if (i == 0)
7584			continue;
7585
7586		kvfree(slot->arch.lpage_info[i - 1]);
7587		slot->arch.lpage_info[i - 1] = NULL;
7588	}
7589	return -ENOMEM;
7590}
7591
7592void kvm_arch_memslots_updated(struct kvm *kvm)
7593{
7594	/*
7595	 * memslots->generation has been incremented.
7596	 * mmio generation may have reached its maximum value.
7597	 */
7598	kvm_mmu_invalidate_mmio_sptes(kvm);
7599}
7600
7601int kvm_arch_prepare_memory_region(struct kvm *kvm,
7602				struct kvm_memory_slot *memslot,
7603				struct kvm_userspace_memory_region *mem,
7604				enum kvm_mr_change change)
7605{
7606	/*
7607	 * Only private memory slots need to be mapped here since
7608	 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7609	 */
7610	if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7611		unsigned long userspace_addr;
7612
7613		/*
7614		 * MAP_SHARED to prevent internal slot pages from being moved
7615		 * by fork()/COW.
7616		 */
7617		userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7618					 PROT_READ | PROT_WRITE,
7619					 MAP_SHARED | MAP_ANONYMOUS, 0);
7620
7621		if (IS_ERR((void *)userspace_addr))
7622			return PTR_ERR((void *)userspace_addr);
7623
7624		memslot->userspace_addr = userspace_addr;
7625	}
7626
7627	return 0;
7628}
7629
7630static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7631				     struct kvm_memory_slot *new)
7632{
7633	/* Still write protect RO slot */
7634	if (new->flags & KVM_MEM_READONLY) {
7635		kvm_mmu_slot_remove_write_access(kvm, new);
7636		return;
7637	}
7638
7639	/*
7640	 * Call kvm_x86_ops dirty logging hooks when they are valid.
7641	 *
7642	 * kvm_x86_ops->slot_disable_log_dirty is called when:
7643	 *
7644	 *  - KVM_MR_CREATE with dirty logging is disabled
7645	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7646	 *
7647	 * The reason is, in case of PML, we need to set D-bit for any slots
7648	 * with dirty logging disabled in order to eliminate unnecessary GPA
7649	 * logging in PML buffer (and potential PML buffer full VMEXT). This
7650	 * guarantees leaving PML enabled during guest's lifetime won't have
7651	 * any additonal overhead from PML when guest is running with dirty
7652	 * logging disabled for memory slots.
7653	 *
7654	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7655	 * to dirty logging mode.
7656	 *
7657	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7658	 *
7659	 * In case of write protect:
7660	 *
7661	 * Write protect all pages for dirty logging.
7662	 *
7663	 * All the sptes including the large sptes which point to this
7664	 * slot are set to readonly. We can not create any new large
7665	 * spte on this slot until the end of the logging.
7666	 *
7667	 * See the comments in fast_page_fault().
7668	 */
7669	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7670		if (kvm_x86_ops->slot_enable_log_dirty)
7671			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7672		else
7673			kvm_mmu_slot_remove_write_access(kvm, new);
7674	} else {
7675		if (kvm_x86_ops->slot_disable_log_dirty)
7676			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7677	}
7678}
7679
7680void kvm_arch_commit_memory_region(struct kvm *kvm,
7681				struct kvm_userspace_memory_region *mem,
7682				const struct kvm_memory_slot *old,
7683				enum kvm_mr_change change)
7684{
7685	struct kvm_memory_slot *new;
7686	int nr_mmu_pages = 0;
7687
7688	if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7689		int ret;
7690
7691		ret = vm_munmap(old->userspace_addr,
7692				old->npages * PAGE_SIZE);
7693		if (ret < 0)
7694			printk(KERN_WARNING
7695			       "kvm_vm_ioctl_set_memory_region: "
7696			       "failed to munmap memory\n");
7697	}
7698
7699	if (!kvm->arch.n_requested_mmu_pages)
7700		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7701
7702	if (nr_mmu_pages)
7703		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7704
7705	/* It's OK to get 'new' slot here as it has already been installed */
7706	new = id_to_memslot(kvm->memslots, mem->slot);
7707
7708	/*
7709	 * Dirty logging tracks sptes in 4k granularity, meaning that large
7710	 * sptes have to be split.  If live migration is successful, the guest
7711	 * in the source machine will be destroyed and large sptes will be
7712	 * created in the destination. However, if the guest continues to run
7713	 * in the source machine (for example if live migration fails), small
7714	 * sptes will remain around and cause bad performance.
7715	 *
7716	 * Scan sptes if dirty logging has been stopped, dropping those
7717	 * which can be collapsed into a single large-page spte.  Later
7718	 * page faults will create the large-page sptes.
7719	 */
7720	if ((change != KVM_MR_DELETE) &&
7721		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7722		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7723		kvm_mmu_zap_collapsible_sptes(kvm, new);
7724
7725	/*
7726	 * Set up write protection and/or dirty logging for the new slot.
7727	 *
7728	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7729	 * been zapped so no dirty logging staff is needed for old slot. For
7730	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7731	 * new and it's also covered when dealing with the new slot.
7732	 */
7733	if (change != KVM_MR_DELETE)
7734		kvm_mmu_slot_apply_flags(kvm, new);
7735}
7736
7737void kvm_arch_flush_shadow_all(struct kvm *kvm)
7738{
7739	kvm_mmu_invalidate_zap_all_pages(kvm);
7740}
7741
7742void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7743				   struct kvm_memory_slot *slot)
7744{
7745	kvm_mmu_invalidate_zap_all_pages(kvm);
7746}
7747
7748int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7749{
7750	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7751		kvm_x86_ops->check_nested_events(vcpu, false);
7752
7753	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7754		!vcpu->arch.apf.halted)
7755		|| !list_empty_careful(&vcpu->async_pf.done)
7756		|| kvm_apic_has_events(vcpu)
7757		|| vcpu->arch.pv.pv_unhalted
7758		|| atomic_read(&vcpu->arch.nmi_queued) ||
7759		(kvm_arch_interrupt_allowed(vcpu) &&
7760		 kvm_cpu_has_interrupt(vcpu));
7761}
7762
7763int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7764{
7765	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7766}
7767
7768int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7769{
7770	return kvm_x86_ops->interrupt_allowed(vcpu);
7771}
7772
7773unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7774{
7775	if (is_64_bit_mode(vcpu))
7776		return kvm_rip_read(vcpu);
7777	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7778		     kvm_rip_read(vcpu));
7779}
7780EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7781
7782bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7783{
7784	return kvm_get_linear_rip(vcpu) == linear_rip;
7785}
7786EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7787
7788unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7789{
7790	unsigned long rflags;
7791
7792	rflags = kvm_x86_ops->get_rflags(vcpu);
7793	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7794		rflags &= ~X86_EFLAGS_TF;
7795	return rflags;
7796}
7797EXPORT_SYMBOL_GPL(kvm_get_rflags);
7798
7799static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7800{
7801	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7802	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7803		rflags |= X86_EFLAGS_TF;
7804	kvm_x86_ops->set_rflags(vcpu, rflags);
7805}
7806
7807void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7808{
7809	__kvm_set_rflags(vcpu, rflags);
7810	kvm_make_request(KVM_REQ_EVENT, vcpu);
7811}
7812EXPORT_SYMBOL_GPL(kvm_set_rflags);
7813
7814void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7815{
7816	int r;
7817
7818	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7819	      work->wakeup_all)
7820		return;
7821
7822	r = kvm_mmu_reload(vcpu);
7823	if (unlikely(r))
7824		return;
7825
7826	if (!vcpu->arch.mmu.direct_map &&
7827	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7828		return;
7829
7830	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7831}
7832
7833static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7834{
7835	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7836}
7837
7838static inline u32 kvm_async_pf_next_probe(u32 key)
7839{
7840	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7841}
7842
7843static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7844{
7845	u32 key = kvm_async_pf_hash_fn(gfn);
7846
7847	while (vcpu->arch.apf.gfns[key] != ~0)
7848		key = kvm_async_pf_next_probe(key);
7849
7850	vcpu->arch.apf.gfns[key] = gfn;
7851}
7852
7853static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7854{
7855	int i;
7856	u32 key = kvm_async_pf_hash_fn(gfn);
7857
7858	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7859		     (vcpu->arch.apf.gfns[key] != gfn &&
7860		      vcpu->arch.apf.gfns[key] != ~0); i++)
7861		key = kvm_async_pf_next_probe(key);
7862
7863	return key;
7864}
7865
7866bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7867{
7868	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7869}
7870
7871static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7872{
7873	u32 i, j, k;
7874
7875	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7876	while (true) {
7877		vcpu->arch.apf.gfns[i] = ~0;
7878		do {
7879			j = kvm_async_pf_next_probe(j);
7880			if (vcpu->arch.apf.gfns[j] == ~0)
7881				return;
7882			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7883			/*
7884			 * k lies cyclically in ]i,j]
7885			 * |    i.k.j |
7886			 * |....j i.k.| or  |.k..j i...|
7887			 */
7888		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7889		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7890		i = j;
7891	}
7892}
7893
7894static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7895{
7896
7897	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7898				      sizeof(val));
7899}
7900
7901void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7902				     struct kvm_async_pf *work)
7903{
7904	struct x86_exception fault;
7905
7906	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7907	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7908
7909	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7910	    (vcpu->arch.apf.send_user_only &&
7911	     kvm_x86_ops->get_cpl(vcpu) == 0))
7912		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7913	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7914		fault.vector = PF_VECTOR;
7915		fault.error_code_valid = true;
7916		fault.error_code = 0;
7917		fault.nested_page_fault = false;
7918		fault.address = work->arch.token;
7919		kvm_inject_page_fault(vcpu, &fault);
7920	}
7921}
7922
7923void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7924				 struct kvm_async_pf *work)
7925{
7926	struct x86_exception fault;
7927
7928	trace_kvm_async_pf_ready(work->arch.token, work->gva);
7929	if (work->wakeup_all)
7930		work->arch.token = ~0; /* broadcast wakeup */
7931	else
7932		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7933
7934	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7935	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7936		fault.vector = PF_VECTOR;
7937		fault.error_code_valid = true;
7938		fault.error_code = 0;
7939		fault.nested_page_fault = false;
7940		fault.address = work->arch.token;
7941		kvm_inject_page_fault(vcpu, &fault);
7942	}
7943	vcpu->arch.apf.halted = false;
7944	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7945}
7946
7947bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7948{
7949	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7950		return true;
7951	else
7952		return !kvm_event_needs_reinjection(vcpu) &&
7953			kvm_x86_ops->interrupt_allowed(vcpu);
7954}
7955
7956void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7957{
7958	atomic_inc(&kvm->arch.noncoherent_dma_count);
7959}
7960EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7961
7962void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7963{
7964	atomic_dec(&kvm->arch.noncoherent_dma_count);
7965}
7966EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7967
7968bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7969{
7970	return atomic_read(&kvm->arch.noncoherent_dma_count);
7971}
7972EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7973
7974EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7975EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7976EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7977EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7978EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7979EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7980EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7981EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7982EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7983EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7984EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7985EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7986EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7987EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
7988EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
7989