xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 4da722ca19f30f7db250db808d1ab1703607a932)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 
58 #include <trace/events/kvm.h>
59 
60 #include <asm/debugreg.h>
61 #include <asm/msr.h>
62 #include <asm/desc.h>
63 #include <asm/mce.h>
64 #include <linux/kernel_stat.h>
65 #include <asm/fpu/internal.h> /* Ugh! */
66 #include <asm/pvclock.h>
67 #include <asm/div64.h>
68 #include <asm/irq_remapping.h>
69 
70 #define CREATE_TRACE_POINTS
71 #include "trace.h"
72 
73 #define MAX_IO_MSRS 256
74 #define KVM_MAX_MCE_BANKS 32
75 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
76 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
77 
78 #define emul_to_vcpu(ctxt) \
79 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
80 
81 /* EFER defaults:
82  * - enable syscall per default because its emulated by KVM
83  * - enable LME and LMA per default on 64 bit KVM
84  */
85 #ifdef CONFIG_X86_64
86 static
87 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
88 #else
89 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
90 #endif
91 
92 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
93 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
94 
95 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
96                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
97 
98 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
99 static void process_nmi(struct kvm_vcpu *vcpu);
100 static void enter_smm(struct kvm_vcpu *vcpu);
101 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
102 
103 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
104 EXPORT_SYMBOL_GPL(kvm_x86_ops);
105 
106 static bool __read_mostly ignore_msrs = 0;
107 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
108 
109 unsigned int min_timer_period_us = 500;
110 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
111 
112 static bool __read_mostly kvmclock_periodic_sync = true;
113 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
114 
115 bool __read_mostly kvm_has_tsc_control;
116 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
117 u32  __read_mostly kvm_max_guest_tsc_khz;
118 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
119 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
120 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
121 u64  __read_mostly kvm_max_tsc_scaling_ratio;
122 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
123 u64 __read_mostly kvm_default_tsc_scaling_ratio;
124 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
125 
126 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
127 static u32 __read_mostly tsc_tolerance_ppm = 250;
128 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
129 
130 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
131 unsigned int __read_mostly lapic_timer_advance_ns = 0;
132 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
133 
134 static bool __read_mostly vector_hashing = true;
135 module_param(vector_hashing, bool, S_IRUGO);
136 
137 #define KVM_NR_SHARED_MSRS 16
138 
139 struct kvm_shared_msrs_global {
140 	int nr;
141 	u32 msrs[KVM_NR_SHARED_MSRS];
142 };
143 
144 struct kvm_shared_msrs {
145 	struct user_return_notifier urn;
146 	bool registered;
147 	struct kvm_shared_msr_values {
148 		u64 host;
149 		u64 curr;
150 	} values[KVM_NR_SHARED_MSRS];
151 };
152 
153 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
154 static struct kvm_shared_msrs __percpu *shared_msrs;
155 
156 struct kvm_stats_debugfs_item debugfs_entries[] = {
157 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
158 	{ "pf_guest", VCPU_STAT(pf_guest) },
159 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
160 	{ "invlpg", VCPU_STAT(invlpg) },
161 	{ "exits", VCPU_STAT(exits) },
162 	{ "io_exits", VCPU_STAT(io_exits) },
163 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
164 	{ "signal_exits", VCPU_STAT(signal_exits) },
165 	{ "irq_window", VCPU_STAT(irq_window_exits) },
166 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
167 	{ "halt_exits", VCPU_STAT(halt_exits) },
168 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
169 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
170 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
171 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
172 	{ "hypercalls", VCPU_STAT(hypercalls) },
173 	{ "request_irq", VCPU_STAT(request_irq_exits) },
174 	{ "irq_exits", VCPU_STAT(irq_exits) },
175 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
176 	{ "efer_reload", VCPU_STAT(efer_reload) },
177 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
178 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
179 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
180 	{ "irq_injections", VCPU_STAT(irq_injections) },
181 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
182 	{ "req_event", VCPU_STAT(req_event) },
183 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
184 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
185 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
186 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
187 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
188 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
189 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
190 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
191 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
192 	{ "largepages", VM_STAT(lpages) },
193 	{ "max_mmu_page_hash_collisions",
194 		VM_STAT(max_mmu_page_hash_collisions) },
195 	{ NULL }
196 };
197 
198 u64 __read_mostly host_xcr0;
199 
200 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
201 
202 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
203 {
204 	int i;
205 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
206 		vcpu->arch.apf.gfns[i] = ~0;
207 }
208 
209 static void kvm_on_user_return(struct user_return_notifier *urn)
210 {
211 	unsigned slot;
212 	struct kvm_shared_msrs *locals
213 		= container_of(urn, struct kvm_shared_msrs, urn);
214 	struct kvm_shared_msr_values *values;
215 	unsigned long flags;
216 
217 	/*
218 	 * Disabling irqs at this point since the following code could be
219 	 * interrupted and executed through kvm_arch_hardware_disable()
220 	 */
221 	local_irq_save(flags);
222 	if (locals->registered) {
223 		locals->registered = false;
224 		user_return_notifier_unregister(urn);
225 	}
226 	local_irq_restore(flags);
227 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
228 		values = &locals->values[slot];
229 		if (values->host != values->curr) {
230 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
231 			values->curr = values->host;
232 		}
233 	}
234 }
235 
236 static void shared_msr_update(unsigned slot, u32 msr)
237 {
238 	u64 value;
239 	unsigned int cpu = smp_processor_id();
240 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
241 
242 	/* only read, and nobody should modify it at this time,
243 	 * so don't need lock */
244 	if (slot >= shared_msrs_global.nr) {
245 		printk(KERN_ERR "kvm: invalid MSR slot!");
246 		return;
247 	}
248 	rdmsrl_safe(msr, &value);
249 	smsr->values[slot].host = value;
250 	smsr->values[slot].curr = value;
251 }
252 
253 void kvm_define_shared_msr(unsigned slot, u32 msr)
254 {
255 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
256 	shared_msrs_global.msrs[slot] = msr;
257 	if (slot >= shared_msrs_global.nr)
258 		shared_msrs_global.nr = slot + 1;
259 }
260 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
261 
262 static void kvm_shared_msr_cpu_online(void)
263 {
264 	unsigned i;
265 
266 	for (i = 0; i < shared_msrs_global.nr; ++i)
267 		shared_msr_update(i, shared_msrs_global.msrs[i]);
268 }
269 
270 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
271 {
272 	unsigned int cpu = smp_processor_id();
273 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
274 	int err;
275 
276 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
277 		return 0;
278 	smsr->values[slot].curr = value;
279 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
280 	if (err)
281 		return 1;
282 
283 	if (!smsr->registered) {
284 		smsr->urn.on_user_return = kvm_on_user_return;
285 		user_return_notifier_register(&smsr->urn);
286 		smsr->registered = true;
287 	}
288 	return 0;
289 }
290 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
291 
292 static void drop_user_return_notifiers(void)
293 {
294 	unsigned int cpu = smp_processor_id();
295 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
296 
297 	if (smsr->registered)
298 		kvm_on_user_return(&smsr->urn);
299 }
300 
301 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
302 {
303 	return vcpu->arch.apic_base;
304 }
305 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
306 
307 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
308 {
309 	u64 old_state = vcpu->arch.apic_base &
310 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
311 	u64 new_state = msr_info->data &
312 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
313 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
314 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
315 
316 	if (!msr_info->host_initiated &&
317 	    ((msr_info->data & reserved_bits) != 0 ||
318 	     new_state == X2APIC_ENABLE ||
319 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
320 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
321 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
322 	      old_state == 0)))
323 		return 1;
324 
325 	kvm_lapic_set_base(vcpu, msr_info->data);
326 	return 0;
327 }
328 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
329 
330 asmlinkage __visible void kvm_spurious_fault(void)
331 {
332 	/* Fault while not rebooting.  We want the trace. */
333 	BUG();
334 }
335 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
336 
337 #define EXCPT_BENIGN		0
338 #define EXCPT_CONTRIBUTORY	1
339 #define EXCPT_PF		2
340 
341 static int exception_class(int vector)
342 {
343 	switch (vector) {
344 	case PF_VECTOR:
345 		return EXCPT_PF;
346 	case DE_VECTOR:
347 	case TS_VECTOR:
348 	case NP_VECTOR:
349 	case SS_VECTOR:
350 	case GP_VECTOR:
351 		return EXCPT_CONTRIBUTORY;
352 	default:
353 		break;
354 	}
355 	return EXCPT_BENIGN;
356 }
357 
358 #define EXCPT_FAULT		0
359 #define EXCPT_TRAP		1
360 #define EXCPT_ABORT		2
361 #define EXCPT_INTERRUPT		3
362 
363 static int exception_type(int vector)
364 {
365 	unsigned int mask;
366 
367 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
368 		return EXCPT_INTERRUPT;
369 
370 	mask = 1 << vector;
371 
372 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
373 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
374 		return EXCPT_TRAP;
375 
376 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
377 		return EXCPT_ABORT;
378 
379 	/* Reserved exceptions will result in fault */
380 	return EXCPT_FAULT;
381 }
382 
383 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
384 		unsigned nr, bool has_error, u32 error_code,
385 		bool reinject)
386 {
387 	u32 prev_nr;
388 	int class1, class2;
389 
390 	kvm_make_request(KVM_REQ_EVENT, vcpu);
391 
392 	if (!vcpu->arch.exception.pending) {
393 	queue:
394 		if (has_error && !is_protmode(vcpu))
395 			has_error = false;
396 		vcpu->arch.exception.pending = true;
397 		vcpu->arch.exception.has_error_code = has_error;
398 		vcpu->arch.exception.nr = nr;
399 		vcpu->arch.exception.error_code = error_code;
400 		vcpu->arch.exception.reinject = reinject;
401 		return;
402 	}
403 
404 	/* to check exception */
405 	prev_nr = vcpu->arch.exception.nr;
406 	if (prev_nr == DF_VECTOR) {
407 		/* triple fault -> shutdown */
408 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
409 		return;
410 	}
411 	class1 = exception_class(prev_nr);
412 	class2 = exception_class(nr);
413 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
414 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
415 		/* generate double fault per SDM Table 5-5 */
416 		vcpu->arch.exception.pending = true;
417 		vcpu->arch.exception.has_error_code = true;
418 		vcpu->arch.exception.nr = DF_VECTOR;
419 		vcpu->arch.exception.error_code = 0;
420 	} else
421 		/* replace previous exception with a new one in a hope
422 		   that instruction re-execution will regenerate lost
423 		   exception */
424 		goto queue;
425 }
426 
427 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
428 {
429 	kvm_multiple_exception(vcpu, nr, false, 0, false);
430 }
431 EXPORT_SYMBOL_GPL(kvm_queue_exception);
432 
433 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
434 {
435 	kvm_multiple_exception(vcpu, nr, false, 0, true);
436 }
437 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
438 
439 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
440 {
441 	if (err)
442 		kvm_inject_gp(vcpu, 0);
443 	else
444 		return kvm_skip_emulated_instruction(vcpu);
445 
446 	return 1;
447 }
448 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
449 
450 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
451 {
452 	++vcpu->stat.pf_guest;
453 	vcpu->arch.exception.nested_apf =
454 		is_guest_mode(vcpu) && fault->async_page_fault;
455 	if (vcpu->arch.exception.nested_apf)
456 		vcpu->arch.apf.nested_apf_token = fault->address;
457 	else
458 		vcpu->arch.cr2 = fault->address;
459 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
460 }
461 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
462 
463 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
464 {
465 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
466 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
467 	else
468 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
469 
470 	return fault->nested_page_fault;
471 }
472 
473 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
474 {
475 	atomic_inc(&vcpu->arch.nmi_queued);
476 	kvm_make_request(KVM_REQ_NMI, vcpu);
477 }
478 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
479 
480 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
481 {
482 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
483 }
484 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
485 
486 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
487 {
488 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
489 }
490 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
491 
492 /*
493  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
494  * a #GP and return false.
495  */
496 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
497 {
498 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
499 		return true;
500 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
501 	return false;
502 }
503 EXPORT_SYMBOL_GPL(kvm_require_cpl);
504 
505 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
506 {
507 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
508 		return true;
509 
510 	kvm_queue_exception(vcpu, UD_VECTOR);
511 	return false;
512 }
513 EXPORT_SYMBOL_GPL(kvm_require_dr);
514 
515 /*
516  * This function will be used to read from the physical memory of the currently
517  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
518  * can read from guest physical or from the guest's guest physical memory.
519  */
520 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
521 			    gfn_t ngfn, void *data, int offset, int len,
522 			    u32 access)
523 {
524 	struct x86_exception exception;
525 	gfn_t real_gfn;
526 	gpa_t ngpa;
527 
528 	ngpa     = gfn_to_gpa(ngfn);
529 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
530 	if (real_gfn == UNMAPPED_GVA)
531 		return -EFAULT;
532 
533 	real_gfn = gpa_to_gfn(real_gfn);
534 
535 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
536 }
537 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
538 
539 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
540 			       void *data, int offset, int len, u32 access)
541 {
542 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
543 				       data, offset, len, access);
544 }
545 
546 /*
547  * Load the pae pdptrs.  Return true is they are all valid.
548  */
549 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
550 {
551 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
552 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
553 	int i;
554 	int ret;
555 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
556 
557 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
558 				      offset * sizeof(u64), sizeof(pdpte),
559 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
560 	if (ret < 0) {
561 		ret = 0;
562 		goto out;
563 	}
564 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
565 		if ((pdpte[i] & PT_PRESENT_MASK) &&
566 		    (pdpte[i] &
567 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
568 			ret = 0;
569 			goto out;
570 		}
571 	}
572 	ret = 1;
573 
574 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
575 	__set_bit(VCPU_EXREG_PDPTR,
576 		  (unsigned long *)&vcpu->arch.regs_avail);
577 	__set_bit(VCPU_EXREG_PDPTR,
578 		  (unsigned long *)&vcpu->arch.regs_dirty);
579 out:
580 
581 	return ret;
582 }
583 EXPORT_SYMBOL_GPL(load_pdptrs);
584 
585 bool pdptrs_changed(struct kvm_vcpu *vcpu)
586 {
587 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
588 	bool changed = true;
589 	int offset;
590 	gfn_t gfn;
591 	int r;
592 
593 	if (is_long_mode(vcpu) || !is_pae(vcpu))
594 		return false;
595 
596 	if (!test_bit(VCPU_EXREG_PDPTR,
597 		      (unsigned long *)&vcpu->arch.regs_avail))
598 		return true;
599 
600 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
601 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
602 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
603 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
604 	if (r < 0)
605 		goto out;
606 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
607 out:
608 
609 	return changed;
610 }
611 EXPORT_SYMBOL_GPL(pdptrs_changed);
612 
613 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
614 {
615 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
616 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
617 
618 	cr0 |= X86_CR0_ET;
619 
620 #ifdef CONFIG_X86_64
621 	if (cr0 & 0xffffffff00000000UL)
622 		return 1;
623 #endif
624 
625 	cr0 &= ~CR0_RESERVED_BITS;
626 
627 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
628 		return 1;
629 
630 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
631 		return 1;
632 
633 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
634 #ifdef CONFIG_X86_64
635 		if ((vcpu->arch.efer & EFER_LME)) {
636 			int cs_db, cs_l;
637 
638 			if (!is_pae(vcpu))
639 				return 1;
640 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
641 			if (cs_l)
642 				return 1;
643 		} else
644 #endif
645 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
646 						 kvm_read_cr3(vcpu)))
647 			return 1;
648 	}
649 
650 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
651 		return 1;
652 
653 	kvm_x86_ops->set_cr0(vcpu, cr0);
654 
655 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
656 		kvm_clear_async_pf_completion_queue(vcpu);
657 		kvm_async_pf_hash_reset(vcpu);
658 	}
659 
660 	if ((cr0 ^ old_cr0) & update_bits)
661 		kvm_mmu_reset_context(vcpu);
662 
663 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
664 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
665 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
666 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
667 
668 	return 0;
669 }
670 EXPORT_SYMBOL_GPL(kvm_set_cr0);
671 
672 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
673 {
674 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
675 }
676 EXPORT_SYMBOL_GPL(kvm_lmsw);
677 
678 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
679 {
680 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
681 			!vcpu->guest_xcr0_loaded) {
682 		/* kvm_set_xcr() also depends on this */
683 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
684 		vcpu->guest_xcr0_loaded = 1;
685 	}
686 }
687 
688 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
689 {
690 	if (vcpu->guest_xcr0_loaded) {
691 		if (vcpu->arch.xcr0 != host_xcr0)
692 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
693 		vcpu->guest_xcr0_loaded = 0;
694 	}
695 }
696 
697 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
698 {
699 	u64 xcr0 = xcr;
700 	u64 old_xcr0 = vcpu->arch.xcr0;
701 	u64 valid_bits;
702 
703 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
704 	if (index != XCR_XFEATURE_ENABLED_MASK)
705 		return 1;
706 	if (!(xcr0 & XFEATURE_MASK_FP))
707 		return 1;
708 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
709 		return 1;
710 
711 	/*
712 	 * Do not allow the guest to set bits that we do not support
713 	 * saving.  However, xcr0 bit 0 is always set, even if the
714 	 * emulated CPU does not support XSAVE (see fx_init).
715 	 */
716 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
717 	if (xcr0 & ~valid_bits)
718 		return 1;
719 
720 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
721 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
722 		return 1;
723 
724 	if (xcr0 & XFEATURE_MASK_AVX512) {
725 		if (!(xcr0 & XFEATURE_MASK_YMM))
726 			return 1;
727 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
728 			return 1;
729 	}
730 	vcpu->arch.xcr0 = xcr0;
731 
732 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
733 		kvm_update_cpuid(vcpu);
734 	return 0;
735 }
736 
737 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
738 {
739 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
740 	    __kvm_set_xcr(vcpu, index, xcr)) {
741 		kvm_inject_gp(vcpu, 0);
742 		return 1;
743 	}
744 	return 0;
745 }
746 EXPORT_SYMBOL_GPL(kvm_set_xcr);
747 
748 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
749 {
750 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
751 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
752 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
753 
754 	if (cr4 & CR4_RESERVED_BITS)
755 		return 1;
756 
757 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
758 		return 1;
759 
760 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
761 		return 1;
762 
763 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
764 		return 1;
765 
766 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
767 		return 1;
768 
769 	if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
770 		return 1;
771 
772 	if (is_long_mode(vcpu)) {
773 		if (!(cr4 & X86_CR4_PAE))
774 			return 1;
775 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
776 		   && ((cr4 ^ old_cr4) & pdptr_bits)
777 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
778 				   kvm_read_cr3(vcpu)))
779 		return 1;
780 
781 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
782 		if (!guest_cpuid_has_pcid(vcpu))
783 			return 1;
784 
785 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
786 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
787 			return 1;
788 	}
789 
790 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
791 		return 1;
792 
793 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
794 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
795 		kvm_mmu_reset_context(vcpu);
796 
797 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
798 		kvm_update_cpuid(vcpu);
799 
800 	return 0;
801 }
802 EXPORT_SYMBOL_GPL(kvm_set_cr4);
803 
804 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
805 {
806 #ifdef CONFIG_X86_64
807 	cr3 &= ~CR3_PCID_INVD;
808 #endif
809 
810 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
811 		kvm_mmu_sync_roots(vcpu);
812 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
813 		return 0;
814 	}
815 
816 	if (is_long_mode(vcpu)) {
817 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
818 			return 1;
819 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
820 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
821 		return 1;
822 
823 	vcpu->arch.cr3 = cr3;
824 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
825 	kvm_mmu_new_cr3(vcpu);
826 	return 0;
827 }
828 EXPORT_SYMBOL_GPL(kvm_set_cr3);
829 
830 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
831 {
832 	if (cr8 & CR8_RESERVED_BITS)
833 		return 1;
834 	if (lapic_in_kernel(vcpu))
835 		kvm_lapic_set_tpr(vcpu, cr8);
836 	else
837 		vcpu->arch.cr8 = cr8;
838 	return 0;
839 }
840 EXPORT_SYMBOL_GPL(kvm_set_cr8);
841 
842 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
843 {
844 	if (lapic_in_kernel(vcpu))
845 		return kvm_lapic_get_cr8(vcpu);
846 	else
847 		return vcpu->arch.cr8;
848 }
849 EXPORT_SYMBOL_GPL(kvm_get_cr8);
850 
851 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
852 {
853 	int i;
854 
855 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
856 		for (i = 0; i < KVM_NR_DB_REGS; i++)
857 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
858 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
859 	}
860 }
861 
862 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
863 {
864 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
865 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
866 }
867 
868 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
869 {
870 	unsigned long dr7;
871 
872 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
873 		dr7 = vcpu->arch.guest_debug_dr7;
874 	else
875 		dr7 = vcpu->arch.dr7;
876 	kvm_x86_ops->set_dr7(vcpu, dr7);
877 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
878 	if (dr7 & DR7_BP_EN_MASK)
879 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
880 }
881 
882 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
883 {
884 	u64 fixed = DR6_FIXED_1;
885 
886 	if (!guest_cpuid_has_rtm(vcpu))
887 		fixed |= DR6_RTM;
888 	return fixed;
889 }
890 
891 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
892 {
893 	switch (dr) {
894 	case 0 ... 3:
895 		vcpu->arch.db[dr] = val;
896 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
897 			vcpu->arch.eff_db[dr] = val;
898 		break;
899 	case 4:
900 		/* fall through */
901 	case 6:
902 		if (val & 0xffffffff00000000ULL)
903 			return -1; /* #GP */
904 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
905 		kvm_update_dr6(vcpu);
906 		break;
907 	case 5:
908 		/* fall through */
909 	default: /* 7 */
910 		if (val & 0xffffffff00000000ULL)
911 			return -1; /* #GP */
912 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
913 		kvm_update_dr7(vcpu);
914 		break;
915 	}
916 
917 	return 0;
918 }
919 
920 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
921 {
922 	if (__kvm_set_dr(vcpu, dr, val)) {
923 		kvm_inject_gp(vcpu, 0);
924 		return 1;
925 	}
926 	return 0;
927 }
928 EXPORT_SYMBOL_GPL(kvm_set_dr);
929 
930 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
931 {
932 	switch (dr) {
933 	case 0 ... 3:
934 		*val = vcpu->arch.db[dr];
935 		break;
936 	case 4:
937 		/* fall through */
938 	case 6:
939 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
940 			*val = vcpu->arch.dr6;
941 		else
942 			*val = kvm_x86_ops->get_dr6(vcpu);
943 		break;
944 	case 5:
945 		/* fall through */
946 	default: /* 7 */
947 		*val = vcpu->arch.dr7;
948 		break;
949 	}
950 	return 0;
951 }
952 EXPORT_SYMBOL_GPL(kvm_get_dr);
953 
954 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
955 {
956 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
957 	u64 data;
958 	int err;
959 
960 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
961 	if (err)
962 		return err;
963 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
964 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
965 	return err;
966 }
967 EXPORT_SYMBOL_GPL(kvm_rdpmc);
968 
969 /*
970  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
971  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
972  *
973  * This list is modified at module load time to reflect the
974  * capabilities of the host cpu. This capabilities test skips MSRs that are
975  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
976  * may depend on host virtualization features rather than host cpu features.
977  */
978 
979 static u32 msrs_to_save[] = {
980 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
981 	MSR_STAR,
982 #ifdef CONFIG_X86_64
983 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
984 #endif
985 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
986 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
987 };
988 
989 static unsigned num_msrs_to_save;
990 
991 static u32 emulated_msrs[] = {
992 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
993 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
994 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
995 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
996 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
997 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
998 	HV_X64_MSR_RESET,
999 	HV_X64_MSR_VP_INDEX,
1000 	HV_X64_MSR_VP_RUNTIME,
1001 	HV_X64_MSR_SCONTROL,
1002 	HV_X64_MSR_STIMER0_CONFIG,
1003 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1004 	MSR_KVM_PV_EOI_EN,
1005 
1006 	MSR_IA32_TSC_ADJUST,
1007 	MSR_IA32_TSCDEADLINE,
1008 	MSR_IA32_MISC_ENABLE,
1009 	MSR_IA32_MCG_STATUS,
1010 	MSR_IA32_MCG_CTL,
1011 	MSR_IA32_MCG_EXT_CTL,
1012 	MSR_IA32_SMBASE,
1013 	MSR_PLATFORM_INFO,
1014 	MSR_MISC_FEATURES_ENABLES,
1015 };
1016 
1017 static unsigned num_emulated_msrs;
1018 
1019 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1020 {
1021 	if (efer & efer_reserved_bits)
1022 		return false;
1023 
1024 	if (efer & EFER_FFXSR) {
1025 		struct kvm_cpuid_entry2 *feat;
1026 
1027 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1028 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1029 			return false;
1030 	}
1031 
1032 	if (efer & EFER_SVME) {
1033 		struct kvm_cpuid_entry2 *feat;
1034 
1035 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1036 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1037 			return false;
1038 	}
1039 
1040 	return true;
1041 }
1042 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1043 
1044 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1045 {
1046 	u64 old_efer = vcpu->arch.efer;
1047 
1048 	if (!kvm_valid_efer(vcpu, efer))
1049 		return 1;
1050 
1051 	if (is_paging(vcpu)
1052 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1053 		return 1;
1054 
1055 	efer &= ~EFER_LMA;
1056 	efer |= vcpu->arch.efer & EFER_LMA;
1057 
1058 	kvm_x86_ops->set_efer(vcpu, efer);
1059 
1060 	/* Update reserved bits */
1061 	if ((efer ^ old_efer) & EFER_NX)
1062 		kvm_mmu_reset_context(vcpu);
1063 
1064 	return 0;
1065 }
1066 
1067 void kvm_enable_efer_bits(u64 mask)
1068 {
1069        efer_reserved_bits &= ~mask;
1070 }
1071 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1072 
1073 /*
1074  * Writes msr value into into the appropriate "register".
1075  * Returns 0 on success, non-0 otherwise.
1076  * Assumes vcpu_load() was already called.
1077  */
1078 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1079 {
1080 	switch (msr->index) {
1081 	case MSR_FS_BASE:
1082 	case MSR_GS_BASE:
1083 	case MSR_KERNEL_GS_BASE:
1084 	case MSR_CSTAR:
1085 	case MSR_LSTAR:
1086 		if (is_noncanonical_address(msr->data))
1087 			return 1;
1088 		break;
1089 	case MSR_IA32_SYSENTER_EIP:
1090 	case MSR_IA32_SYSENTER_ESP:
1091 		/*
1092 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1093 		 * non-canonical address is written on Intel but not on
1094 		 * AMD (which ignores the top 32-bits, because it does
1095 		 * not implement 64-bit SYSENTER).
1096 		 *
1097 		 * 64-bit code should hence be able to write a non-canonical
1098 		 * value on AMD.  Making the address canonical ensures that
1099 		 * vmentry does not fail on Intel after writing a non-canonical
1100 		 * value, and that something deterministic happens if the guest
1101 		 * invokes 64-bit SYSENTER.
1102 		 */
1103 		msr->data = get_canonical(msr->data);
1104 	}
1105 	return kvm_x86_ops->set_msr(vcpu, msr);
1106 }
1107 EXPORT_SYMBOL_GPL(kvm_set_msr);
1108 
1109 /*
1110  * Adapt set_msr() to msr_io()'s calling convention
1111  */
1112 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1113 {
1114 	struct msr_data msr;
1115 	int r;
1116 
1117 	msr.index = index;
1118 	msr.host_initiated = true;
1119 	r = kvm_get_msr(vcpu, &msr);
1120 	if (r)
1121 		return r;
1122 
1123 	*data = msr.data;
1124 	return 0;
1125 }
1126 
1127 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1128 {
1129 	struct msr_data msr;
1130 
1131 	msr.data = *data;
1132 	msr.index = index;
1133 	msr.host_initiated = true;
1134 	return kvm_set_msr(vcpu, &msr);
1135 }
1136 
1137 #ifdef CONFIG_X86_64
1138 struct pvclock_gtod_data {
1139 	seqcount_t	seq;
1140 
1141 	struct { /* extract of a clocksource struct */
1142 		int vclock_mode;
1143 		u64	cycle_last;
1144 		u64	mask;
1145 		u32	mult;
1146 		u32	shift;
1147 	} clock;
1148 
1149 	u64		boot_ns;
1150 	u64		nsec_base;
1151 	u64		wall_time_sec;
1152 };
1153 
1154 static struct pvclock_gtod_data pvclock_gtod_data;
1155 
1156 static void update_pvclock_gtod(struct timekeeper *tk)
1157 {
1158 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1159 	u64 boot_ns;
1160 
1161 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1162 
1163 	write_seqcount_begin(&vdata->seq);
1164 
1165 	/* copy pvclock gtod data */
1166 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1167 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1168 	vdata->clock.mask		= tk->tkr_mono.mask;
1169 	vdata->clock.mult		= tk->tkr_mono.mult;
1170 	vdata->clock.shift		= tk->tkr_mono.shift;
1171 
1172 	vdata->boot_ns			= boot_ns;
1173 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1174 
1175 	vdata->wall_time_sec            = tk->xtime_sec;
1176 
1177 	write_seqcount_end(&vdata->seq);
1178 }
1179 #endif
1180 
1181 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1182 {
1183 	/*
1184 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1185 	 * vcpu_enter_guest.  This function is only called from
1186 	 * the physical CPU that is running vcpu.
1187 	 */
1188 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1189 }
1190 
1191 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1192 {
1193 	int version;
1194 	int r;
1195 	struct pvclock_wall_clock wc;
1196 	struct timespec64 boot;
1197 
1198 	if (!wall_clock)
1199 		return;
1200 
1201 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1202 	if (r)
1203 		return;
1204 
1205 	if (version & 1)
1206 		++version;  /* first time write, random junk */
1207 
1208 	++version;
1209 
1210 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1211 		return;
1212 
1213 	/*
1214 	 * The guest calculates current wall clock time by adding
1215 	 * system time (updated by kvm_guest_time_update below) to the
1216 	 * wall clock specified here.  guest system time equals host
1217 	 * system time for us, thus we must fill in host boot time here.
1218 	 */
1219 	getboottime64(&boot);
1220 
1221 	if (kvm->arch.kvmclock_offset) {
1222 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1223 		boot = timespec64_sub(boot, ts);
1224 	}
1225 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1226 	wc.nsec = boot.tv_nsec;
1227 	wc.version = version;
1228 
1229 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1230 
1231 	version++;
1232 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1233 }
1234 
1235 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1236 {
1237 	do_shl32_div32(dividend, divisor);
1238 	return dividend;
1239 }
1240 
1241 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1242 			       s8 *pshift, u32 *pmultiplier)
1243 {
1244 	uint64_t scaled64;
1245 	int32_t  shift = 0;
1246 	uint64_t tps64;
1247 	uint32_t tps32;
1248 
1249 	tps64 = base_hz;
1250 	scaled64 = scaled_hz;
1251 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1252 		tps64 >>= 1;
1253 		shift--;
1254 	}
1255 
1256 	tps32 = (uint32_t)tps64;
1257 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1258 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1259 			scaled64 >>= 1;
1260 		else
1261 			tps32 <<= 1;
1262 		shift++;
1263 	}
1264 
1265 	*pshift = shift;
1266 	*pmultiplier = div_frac(scaled64, tps32);
1267 
1268 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1269 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1270 }
1271 
1272 #ifdef CONFIG_X86_64
1273 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1274 #endif
1275 
1276 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1277 static unsigned long max_tsc_khz;
1278 
1279 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1280 {
1281 	u64 v = (u64)khz * (1000000 + ppm);
1282 	do_div(v, 1000000);
1283 	return v;
1284 }
1285 
1286 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1287 {
1288 	u64 ratio;
1289 
1290 	/* Guest TSC same frequency as host TSC? */
1291 	if (!scale) {
1292 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1293 		return 0;
1294 	}
1295 
1296 	/* TSC scaling supported? */
1297 	if (!kvm_has_tsc_control) {
1298 		if (user_tsc_khz > tsc_khz) {
1299 			vcpu->arch.tsc_catchup = 1;
1300 			vcpu->arch.tsc_always_catchup = 1;
1301 			return 0;
1302 		} else {
1303 			WARN(1, "user requested TSC rate below hardware speed\n");
1304 			return -1;
1305 		}
1306 	}
1307 
1308 	/* TSC scaling required  - calculate ratio */
1309 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1310 				user_tsc_khz, tsc_khz);
1311 
1312 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1313 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1314 			  user_tsc_khz);
1315 		return -1;
1316 	}
1317 
1318 	vcpu->arch.tsc_scaling_ratio = ratio;
1319 	return 0;
1320 }
1321 
1322 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1323 {
1324 	u32 thresh_lo, thresh_hi;
1325 	int use_scaling = 0;
1326 
1327 	/* tsc_khz can be zero if TSC calibration fails */
1328 	if (user_tsc_khz == 0) {
1329 		/* set tsc_scaling_ratio to a safe value */
1330 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1331 		return -1;
1332 	}
1333 
1334 	/* Compute a scale to convert nanoseconds in TSC cycles */
1335 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1336 			   &vcpu->arch.virtual_tsc_shift,
1337 			   &vcpu->arch.virtual_tsc_mult);
1338 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1339 
1340 	/*
1341 	 * Compute the variation in TSC rate which is acceptable
1342 	 * within the range of tolerance and decide if the
1343 	 * rate being applied is within that bounds of the hardware
1344 	 * rate.  If so, no scaling or compensation need be done.
1345 	 */
1346 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1347 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1348 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1349 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1350 		use_scaling = 1;
1351 	}
1352 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1353 }
1354 
1355 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1356 {
1357 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1358 				      vcpu->arch.virtual_tsc_mult,
1359 				      vcpu->arch.virtual_tsc_shift);
1360 	tsc += vcpu->arch.this_tsc_write;
1361 	return tsc;
1362 }
1363 
1364 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1365 {
1366 #ifdef CONFIG_X86_64
1367 	bool vcpus_matched;
1368 	struct kvm_arch *ka = &vcpu->kvm->arch;
1369 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1370 
1371 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1372 			 atomic_read(&vcpu->kvm->online_vcpus));
1373 
1374 	/*
1375 	 * Once the masterclock is enabled, always perform request in
1376 	 * order to update it.
1377 	 *
1378 	 * In order to enable masterclock, the host clocksource must be TSC
1379 	 * and the vcpus need to have matched TSCs.  When that happens,
1380 	 * perform request to enable masterclock.
1381 	 */
1382 	if (ka->use_master_clock ||
1383 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1384 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1385 
1386 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1387 			    atomic_read(&vcpu->kvm->online_vcpus),
1388 		            ka->use_master_clock, gtod->clock.vclock_mode);
1389 #endif
1390 }
1391 
1392 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1393 {
1394 	u64 curr_offset = vcpu->arch.tsc_offset;
1395 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1396 }
1397 
1398 /*
1399  * Multiply tsc by a fixed point number represented by ratio.
1400  *
1401  * The most significant 64-N bits (mult) of ratio represent the
1402  * integral part of the fixed point number; the remaining N bits
1403  * (frac) represent the fractional part, ie. ratio represents a fixed
1404  * point number (mult + frac * 2^(-N)).
1405  *
1406  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1407  */
1408 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1409 {
1410 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1411 }
1412 
1413 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1414 {
1415 	u64 _tsc = tsc;
1416 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1417 
1418 	if (ratio != kvm_default_tsc_scaling_ratio)
1419 		_tsc = __scale_tsc(ratio, tsc);
1420 
1421 	return _tsc;
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1424 
1425 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1426 {
1427 	u64 tsc;
1428 
1429 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1430 
1431 	return target_tsc - tsc;
1432 }
1433 
1434 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1435 {
1436 	return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1439 
1440 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1441 {
1442 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1443 	vcpu->arch.tsc_offset = offset;
1444 }
1445 
1446 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1447 {
1448 	struct kvm *kvm = vcpu->kvm;
1449 	u64 offset, ns, elapsed;
1450 	unsigned long flags;
1451 	bool matched;
1452 	bool already_matched;
1453 	u64 data = msr->data;
1454 	bool synchronizing = false;
1455 
1456 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1457 	offset = kvm_compute_tsc_offset(vcpu, data);
1458 	ns = ktime_get_boot_ns();
1459 	elapsed = ns - kvm->arch.last_tsc_nsec;
1460 
1461 	if (vcpu->arch.virtual_tsc_khz) {
1462 		if (data == 0 && msr->host_initiated) {
1463 			/*
1464 			 * detection of vcpu initialization -- need to sync
1465 			 * with other vCPUs. This particularly helps to keep
1466 			 * kvm_clock stable after CPU hotplug
1467 			 */
1468 			synchronizing = true;
1469 		} else {
1470 			u64 tsc_exp = kvm->arch.last_tsc_write +
1471 						nsec_to_cycles(vcpu, elapsed);
1472 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1473 			/*
1474 			 * Special case: TSC write with a small delta (1 second)
1475 			 * of virtual cycle time against real time is
1476 			 * interpreted as an attempt to synchronize the CPU.
1477 			 */
1478 			synchronizing = data < tsc_exp + tsc_hz &&
1479 					data + tsc_hz > tsc_exp;
1480 		}
1481 	}
1482 
1483 	/*
1484 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1485 	 * TSC, we add elapsed time in this computation.  We could let the
1486 	 * compensation code attempt to catch up if we fall behind, but
1487 	 * it's better to try to match offsets from the beginning.
1488          */
1489 	if (synchronizing &&
1490 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1491 		if (!check_tsc_unstable()) {
1492 			offset = kvm->arch.cur_tsc_offset;
1493 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1494 		} else {
1495 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1496 			data += delta;
1497 			offset = kvm_compute_tsc_offset(vcpu, data);
1498 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1499 		}
1500 		matched = true;
1501 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1502 	} else {
1503 		/*
1504 		 * We split periods of matched TSC writes into generations.
1505 		 * For each generation, we track the original measured
1506 		 * nanosecond time, offset, and write, so if TSCs are in
1507 		 * sync, we can match exact offset, and if not, we can match
1508 		 * exact software computation in compute_guest_tsc()
1509 		 *
1510 		 * These values are tracked in kvm->arch.cur_xxx variables.
1511 		 */
1512 		kvm->arch.cur_tsc_generation++;
1513 		kvm->arch.cur_tsc_nsec = ns;
1514 		kvm->arch.cur_tsc_write = data;
1515 		kvm->arch.cur_tsc_offset = offset;
1516 		matched = false;
1517 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1518 			 kvm->arch.cur_tsc_generation, data);
1519 	}
1520 
1521 	/*
1522 	 * We also track th most recent recorded KHZ, write and time to
1523 	 * allow the matching interval to be extended at each write.
1524 	 */
1525 	kvm->arch.last_tsc_nsec = ns;
1526 	kvm->arch.last_tsc_write = data;
1527 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1528 
1529 	vcpu->arch.last_guest_tsc = data;
1530 
1531 	/* Keep track of which generation this VCPU has synchronized to */
1532 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1533 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1534 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1535 
1536 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1537 		update_ia32_tsc_adjust_msr(vcpu, offset);
1538 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1539 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1540 
1541 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1542 	if (!matched) {
1543 		kvm->arch.nr_vcpus_matched_tsc = 0;
1544 	} else if (!already_matched) {
1545 		kvm->arch.nr_vcpus_matched_tsc++;
1546 	}
1547 
1548 	kvm_track_tsc_matching(vcpu);
1549 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1550 }
1551 
1552 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1553 
1554 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1555 					   s64 adjustment)
1556 {
1557 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1558 }
1559 
1560 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1561 {
1562 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1563 		WARN_ON(adjustment < 0);
1564 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1565 	adjust_tsc_offset_guest(vcpu, adjustment);
1566 }
1567 
1568 #ifdef CONFIG_X86_64
1569 
1570 static u64 read_tsc(void)
1571 {
1572 	u64 ret = (u64)rdtsc_ordered();
1573 	u64 last = pvclock_gtod_data.clock.cycle_last;
1574 
1575 	if (likely(ret >= last))
1576 		return ret;
1577 
1578 	/*
1579 	 * GCC likes to generate cmov here, but this branch is extremely
1580 	 * predictable (it's just a function of time and the likely is
1581 	 * very likely) and there's a data dependence, so force GCC
1582 	 * to generate a branch instead.  I don't barrier() because
1583 	 * we don't actually need a barrier, and if this function
1584 	 * ever gets inlined it will generate worse code.
1585 	 */
1586 	asm volatile ("");
1587 	return last;
1588 }
1589 
1590 static inline u64 vgettsc(u64 *cycle_now)
1591 {
1592 	long v;
1593 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1594 
1595 	*cycle_now = read_tsc();
1596 
1597 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1598 	return v * gtod->clock.mult;
1599 }
1600 
1601 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1602 {
1603 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1604 	unsigned long seq;
1605 	int mode;
1606 	u64 ns;
1607 
1608 	do {
1609 		seq = read_seqcount_begin(&gtod->seq);
1610 		mode = gtod->clock.vclock_mode;
1611 		ns = gtod->nsec_base;
1612 		ns += vgettsc(cycle_now);
1613 		ns >>= gtod->clock.shift;
1614 		ns += gtod->boot_ns;
1615 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1616 	*t = ns;
1617 
1618 	return mode;
1619 }
1620 
1621 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1622 {
1623 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1624 	unsigned long seq;
1625 	int mode;
1626 	u64 ns;
1627 
1628 	do {
1629 		seq = read_seqcount_begin(&gtod->seq);
1630 		mode = gtod->clock.vclock_mode;
1631 		ts->tv_sec = gtod->wall_time_sec;
1632 		ns = gtod->nsec_base;
1633 		ns += vgettsc(cycle_now);
1634 		ns >>= gtod->clock.shift;
1635 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1636 
1637 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1638 	ts->tv_nsec = ns;
1639 
1640 	return mode;
1641 }
1642 
1643 /* returns true if host is using tsc clocksource */
1644 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1645 {
1646 	/* checked again under seqlock below */
1647 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1648 		return false;
1649 
1650 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1651 }
1652 
1653 /* returns true if host is using tsc clocksource */
1654 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1655 					   u64 *cycle_now)
1656 {
1657 	/* checked again under seqlock below */
1658 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1659 		return false;
1660 
1661 	return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1662 }
1663 #endif
1664 
1665 /*
1666  *
1667  * Assuming a stable TSC across physical CPUS, and a stable TSC
1668  * across virtual CPUs, the following condition is possible.
1669  * Each numbered line represents an event visible to both
1670  * CPUs at the next numbered event.
1671  *
1672  * "timespecX" represents host monotonic time. "tscX" represents
1673  * RDTSC value.
1674  *
1675  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1676  *
1677  * 1.  read timespec0,tsc0
1678  * 2.					| timespec1 = timespec0 + N
1679  * 					| tsc1 = tsc0 + M
1680  * 3. transition to guest		| transition to guest
1681  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1682  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1683  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1684  *
1685  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1686  *
1687  * 	- ret0 < ret1
1688  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1689  *		...
1690  *	- 0 < N - M => M < N
1691  *
1692  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1693  * always the case (the difference between two distinct xtime instances
1694  * might be smaller then the difference between corresponding TSC reads,
1695  * when updating guest vcpus pvclock areas).
1696  *
1697  * To avoid that problem, do not allow visibility of distinct
1698  * system_timestamp/tsc_timestamp values simultaneously: use a master
1699  * copy of host monotonic time values. Update that master copy
1700  * in lockstep.
1701  *
1702  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1703  *
1704  */
1705 
1706 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1707 {
1708 #ifdef CONFIG_X86_64
1709 	struct kvm_arch *ka = &kvm->arch;
1710 	int vclock_mode;
1711 	bool host_tsc_clocksource, vcpus_matched;
1712 
1713 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1714 			atomic_read(&kvm->online_vcpus));
1715 
1716 	/*
1717 	 * If the host uses TSC clock, then passthrough TSC as stable
1718 	 * to the guest.
1719 	 */
1720 	host_tsc_clocksource = kvm_get_time_and_clockread(
1721 					&ka->master_kernel_ns,
1722 					&ka->master_cycle_now);
1723 
1724 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1725 				&& !ka->backwards_tsc_observed
1726 				&& !ka->boot_vcpu_runs_old_kvmclock;
1727 
1728 	if (ka->use_master_clock)
1729 		atomic_set(&kvm_guest_has_master_clock, 1);
1730 
1731 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1732 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1733 					vcpus_matched);
1734 #endif
1735 }
1736 
1737 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1738 {
1739 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1740 }
1741 
1742 static void kvm_gen_update_masterclock(struct kvm *kvm)
1743 {
1744 #ifdef CONFIG_X86_64
1745 	int i;
1746 	struct kvm_vcpu *vcpu;
1747 	struct kvm_arch *ka = &kvm->arch;
1748 
1749 	spin_lock(&ka->pvclock_gtod_sync_lock);
1750 	kvm_make_mclock_inprogress_request(kvm);
1751 	/* no guest entries from this point */
1752 	pvclock_update_vm_gtod_copy(kvm);
1753 
1754 	kvm_for_each_vcpu(i, vcpu, kvm)
1755 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1756 
1757 	/* guest entries allowed */
1758 	kvm_for_each_vcpu(i, vcpu, kvm)
1759 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1760 
1761 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1762 #endif
1763 }
1764 
1765 u64 get_kvmclock_ns(struct kvm *kvm)
1766 {
1767 	struct kvm_arch *ka = &kvm->arch;
1768 	struct pvclock_vcpu_time_info hv_clock;
1769 	u64 ret;
1770 
1771 	spin_lock(&ka->pvclock_gtod_sync_lock);
1772 	if (!ka->use_master_clock) {
1773 		spin_unlock(&ka->pvclock_gtod_sync_lock);
1774 		return ktime_get_boot_ns() + ka->kvmclock_offset;
1775 	}
1776 
1777 	hv_clock.tsc_timestamp = ka->master_cycle_now;
1778 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1779 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1780 
1781 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
1782 	get_cpu();
1783 
1784 	kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1785 			   &hv_clock.tsc_shift,
1786 			   &hv_clock.tsc_to_system_mul);
1787 	ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1788 
1789 	put_cpu();
1790 
1791 	return ret;
1792 }
1793 
1794 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1795 {
1796 	struct kvm_vcpu_arch *vcpu = &v->arch;
1797 	struct pvclock_vcpu_time_info guest_hv_clock;
1798 
1799 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1800 		&guest_hv_clock, sizeof(guest_hv_clock))))
1801 		return;
1802 
1803 	/* This VCPU is paused, but it's legal for a guest to read another
1804 	 * VCPU's kvmclock, so we really have to follow the specification where
1805 	 * it says that version is odd if data is being modified, and even after
1806 	 * it is consistent.
1807 	 *
1808 	 * Version field updates must be kept separate.  This is because
1809 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1810 	 * writes within a string instruction are weakly ordered.  So there
1811 	 * are three writes overall.
1812 	 *
1813 	 * As a small optimization, only write the version field in the first
1814 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1815 	 * version field is the first in the struct.
1816 	 */
1817 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1818 
1819 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1820 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1821 				&vcpu->hv_clock,
1822 				sizeof(vcpu->hv_clock.version));
1823 
1824 	smp_wmb();
1825 
1826 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1827 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1828 
1829 	if (vcpu->pvclock_set_guest_stopped_request) {
1830 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1831 		vcpu->pvclock_set_guest_stopped_request = false;
1832 	}
1833 
1834 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1835 
1836 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1837 				&vcpu->hv_clock,
1838 				sizeof(vcpu->hv_clock));
1839 
1840 	smp_wmb();
1841 
1842 	vcpu->hv_clock.version++;
1843 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1844 				&vcpu->hv_clock,
1845 				sizeof(vcpu->hv_clock.version));
1846 }
1847 
1848 static int kvm_guest_time_update(struct kvm_vcpu *v)
1849 {
1850 	unsigned long flags, tgt_tsc_khz;
1851 	struct kvm_vcpu_arch *vcpu = &v->arch;
1852 	struct kvm_arch *ka = &v->kvm->arch;
1853 	s64 kernel_ns;
1854 	u64 tsc_timestamp, host_tsc;
1855 	u8 pvclock_flags;
1856 	bool use_master_clock;
1857 
1858 	kernel_ns = 0;
1859 	host_tsc = 0;
1860 
1861 	/*
1862 	 * If the host uses TSC clock, then passthrough TSC as stable
1863 	 * to the guest.
1864 	 */
1865 	spin_lock(&ka->pvclock_gtod_sync_lock);
1866 	use_master_clock = ka->use_master_clock;
1867 	if (use_master_clock) {
1868 		host_tsc = ka->master_cycle_now;
1869 		kernel_ns = ka->master_kernel_ns;
1870 	}
1871 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1872 
1873 	/* Keep irq disabled to prevent changes to the clock */
1874 	local_irq_save(flags);
1875 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1876 	if (unlikely(tgt_tsc_khz == 0)) {
1877 		local_irq_restore(flags);
1878 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1879 		return 1;
1880 	}
1881 	if (!use_master_clock) {
1882 		host_tsc = rdtsc();
1883 		kernel_ns = ktime_get_boot_ns();
1884 	}
1885 
1886 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1887 
1888 	/*
1889 	 * We may have to catch up the TSC to match elapsed wall clock
1890 	 * time for two reasons, even if kvmclock is used.
1891 	 *   1) CPU could have been running below the maximum TSC rate
1892 	 *   2) Broken TSC compensation resets the base at each VCPU
1893 	 *      entry to avoid unknown leaps of TSC even when running
1894 	 *      again on the same CPU.  This may cause apparent elapsed
1895 	 *      time to disappear, and the guest to stand still or run
1896 	 *	very slowly.
1897 	 */
1898 	if (vcpu->tsc_catchup) {
1899 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1900 		if (tsc > tsc_timestamp) {
1901 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1902 			tsc_timestamp = tsc;
1903 		}
1904 	}
1905 
1906 	local_irq_restore(flags);
1907 
1908 	/* With all the info we got, fill in the values */
1909 
1910 	if (kvm_has_tsc_control)
1911 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1912 
1913 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1914 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1915 				   &vcpu->hv_clock.tsc_shift,
1916 				   &vcpu->hv_clock.tsc_to_system_mul);
1917 		vcpu->hw_tsc_khz = tgt_tsc_khz;
1918 	}
1919 
1920 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1921 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1922 	vcpu->last_guest_tsc = tsc_timestamp;
1923 
1924 	/* If the host uses TSC clocksource, then it is stable */
1925 	pvclock_flags = 0;
1926 	if (use_master_clock)
1927 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1928 
1929 	vcpu->hv_clock.flags = pvclock_flags;
1930 
1931 	if (vcpu->pv_time_enabled)
1932 		kvm_setup_pvclock_page(v);
1933 	if (v == kvm_get_vcpu(v->kvm, 0))
1934 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1935 	return 0;
1936 }
1937 
1938 /*
1939  * kvmclock updates which are isolated to a given vcpu, such as
1940  * vcpu->cpu migration, should not allow system_timestamp from
1941  * the rest of the vcpus to remain static. Otherwise ntp frequency
1942  * correction applies to one vcpu's system_timestamp but not
1943  * the others.
1944  *
1945  * So in those cases, request a kvmclock update for all vcpus.
1946  * We need to rate-limit these requests though, as they can
1947  * considerably slow guests that have a large number of vcpus.
1948  * The time for a remote vcpu to update its kvmclock is bound
1949  * by the delay we use to rate-limit the updates.
1950  */
1951 
1952 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1953 
1954 static void kvmclock_update_fn(struct work_struct *work)
1955 {
1956 	int i;
1957 	struct delayed_work *dwork = to_delayed_work(work);
1958 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1959 					   kvmclock_update_work);
1960 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1961 	struct kvm_vcpu *vcpu;
1962 
1963 	kvm_for_each_vcpu(i, vcpu, kvm) {
1964 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1965 		kvm_vcpu_kick(vcpu);
1966 	}
1967 }
1968 
1969 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1970 {
1971 	struct kvm *kvm = v->kvm;
1972 
1973 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1974 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1975 					KVMCLOCK_UPDATE_DELAY);
1976 }
1977 
1978 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1979 
1980 static void kvmclock_sync_fn(struct work_struct *work)
1981 {
1982 	struct delayed_work *dwork = to_delayed_work(work);
1983 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1984 					   kvmclock_sync_work);
1985 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1986 
1987 	if (!kvmclock_periodic_sync)
1988 		return;
1989 
1990 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1991 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1992 					KVMCLOCK_SYNC_PERIOD);
1993 }
1994 
1995 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1996 {
1997 	u64 mcg_cap = vcpu->arch.mcg_cap;
1998 	unsigned bank_num = mcg_cap & 0xff;
1999 
2000 	switch (msr) {
2001 	case MSR_IA32_MCG_STATUS:
2002 		vcpu->arch.mcg_status = data;
2003 		break;
2004 	case MSR_IA32_MCG_CTL:
2005 		if (!(mcg_cap & MCG_CTL_P))
2006 			return 1;
2007 		if (data != 0 && data != ~(u64)0)
2008 			return -1;
2009 		vcpu->arch.mcg_ctl = data;
2010 		break;
2011 	default:
2012 		if (msr >= MSR_IA32_MC0_CTL &&
2013 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2014 			u32 offset = msr - MSR_IA32_MC0_CTL;
2015 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2016 			 * some Linux kernels though clear bit 10 in bank 4 to
2017 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2018 			 * this to avoid an uncatched #GP in the guest
2019 			 */
2020 			if ((offset & 0x3) == 0 &&
2021 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2022 				return -1;
2023 			vcpu->arch.mce_banks[offset] = data;
2024 			break;
2025 		}
2026 		return 1;
2027 	}
2028 	return 0;
2029 }
2030 
2031 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2032 {
2033 	struct kvm *kvm = vcpu->kvm;
2034 	int lm = is_long_mode(vcpu);
2035 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2036 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2037 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2038 		: kvm->arch.xen_hvm_config.blob_size_32;
2039 	u32 page_num = data & ~PAGE_MASK;
2040 	u64 page_addr = data & PAGE_MASK;
2041 	u8 *page;
2042 	int r;
2043 
2044 	r = -E2BIG;
2045 	if (page_num >= blob_size)
2046 		goto out;
2047 	r = -ENOMEM;
2048 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2049 	if (IS_ERR(page)) {
2050 		r = PTR_ERR(page);
2051 		goto out;
2052 	}
2053 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2054 		goto out_free;
2055 	r = 0;
2056 out_free:
2057 	kfree(page);
2058 out:
2059 	return r;
2060 }
2061 
2062 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2063 {
2064 	gpa_t gpa = data & ~0x3f;
2065 
2066 	/* Bits 3:5 are reserved, Should be zero */
2067 	if (data & 0x38)
2068 		return 1;
2069 
2070 	vcpu->arch.apf.msr_val = data;
2071 
2072 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2073 		kvm_clear_async_pf_completion_queue(vcpu);
2074 		kvm_async_pf_hash_reset(vcpu);
2075 		return 0;
2076 	}
2077 
2078 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2079 					sizeof(u32)))
2080 		return 1;
2081 
2082 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2083 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2084 	kvm_async_pf_wakeup_all(vcpu);
2085 	return 0;
2086 }
2087 
2088 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2089 {
2090 	vcpu->arch.pv_time_enabled = false;
2091 }
2092 
2093 static void record_steal_time(struct kvm_vcpu *vcpu)
2094 {
2095 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2096 		return;
2097 
2098 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2099 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2100 		return;
2101 
2102 	vcpu->arch.st.steal.preempted = 0;
2103 
2104 	if (vcpu->arch.st.steal.version & 1)
2105 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2106 
2107 	vcpu->arch.st.steal.version += 1;
2108 
2109 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2110 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2111 
2112 	smp_wmb();
2113 
2114 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2115 		vcpu->arch.st.last_steal;
2116 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2117 
2118 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2119 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2120 
2121 	smp_wmb();
2122 
2123 	vcpu->arch.st.steal.version += 1;
2124 
2125 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2126 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2127 }
2128 
2129 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2130 {
2131 	bool pr = false;
2132 	u32 msr = msr_info->index;
2133 	u64 data = msr_info->data;
2134 
2135 	switch (msr) {
2136 	case MSR_AMD64_NB_CFG:
2137 	case MSR_IA32_UCODE_REV:
2138 	case MSR_IA32_UCODE_WRITE:
2139 	case MSR_VM_HSAVE_PA:
2140 	case MSR_AMD64_PATCH_LOADER:
2141 	case MSR_AMD64_BU_CFG2:
2142 	case MSR_AMD64_DC_CFG:
2143 		break;
2144 
2145 	case MSR_EFER:
2146 		return set_efer(vcpu, data);
2147 	case MSR_K7_HWCR:
2148 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2149 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2150 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2151 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2152 		if (data != 0) {
2153 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2154 				    data);
2155 			return 1;
2156 		}
2157 		break;
2158 	case MSR_FAM10H_MMIO_CONF_BASE:
2159 		if (data != 0) {
2160 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2161 				    "0x%llx\n", data);
2162 			return 1;
2163 		}
2164 		break;
2165 	case MSR_IA32_DEBUGCTLMSR:
2166 		if (!data) {
2167 			/* We support the non-activated case already */
2168 			break;
2169 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2170 			/* Values other than LBR and BTF are vendor-specific,
2171 			   thus reserved and should throw a #GP */
2172 			return 1;
2173 		}
2174 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2175 			    __func__, data);
2176 		break;
2177 	case 0x200 ... 0x2ff:
2178 		return kvm_mtrr_set_msr(vcpu, msr, data);
2179 	case MSR_IA32_APICBASE:
2180 		return kvm_set_apic_base(vcpu, msr_info);
2181 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2182 		return kvm_x2apic_msr_write(vcpu, msr, data);
2183 	case MSR_IA32_TSCDEADLINE:
2184 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2185 		break;
2186 	case MSR_IA32_TSC_ADJUST:
2187 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2188 			if (!msr_info->host_initiated) {
2189 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2190 				adjust_tsc_offset_guest(vcpu, adj);
2191 			}
2192 			vcpu->arch.ia32_tsc_adjust_msr = data;
2193 		}
2194 		break;
2195 	case MSR_IA32_MISC_ENABLE:
2196 		vcpu->arch.ia32_misc_enable_msr = data;
2197 		break;
2198 	case MSR_IA32_SMBASE:
2199 		if (!msr_info->host_initiated)
2200 			return 1;
2201 		vcpu->arch.smbase = data;
2202 		break;
2203 	case MSR_KVM_WALL_CLOCK_NEW:
2204 	case MSR_KVM_WALL_CLOCK:
2205 		vcpu->kvm->arch.wall_clock = data;
2206 		kvm_write_wall_clock(vcpu->kvm, data);
2207 		break;
2208 	case MSR_KVM_SYSTEM_TIME_NEW:
2209 	case MSR_KVM_SYSTEM_TIME: {
2210 		struct kvm_arch *ka = &vcpu->kvm->arch;
2211 
2212 		kvmclock_reset(vcpu);
2213 
2214 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2215 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2216 
2217 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2218 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2219 
2220 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2221 		}
2222 
2223 		vcpu->arch.time = data;
2224 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2225 
2226 		/* we verify if the enable bit is set... */
2227 		if (!(data & 1))
2228 			break;
2229 
2230 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2231 		     &vcpu->arch.pv_time, data & ~1ULL,
2232 		     sizeof(struct pvclock_vcpu_time_info)))
2233 			vcpu->arch.pv_time_enabled = false;
2234 		else
2235 			vcpu->arch.pv_time_enabled = true;
2236 
2237 		break;
2238 	}
2239 	case MSR_KVM_ASYNC_PF_EN:
2240 		if (kvm_pv_enable_async_pf(vcpu, data))
2241 			return 1;
2242 		break;
2243 	case MSR_KVM_STEAL_TIME:
2244 
2245 		if (unlikely(!sched_info_on()))
2246 			return 1;
2247 
2248 		if (data & KVM_STEAL_RESERVED_MASK)
2249 			return 1;
2250 
2251 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2252 						data & KVM_STEAL_VALID_BITS,
2253 						sizeof(struct kvm_steal_time)))
2254 			return 1;
2255 
2256 		vcpu->arch.st.msr_val = data;
2257 
2258 		if (!(data & KVM_MSR_ENABLED))
2259 			break;
2260 
2261 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2262 
2263 		break;
2264 	case MSR_KVM_PV_EOI_EN:
2265 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2266 			return 1;
2267 		break;
2268 
2269 	case MSR_IA32_MCG_CTL:
2270 	case MSR_IA32_MCG_STATUS:
2271 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2272 		return set_msr_mce(vcpu, msr, data);
2273 
2274 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2275 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2276 		pr = true; /* fall through */
2277 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2278 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2279 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2280 			return kvm_pmu_set_msr(vcpu, msr_info);
2281 
2282 		if (pr || data != 0)
2283 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2284 				    "0x%x data 0x%llx\n", msr, data);
2285 		break;
2286 	case MSR_K7_CLK_CTL:
2287 		/*
2288 		 * Ignore all writes to this no longer documented MSR.
2289 		 * Writes are only relevant for old K7 processors,
2290 		 * all pre-dating SVM, but a recommended workaround from
2291 		 * AMD for these chips. It is possible to specify the
2292 		 * affected processor models on the command line, hence
2293 		 * the need to ignore the workaround.
2294 		 */
2295 		break;
2296 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2297 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2298 	case HV_X64_MSR_CRASH_CTL:
2299 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2300 		return kvm_hv_set_msr_common(vcpu, msr, data,
2301 					     msr_info->host_initiated);
2302 	case MSR_IA32_BBL_CR_CTL3:
2303 		/* Drop writes to this legacy MSR -- see rdmsr
2304 		 * counterpart for further detail.
2305 		 */
2306 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2307 		break;
2308 	case MSR_AMD64_OSVW_ID_LENGTH:
2309 		if (!guest_cpuid_has_osvw(vcpu))
2310 			return 1;
2311 		vcpu->arch.osvw.length = data;
2312 		break;
2313 	case MSR_AMD64_OSVW_STATUS:
2314 		if (!guest_cpuid_has_osvw(vcpu))
2315 			return 1;
2316 		vcpu->arch.osvw.status = data;
2317 		break;
2318 	case MSR_PLATFORM_INFO:
2319 		if (!msr_info->host_initiated ||
2320 		    data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2321 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2322 		     cpuid_fault_enabled(vcpu)))
2323 			return 1;
2324 		vcpu->arch.msr_platform_info = data;
2325 		break;
2326 	case MSR_MISC_FEATURES_ENABLES:
2327 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2328 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2329 		     !supports_cpuid_fault(vcpu)))
2330 			return 1;
2331 		vcpu->arch.msr_misc_features_enables = data;
2332 		break;
2333 	default:
2334 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2335 			return xen_hvm_config(vcpu, data);
2336 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2337 			return kvm_pmu_set_msr(vcpu, msr_info);
2338 		if (!ignore_msrs) {
2339 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2340 				    msr, data);
2341 			return 1;
2342 		} else {
2343 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2344 				    msr, data);
2345 			break;
2346 		}
2347 	}
2348 	return 0;
2349 }
2350 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2351 
2352 
2353 /*
2354  * Reads an msr value (of 'msr_index') into 'pdata'.
2355  * Returns 0 on success, non-0 otherwise.
2356  * Assumes vcpu_load() was already called.
2357  */
2358 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2359 {
2360 	return kvm_x86_ops->get_msr(vcpu, msr);
2361 }
2362 EXPORT_SYMBOL_GPL(kvm_get_msr);
2363 
2364 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2365 {
2366 	u64 data;
2367 	u64 mcg_cap = vcpu->arch.mcg_cap;
2368 	unsigned bank_num = mcg_cap & 0xff;
2369 
2370 	switch (msr) {
2371 	case MSR_IA32_P5_MC_ADDR:
2372 	case MSR_IA32_P5_MC_TYPE:
2373 		data = 0;
2374 		break;
2375 	case MSR_IA32_MCG_CAP:
2376 		data = vcpu->arch.mcg_cap;
2377 		break;
2378 	case MSR_IA32_MCG_CTL:
2379 		if (!(mcg_cap & MCG_CTL_P))
2380 			return 1;
2381 		data = vcpu->arch.mcg_ctl;
2382 		break;
2383 	case MSR_IA32_MCG_STATUS:
2384 		data = vcpu->arch.mcg_status;
2385 		break;
2386 	default:
2387 		if (msr >= MSR_IA32_MC0_CTL &&
2388 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2389 			u32 offset = msr - MSR_IA32_MC0_CTL;
2390 			data = vcpu->arch.mce_banks[offset];
2391 			break;
2392 		}
2393 		return 1;
2394 	}
2395 	*pdata = data;
2396 	return 0;
2397 }
2398 
2399 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2400 {
2401 	switch (msr_info->index) {
2402 	case MSR_IA32_PLATFORM_ID:
2403 	case MSR_IA32_EBL_CR_POWERON:
2404 	case MSR_IA32_DEBUGCTLMSR:
2405 	case MSR_IA32_LASTBRANCHFROMIP:
2406 	case MSR_IA32_LASTBRANCHTOIP:
2407 	case MSR_IA32_LASTINTFROMIP:
2408 	case MSR_IA32_LASTINTTOIP:
2409 	case MSR_K8_SYSCFG:
2410 	case MSR_K8_TSEG_ADDR:
2411 	case MSR_K8_TSEG_MASK:
2412 	case MSR_K7_HWCR:
2413 	case MSR_VM_HSAVE_PA:
2414 	case MSR_K8_INT_PENDING_MSG:
2415 	case MSR_AMD64_NB_CFG:
2416 	case MSR_FAM10H_MMIO_CONF_BASE:
2417 	case MSR_AMD64_BU_CFG2:
2418 	case MSR_IA32_PERF_CTL:
2419 	case MSR_AMD64_DC_CFG:
2420 		msr_info->data = 0;
2421 		break;
2422 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2423 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2424 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2425 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2426 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2427 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2428 		msr_info->data = 0;
2429 		break;
2430 	case MSR_IA32_UCODE_REV:
2431 		msr_info->data = 0x100000000ULL;
2432 		break;
2433 	case MSR_MTRRcap:
2434 	case 0x200 ... 0x2ff:
2435 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2436 	case 0xcd: /* fsb frequency */
2437 		msr_info->data = 3;
2438 		break;
2439 		/*
2440 		 * MSR_EBC_FREQUENCY_ID
2441 		 * Conservative value valid for even the basic CPU models.
2442 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2443 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2444 		 * and 266MHz for model 3, or 4. Set Core Clock
2445 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2446 		 * 31:24) even though these are only valid for CPU
2447 		 * models > 2, however guests may end up dividing or
2448 		 * multiplying by zero otherwise.
2449 		 */
2450 	case MSR_EBC_FREQUENCY_ID:
2451 		msr_info->data = 1 << 24;
2452 		break;
2453 	case MSR_IA32_APICBASE:
2454 		msr_info->data = kvm_get_apic_base(vcpu);
2455 		break;
2456 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2457 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2458 		break;
2459 	case MSR_IA32_TSCDEADLINE:
2460 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2461 		break;
2462 	case MSR_IA32_TSC_ADJUST:
2463 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2464 		break;
2465 	case MSR_IA32_MISC_ENABLE:
2466 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2467 		break;
2468 	case MSR_IA32_SMBASE:
2469 		if (!msr_info->host_initiated)
2470 			return 1;
2471 		msr_info->data = vcpu->arch.smbase;
2472 		break;
2473 	case MSR_IA32_PERF_STATUS:
2474 		/* TSC increment by tick */
2475 		msr_info->data = 1000ULL;
2476 		/* CPU multiplier */
2477 		msr_info->data |= (((uint64_t)4ULL) << 40);
2478 		break;
2479 	case MSR_EFER:
2480 		msr_info->data = vcpu->arch.efer;
2481 		break;
2482 	case MSR_KVM_WALL_CLOCK:
2483 	case MSR_KVM_WALL_CLOCK_NEW:
2484 		msr_info->data = vcpu->kvm->arch.wall_clock;
2485 		break;
2486 	case MSR_KVM_SYSTEM_TIME:
2487 	case MSR_KVM_SYSTEM_TIME_NEW:
2488 		msr_info->data = vcpu->arch.time;
2489 		break;
2490 	case MSR_KVM_ASYNC_PF_EN:
2491 		msr_info->data = vcpu->arch.apf.msr_val;
2492 		break;
2493 	case MSR_KVM_STEAL_TIME:
2494 		msr_info->data = vcpu->arch.st.msr_val;
2495 		break;
2496 	case MSR_KVM_PV_EOI_EN:
2497 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2498 		break;
2499 	case MSR_IA32_P5_MC_ADDR:
2500 	case MSR_IA32_P5_MC_TYPE:
2501 	case MSR_IA32_MCG_CAP:
2502 	case MSR_IA32_MCG_CTL:
2503 	case MSR_IA32_MCG_STATUS:
2504 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2505 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2506 	case MSR_K7_CLK_CTL:
2507 		/*
2508 		 * Provide expected ramp-up count for K7. All other
2509 		 * are set to zero, indicating minimum divisors for
2510 		 * every field.
2511 		 *
2512 		 * This prevents guest kernels on AMD host with CPU
2513 		 * type 6, model 8 and higher from exploding due to
2514 		 * the rdmsr failing.
2515 		 */
2516 		msr_info->data = 0x20000000;
2517 		break;
2518 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2519 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2520 	case HV_X64_MSR_CRASH_CTL:
2521 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2522 		return kvm_hv_get_msr_common(vcpu,
2523 					     msr_info->index, &msr_info->data);
2524 		break;
2525 	case MSR_IA32_BBL_CR_CTL3:
2526 		/* This legacy MSR exists but isn't fully documented in current
2527 		 * silicon.  It is however accessed by winxp in very narrow
2528 		 * scenarios where it sets bit #19, itself documented as
2529 		 * a "reserved" bit.  Best effort attempt to source coherent
2530 		 * read data here should the balance of the register be
2531 		 * interpreted by the guest:
2532 		 *
2533 		 * L2 cache control register 3: 64GB range, 256KB size,
2534 		 * enabled, latency 0x1, configured
2535 		 */
2536 		msr_info->data = 0xbe702111;
2537 		break;
2538 	case MSR_AMD64_OSVW_ID_LENGTH:
2539 		if (!guest_cpuid_has_osvw(vcpu))
2540 			return 1;
2541 		msr_info->data = vcpu->arch.osvw.length;
2542 		break;
2543 	case MSR_AMD64_OSVW_STATUS:
2544 		if (!guest_cpuid_has_osvw(vcpu))
2545 			return 1;
2546 		msr_info->data = vcpu->arch.osvw.status;
2547 		break;
2548 	case MSR_PLATFORM_INFO:
2549 		msr_info->data = vcpu->arch.msr_platform_info;
2550 		break;
2551 	case MSR_MISC_FEATURES_ENABLES:
2552 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2553 		break;
2554 	default:
2555 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2556 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2557 		if (!ignore_msrs) {
2558 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2559 					       msr_info->index);
2560 			return 1;
2561 		} else {
2562 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2563 			msr_info->data = 0;
2564 		}
2565 		break;
2566 	}
2567 	return 0;
2568 }
2569 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2570 
2571 /*
2572  * Read or write a bunch of msrs. All parameters are kernel addresses.
2573  *
2574  * @return number of msrs set successfully.
2575  */
2576 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2577 		    struct kvm_msr_entry *entries,
2578 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2579 				  unsigned index, u64 *data))
2580 {
2581 	int i, idx;
2582 
2583 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2584 	for (i = 0; i < msrs->nmsrs; ++i)
2585 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2586 			break;
2587 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2588 
2589 	return i;
2590 }
2591 
2592 /*
2593  * Read or write a bunch of msrs. Parameters are user addresses.
2594  *
2595  * @return number of msrs set successfully.
2596  */
2597 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2598 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2599 				unsigned index, u64 *data),
2600 		  int writeback)
2601 {
2602 	struct kvm_msrs msrs;
2603 	struct kvm_msr_entry *entries;
2604 	int r, n;
2605 	unsigned size;
2606 
2607 	r = -EFAULT;
2608 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2609 		goto out;
2610 
2611 	r = -E2BIG;
2612 	if (msrs.nmsrs >= MAX_IO_MSRS)
2613 		goto out;
2614 
2615 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2616 	entries = memdup_user(user_msrs->entries, size);
2617 	if (IS_ERR(entries)) {
2618 		r = PTR_ERR(entries);
2619 		goto out;
2620 	}
2621 
2622 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2623 	if (r < 0)
2624 		goto out_free;
2625 
2626 	r = -EFAULT;
2627 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2628 		goto out_free;
2629 
2630 	r = n;
2631 
2632 out_free:
2633 	kfree(entries);
2634 out:
2635 	return r;
2636 }
2637 
2638 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2639 {
2640 	int r;
2641 
2642 	switch (ext) {
2643 	case KVM_CAP_IRQCHIP:
2644 	case KVM_CAP_HLT:
2645 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2646 	case KVM_CAP_SET_TSS_ADDR:
2647 	case KVM_CAP_EXT_CPUID:
2648 	case KVM_CAP_EXT_EMUL_CPUID:
2649 	case KVM_CAP_CLOCKSOURCE:
2650 	case KVM_CAP_PIT:
2651 	case KVM_CAP_NOP_IO_DELAY:
2652 	case KVM_CAP_MP_STATE:
2653 	case KVM_CAP_SYNC_MMU:
2654 	case KVM_CAP_USER_NMI:
2655 	case KVM_CAP_REINJECT_CONTROL:
2656 	case KVM_CAP_IRQ_INJECT_STATUS:
2657 	case KVM_CAP_IOEVENTFD:
2658 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2659 	case KVM_CAP_PIT2:
2660 	case KVM_CAP_PIT_STATE2:
2661 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2662 	case KVM_CAP_XEN_HVM:
2663 	case KVM_CAP_VCPU_EVENTS:
2664 	case KVM_CAP_HYPERV:
2665 	case KVM_CAP_HYPERV_VAPIC:
2666 	case KVM_CAP_HYPERV_SPIN:
2667 	case KVM_CAP_HYPERV_SYNIC:
2668 	case KVM_CAP_HYPERV_SYNIC2:
2669 	case KVM_CAP_HYPERV_VP_INDEX:
2670 	case KVM_CAP_PCI_SEGMENT:
2671 	case KVM_CAP_DEBUGREGS:
2672 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2673 	case KVM_CAP_XSAVE:
2674 	case KVM_CAP_ASYNC_PF:
2675 	case KVM_CAP_GET_TSC_KHZ:
2676 	case KVM_CAP_KVMCLOCK_CTRL:
2677 	case KVM_CAP_READONLY_MEM:
2678 	case KVM_CAP_HYPERV_TIME:
2679 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2680 	case KVM_CAP_TSC_DEADLINE_TIMER:
2681 	case KVM_CAP_ENABLE_CAP_VM:
2682 	case KVM_CAP_DISABLE_QUIRKS:
2683 	case KVM_CAP_SET_BOOT_CPU_ID:
2684  	case KVM_CAP_SPLIT_IRQCHIP:
2685 	case KVM_CAP_IMMEDIATE_EXIT:
2686 		r = 1;
2687 		break;
2688 	case KVM_CAP_ADJUST_CLOCK:
2689 		r = KVM_CLOCK_TSC_STABLE;
2690 		break;
2691 	case KVM_CAP_X86_GUEST_MWAIT:
2692 		r = kvm_mwait_in_guest();
2693 		break;
2694 	case KVM_CAP_X86_SMM:
2695 		/* SMBASE is usually relocated above 1M on modern chipsets,
2696 		 * and SMM handlers might indeed rely on 4G segment limits,
2697 		 * so do not report SMM to be available if real mode is
2698 		 * emulated via vm86 mode.  Still, do not go to great lengths
2699 		 * to avoid userspace's usage of the feature, because it is a
2700 		 * fringe case that is not enabled except via specific settings
2701 		 * of the module parameters.
2702 		 */
2703 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2704 		break;
2705 	case KVM_CAP_VAPIC:
2706 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2707 		break;
2708 	case KVM_CAP_NR_VCPUS:
2709 		r = KVM_SOFT_MAX_VCPUS;
2710 		break;
2711 	case KVM_CAP_MAX_VCPUS:
2712 		r = KVM_MAX_VCPUS;
2713 		break;
2714 	case KVM_CAP_NR_MEMSLOTS:
2715 		r = KVM_USER_MEM_SLOTS;
2716 		break;
2717 	case KVM_CAP_PV_MMU:	/* obsolete */
2718 		r = 0;
2719 		break;
2720 	case KVM_CAP_MCE:
2721 		r = KVM_MAX_MCE_BANKS;
2722 		break;
2723 	case KVM_CAP_XCRS:
2724 		r = boot_cpu_has(X86_FEATURE_XSAVE);
2725 		break;
2726 	case KVM_CAP_TSC_CONTROL:
2727 		r = kvm_has_tsc_control;
2728 		break;
2729 	case KVM_CAP_X2APIC_API:
2730 		r = KVM_X2APIC_API_VALID_FLAGS;
2731 		break;
2732 	default:
2733 		r = 0;
2734 		break;
2735 	}
2736 	return r;
2737 
2738 }
2739 
2740 long kvm_arch_dev_ioctl(struct file *filp,
2741 			unsigned int ioctl, unsigned long arg)
2742 {
2743 	void __user *argp = (void __user *)arg;
2744 	long r;
2745 
2746 	switch (ioctl) {
2747 	case KVM_GET_MSR_INDEX_LIST: {
2748 		struct kvm_msr_list __user *user_msr_list = argp;
2749 		struct kvm_msr_list msr_list;
2750 		unsigned n;
2751 
2752 		r = -EFAULT;
2753 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2754 			goto out;
2755 		n = msr_list.nmsrs;
2756 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2757 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2758 			goto out;
2759 		r = -E2BIG;
2760 		if (n < msr_list.nmsrs)
2761 			goto out;
2762 		r = -EFAULT;
2763 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2764 				 num_msrs_to_save * sizeof(u32)))
2765 			goto out;
2766 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2767 				 &emulated_msrs,
2768 				 num_emulated_msrs * sizeof(u32)))
2769 			goto out;
2770 		r = 0;
2771 		break;
2772 	}
2773 	case KVM_GET_SUPPORTED_CPUID:
2774 	case KVM_GET_EMULATED_CPUID: {
2775 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2776 		struct kvm_cpuid2 cpuid;
2777 
2778 		r = -EFAULT;
2779 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2780 			goto out;
2781 
2782 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2783 					    ioctl);
2784 		if (r)
2785 			goto out;
2786 
2787 		r = -EFAULT;
2788 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2789 			goto out;
2790 		r = 0;
2791 		break;
2792 	}
2793 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2794 		r = -EFAULT;
2795 		if (copy_to_user(argp, &kvm_mce_cap_supported,
2796 				 sizeof(kvm_mce_cap_supported)))
2797 			goto out;
2798 		r = 0;
2799 		break;
2800 	}
2801 	default:
2802 		r = -EINVAL;
2803 	}
2804 out:
2805 	return r;
2806 }
2807 
2808 static void wbinvd_ipi(void *garbage)
2809 {
2810 	wbinvd();
2811 }
2812 
2813 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2814 {
2815 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2816 }
2817 
2818 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2819 {
2820 	/* Address WBINVD may be executed by guest */
2821 	if (need_emulate_wbinvd(vcpu)) {
2822 		if (kvm_x86_ops->has_wbinvd_exit())
2823 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2824 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2825 			smp_call_function_single(vcpu->cpu,
2826 					wbinvd_ipi, NULL, 1);
2827 	}
2828 
2829 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2830 
2831 	/* Apply any externally detected TSC adjustments (due to suspend) */
2832 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2833 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2834 		vcpu->arch.tsc_offset_adjustment = 0;
2835 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2836 	}
2837 
2838 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2839 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2840 				rdtsc() - vcpu->arch.last_host_tsc;
2841 		if (tsc_delta < 0)
2842 			mark_tsc_unstable("KVM discovered backwards TSC");
2843 
2844 		if (check_tsc_unstable()) {
2845 			u64 offset = kvm_compute_tsc_offset(vcpu,
2846 						vcpu->arch.last_guest_tsc);
2847 			kvm_vcpu_write_tsc_offset(vcpu, offset);
2848 			vcpu->arch.tsc_catchup = 1;
2849 		}
2850 
2851 		if (kvm_lapic_hv_timer_in_use(vcpu))
2852 			kvm_lapic_restart_hv_timer(vcpu);
2853 
2854 		/*
2855 		 * On a host with synchronized TSC, there is no need to update
2856 		 * kvmclock on vcpu->cpu migration
2857 		 */
2858 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2859 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2860 		if (vcpu->cpu != cpu)
2861 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2862 		vcpu->cpu = cpu;
2863 	}
2864 
2865 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2866 }
2867 
2868 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2869 {
2870 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2871 		return;
2872 
2873 	vcpu->arch.st.steal.preempted = 1;
2874 
2875 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2876 			&vcpu->arch.st.steal.preempted,
2877 			offsetof(struct kvm_steal_time, preempted),
2878 			sizeof(vcpu->arch.st.steal.preempted));
2879 }
2880 
2881 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2882 {
2883 	int idx;
2884 	/*
2885 	 * Disable page faults because we're in atomic context here.
2886 	 * kvm_write_guest_offset_cached() would call might_fault()
2887 	 * that relies on pagefault_disable() to tell if there's a
2888 	 * bug. NOTE: the write to guest memory may not go through if
2889 	 * during postcopy live migration or if there's heavy guest
2890 	 * paging.
2891 	 */
2892 	pagefault_disable();
2893 	/*
2894 	 * kvm_memslots() will be called by
2895 	 * kvm_write_guest_offset_cached() so take the srcu lock.
2896 	 */
2897 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2898 	kvm_steal_time_set_preempted(vcpu);
2899 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2900 	pagefault_enable();
2901 	kvm_x86_ops->vcpu_put(vcpu);
2902 	kvm_put_guest_fpu(vcpu);
2903 	vcpu->arch.last_host_tsc = rdtsc();
2904 }
2905 
2906 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2907 				    struct kvm_lapic_state *s)
2908 {
2909 	if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2910 		kvm_x86_ops->sync_pir_to_irr(vcpu);
2911 
2912 	return kvm_apic_get_state(vcpu, s);
2913 }
2914 
2915 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2916 				    struct kvm_lapic_state *s)
2917 {
2918 	int r;
2919 
2920 	r = kvm_apic_set_state(vcpu, s);
2921 	if (r)
2922 		return r;
2923 	update_cr8_intercept(vcpu);
2924 
2925 	return 0;
2926 }
2927 
2928 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2929 {
2930 	return (!lapic_in_kernel(vcpu) ||
2931 		kvm_apic_accept_pic_intr(vcpu));
2932 }
2933 
2934 /*
2935  * if userspace requested an interrupt window, check that the
2936  * interrupt window is open.
2937  *
2938  * No need to exit to userspace if we already have an interrupt queued.
2939  */
2940 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2941 {
2942 	return kvm_arch_interrupt_allowed(vcpu) &&
2943 		!kvm_cpu_has_interrupt(vcpu) &&
2944 		!kvm_event_needs_reinjection(vcpu) &&
2945 		kvm_cpu_accept_dm_intr(vcpu);
2946 }
2947 
2948 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2949 				    struct kvm_interrupt *irq)
2950 {
2951 	if (irq->irq >= KVM_NR_INTERRUPTS)
2952 		return -EINVAL;
2953 
2954 	if (!irqchip_in_kernel(vcpu->kvm)) {
2955 		kvm_queue_interrupt(vcpu, irq->irq, false);
2956 		kvm_make_request(KVM_REQ_EVENT, vcpu);
2957 		return 0;
2958 	}
2959 
2960 	/*
2961 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2962 	 * fail for in-kernel 8259.
2963 	 */
2964 	if (pic_in_kernel(vcpu->kvm))
2965 		return -ENXIO;
2966 
2967 	if (vcpu->arch.pending_external_vector != -1)
2968 		return -EEXIST;
2969 
2970 	vcpu->arch.pending_external_vector = irq->irq;
2971 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2972 	return 0;
2973 }
2974 
2975 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2976 {
2977 	kvm_inject_nmi(vcpu);
2978 
2979 	return 0;
2980 }
2981 
2982 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2983 {
2984 	kvm_make_request(KVM_REQ_SMI, vcpu);
2985 
2986 	return 0;
2987 }
2988 
2989 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2990 					   struct kvm_tpr_access_ctl *tac)
2991 {
2992 	if (tac->flags)
2993 		return -EINVAL;
2994 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2995 	return 0;
2996 }
2997 
2998 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2999 					u64 mcg_cap)
3000 {
3001 	int r;
3002 	unsigned bank_num = mcg_cap & 0xff, bank;
3003 
3004 	r = -EINVAL;
3005 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3006 		goto out;
3007 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3008 		goto out;
3009 	r = 0;
3010 	vcpu->arch.mcg_cap = mcg_cap;
3011 	/* Init IA32_MCG_CTL to all 1s */
3012 	if (mcg_cap & MCG_CTL_P)
3013 		vcpu->arch.mcg_ctl = ~(u64)0;
3014 	/* Init IA32_MCi_CTL to all 1s */
3015 	for (bank = 0; bank < bank_num; bank++)
3016 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3017 
3018 	if (kvm_x86_ops->setup_mce)
3019 		kvm_x86_ops->setup_mce(vcpu);
3020 out:
3021 	return r;
3022 }
3023 
3024 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3025 				      struct kvm_x86_mce *mce)
3026 {
3027 	u64 mcg_cap = vcpu->arch.mcg_cap;
3028 	unsigned bank_num = mcg_cap & 0xff;
3029 	u64 *banks = vcpu->arch.mce_banks;
3030 
3031 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3032 		return -EINVAL;
3033 	/*
3034 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3035 	 * reporting is disabled
3036 	 */
3037 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3038 	    vcpu->arch.mcg_ctl != ~(u64)0)
3039 		return 0;
3040 	banks += 4 * mce->bank;
3041 	/*
3042 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3043 	 * reporting is disabled for the bank
3044 	 */
3045 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3046 		return 0;
3047 	if (mce->status & MCI_STATUS_UC) {
3048 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3049 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3050 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3051 			return 0;
3052 		}
3053 		if (banks[1] & MCI_STATUS_VAL)
3054 			mce->status |= MCI_STATUS_OVER;
3055 		banks[2] = mce->addr;
3056 		banks[3] = mce->misc;
3057 		vcpu->arch.mcg_status = mce->mcg_status;
3058 		banks[1] = mce->status;
3059 		kvm_queue_exception(vcpu, MC_VECTOR);
3060 	} else if (!(banks[1] & MCI_STATUS_VAL)
3061 		   || !(banks[1] & MCI_STATUS_UC)) {
3062 		if (banks[1] & MCI_STATUS_VAL)
3063 			mce->status |= MCI_STATUS_OVER;
3064 		banks[2] = mce->addr;
3065 		banks[3] = mce->misc;
3066 		banks[1] = mce->status;
3067 	} else
3068 		banks[1] |= MCI_STATUS_OVER;
3069 	return 0;
3070 }
3071 
3072 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3073 					       struct kvm_vcpu_events *events)
3074 {
3075 	process_nmi(vcpu);
3076 	events->exception.injected =
3077 		vcpu->arch.exception.pending &&
3078 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3079 	events->exception.nr = vcpu->arch.exception.nr;
3080 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3081 	events->exception.pad = 0;
3082 	events->exception.error_code = vcpu->arch.exception.error_code;
3083 
3084 	events->interrupt.injected =
3085 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3086 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3087 	events->interrupt.soft = 0;
3088 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3089 
3090 	events->nmi.injected = vcpu->arch.nmi_injected;
3091 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3092 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3093 	events->nmi.pad = 0;
3094 
3095 	events->sipi_vector = 0; /* never valid when reporting to user space */
3096 
3097 	events->smi.smm = is_smm(vcpu);
3098 	events->smi.pending = vcpu->arch.smi_pending;
3099 	events->smi.smm_inside_nmi =
3100 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3101 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3102 
3103 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3104 			 | KVM_VCPUEVENT_VALID_SHADOW
3105 			 | KVM_VCPUEVENT_VALID_SMM);
3106 	memset(&events->reserved, 0, sizeof(events->reserved));
3107 }
3108 
3109 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3110 
3111 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3112 					      struct kvm_vcpu_events *events)
3113 {
3114 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3115 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3116 			      | KVM_VCPUEVENT_VALID_SHADOW
3117 			      | KVM_VCPUEVENT_VALID_SMM))
3118 		return -EINVAL;
3119 
3120 	if (events->exception.injected &&
3121 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3122 	     is_guest_mode(vcpu)))
3123 		return -EINVAL;
3124 
3125 	/* INITs are latched while in SMM */
3126 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3127 	    (events->smi.smm || events->smi.pending) &&
3128 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
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 	    lapic_in_kernel(vcpu))
3151 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3152 
3153 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3154 		u32 hflags = vcpu->arch.hflags;
3155 		if (events->smi.smm)
3156 			hflags |= HF_SMM_MASK;
3157 		else
3158 			hflags &= ~HF_SMM_MASK;
3159 		kvm_set_hflags(vcpu, hflags);
3160 
3161 		vcpu->arch.smi_pending = events->smi.pending;
3162 		if (events->smi.smm_inside_nmi)
3163 			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3164 		else
3165 			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3166 		if (lapic_in_kernel(vcpu)) {
3167 			if (events->smi.latched_init)
3168 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3169 			else
3170 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3171 		}
3172 	}
3173 
3174 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3175 
3176 	return 0;
3177 }
3178 
3179 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3180 					     struct kvm_debugregs *dbgregs)
3181 {
3182 	unsigned long val;
3183 
3184 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3185 	kvm_get_dr(vcpu, 6, &val);
3186 	dbgregs->dr6 = val;
3187 	dbgregs->dr7 = vcpu->arch.dr7;
3188 	dbgregs->flags = 0;
3189 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3190 }
3191 
3192 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3193 					    struct kvm_debugregs *dbgregs)
3194 {
3195 	if (dbgregs->flags)
3196 		return -EINVAL;
3197 
3198 	if (dbgregs->dr6 & ~0xffffffffull)
3199 		return -EINVAL;
3200 	if (dbgregs->dr7 & ~0xffffffffull)
3201 		return -EINVAL;
3202 
3203 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3204 	kvm_update_dr0123(vcpu);
3205 	vcpu->arch.dr6 = dbgregs->dr6;
3206 	kvm_update_dr6(vcpu);
3207 	vcpu->arch.dr7 = dbgregs->dr7;
3208 	kvm_update_dr7(vcpu);
3209 
3210 	return 0;
3211 }
3212 
3213 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3214 
3215 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3216 {
3217 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3218 	u64 xstate_bv = xsave->header.xfeatures;
3219 	u64 valid;
3220 
3221 	/*
3222 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3223 	 * leaves 0 and 1 in the loop below.
3224 	 */
3225 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3226 
3227 	/* Set XSTATE_BV */
3228 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3229 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3230 
3231 	/*
3232 	 * Copy each region from the possibly compacted offset to the
3233 	 * non-compacted offset.
3234 	 */
3235 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3236 	while (valid) {
3237 		u64 feature = valid & -valid;
3238 		int index = fls64(feature) - 1;
3239 		void *src = get_xsave_addr(xsave, feature);
3240 
3241 		if (src) {
3242 			u32 size, offset, ecx, edx;
3243 			cpuid_count(XSTATE_CPUID, index,
3244 				    &size, &offset, &ecx, &edx);
3245 			memcpy(dest + offset, src, size);
3246 		}
3247 
3248 		valid -= feature;
3249 	}
3250 }
3251 
3252 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3253 {
3254 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3255 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3256 	u64 valid;
3257 
3258 	/*
3259 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3260 	 * leaves 0 and 1 in the loop below.
3261 	 */
3262 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3263 
3264 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3265 	xsave->header.xfeatures = xstate_bv;
3266 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3267 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3268 
3269 	/*
3270 	 * Copy each region from the non-compacted offset to the
3271 	 * possibly compacted offset.
3272 	 */
3273 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3274 	while (valid) {
3275 		u64 feature = valid & -valid;
3276 		int index = fls64(feature) - 1;
3277 		void *dest = get_xsave_addr(xsave, feature);
3278 
3279 		if (dest) {
3280 			u32 size, offset, ecx, edx;
3281 			cpuid_count(XSTATE_CPUID, index,
3282 				    &size, &offset, &ecx, &edx);
3283 			memcpy(dest, src + offset, size);
3284 		}
3285 
3286 		valid -= feature;
3287 	}
3288 }
3289 
3290 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3291 					 struct kvm_xsave *guest_xsave)
3292 {
3293 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3294 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3295 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3296 	} else {
3297 		memcpy(guest_xsave->region,
3298 			&vcpu->arch.guest_fpu.state.fxsave,
3299 			sizeof(struct fxregs_state));
3300 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3301 			XFEATURE_MASK_FPSSE;
3302 	}
3303 }
3304 
3305 #define XSAVE_MXCSR_OFFSET 24
3306 
3307 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3308 					struct kvm_xsave *guest_xsave)
3309 {
3310 	u64 xstate_bv =
3311 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3312 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3313 
3314 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3315 		/*
3316 		 * Here we allow setting states that are not present in
3317 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3318 		 * with old userspace.
3319 		 */
3320 		if (xstate_bv & ~kvm_supported_xcr0() ||
3321 			mxcsr & ~mxcsr_feature_mask)
3322 			return -EINVAL;
3323 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3324 	} else {
3325 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3326 			mxcsr & ~mxcsr_feature_mask)
3327 			return -EINVAL;
3328 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3329 			guest_xsave->region, sizeof(struct fxregs_state));
3330 	}
3331 	return 0;
3332 }
3333 
3334 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3335 					struct kvm_xcrs *guest_xcrs)
3336 {
3337 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3338 		guest_xcrs->nr_xcrs = 0;
3339 		return;
3340 	}
3341 
3342 	guest_xcrs->nr_xcrs = 1;
3343 	guest_xcrs->flags = 0;
3344 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3345 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3346 }
3347 
3348 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3349 				       struct kvm_xcrs *guest_xcrs)
3350 {
3351 	int i, r = 0;
3352 
3353 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3354 		return -EINVAL;
3355 
3356 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3357 		return -EINVAL;
3358 
3359 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3360 		/* Only support XCR0 currently */
3361 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3362 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3363 				guest_xcrs->xcrs[i].value);
3364 			break;
3365 		}
3366 	if (r)
3367 		r = -EINVAL;
3368 	return r;
3369 }
3370 
3371 /*
3372  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3373  * stopped by the hypervisor.  This function will be called from the host only.
3374  * EINVAL is returned when the host attempts to set the flag for a guest that
3375  * does not support pv clocks.
3376  */
3377 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3378 {
3379 	if (!vcpu->arch.pv_time_enabled)
3380 		return -EINVAL;
3381 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3382 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3383 	return 0;
3384 }
3385 
3386 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3387 				     struct kvm_enable_cap *cap)
3388 {
3389 	if (cap->flags)
3390 		return -EINVAL;
3391 
3392 	switch (cap->cap) {
3393 	case KVM_CAP_HYPERV_SYNIC2:
3394 		if (cap->args[0])
3395 			return -EINVAL;
3396 	case KVM_CAP_HYPERV_SYNIC:
3397 		if (!irqchip_in_kernel(vcpu->kvm))
3398 			return -EINVAL;
3399 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3400 					     KVM_CAP_HYPERV_SYNIC2);
3401 	default:
3402 		return -EINVAL;
3403 	}
3404 }
3405 
3406 long kvm_arch_vcpu_ioctl(struct file *filp,
3407 			 unsigned int ioctl, unsigned long arg)
3408 {
3409 	struct kvm_vcpu *vcpu = filp->private_data;
3410 	void __user *argp = (void __user *)arg;
3411 	int r;
3412 	union {
3413 		struct kvm_lapic_state *lapic;
3414 		struct kvm_xsave *xsave;
3415 		struct kvm_xcrs *xcrs;
3416 		void *buffer;
3417 	} u;
3418 
3419 	u.buffer = NULL;
3420 	switch (ioctl) {
3421 	case KVM_GET_LAPIC: {
3422 		r = -EINVAL;
3423 		if (!lapic_in_kernel(vcpu))
3424 			goto out;
3425 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3426 
3427 		r = -ENOMEM;
3428 		if (!u.lapic)
3429 			goto out;
3430 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3431 		if (r)
3432 			goto out;
3433 		r = -EFAULT;
3434 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3435 			goto out;
3436 		r = 0;
3437 		break;
3438 	}
3439 	case KVM_SET_LAPIC: {
3440 		r = -EINVAL;
3441 		if (!lapic_in_kernel(vcpu))
3442 			goto out;
3443 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3444 		if (IS_ERR(u.lapic))
3445 			return PTR_ERR(u.lapic);
3446 
3447 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3448 		break;
3449 	}
3450 	case KVM_INTERRUPT: {
3451 		struct kvm_interrupt irq;
3452 
3453 		r = -EFAULT;
3454 		if (copy_from_user(&irq, argp, sizeof irq))
3455 			goto out;
3456 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3457 		break;
3458 	}
3459 	case KVM_NMI: {
3460 		r = kvm_vcpu_ioctl_nmi(vcpu);
3461 		break;
3462 	}
3463 	case KVM_SMI: {
3464 		r = kvm_vcpu_ioctl_smi(vcpu);
3465 		break;
3466 	}
3467 	case KVM_SET_CPUID: {
3468 		struct kvm_cpuid __user *cpuid_arg = argp;
3469 		struct kvm_cpuid cpuid;
3470 
3471 		r = -EFAULT;
3472 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3473 			goto out;
3474 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3475 		break;
3476 	}
3477 	case KVM_SET_CPUID2: {
3478 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3479 		struct kvm_cpuid2 cpuid;
3480 
3481 		r = -EFAULT;
3482 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3483 			goto out;
3484 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3485 					      cpuid_arg->entries);
3486 		break;
3487 	}
3488 	case KVM_GET_CPUID2: {
3489 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3490 		struct kvm_cpuid2 cpuid;
3491 
3492 		r = -EFAULT;
3493 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3494 			goto out;
3495 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3496 					      cpuid_arg->entries);
3497 		if (r)
3498 			goto out;
3499 		r = -EFAULT;
3500 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3501 			goto out;
3502 		r = 0;
3503 		break;
3504 	}
3505 	case KVM_GET_MSRS:
3506 		r = msr_io(vcpu, argp, do_get_msr, 1);
3507 		break;
3508 	case KVM_SET_MSRS:
3509 		r = msr_io(vcpu, argp, do_set_msr, 0);
3510 		break;
3511 	case KVM_TPR_ACCESS_REPORTING: {
3512 		struct kvm_tpr_access_ctl tac;
3513 
3514 		r = -EFAULT;
3515 		if (copy_from_user(&tac, argp, sizeof tac))
3516 			goto out;
3517 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3518 		if (r)
3519 			goto out;
3520 		r = -EFAULT;
3521 		if (copy_to_user(argp, &tac, sizeof tac))
3522 			goto out;
3523 		r = 0;
3524 		break;
3525 	};
3526 	case KVM_SET_VAPIC_ADDR: {
3527 		struct kvm_vapic_addr va;
3528 		int idx;
3529 
3530 		r = -EINVAL;
3531 		if (!lapic_in_kernel(vcpu))
3532 			goto out;
3533 		r = -EFAULT;
3534 		if (copy_from_user(&va, argp, sizeof va))
3535 			goto out;
3536 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3537 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3538 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3539 		break;
3540 	}
3541 	case KVM_X86_SETUP_MCE: {
3542 		u64 mcg_cap;
3543 
3544 		r = -EFAULT;
3545 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3546 			goto out;
3547 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3548 		break;
3549 	}
3550 	case KVM_X86_SET_MCE: {
3551 		struct kvm_x86_mce mce;
3552 
3553 		r = -EFAULT;
3554 		if (copy_from_user(&mce, argp, sizeof mce))
3555 			goto out;
3556 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3557 		break;
3558 	}
3559 	case KVM_GET_VCPU_EVENTS: {
3560 		struct kvm_vcpu_events events;
3561 
3562 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3563 
3564 		r = -EFAULT;
3565 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3566 			break;
3567 		r = 0;
3568 		break;
3569 	}
3570 	case KVM_SET_VCPU_EVENTS: {
3571 		struct kvm_vcpu_events events;
3572 
3573 		r = -EFAULT;
3574 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3575 			break;
3576 
3577 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3578 		break;
3579 	}
3580 	case KVM_GET_DEBUGREGS: {
3581 		struct kvm_debugregs dbgregs;
3582 
3583 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3584 
3585 		r = -EFAULT;
3586 		if (copy_to_user(argp, &dbgregs,
3587 				 sizeof(struct kvm_debugregs)))
3588 			break;
3589 		r = 0;
3590 		break;
3591 	}
3592 	case KVM_SET_DEBUGREGS: {
3593 		struct kvm_debugregs dbgregs;
3594 
3595 		r = -EFAULT;
3596 		if (copy_from_user(&dbgregs, argp,
3597 				   sizeof(struct kvm_debugregs)))
3598 			break;
3599 
3600 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3601 		break;
3602 	}
3603 	case KVM_GET_XSAVE: {
3604 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3605 		r = -ENOMEM;
3606 		if (!u.xsave)
3607 			break;
3608 
3609 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3610 
3611 		r = -EFAULT;
3612 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3613 			break;
3614 		r = 0;
3615 		break;
3616 	}
3617 	case KVM_SET_XSAVE: {
3618 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3619 		if (IS_ERR(u.xsave))
3620 			return PTR_ERR(u.xsave);
3621 
3622 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3623 		break;
3624 	}
3625 	case KVM_GET_XCRS: {
3626 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3627 		r = -ENOMEM;
3628 		if (!u.xcrs)
3629 			break;
3630 
3631 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3632 
3633 		r = -EFAULT;
3634 		if (copy_to_user(argp, u.xcrs,
3635 				 sizeof(struct kvm_xcrs)))
3636 			break;
3637 		r = 0;
3638 		break;
3639 	}
3640 	case KVM_SET_XCRS: {
3641 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3642 		if (IS_ERR(u.xcrs))
3643 			return PTR_ERR(u.xcrs);
3644 
3645 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3646 		break;
3647 	}
3648 	case KVM_SET_TSC_KHZ: {
3649 		u32 user_tsc_khz;
3650 
3651 		r = -EINVAL;
3652 		user_tsc_khz = (u32)arg;
3653 
3654 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3655 			goto out;
3656 
3657 		if (user_tsc_khz == 0)
3658 			user_tsc_khz = tsc_khz;
3659 
3660 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3661 			r = 0;
3662 
3663 		goto out;
3664 	}
3665 	case KVM_GET_TSC_KHZ: {
3666 		r = vcpu->arch.virtual_tsc_khz;
3667 		goto out;
3668 	}
3669 	case KVM_KVMCLOCK_CTRL: {
3670 		r = kvm_set_guest_paused(vcpu);
3671 		goto out;
3672 	}
3673 	case KVM_ENABLE_CAP: {
3674 		struct kvm_enable_cap cap;
3675 
3676 		r = -EFAULT;
3677 		if (copy_from_user(&cap, argp, sizeof(cap)))
3678 			goto out;
3679 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3680 		break;
3681 	}
3682 	default:
3683 		r = -EINVAL;
3684 	}
3685 out:
3686 	kfree(u.buffer);
3687 	return r;
3688 }
3689 
3690 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3691 {
3692 	return VM_FAULT_SIGBUS;
3693 }
3694 
3695 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3696 {
3697 	int ret;
3698 
3699 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3700 		return -EINVAL;
3701 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3702 	return ret;
3703 }
3704 
3705 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3706 					      u64 ident_addr)
3707 {
3708 	kvm->arch.ept_identity_map_addr = ident_addr;
3709 	return 0;
3710 }
3711 
3712 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3713 					  u32 kvm_nr_mmu_pages)
3714 {
3715 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3716 		return -EINVAL;
3717 
3718 	mutex_lock(&kvm->slots_lock);
3719 
3720 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3721 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3722 
3723 	mutex_unlock(&kvm->slots_lock);
3724 	return 0;
3725 }
3726 
3727 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3728 {
3729 	return kvm->arch.n_max_mmu_pages;
3730 }
3731 
3732 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3733 {
3734 	struct kvm_pic *pic = kvm->arch.vpic;
3735 	int r;
3736 
3737 	r = 0;
3738 	switch (chip->chip_id) {
3739 	case KVM_IRQCHIP_PIC_MASTER:
3740 		memcpy(&chip->chip.pic, &pic->pics[0],
3741 			sizeof(struct kvm_pic_state));
3742 		break;
3743 	case KVM_IRQCHIP_PIC_SLAVE:
3744 		memcpy(&chip->chip.pic, &pic->pics[1],
3745 			sizeof(struct kvm_pic_state));
3746 		break;
3747 	case KVM_IRQCHIP_IOAPIC:
3748 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
3749 		break;
3750 	default:
3751 		r = -EINVAL;
3752 		break;
3753 	}
3754 	return r;
3755 }
3756 
3757 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3758 {
3759 	struct kvm_pic *pic = kvm->arch.vpic;
3760 	int r;
3761 
3762 	r = 0;
3763 	switch (chip->chip_id) {
3764 	case KVM_IRQCHIP_PIC_MASTER:
3765 		spin_lock(&pic->lock);
3766 		memcpy(&pic->pics[0], &chip->chip.pic,
3767 			sizeof(struct kvm_pic_state));
3768 		spin_unlock(&pic->lock);
3769 		break;
3770 	case KVM_IRQCHIP_PIC_SLAVE:
3771 		spin_lock(&pic->lock);
3772 		memcpy(&pic->pics[1], &chip->chip.pic,
3773 			sizeof(struct kvm_pic_state));
3774 		spin_unlock(&pic->lock);
3775 		break;
3776 	case KVM_IRQCHIP_IOAPIC:
3777 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
3778 		break;
3779 	default:
3780 		r = -EINVAL;
3781 		break;
3782 	}
3783 	kvm_pic_update_irq(pic);
3784 	return r;
3785 }
3786 
3787 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3788 {
3789 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3790 
3791 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3792 
3793 	mutex_lock(&kps->lock);
3794 	memcpy(ps, &kps->channels, sizeof(*ps));
3795 	mutex_unlock(&kps->lock);
3796 	return 0;
3797 }
3798 
3799 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3800 {
3801 	int i;
3802 	struct kvm_pit *pit = kvm->arch.vpit;
3803 
3804 	mutex_lock(&pit->pit_state.lock);
3805 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3806 	for (i = 0; i < 3; i++)
3807 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3808 	mutex_unlock(&pit->pit_state.lock);
3809 	return 0;
3810 }
3811 
3812 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3813 {
3814 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3815 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3816 		sizeof(ps->channels));
3817 	ps->flags = kvm->arch.vpit->pit_state.flags;
3818 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3819 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3820 	return 0;
3821 }
3822 
3823 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3824 {
3825 	int start = 0;
3826 	int i;
3827 	u32 prev_legacy, cur_legacy;
3828 	struct kvm_pit *pit = kvm->arch.vpit;
3829 
3830 	mutex_lock(&pit->pit_state.lock);
3831 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3832 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3833 	if (!prev_legacy && cur_legacy)
3834 		start = 1;
3835 	memcpy(&pit->pit_state.channels, &ps->channels,
3836 	       sizeof(pit->pit_state.channels));
3837 	pit->pit_state.flags = ps->flags;
3838 	for (i = 0; i < 3; i++)
3839 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3840 				   start && i == 0);
3841 	mutex_unlock(&pit->pit_state.lock);
3842 	return 0;
3843 }
3844 
3845 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3846 				 struct kvm_reinject_control *control)
3847 {
3848 	struct kvm_pit *pit = kvm->arch.vpit;
3849 
3850 	if (!pit)
3851 		return -ENXIO;
3852 
3853 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
3854 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3855 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3856 	 */
3857 	mutex_lock(&pit->pit_state.lock);
3858 	kvm_pit_set_reinject(pit, control->pit_reinject);
3859 	mutex_unlock(&pit->pit_state.lock);
3860 
3861 	return 0;
3862 }
3863 
3864 /**
3865  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3866  * @kvm: kvm instance
3867  * @log: slot id and address to which we copy the log
3868  *
3869  * Steps 1-4 below provide general overview of dirty page logging. See
3870  * kvm_get_dirty_log_protect() function description for additional details.
3871  *
3872  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3873  * always flush the TLB (step 4) even if previous step failed  and the dirty
3874  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3875  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3876  * writes will be marked dirty for next log read.
3877  *
3878  *   1. Take a snapshot of the bit and clear it if needed.
3879  *   2. Write protect the corresponding page.
3880  *   3. Copy the snapshot to the userspace.
3881  *   4. Flush TLB's if needed.
3882  */
3883 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3884 {
3885 	bool is_dirty = false;
3886 	int r;
3887 
3888 	mutex_lock(&kvm->slots_lock);
3889 
3890 	/*
3891 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3892 	 */
3893 	if (kvm_x86_ops->flush_log_dirty)
3894 		kvm_x86_ops->flush_log_dirty(kvm);
3895 
3896 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3897 
3898 	/*
3899 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3900 	 * kvm_mmu_slot_remove_write_access().
3901 	 */
3902 	lockdep_assert_held(&kvm->slots_lock);
3903 	if (is_dirty)
3904 		kvm_flush_remote_tlbs(kvm);
3905 
3906 	mutex_unlock(&kvm->slots_lock);
3907 	return r;
3908 }
3909 
3910 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3911 			bool line_status)
3912 {
3913 	if (!irqchip_in_kernel(kvm))
3914 		return -ENXIO;
3915 
3916 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3917 					irq_event->irq, irq_event->level,
3918 					line_status);
3919 	return 0;
3920 }
3921 
3922 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3923 				   struct kvm_enable_cap *cap)
3924 {
3925 	int r;
3926 
3927 	if (cap->flags)
3928 		return -EINVAL;
3929 
3930 	switch (cap->cap) {
3931 	case KVM_CAP_DISABLE_QUIRKS:
3932 		kvm->arch.disabled_quirks = cap->args[0];
3933 		r = 0;
3934 		break;
3935 	case KVM_CAP_SPLIT_IRQCHIP: {
3936 		mutex_lock(&kvm->lock);
3937 		r = -EINVAL;
3938 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3939 			goto split_irqchip_unlock;
3940 		r = -EEXIST;
3941 		if (irqchip_in_kernel(kvm))
3942 			goto split_irqchip_unlock;
3943 		if (kvm->created_vcpus)
3944 			goto split_irqchip_unlock;
3945 		r = kvm_setup_empty_irq_routing(kvm);
3946 		if (r)
3947 			goto split_irqchip_unlock;
3948 		/* Pairs with irqchip_in_kernel. */
3949 		smp_wmb();
3950 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3951 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3952 		r = 0;
3953 split_irqchip_unlock:
3954 		mutex_unlock(&kvm->lock);
3955 		break;
3956 	}
3957 	case KVM_CAP_X2APIC_API:
3958 		r = -EINVAL;
3959 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3960 			break;
3961 
3962 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3963 			kvm->arch.x2apic_format = true;
3964 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3965 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
3966 
3967 		r = 0;
3968 		break;
3969 	default:
3970 		r = -EINVAL;
3971 		break;
3972 	}
3973 	return r;
3974 }
3975 
3976 long kvm_arch_vm_ioctl(struct file *filp,
3977 		       unsigned int ioctl, unsigned long arg)
3978 {
3979 	struct kvm *kvm = filp->private_data;
3980 	void __user *argp = (void __user *)arg;
3981 	int r = -ENOTTY;
3982 	/*
3983 	 * This union makes it completely explicit to gcc-3.x
3984 	 * that these two variables' stack usage should be
3985 	 * combined, not added together.
3986 	 */
3987 	union {
3988 		struct kvm_pit_state ps;
3989 		struct kvm_pit_state2 ps2;
3990 		struct kvm_pit_config pit_config;
3991 	} u;
3992 
3993 	switch (ioctl) {
3994 	case KVM_SET_TSS_ADDR:
3995 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3996 		break;
3997 	case KVM_SET_IDENTITY_MAP_ADDR: {
3998 		u64 ident_addr;
3999 
4000 		r = -EFAULT;
4001 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4002 			goto out;
4003 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4004 		break;
4005 	}
4006 	case KVM_SET_NR_MMU_PAGES:
4007 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4008 		break;
4009 	case KVM_GET_NR_MMU_PAGES:
4010 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4011 		break;
4012 	case KVM_CREATE_IRQCHIP: {
4013 		mutex_lock(&kvm->lock);
4014 
4015 		r = -EEXIST;
4016 		if (irqchip_in_kernel(kvm))
4017 			goto create_irqchip_unlock;
4018 
4019 		r = -EINVAL;
4020 		if (kvm->created_vcpus)
4021 			goto create_irqchip_unlock;
4022 
4023 		r = kvm_pic_init(kvm);
4024 		if (r)
4025 			goto create_irqchip_unlock;
4026 
4027 		r = kvm_ioapic_init(kvm);
4028 		if (r) {
4029 			kvm_pic_destroy(kvm);
4030 			goto create_irqchip_unlock;
4031 		}
4032 
4033 		r = kvm_setup_default_irq_routing(kvm);
4034 		if (r) {
4035 			kvm_ioapic_destroy(kvm);
4036 			kvm_pic_destroy(kvm);
4037 			goto create_irqchip_unlock;
4038 		}
4039 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4040 		smp_wmb();
4041 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4042 	create_irqchip_unlock:
4043 		mutex_unlock(&kvm->lock);
4044 		break;
4045 	}
4046 	case KVM_CREATE_PIT:
4047 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4048 		goto create_pit;
4049 	case KVM_CREATE_PIT2:
4050 		r = -EFAULT;
4051 		if (copy_from_user(&u.pit_config, argp,
4052 				   sizeof(struct kvm_pit_config)))
4053 			goto out;
4054 	create_pit:
4055 		mutex_lock(&kvm->lock);
4056 		r = -EEXIST;
4057 		if (kvm->arch.vpit)
4058 			goto create_pit_unlock;
4059 		r = -ENOMEM;
4060 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4061 		if (kvm->arch.vpit)
4062 			r = 0;
4063 	create_pit_unlock:
4064 		mutex_unlock(&kvm->lock);
4065 		break;
4066 	case KVM_GET_IRQCHIP: {
4067 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4068 		struct kvm_irqchip *chip;
4069 
4070 		chip = memdup_user(argp, sizeof(*chip));
4071 		if (IS_ERR(chip)) {
4072 			r = PTR_ERR(chip);
4073 			goto out;
4074 		}
4075 
4076 		r = -ENXIO;
4077 		if (!irqchip_kernel(kvm))
4078 			goto get_irqchip_out;
4079 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4080 		if (r)
4081 			goto get_irqchip_out;
4082 		r = -EFAULT;
4083 		if (copy_to_user(argp, chip, sizeof *chip))
4084 			goto get_irqchip_out;
4085 		r = 0;
4086 	get_irqchip_out:
4087 		kfree(chip);
4088 		break;
4089 	}
4090 	case KVM_SET_IRQCHIP: {
4091 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4092 		struct kvm_irqchip *chip;
4093 
4094 		chip = memdup_user(argp, sizeof(*chip));
4095 		if (IS_ERR(chip)) {
4096 			r = PTR_ERR(chip);
4097 			goto out;
4098 		}
4099 
4100 		r = -ENXIO;
4101 		if (!irqchip_kernel(kvm))
4102 			goto set_irqchip_out;
4103 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4104 		if (r)
4105 			goto set_irqchip_out;
4106 		r = 0;
4107 	set_irqchip_out:
4108 		kfree(chip);
4109 		break;
4110 	}
4111 	case KVM_GET_PIT: {
4112 		r = -EFAULT;
4113 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4114 			goto out;
4115 		r = -ENXIO;
4116 		if (!kvm->arch.vpit)
4117 			goto out;
4118 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4119 		if (r)
4120 			goto out;
4121 		r = -EFAULT;
4122 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4123 			goto out;
4124 		r = 0;
4125 		break;
4126 	}
4127 	case KVM_SET_PIT: {
4128 		r = -EFAULT;
4129 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4130 			goto out;
4131 		r = -ENXIO;
4132 		if (!kvm->arch.vpit)
4133 			goto out;
4134 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4135 		break;
4136 	}
4137 	case KVM_GET_PIT2: {
4138 		r = -ENXIO;
4139 		if (!kvm->arch.vpit)
4140 			goto out;
4141 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4142 		if (r)
4143 			goto out;
4144 		r = -EFAULT;
4145 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4146 			goto out;
4147 		r = 0;
4148 		break;
4149 	}
4150 	case KVM_SET_PIT2: {
4151 		r = -EFAULT;
4152 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4153 			goto out;
4154 		r = -ENXIO;
4155 		if (!kvm->arch.vpit)
4156 			goto out;
4157 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4158 		break;
4159 	}
4160 	case KVM_REINJECT_CONTROL: {
4161 		struct kvm_reinject_control control;
4162 		r =  -EFAULT;
4163 		if (copy_from_user(&control, argp, sizeof(control)))
4164 			goto out;
4165 		r = kvm_vm_ioctl_reinject(kvm, &control);
4166 		break;
4167 	}
4168 	case KVM_SET_BOOT_CPU_ID:
4169 		r = 0;
4170 		mutex_lock(&kvm->lock);
4171 		if (kvm->created_vcpus)
4172 			r = -EBUSY;
4173 		else
4174 			kvm->arch.bsp_vcpu_id = arg;
4175 		mutex_unlock(&kvm->lock);
4176 		break;
4177 	case KVM_XEN_HVM_CONFIG: {
4178 		r = -EFAULT;
4179 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4180 				   sizeof(struct kvm_xen_hvm_config)))
4181 			goto out;
4182 		r = -EINVAL;
4183 		if (kvm->arch.xen_hvm_config.flags)
4184 			goto out;
4185 		r = 0;
4186 		break;
4187 	}
4188 	case KVM_SET_CLOCK: {
4189 		struct kvm_clock_data user_ns;
4190 		u64 now_ns;
4191 
4192 		r = -EFAULT;
4193 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4194 			goto out;
4195 
4196 		r = -EINVAL;
4197 		if (user_ns.flags)
4198 			goto out;
4199 
4200 		r = 0;
4201 		/*
4202 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4203 		 * kvm_gen_update_masterclock() can be cut down to locked
4204 		 * pvclock_update_vm_gtod_copy().
4205 		 */
4206 		kvm_gen_update_masterclock(kvm);
4207 		now_ns = get_kvmclock_ns(kvm);
4208 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4209 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4210 		break;
4211 	}
4212 	case KVM_GET_CLOCK: {
4213 		struct kvm_clock_data user_ns;
4214 		u64 now_ns;
4215 
4216 		now_ns = get_kvmclock_ns(kvm);
4217 		user_ns.clock = now_ns;
4218 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4219 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4220 
4221 		r = -EFAULT;
4222 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4223 			goto out;
4224 		r = 0;
4225 		break;
4226 	}
4227 	case KVM_ENABLE_CAP: {
4228 		struct kvm_enable_cap cap;
4229 
4230 		r = -EFAULT;
4231 		if (copy_from_user(&cap, argp, sizeof(cap)))
4232 			goto out;
4233 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4234 		break;
4235 	}
4236 	default:
4237 		r = -ENOTTY;
4238 	}
4239 out:
4240 	return r;
4241 }
4242 
4243 static void kvm_init_msr_list(void)
4244 {
4245 	u32 dummy[2];
4246 	unsigned i, j;
4247 
4248 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4249 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4250 			continue;
4251 
4252 		/*
4253 		 * Even MSRs that are valid in the host may not be exposed
4254 		 * to the guests in some cases.
4255 		 */
4256 		switch (msrs_to_save[i]) {
4257 		case MSR_IA32_BNDCFGS:
4258 			if (!kvm_x86_ops->mpx_supported())
4259 				continue;
4260 			break;
4261 		case MSR_TSC_AUX:
4262 			if (!kvm_x86_ops->rdtscp_supported())
4263 				continue;
4264 			break;
4265 		default:
4266 			break;
4267 		}
4268 
4269 		if (j < i)
4270 			msrs_to_save[j] = msrs_to_save[i];
4271 		j++;
4272 	}
4273 	num_msrs_to_save = j;
4274 
4275 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4276 		switch (emulated_msrs[i]) {
4277 		case MSR_IA32_SMBASE:
4278 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4279 				continue;
4280 			break;
4281 		default:
4282 			break;
4283 		}
4284 
4285 		if (j < i)
4286 			emulated_msrs[j] = emulated_msrs[i];
4287 		j++;
4288 	}
4289 	num_emulated_msrs = j;
4290 }
4291 
4292 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4293 			   const void *v)
4294 {
4295 	int handled = 0;
4296 	int n;
4297 
4298 	do {
4299 		n = min(len, 8);
4300 		if (!(lapic_in_kernel(vcpu) &&
4301 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4302 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4303 			break;
4304 		handled += n;
4305 		addr += n;
4306 		len -= n;
4307 		v += n;
4308 	} while (len);
4309 
4310 	return handled;
4311 }
4312 
4313 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4314 {
4315 	int handled = 0;
4316 	int n;
4317 
4318 	do {
4319 		n = min(len, 8);
4320 		if (!(lapic_in_kernel(vcpu) &&
4321 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4322 					 addr, n, v))
4323 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4324 			break;
4325 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4326 		handled += n;
4327 		addr += n;
4328 		len -= n;
4329 		v += n;
4330 	} while (len);
4331 
4332 	return handled;
4333 }
4334 
4335 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4336 			struct kvm_segment *var, int seg)
4337 {
4338 	kvm_x86_ops->set_segment(vcpu, var, seg);
4339 }
4340 
4341 void kvm_get_segment(struct kvm_vcpu *vcpu,
4342 		     struct kvm_segment *var, int seg)
4343 {
4344 	kvm_x86_ops->get_segment(vcpu, var, seg);
4345 }
4346 
4347 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4348 			   struct x86_exception *exception)
4349 {
4350 	gpa_t t_gpa;
4351 
4352 	BUG_ON(!mmu_is_nested(vcpu));
4353 
4354 	/* NPT walks are always user-walks */
4355 	access |= PFERR_USER_MASK;
4356 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4357 
4358 	return t_gpa;
4359 }
4360 
4361 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4362 			      struct x86_exception *exception)
4363 {
4364 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4365 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4366 }
4367 
4368  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4369 				struct x86_exception *exception)
4370 {
4371 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4372 	access |= PFERR_FETCH_MASK;
4373 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4374 }
4375 
4376 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4377 			       struct x86_exception *exception)
4378 {
4379 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4380 	access |= PFERR_WRITE_MASK;
4381 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4382 }
4383 
4384 /* uses this to access any guest's mapped memory without checking CPL */
4385 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4386 				struct x86_exception *exception)
4387 {
4388 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4389 }
4390 
4391 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4392 				      struct kvm_vcpu *vcpu, u32 access,
4393 				      struct x86_exception *exception)
4394 {
4395 	void *data = val;
4396 	int r = X86EMUL_CONTINUE;
4397 
4398 	while (bytes) {
4399 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4400 							    exception);
4401 		unsigned offset = addr & (PAGE_SIZE-1);
4402 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4403 		int ret;
4404 
4405 		if (gpa == UNMAPPED_GVA)
4406 			return X86EMUL_PROPAGATE_FAULT;
4407 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4408 					       offset, toread);
4409 		if (ret < 0) {
4410 			r = X86EMUL_IO_NEEDED;
4411 			goto out;
4412 		}
4413 
4414 		bytes -= toread;
4415 		data += toread;
4416 		addr += toread;
4417 	}
4418 out:
4419 	return r;
4420 }
4421 
4422 /* used for instruction fetching */
4423 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4424 				gva_t addr, void *val, unsigned int bytes,
4425 				struct x86_exception *exception)
4426 {
4427 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4428 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4429 	unsigned offset;
4430 	int ret;
4431 
4432 	/* Inline kvm_read_guest_virt_helper for speed.  */
4433 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4434 						    exception);
4435 	if (unlikely(gpa == UNMAPPED_GVA))
4436 		return X86EMUL_PROPAGATE_FAULT;
4437 
4438 	offset = addr & (PAGE_SIZE-1);
4439 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4440 		bytes = (unsigned)PAGE_SIZE - offset;
4441 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4442 				       offset, bytes);
4443 	if (unlikely(ret < 0))
4444 		return X86EMUL_IO_NEEDED;
4445 
4446 	return X86EMUL_CONTINUE;
4447 }
4448 
4449 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4450 			       gva_t addr, void *val, unsigned int bytes,
4451 			       struct x86_exception *exception)
4452 {
4453 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4454 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4455 
4456 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4457 					  exception);
4458 }
4459 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4460 
4461 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4462 				      gva_t addr, void *val, unsigned int bytes,
4463 				      struct x86_exception *exception)
4464 {
4465 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4466 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4467 }
4468 
4469 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4470 		unsigned long addr, void *val, unsigned int bytes)
4471 {
4472 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4473 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4474 
4475 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4476 }
4477 
4478 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4479 				       gva_t addr, void *val,
4480 				       unsigned int bytes,
4481 				       struct x86_exception *exception)
4482 {
4483 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4484 	void *data = val;
4485 	int r = X86EMUL_CONTINUE;
4486 
4487 	while (bytes) {
4488 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4489 							     PFERR_WRITE_MASK,
4490 							     exception);
4491 		unsigned offset = addr & (PAGE_SIZE-1);
4492 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4493 		int ret;
4494 
4495 		if (gpa == UNMAPPED_GVA)
4496 			return X86EMUL_PROPAGATE_FAULT;
4497 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4498 		if (ret < 0) {
4499 			r = X86EMUL_IO_NEEDED;
4500 			goto out;
4501 		}
4502 
4503 		bytes -= towrite;
4504 		data += towrite;
4505 		addr += towrite;
4506 	}
4507 out:
4508 	return r;
4509 }
4510 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4511 
4512 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4513 			    gpa_t gpa, bool write)
4514 {
4515 	/* For APIC access vmexit */
4516 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4517 		return 1;
4518 
4519 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4520 		trace_vcpu_match_mmio(gva, gpa, write, true);
4521 		return 1;
4522 	}
4523 
4524 	return 0;
4525 }
4526 
4527 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4528 				gpa_t *gpa, struct x86_exception *exception,
4529 				bool write)
4530 {
4531 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4532 		| (write ? PFERR_WRITE_MASK : 0);
4533 
4534 	/*
4535 	 * currently PKRU is only applied to ept enabled guest so
4536 	 * there is no pkey in EPT page table for L1 guest or EPT
4537 	 * shadow page table for L2 guest.
4538 	 */
4539 	if (vcpu_match_mmio_gva(vcpu, gva)
4540 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4541 				 vcpu->arch.access, 0, access)) {
4542 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4543 					(gva & (PAGE_SIZE - 1));
4544 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4545 		return 1;
4546 	}
4547 
4548 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4549 
4550 	if (*gpa == UNMAPPED_GVA)
4551 		return -1;
4552 
4553 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4554 }
4555 
4556 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4557 			const void *val, int bytes)
4558 {
4559 	int ret;
4560 
4561 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4562 	if (ret < 0)
4563 		return 0;
4564 	kvm_page_track_write(vcpu, gpa, val, bytes);
4565 	return 1;
4566 }
4567 
4568 struct read_write_emulator_ops {
4569 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4570 				  int bytes);
4571 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4572 				  void *val, int bytes);
4573 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4574 			       int bytes, void *val);
4575 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4576 				    void *val, int bytes);
4577 	bool write;
4578 };
4579 
4580 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4581 {
4582 	if (vcpu->mmio_read_completed) {
4583 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4584 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4585 		vcpu->mmio_read_completed = 0;
4586 		return 1;
4587 	}
4588 
4589 	return 0;
4590 }
4591 
4592 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4593 			void *val, int bytes)
4594 {
4595 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4596 }
4597 
4598 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4599 			 void *val, int bytes)
4600 {
4601 	return emulator_write_phys(vcpu, gpa, val, bytes);
4602 }
4603 
4604 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4605 {
4606 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4607 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4608 }
4609 
4610 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4611 			  void *val, int bytes)
4612 {
4613 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4614 	return X86EMUL_IO_NEEDED;
4615 }
4616 
4617 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4618 			   void *val, int bytes)
4619 {
4620 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4621 
4622 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4623 	return X86EMUL_CONTINUE;
4624 }
4625 
4626 static const struct read_write_emulator_ops read_emultor = {
4627 	.read_write_prepare = read_prepare,
4628 	.read_write_emulate = read_emulate,
4629 	.read_write_mmio = vcpu_mmio_read,
4630 	.read_write_exit_mmio = read_exit_mmio,
4631 };
4632 
4633 static const struct read_write_emulator_ops write_emultor = {
4634 	.read_write_emulate = write_emulate,
4635 	.read_write_mmio = write_mmio,
4636 	.read_write_exit_mmio = write_exit_mmio,
4637 	.write = true,
4638 };
4639 
4640 static int emulator_read_write_onepage(unsigned long addr, void *val,
4641 				       unsigned int bytes,
4642 				       struct x86_exception *exception,
4643 				       struct kvm_vcpu *vcpu,
4644 				       const struct read_write_emulator_ops *ops)
4645 {
4646 	gpa_t gpa;
4647 	int handled, ret;
4648 	bool write = ops->write;
4649 	struct kvm_mmio_fragment *frag;
4650 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4651 
4652 	/*
4653 	 * If the exit was due to a NPF we may already have a GPA.
4654 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4655 	 * Note, this cannot be used on string operations since string
4656 	 * operation using rep will only have the initial GPA from the NPF
4657 	 * occurred.
4658 	 */
4659 	if (vcpu->arch.gpa_available &&
4660 	    emulator_can_use_gpa(ctxt) &&
4661 	    vcpu_is_mmio_gpa(vcpu, addr, exception->address, write) &&
4662 	    (addr & ~PAGE_MASK) == (exception->address & ~PAGE_MASK)) {
4663 		gpa = exception->address;
4664 		goto mmio;
4665 	}
4666 
4667 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4668 
4669 	if (ret < 0)
4670 		return X86EMUL_PROPAGATE_FAULT;
4671 
4672 	/* For APIC access vmexit */
4673 	if (ret)
4674 		goto mmio;
4675 
4676 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4677 		return X86EMUL_CONTINUE;
4678 
4679 mmio:
4680 	/*
4681 	 * Is this MMIO handled locally?
4682 	 */
4683 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4684 	if (handled == bytes)
4685 		return X86EMUL_CONTINUE;
4686 
4687 	gpa += handled;
4688 	bytes -= handled;
4689 	val += handled;
4690 
4691 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4692 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4693 	frag->gpa = gpa;
4694 	frag->data = val;
4695 	frag->len = bytes;
4696 	return X86EMUL_CONTINUE;
4697 }
4698 
4699 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4700 			unsigned long addr,
4701 			void *val, unsigned int bytes,
4702 			struct x86_exception *exception,
4703 			const struct read_write_emulator_ops *ops)
4704 {
4705 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4706 	gpa_t gpa;
4707 	int rc;
4708 
4709 	if (ops->read_write_prepare &&
4710 		  ops->read_write_prepare(vcpu, val, bytes))
4711 		return X86EMUL_CONTINUE;
4712 
4713 	vcpu->mmio_nr_fragments = 0;
4714 
4715 	/* Crossing a page boundary? */
4716 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4717 		int now;
4718 
4719 		now = -addr & ~PAGE_MASK;
4720 		rc = emulator_read_write_onepage(addr, val, now, exception,
4721 						 vcpu, ops);
4722 
4723 		if (rc != X86EMUL_CONTINUE)
4724 			return rc;
4725 		addr += now;
4726 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4727 			addr = (u32)addr;
4728 		val += now;
4729 		bytes -= now;
4730 	}
4731 
4732 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4733 					 vcpu, ops);
4734 	if (rc != X86EMUL_CONTINUE)
4735 		return rc;
4736 
4737 	if (!vcpu->mmio_nr_fragments)
4738 		return rc;
4739 
4740 	gpa = vcpu->mmio_fragments[0].gpa;
4741 
4742 	vcpu->mmio_needed = 1;
4743 	vcpu->mmio_cur_fragment = 0;
4744 
4745 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4746 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4747 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4748 	vcpu->run->mmio.phys_addr = gpa;
4749 
4750 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4751 }
4752 
4753 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4754 				  unsigned long addr,
4755 				  void *val,
4756 				  unsigned int bytes,
4757 				  struct x86_exception *exception)
4758 {
4759 	return emulator_read_write(ctxt, addr, val, bytes,
4760 				   exception, &read_emultor);
4761 }
4762 
4763 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4764 			    unsigned long addr,
4765 			    const void *val,
4766 			    unsigned int bytes,
4767 			    struct x86_exception *exception)
4768 {
4769 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4770 				   exception, &write_emultor);
4771 }
4772 
4773 #define CMPXCHG_TYPE(t, ptr, old, new) \
4774 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4775 
4776 #ifdef CONFIG_X86_64
4777 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4778 #else
4779 #  define CMPXCHG64(ptr, old, new) \
4780 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4781 #endif
4782 
4783 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4784 				     unsigned long addr,
4785 				     const void *old,
4786 				     const void *new,
4787 				     unsigned int bytes,
4788 				     struct x86_exception *exception)
4789 {
4790 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4791 	gpa_t gpa;
4792 	struct page *page;
4793 	char *kaddr;
4794 	bool exchanged;
4795 
4796 	/* guests cmpxchg8b have to be emulated atomically */
4797 	if (bytes > 8 || (bytes & (bytes - 1)))
4798 		goto emul_write;
4799 
4800 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4801 
4802 	if (gpa == UNMAPPED_GVA ||
4803 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4804 		goto emul_write;
4805 
4806 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4807 		goto emul_write;
4808 
4809 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4810 	if (is_error_page(page))
4811 		goto emul_write;
4812 
4813 	kaddr = kmap_atomic(page);
4814 	kaddr += offset_in_page(gpa);
4815 	switch (bytes) {
4816 	case 1:
4817 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4818 		break;
4819 	case 2:
4820 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4821 		break;
4822 	case 4:
4823 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4824 		break;
4825 	case 8:
4826 		exchanged = CMPXCHG64(kaddr, old, new);
4827 		break;
4828 	default:
4829 		BUG();
4830 	}
4831 	kunmap_atomic(kaddr);
4832 	kvm_release_page_dirty(page);
4833 
4834 	if (!exchanged)
4835 		return X86EMUL_CMPXCHG_FAILED;
4836 
4837 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4838 	kvm_page_track_write(vcpu, gpa, new, bytes);
4839 
4840 	return X86EMUL_CONTINUE;
4841 
4842 emul_write:
4843 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4844 
4845 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4846 }
4847 
4848 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4849 {
4850 	int r = 0, i;
4851 
4852 	for (i = 0; i < vcpu->arch.pio.count; i++) {
4853 		if (vcpu->arch.pio.in)
4854 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4855 					    vcpu->arch.pio.size, pd);
4856 		else
4857 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4858 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
4859 					     pd);
4860 		if (r)
4861 			break;
4862 		pd += vcpu->arch.pio.size;
4863 	}
4864 	return r;
4865 }
4866 
4867 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4868 			       unsigned short port, void *val,
4869 			       unsigned int count, bool in)
4870 {
4871 	vcpu->arch.pio.port = port;
4872 	vcpu->arch.pio.in = in;
4873 	vcpu->arch.pio.count  = count;
4874 	vcpu->arch.pio.size = size;
4875 
4876 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4877 		vcpu->arch.pio.count = 0;
4878 		return 1;
4879 	}
4880 
4881 	vcpu->run->exit_reason = KVM_EXIT_IO;
4882 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4883 	vcpu->run->io.size = size;
4884 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4885 	vcpu->run->io.count = count;
4886 	vcpu->run->io.port = port;
4887 
4888 	return 0;
4889 }
4890 
4891 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4892 				    int size, unsigned short port, void *val,
4893 				    unsigned int count)
4894 {
4895 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4896 	int ret;
4897 
4898 	if (vcpu->arch.pio.count)
4899 		goto data_avail;
4900 
4901 	memset(vcpu->arch.pio_data, 0, size * count);
4902 
4903 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4904 	if (ret) {
4905 data_avail:
4906 		memcpy(val, vcpu->arch.pio_data, size * count);
4907 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4908 		vcpu->arch.pio.count = 0;
4909 		return 1;
4910 	}
4911 
4912 	return 0;
4913 }
4914 
4915 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4916 				     int size, unsigned short port,
4917 				     const void *val, unsigned int count)
4918 {
4919 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4920 
4921 	memcpy(vcpu->arch.pio_data, val, size * count);
4922 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4923 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4924 }
4925 
4926 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4927 {
4928 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4929 }
4930 
4931 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4932 {
4933 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4934 }
4935 
4936 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4937 {
4938 	if (!need_emulate_wbinvd(vcpu))
4939 		return X86EMUL_CONTINUE;
4940 
4941 	if (kvm_x86_ops->has_wbinvd_exit()) {
4942 		int cpu = get_cpu();
4943 
4944 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4945 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4946 				wbinvd_ipi, NULL, 1);
4947 		put_cpu();
4948 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4949 	} else
4950 		wbinvd();
4951 	return X86EMUL_CONTINUE;
4952 }
4953 
4954 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4955 {
4956 	kvm_emulate_wbinvd_noskip(vcpu);
4957 	return kvm_skip_emulated_instruction(vcpu);
4958 }
4959 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4960 
4961 
4962 
4963 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4964 {
4965 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4966 }
4967 
4968 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4969 			   unsigned long *dest)
4970 {
4971 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4972 }
4973 
4974 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4975 			   unsigned long value)
4976 {
4977 
4978 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4979 }
4980 
4981 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4982 {
4983 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4984 }
4985 
4986 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4987 {
4988 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4989 	unsigned long value;
4990 
4991 	switch (cr) {
4992 	case 0:
4993 		value = kvm_read_cr0(vcpu);
4994 		break;
4995 	case 2:
4996 		value = vcpu->arch.cr2;
4997 		break;
4998 	case 3:
4999 		value = kvm_read_cr3(vcpu);
5000 		break;
5001 	case 4:
5002 		value = kvm_read_cr4(vcpu);
5003 		break;
5004 	case 8:
5005 		value = kvm_get_cr8(vcpu);
5006 		break;
5007 	default:
5008 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5009 		return 0;
5010 	}
5011 
5012 	return value;
5013 }
5014 
5015 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5016 {
5017 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5018 	int res = 0;
5019 
5020 	switch (cr) {
5021 	case 0:
5022 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5023 		break;
5024 	case 2:
5025 		vcpu->arch.cr2 = val;
5026 		break;
5027 	case 3:
5028 		res = kvm_set_cr3(vcpu, val);
5029 		break;
5030 	case 4:
5031 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5032 		break;
5033 	case 8:
5034 		res = kvm_set_cr8(vcpu, val);
5035 		break;
5036 	default:
5037 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5038 		res = -1;
5039 	}
5040 
5041 	return res;
5042 }
5043 
5044 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5045 {
5046 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5047 }
5048 
5049 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5050 {
5051 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5052 }
5053 
5054 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5055 {
5056 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5057 }
5058 
5059 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5060 {
5061 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5062 }
5063 
5064 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5065 {
5066 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5067 }
5068 
5069 static unsigned long emulator_get_cached_segment_base(
5070 	struct x86_emulate_ctxt *ctxt, int seg)
5071 {
5072 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5073 }
5074 
5075 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5076 				 struct desc_struct *desc, u32 *base3,
5077 				 int seg)
5078 {
5079 	struct kvm_segment var;
5080 
5081 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5082 	*selector = var.selector;
5083 
5084 	if (var.unusable) {
5085 		memset(desc, 0, sizeof(*desc));
5086 		if (base3)
5087 			*base3 = 0;
5088 		return false;
5089 	}
5090 
5091 	if (var.g)
5092 		var.limit >>= 12;
5093 	set_desc_limit(desc, var.limit);
5094 	set_desc_base(desc, (unsigned long)var.base);
5095 #ifdef CONFIG_X86_64
5096 	if (base3)
5097 		*base3 = var.base >> 32;
5098 #endif
5099 	desc->type = var.type;
5100 	desc->s = var.s;
5101 	desc->dpl = var.dpl;
5102 	desc->p = var.present;
5103 	desc->avl = var.avl;
5104 	desc->l = var.l;
5105 	desc->d = var.db;
5106 	desc->g = var.g;
5107 
5108 	return true;
5109 }
5110 
5111 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5112 				 struct desc_struct *desc, u32 base3,
5113 				 int seg)
5114 {
5115 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5116 	struct kvm_segment var;
5117 
5118 	var.selector = selector;
5119 	var.base = get_desc_base(desc);
5120 #ifdef CONFIG_X86_64
5121 	var.base |= ((u64)base3) << 32;
5122 #endif
5123 	var.limit = get_desc_limit(desc);
5124 	if (desc->g)
5125 		var.limit = (var.limit << 12) | 0xfff;
5126 	var.type = desc->type;
5127 	var.dpl = desc->dpl;
5128 	var.db = desc->d;
5129 	var.s = desc->s;
5130 	var.l = desc->l;
5131 	var.g = desc->g;
5132 	var.avl = desc->avl;
5133 	var.present = desc->p;
5134 	var.unusable = !var.present;
5135 	var.padding = 0;
5136 
5137 	kvm_set_segment(vcpu, &var, seg);
5138 	return;
5139 }
5140 
5141 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5142 			    u32 msr_index, u64 *pdata)
5143 {
5144 	struct msr_data msr;
5145 	int r;
5146 
5147 	msr.index = msr_index;
5148 	msr.host_initiated = false;
5149 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5150 	if (r)
5151 		return r;
5152 
5153 	*pdata = msr.data;
5154 	return 0;
5155 }
5156 
5157 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5158 			    u32 msr_index, u64 data)
5159 {
5160 	struct msr_data msr;
5161 
5162 	msr.data = data;
5163 	msr.index = msr_index;
5164 	msr.host_initiated = false;
5165 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5166 }
5167 
5168 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5169 {
5170 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5171 
5172 	return vcpu->arch.smbase;
5173 }
5174 
5175 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5176 {
5177 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5178 
5179 	vcpu->arch.smbase = smbase;
5180 }
5181 
5182 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5183 			      u32 pmc)
5184 {
5185 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5186 }
5187 
5188 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5189 			     u32 pmc, u64 *pdata)
5190 {
5191 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5192 }
5193 
5194 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5195 {
5196 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5197 }
5198 
5199 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5200 {
5201 	preempt_disable();
5202 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5203 }
5204 
5205 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5206 {
5207 	preempt_enable();
5208 }
5209 
5210 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5211 			      struct x86_instruction_info *info,
5212 			      enum x86_intercept_stage stage)
5213 {
5214 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5215 }
5216 
5217 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5218 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5219 {
5220 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5221 }
5222 
5223 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5224 {
5225 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5226 }
5227 
5228 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5229 {
5230 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5231 }
5232 
5233 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5234 {
5235 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5236 }
5237 
5238 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5239 {
5240 	return emul_to_vcpu(ctxt)->arch.hflags;
5241 }
5242 
5243 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5244 {
5245 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5246 }
5247 
5248 static const struct x86_emulate_ops emulate_ops = {
5249 	.read_gpr            = emulator_read_gpr,
5250 	.write_gpr           = emulator_write_gpr,
5251 	.read_std            = kvm_read_guest_virt_system,
5252 	.write_std           = kvm_write_guest_virt_system,
5253 	.read_phys           = kvm_read_guest_phys_system,
5254 	.fetch               = kvm_fetch_guest_virt,
5255 	.read_emulated       = emulator_read_emulated,
5256 	.write_emulated      = emulator_write_emulated,
5257 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5258 	.invlpg              = emulator_invlpg,
5259 	.pio_in_emulated     = emulator_pio_in_emulated,
5260 	.pio_out_emulated    = emulator_pio_out_emulated,
5261 	.get_segment         = emulator_get_segment,
5262 	.set_segment         = emulator_set_segment,
5263 	.get_cached_segment_base = emulator_get_cached_segment_base,
5264 	.get_gdt             = emulator_get_gdt,
5265 	.get_idt	     = emulator_get_idt,
5266 	.set_gdt             = emulator_set_gdt,
5267 	.set_idt	     = emulator_set_idt,
5268 	.get_cr              = emulator_get_cr,
5269 	.set_cr              = emulator_set_cr,
5270 	.cpl                 = emulator_get_cpl,
5271 	.get_dr              = emulator_get_dr,
5272 	.set_dr              = emulator_set_dr,
5273 	.get_smbase          = emulator_get_smbase,
5274 	.set_smbase          = emulator_set_smbase,
5275 	.set_msr             = emulator_set_msr,
5276 	.get_msr             = emulator_get_msr,
5277 	.check_pmc	     = emulator_check_pmc,
5278 	.read_pmc            = emulator_read_pmc,
5279 	.halt                = emulator_halt,
5280 	.wbinvd              = emulator_wbinvd,
5281 	.fix_hypercall       = emulator_fix_hypercall,
5282 	.get_fpu             = emulator_get_fpu,
5283 	.put_fpu             = emulator_put_fpu,
5284 	.intercept           = emulator_intercept,
5285 	.get_cpuid           = emulator_get_cpuid,
5286 	.set_nmi_mask        = emulator_set_nmi_mask,
5287 	.get_hflags          = emulator_get_hflags,
5288 	.set_hflags          = emulator_set_hflags,
5289 };
5290 
5291 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5292 {
5293 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5294 	/*
5295 	 * an sti; sti; sequence only disable interrupts for the first
5296 	 * instruction. So, if the last instruction, be it emulated or
5297 	 * not, left the system with the INT_STI flag enabled, it
5298 	 * means that the last instruction is an sti. We should not
5299 	 * leave the flag on in this case. The same goes for mov ss
5300 	 */
5301 	if (int_shadow & mask)
5302 		mask = 0;
5303 	if (unlikely(int_shadow || mask)) {
5304 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5305 		if (!mask)
5306 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5307 	}
5308 }
5309 
5310 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5311 {
5312 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5313 	if (ctxt->exception.vector == PF_VECTOR)
5314 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5315 
5316 	if (ctxt->exception.error_code_valid)
5317 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5318 				      ctxt->exception.error_code);
5319 	else
5320 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5321 	return false;
5322 }
5323 
5324 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5325 {
5326 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5327 	int cs_db, cs_l;
5328 
5329 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5330 
5331 	ctxt->eflags = kvm_get_rflags(vcpu);
5332 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5333 
5334 	ctxt->eip = kvm_rip_read(vcpu);
5335 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5336 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5337 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5338 		     cs_db				? X86EMUL_MODE_PROT32 :
5339 							  X86EMUL_MODE_PROT16;
5340 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5341 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5342 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5343 
5344 	init_decode_cache(ctxt);
5345 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5346 }
5347 
5348 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5349 {
5350 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5351 	int ret;
5352 
5353 	init_emulate_ctxt(vcpu);
5354 
5355 	ctxt->op_bytes = 2;
5356 	ctxt->ad_bytes = 2;
5357 	ctxt->_eip = ctxt->eip + inc_eip;
5358 	ret = emulate_int_real(ctxt, irq);
5359 
5360 	if (ret != X86EMUL_CONTINUE)
5361 		return EMULATE_FAIL;
5362 
5363 	ctxt->eip = ctxt->_eip;
5364 	kvm_rip_write(vcpu, ctxt->eip);
5365 	kvm_set_rflags(vcpu, ctxt->eflags);
5366 
5367 	if (irq == NMI_VECTOR)
5368 		vcpu->arch.nmi_pending = 0;
5369 	else
5370 		vcpu->arch.interrupt.pending = false;
5371 
5372 	return EMULATE_DONE;
5373 }
5374 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5375 
5376 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5377 {
5378 	int r = EMULATE_DONE;
5379 
5380 	++vcpu->stat.insn_emulation_fail;
5381 	trace_kvm_emulate_insn_failed(vcpu);
5382 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5383 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5384 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5385 		vcpu->run->internal.ndata = 0;
5386 		r = EMULATE_FAIL;
5387 	}
5388 	kvm_queue_exception(vcpu, UD_VECTOR);
5389 
5390 	return r;
5391 }
5392 
5393 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5394 				  bool write_fault_to_shadow_pgtable,
5395 				  int emulation_type)
5396 {
5397 	gpa_t gpa = cr2;
5398 	kvm_pfn_t pfn;
5399 
5400 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5401 		return false;
5402 
5403 	if (!vcpu->arch.mmu.direct_map) {
5404 		/*
5405 		 * Write permission should be allowed since only
5406 		 * write access need to be emulated.
5407 		 */
5408 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5409 
5410 		/*
5411 		 * If the mapping is invalid in guest, let cpu retry
5412 		 * it to generate fault.
5413 		 */
5414 		if (gpa == UNMAPPED_GVA)
5415 			return true;
5416 	}
5417 
5418 	/*
5419 	 * Do not retry the unhandleable instruction if it faults on the
5420 	 * readonly host memory, otherwise it will goto a infinite loop:
5421 	 * retry instruction -> write #PF -> emulation fail -> retry
5422 	 * instruction -> ...
5423 	 */
5424 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5425 
5426 	/*
5427 	 * If the instruction failed on the error pfn, it can not be fixed,
5428 	 * report the error to userspace.
5429 	 */
5430 	if (is_error_noslot_pfn(pfn))
5431 		return false;
5432 
5433 	kvm_release_pfn_clean(pfn);
5434 
5435 	/* The instructions are well-emulated on direct mmu. */
5436 	if (vcpu->arch.mmu.direct_map) {
5437 		unsigned int indirect_shadow_pages;
5438 
5439 		spin_lock(&vcpu->kvm->mmu_lock);
5440 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5441 		spin_unlock(&vcpu->kvm->mmu_lock);
5442 
5443 		if (indirect_shadow_pages)
5444 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5445 
5446 		return true;
5447 	}
5448 
5449 	/*
5450 	 * if emulation was due to access to shadowed page table
5451 	 * and it failed try to unshadow page and re-enter the
5452 	 * guest to let CPU execute the instruction.
5453 	 */
5454 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5455 
5456 	/*
5457 	 * If the access faults on its page table, it can not
5458 	 * be fixed by unprotecting shadow page and it should
5459 	 * be reported to userspace.
5460 	 */
5461 	return !write_fault_to_shadow_pgtable;
5462 }
5463 
5464 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5465 			      unsigned long cr2,  int emulation_type)
5466 {
5467 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5468 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5469 
5470 	last_retry_eip = vcpu->arch.last_retry_eip;
5471 	last_retry_addr = vcpu->arch.last_retry_addr;
5472 
5473 	/*
5474 	 * If the emulation is caused by #PF and it is non-page_table
5475 	 * writing instruction, it means the VM-EXIT is caused by shadow
5476 	 * page protected, we can zap the shadow page and retry this
5477 	 * instruction directly.
5478 	 *
5479 	 * Note: if the guest uses a non-page-table modifying instruction
5480 	 * on the PDE that points to the instruction, then we will unmap
5481 	 * the instruction and go to an infinite loop. So, we cache the
5482 	 * last retried eip and the last fault address, if we meet the eip
5483 	 * and the address again, we can break out of the potential infinite
5484 	 * loop.
5485 	 */
5486 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5487 
5488 	if (!(emulation_type & EMULTYPE_RETRY))
5489 		return false;
5490 
5491 	if (x86_page_table_writing_insn(ctxt))
5492 		return false;
5493 
5494 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5495 		return false;
5496 
5497 	vcpu->arch.last_retry_eip = ctxt->eip;
5498 	vcpu->arch.last_retry_addr = cr2;
5499 
5500 	if (!vcpu->arch.mmu.direct_map)
5501 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5502 
5503 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5504 
5505 	return true;
5506 }
5507 
5508 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5509 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5510 
5511 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5512 {
5513 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5514 		/* This is a good place to trace that we are exiting SMM.  */
5515 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5516 
5517 		/* Process a latched INIT or SMI, if any.  */
5518 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5519 	}
5520 
5521 	kvm_mmu_reset_context(vcpu);
5522 }
5523 
5524 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5525 {
5526 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5527 
5528 	vcpu->arch.hflags = emul_flags;
5529 
5530 	if (changed & HF_SMM_MASK)
5531 		kvm_smm_changed(vcpu);
5532 }
5533 
5534 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5535 				unsigned long *db)
5536 {
5537 	u32 dr6 = 0;
5538 	int i;
5539 	u32 enable, rwlen;
5540 
5541 	enable = dr7;
5542 	rwlen = dr7 >> 16;
5543 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5544 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5545 			dr6 |= (1 << i);
5546 	return dr6;
5547 }
5548 
5549 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5550 {
5551 	struct kvm_run *kvm_run = vcpu->run;
5552 
5553 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5554 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5555 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5556 		kvm_run->debug.arch.exception = DB_VECTOR;
5557 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5558 		*r = EMULATE_USER_EXIT;
5559 	} else {
5560 		/*
5561 		 * "Certain debug exceptions may clear bit 0-3.  The
5562 		 * remaining contents of the DR6 register are never
5563 		 * cleared by the processor".
5564 		 */
5565 		vcpu->arch.dr6 &= ~15;
5566 		vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5567 		kvm_queue_exception(vcpu, DB_VECTOR);
5568 	}
5569 }
5570 
5571 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5572 {
5573 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5574 	int r = EMULATE_DONE;
5575 
5576 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5577 
5578 	/*
5579 	 * rflags is the old, "raw" value of the flags.  The new value has
5580 	 * not been saved yet.
5581 	 *
5582 	 * This is correct even for TF set by the guest, because "the
5583 	 * processor will not generate this exception after the instruction
5584 	 * that sets the TF flag".
5585 	 */
5586 	if (unlikely(rflags & X86_EFLAGS_TF))
5587 		kvm_vcpu_do_singlestep(vcpu, &r);
5588 	return r == EMULATE_DONE;
5589 }
5590 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5591 
5592 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5593 {
5594 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5595 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5596 		struct kvm_run *kvm_run = vcpu->run;
5597 		unsigned long eip = kvm_get_linear_rip(vcpu);
5598 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5599 					   vcpu->arch.guest_debug_dr7,
5600 					   vcpu->arch.eff_db);
5601 
5602 		if (dr6 != 0) {
5603 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5604 			kvm_run->debug.arch.pc = eip;
5605 			kvm_run->debug.arch.exception = DB_VECTOR;
5606 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5607 			*r = EMULATE_USER_EXIT;
5608 			return true;
5609 		}
5610 	}
5611 
5612 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5613 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5614 		unsigned long eip = kvm_get_linear_rip(vcpu);
5615 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5616 					   vcpu->arch.dr7,
5617 					   vcpu->arch.db);
5618 
5619 		if (dr6 != 0) {
5620 			vcpu->arch.dr6 &= ~15;
5621 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5622 			kvm_queue_exception(vcpu, DB_VECTOR);
5623 			*r = EMULATE_DONE;
5624 			return true;
5625 		}
5626 	}
5627 
5628 	return false;
5629 }
5630 
5631 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5632 			    unsigned long cr2,
5633 			    int emulation_type,
5634 			    void *insn,
5635 			    int insn_len)
5636 {
5637 	int r;
5638 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5639 	bool writeback = true;
5640 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5641 
5642 	/*
5643 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5644 	 * never reused.
5645 	 */
5646 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5647 	kvm_clear_exception_queue(vcpu);
5648 
5649 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5650 		init_emulate_ctxt(vcpu);
5651 
5652 		/*
5653 		 * We will reenter on the same instruction since
5654 		 * we do not set complete_userspace_io.  This does not
5655 		 * handle watchpoints yet, those would be handled in
5656 		 * the emulate_ops.
5657 		 */
5658 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5659 			return r;
5660 
5661 		ctxt->interruptibility = 0;
5662 		ctxt->have_exception = false;
5663 		ctxt->exception.vector = -1;
5664 		ctxt->perm_ok = false;
5665 
5666 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5667 
5668 		r = x86_decode_insn(ctxt, insn, insn_len);
5669 
5670 		trace_kvm_emulate_insn_start(vcpu);
5671 		++vcpu->stat.insn_emulation;
5672 		if (r != EMULATION_OK)  {
5673 			if (emulation_type & EMULTYPE_TRAP_UD)
5674 				return EMULATE_FAIL;
5675 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5676 						emulation_type))
5677 				return EMULATE_DONE;
5678 			if (emulation_type & EMULTYPE_SKIP)
5679 				return EMULATE_FAIL;
5680 			return handle_emulation_failure(vcpu);
5681 		}
5682 	}
5683 
5684 	if (emulation_type & EMULTYPE_SKIP) {
5685 		kvm_rip_write(vcpu, ctxt->_eip);
5686 		if (ctxt->eflags & X86_EFLAGS_RF)
5687 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5688 		return EMULATE_DONE;
5689 	}
5690 
5691 	if (retry_instruction(ctxt, cr2, emulation_type))
5692 		return EMULATE_DONE;
5693 
5694 	/* this is needed for vmware backdoor interface to work since it
5695 	   changes registers values  during IO operation */
5696 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5697 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5698 		emulator_invalidate_register_cache(ctxt);
5699 	}
5700 
5701 restart:
5702 	/* Save the faulting GPA (cr2) in the address field */
5703 	ctxt->exception.address = cr2;
5704 
5705 	r = x86_emulate_insn(ctxt);
5706 
5707 	if (r == EMULATION_INTERCEPTED)
5708 		return EMULATE_DONE;
5709 
5710 	if (r == EMULATION_FAILED) {
5711 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5712 					emulation_type))
5713 			return EMULATE_DONE;
5714 
5715 		return handle_emulation_failure(vcpu);
5716 	}
5717 
5718 	if (ctxt->have_exception) {
5719 		r = EMULATE_DONE;
5720 		if (inject_emulated_exception(vcpu))
5721 			return r;
5722 	} else if (vcpu->arch.pio.count) {
5723 		if (!vcpu->arch.pio.in) {
5724 			/* FIXME: return into emulator if single-stepping.  */
5725 			vcpu->arch.pio.count = 0;
5726 		} else {
5727 			writeback = false;
5728 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5729 		}
5730 		r = EMULATE_USER_EXIT;
5731 	} else if (vcpu->mmio_needed) {
5732 		if (!vcpu->mmio_is_write)
5733 			writeback = false;
5734 		r = EMULATE_USER_EXIT;
5735 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5736 	} else if (r == EMULATION_RESTART)
5737 		goto restart;
5738 	else
5739 		r = EMULATE_DONE;
5740 
5741 	if (writeback) {
5742 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5743 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5744 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5745 		kvm_rip_write(vcpu, ctxt->eip);
5746 		if (r == EMULATE_DONE &&
5747 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5748 			kvm_vcpu_do_singlestep(vcpu, &r);
5749 		if (!ctxt->have_exception ||
5750 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5751 			__kvm_set_rflags(vcpu, ctxt->eflags);
5752 
5753 		/*
5754 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5755 		 * do nothing, and it will be requested again as soon as
5756 		 * the shadow expires.  But we still need to check here,
5757 		 * because POPF has no interrupt shadow.
5758 		 */
5759 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5760 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5761 	} else
5762 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5763 
5764 	return r;
5765 }
5766 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5767 
5768 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5769 {
5770 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5771 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5772 					    size, port, &val, 1);
5773 	/* do not return to emulator after return from userspace */
5774 	vcpu->arch.pio.count = 0;
5775 	return ret;
5776 }
5777 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5778 
5779 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5780 {
5781 	unsigned long val;
5782 
5783 	/* We should only ever be called with arch.pio.count equal to 1 */
5784 	BUG_ON(vcpu->arch.pio.count != 1);
5785 
5786 	/* For size less than 4 we merge, else we zero extend */
5787 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5788 					: 0;
5789 
5790 	/*
5791 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5792 	 * the copy and tracing
5793 	 */
5794 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5795 				 vcpu->arch.pio.port, &val, 1);
5796 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5797 
5798 	return 1;
5799 }
5800 
5801 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5802 {
5803 	unsigned long val;
5804 	int ret;
5805 
5806 	/* For size less than 4 we merge, else we zero extend */
5807 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5808 
5809 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5810 				       &val, 1);
5811 	if (ret) {
5812 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5813 		return ret;
5814 	}
5815 
5816 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5817 
5818 	return 0;
5819 }
5820 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5821 
5822 static int kvmclock_cpu_down_prep(unsigned int cpu)
5823 {
5824 	__this_cpu_write(cpu_tsc_khz, 0);
5825 	return 0;
5826 }
5827 
5828 static void tsc_khz_changed(void *data)
5829 {
5830 	struct cpufreq_freqs *freq = data;
5831 	unsigned long khz = 0;
5832 
5833 	if (data)
5834 		khz = freq->new;
5835 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5836 		khz = cpufreq_quick_get(raw_smp_processor_id());
5837 	if (!khz)
5838 		khz = tsc_khz;
5839 	__this_cpu_write(cpu_tsc_khz, khz);
5840 }
5841 
5842 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5843 				     void *data)
5844 {
5845 	struct cpufreq_freqs *freq = data;
5846 	struct kvm *kvm;
5847 	struct kvm_vcpu *vcpu;
5848 	int i, send_ipi = 0;
5849 
5850 	/*
5851 	 * We allow guests to temporarily run on slowing clocks,
5852 	 * provided we notify them after, or to run on accelerating
5853 	 * clocks, provided we notify them before.  Thus time never
5854 	 * goes backwards.
5855 	 *
5856 	 * However, we have a problem.  We can't atomically update
5857 	 * the frequency of a given CPU from this function; it is
5858 	 * merely a notifier, which can be called from any CPU.
5859 	 * Changing the TSC frequency at arbitrary points in time
5860 	 * requires a recomputation of local variables related to
5861 	 * the TSC for each VCPU.  We must flag these local variables
5862 	 * to be updated and be sure the update takes place with the
5863 	 * new frequency before any guests proceed.
5864 	 *
5865 	 * Unfortunately, the combination of hotplug CPU and frequency
5866 	 * change creates an intractable locking scenario; the order
5867 	 * of when these callouts happen is undefined with respect to
5868 	 * CPU hotplug, and they can race with each other.  As such,
5869 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5870 	 * undefined; you can actually have a CPU frequency change take
5871 	 * place in between the computation of X and the setting of the
5872 	 * variable.  To protect against this problem, all updates of
5873 	 * the per_cpu tsc_khz variable are done in an interrupt
5874 	 * protected IPI, and all callers wishing to update the value
5875 	 * must wait for a synchronous IPI to complete (which is trivial
5876 	 * if the caller is on the CPU already).  This establishes the
5877 	 * necessary total order on variable updates.
5878 	 *
5879 	 * Note that because a guest time update may take place
5880 	 * anytime after the setting of the VCPU's request bit, the
5881 	 * correct TSC value must be set before the request.  However,
5882 	 * to ensure the update actually makes it to any guest which
5883 	 * starts running in hardware virtualization between the set
5884 	 * and the acquisition of the spinlock, we must also ping the
5885 	 * CPU after setting the request bit.
5886 	 *
5887 	 */
5888 
5889 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5890 		return 0;
5891 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5892 		return 0;
5893 
5894 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5895 
5896 	spin_lock(&kvm_lock);
5897 	list_for_each_entry(kvm, &vm_list, vm_list) {
5898 		kvm_for_each_vcpu(i, vcpu, kvm) {
5899 			if (vcpu->cpu != freq->cpu)
5900 				continue;
5901 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5902 			if (vcpu->cpu != smp_processor_id())
5903 				send_ipi = 1;
5904 		}
5905 	}
5906 	spin_unlock(&kvm_lock);
5907 
5908 	if (freq->old < freq->new && send_ipi) {
5909 		/*
5910 		 * We upscale the frequency.  Must make the guest
5911 		 * doesn't see old kvmclock values while running with
5912 		 * the new frequency, otherwise we risk the guest sees
5913 		 * time go backwards.
5914 		 *
5915 		 * In case we update the frequency for another cpu
5916 		 * (which might be in guest context) send an interrupt
5917 		 * to kick the cpu out of guest context.  Next time
5918 		 * guest context is entered kvmclock will be updated,
5919 		 * so the guest will not see stale values.
5920 		 */
5921 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5922 	}
5923 	return 0;
5924 }
5925 
5926 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5927 	.notifier_call  = kvmclock_cpufreq_notifier
5928 };
5929 
5930 static int kvmclock_cpu_online(unsigned int cpu)
5931 {
5932 	tsc_khz_changed(NULL);
5933 	return 0;
5934 }
5935 
5936 static void kvm_timer_init(void)
5937 {
5938 	max_tsc_khz = tsc_khz;
5939 
5940 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5941 #ifdef CONFIG_CPU_FREQ
5942 		struct cpufreq_policy policy;
5943 		int cpu;
5944 
5945 		memset(&policy, 0, sizeof(policy));
5946 		cpu = get_cpu();
5947 		cpufreq_get_policy(&policy, cpu);
5948 		if (policy.cpuinfo.max_freq)
5949 			max_tsc_khz = policy.cpuinfo.max_freq;
5950 		put_cpu();
5951 #endif
5952 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5953 					  CPUFREQ_TRANSITION_NOTIFIER);
5954 	}
5955 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5956 
5957 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5958 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
5959 }
5960 
5961 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5962 
5963 int kvm_is_in_guest(void)
5964 {
5965 	return __this_cpu_read(current_vcpu) != NULL;
5966 }
5967 
5968 static int kvm_is_user_mode(void)
5969 {
5970 	int user_mode = 3;
5971 
5972 	if (__this_cpu_read(current_vcpu))
5973 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5974 
5975 	return user_mode != 0;
5976 }
5977 
5978 static unsigned long kvm_get_guest_ip(void)
5979 {
5980 	unsigned long ip = 0;
5981 
5982 	if (__this_cpu_read(current_vcpu))
5983 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5984 
5985 	return ip;
5986 }
5987 
5988 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5989 	.is_in_guest		= kvm_is_in_guest,
5990 	.is_user_mode		= kvm_is_user_mode,
5991 	.get_guest_ip		= kvm_get_guest_ip,
5992 };
5993 
5994 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5995 {
5996 	__this_cpu_write(current_vcpu, vcpu);
5997 }
5998 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5999 
6000 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6001 {
6002 	__this_cpu_write(current_vcpu, NULL);
6003 }
6004 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6005 
6006 static void kvm_set_mmio_spte_mask(void)
6007 {
6008 	u64 mask;
6009 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6010 
6011 	/*
6012 	 * Set the reserved bits and the present bit of an paging-structure
6013 	 * entry to generate page fault with PFER.RSV = 1.
6014 	 */
6015 	 /* Mask the reserved physical address bits. */
6016 	mask = rsvd_bits(maxphyaddr, 51);
6017 
6018 	/* Set the present bit. */
6019 	mask |= 1ull;
6020 
6021 #ifdef CONFIG_X86_64
6022 	/*
6023 	 * If reserved bit is not supported, clear the present bit to disable
6024 	 * mmio page fault.
6025 	 */
6026 	if (maxphyaddr == 52)
6027 		mask &= ~1ull;
6028 #endif
6029 
6030 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6031 }
6032 
6033 #ifdef CONFIG_X86_64
6034 static void pvclock_gtod_update_fn(struct work_struct *work)
6035 {
6036 	struct kvm *kvm;
6037 
6038 	struct kvm_vcpu *vcpu;
6039 	int i;
6040 
6041 	spin_lock(&kvm_lock);
6042 	list_for_each_entry(kvm, &vm_list, vm_list)
6043 		kvm_for_each_vcpu(i, vcpu, kvm)
6044 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6045 	atomic_set(&kvm_guest_has_master_clock, 0);
6046 	spin_unlock(&kvm_lock);
6047 }
6048 
6049 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6050 
6051 /*
6052  * Notification about pvclock gtod data update.
6053  */
6054 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6055 			       void *priv)
6056 {
6057 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6058 	struct timekeeper *tk = priv;
6059 
6060 	update_pvclock_gtod(tk);
6061 
6062 	/* disable master clock if host does not trust, or does not
6063 	 * use, TSC clocksource
6064 	 */
6065 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6066 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6067 		queue_work(system_long_wq, &pvclock_gtod_work);
6068 
6069 	return 0;
6070 }
6071 
6072 static struct notifier_block pvclock_gtod_notifier = {
6073 	.notifier_call = pvclock_gtod_notify,
6074 };
6075 #endif
6076 
6077 int kvm_arch_init(void *opaque)
6078 {
6079 	int r;
6080 	struct kvm_x86_ops *ops = opaque;
6081 
6082 	if (kvm_x86_ops) {
6083 		printk(KERN_ERR "kvm: already loaded the other module\n");
6084 		r = -EEXIST;
6085 		goto out;
6086 	}
6087 
6088 	if (!ops->cpu_has_kvm_support()) {
6089 		printk(KERN_ERR "kvm: no hardware support\n");
6090 		r = -EOPNOTSUPP;
6091 		goto out;
6092 	}
6093 	if (ops->disabled_by_bios()) {
6094 		printk(KERN_ERR "kvm: disabled by bios\n");
6095 		r = -EOPNOTSUPP;
6096 		goto out;
6097 	}
6098 
6099 	r = -ENOMEM;
6100 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6101 	if (!shared_msrs) {
6102 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6103 		goto out;
6104 	}
6105 
6106 	r = kvm_mmu_module_init();
6107 	if (r)
6108 		goto out_free_percpu;
6109 
6110 	kvm_set_mmio_spte_mask();
6111 
6112 	kvm_x86_ops = ops;
6113 
6114 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6115 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6116 			PT_PRESENT_MASK, 0);
6117 	kvm_timer_init();
6118 
6119 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6120 
6121 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6122 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6123 
6124 	kvm_lapic_init();
6125 #ifdef CONFIG_X86_64
6126 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6127 #endif
6128 
6129 	return 0;
6130 
6131 out_free_percpu:
6132 	free_percpu(shared_msrs);
6133 out:
6134 	return r;
6135 }
6136 
6137 void kvm_arch_exit(void)
6138 {
6139 	kvm_lapic_exit();
6140 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6141 
6142 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6143 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6144 					    CPUFREQ_TRANSITION_NOTIFIER);
6145 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6146 #ifdef CONFIG_X86_64
6147 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6148 #endif
6149 	kvm_x86_ops = NULL;
6150 	kvm_mmu_module_exit();
6151 	free_percpu(shared_msrs);
6152 }
6153 
6154 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6155 {
6156 	++vcpu->stat.halt_exits;
6157 	if (lapic_in_kernel(vcpu)) {
6158 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6159 		return 1;
6160 	} else {
6161 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6162 		return 0;
6163 	}
6164 }
6165 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6166 
6167 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6168 {
6169 	int ret = kvm_skip_emulated_instruction(vcpu);
6170 	/*
6171 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6172 	 * KVM_EXIT_DEBUG here.
6173 	 */
6174 	return kvm_vcpu_halt(vcpu) && ret;
6175 }
6176 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6177 
6178 #ifdef CONFIG_X86_64
6179 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6180 			        unsigned long clock_type)
6181 {
6182 	struct kvm_clock_pairing clock_pairing;
6183 	struct timespec ts;
6184 	u64 cycle;
6185 	int ret;
6186 
6187 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6188 		return -KVM_EOPNOTSUPP;
6189 
6190 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6191 		return -KVM_EOPNOTSUPP;
6192 
6193 	clock_pairing.sec = ts.tv_sec;
6194 	clock_pairing.nsec = ts.tv_nsec;
6195 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6196 	clock_pairing.flags = 0;
6197 
6198 	ret = 0;
6199 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6200 			    sizeof(struct kvm_clock_pairing)))
6201 		ret = -KVM_EFAULT;
6202 
6203 	return ret;
6204 }
6205 #endif
6206 
6207 /*
6208  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6209  *
6210  * @apicid - apicid of vcpu to be kicked.
6211  */
6212 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6213 {
6214 	struct kvm_lapic_irq lapic_irq;
6215 
6216 	lapic_irq.shorthand = 0;
6217 	lapic_irq.dest_mode = 0;
6218 	lapic_irq.dest_id = apicid;
6219 	lapic_irq.msi_redir_hint = false;
6220 
6221 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6222 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6223 }
6224 
6225 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6226 {
6227 	vcpu->arch.apicv_active = false;
6228 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6229 }
6230 
6231 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6232 {
6233 	unsigned long nr, a0, a1, a2, a3, ret;
6234 	int op_64_bit, r;
6235 
6236 	r = kvm_skip_emulated_instruction(vcpu);
6237 
6238 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6239 		return kvm_hv_hypercall(vcpu);
6240 
6241 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6242 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6243 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6244 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6245 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6246 
6247 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6248 
6249 	op_64_bit = is_64_bit_mode(vcpu);
6250 	if (!op_64_bit) {
6251 		nr &= 0xFFFFFFFF;
6252 		a0 &= 0xFFFFFFFF;
6253 		a1 &= 0xFFFFFFFF;
6254 		a2 &= 0xFFFFFFFF;
6255 		a3 &= 0xFFFFFFFF;
6256 	}
6257 
6258 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6259 		ret = -KVM_EPERM;
6260 		goto out;
6261 	}
6262 
6263 	switch (nr) {
6264 	case KVM_HC_VAPIC_POLL_IRQ:
6265 		ret = 0;
6266 		break;
6267 	case KVM_HC_KICK_CPU:
6268 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6269 		ret = 0;
6270 		break;
6271 #ifdef CONFIG_X86_64
6272 	case KVM_HC_CLOCK_PAIRING:
6273 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6274 		break;
6275 #endif
6276 	default:
6277 		ret = -KVM_ENOSYS;
6278 		break;
6279 	}
6280 out:
6281 	if (!op_64_bit)
6282 		ret = (u32)ret;
6283 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6284 	++vcpu->stat.hypercalls;
6285 	return r;
6286 }
6287 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6288 
6289 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6290 {
6291 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6292 	char instruction[3];
6293 	unsigned long rip = kvm_rip_read(vcpu);
6294 
6295 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6296 
6297 	return emulator_write_emulated(ctxt, rip, instruction, 3,
6298 		&ctxt->exception);
6299 }
6300 
6301 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6302 {
6303 	return vcpu->run->request_interrupt_window &&
6304 		likely(!pic_in_kernel(vcpu->kvm));
6305 }
6306 
6307 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6308 {
6309 	struct kvm_run *kvm_run = vcpu->run;
6310 
6311 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6312 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6313 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6314 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6315 	kvm_run->ready_for_interrupt_injection =
6316 		pic_in_kernel(vcpu->kvm) ||
6317 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6318 }
6319 
6320 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6321 {
6322 	int max_irr, tpr;
6323 
6324 	if (!kvm_x86_ops->update_cr8_intercept)
6325 		return;
6326 
6327 	if (!lapic_in_kernel(vcpu))
6328 		return;
6329 
6330 	if (vcpu->arch.apicv_active)
6331 		return;
6332 
6333 	if (!vcpu->arch.apic->vapic_addr)
6334 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6335 	else
6336 		max_irr = -1;
6337 
6338 	if (max_irr != -1)
6339 		max_irr >>= 4;
6340 
6341 	tpr = kvm_lapic_get_cr8(vcpu);
6342 
6343 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6344 }
6345 
6346 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6347 {
6348 	int r;
6349 
6350 	/* try to reinject previous events if any */
6351 	if (vcpu->arch.exception.pending) {
6352 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6353 					vcpu->arch.exception.has_error_code,
6354 					vcpu->arch.exception.error_code);
6355 
6356 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6357 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6358 					     X86_EFLAGS_RF);
6359 
6360 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6361 		    (vcpu->arch.dr7 & DR7_GD)) {
6362 			vcpu->arch.dr7 &= ~DR7_GD;
6363 			kvm_update_dr7(vcpu);
6364 		}
6365 
6366 		kvm_x86_ops->queue_exception(vcpu);
6367 		return 0;
6368 	}
6369 
6370 	if (vcpu->arch.nmi_injected) {
6371 		kvm_x86_ops->set_nmi(vcpu);
6372 		return 0;
6373 	}
6374 
6375 	if (vcpu->arch.interrupt.pending) {
6376 		kvm_x86_ops->set_irq(vcpu);
6377 		return 0;
6378 	}
6379 
6380 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6381 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6382 		if (r != 0)
6383 			return r;
6384 	}
6385 
6386 	/* try to inject new event if pending */
6387 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6388 		vcpu->arch.smi_pending = false;
6389 		enter_smm(vcpu);
6390 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6391 		--vcpu->arch.nmi_pending;
6392 		vcpu->arch.nmi_injected = true;
6393 		kvm_x86_ops->set_nmi(vcpu);
6394 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6395 		/*
6396 		 * Because interrupts can be injected asynchronously, we are
6397 		 * calling check_nested_events again here to avoid a race condition.
6398 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6399 		 * proposal and current concerns.  Perhaps we should be setting
6400 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6401 		 */
6402 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6403 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6404 			if (r != 0)
6405 				return r;
6406 		}
6407 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6408 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6409 					    false);
6410 			kvm_x86_ops->set_irq(vcpu);
6411 		}
6412 	}
6413 
6414 	return 0;
6415 }
6416 
6417 static void process_nmi(struct kvm_vcpu *vcpu)
6418 {
6419 	unsigned limit = 2;
6420 
6421 	/*
6422 	 * x86 is limited to one NMI running, and one NMI pending after it.
6423 	 * If an NMI is already in progress, limit further NMIs to just one.
6424 	 * Otherwise, allow two (and we'll inject the first one immediately).
6425 	 */
6426 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6427 		limit = 1;
6428 
6429 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6430 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6431 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6432 }
6433 
6434 #define put_smstate(type, buf, offset, val)			  \
6435 	*(type *)((buf) + (offset) - 0x7e00) = val
6436 
6437 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6438 {
6439 	u32 flags = 0;
6440 	flags |= seg->g       << 23;
6441 	flags |= seg->db      << 22;
6442 	flags |= seg->l       << 21;
6443 	flags |= seg->avl     << 20;
6444 	flags |= seg->present << 15;
6445 	flags |= seg->dpl     << 13;
6446 	flags |= seg->s       << 12;
6447 	flags |= seg->type    << 8;
6448 	return flags;
6449 }
6450 
6451 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6452 {
6453 	struct kvm_segment seg;
6454 	int offset;
6455 
6456 	kvm_get_segment(vcpu, &seg, n);
6457 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6458 
6459 	if (n < 3)
6460 		offset = 0x7f84 + n * 12;
6461 	else
6462 		offset = 0x7f2c + (n - 3) * 12;
6463 
6464 	put_smstate(u32, buf, offset + 8, seg.base);
6465 	put_smstate(u32, buf, offset + 4, seg.limit);
6466 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6467 }
6468 
6469 #ifdef CONFIG_X86_64
6470 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6471 {
6472 	struct kvm_segment seg;
6473 	int offset;
6474 	u16 flags;
6475 
6476 	kvm_get_segment(vcpu, &seg, n);
6477 	offset = 0x7e00 + n * 16;
6478 
6479 	flags = enter_smm_get_segment_flags(&seg) >> 8;
6480 	put_smstate(u16, buf, offset, seg.selector);
6481 	put_smstate(u16, buf, offset + 2, flags);
6482 	put_smstate(u32, buf, offset + 4, seg.limit);
6483 	put_smstate(u64, buf, offset + 8, seg.base);
6484 }
6485 #endif
6486 
6487 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6488 {
6489 	struct desc_ptr dt;
6490 	struct kvm_segment seg;
6491 	unsigned long val;
6492 	int i;
6493 
6494 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6495 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6496 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6497 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6498 
6499 	for (i = 0; i < 8; i++)
6500 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6501 
6502 	kvm_get_dr(vcpu, 6, &val);
6503 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6504 	kvm_get_dr(vcpu, 7, &val);
6505 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6506 
6507 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6508 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6509 	put_smstate(u32, buf, 0x7f64, seg.base);
6510 	put_smstate(u32, buf, 0x7f60, seg.limit);
6511 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6512 
6513 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6514 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6515 	put_smstate(u32, buf, 0x7f80, seg.base);
6516 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6517 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6518 
6519 	kvm_x86_ops->get_gdt(vcpu, &dt);
6520 	put_smstate(u32, buf, 0x7f74, dt.address);
6521 	put_smstate(u32, buf, 0x7f70, dt.size);
6522 
6523 	kvm_x86_ops->get_idt(vcpu, &dt);
6524 	put_smstate(u32, buf, 0x7f58, dt.address);
6525 	put_smstate(u32, buf, 0x7f54, dt.size);
6526 
6527 	for (i = 0; i < 6; i++)
6528 		enter_smm_save_seg_32(vcpu, buf, i);
6529 
6530 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6531 
6532 	/* revision id */
6533 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6534 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6535 }
6536 
6537 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6538 {
6539 #ifdef CONFIG_X86_64
6540 	struct desc_ptr dt;
6541 	struct kvm_segment seg;
6542 	unsigned long val;
6543 	int i;
6544 
6545 	for (i = 0; i < 16; i++)
6546 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6547 
6548 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6549 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6550 
6551 	kvm_get_dr(vcpu, 6, &val);
6552 	put_smstate(u64, buf, 0x7f68, val);
6553 	kvm_get_dr(vcpu, 7, &val);
6554 	put_smstate(u64, buf, 0x7f60, val);
6555 
6556 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6557 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6558 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6559 
6560 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6561 
6562 	/* revision id */
6563 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6564 
6565 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6566 
6567 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6568 	put_smstate(u16, buf, 0x7e90, seg.selector);
6569 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6570 	put_smstate(u32, buf, 0x7e94, seg.limit);
6571 	put_smstate(u64, buf, 0x7e98, seg.base);
6572 
6573 	kvm_x86_ops->get_idt(vcpu, &dt);
6574 	put_smstate(u32, buf, 0x7e84, dt.size);
6575 	put_smstate(u64, buf, 0x7e88, dt.address);
6576 
6577 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6578 	put_smstate(u16, buf, 0x7e70, seg.selector);
6579 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6580 	put_smstate(u32, buf, 0x7e74, seg.limit);
6581 	put_smstate(u64, buf, 0x7e78, seg.base);
6582 
6583 	kvm_x86_ops->get_gdt(vcpu, &dt);
6584 	put_smstate(u32, buf, 0x7e64, dt.size);
6585 	put_smstate(u64, buf, 0x7e68, dt.address);
6586 
6587 	for (i = 0; i < 6; i++)
6588 		enter_smm_save_seg_64(vcpu, buf, i);
6589 #else
6590 	WARN_ON_ONCE(1);
6591 #endif
6592 }
6593 
6594 static void enter_smm(struct kvm_vcpu *vcpu)
6595 {
6596 	struct kvm_segment cs, ds;
6597 	struct desc_ptr dt;
6598 	char buf[512];
6599 	u32 cr0;
6600 
6601 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6602 	vcpu->arch.hflags |= HF_SMM_MASK;
6603 	memset(buf, 0, 512);
6604 	if (guest_cpuid_has_longmode(vcpu))
6605 		enter_smm_save_state_64(vcpu, buf);
6606 	else
6607 		enter_smm_save_state_32(vcpu, buf);
6608 
6609 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6610 
6611 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6612 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6613 	else
6614 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6615 
6616 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6617 	kvm_rip_write(vcpu, 0x8000);
6618 
6619 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6620 	kvm_x86_ops->set_cr0(vcpu, cr0);
6621 	vcpu->arch.cr0 = cr0;
6622 
6623 	kvm_x86_ops->set_cr4(vcpu, 0);
6624 
6625 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6626 	dt.address = dt.size = 0;
6627 	kvm_x86_ops->set_idt(vcpu, &dt);
6628 
6629 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6630 
6631 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6632 	cs.base = vcpu->arch.smbase;
6633 
6634 	ds.selector = 0;
6635 	ds.base = 0;
6636 
6637 	cs.limit    = ds.limit = 0xffffffff;
6638 	cs.type     = ds.type = 0x3;
6639 	cs.dpl      = ds.dpl = 0;
6640 	cs.db       = ds.db = 0;
6641 	cs.s        = ds.s = 1;
6642 	cs.l        = ds.l = 0;
6643 	cs.g        = ds.g = 1;
6644 	cs.avl      = ds.avl = 0;
6645 	cs.present  = ds.present = 1;
6646 	cs.unusable = ds.unusable = 0;
6647 	cs.padding  = ds.padding = 0;
6648 
6649 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6650 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6651 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6652 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6653 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6654 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6655 
6656 	if (guest_cpuid_has_longmode(vcpu))
6657 		kvm_x86_ops->set_efer(vcpu, 0);
6658 
6659 	kvm_update_cpuid(vcpu);
6660 	kvm_mmu_reset_context(vcpu);
6661 }
6662 
6663 static void process_smi(struct kvm_vcpu *vcpu)
6664 {
6665 	vcpu->arch.smi_pending = true;
6666 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6667 }
6668 
6669 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6670 {
6671 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6672 }
6673 
6674 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6675 {
6676 	u64 eoi_exit_bitmap[4];
6677 
6678 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6679 		return;
6680 
6681 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6682 
6683 	if (irqchip_split(vcpu->kvm))
6684 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6685 	else {
6686 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6687 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6688 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6689 	}
6690 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6691 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
6692 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6693 }
6694 
6695 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6696 {
6697 	++vcpu->stat.tlb_flush;
6698 	kvm_x86_ops->tlb_flush(vcpu);
6699 }
6700 
6701 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6702 {
6703 	struct page *page = NULL;
6704 
6705 	if (!lapic_in_kernel(vcpu))
6706 		return;
6707 
6708 	if (!kvm_x86_ops->set_apic_access_page_addr)
6709 		return;
6710 
6711 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6712 	if (is_error_page(page))
6713 		return;
6714 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6715 
6716 	/*
6717 	 * Do not pin apic access page in memory, the MMU notifier
6718 	 * will call us again if it is migrated or swapped out.
6719 	 */
6720 	put_page(page);
6721 }
6722 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6723 
6724 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6725 					   unsigned long address)
6726 {
6727 	/*
6728 	 * The physical address of apic access page is stored in the VMCS.
6729 	 * Update it when it becomes invalid.
6730 	 */
6731 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6732 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6733 }
6734 
6735 /*
6736  * Returns 1 to let vcpu_run() continue the guest execution loop without
6737  * exiting to the userspace.  Otherwise, the value will be returned to the
6738  * userspace.
6739  */
6740 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6741 {
6742 	int r;
6743 	bool req_int_win =
6744 		dm_request_for_irq_injection(vcpu) &&
6745 		kvm_cpu_accept_dm_intr(vcpu);
6746 
6747 	bool req_immediate_exit = false;
6748 
6749 	if (kvm_request_pending(vcpu)) {
6750 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6751 			kvm_mmu_unload(vcpu);
6752 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6753 			__kvm_migrate_timers(vcpu);
6754 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6755 			kvm_gen_update_masterclock(vcpu->kvm);
6756 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6757 			kvm_gen_kvmclock_update(vcpu);
6758 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6759 			r = kvm_guest_time_update(vcpu);
6760 			if (unlikely(r))
6761 				goto out;
6762 		}
6763 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6764 			kvm_mmu_sync_roots(vcpu);
6765 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6766 			kvm_vcpu_flush_tlb(vcpu);
6767 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6768 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6769 			r = 0;
6770 			goto out;
6771 		}
6772 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6773 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6774 			r = 0;
6775 			goto out;
6776 		}
6777 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6778 			/* Page is swapped out. Do synthetic halt */
6779 			vcpu->arch.apf.halted = true;
6780 			r = 1;
6781 			goto out;
6782 		}
6783 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6784 			record_steal_time(vcpu);
6785 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6786 			process_smi(vcpu);
6787 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6788 			process_nmi(vcpu);
6789 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6790 			kvm_pmu_handle_event(vcpu);
6791 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6792 			kvm_pmu_deliver_pmi(vcpu);
6793 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6794 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6795 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
6796 				     vcpu->arch.ioapic_handled_vectors)) {
6797 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6798 				vcpu->run->eoi.vector =
6799 						vcpu->arch.pending_ioapic_eoi;
6800 				r = 0;
6801 				goto out;
6802 			}
6803 		}
6804 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6805 			vcpu_scan_ioapic(vcpu);
6806 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6807 			kvm_vcpu_reload_apic_access_page(vcpu);
6808 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6809 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6810 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6811 			r = 0;
6812 			goto out;
6813 		}
6814 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6815 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6816 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6817 			r = 0;
6818 			goto out;
6819 		}
6820 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6821 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6822 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6823 			r = 0;
6824 			goto out;
6825 		}
6826 
6827 		/*
6828 		 * KVM_REQ_HV_STIMER has to be processed after
6829 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6830 		 * depend on the guest clock being up-to-date
6831 		 */
6832 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6833 			kvm_hv_process_stimers(vcpu);
6834 	}
6835 
6836 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6837 		++vcpu->stat.req_event;
6838 		kvm_apic_accept_events(vcpu);
6839 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6840 			r = 1;
6841 			goto out;
6842 		}
6843 
6844 		if (inject_pending_event(vcpu, req_int_win) != 0)
6845 			req_immediate_exit = true;
6846 		else {
6847 			/* Enable NMI/IRQ window open exits if needed.
6848 			 *
6849 			 * SMIs have two cases: 1) they can be nested, and
6850 			 * then there is nothing to do here because RSM will
6851 			 * cause a vmexit anyway; 2) or the SMI can be pending
6852 			 * because inject_pending_event has completed the
6853 			 * injection of an IRQ or NMI from the previous vmexit,
6854 			 * and then we request an immediate exit to inject the SMI.
6855 			 */
6856 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
6857 				req_immediate_exit = true;
6858 			if (vcpu->arch.nmi_pending)
6859 				kvm_x86_ops->enable_nmi_window(vcpu);
6860 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6861 				kvm_x86_ops->enable_irq_window(vcpu);
6862 		}
6863 
6864 		if (kvm_lapic_enabled(vcpu)) {
6865 			update_cr8_intercept(vcpu);
6866 			kvm_lapic_sync_to_vapic(vcpu);
6867 		}
6868 	}
6869 
6870 	r = kvm_mmu_reload(vcpu);
6871 	if (unlikely(r)) {
6872 		goto cancel_injection;
6873 	}
6874 
6875 	preempt_disable();
6876 
6877 	kvm_x86_ops->prepare_guest_switch(vcpu);
6878 	kvm_load_guest_fpu(vcpu);
6879 
6880 	/*
6881 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
6882 	 * IPI are then delayed after guest entry, which ensures that they
6883 	 * result in virtual interrupt delivery.
6884 	 */
6885 	local_irq_disable();
6886 	vcpu->mode = IN_GUEST_MODE;
6887 
6888 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6889 
6890 	/*
6891 	 * 1) We should set ->mode before checking ->requests.  Please see
6892 	 * the comment in kvm_vcpu_exiting_guest_mode().
6893 	 *
6894 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
6895 	 * pairs with the memory barrier implicit in pi_test_and_set_on
6896 	 * (see vmx_deliver_posted_interrupt).
6897 	 *
6898 	 * 3) This also orders the write to mode from any reads to the page
6899 	 * tables done while the VCPU is running.  Please see the comment
6900 	 * in kvm_flush_remote_tlbs.
6901 	 */
6902 	smp_mb__after_srcu_read_unlock();
6903 
6904 	/*
6905 	 * This handles the case where a posted interrupt was
6906 	 * notified with kvm_vcpu_kick.
6907 	 */
6908 	if (kvm_lapic_enabled(vcpu)) {
6909 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6910 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6911 	}
6912 
6913 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
6914 	    || need_resched() || signal_pending(current)) {
6915 		vcpu->mode = OUTSIDE_GUEST_MODE;
6916 		smp_wmb();
6917 		local_irq_enable();
6918 		preempt_enable();
6919 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6920 		r = 1;
6921 		goto cancel_injection;
6922 	}
6923 
6924 	kvm_load_guest_xcr0(vcpu);
6925 
6926 	if (req_immediate_exit) {
6927 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6928 		smp_send_reschedule(vcpu->cpu);
6929 	}
6930 
6931 	trace_kvm_entry(vcpu->vcpu_id);
6932 	wait_lapic_expire(vcpu);
6933 	guest_enter_irqoff();
6934 
6935 	if (unlikely(vcpu->arch.switch_db_regs)) {
6936 		set_debugreg(0, 7);
6937 		set_debugreg(vcpu->arch.eff_db[0], 0);
6938 		set_debugreg(vcpu->arch.eff_db[1], 1);
6939 		set_debugreg(vcpu->arch.eff_db[2], 2);
6940 		set_debugreg(vcpu->arch.eff_db[3], 3);
6941 		set_debugreg(vcpu->arch.dr6, 6);
6942 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6943 	}
6944 
6945 	kvm_x86_ops->run(vcpu);
6946 
6947 	/*
6948 	 * Do this here before restoring debug registers on the host.  And
6949 	 * since we do this before handling the vmexit, a DR access vmexit
6950 	 * can (a) read the correct value of the debug registers, (b) set
6951 	 * KVM_DEBUGREG_WONT_EXIT again.
6952 	 */
6953 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6954 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6955 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6956 		kvm_update_dr0123(vcpu);
6957 		kvm_update_dr6(vcpu);
6958 		kvm_update_dr7(vcpu);
6959 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6960 	}
6961 
6962 	/*
6963 	 * If the guest has used debug registers, at least dr7
6964 	 * will be disabled while returning to the host.
6965 	 * If we don't have active breakpoints in the host, we don't
6966 	 * care about the messed up debug address registers. But if
6967 	 * we have some of them active, restore the old state.
6968 	 */
6969 	if (hw_breakpoint_active())
6970 		hw_breakpoint_restore();
6971 
6972 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6973 
6974 	vcpu->mode = OUTSIDE_GUEST_MODE;
6975 	smp_wmb();
6976 
6977 	kvm_put_guest_xcr0(vcpu);
6978 
6979 	kvm_x86_ops->handle_external_intr(vcpu);
6980 
6981 	++vcpu->stat.exits;
6982 
6983 	guest_exit_irqoff();
6984 
6985 	local_irq_enable();
6986 	preempt_enable();
6987 
6988 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6989 
6990 	/*
6991 	 * Profile KVM exit RIPs:
6992 	 */
6993 	if (unlikely(prof_on == KVM_PROFILING)) {
6994 		unsigned long rip = kvm_rip_read(vcpu);
6995 		profile_hit(KVM_PROFILING, (void *)rip);
6996 	}
6997 
6998 	if (unlikely(vcpu->arch.tsc_always_catchup))
6999 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7000 
7001 	if (vcpu->arch.apic_attention)
7002 		kvm_lapic_sync_from_vapic(vcpu);
7003 
7004 	r = kvm_x86_ops->handle_exit(vcpu);
7005 	return r;
7006 
7007 cancel_injection:
7008 	kvm_x86_ops->cancel_injection(vcpu);
7009 	if (unlikely(vcpu->arch.apic_attention))
7010 		kvm_lapic_sync_from_vapic(vcpu);
7011 out:
7012 	return r;
7013 }
7014 
7015 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7016 {
7017 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7018 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7019 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7020 		kvm_vcpu_block(vcpu);
7021 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7022 
7023 		if (kvm_x86_ops->post_block)
7024 			kvm_x86_ops->post_block(vcpu);
7025 
7026 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7027 			return 1;
7028 	}
7029 
7030 	kvm_apic_accept_events(vcpu);
7031 	switch(vcpu->arch.mp_state) {
7032 	case KVM_MP_STATE_HALTED:
7033 		vcpu->arch.pv.pv_unhalted = false;
7034 		vcpu->arch.mp_state =
7035 			KVM_MP_STATE_RUNNABLE;
7036 	case KVM_MP_STATE_RUNNABLE:
7037 		vcpu->arch.apf.halted = false;
7038 		break;
7039 	case KVM_MP_STATE_INIT_RECEIVED:
7040 		break;
7041 	default:
7042 		return -EINTR;
7043 		break;
7044 	}
7045 	return 1;
7046 }
7047 
7048 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7049 {
7050 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7051 		kvm_x86_ops->check_nested_events(vcpu, false);
7052 
7053 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7054 		!vcpu->arch.apf.halted);
7055 }
7056 
7057 static int vcpu_run(struct kvm_vcpu *vcpu)
7058 {
7059 	int r;
7060 	struct kvm *kvm = vcpu->kvm;
7061 
7062 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7063 
7064 	for (;;) {
7065 		if (kvm_vcpu_running(vcpu)) {
7066 			r = vcpu_enter_guest(vcpu);
7067 		} else {
7068 			r = vcpu_block(kvm, vcpu);
7069 		}
7070 
7071 		if (r <= 0)
7072 			break;
7073 
7074 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7075 		if (kvm_cpu_has_pending_timer(vcpu))
7076 			kvm_inject_pending_timer_irqs(vcpu);
7077 
7078 		if (dm_request_for_irq_injection(vcpu) &&
7079 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7080 			r = 0;
7081 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7082 			++vcpu->stat.request_irq_exits;
7083 			break;
7084 		}
7085 
7086 		kvm_check_async_pf_completion(vcpu);
7087 
7088 		if (signal_pending(current)) {
7089 			r = -EINTR;
7090 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7091 			++vcpu->stat.signal_exits;
7092 			break;
7093 		}
7094 		if (need_resched()) {
7095 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7096 			cond_resched();
7097 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7098 		}
7099 	}
7100 
7101 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7102 
7103 	return r;
7104 }
7105 
7106 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7107 {
7108 	int r;
7109 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7110 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7111 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7112 	if (r != EMULATE_DONE)
7113 		return 0;
7114 	return 1;
7115 }
7116 
7117 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7118 {
7119 	BUG_ON(!vcpu->arch.pio.count);
7120 
7121 	return complete_emulated_io(vcpu);
7122 }
7123 
7124 /*
7125  * Implements the following, as a state machine:
7126  *
7127  * read:
7128  *   for each fragment
7129  *     for each mmio piece in the fragment
7130  *       write gpa, len
7131  *       exit
7132  *       copy data
7133  *   execute insn
7134  *
7135  * write:
7136  *   for each fragment
7137  *     for each mmio piece in the fragment
7138  *       write gpa, len
7139  *       copy data
7140  *       exit
7141  */
7142 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7143 {
7144 	struct kvm_run *run = vcpu->run;
7145 	struct kvm_mmio_fragment *frag;
7146 	unsigned len;
7147 
7148 	BUG_ON(!vcpu->mmio_needed);
7149 
7150 	/* Complete previous fragment */
7151 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7152 	len = min(8u, frag->len);
7153 	if (!vcpu->mmio_is_write)
7154 		memcpy(frag->data, run->mmio.data, len);
7155 
7156 	if (frag->len <= 8) {
7157 		/* Switch to the next fragment. */
7158 		frag++;
7159 		vcpu->mmio_cur_fragment++;
7160 	} else {
7161 		/* Go forward to the next mmio piece. */
7162 		frag->data += len;
7163 		frag->gpa += len;
7164 		frag->len -= len;
7165 	}
7166 
7167 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7168 		vcpu->mmio_needed = 0;
7169 
7170 		/* FIXME: return into emulator if single-stepping.  */
7171 		if (vcpu->mmio_is_write)
7172 			return 1;
7173 		vcpu->mmio_read_completed = 1;
7174 		return complete_emulated_io(vcpu);
7175 	}
7176 
7177 	run->exit_reason = KVM_EXIT_MMIO;
7178 	run->mmio.phys_addr = frag->gpa;
7179 	if (vcpu->mmio_is_write)
7180 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7181 	run->mmio.len = min(8u, frag->len);
7182 	run->mmio.is_write = vcpu->mmio_is_write;
7183 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7184 	return 0;
7185 }
7186 
7187 
7188 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7189 {
7190 	struct fpu *fpu = &current->thread.fpu;
7191 	int r;
7192 	sigset_t sigsaved;
7193 
7194 	fpu__activate_curr(fpu);
7195 
7196 	if (vcpu->sigset_active)
7197 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7198 
7199 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7200 		kvm_vcpu_block(vcpu);
7201 		kvm_apic_accept_events(vcpu);
7202 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7203 		r = -EAGAIN;
7204 		goto out;
7205 	}
7206 
7207 	/* re-sync apic's tpr */
7208 	if (!lapic_in_kernel(vcpu)) {
7209 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7210 			r = -EINVAL;
7211 			goto out;
7212 		}
7213 	}
7214 
7215 	if (unlikely(vcpu->arch.complete_userspace_io)) {
7216 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7217 		vcpu->arch.complete_userspace_io = NULL;
7218 		r = cui(vcpu);
7219 		if (r <= 0)
7220 			goto out;
7221 	} else
7222 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7223 
7224 	if (kvm_run->immediate_exit)
7225 		r = -EINTR;
7226 	else
7227 		r = vcpu_run(vcpu);
7228 
7229 out:
7230 	post_kvm_run_save(vcpu);
7231 	if (vcpu->sigset_active)
7232 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7233 
7234 	return r;
7235 }
7236 
7237 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7238 {
7239 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7240 		/*
7241 		 * We are here if userspace calls get_regs() in the middle of
7242 		 * instruction emulation. Registers state needs to be copied
7243 		 * back from emulation context to vcpu. Userspace shouldn't do
7244 		 * that usually, but some bad designed PV devices (vmware
7245 		 * backdoor interface) need this to work
7246 		 */
7247 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7248 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7249 	}
7250 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7251 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7252 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7253 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7254 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7255 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7256 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7257 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7258 #ifdef CONFIG_X86_64
7259 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7260 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7261 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7262 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7263 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7264 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7265 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7266 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7267 #endif
7268 
7269 	regs->rip = kvm_rip_read(vcpu);
7270 	regs->rflags = kvm_get_rflags(vcpu);
7271 
7272 	return 0;
7273 }
7274 
7275 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7276 {
7277 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7278 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7279 
7280 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7281 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7282 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7283 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7284 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7285 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7286 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7287 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7288 #ifdef CONFIG_X86_64
7289 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7290 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7291 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7292 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7293 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7294 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7295 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7296 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7297 #endif
7298 
7299 	kvm_rip_write(vcpu, regs->rip);
7300 	kvm_set_rflags(vcpu, regs->rflags);
7301 
7302 	vcpu->arch.exception.pending = false;
7303 
7304 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7305 
7306 	return 0;
7307 }
7308 
7309 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7310 {
7311 	struct kvm_segment cs;
7312 
7313 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7314 	*db = cs.db;
7315 	*l = cs.l;
7316 }
7317 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7318 
7319 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7320 				  struct kvm_sregs *sregs)
7321 {
7322 	struct desc_ptr dt;
7323 
7324 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7325 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7326 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7327 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7328 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7329 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7330 
7331 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7332 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7333 
7334 	kvm_x86_ops->get_idt(vcpu, &dt);
7335 	sregs->idt.limit = dt.size;
7336 	sregs->idt.base = dt.address;
7337 	kvm_x86_ops->get_gdt(vcpu, &dt);
7338 	sregs->gdt.limit = dt.size;
7339 	sregs->gdt.base = dt.address;
7340 
7341 	sregs->cr0 = kvm_read_cr0(vcpu);
7342 	sregs->cr2 = vcpu->arch.cr2;
7343 	sregs->cr3 = kvm_read_cr3(vcpu);
7344 	sregs->cr4 = kvm_read_cr4(vcpu);
7345 	sregs->cr8 = kvm_get_cr8(vcpu);
7346 	sregs->efer = vcpu->arch.efer;
7347 	sregs->apic_base = kvm_get_apic_base(vcpu);
7348 
7349 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7350 
7351 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7352 		set_bit(vcpu->arch.interrupt.nr,
7353 			(unsigned long *)sregs->interrupt_bitmap);
7354 
7355 	return 0;
7356 }
7357 
7358 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7359 				    struct kvm_mp_state *mp_state)
7360 {
7361 	kvm_apic_accept_events(vcpu);
7362 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7363 					vcpu->arch.pv.pv_unhalted)
7364 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7365 	else
7366 		mp_state->mp_state = vcpu->arch.mp_state;
7367 
7368 	return 0;
7369 }
7370 
7371 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7372 				    struct kvm_mp_state *mp_state)
7373 {
7374 	if (!lapic_in_kernel(vcpu) &&
7375 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7376 		return -EINVAL;
7377 
7378 	/* INITs are latched while in SMM */
7379 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7380 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7381 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7382 		return -EINVAL;
7383 
7384 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7385 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7386 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7387 	} else
7388 		vcpu->arch.mp_state = mp_state->mp_state;
7389 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7390 	return 0;
7391 }
7392 
7393 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7394 		    int reason, bool has_error_code, u32 error_code)
7395 {
7396 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7397 	int ret;
7398 
7399 	init_emulate_ctxt(vcpu);
7400 
7401 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7402 				   has_error_code, error_code);
7403 
7404 	if (ret)
7405 		return EMULATE_FAIL;
7406 
7407 	kvm_rip_write(vcpu, ctxt->eip);
7408 	kvm_set_rflags(vcpu, ctxt->eflags);
7409 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7410 	return EMULATE_DONE;
7411 }
7412 EXPORT_SYMBOL_GPL(kvm_task_switch);
7413 
7414 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7415 				  struct kvm_sregs *sregs)
7416 {
7417 	struct msr_data apic_base_msr;
7418 	int mmu_reset_needed = 0;
7419 	int pending_vec, max_bits, idx;
7420 	struct desc_ptr dt;
7421 
7422 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7423 		return -EINVAL;
7424 
7425 	dt.size = sregs->idt.limit;
7426 	dt.address = sregs->idt.base;
7427 	kvm_x86_ops->set_idt(vcpu, &dt);
7428 	dt.size = sregs->gdt.limit;
7429 	dt.address = sregs->gdt.base;
7430 	kvm_x86_ops->set_gdt(vcpu, &dt);
7431 
7432 	vcpu->arch.cr2 = sregs->cr2;
7433 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7434 	vcpu->arch.cr3 = sregs->cr3;
7435 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7436 
7437 	kvm_set_cr8(vcpu, sregs->cr8);
7438 
7439 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7440 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7441 	apic_base_msr.data = sregs->apic_base;
7442 	apic_base_msr.host_initiated = true;
7443 	kvm_set_apic_base(vcpu, &apic_base_msr);
7444 
7445 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7446 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7447 	vcpu->arch.cr0 = sregs->cr0;
7448 
7449 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7450 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7451 	if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7452 		kvm_update_cpuid(vcpu);
7453 
7454 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7455 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7456 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7457 		mmu_reset_needed = 1;
7458 	}
7459 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7460 
7461 	if (mmu_reset_needed)
7462 		kvm_mmu_reset_context(vcpu);
7463 
7464 	max_bits = KVM_NR_INTERRUPTS;
7465 	pending_vec = find_first_bit(
7466 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7467 	if (pending_vec < max_bits) {
7468 		kvm_queue_interrupt(vcpu, pending_vec, false);
7469 		pr_debug("Set back pending irq %d\n", pending_vec);
7470 	}
7471 
7472 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7473 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7474 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7475 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7476 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7477 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7478 
7479 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7480 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7481 
7482 	update_cr8_intercept(vcpu);
7483 
7484 	/* Older userspace won't unhalt the vcpu on reset. */
7485 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7486 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7487 	    !is_protmode(vcpu))
7488 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7489 
7490 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7491 
7492 	return 0;
7493 }
7494 
7495 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7496 					struct kvm_guest_debug *dbg)
7497 {
7498 	unsigned long rflags;
7499 	int i, r;
7500 
7501 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7502 		r = -EBUSY;
7503 		if (vcpu->arch.exception.pending)
7504 			goto out;
7505 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7506 			kvm_queue_exception(vcpu, DB_VECTOR);
7507 		else
7508 			kvm_queue_exception(vcpu, BP_VECTOR);
7509 	}
7510 
7511 	/*
7512 	 * Read rflags as long as potentially injected trace flags are still
7513 	 * filtered out.
7514 	 */
7515 	rflags = kvm_get_rflags(vcpu);
7516 
7517 	vcpu->guest_debug = dbg->control;
7518 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7519 		vcpu->guest_debug = 0;
7520 
7521 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7522 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7523 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7524 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7525 	} else {
7526 		for (i = 0; i < KVM_NR_DB_REGS; i++)
7527 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7528 	}
7529 	kvm_update_dr7(vcpu);
7530 
7531 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7532 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7533 			get_segment_base(vcpu, VCPU_SREG_CS);
7534 
7535 	/*
7536 	 * Trigger an rflags update that will inject or remove the trace
7537 	 * flags.
7538 	 */
7539 	kvm_set_rflags(vcpu, rflags);
7540 
7541 	kvm_x86_ops->update_bp_intercept(vcpu);
7542 
7543 	r = 0;
7544 
7545 out:
7546 
7547 	return r;
7548 }
7549 
7550 /*
7551  * Translate a guest virtual address to a guest physical address.
7552  */
7553 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7554 				    struct kvm_translation *tr)
7555 {
7556 	unsigned long vaddr = tr->linear_address;
7557 	gpa_t gpa;
7558 	int idx;
7559 
7560 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7561 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7562 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7563 	tr->physical_address = gpa;
7564 	tr->valid = gpa != UNMAPPED_GVA;
7565 	tr->writeable = 1;
7566 	tr->usermode = 0;
7567 
7568 	return 0;
7569 }
7570 
7571 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7572 {
7573 	struct fxregs_state *fxsave =
7574 			&vcpu->arch.guest_fpu.state.fxsave;
7575 
7576 	memcpy(fpu->fpr, fxsave->st_space, 128);
7577 	fpu->fcw = fxsave->cwd;
7578 	fpu->fsw = fxsave->swd;
7579 	fpu->ftwx = fxsave->twd;
7580 	fpu->last_opcode = fxsave->fop;
7581 	fpu->last_ip = fxsave->rip;
7582 	fpu->last_dp = fxsave->rdp;
7583 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7584 
7585 	return 0;
7586 }
7587 
7588 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7589 {
7590 	struct fxregs_state *fxsave =
7591 			&vcpu->arch.guest_fpu.state.fxsave;
7592 
7593 	memcpy(fxsave->st_space, fpu->fpr, 128);
7594 	fxsave->cwd = fpu->fcw;
7595 	fxsave->swd = fpu->fsw;
7596 	fxsave->twd = fpu->ftwx;
7597 	fxsave->fop = fpu->last_opcode;
7598 	fxsave->rip = fpu->last_ip;
7599 	fxsave->rdp = fpu->last_dp;
7600 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7601 
7602 	return 0;
7603 }
7604 
7605 static void fx_init(struct kvm_vcpu *vcpu)
7606 {
7607 	fpstate_init(&vcpu->arch.guest_fpu.state);
7608 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7609 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7610 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7611 
7612 	/*
7613 	 * Ensure guest xcr0 is valid for loading
7614 	 */
7615 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7616 
7617 	vcpu->arch.cr0 |= X86_CR0_ET;
7618 }
7619 
7620 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7621 {
7622 	if (vcpu->guest_fpu_loaded)
7623 		return;
7624 
7625 	/*
7626 	 * Restore all possible states in the guest,
7627 	 * and assume host would use all available bits.
7628 	 * Guest xcr0 would be loaded later.
7629 	 */
7630 	vcpu->guest_fpu_loaded = 1;
7631 	__kernel_fpu_begin();
7632 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7633 	trace_kvm_fpu(1);
7634 }
7635 
7636 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7637 {
7638 	if (!vcpu->guest_fpu_loaded)
7639 		return;
7640 
7641 	vcpu->guest_fpu_loaded = 0;
7642 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7643 	__kernel_fpu_end();
7644 	++vcpu->stat.fpu_reload;
7645 	trace_kvm_fpu(0);
7646 }
7647 
7648 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7649 {
7650 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7651 
7652 	kvmclock_reset(vcpu);
7653 
7654 	kvm_x86_ops->vcpu_free(vcpu);
7655 	free_cpumask_var(wbinvd_dirty_mask);
7656 }
7657 
7658 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7659 						unsigned int id)
7660 {
7661 	struct kvm_vcpu *vcpu;
7662 
7663 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7664 		printk_once(KERN_WARNING
7665 		"kvm: SMP vm created on host with unstable TSC; "
7666 		"guest TSC will not be reliable\n");
7667 
7668 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7669 
7670 	return vcpu;
7671 }
7672 
7673 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7674 {
7675 	int r;
7676 
7677 	kvm_vcpu_mtrr_init(vcpu);
7678 	r = vcpu_load(vcpu);
7679 	if (r)
7680 		return r;
7681 	kvm_vcpu_reset(vcpu, false);
7682 	kvm_mmu_setup(vcpu);
7683 	vcpu_put(vcpu);
7684 	return r;
7685 }
7686 
7687 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7688 {
7689 	struct msr_data msr;
7690 	struct kvm *kvm = vcpu->kvm;
7691 
7692 	kvm_hv_vcpu_postcreate(vcpu);
7693 
7694 	if (vcpu_load(vcpu))
7695 		return;
7696 	msr.data = 0x0;
7697 	msr.index = MSR_IA32_TSC;
7698 	msr.host_initiated = true;
7699 	kvm_write_tsc(vcpu, &msr);
7700 	vcpu_put(vcpu);
7701 
7702 	if (!kvmclock_periodic_sync)
7703 		return;
7704 
7705 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7706 					KVMCLOCK_SYNC_PERIOD);
7707 }
7708 
7709 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7710 {
7711 	int r;
7712 	vcpu->arch.apf.msr_val = 0;
7713 
7714 	r = vcpu_load(vcpu);
7715 	BUG_ON(r);
7716 	kvm_mmu_unload(vcpu);
7717 	vcpu_put(vcpu);
7718 
7719 	kvm_x86_ops->vcpu_free(vcpu);
7720 }
7721 
7722 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7723 {
7724 	vcpu->arch.hflags = 0;
7725 
7726 	vcpu->arch.smi_pending = 0;
7727 	atomic_set(&vcpu->arch.nmi_queued, 0);
7728 	vcpu->arch.nmi_pending = 0;
7729 	vcpu->arch.nmi_injected = false;
7730 	kvm_clear_interrupt_queue(vcpu);
7731 	kvm_clear_exception_queue(vcpu);
7732 
7733 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7734 	kvm_update_dr0123(vcpu);
7735 	vcpu->arch.dr6 = DR6_INIT;
7736 	kvm_update_dr6(vcpu);
7737 	vcpu->arch.dr7 = DR7_FIXED_1;
7738 	kvm_update_dr7(vcpu);
7739 
7740 	vcpu->arch.cr2 = 0;
7741 
7742 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7743 	vcpu->arch.apf.msr_val = 0;
7744 	vcpu->arch.st.msr_val = 0;
7745 
7746 	kvmclock_reset(vcpu);
7747 
7748 	kvm_clear_async_pf_completion_queue(vcpu);
7749 	kvm_async_pf_hash_reset(vcpu);
7750 	vcpu->arch.apf.halted = false;
7751 
7752 	if (!init_event) {
7753 		kvm_pmu_reset(vcpu);
7754 		vcpu->arch.smbase = 0x30000;
7755 
7756 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7757 		vcpu->arch.msr_misc_features_enables = 0;
7758 	}
7759 
7760 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7761 	vcpu->arch.regs_avail = ~0;
7762 	vcpu->arch.regs_dirty = ~0;
7763 
7764 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7765 }
7766 
7767 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7768 {
7769 	struct kvm_segment cs;
7770 
7771 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7772 	cs.selector = vector << 8;
7773 	cs.base = vector << 12;
7774 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7775 	kvm_rip_write(vcpu, 0);
7776 }
7777 
7778 int kvm_arch_hardware_enable(void)
7779 {
7780 	struct kvm *kvm;
7781 	struct kvm_vcpu *vcpu;
7782 	int i;
7783 	int ret;
7784 	u64 local_tsc;
7785 	u64 max_tsc = 0;
7786 	bool stable, backwards_tsc = false;
7787 
7788 	kvm_shared_msr_cpu_online();
7789 	ret = kvm_x86_ops->hardware_enable();
7790 	if (ret != 0)
7791 		return ret;
7792 
7793 	local_tsc = rdtsc();
7794 	stable = !check_tsc_unstable();
7795 	list_for_each_entry(kvm, &vm_list, vm_list) {
7796 		kvm_for_each_vcpu(i, vcpu, kvm) {
7797 			if (!stable && vcpu->cpu == smp_processor_id())
7798 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7799 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7800 				backwards_tsc = true;
7801 				if (vcpu->arch.last_host_tsc > max_tsc)
7802 					max_tsc = vcpu->arch.last_host_tsc;
7803 			}
7804 		}
7805 	}
7806 
7807 	/*
7808 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7809 	 * platforms that reset TSC during suspend or hibernate actions, but
7810 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7811 	 * detect that condition here, which happens early in CPU bringup,
7812 	 * before any KVM threads can be running.  Unfortunately, we can't
7813 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7814 	 * enough into CPU bringup that we know how much real time has actually
7815 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7816 	 * variables that haven't been updated yet.
7817 	 *
7818 	 * So we simply find the maximum observed TSC above, then record the
7819 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7820 	 * the adjustment will be applied.  Note that we accumulate
7821 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7822 	 * gets a chance to run again.  In the event that no KVM threads get a
7823 	 * chance to run, we will miss the entire elapsed period, as we'll have
7824 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7825 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7826 	 * uniform across all VCPUs (not to mention the scenario is extremely
7827 	 * unlikely). It is possible that a second hibernate recovery happens
7828 	 * much faster than a first, causing the observed TSC here to be
7829 	 * smaller; this would require additional padding adjustment, which is
7830 	 * why we set last_host_tsc to the local tsc observed here.
7831 	 *
7832 	 * N.B. - this code below runs only on platforms with reliable TSC,
7833 	 * as that is the only way backwards_tsc is set above.  Also note
7834 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7835 	 * have the same delta_cyc adjustment applied if backwards_tsc
7836 	 * is detected.  Note further, this adjustment is only done once,
7837 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7838 	 * called multiple times (one for each physical CPU bringup).
7839 	 *
7840 	 * Platforms with unreliable TSCs don't have to deal with this, they
7841 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7842 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7843 	 * guarantee that they stay in perfect synchronization.
7844 	 */
7845 	if (backwards_tsc) {
7846 		u64 delta_cyc = max_tsc - local_tsc;
7847 		list_for_each_entry(kvm, &vm_list, vm_list) {
7848 			kvm->arch.backwards_tsc_observed = true;
7849 			kvm_for_each_vcpu(i, vcpu, kvm) {
7850 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7851 				vcpu->arch.last_host_tsc = local_tsc;
7852 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7853 			}
7854 
7855 			/*
7856 			 * We have to disable TSC offset matching.. if you were
7857 			 * booting a VM while issuing an S4 host suspend....
7858 			 * you may have some problem.  Solving this issue is
7859 			 * left as an exercise to the reader.
7860 			 */
7861 			kvm->arch.last_tsc_nsec = 0;
7862 			kvm->arch.last_tsc_write = 0;
7863 		}
7864 
7865 	}
7866 	return 0;
7867 }
7868 
7869 void kvm_arch_hardware_disable(void)
7870 {
7871 	kvm_x86_ops->hardware_disable();
7872 	drop_user_return_notifiers();
7873 }
7874 
7875 int kvm_arch_hardware_setup(void)
7876 {
7877 	int r;
7878 
7879 	r = kvm_x86_ops->hardware_setup();
7880 	if (r != 0)
7881 		return r;
7882 
7883 	if (kvm_has_tsc_control) {
7884 		/*
7885 		 * Make sure the user can only configure tsc_khz values that
7886 		 * fit into a signed integer.
7887 		 * A min value is not calculated needed because it will always
7888 		 * be 1 on all machines.
7889 		 */
7890 		u64 max = min(0x7fffffffULL,
7891 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7892 		kvm_max_guest_tsc_khz = max;
7893 
7894 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7895 	}
7896 
7897 	kvm_init_msr_list();
7898 	return 0;
7899 }
7900 
7901 void kvm_arch_hardware_unsetup(void)
7902 {
7903 	kvm_x86_ops->hardware_unsetup();
7904 }
7905 
7906 void kvm_arch_check_processor_compat(void *rtn)
7907 {
7908 	kvm_x86_ops->check_processor_compatibility(rtn);
7909 }
7910 
7911 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7912 {
7913 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7914 }
7915 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7916 
7917 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7918 {
7919 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7920 }
7921 
7922 struct static_key kvm_no_apic_vcpu __read_mostly;
7923 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7924 
7925 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7926 {
7927 	struct page *page;
7928 	struct kvm *kvm;
7929 	int r;
7930 
7931 	BUG_ON(vcpu->kvm == NULL);
7932 	kvm = vcpu->kvm;
7933 
7934 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7935 	vcpu->arch.pv.pv_unhalted = false;
7936 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7937 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7938 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7939 	else
7940 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7941 
7942 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7943 	if (!page) {
7944 		r = -ENOMEM;
7945 		goto fail;
7946 	}
7947 	vcpu->arch.pio_data = page_address(page);
7948 
7949 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7950 
7951 	r = kvm_mmu_create(vcpu);
7952 	if (r < 0)
7953 		goto fail_free_pio_data;
7954 
7955 	if (irqchip_in_kernel(kvm)) {
7956 		r = kvm_create_lapic(vcpu);
7957 		if (r < 0)
7958 			goto fail_mmu_destroy;
7959 	} else
7960 		static_key_slow_inc(&kvm_no_apic_vcpu);
7961 
7962 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7963 				       GFP_KERNEL);
7964 	if (!vcpu->arch.mce_banks) {
7965 		r = -ENOMEM;
7966 		goto fail_free_lapic;
7967 	}
7968 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7969 
7970 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7971 		r = -ENOMEM;
7972 		goto fail_free_mce_banks;
7973 	}
7974 
7975 	fx_init(vcpu);
7976 
7977 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7978 	vcpu->arch.pv_time_enabled = false;
7979 
7980 	vcpu->arch.guest_supported_xcr0 = 0;
7981 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7982 
7983 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7984 
7985 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7986 
7987 	kvm_async_pf_hash_reset(vcpu);
7988 	kvm_pmu_init(vcpu);
7989 
7990 	vcpu->arch.pending_external_vector = -1;
7991 
7992 	kvm_hv_vcpu_init(vcpu);
7993 
7994 	return 0;
7995 
7996 fail_free_mce_banks:
7997 	kfree(vcpu->arch.mce_banks);
7998 fail_free_lapic:
7999 	kvm_free_lapic(vcpu);
8000 fail_mmu_destroy:
8001 	kvm_mmu_destroy(vcpu);
8002 fail_free_pio_data:
8003 	free_page((unsigned long)vcpu->arch.pio_data);
8004 fail:
8005 	return r;
8006 }
8007 
8008 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8009 {
8010 	int idx;
8011 
8012 	kvm_hv_vcpu_uninit(vcpu);
8013 	kvm_pmu_destroy(vcpu);
8014 	kfree(vcpu->arch.mce_banks);
8015 	kvm_free_lapic(vcpu);
8016 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8017 	kvm_mmu_destroy(vcpu);
8018 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8019 	free_page((unsigned long)vcpu->arch.pio_data);
8020 	if (!lapic_in_kernel(vcpu))
8021 		static_key_slow_dec(&kvm_no_apic_vcpu);
8022 }
8023 
8024 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8025 {
8026 	kvm_x86_ops->sched_in(vcpu, cpu);
8027 }
8028 
8029 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8030 {
8031 	if (type)
8032 		return -EINVAL;
8033 
8034 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8035 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8036 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8037 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8038 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8039 
8040 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8041 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8042 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8043 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8044 		&kvm->arch.irq_sources_bitmap);
8045 
8046 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8047 	mutex_init(&kvm->arch.apic_map_lock);
8048 	mutex_init(&kvm->arch.hyperv.hv_lock);
8049 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8050 
8051 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8052 	pvclock_update_vm_gtod_copy(kvm);
8053 
8054 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8055 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8056 
8057 	kvm_page_track_init(kvm);
8058 	kvm_mmu_init_vm(kvm);
8059 
8060 	if (kvm_x86_ops->vm_init)
8061 		return kvm_x86_ops->vm_init(kvm);
8062 
8063 	return 0;
8064 }
8065 
8066 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8067 {
8068 	int r;
8069 	r = vcpu_load(vcpu);
8070 	BUG_ON(r);
8071 	kvm_mmu_unload(vcpu);
8072 	vcpu_put(vcpu);
8073 }
8074 
8075 static void kvm_free_vcpus(struct kvm *kvm)
8076 {
8077 	unsigned int i;
8078 	struct kvm_vcpu *vcpu;
8079 
8080 	/*
8081 	 * Unpin any mmu pages first.
8082 	 */
8083 	kvm_for_each_vcpu(i, vcpu, kvm) {
8084 		kvm_clear_async_pf_completion_queue(vcpu);
8085 		kvm_unload_vcpu_mmu(vcpu);
8086 	}
8087 	kvm_for_each_vcpu(i, vcpu, kvm)
8088 		kvm_arch_vcpu_free(vcpu);
8089 
8090 	mutex_lock(&kvm->lock);
8091 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8092 		kvm->vcpus[i] = NULL;
8093 
8094 	atomic_set(&kvm->online_vcpus, 0);
8095 	mutex_unlock(&kvm->lock);
8096 }
8097 
8098 void kvm_arch_sync_events(struct kvm *kvm)
8099 {
8100 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8101 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8102 	kvm_free_pit(kvm);
8103 }
8104 
8105 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8106 {
8107 	int i, r;
8108 	unsigned long hva;
8109 	struct kvm_memslots *slots = kvm_memslots(kvm);
8110 	struct kvm_memory_slot *slot, old;
8111 
8112 	/* Called with kvm->slots_lock held.  */
8113 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8114 		return -EINVAL;
8115 
8116 	slot = id_to_memslot(slots, id);
8117 	if (size) {
8118 		if (slot->npages)
8119 			return -EEXIST;
8120 
8121 		/*
8122 		 * MAP_SHARED to prevent internal slot pages from being moved
8123 		 * by fork()/COW.
8124 		 */
8125 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8126 			      MAP_SHARED | MAP_ANONYMOUS, 0);
8127 		if (IS_ERR((void *)hva))
8128 			return PTR_ERR((void *)hva);
8129 	} else {
8130 		if (!slot->npages)
8131 			return 0;
8132 
8133 		hva = 0;
8134 	}
8135 
8136 	old = *slot;
8137 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8138 		struct kvm_userspace_memory_region m;
8139 
8140 		m.slot = id | (i << 16);
8141 		m.flags = 0;
8142 		m.guest_phys_addr = gpa;
8143 		m.userspace_addr = hva;
8144 		m.memory_size = size;
8145 		r = __kvm_set_memory_region(kvm, &m);
8146 		if (r < 0)
8147 			return r;
8148 	}
8149 
8150 	if (!size) {
8151 		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8152 		WARN_ON(r < 0);
8153 	}
8154 
8155 	return 0;
8156 }
8157 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8158 
8159 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8160 {
8161 	int r;
8162 
8163 	mutex_lock(&kvm->slots_lock);
8164 	r = __x86_set_memory_region(kvm, id, gpa, size);
8165 	mutex_unlock(&kvm->slots_lock);
8166 
8167 	return r;
8168 }
8169 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8170 
8171 void kvm_arch_destroy_vm(struct kvm *kvm)
8172 {
8173 	if (current->mm == kvm->mm) {
8174 		/*
8175 		 * Free memory regions allocated on behalf of userspace,
8176 		 * unless the the memory map has changed due to process exit
8177 		 * or fd copying.
8178 		 */
8179 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8180 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8181 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8182 	}
8183 	if (kvm_x86_ops->vm_destroy)
8184 		kvm_x86_ops->vm_destroy(kvm);
8185 	kvm_pic_destroy(kvm);
8186 	kvm_ioapic_destroy(kvm);
8187 	kvm_free_vcpus(kvm);
8188 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8189 	kvm_mmu_uninit_vm(kvm);
8190 	kvm_page_track_cleanup(kvm);
8191 }
8192 
8193 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8194 			   struct kvm_memory_slot *dont)
8195 {
8196 	int i;
8197 
8198 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8199 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8200 			kvfree(free->arch.rmap[i]);
8201 			free->arch.rmap[i] = NULL;
8202 		}
8203 		if (i == 0)
8204 			continue;
8205 
8206 		if (!dont || free->arch.lpage_info[i - 1] !=
8207 			     dont->arch.lpage_info[i - 1]) {
8208 			kvfree(free->arch.lpage_info[i - 1]);
8209 			free->arch.lpage_info[i - 1] = NULL;
8210 		}
8211 	}
8212 
8213 	kvm_page_track_free_memslot(free, dont);
8214 }
8215 
8216 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8217 			    unsigned long npages)
8218 {
8219 	int i;
8220 
8221 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8222 		struct kvm_lpage_info *linfo;
8223 		unsigned long ugfn;
8224 		int lpages;
8225 		int level = i + 1;
8226 
8227 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
8228 				      slot->base_gfn, level) + 1;
8229 
8230 		slot->arch.rmap[i] =
8231 			kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8232 		if (!slot->arch.rmap[i])
8233 			goto out_free;
8234 		if (i == 0)
8235 			continue;
8236 
8237 		linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8238 		if (!linfo)
8239 			goto out_free;
8240 
8241 		slot->arch.lpage_info[i - 1] = linfo;
8242 
8243 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8244 			linfo[0].disallow_lpage = 1;
8245 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8246 			linfo[lpages - 1].disallow_lpage = 1;
8247 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
8248 		/*
8249 		 * If the gfn and userspace address are not aligned wrt each
8250 		 * other, or if explicitly asked to, disable large page
8251 		 * support for this slot
8252 		 */
8253 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8254 		    !kvm_largepages_enabled()) {
8255 			unsigned long j;
8256 
8257 			for (j = 0; j < lpages; ++j)
8258 				linfo[j].disallow_lpage = 1;
8259 		}
8260 	}
8261 
8262 	if (kvm_page_track_create_memslot(slot, npages))
8263 		goto out_free;
8264 
8265 	return 0;
8266 
8267 out_free:
8268 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8269 		kvfree(slot->arch.rmap[i]);
8270 		slot->arch.rmap[i] = NULL;
8271 		if (i == 0)
8272 			continue;
8273 
8274 		kvfree(slot->arch.lpage_info[i - 1]);
8275 		slot->arch.lpage_info[i - 1] = NULL;
8276 	}
8277 	return -ENOMEM;
8278 }
8279 
8280 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8281 {
8282 	/*
8283 	 * memslots->generation has been incremented.
8284 	 * mmio generation may have reached its maximum value.
8285 	 */
8286 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8287 }
8288 
8289 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8290 				struct kvm_memory_slot *memslot,
8291 				const struct kvm_userspace_memory_region *mem,
8292 				enum kvm_mr_change change)
8293 {
8294 	return 0;
8295 }
8296 
8297 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8298 				     struct kvm_memory_slot *new)
8299 {
8300 	/* Still write protect RO slot */
8301 	if (new->flags & KVM_MEM_READONLY) {
8302 		kvm_mmu_slot_remove_write_access(kvm, new);
8303 		return;
8304 	}
8305 
8306 	/*
8307 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
8308 	 *
8309 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
8310 	 *
8311 	 *  - KVM_MR_CREATE with dirty logging is disabled
8312 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8313 	 *
8314 	 * The reason is, in case of PML, we need to set D-bit for any slots
8315 	 * with dirty logging disabled in order to eliminate unnecessary GPA
8316 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
8317 	 * guarantees leaving PML enabled during guest's lifetime won't have
8318 	 * any additonal overhead from PML when guest is running with dirty
8319 	 * logging disabled for memory slots.
8320 	 *
8321 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8322 	 * to dirty logging mode.
8323 	 *
8324 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8325 	 *
8326 	 * In case of write protect:
8327 	 *
8328 	 * Write protect all pages for dirty logging.
8329 	 *
8330 	 * All the sptes including the large sptes which point to this
8331 	 * slot are set to readonly. We can not create any new large
8332 	 * spte on this slot until the end of the logging.
8333 	 *
8334 	 * See the comments in fast_page_fault().
8335 	 */
8336 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8337 		if (kvm_x86_ops->slot_enable_log_dirty)
8338 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8339 		else
8340 			kvm_mmu_slot_remove_write_access(kvm, new);
8341 	} else {
8342 		if (kvm_x86_ops->slot_disable_log_dirty)
8343 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8344 	}
8345 }
8346 
8347 void kvm_arch_commit_memory_region(struct kvm *kvm,
8348 				const struct kvm_userspace_memory_region *mem,
8349 				const struct kvm_memory_slot *old,
8350 				const struct kvm_memory_slot *new,
8351 				enum kvm_mr_change change)
8352 {
8353 	int nr_mmu_pages = 0;
8354 
8355 	if (!kvm->arch.n_requested_mmu_pages)
8356 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8357 
8358 	if (nr_mmu_pages)
8359 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8360 
8361 	/*
8362 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
8363 	 * sptes have to be split.  If live migration is successful, the guest
8364 	 * in the source machine will be destroyed and large sptes will be
8365 	 * created in the destination. However, if the guest continues to run
8366 	 * in the source machine (for example if live migration fails), small
8367 	 * sptes will remain around and cause bad performance.
8368 	 *
8369 	 * Scan sptes if dirty logging has been stopped, dropping those
8370 	 * which can be collapsed into a single large-page spte.  Later
8371 	 * page faults will create the large-page sptes.
8372 	 */
8373 	if ((change != KVM_MR_DELETE) &&
8374 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8375 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8376 		kvm_mmu_zap_collapsible_sptes(kvm, new);
8377 
8378 	/*
8379 	 * Set up write protection and/or dirty logging for the new slot.
8380 	 *
8381 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8382 	 * been zapped so no dirty logging staff is needed for old slot. For
8383 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8384 	 * new and it's also covered when dealing with the new slot.
8385 	 *
8386 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
8387 	 */
8388 	if (change != KVM_MR_DELETE)
8389 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8390 }
8391 
8392 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8393 {
8394 	kvm_mmu_invalidate_zap_all_pages(kvm);
8395 }
8396 
8397 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8398 				   struct kvm_memory_slot *slot)
8399 {
8400 	kvm_page_track_flush_slot(kvm, slot);
8401 }
8402 
8403 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8404 {
8405 	if (!list_empty_careful(&vcpu->async_pf.done))
8406 		return true;
8407 
8408 	if (kvm_apic_has_events(vcpu))
8409 		return true;
8410 
8411 	if (vcpu->arch.pv.pv_unhalted)
8412 		return true;
8413 
8414 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8415 	    (vcpu->arch.nmi_pending &&
8416 	     kvm_x86_ops->nmi_allowed(vcpu)))
8417 		return true;
8418 
8419 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8420 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
8421 		return true;
8422 
8423 	if (kvm_arch_interrupt_allowed(vcpu) &&
8424 	    kvm_cpu_has_interrupt(vcpu))
8425 		return true;
8426 
8427 	if (kvm_hv_has_stimer_pending(vcpu))
8428 		return true;
8429 
8430 	return false;
8431 }
8432 
8433 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8434 {
8435 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8436 }
8437 
8438 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8439 {
8440 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8441 }
8442 
8443 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8444 {
8445 	return kvm_x86_ops->interrupt_allowed(vcpu);
8446 }
8447 
8448 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8449 {
8450 	if (is_64_bit_mode(vcpu))
8451 		return kvm_rip_read(vcpu);
8452 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8453 		     kvm_rip_read(vcpu));
8454 }
8455 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8456 
8457 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8458 {
8459 	return kvm_get_linear_rip(vcpu) == linear_rip;
8460 }
8461 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8462 
8463 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8464 {
8465 	unsigned long rflags;
8466 
8467 	rflags = kvm_x86_ops->get_rflags(vcpu);
8468 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8469 		rflags &= ~X86_EFLAGS_TF;
8470 	return rflags;
8471 }
8472 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8473 
8474 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8475 {
8476 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8477 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8478 		rflags |= X86_EFLAGS_TF;
8479 	kvm_x86_ops->set_rflags(vcpu, rflags);
8480 }
8481 
8482 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8483 {
8484 	__kvm_set_rflags(vcpu, rflags);
8485 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8486 }
8487 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8488 
8489 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8490 {
8491 	int r;
8492 
8493 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8494 	      work->wakeup_all)
8495 		return;
8496 
8497 	r = kvm_mmu_reload(vcpu);
8498 	if (unlikely(r))
8499 		return;
8500 
8501 	if (!vcpu->arch.mmu.direct_map &&
8502 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8503 		return;
8504 
8505 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8506 }
8507 
8508 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8509 {
8510 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8511 }
8512 
8513 static inline u32 kvm_async_pf_next_probe(u32 key)
8514 {
8515 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8516 }
8517 
8518 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8519 {
8520 	u32 key = kvm_async_pf_hash_fn(gfn);
8521 
8522 	while (vcpu->arch.apf.gfns[key] != ~0)
8523 		key = kvm_async_pf_next_probe(key);
8524 
8525 	vcpu->arch.apf.gfns[key] = gfn;
8526 }
8527 
8528 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8529 {
8530 	int i;
8531 	u32 key = kvm_async_pf_hash_fn(gfn);
8532 
8533 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8534 		     (vcpu->arch.apf.gfns[key] != gfn &&
8535 		      vcpu->arch.apf.gfns[key] != ~0); i++)
8536 		key = kvm_async_pf_next_probe(key);
8537 
8538 	return key;
8539 }
8540 
8541 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8542 {
8543 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8544 }
8545 
8546 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8547 {
8548 	u32 i, j, k;
8549 
8550 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8551 	while (true) {
8552 		vcpu->arch.apf.gfns[i] = ~0;
8553 		do {
8554 			j = kvm_async_pf_next_probe(j);
8555 			if (vcpu->arch.apf.gfns[j] == ~0)
8556 				return;
8557 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8558 			/*
8559 			 * k lies cyclically in ]i,j]
8560 			 * |    i.k.j |
8561 			 * |....j i.k.| or  |.k..j i...|
8562 			 */
8563 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8564 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8565 		i = j;
8566 	}
8567 }
8568 
8569 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8570 {
8571 
8572 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8573 				      sizeof(val));
8574 }
8575 
8576 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8577 				     struct kvm_async_pf *work)
8578 {
8579 	struct x86_exception fault;
8580 
8581 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8582 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8583 
8584 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8585 	    (vcpu->arch.apf.send_user_only &&
8586 	     kvm_x86_ops->get_cpl(vcpu) == 0))
8587 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8588 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8589 		fault.vector = PF_VECTOR;
8590 		fault.error_code_valid = true;
8591 		fault.error_code = 0;
8592 		fault.nested_page_fault = false;
8593 		fault.address = work->arch.token;
8594 		fault.async_page_fault = true;
8595 		kvm_inject_page_fault(vcpu, &fault);
8596 	}
8597 }
8598 
8599 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8600 				 struct kvm_async_pf *work)
8601 {
8602 	struct x86_exception fault;
8603 
8604 	if (work->wakeup_all)
8605 		work->arch.token = ~0; /* broadcast wakeup */
8606 	else
8607 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8608 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8609 
8610 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8611 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8612 		fault.vector = PF_VECTOR;
8613 		fault.error_code_valid = true;
8614 		fault.error_code = 0;
8615 		fault.nested_page_fault = false;
8616 		fault.address = work->arch.token;
8617 		fault.async_page_fault = true;
8618 		kvm_inject_page_fault(vcpu, &fault);
8619 	}
8620 	vcpu->arch.apf.halted = false;
8621 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8622 }
8623 
8624 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8625 {
8626 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8627 		return true;
8628 	else
8629 		return kvm_can_do_async_pf(vcpu);
8630 }
8631 
8632 void kvm_arch_start_assignment(struct kvm *kvm)
8633 {
8634 	atomic_inc(&kvm->arch.assigned_device_count);
8635 }
8636 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8637 
8638 void kvm_arch_end_assignment(struct kvm *kvm)
8639 {
8640 	atomic_dec(&kvm->arch.assigned_device_count);
8641 }
8642 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8643 
8644 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8645 {
8646 	return atomic_read(&kvm->arch.assigned_device_count);
8647 }
8648 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8649 
8650 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8651 {
8652 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8653 }
8654 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8655 
8656 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8657 {
8658 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8659 }
8660 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8661 
8662 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8663 {
8664 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8665 }
8666 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8667 
8668 bool kvm_arch_has_irq_bypass(void)
8669 {
8670 	return kvm_x86_ops->update_pi_irte != NULL;
8671 }
8672 
8673 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8674 				      struct irq_bypass_producer *prod)
8675 {
8676 	struct kvm_kernel_irqfd *irqfd =
8677 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8678 
8679 	irqfd->producer = prod;
8680 
8681 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8682 					   prod->irq, irqfd->gsi, 1);
8683 }
8684 
8685 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8686 				      struct irq_bypass_producer *prod)
8687 {
8688 	int ret;
8689 	struct kvm_kernel_irqfd *irqfd =
8690 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8691 
8692 	WARN_ON(irqfd->producer != prod);
8693 	irqfd->producer = NULL;
8694 
8695 	/*
8696 	 * When producer of consumer is unregistered, we change back to
8697 	 * remapped mode, so we can re-use the current implementation
8698 	 * when the irq is masked/disabled or the consumer side (KVM
8699 	 * int this case doesn't want to receive the interrupts.
8700 	*/
8701 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8702 	if (ret)
8703 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8704 		       " fails: %d\n", irqfd->consumer.token, ret);
8705 }
8706 
8707 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8708 				   uint32_t guest_irq, bool set)
8709 {
8710 	if (!kvm_x86_ops->update_pi_irte)
8711 		return -EINVAL;
8712 
8713 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8714 }
8715 
8716 bool kvm_vector_hashing_enabled(void)
8717 {
8718 	return vector_hashing;
8719 }
8720 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8721 
8722 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8723 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8724 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8725 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8726 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8727 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8728 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8729 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8730 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8731 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8732 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8733 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8738 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8739 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8740 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
8741