xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 171fa692)
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 static bool __read_mostly backwards_tsc_observed = false;
138 
139 #define KVM_NR_SHARED_MSRS 16
140 
141 struct kvm_shared_msrs_global {
142 	int nr;
143 	u32 msrs[KVM_NR_SHARED_MSRS];
144 };
145 
146 struct kvm_shared_msrs {
147 	struct user_return_notifier urn;
148 	bool registered;
149 	struct kvm_shared_msr_values {
150 		u64 host;
151 		u64 curr;
152 	} values[KVM_NR_SHARED_MSRS];
153 };
154 
155 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
156 static struct kvm_shared_msrs __percpu *shared_msrs;
157 
158 struct kvm_stats_debugfs_item debugfs_entries[] = {
159 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
160 	{ "pf_guest", VCPU_STAT(pf_guest) },
161 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
162 	{ "invlpg", VCPU_STAT(invlpg) },
163 	{ "exits", VCPU_STAT(exits) },
164 	{ "io_exits", VCPU_STAT(io_exits) },
165 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
166 	{ "signal_exits", VCPU_STAT(signal_exits) },
167 	{ "irq_window", VCPU_STAT(irq_window_exits) },
168 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
169 	{ "halt_exits", VCPU_STAT(halt_exits) },
170 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
171 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
172 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
173 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
174 	{ "hypercalls", VCPU_STAT(hypercalls) },
175 	{ "request_irq", VCPU_STAT(request_irq_exits) },
176 	{ "irq_exits", VCPU_STAT(irq_exits) },
177 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
178 	{ "efer_reload", VCPU_STAT(efer_reload) },
179 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
180 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
181 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
182 	{ "irq_injections", VCPU_STAT(irq_injections) },
183 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
184 	{ "req_event", VCPU_STAT(req_event) },
185 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
186 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
187 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
188 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
189 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
190 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
191 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
192 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
193 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
194 	{ "largepages", VM_STAT(lpages) },
195 	{ "max_mmu_page_hash_collisions",
196 		VM_STAT(max_mmu_page_hash_collisions) },
197 	{ NULL }
198 };
199 
200 u64 __read_mostly host_xcr0;
201 
202 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
203 
204 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
205 {
206 	int i;
207 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
208 		vcpu->arch.apf.gfns[i] = ~0;
209 }
210 
211 static void kvm_on_user_return(struct user_return_notifier *urn)
212 {
213 	unsigned slot;
214 	struct kvm_shared_msrs *locals
215 		= container_of(urn, struct kvm_shared_msrs, urn);
216 	struct kvm_shared_msr_values *values;
217 	unsigned long flags;
218 
219 	/*
220 	 * Disabling irqs at this point since the following code could be
221 	 * interrupted and executed through kvm_arch_hardware_disable()
222 	 */
223 	local_irq_save(flags);
224 	if (locals->registered) {
225 		locals->registered = false;
226 		user_return_notifier_unregister(urn);
227 	}
228 	local_irq_restore(flags);
229 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
230 		values = &locals->values[slot];
231 		if (values->host != values->curr) {
232 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
233 			values->curr = values->host;
234 		}
235 	}
236 }
237 
238 static void shared_msr_update(unsigned slot, u32 msr)
239 {
240 	u64 value;
241 	unsigned int cpu = smp_processor_id();
242 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
243 
244 	/* only read, and nobody should modify it at this time,
245 	 * so don't need lock */
246 	if (slot >= shared_msrs_global.nr) {
247 		printk(KERN_ERR "kvm: invalid MSR slot!");
248 		return;
249 	}
250 	rdmsrl_safe(msr, &value);
251 	smsr->values[slot].host = value;
252 	smsr->values[slot].curr = value;
253 }
254 
255 void kvm_define_shared_msr(unsigned slot, u32 msr)
256 {
257 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
258 	shared_msrs_global.msrs[slot] = msr;
259 	if (slot >= shared_msrs_global.nr)
260 		shared_msrs_global.nr = slot + 1;
261 }
262 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
263 
264 static void kvm_shared_msr_cpu_online(void)
265 {
266 	unsigned i;
267 
268 	for (i = 0; i < shared_msrs_global.nr; ++i)
269 		shared_msr_update(i, shared_msrs_global.msrs[i]);
270 }
271 
272 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
273 {
274 	unsigned int cpu = smp_processor_id();
275 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
276 	int err;
277 
278 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
279 		return 0;
280 	smsr->values[slot].curr = value;
281 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
282 	if (err)
283 		return 1;
284 
285 	if (!smsr->registered) {
286 		smsr->urn.on_user_return = kvm_on_user_return;
287 		user_return_notifier_register(&smsr->urn);
288 		smsr->registered = true;
289 	}
290 	return 0;
291 }
292 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
293 
294 static void drop_user_return_notifiers(void)
295 {
296 	unsigned int cpu = smp_processor_id();
297 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
298 
299 	if (smsr->registered)
300 		kvm_on_user_return(&smsr->urn);
301 }
302 
303 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
304 {
305 	return vcpu->arch.apic_base;
306 }
307 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
308 
309 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
310 {
311 	u64 old_state = vcpu->arch.apic_base &
312 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
313 	u64 new_state = msr_info->data &
314 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
316 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
317 
318 	if (!msr_info->host_initiated &&
319 	    ((msr_info->data & reserved_bits) != 0 ||
320 	     new_state == X2APIC_ENABLE ||
321 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
322 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
323 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
324 	      old_state == 0)))
325 		return 1;
326 
327 	kvm_lapic_set_base(vcpu, msr_info->data);
328 	return 0;
329 }
330 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
331 
332 asmlinkage __visible void kvm_spurious_fault(void)
333 {
334 	/* Fault while not rebooting.  We want the trace. */
335 	BUG();
336 }
337 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
338 
339 #define EXCPT_BENIGN		0
340 #define EXCPT_CONTRIBUTORY	1
341 #define EXCPT_PF		2
342 
343 static int exception_class(int vector)
344 {
345 	switch (vector) {
346 	case PF_VECTOR:
347 		return EXCPT_PF;
348 	case DE_VECTOR:
349 	case TS_VECTOR:
350 	case NP_VECTOR:
351 	case SS_VECTOR:
352 	case GP_VECTOR:
353 		return EXCPT_CONTRIBUTORY;
354 	default:
355 		break;
356 	}
357 	return EXCPT_BENIGN;
358 }
359 
360 #define EXCPT_FAULT		0
361 #define EXCPT_TRAP		1
362 #define EXCPT_ABORT		2
363 #define EXCPT_INTERRUPT		3
364 
365 static int exception_type(int vector)
366 {
367 	unsigned int mask;
368 
369 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
370 		return EXCPT_INTERRUPT;
371 
372 	mask = 1 << vector;
373 
374 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
375 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
376 		return EXCPT_TRAP;
377 
378 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
379 		return EXCPT_ABORT;
380 
381 	/* Reserved exceptions will result in fault */
382 	return EXCPT_FAULT;
383 }
384 
385 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
386 		unsigned nr, bool has_error, u32 error_code,
387 		bool reinject)
388 {
389 	u32 prev_nr;
390 	int class1, class2;
391 
392 	kvm_make_request(KVM_REQ_EVENT, vcpu);
393 
394 	if (!vcpu->arch.exception.pending) {
395 	queue:
396 		if (has_error && !is_protmode(vcpu))
397 			has_error = false;
398 		vcpu->arch.exception.pending = true;
399 		vcpu->arch.exception.has_error_code = has_error;
400 		vcpu->arch.exception.nr = nr;
401 		vcpu->arch.exception.error_code = error_code;
402 		vcpu->arch.exception.reinject = reinject;
403 		return;
404 	}
405 
406 	/* to check exception */
407 	prev_nr = vcpu->arch.exception.nr;
408 	if (prev_nr == DF_VECTOR) {
409 		/* triple fault -> shutdown */
410 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
411 		return;
412 	}
413 	class1 = exception_class(prev_nr);
414 	class2 = exception_class(nr);
415 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
416 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
417 		/* generate double fault per SDM Table 5-5 */
418 		vcpu->arch.exception.pending = true;
419 		vcpu->arch.exception.has_error_code = true;
420 		vcpu->arch.exception.nr = DF_VECTOR;
421 		vcpu->arch.exception.error_code = 0;
422 	} else
423 		/* replace previous exception with a new one in a hope
424 		   that instruction re-execution will regenerate lost
425 		   exception */
426 		goto queue;
427 }
428 
429 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
430 {
431 	kvm_multiple_exception(vcpu, nr, false, 0, false);
432 }
433 EXPORT_SYMBOL_GPL(kvm_queue_exception);
434 
435 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
436 {
437 	kvm_multiple_exception(vcpu, nr, false, 0, true);
438 }
439 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
440 
441 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
442 {
443 	if (err)
444 		kvm_inject_gp(vcpu, 0);
445 	else
446 		return kvm_skip_emulated_instruction(vcpu);
447 
448 	return 1;
449 }
450 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
451 
452 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
453 {
454 	++vcpu->stat.pf_guest;
455 	vcpu->arch.cr2 = fault->address;
456 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
457 }
458 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
459 
460 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
461 {
462 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
463 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
464 	else
465 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
466 
467 	return fault->nested_page_fault;
468 }
469 
470 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
471 {
472 	atomic_inc(&vcpu->arch.nmi_queued);
473 	kvm_make_request(KVM_REQ_NMI, vcpu);
474 }
475 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
476 
477 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
478 {
479 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
480 }
481 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
482 
483 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
484 {
485 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
486 }
487 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
488 
489 /*
490  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
491  * a #GP and return false.
492  */
493 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
494 {
495 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
496 		return true;
497 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
498 	return false;
499 }
500 EXPORT_SYMBOL_GPL(kvm_require_cpl);
501 
502 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
503 {
504 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
505 		return true;
506 
507 	kvm_queue_exception(vcpu, UD_VECTOR);
508 	return false;
509 }
510 EXPORT_SYMBOL_GPL(kvm_require_dr);
511 
512 /*
513  * This function will be used to read from the physical memory of the currently
514  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
515  * can read from guest physical or from the guest's guest physical memory.
516  */
517 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
518 			    gfn_t ngfn, void *data, int offset, int len,
519 			    u32 access)
520 {
521 	struct x86_exception exception;
522 	gfn_t real_gfn;
523 	gpa_t ngpa;
524 
525 	ngpa     = gfn_to_gpa(ngfn);
526 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
527 	if (real_gfn == UNMAPPED_GVA)
528 		return -EFAULT;
529 
530 	real_gfn = gpa_to_gfn(real_gfn);
531 
532 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
533 }
534 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
535 
536 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
537 			       void *data, int offset, int len, u32 access)
538 {
539 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
540 				       data, offset, len, access);
541 }
542 
543 /*
544  * Load the pae pdptrs.  Return true is they are all valid.
545  */
546 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
547 {
548 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
549 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
550 	int i;
551 	int ret;
552 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
553 
554 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
555 				      offset * sizeof(u64), sizeof(pdpte),
556 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
557 	if (ret < 0) {
558 		ret = 0;
559 		goto out;
560 	}
561 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
562 		if ((pdpte[i] & PT_PRESENT_MASK) &&
563 		    (pdpte[i] &
564 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
565 			ret = 0;
566 			goto out;
567 		}
568 	}
569 	ret = 1;
570 
571 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
572 	__set_bit(VCPU_EXREG_PDPTR,
573 		  (unsigned long *)&vcpu->arch.regs_avail);
574 	__set_bit(VCPU_EXREG_PDPTR,
575 		  (unsigned long *)&vcpu->arch.regs_dirty);
576 out:
577 
578 	return ret;
579 }
580 EXPORT_SYMBOL_GPL(load_pdptrs);
581 
582 bool pdptrs_changed(struct kvm_vcpu *vcpu)
583 {
584 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
585 	bool changed = true;
586 	int offset;
587 	gfn_t gfn;
588 	int r;
589 
590 	if (is_long_mode(vcpu) || !is_pae(vcpu))
591 		return false;
592 
593 	if (!test_bit(VCPU_EXREG_PDPTR,
594 		      (unsigned long *)&vcpu->arch.regs_avail))
595 		return true;
596 
597 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
598 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
599 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
600 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
601 	if (r < 0)
602 		goto out;
603 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
604 out:
605 
606 	return changed;
607 }
608 EXPORT_SYMBOL_GPL(pdptrs_changed);
609 
610 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
611 {
612 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
613 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
614 
615 	cr0 |= X86_CR0_ET;
616 
617 #ifdef CONFIG_X86_64
618 	if (cr0 & 0xffffffff00000000UL)
619 		return 1;
620 #endif
621 
622 	cr0 &= ~CR0_RESERVED_BITS;
623 
624 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
625 		return 1;
626 
627 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
628 		return 1;
629 
630 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
631 #ifdef CONFIG_X86_64
632 		if ((vcpu->arch.efer & EFER_LME)) {
633 			int cs_db, cs_l;
634 
635 			if (!is_pae(vcpu))
636 				return 1;
637 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
638 			if (cs_l)
639 				return 1;
640 		} else
641 #endif
642 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
643 						 kvm_read_cr3(vcpu)))
644 			return 1;
645 	}
646 
647 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
648 		return 1;
649 
650 	kvm_x86_ops->set_cr0(vcpu, cr0);
651 
652 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
653 		kvm_clear_async_pf_completion_queue(vcpu);
654 		kvm_async_pf_hash_reset(vcpu);
655 	}
656 
657 	if ((cr0 ^ old_cr0) & update_bits)
658 		kvm_mmu_reset_context(vcpu);
659 
660 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
661 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
662 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
663 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
664 
665 	return 0;
666 }
667 EXPORT_SYMBOL_GPL(kvm_set_cr0);
668 
669 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
670 {
671 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
672 }
673 EXPORT_SYMBOL_GPL(kvm_lmsw);
674 
675 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
676 {
677 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
678 			!vcpu->guest_xcr0_loaded) {
679 		/* kvm_set_xcr() also depends on this */
680 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
681 		vcpu->guest_xcr0_loaded = 1;
682 	}
683 }
684 
685 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
686 {
687 	if (vcpu->guest_xcr0_loaded) {
688 		if (vcpu->arch.xcr0 != host_xcr0)
689 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
690 		vcpu->guest_xcr0_loaded = 0;
691 	}
692 }
693 
694 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
695 {
696 	u64 xcr0 = xcr;
697 	u64 old_xcr0 = vcpu->arch.xcr0;
698 	u64 valid_bits;
699 
700 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
701 	if (index != XCR_XFEATURE_ENABLED_MASK)
702 		return 1;
703 	if (!(xcr0 & XFEATURE_MASK_FP))
704 		return 1;
705 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
706 		return 1;
707 
708 	/*
709 	 * Do not allow the guest to set bits that we do not support
710 	 * saving.  However, xcr0 bit 0 is always set, even if the
711 	 * emulated CPU does not support XSAVE (see fx_init).
712 	 */
713 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
714 	if (xcr0 & ~valid_bits)
715 		return 1;
716 
717 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
718 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
719 		return 1;
720 
721 	if (xcr0 & XFEATURE_MASK_AVX512) {
722 		if (!(xcr0 & XFEATURE_MASK_YMM))
723 			return 1;
724 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
725 			return 1;
726 	}
727 	vcpu->arch.xcr0 = xcr0;
728 
729 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
730 		kvm_update_cpuid(vcpu);
731 	return 0;
732 }
733 
734 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
735 {
736 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
737 	    __kvm_set_xcr(vcpu, index, xcr)) {
738 		kvm_inject_gp(vcpu, 0);
739 		return 1;
740 	}
741 	return 0;
742 }
743 EXPORT_SYMBOL_GPL(kvm_set_xcr);
744 
745 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
746 {
747 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
748 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
749 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
750 
751 	if (cr4 & CR4_RESERVED_BITS)
752 		return 1;
753 
754 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
755 		return 1;
756 
757 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
758 		return 1;
759 
760 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
761 		return 1;
762 
763 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
764 		return 1;
765 
766 	if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
767 		return 1;
768 
769 	if (is_long_mode(vcpu)) {
770 		if (!(cr4 & X86_CR4_PAE))
771 			return 1;
772 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
773 		   && ((cr4 ^ old_cr4) & pdptr_bits)
774 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
775 				   kvm_read_cr3(vcpu)))
776 		return 1;
777 
778 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
779 		if (!guest_cpuid_has_pcid(vcpu))
780 			return 1;
781 
782 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
783 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
784 			return 1;
785 	}
786 
787 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
788 		return 1;
789 
790 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
791 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
792 		kvm_mmu_reset_context(vcpu);
793 
794 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
795 		kvm_update_cpuid(vcpu);
796 
797 	return 0;
798 }
799 EXPORT_SYMBOL_GPL(kvm_set_cr4);
800 
801 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
802 {
803 #ifdef CONFIG_X86_64
804 	cr3 &= ~CR3_PCID_INVD;
805 #endif
806 
807 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
808 		kvm_mmu_sync_roots(vcpu);
809 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
810 		return 0;
811 	}
812 
813 	if (is_long_mode(vcpu)) {
814 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
815 			return 1;
816 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
817 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
818 		return 1;
819 
820 	vcpu->arch.cr3 = cr3;
821 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
822 	kvm_mmu_new_cr3(vcpu);
823 	return 0;
824 }
825 EXPORT_SYMBOL_GPL(kvm_set_cr3);
826 
827 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
828 {
829 	if (cr8 & CR8_RESERVED_BITS)
830 		return 1;
831 	if (lapic_in_kernel(vcpu))
832 		kvm_lapic_set_tpr(vcpu, cr8);
833 	else
834 		vcpu->arch.cr8 = cr8;
835 	return 0;
836 }
837 EXPORT_SYMBOL_GPL(kvm_set_cr8);
838 
839 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
840 {
841 	if (lapic_in_kernel(vcpu))
842 		return kvm_lapic_get_cr8(vcpu);
843 	else
844 		return vcpu->arch.cr8;
845 }
846 EXPORT_SYMBOL_GPL(kvm_get_cr8);
847 
848 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
849 {
850 	int i;
851 
852 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
853 		for (i = 0; i < KVM_NR_DB_REGS; i++)
854 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
855 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
856 	}
857 }
858 
859 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
860 {
861 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
862 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
863 }
864 
865 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
866 {
867 	unsigned long dr7;
868 
869 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
870 		dr7 = vcpu->arch.guest_debug_dr7;
871 	else
872 		dr7 = vcpu->arch.dr7;
873 	kvm_x86_ops->set_dr7(vcpu, dr7);
874 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
875 	if (dr7 & DR7_BP_EN_MASK)
876 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
877 }
878 
879 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
880 {
881 	u64 fixed = DR6_FIXED_1;
882 
883 	if (!guest_cpuid_has_rtm(vcpu))
884 		fixed |= DR6_RTM;
885 	return fixed;
886 }
887 
888 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
889 {
890 	switch (dr) {
891 	case 0 ... 3:
892 		vcpu->arch.db[dr] = val;
893 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
894 			vcpu->arch.eff_db[dr] = val;
895 		break;
896 	case 4:
897 		/* fall through */
898 	case 6:
899 		if (val & 0xffffffff00000000ULL)
900 			return -1; /* #GP */
901 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
902 		kvm_update_dr6(vcpu);
903 		break;
904 	case 5:
905 		/* fall through */
906 	default: /* 7 */
907 		if (val & 0xffffffff00000000ULL)
908 			return -1; /* #GP */
909 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
910 		kvm_update_dr7(vcpu);
911 		break;
912 	}
913 
914 	return 0;
915 }
916 
917 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
918 {
919 	if (__kvm_set_dr(vcpu, dr, val)) {
920 		kvm_inject_gp(vcpu, 0);
921 		return 1;
922 	}
923 	return 0;
924 }
925 EXPORT_SYMBOL_GPL(kvm_set_dr);
926 
927 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
928 {
929 	switch (dr) {
930 	case 0 ... 3:
931 		*val = vcpu->arch.db[dr];
932 		break;
933 	case 4:
934 		/* fall through */
935 	case 6:
936 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
937 			*val = vcpu->arch.dr6;
938 		else
939 			*val = kvm_x86_ops->get_dr6(vcpu);
940 		break;
941 	case 5:
942 		/* fall through */
943 	default: /* 7 */
944 		*val = vcpu->arch.dr7;
945 		break;
946 	}
947 	return 0;
948 }
949 EXPORT_SYMBOL_GPL(kvm_get_dr);
950 
951 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
952 {
953 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
954 	u64 data;
955 	int err;
956 
957 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
958 	if (err)
959 		return err;
960 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
961 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
962 	return err;
963 }
964 EXPORT_SYMBOL_GPL(kvm_rdpmc);
965 
966 /*
967  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
968  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
969  *
970  * This list is modified at module load time to reflect the
971  * capabilities of the host cpu. This capabilities test skips MSRs that are
972  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
973  * may depend on host virtualization features rather than host cpu features.
974  */
975 
976 static u32 msrs_to_save[] = {
977 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
978 	MSR_STAR,
979 #ifdef CONFIG_X86_64
980 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
981 #endif
982 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
983 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
984 };
985 
986 static unsigned num_msrs_to_save;
987 
988 static u32 emulated_msrs[] = {
989 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
990 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
991 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
992 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
993 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
994 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
995 	HV_X64_MSR_RESET,
996 	HV_X64_MSR_VP_INDEX,
997 	HV_X64_MSR_VP_RUNTIME,
998 	HV_X64_MSR_SCONTROL,
999 	HV_X64_MSR_STIMER0_CONFIG,
1000 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1001 	MSR_KVM_PV_EOI_EN,
1002 
1003 	MSR_IA32_TSC_ADJUST,
1004 	MSR_IA32_TSCDEADLINE,
1005 	MSR_IA32_MISC_ENABLE,
1006 	MSR_IA32_MCG_STATUS,
1007 	MSR_IA32_MCG_CTL,
1008 	MSR_IA32_MCG_EXT_CTL,
1009 	MSR_IA32_SMBASE,
1010 	MSR_PLATFORM_INFO,
1011 	MSR_MISC_FEATURES_ENABLES,
1012 };
1013 
1014 static unsigned num_emulated_msrs;
1015 
1016 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1017 {
1018 	if (efer & efer_reserved_bits)
1019 		return false;
1020 
1021 	if (efer & EFER_FFXSR) {
1022 		struct kvm_cpuid_entry2 *feat;
1023 
1024 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1025 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1026 			return false;
1027 	}
1028 
1029 	if (efer & EFER_SVME) {
1030 		struct kvm_cpuid_entry2 *feat;
1031 
1032 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1033 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1034 			return false;
1035 	}
1036 
1037 	return true;
1038 }
1039 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1040 
1041 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1042 {
1043 	u64 old_efer = vcpu->arch.efer;
1044 
1045 	if (!kvm_valid_efer(vcpu, efer))
1046 		return 1;
1047 
1048 	if (is_paging(vcpu)
1049 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1050 		return 1;
1051 
1052 	efer &= ~EFER_LMA;
1053 	efer |= vcpu->arch.efer & EFER_LMA;
1054 
1055 	kvm_x86_ops->set_efer(vcpu, efer);
1056 
1057 	/* Update reserved bits */
1058 	if ((efer ^ old_efer) & EFER_NX)
1059 		kvm_mmu_reset_context(vcpu);
1060 
1061 	return 0;
1062 }
1063 
1064 void kvm_enable_efer_bits(u64 mask)
1065 {
1066        efer_reserved_bits &= ~mask;
1067 }
1068 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1069 
1070 /*
1071  * Writes msr value into into the appropriate "register".
1072  * Returns 0 on success, non-0 otherwise.
1073  * Assumes vcpu_load() was already called.
1074  */
1075 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1076 {
1077 	switch (msr->index) {
1078 	case MSR_FS_BASE:
1079 	case MSR_GS_BASE:
1080 	case MSR_KERNEL_GS_BASE:
1081 	case MSR_CSTAR:
1082 	case MSR_LSTAR:
1083 		if (is_noncanonical_address(msr->data))
1084 			return 1;
1085 		break;
1086 	case MSR_IA32_SYSENTER_EIP:
1087 	case MSR_IA32_SYSENTER_ESP:
1088 		/*
1089 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1090 		 * non-canonical address is written on Intel but not on
1091 		 * AMD (which ignores the top 32-bits, because it does
1092 		 * not implement 64-bit SYSENTER).
1093 		 *
1094 		 * 64-bit code should hence be able to write a non-canonical
1095 		 * value on AMD.  Making the address canonical ensures that
1096 		 * vmentry does not fail on Intel after writing a non-canonical
1097 		 * value, and that something deterministic happens if the guest
1098 		 * invokes 64-bit SYSENTER.
1099 		 */
1100 		msr->data = get_canonical(msr->data);
1101 	}
1102 	return kvm_x86_ops->set_msr(vcpu, msr);
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_set_msr);
1105 
1106 /*
1107  * Adapt set_msr() to msr_io()'s calling convention
1108  */
1109 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1110 {
1111 	struct msr_data msr;
1112 	int r;
1113 
1114 	msr.index = index;
1115 	msr.host_initiated = true;
1116 	r = kvm_get_msr(vcpu, &msr);
1117 	if (r)
1118 		return r;
1119 
1120 	*data = msr.data;
1121 	return 0;
1122 }
1123 
1124 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1125 {
1126 	struct msr_data msr;
1127 
1128 	msr.data = *data;
1129 	msr.index = index;
1130 	msr.host_initiated = true;
1131 	return kvm_set_msr(vcpu, &msr);
1132 }
1133 
1134 #ifdef CONFIG_X86_64
1135 struct pvclock_gtod_data {
1136 	seqcount_t	seq;
1137 
1138 	struct { /* extract of a clocksource struct */
1139 		int vclock_mode;
1140 		u64	cycle_last;
1141 		u64	mask;
1142 		u32	mult;
1143 		u32	shift;
1144 	} clock;
1145 
1146 	u64		boot_ns;
1147 	u64		nsec_base;
1148 	u64		wall_time_sec;
1149 };
1150 
1151 static struct pvclock_gtod_data pvclock_gtod_data;
1152 
1153 static void update_pvclock_gtod(struct timekeeper *tk)
1154 {
1155 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1156 	u64 boot_ns;
1157 
1158 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1159 
1160 	write_seqcount_begin(&vdata->seq);
1161 
1162 	/* copy pvclock gtod data */
1163 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1164 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1165 	vdata->clock.mask		= tk->tkr_mono.mask;
1166 	vdata->clock.mult		= tk->tkr_mono.mult;
1167 	vdata->clock.shift		= tk->tkr_mono.shift;
1168 
1169 	vdata->boot_ns			= boot_ns;
1170 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1171 
1172 	vdata->wall_time_sec            = tk->xtime_sec;
1173 
1174 	write_seqcount_end(&vdata->seq);
1175 }
1176 #endif
1177 
1178 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1179 {
1180 	/*
1181 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1182 	 * vcpu_enter_guest.  This function is only called from
1183 	 * the physical CPU that is running vcpu.
1184 	 */
1185 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1186 }
1187 
1188 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1189 {
1190 	int version;
1191 	int r;
1192 	struct pvclock_wall_clock wc;
1193 	struct timespec64 boot;
1194 
1195 	if (!wall_clock)
1196 		return;
1197 
1198 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1199 	if (r)
1200 		return;
1201 
1202 	if (version & 1)
1203 		++version;  /* first time write, random junk */
1204 
1205 	++version;
1206 
1207 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1208 		return;
1209 
1210 	/*
1211 	 * The guest calculates current wall clock time by adding
1212 	 * system time (updated by kvm_guest_time_update below) to the
1213 	 * wall clock specified here.  guest system time equals host
1214 	 * system time for us, thus we must fill in host boot time here.
1215 	 */
1216 	getboottime64(&boot);
1217 
1218 	if (kvm->arch.kvmclock_offset) {
1219 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1220 		boot = timespec64_sub(boot, ts);
1221 	}
1222 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1223 	wc.nsec = boot.tv_nsec;
1224 	wc.version = version;
1225 
1226 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1227 
1228 	version++;
1229 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1230 }
1231 
1232 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1233 {
1234 	do_shl32_div32(dividend, divisor);
1235 	return dividend;
1236 }
1237 
1238 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1239 			       s8 *pshift, u32 *pmultiplier)
1240 {
1241 	uint64_t scaled64;
1242 	int32_t  shift = 0;
1243 	uint64_t tps64;
1244 	uint32_t tps32;
1245 
1246 	tps64 = base_hz;
1247 	scaled64 = scaled_hz;
1248 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1249 		tps64 >>= 1;
1250 		shift--;
1251 	}
1252 
1253 	tps32 = (uint32_t)tps64;
1254 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1255 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1256 			scaled64 >>= 1;
1257 		else
1258 			tps32 <<= 1;
1259 		shift++;
1260 	}
1261 
1262 	*pshift = shift;
1263 	*pmultiplier = div_frac(scaled64, tps32);
1264 
1265 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1266 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1267 }
1268 
1269 #ifdef CONFIG_X86_64
1270 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1271 #endif
1272 
1273 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1274 static unsigned long max_tsc_khz;
1275 
1276 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1277 {
1278 	u64 v = (u64)khz * (1000000 + ppm);
1279 	do_div(v, 1000000);
1280 	return v;
1281 }
1282 
1283 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1284 {
1285 	u64 ratio;
1286 
1287 	/* Guest TSC same frequency as host TSC? */
1288 	if (!scale) {
1289 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1290 		return 0;
1291 	}
1292 
1293 	/* TSC scaling supported? */
1294 	if (!kvm_has_tsc_control) {
1295 		if (user_tsc_khz > tsc_khz) {
1296 			vcpu->arch.tsc_catchup = 1;
1297 			vcpu->arch.tsc_always_catchup = 1;
1298 			return 0;
1299 		} else {
1300 			WARN(1, "user requested TSC rate below hardware speed\n");
1301 			return -1;
1302 		}
1303 	}
1304 
1305 	/* TSC scaling required  - calculate ratio */
1306 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1307 				user_tsc_khz, tsc_khz);
1308 
1309 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1310 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1311 			  user_tsc_khz);
1312 		return -1;
1313 	}
1314 
1315 	vcpu->arch.tsc_scaling_ratio = ratio;
1316 	return 0;
1317 }
1318 
1319 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1320 {
1321 	u32 thresh_lo, thresh_hi;
1322 	int use_scaling = 0;
1323 
1324 	/* tsc_khz can be zero if TSC calibration fails */
1325 	if (user_tsc_khz == 0) {
1326 		/* set tsc_scaling_ratio to a safe value */
1327 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1328 		return -1;
1329 	}
1330 
1331 	/* Compute a scale to convert nanoseconds in TSC cycles */
1332 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1333 			   &vcpu->arch.virtual_tsc_shift,
1334 			   &vcpu->arch.virtual_tsc_mult);
1335 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1336 
1337 	/*
1338 	 * Compute the variation in TSC rate which is acceptable
1339 	 * within the range of tolerance and decide if the
1340 	 * rate being applied is within that bounds of the hardware
1341 	 * rate.  If so, no scaling or compensation need be done.
1342 	 */
1343 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1344 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1345 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1346 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1347 		use_scaling = 1;
1348 	}
1349 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1350 }
1351 
1352 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1353 {
1354 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1355 				      vcpu->arch.virtual_tsc_mult,
1356 				      vcpu->arch.virtual_tsc_shift);
1357 	tsc += vcpu->arch.this_tsc_write;
1358 	return tsc;
1359 }
1360 
1361 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1362 {
1363 #ifdef CONFIG_X86_64
1364 	bool vcpus_matched;
1365 	struct kvm_arch *ka = &vcpu->kvm->arch;
1366 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1367 
1368 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1369 			 atomic_read(&vcpu->kvm->online_vcpus));
1370 
1371 	/*
1372 	 * Once the masterclock is enabled, always perform request in
1373 	 * order to update it.
1374 	 *
1375 	 * In order to enable masterclock, the host clocksource must be TSC
1376 	 * and the vcpus need to have matched TSCs.  When that happens,
1377 	 * perform request to enable masterclock.
1378 	 */
1379 	if (ka->use_master_clock ||
1380 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1381 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1382 
1383 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1384 			    atomic_read(&vcpu->kvm->online_vcpus),
1385 		            ka->use_master_clock, gtod->clock.vclock_mode);
1386 #endif
1387 }
1388 
1389 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1390 {
1391 	u64 curr_offset = vcpu->arch.tsc_offset;
1392 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1393 }
1394 
1395 /*
1396  * Multiply tsc by a fixed point number represented by ratio.
1397  *
1398  * The most significant 64-N bits (mult) of ratio represent the
1399  * integral part of the fixed point number; the remaining N bits
1400  * (frac) represent the fractional part, ie. ratio represents a fixed
1401  * point number (mult + frac * 2^(-N)).
1402  *
1403  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1404  */
1405 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1406 {
1407 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1408 }
1409 
1410 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1411 {
1412 	u64 _tsc = tsc;
1413 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1414 
1415 	if (ratio != kvm_default_tsc_scaling_ratio)
1416 		_tsc = __scale_tsc(ratio, tsc);
1417 
1418 	return _tsc;
1419 }
1420 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1421 
1422 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1423 {
1424 	u64 tsc;
1425 
1426 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1427 
1428 	return target_tsc - tsc;
1429 }
1430 
1431 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1432 {
1433 	return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1436 
1437 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1438 {
1439 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1440 	vcpu->arch.tsc_offset = offset;
1441 }
1442 
1443 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1444 {
1445 	struct kvm *kvm = vcpu->kvm;
1446 	u64 offset, ns, elapsed;
1447 	unsigned long flags;
1448 	bool matched;
1449 	bool already_matched;
1450 	u64 data = msr->data;
1451 	bool synchronizing = false;
1452 
1453 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1454 	offset = kvm_compute_tsc_offset(vcpu, data);
1455 	ns = ktime_get_boot_ns();
1456 	elapsed = ns - kvm->arch.last_tsc_nsec;
1457 
1458 	if (vcpu->arch.virtual_tsc_khz) {
1459 		if (data == 0 && msr->host_initiated) {
1460 			/*
1461 			 * detection of vcpu initialization -- need to sync
1462 			 * with other vCPUs. This particularly helps to keep
1463 			 * kvm_clock stable after CPU hotplug
1464 			 */
1465 			synchronizing = true;
1466 		} else {
1467 			u64 tsc_exp = kvm->arch.last_tsc_write +
1468 						nsec_to_cycles(vcpu, elapsed);
1469 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1470 			/*
1471 			 * Special case: TSC write with a small delta (1 second)
1472 			 * of virtual cycle time against real time is
1473 			 * interpreted as an attempt to synchronize the CPU.
1474 			 */
1475 			synchronizing = data < tsc_exp + tsc_hz &&
1476 					data + tsc_hz > tsc_exp;
1477 		}
1478 	}
1479 
1480 	/*
1481 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1482 	 * TSC, we add elapsed time in this computation.  We could let the
1483 	 * compensation code attempt to catch up if we fall behind, but
1484 	 * it's better to try to match offsets from the beginning.
1485          */
1486 	if (synchronizing &&
1487 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1488 		if (!check_tsc_unstable()) {
1489 			offset = kvm->arch.cur_tsc_offset;
1490 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1491 		} else {
1492 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1493 			data += delta;
1494 			offset = kvm_compute_tsc_offset(vcpu, data);
1495 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1496 		}
1497 		matched = true;
1498 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1499 	} else {
1500 		/*
1501 		 * We split periods of matched TSC writes into generations.
1502 		 * For each generation, we track the original measured
1503 		 * nanosecond time, offset, and write, so if TSCs are in
1504 		 * sync, we can match exact offset, and if not, we can match
1505 		 * exact software computation in compute_guest_tsc()
1506 		 *
1507 		 * These values are tracked in kvm->arch.cur_xxx variables.
1508 		 */
1509 		kvm->arch.cur_tsc_generation++;
1510 		kvm->arch.cur_tsc_nsec = ns;
1511 		kvm->arch.cur_tsc_write = data;
1512 		kvm->arch.cur_tsc_offset = offset;
1513 		matched = false;
1514 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1515 			 kvm->arch.cur_tsc_generation, data);
1516 	}
1517 
1518 	/*
1519 	 * We also track th most recent recorded KHZ, write and time to
1520 	 * allow the matching interval to be extended at each write.
1521 	 */
1522 	kvm->arch.last_tsc_nsec = ns;
1523 	kvm->arch.last_tsc_write = data;
1524 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1525 
1526 	vcpu->arch.last_guest_tsc = data;
1527 
1528 	/* Keep track of which generation this VCPU has synchronized to */
1529 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1530 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1531 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1532 
1533 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1534 		update_ia32_tsc_adjust_msr(vcpu, offset);
1535 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1536 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1537 
1538 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1539 	if (!matched) {
1540 		kvm->arch.nr_vcpus_matched_tsc = 0;
1541 	} else if (!already_matched) {
1542 		kvm->arch.nr_vcpus_matched_tsc++;
1543 	}
1544 
1545 	kvm_track_tsc_matching(vcpu);
1546 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1547 }
1548 
1549 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1550 
1551 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1552 					   s64 adjustment)
1553 {
1554 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1555 }
1556 
1557 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1558 {
1559 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1560 		WARN_ON(adjustment < 0);
1561 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1562 	adjust_tsc_offset_guest(vcpu, adjustment);
1563 }
1564 
1565 #ifdef CONFIG_X86_64
1566 
1567 static u64 read_tsc(void)
1568 {
1569 	u64 ret = (u64)rdtsc_ordered();
1570 	u64 last = pvclock_gtod_data.clock.cycle_last;
1571 
1572 	if (likely(ret >= last))
1573 		return ret;
1574 
1575 	/*
1576 	 * GCC likes to generate cmov here, but this branch is extremely
1577 	 * predictable (it's just a function of time and the likely is
1578 	 * very likely) and there's a data dependence, so force GCC
1579 	 * to generate a branch instead.  I don't barrier() because
1580 	 * we don't actually need a barrier, and if this function
1581 	 * ever gets inlined it will generate worse code.
1582 	 */
1583 	asm volatile ("");
1584 	return last;
1585 }
1586 
1587 static inline u64 vgettsc(u64 *cycle_now)
1588 {
1589 	long v;
1590 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1591 
1592 	*cycle_now = read_tsc();
1593 
1594 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1595 	return v * gtod->clock.mult;
1596 }
1597 
1598 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1599 {
1600 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1601 	unsigned long seq;
1602 	int mode;
1603 	u64 ns;
1604 
1605 	do {
1606 		seq = read_seqcount_begin(&gtod->seq);
1607 		mode = gtod->clock.vclock_mode;
1608 		ns = gtod->nsec_base;
1609 		ns += vgettsc(cycle_now);
1610 		ns >>= gtod->clock.shift;
1611 		ns += gtod->boot_ns;
1612 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1613 	*t = ns;
1614 
1615 	return mode;
1616 }
1617 
1618 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1619 {
1620 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1621 	unsigned long seq;
1622 	int mode;
1623 	u64 ns;
1624 
1625 	do {
1626 		seq = read_seqcount_begin(&gtod->seq);
1627 		mode = gtod->clock.vclock_mode;
1628 		ts->tv_sec = gtod->wall_time_sec;
1629 		ns = gtod->nsec_base;
1630 		ns += vgettsc(cycle_now);
1631 		ns >>= gtod->clock.shift;
1632 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1633 
1634 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1635 	ts->tv_nsec = ns;
1636 
1637 	return mode;
1638 }
1639 
1640 /* returns true if host is using tsc clocksource */
1641 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1642 {
1643 	/* checked again under seqlock below */
1644 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1645 		return false;
1646 
1647 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1648 }
1649 
1650 /* returns true if host is using tsc clocksource */
1651 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1652 					   u64 *cycle_now)
1653 {
1654 	/* checked again under seqlock below */
1655 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1656 		return false;
1657 
1658 	return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1659 }
1660 #endif
1661 
1662 /*
1663  *
1664  * Assuming a stable TSC across physical CPUS, and a stable TSC
1665  * across virtual CPUs, the following condition is possible.
1666  * Each numbered line represents an event visible to both
1667  * CPUs at the next numbered event.
1668  *
1669  * "timespecX" represents host monotonic time. "tscX" represents
1670  * RDTSC value.
1671  *
1672  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1673  *
1674  * 1.  read timespec0,tsc0
1675  * 2.					| timespec1 = timespec0 + N
1676  * 					| tsc1 = tsc0 + M
1677  * 3. transition to guest		| transition to guest
1678  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1679  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1680  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1681  *
1682  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1683  *
1684  * 	- ret0 < ret1
1685  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1686  *		...
1687  *	- 0 < N - M => M < N
1688  *
1689  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1690  * always the case (the difference between two distinct xtime instances
1691  * might be smaller then the difference between corresponding TSC reads,
1692  * when updating guest vcpus pvclock areas).
1693  *
1694  * To avoid that problem, do not allow visibility of distinct
1695  * system_timestamp/tsc_timestamp values simultaneously: use a master
1696  * copy of host monotonic time values. Update that master copy
1697  * in lockstep.
1698  *
1699  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1700  *
1701  */
1702 
1703 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1704 {
1705 #ifdef CONFIG_X86_64
1706 	struct kvm_arch *ka = &kvm->arch;
1707 	int vclock_mode;
1708 	bool host_tsc_clocksource, vcpus_matched;
1709 
1710 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1711 			atomic_read(&kvm->online_vcpus));
1712 
1713 	/*
1714 	 * If the host uses TSC clock, then passthrough TSC as stable
1715 	 * to the guest.
1716 	 */
1717 	host_tsc_clocksource = kvm_get_time_and_clockread(
1718 					&ka->master_kernel_ns,
1719 					&ka->master_cycle_now);
1720 
1721 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1722 				&& !backwards_tsc_observed
1723 				&& !ka->boot_vcpu_runs_old_kvmclock;
1724 
1725 	if (ka->use_master_clock)
1726 		atomic_set(&kvm_guest_has_master_clock, 1);
1727 
1728 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1729 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1730 					vcpus_matched);
1731 #endif
1732 }
1733 
1734 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1735 {
1736 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1737 }
1738 
1739 static void kvm_gen_update_masterclock(struct kvm *kvm)
1740 {
1741 #ifdef CONFIG_X86_64
1742 	int i;
1743 	struct kvm_vcpu *vcpu;
1744 	struct kvm_arch *ka = &kvm->arch;
1745 
1746 	spin_lock(&ka->pvclock_gtod_sync_lock);
1747 	kvm_make_mclock_inprogress_request(kvm);
1748 	/* no guest entries from this point */
1749 	pvclock_update_vm_gtod_copy(kvm);
1750 
1751 	kvm_for_each_vcpu(i, vcpu, kvm)
1752 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1753 
1754 	/* guest entries allowed */
1755 	kvm_for_each_vcpu(i, vcpu, kvm)
1756 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1757 
1758 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1759 #endif
1760 }
1761 
1762 u64 get_kvmclock_ns(struct kvm *kvm)
1763 {
1764 	struct kvm_arch *ka = &kvm->arch;
1765 	struct pvclock_vcpu_time_info hv_clock;
1766 	u64 ret;
1767 
1768 	spin_lock(&ka->pvclock_gtod_sync_lock);
1769 	if (!ka->use_master_clock) {
1770 		spin_unlock(&ka->pvclock_gtod_sync_lock);
1771 		return ktime_get_boot_ns() + ka->kvmclock_offset;
1772 	}
1773 
1774 	hv_clock.tsc_timestamp = ka->master_cycle_now;
1775 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1776 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1777 
1778 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
1779 	get_cpu();
1780 
1781 	kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1782 			   &hv_clock.tsc_shift,
1783 			   &hv_clock.tsc_to_system_mul);
1784 	ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1785 
1786 	put_cpu();
1787 
1788 	return ret;
1789 }
1790 
1791 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1792 {
1793 	struct kvm_vcpu_arch *vcpu = &v->arch;
1794 	struct pvclock_vcpu_time_info guest_hv_clock;
1795 
1796 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1797 		&guest_hv_clock, sizeof(guest_hv_clock))))
1798 		return;
1799 
1800 	/* This VCPU is paused, but it's legal for a guest to read another
1801 	 * VCPU's kvmclock, so we really have to follow the specification where
1802 	 * it says that version is odd if data is being modified, and even after
1803 	 * it is consistent.
1804 	 *
1805 	 * Version field updates must be kept separate.  This is because
1806 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1807 	 * writes within a string instruction are weakly ordered.  So there
1808 	 * are three writes overall.
1809 	 *
1810 	 * As a small optimization, only write the version field in the first
1811 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1812 	 * version field is the first in the struct.
1813 	 */
1814 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1815 
1816 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1817 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1818 				&vcpu->hv_clock,
1819 				sizeof(vcpu->hv_clock.version));
1820 
1821 	smp_wmb();
1822 
1823 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1824 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1825 
1826 	if (vcpu->pvclock_set_guest_stopped_request) {
1827 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1828 		vcpu->pvclock_set_guest_stopped_request = false;
1829 	}
1830 
1831 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1832 
1833 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1834 				&vcpu->hv_clock,
1835 				sizeof(vcpu->hv_clock));
1836 
1837 	smp_wmb();
1838 
1839 	vcpu->hv_clock.version++;
1840 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1841 				&vcpu->hv_clock,
1842 				sizeof(vcpu->hv_clock.version));
1843 }
1844 
1845 static int kvm_guest_time_update(struct kvm_vcpu *v)
1846 {
1847 	unsigned long flags, tgt_tsc_khz;
1848 	struct kvm_vcpu_arch *vcpu = &v->arch;
1849 	struct kvm_arch *ka = &v->kvm->arch;
1850 	s64 kernel_ns;
1851 	u64 tsc_timestamp, host_tsc;
1852 	u8 pvclock_flags;
1853 	bool use_master_clock;
1854 
1855 	kernel_ns = 0;
1856 	host_tsc = 0;
1857 
1858 	/*
1859 	 * If the host uses TSC clock, then passthrough TSC as stable
1860 	 * to the guest.
1861 	 */
1862 	spin_lock(&ka->pvclock_gtod_sync_lock);
1863 	use_master_clock = ka->use_master_clock;
1864 	if (use_master_clock) {
1865 		host_tsc = ka->master_cycle_now;
1866 		kernel_ns = ka->master_kernel_ns;
1867 	}
1868 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1869 
1870 	/* Keep irq disabled to prevent changes to the clock */
1871 	local_irq_save(flags);
1872 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1873 	if (unlikely(tgt_tsc_khz == 0)) {
1874 		local_irq_restore(flags);
1875 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1876 		return 1;
1877 	}
1878 	if (!use_master_clock) {
1879 		host_tsc = rdtsc();
1880 		kernel_ns = ktime_get_boot_ns();
1881 	}
1882 
1883 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1884 
1885 	/*
1886 	 * We may have to catch up the TSC to match elapsed wall clock
1887 	 * time for two reasons, even if kvmclock is used.
1888 	 *   1) CPU could have been running below the maximum TSC rate
1889 	 *   2) Broken TSC compensation resets the base at each VCPU
1890 	 *      entry to avoid unknown leaps of TSC even when running
1891 	 *      again on the same CPU.  This may cause apparent elapsed
1892 	 *      time to disappear, and the guest to stand still or run
1893 	 *	very slowly.
1894 	 */
1895 	if (vcpu->tsc_catchup) {
1896 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1897 		if (tsc > tsc_timestamp) {
1898 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1899 			tsc_timestamp = tsc;
1900 		}
1901 	}
1902 
1903 	local_irq_restore(flags);
1904 
1905 	/* With all the info we got, fill in the values */
1906 
1907 	if (kvm_has_tsc_control)
1908 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1909 
1910 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1911 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1912 				   &vcpu->hv_clock.tsc_shift,
1913 				   &vcpu->hv_clock.tsc_to_system_mul);
1914 		vcpu->hw_tsc_khz = tgt_tsc_khz;
1915 	}
1916 
1917 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1918 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1919 	vcpu->last_guest_tsc = tsc_timestamp;
1920 
1921 	/* If the host uses TSC clocksource, then it is stable */
1922 	pvclock_flags = 0;
1923 	if (use_master_clock)
1924 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1925 
1926 	vcpu->hv_clock.flags = pvclock_flags;
1927 
1928 	if (vcpu->pv_time_enabled)
1929 		kvm_setup_pvclock_page(v);
1930 	if (v == kvm_get_vcpu(v->kvm, 0))
1931 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1932 	return 0;
1933 }
1934 
1935 /*
1936  * kvmclock updates which are isolated to a given vcpu, such as
1937  * vcpu->cpu migration, should not allow system_timestamp from
1938  * the rest of the vcpus to remain static. Otherwise ntp frequency
1939  * correction applies to one vcpu's system_timestamp but not
1940  * the others.
1941  *
1942  * So in those cases, request a kvmclock update for all vcpus.
1943  * We need to rate-limit these requests though, as they can
1944  * considerably slow guests that have a large number of vcpus.
1945  * The time for a remote vcpu to update its kvmclock is bound
1946  * by the delay we use to rate-limit the updates.
1947  */
1948 
1949 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1950 
1951 static void kvmclock_update_fn(struct work_struct *work)
1952 {
1953 	int i;
1954 	struct delayed_work *dwork = to_delayed_work(work);
1955 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1956 					   kvmclock_update_work);
1957 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1958 	struct kvm_vcpu *vcpu;
1959 
1960 	kvm_for_each_vcpu(i, vcpu, kvm) {
1961 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1962 		kvm_vcpu_kick(vcpu);
1963 	}
1964 }
1965 
1966 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1967 {
1968 	struct kvm *kvm = v->kvm;
1969 
1970 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1971 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1972 					KVMCLOCK_UPDATE_DELAY);
1973 }
1974 
1975 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1976 
1977 static void kvmclock_sync_fn(struct work_struct *work)
1978 {
1979 	struct delayed_work *dwork = to_delayed_work(work);
1980 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1981 					   kvmclock_sync_work);
1982 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1983 
1984 	if (!kvmclock_periodic_sync)
1985 		return;
1986 
1987 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1988 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1989 					KVMCLOCK_SYNC_PERIOD);
1990 }
1991 
1992 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1993 {
1994 	u64 mcg_cap = vcpu->arch.mcg_cap;
1995 	unsigned bank_num = mcg_cap & 0xff;
1996 
1997 	switch (msr) {
1998 	case MSR_IA32_MCG_STATUS:
1999 		vcpu->arch.mcg_status = data;
2000 		break;
2001 	case MSR_IA32_MCG_CTL:
2002 		if (!(mcg_cap & MCG_CTL_P))
2003 			return 1;
2004 		if (data != 0 && data != ~(u64)0)
2005 			return -1;
2006 		vcpu->arch.mcg_ctl = data;
2007 		break;
2008 	default:
2009 		if (msr >= MSR_IA32_MC0_CTL &&
2010 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2011 			u32 offset = msr - MSR_IA32_MC0_CTL;
2012 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2013 			 * some Linux kernels though clear bit 10 in bank 4 to
2014 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2015 			 * this to avoid an uncatched #GP in the guest
2016 			 */
2017 			if ((offset & 0x3) == 0 &&
2018 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2019 				return -1;
2020 			vcpu->arch.mce_banks[offset] = data;
2021 			break;
2022 		}
2023 		return 1;
2024 	}
2025 	return 0;
2026 }
2027 
2028 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2029 {
2030 	struct kvm *kvm = vcpu->kvm;
2031 	int lm = is_long_mode(vcpu);
2032 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2033 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2034 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2035 		: kvm->arch.xen_hvm_config.blob_size_32;
2036 	u32 page_num = data & ~PAGE_MASK;
2037 	u64 page_addr = data & PAGE_MASK;
2038 	u8 *page;
2039 	int r;
2040 
2041 	r = -E2BIG;
2042 	if (page_num >= blob_size)
2043 		goto out;
2044 	r = -ENOMEM;
2045 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2046 	if (IS_ERR(page)) {
2047 		r = PTR_ERR(page);
2048 		goto out;
2049 	}
2050 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2051 		goto out_free;
2052 	r = 0;
2053 out_free:
2054 	kfree(page);
2055 out:
2056 	return r;
2057 }
2058 
2059 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2060 {
2061 	gpa_t gpa = data & ~0x3f;
2062 
2063 	/* Bits 2:5 are reserved, Should be zero */
2064 	if (data & 0x3c)
2065 		return 1;
2066 
2067 	vcpu->arch.apf.msr_val = data;
2068 
2069 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2070 		kvm_clear_async_pf_completion_queue(vcpu);
2071 		kvm_async_pf_hash_reset(vcpu);
2072 		return 0;
2073 	}
2074 
2075 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2076 					sizeof(u32)))
2077 		return 1;
2078 
2079 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2080 	kvm_async_pf_wakeup_all(vcpu);
2081 	return 0;
2082 }
2083 
2084 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2085 {
2086 	vcpu->arch.pv_time_enabled = false;
2087 }
2088 
2089 static void record_steal_time(struct kvm_vcpu *vcpu)
2090 {
2091 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2092 		return;
2093 
2094 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2095 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2096 		return;
2097 
2098 	vcpu->arch.st.steal.preempted = 0;
2099 
2100 	if (vcpu->arch.st.steal.version & 1)
2101 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2102 
2103 	vcpu->arch.st.steal.version += 1;
2104 
2105 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2106 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2107 
2108 	smp_wmb();
2109 
2110 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2111 		vcpu->arch.st.last_steal;
2112 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2113 
2114 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2115 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2116 
2117 	smp_wmb();
2118 
2119 	vcpu->arch.st.steal.version += 1;
2120 
2121 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2122 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2123 }
2124 
2125 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2126 {
2127 	bool pr = false;
2128 	u32 msr = msr_info->index;
2129 	u64 data = msr_info->data;
2130 
2131 	switch (msr) {
2132 	case MSR_AMD64_NB_CFG:
2133 	case MSR_IA32_UCODE_REV:
2134 	case MSR_IA32_UCODE_WRITE:
2135 	case MSR_VM_HSAVE_PA:
2136 	case MSR_AMD64_PATCH_LOADER:
2137 	case MSR_AMD64_BU_CFG2:
2138 	case MSR_AMD64_DC_CFG:
2139 		break;
2140 
2141 	case MSR_EFER:
2142 		return set_efer(vcpu, data);
2143 	case MSR_K7_HWCR:
2144 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2145 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2146 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2147 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2148 		if (data != 0) {
2149 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2150 				    data);
2151 			return 1;
2152 		}
2153 		break;
2154 	case MSR_FAM10H_MMIO_CONF_BASE:
2155 		if (data != 0) {
2156 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2157 				    "0x%llx\n", data);
2158 			return 1;
2159 		}
2160 		break;
2161 	case MSR_IA32_DEBUGCTLMSR:
2162 		if (!data) {
2163 			/* We support the non-activated case already */
2164 			break;
2165 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2166 			/* Values other than LBR and BTF are vendor-specific,
2167 			   thus reserved and should throw a #GP */
2168 			return 1;
2169 		}
2170 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2171 			    __func__, data);
2172 		break;
2173 	case 0x200 ... 0x2ff:
2174 		return kvm_mtrr_set_msr(vcpu, msr, data);
2175 	case MSR_IA32_APICBASE:
2176 		return kvm_set_apic_base(vcpu, msr_info);
2177 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2178 		return kvm_x2apic_msr_write(vcpu, msr, data);
2179 	case MSR_IA32_TSCDEADLINE:
2180 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2181 		break;
2182 	case MSR_IA32_TSC_ADJUST:
2183 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2184 			if (!msr_info->host_initiated) {
2185 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2186 				adjust_tsc_offset_guest(vcpu, adj);
2187 			}
2188 			vcpu->arch.ia32_tsc_adjust_msr = data;
2189 		}
2190 		break;
2191 	case MSR_IA32_MISC_ENABLE:
2192 		vcpu->arch.ia32_misc_enable_msr = data;
2193 		break;
2194 	case MSR_IA32_SMBASE:
2195 		if (!msr_info->host_initiated)
2196 			return 1;
2197 		vcpu->arch.smbase = data;
2198 		break;
2199 	case MSR_KVM_WALL_CLOCK_NEW:
2200 	case MSR_KVM_WALL_CLOCK:
2201 		vcpu->kvm->arch.wall_clock = data;
2202 		kvm_write_wall_clock(vcpu->kvm, data);
2203 		break;
2204 	case MSR_KVM_SYSTEM_TIME_NEW:
2205 	case MSR_KVM_SYSTEM_TIME: {
2206 		struct kvm_arch *ka = &vcpu->kvm->arch;
2207 
2208 		kvmclock_reset(vcpu);
2209 
2210 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2211 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2212 
2213 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2214 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2215 
2216 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2217 		}
2218 
2219 		vcpu->arch.time = data;
2220 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2221 
2222 		/* we verify if the enable bit is set... */
2223 		if (!(data & 1))
2224 			break;
2225 
2226 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2227 		     &vcpu->arch.pv_time, data & ~1ULL,
2228 		     sizeof(struct pvclock_vcpu_time_info)))
2229 			vcpu->arch.pv_time_enabled = false;
2230 		else
2231 			vcpu->arch.pv_time_enabled = true;
2232 
2233 		break;
2234 	}
2235 	case MSR_KVM_ASYNC_PF_EN:
2236 		if (kvm_pv_enable_async_pf(vcpu, data))
2237 			return 1;
2238 		break;
2239 	case MSR_KVM_STEAL_TIME:
2240 
2241 		if (unlikely(!sched_info_on()))
2242 			return 1;
2243 
2244 		if (data & KVM_STEAL_RESERVED_MASK)
2245 			return 1;
2246 
2247 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2248 						data & KVM_STEAL_VALID_BITS,
2249 						sizeof(struct kvm_steal_time)))
2250 			return 1;
2251 
2252 		vcpu->arch.st.msr_val = data;
2253 
2254 		if (!(data & KVM_MSR_ENABLED))
2255 			break;
2256 
2257 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2258 
2259 		break;
2260 	case MSR_KVM_PV_EOI_EN:
2261 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2262 			return 1;
2263 		break;
2264 
2265 	case MSR_IA32_MCG_CTL:
2266 	case MSR_IA32_MCG_STATUS:
2267 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2268 		return set_msr_mce(vcpu, msr, data);
2269 
2270 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2271 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2272 		pr = true; /* fall through */
2273 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2274 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2275 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2276 			return kvm_pmu_set_msr(vcpu, msr_info);
2277 
2278 		if (pr || data != 0)
2279 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2280 				    "0x%x data 0x%llx\n", msr, data);
2281 		break;
2282 	case MSR_K7_CLK_CTL:
2283 		/*
2284 		 * Ignore all writes to this no longer documented MSR.
2285 		 * Writes are only relevant for old K7 processors,
2286 		 * all pre-dating SVM, but a recommended workaround from
2287 		 * AMD for these chips. It is possible to specify the
2288 		 * affected processor models on the command line, hence
2289 		 * the need to ignore the workaround.
2290 		 */
2291 		break;
2292 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2293 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2294 	case HV_X64_MSR_CRASH_CTL:
2295 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2296 		return kvm_hv_set_msr_common(vcpu, msr, data,
2297 					     msr_info->host_initiated);
2298 	case MSR_IA32_BBL_CR_CTL3:
2299 		/* Drop writes to this legacy MSR -- see rdmsr
2300 		 * counterpart for further detail.
2301 		 */
2302 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2303 		break;
2304 	case MSR_AMD64_OSVW_ID_LENGTH:
2305 		if (!guest_cpuid_has_osvw(vcpu))
2306 			return 1;
2307 		vcpu->arch.osvw.length = data;
2308 		break;
2309 	case MSR_AMD64_OSVW_STATUS:
2310 		if (!guest_cpuid_has_osvw(vcpu))
2311 			return 1;
2312 		vcpu->arch.osvw.status = data;
2313 		break;
2314 	case MSR_PLATFORM_INFO:
2315 		if (!msr_info->host_initiated ||
2316 		    data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2317 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2318 		     cpuid_fault_enabled(vcpu)))
2319 			return 1;
2320 		vcpu->arch.msr_platform_info = data;
2321 		break;
2322 	case MSR_MISC_FEATURES_ENABLES:
2323 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2324 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2325 		     !supports_cpuid_fault(vcpu)))
2326 			return 1;
2327 		vcpu->arch.msr_misc_features_enables = data;
2328 		break;
2329 	default:
2330 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2331 			return xen_hvm_config(vcpu, data);
2332 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2333 			return kvm_pmu_set_msr(vcpu, msr_info);
2334 		if (!ignore_msrs) {
2335 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2336 				    msr, data);
2337 			return 1;
2338 		} else {
2339 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2340 				    msr, data);
2341 			break;
2342 		}
2343 	}
2344 	return 0;
2345 }
2346 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2347 
2348 
2349 /*
2350  * Reads an msr value (of 'msr_index') into 'pdata'.
2351  * Returns 0 on success, non-0 otherwise.
2352  * Assumes vcpu_load() was already called.
2353  */
2354 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2355 {
2356 	return kvm_x86_ops->get_msr(vcpu, msr);
2357 }
2358 EXPORT_SYMBOL_GPL(kvm_get_msr);
2359 
2360 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2361 {
2362 	u64 data;
2363 	u64 mcg_cap = vcpu->arch.mcg_cap;
2364 	unsigned bank_num = mcg_cap & 0xff;
2365 
2366 	switch (msr) {
2367 	case MSR_IA32_P5_MC_ADDR:
2368 	case MSR_IA32_P5_MC_TYPE:
2369 		data = 0;
2370 		break;
2371 	case MSR_IA32_MCG_CAP:
2372 		data = vcpu->arch.mcg_cap;
2373 		break;
2374 	case MSR_IA32_MCG_CTL:
2375 		if (!(mcg_cap & MCG_CTL_P))
2376 			return 1;
2377 		data = vcpu->arch.mcg_ctl;
2378 		break;
2379 	case MSR_IA32_MCG_STATUS:
2380 		data = vcpu->arch.mcg_status;
2381 		break;
2382 	default:
2383 		if (msr >= MSR_IA32_MC0_CTL &&
2384 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2385 			u32 offset = msr - MSR_IA32_MC0_CTL;
2386 			data = vcpu->arch.mce_banks[offset];
2387 			break;
2388 		}
2389 		return 1;
2390 	}
2391 	*pdata = data;
2392 	return 0;
2393 }
2394 
2395 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2396 {
2397 	switch (msr_info->index) {
2398 	case MSR_IA32_PLATFORM_ID:
2399 	case MSR_IA32_EBL_CR_POWERON:
2400 	case MSR_IA32_DEBUGCTLMSR:
2401 	case MSR_IA32_LASTBRANCHFROMIP:
2402 	case MSR_IA32_LASTBRANCHTOIP:
2403 	case MSR_IA32_LASTINTFROMIP:
2404 	case MSR_IA32_LASTINTTOIP:
2405 	case MSR_K8_SYSCFG:
2406 	case MSR_K8_TSEG_ADDR:
2407 	case MSR_K8_TSEG_MASK:
2408 	case MSR_K7_HWCR:
2409 	case MSR_VM_HSAVE_PA:
2410 	case MSR_K8_INT_PENDING_MSG:
2411 	case MSR_AMD64_NB_CFG:
2412 	case MSR_FAM10H_MMIO_CONF_BASE:
2413 	case MSR_AMD64_BU_CFG2:
2414 	case MSR_IA32_PERF_CTL:
2415 	case MSR_AMD64_DC_CFG:
2416 		msr_info->data = 0;
2417 		break;
2418 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2419 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2420 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2421 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2422 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2423 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2424 		msr_info->data = 0;
2425 		break;
2426 	case MSR_IA32_UCODE_REV:
2427 		msr_info->data = 0x100000000ULL;
2428 		break;
2429 	case MSR_MTRRcap:
2430 	case 0x200 ... 0x2ff:
2431 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2432 	case 0xcd: /* fsb frequency */
2433 		msr_info->data = 3;
2434 		break;
2435 		/*
2436 		 * MSR_EBC_FREQUENCY_ID
2437 		 * Conservative value valid for even the basic CPU models.
2438 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2439 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2440 		 * and 266MHz for model 3, or 4. Set Core Clock
2441 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2442 		 * 31:24) even though these are only valid for CPU
2443 		 * models > 2, however guests may end up dividing or
2444 		 * multiplying by zero otherwise.
2445 		 */
2446 	case MSR_EBC_FREQUENCY_ID:
2447 		msr_info->data = 1 << 24;
2448 		break;
2449 	case MSR_IA32_APICBASE:
2450 		msr_info->data = kvm_get_apic_base(vcpu);
2451 		break;
2452 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2453 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2454 		break;
2455 	case MSR_IA32_TSCDEADLINE:
2456 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2457 		break;
2458 	case MSR_IA32_TSC_ADJUST:
2459 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2460 		break;
2461 	case MSR_IA32_MISC_ENABLE:
2462 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2463 		break;
2464 	case MSR_IA32_SMBASE:
2465 		if (!msr_info->host_initiated)
2466 			return 1;
2467 		msr_info->data = vcpu->arch.smbase;
2468 		break;
2469 	case MSR_IA32_PERF_STATUS:
2470 		/* TSC increment by tick */
2471 		msr_info->data = 1000ULL;
2472 		/* CPU multiplier */
2473 		msr_info->data |= (((uint64_t)4ULL) << 40);
2474 		break;
2475 	case MSR_EFER:
2476 		msr_info->data = vcpu->arch.efer;
2477 		break;
2478 	case MSR_KVM_WALL_CLOCK:
2479 	case MSR_KVM_WALL_CLOCK_NEW:
2480 		msr_info->data = vcpu->kvm->arch.wall_clock;
2481 		break;
2482 	case MSR_KVM_SYSTEM_TIME:
2483 	case MSR_KVM_SYSTEM_TIME_NEW:
2484 		msr_info->data = vcpu->arch.time;
2485 		break;
2486 	case MSR_KVM_ASYNC_PF_EN:
2487 		msr_info->data = vcpu->arch.apf.msr_val;
2488 		break;
2489 	case MSR_KVM_STEAL_TIME:
2490 		msr_info->data = vcpu->arch.st.msr_val;
2491 		break;
2492 	case MSR_KVM_PV_EOI_EN:
2493 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2494 		break;
2495 	case MSR_IA32_P5_MC_ADDR:
2496 	case MSR_IA32_P5_MC_TYPE:
2497 	case MSR_IA32_MCG_CAP:
2498 	case MSR_IA32_MCG_CTL:
2499 	case MSR_IA32_MCG_STATUS:
2500 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2501 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2502 	case MSR_K7_CLK_CTL:
2503 		/*
2504 		 * Provide expected ramp-up count for K7. All other
2505 		 * are set to zero, indicating minimum divisors for
2506 		 * every field.
2507 		 *
2508 		 * This prevents guest kernels on AMD host with CPU
2509 		 * type 6, model 8 and higher from exploding due to
2510 		 * the rdmsr failing.
2511 		 */
2512 		msr_info->data = 0x20000000;
2513 		break;
2514 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2515 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2516 	case HV_X64_MSR_CRASH_CTL:
2517 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2518 		return kvm_hv_get_msr_common(vcpu,
2519 					     msr_info->index, &msr_info->data);
2520 		break;
2521 	case MSR_IA32_BBL_CR_CTL3:
2522 		/* This legacy MSR exists but isn't fully documented in current
2523 		 * silicon.  It is however accessed by winxp in very narrow
2524 		 * scenarios where it sets bit #19, itself documented as
2525 		 * a "reserved" bit.  Best effort attempt to source coherent
2526 		 * read data here should the balance of the register be
2527 		 * interpreted by the guest:
2528 		 *
2529 		 * L2 cache control register 3: 64GB range, 256KB size,
2530 		 * enabled, latency 0x1, configured
2531 		 */
2532 		msr_info->data = 0xbe702111;
2533 		break;
2534 	case MSR_AMD64_OSVW_ID_LENGTH:
2535 		if (!guest_cpuid_has_osvw(vcpu))
2536 			return 1;
2537 		msr_info->data = vcpu->arch.osvw.length;
2538 		break;
2539 	case MSR_AMD64_OSVW_STATUS:
2540 		if (!guest_cpuid_has_osvw(vcpu))
2541 			return 1;
2542 		msr_info->data = vcpu->arch.osvw.status;
2543 		break;
2544 	case MSR_PLATFORM_INFO:
2545 		msr_info->data = vcpu->arch.msr_platform_info;
2546 		break;
2547 	case MSR_MISC_FEATURES_ENABLES:
2548 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2549 		break;
2550 	default:
2551 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2552 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2553 		if (!ignore_msrs) {
2554 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2555 					       msr_info->index);
2556 			return 1;
2557 		} else {
2558 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2559 			msr_info->data = 0;
2560 		}
2561 		break;
2562 	}
2563 	return 0;
2564 }
2565 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2566 
2567 /*
2568  * Read or write a bunch of msrs. All parameters are kernel addresses.
2569  *
2570  * @return number of msrs set successfully.
2571  */
2572 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2573 		    struct kvm_msr_entry *entries,
2574 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2575 				  unsigned index, u64 *data))
2576 {
2577 	int i, idx;
2578 
2579 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2580 	for (i = 0; i < msrs->nmsrs; ++i)
2581 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2582 			break;
2583 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2584 
2585 	return i;
2586 }
2587 
2588 /*
2589  * Read or write a bunch of msrs. Parameters are user addresses.
2590  *
2591  * @return number of msrs set successfully.
2592  */
2593 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2594 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2595 				unsigned index, u64 *data),
2596 		  int writeback)
2597 {
2598 	struct kvm_msrs msrs;
2599 	struct kvm_msr_entry *entries;
2600 	int r, n;
2601 	unsigned size;
2602 
2603 	r = -EFAULT;
2604 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2605 		goto out;
2606 
2607 	r = -E2BIG;
2608 	if (msrs.nmsrs >= MAX_IO_MSRS)
2609 		goto out;
2610 
2611 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2612 	entries = memdup_user(user_msrs->entries, size);
2613 	if (IS_ERR(entries)) {
2614 		r = PTR_ERR(entries);
2615 		goto out;
2616 	}
2617 
2618 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2619 	if (r < 0)
2620 		goto out_free;
2621 
2622 	r = -EFAULT;
2623 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2624 		goto out_free;
2625 
2626 	r = n;
2627 
2628 out_free:
2629 	kfree(entries);
2630 out:
2631 	return r;
2632 }
2633 
2634 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2635 {
2636 	int r;
2637 
2638 	switch (ext) {
2639 	case KVM_CAP_IRQCHIP:
2640 	case KVM_CAP_HLT:
2641 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2642 	case KVM_CAP_SET_TSS_ADDR:
2643 	case KVM_CAP_EXT_CPUID:
2644 	case KVM_CAP_EXT_EMUL_CPUID:
2645 	case KVM_CAP_CLOCKSOURCE:
2646 	case KVM_CAP_PIT:
2647 	case KVM_CAP_NOP_IO_DELAY:
2648 	case KVM_CAP_MP_STATE:
2649 	case KVM_CAP_SYNC_MMU:
2650 	case KVM_CAP_USER_NMI:
2651 	case KVM_CAP_REINJECT_CONTROL:
2652 	case KVM_CAP_IRQ_INJECT_STATUS:
2653 	case KVM_CAP_IOEVENTFD:
2654 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2655 	case KVM_CAP_PIT2:
2656 	case KVM_CAP_PIT_STATE2:
2657 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2658 	case KVM_CAP_XEN_HVM:
2659 	case KVM_CAP_VCPU_EVENTS:
2660 	case KVM_CAP_HYPERV:
2661 	case KVM_CAP_HYPERV_VAPIC:
2662 	case KVM_CAP_HYPERV_SPIN:
2663 	case KVM_CAP_HYPERV_SYNIC:
2664 	case KVM_CAP_PCI_SEGMENT:
2665 	case KVM_CAP_DEBUGREGS:
2666 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2667 	case KVM_CAP_XSAVE:
2668 	case KVM_CAP_ASYNC_PF:
2669 	case KVM_CAP_GET_TSC_KHZ:
2670 	case KVM_CAP_KVMCLOCK_CTRL:
2671 	case KVM_CAP_READONLY_MEM:
2672 	case KVM_CAP_HYPERV_TIME:
2673 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2674 	case KVM_CAP_TSC_DEADLINE_TIMER:
2675 	case KVM_CAP_ENABLE_CAP_VM:
2676 	case KVM_CAP_DISABLE_QUIRKS:
2677 	case KVM_CAP_SET_BOOT_CPU_ID:
2678  	case KVM_CAP_SPLIT_IRQCHIP:
2679 	case KVM_CAP_IMMEDIATE_EXIT:
2680 		r = 1;
2681 		break;
2682 	case KVM_CAP_ADJUST_CLOCK:
2683 		r = KVM_CLOCK_TSC_STABLE;
2684 		break;
2685 	case KVM_CAP_X86_GUEST_MWAIT:
2686 		r = kvm_mwait_in_guest();
2687 		break;
2688 	case KVM_CAP_X86_SMM:
2689 		/* SMBASE is usually relocated above 1M on modern chipsets,
2690 		 * and SMM handlers might indeed rely on 4G segment limits,
2691 		 * so do not report SMM to be available if real mode is
2692 		 * emulated via vm86 mode.  Still, do not go to great lengths
2693 		 * to avoid userspace's usage of the feature, because it is a
2694 		 * fringe case that is not enabled except via specific settings
2695 		 * of the module parameters.
2696 		 */
2697 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2698 		break;
2699 	case KVM_CAP_VAPIC:
2700 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2701 		break;
2702 	case KVM_CAP_NR_VCPUS:
2703 		r = KVM_SOFT_MAX_VCPUS;
2704 		break;
2705 	case KVM_CAP_MAX_VCPUS:
2706 		r = KVM_MAX_VCPUS;
2707 		break;
2708 	case KVM_CAP_NR_MEMSLOTS:
2709 		r = KVM_USER_MEM_SLOTS;
2710 		break;
2711 	case KVM_CAP_PV_MMU:	/* obsolete */
2712 		r = 0;
2713 		break;
2714 	case KVM_CAP_MCE:
2715 		r = KVM_MAX_MCE_BANKS;
2716 		break;
2717 	case KVM_CAP_XCRS:
2718 		r = boot_cpu_has(X86_FEATURE_XSAVE);
2719 		break;
2720 	case KVM_CAP_TSC_CONTROL:
2721 		r = kvm_has_tsc_control;
2722 		break;
2723 	case KVM_CAP_X2APIC_API:
2724 		r = KVM_X2APIC_API_VALID_FLAGS;
2725 		break;
2726 	default:
2727 		r = 0;
2728 		break;
2729 	}
2730 	return r;
2731 
2732 }
2733 
2734 long kvm_arch_dev_ioctl(struct file *filp,
2735 			unsigned int ioctl, unsigned long arg)
2736 {
2737 	void __user *argp = (void __user *)arg;
2738 	long r;
2739 
2740 	switch (ioctl) {
2741 	case KVM_GET_MSR_INDEX_LIST: {
2742 		struct kvm_msr_list __user *user_msr_list = argp;
2743 		struct kvm_msr_list msr_list;
2744 		unsigned n;
2745 
2746 		r = -EFAULT;
2747 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2748 			goto out;
2749 		n = msr_list.nmsrs;
2750 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2751 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2752 			goto out;
2753 		r = -E2BIG;
2754 		if (n < msr_list.nmsrs)
2755 			goto out;
2756 		r = -EFAULT;
2757 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2758 				 num_msrs_to_save * sizeof(u32)))
2759 			goto out;
2760 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2761 				 &emulated_msrs,
2762 				 num_emulated_msrs * sizeof(u32)))
2763 			goto out;
2764 		r = 0;
2765 		break;
2766 	}
2767 	case KVM_GET_SUPPORTED_CPUID:
2768 	case KVM_GET_EMULATED_CPUID: {
2769 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2770 		struct kvm_cpuid2 cpuid;
2771 
2772 		r = -EFAULT;
2773 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2774 			goto out;
2775 
2776 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2777 					    ioctl);
2778 		if (r)
2779 			goto out;
2780 
2781 		r = -EFAULT;
2782 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2783 			goto out;
2784 		r = 0;
2785 		break;
2786 	}
2787 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2788 		r = -EFAULT;
2789 		if (copy_to_user(argp, &kvm_mce_cap_supported,
2790 				 sizeof(kvm_mce_cap_supported)))
2791 			goto out;
2792 		r = 0;
2793 		break;
2794 	}
2795 	default:
2796 		r = -EINVAL;
2797 	}
2798 out:
2799 	return r;
2800 }
2801 
2802 static void wbinvd_ipi(void *garbage)
2803 {
2804 	wbinvd();
2805 }
2806 
2807 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2808 {
2809 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2810 }
2811 
2812 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2813 {
2814 	/* Address WBINVD may be executed by guest */
2815 	if (need_emulate_wbinvd(vcpu)) {
2816 		if (kvm_x86_ops->has_wbinvd_exit())
2817 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2818 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2819 			smp_call_function_single(vcpu->cpu,
2820 					wbinvd_ipi, NULL, 1);
2821 	}
2822 
2823 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2824 
2825 	/* Apply any externally detected TSC adjustments (due to suspend) */
2826 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2827 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2828 		vcpu->arch.tsc_offset_adjustment = 0;
2829 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2830 	}
2831 
2832 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2833 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2834 				rdtsc() - vcpu->arch.last_host_tsc;
2835 		if (tsc_delta < 0)
2836 			mark_tsc_unstable("KVM discovered backwards TSC");
2837 
2838 		if (check_tsc_unstable()) {
2839 			u64 offset = kvm_compute_tsc_offset(vcpu,
2840 						vcpu->arch.last_guest_tsc);
2841 			kvm_vcpu_write_tsc_offset(vcpu, offset);
2842 			vcpu->arch.tsc_catchup = 1;
2843 		}
2844 		if (kvm_lapic_hv_timer_in_use(vcpu) &&
2845 				kvm_x86_ops->set_hv_timer(vcpu,
2846 					kvm_get_lapic_target_expiration_tsc(vcpu)))
2847 			kvm_lapic_switch_to_sw_timer(vcpu);
2848 		/*
2849 		 * On a host with synchronized TSC, there is no need to update
2850 		 * kvmclock on vcpu->cpu migration
2851 		 */
2852 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2853 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2854 		if (vcpu->cpu != cpu)
2855 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2856 		vcpu->cpu = cpu;
2857 	}
2858 
2859 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2860 }
2861 
2862 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2863 {
2864 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2865 		return;
2866 
2867 	vcpu->arch.st.steal.preempted = 1;
2868 
2869 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2870 			&vcpu->arch.st.steal.preempted,
2871 			offsetof(struct kvm_steal_time, preempted),
2872 			sizeof(vcpu->arch.st.steal.preempted));
2873 }
2874 
2875 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2876 {
2877 	int idx;
2878 	/*
2879 	 * Disable page faults because we're in atomic context here.
2880 	 * kvm_write_guest_offset_cached() would call might_fault()
2881 	 * that relies on pagefault_disable() to tell if there's a
2882 	 * bug. NOTE: the write to guest memory may not go through if
2883 	 * during postcopy live migration or if there's heavy guest
2884 	 * paging.
2885 	 */
2886 	pagefault_disable();
2887 	/*
2888 	 * kvm_memslots() will be called by
2889 	 * kvm_write_guest_offset_cached() so take the srcu lock.
2890 	 */
2891 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2892 	kvm_steal_time_set_preempted(vcpu);
2893 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2894 	pagefault_enable();
2895 	kvm_x86_ops->vcpu_put(vcpu);
2896 	kvm_put_guest_fpu(vcpu);
2897 	vcpu->arch.last_host_tsc = rdtsc();
2898 }
2899 
2900 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2901 				    struct kvm_lapic_state *s)
2902 {
2903 	if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2904 		kvm_x86_ops->sync_pir_to_irr(vcpu);
2905 
2906 	return kvm_apic_get_state(vcpu, s);
2907 }
2908 
2909 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2910 				    struct kvm_lapic_state *s)
2911 {
2912 	int r;
2913 
2914 	r = kvm_apic_set_state(vcpu, s);
2915 	if (r)
2916 		return r;
2917 	update_cr8_intercept(vcpu);
2918 
2919 	return 0;
2920 }
2921 
2922 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2923 {
2924 	return (!lapic_in_kernel(vcpu) ||
2925 		kvm_apic_accept_pic_intr(vcpu));
2926 }
2927 
2928 /*
2929  * if userspace requested an interrupt window, check that the
2930  * interrupt window is open.
2931  *
2932  * No need to exit to userspace if we already have an interrupt queued.
2933  */
2934 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2935 {
2936 	return kvm_arch_interrupt_allowed(vcpu) &&
2937 		!kvm_cpu_has_interrupt(vcpu) &&
2938 		!kvm_event_needs_reinjection(vcpu) &&
2939 		kvm_cpu_accept_dm_intr(vcpu);
2940 }
2941 
2942 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2943 				    struct kvm_interrupt *irq)
2944 {
2945 	if (irq->irq >= KVM_NR_INTERRUPTS)
2946 		return -EINVAL;
2947 
2948 	if (!irqchip_in_kernel(vcpu->kvm)) {
2949 		kvm_queue_interrupt(vcpu, irq->irq, false);
2950 		kvm_make_request(KVM_REQ_EVENT, vcpu);
2951 		return 0;
2952 	}
2953 
2954 	/*
2955 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2956 	 * fail for in-kernel 8259.
2957 	 */
2958 	if (pic_in_kernel(vcpu->kvm))
2959 		return -ENXIO;
2960 
2961 	if (vcpu->arch.pending_external_vector != -1)
2962 		return -EEXIST;
2963 
2964 	vcpu->arch.pending_external_vector = irq->irq;
2965 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2966 	return 0;
2967 }
2968 
2969 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2970 {
2971 	kvm_inject_nmi(vcpu);
2972 
2973 	return 0;
2974 }
2975 
2976 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2977 {
2978 	kvm_make_request(KVM_REQ_SMI, vcpu);
2979 
2980 	return 0;
2981 }
2982 
2983 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2984 					   struct kvm_tpr_access_ctl *tac)
2985 {
2986 	if (tac->flags)
2987 		return -EINVAL;
2988 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2989 	return 0;
2990 }
2991 
2992 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2993 					u64 mcg_cap)
2994 {
2995 	int r;
2996 	unsigned bank_num = mcg_cap & 0xff, bank;
2997 
2998 	r = -EINVAL;
2999 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3000 		goto out;
3001 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3002 		goto out;
3003 	r = 0;
3004 	vcpu->arch.mcg_cap = mcg_cap;
3005 	/* Init IA32_MCG_CTL to all 1s */
3006 	if (mcg_cap & MCG_CTL_P)
3007 		vcpu->arch.mcg_ctl = ~(u64)0;
3008 	/* Init IA32_MCi_CTL to all 1s */
3009 	for (bank = 0; bank < bank_num; bank++)
3010 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3011 
3012 	if (kvm_x86_ops->setup_mce)
3013 		kvm_x86_ops->setup_mce(vcpu);
3014 out:
3015 	return r;
3016 }
3017 
3018 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3019 				      struct kvm_x86_mce *mce)
3020 {
3021 	u64 mcg_cap = vcpu->arch.mcg_cap;
3022 	unsigned bank_num = mcg_cap & 0xff;
3023 	u64 *banks = vcpu->arch.mce_banks;
3024 
3025 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3026 		return -EINVAL;
3027 	/*
3028 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3029 	 * reporting is disabled
3030 	 */
3031 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3032 	    vcpu->arch.mcg_ctl != ~(u64)0)
3033 		return 0;
3034 	banks += 4 * mce->bank;
3035 	/*
3036 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3037 	 * reporting is disabled for the bank
3038 	 */
3039 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3040 		return 0;
3041 	if (mce->status & MCI_STATUS_UC) {
3042 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3043 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3044 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3045 			return 0;
3046 		}
3047 		if (banks[1] & MCI_STATUS_VAL)
3048 			mce->status |= MCI_STATUS_OVER;
3049 		banks[2] = mce->addr;
3050 		banks[3] = mce->misc;
3051 		vcpu->arch.mcg_status = mce->mcg_status;
3052 		banks[1] = mce->status;
3053 		kvm_queue_exception(vcpu, MC_VECTOR);
3054 	} else if (!(banks[1] & MCI_STATUS_VAL)
3055 		   || !(banks[1] & MCI_STATUS_UC)) {
3056 		if (banks[1] & MCI_STATUS_VAL)
3057 			mce->status |= MCI_STATUS_OVER;
3058 		banks[2] = mce->addr;
3059 		banks[3] = mce->misc;
3060 		banks[1] = mce->status;
3061 	} else
3062 		banks[1] |= MCI_STATUS_OVER;
3063 	return 0;
3064 }
3065 
3066 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3067 					       struct kvm_vcpu_events *events)
3068 {
3069 	process_nmi(vcpu);
3070 	events->exception.injected =
3071 		vcpu->arch.exception.pending &&
3072 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3073 	events->exception.nr = vcpu->arch.exception.nr;
3074 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3075 	events->exception.pad = 0;
3076 	events->exception.error_code = vcpu->arch.exception.error_code;
3077 
3078 	events->interrupt.injected =
3079 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3080 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3081 	events->interrupt.soft = 0;
3082 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3083 
3084 	events->nmi.injected = vcpu->arch.nmi_injected;
3085 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3086 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3087 	events->nmi.pad = 0;
3088 
3089 	events->sipi_vector = 0; /* never valid when reporting to user space */
3090 
3091 	events->smi.smm = is_smm(vcpu);
3092 	events->smi.pending = vcpu->arch.smi_pending;
3093 	events->smi.smm_inside_nmi =
3094 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3095 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3096 
3097 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3098 			 | KVM_VCPUEVENT_VALID_SHADOW
3099 			 | KVM_VCPUEVENT_VALID_SMM);
3100 	memset(&events->reserved, 0, sizeof(events->reserved));
3101 }
3102 
3103 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3104 
3105 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3106 					      struct kvm_vcpu_events *events)
3107 {
3108 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3109 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3110 			      | KVM_VCPUEVENT_VALID_SHADOW
3111 			      | KVM_VCPUEVENT_VALID_SMM))
3112 		return -EINVAL;
3113 
3114 	if (events->exception.injected &&
3115 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3116 	     is_guest_mode(vcpu)))
3117 		return -EINVAL;
3118 
3119 	/* INITs are latched while in SMM */
3120 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3121 	    (events->smi.smm || events->smi.pending) &&
3122 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3123 		return -EINVAL;
3124 
3125 	process_nmi(vcpu);
3126 	vcpu->arch.exception.pending = events->exception.injected;
3127 	vcpu->arch.exception.nr = events->exception.nr;
3128 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3129 	vcpu->arch.exception.error_code = events->exception.error_code;
3130 
3131 	vcpu->arch.interrupt.pending = events->interrupt.injected;
3132 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3133 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3134 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3135 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3136 						  events->interrupt.shadow);
3137 
3138 	vcpu->arch.nmi_injected = events->nmi.injected;
3139 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3140 		vcpu->arch.nmi_pending = events->nmi.pending;
3141 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3142 
3143 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3144 	    lapic_in_kernel(vcpu))
3145 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3146 
3147 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3148 		u32 hflags = vcpu->arch.hflags;
3149 		if (events->smi.smm)
3150 			hflags |= HF_SMM_MASK;
3151 		else
3152 			hflags &= ~HF_SMM_MASK;
3153 		kvm_set_hflags(vcpu, hflags);
3154 
3155 		vcpu->arch.smi_pending = events->smi.pending;
3156 		if (events->smi.smm_inside_nmi)
3157 			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3158 		else
3159 			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3160 		if (lapic_in_kernel(vcpu)) {
3161 			if (events->smi.latched_init)
3162 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3163 			else
3164 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3165 		}
3166 	}
3167 
3168 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3169 
3170 	return 0;
3171 }
3172 
3173 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3174 					     struct kvm_debugregs *dbgregs)
3175 {
3176 	unsigned long val;
3177 
3178 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3179 	kvm_get_dr(vcpu, 6, &val);
3180 	dbgregs->dr6 = val;
3181 	dbgregs->dr7 = vcpu->arch.dr7;
3182 	dbgregs->flags = 0;
3183 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3184 }
3185 
3186 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3187 					    struct kvm_debugregs *dbgregs)
3188 {
3189 	if (dbgregs->flags)
3190 		return -EINVAL;
3191 
3192 	if (dbgregs->dr6 & ~0xffffffffull)
3193 		return -EINVAL;
3194 	if (dbgregs->dr7 & ~0xffffffffull)
3195 		return -EINVAL;
3196 
3197 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3198 	kvm_update_dr0123(vcpu);
3199 	vcpu->arch.dr6 = dbgregs->dr6;
3200 	kvm_update_dr6(vcpu);
3201 	vcpu->arch.dr7 = dbgregs->dr7;
3202 	kvm_update_dr7(vcpu);
3203 
3204 	return 0;
3205 }
3206 
3207 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3208 
3209 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3210 {
3211 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3212 	u64 xstate_bv = xsave->header.xfeatures;
3213 	u64 valid;
3214 
3215 	/*
3216 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3217 	 * leaves 0 and 1 in the loop below.
3218 	 */
3219 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3220 
3221 	/* Set XSTATE_BV */
3222 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3223 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3224 
3225 	/*
3226 	 * Copy each region from the possibly compacted offset to the
3227 	 * non-compacted offset.
3228 	 */
3229 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3230 	while (valid) {
3231 		u64 feature = valid & -valid;
3232 		int index = fls64(feature) - 1;
3233 		void *src = get_xsave_addr(xsave, feature);
3234 
3235 		if (src) {
3236 			u32 size, offset, ecx, edx;
3237 			cpuid_count(XSTATE_CPUID, index,
3238 				    &size, &offset, &ecx, &edx);
3239 			memcpy(dest + offset, src, size);
3240 		}
3241 
3242 		valid -= feature;
3243 	}
3244 }
3245 
3246 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3247 {
3248 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3249 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3250 	u64 valid;
3251 
3252 	/*
3253 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3254 	 * leaves 0 and 1 in the loop below.
3255 	 */
3256 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3257 
3258 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3259 	xsave->header.xfeatures = xstate_bv;
3260 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3261 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3262 
3263 	/*
3264 	 * Copy each region from the non-compacted offset to the
3265 	 * possibly compacted offset.
3266 	 */
3267 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3268 	while (valid) {
3269 		u64 feature = valid & -valid;
3270 		int index = fls64(feature) - 1;
3271 		void *dest = get_xsave_addr(xsave, feature);
3272 
3273 		if (dest) {
3274 			u32 size, offset, ecx, edx;
3275 			cpuid_count(XSTATE_CPUID, index,
3276 				    &size, &offset, &ecx, &edx);
3277 			memcpy(dest, src + offset, size);
3278 		}
3279 
3280 		valid -= feature;
3281 	}
3282 }
3283 
3284 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3285 					 struct kvm_xsave *guest_xsave)
3286 {
3287 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3288 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3289 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3290 	} else {
3291 		memcpy(guest_xsave->region,
3292 			&vcpu->arch.guest_fpu.state.fxsave,
3293 			sizeof(struct fxregs_state));
3294 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3295 			XFEATURE_MASK_FPSSE;
3296 	}
3297 }
3298 
3299 #define XSAVE_MXCSR_OFFSET 24
3300 
3301 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3302 					struct kvm_xsave *guest_xsave)
3303 {
3304 	u64 xstate_bv =
3305 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3306 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3307 
3308 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3309 		/*
3310 		 * Here we allow setting states that are not present in
3311 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3312 		 * with old userspace.
3313 		 */
3314 		if (xstate_bv & ~kvm_supported_xcr0() ||
3315 			mxcsr & ~mxcsr_feature_mask)
3316 			return -EINVAL;
3317 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3318 	} else {
3319 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3320 			mxcsr & ~mxcsr_feature_mask)
3321 			return -EINVAL;
3322 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3323 			guest_xsave->region, sizeof(struct fxregs_state));
3324 	}
3325 	return 0;
3326 }
3327 
3328 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3329 					struct kvm_xcrs *guest_xcrs)
3330 {
3331 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3332 		guest_xcrs->nr_xcrs = 0;
3333 		return;
3334 	}
3335 
3336 	guest_xcrs->nr_xcrs = 1;
3337 	guest_xcrs->flags = 0;
3338 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3339 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3340 }
3341 
3342 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3343 				       struct kvm_xcrs *guest_xcrs)
3344 {
3345 	int i, r = 0;
3346 
3347 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3348 		return -EINVAL;
3349 
3350 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3351 		return -EINVAL;
3352 
3353 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3354 		/* Only support XCR0 currently */
3355 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3356 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3357 				guest_xcrs->xcrs[i].value);
3358 			break;
3359 		}
3360 	if (r)
3361 		r = -EINVAL;
3362 	return r;
3363 }
3364 
3365 /*
3366  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3367  * stopped by the hypervisor.  This function will be called from the host only.
3368  * EINVAL is returned when the host attempts to set the flag for a guest that
3369  * does not support pv clocks.
3370  */
3371 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3372 {
3373 	if (!vcpu->arch.pv_time_enabled)
3374 		return -EINVAL;
3375 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3376 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3377 	return 0;
3378 }
3379 
3380 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3381 				     struct kvm_enable_cap *cap)
3382 {
3383 	if (cap->flags)
3384 		return -EINVAL;
3385 
3386 	switch (cap->cap) {
3387 	case KVM_CAP_HYPERV_SYNIC:
3388 		if (!irqchip_in_kernel(vcpu->kvm))
3389 			return -EINVAL;
3390 		return kvm_hv_activate_synic(vcpu);
3391 	default:
3392 		return -EINVAL;
3393 	}
3394 }
3395 
3396 long kvm_arch_vcpu_ioctl(struct file *filp,
3397 			 unsigned int ioctl, unsigned long arg)
3398 {
3399 	struct kvm_vcpu *vcpu = filp->private_data;
3400 	void __user *argp = (void __user *)arg;
3401 	int r;
3402 	union {
3403 		struct kvm_lapic_state *lapic;
3404 		struct kvm_xsave *xsave;
3405 		struct kvm_xcrs *xcrs;
3406 		void *buffer;
3407 	} u;
3408 
3409 	u.buffer = NULL;
3410 	switch (ioctl) {
3411 	case KVM_GET_LAPIC: {
3412 		r = -EINVAL;
3413 		if (!lapic_in_kernel(vcpu))
3414 			goto out;
3415 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3416 
3417 		r = -ENOMEM;
3418 		if (!u.lapic)
3419 			goto out;
3420 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3421 		if (r)
3422 			goto out;
3423 		r = -EFAULT;
3424 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3425 			goto out;
3426 		r = 0;
3427 		break;
3428 	}
3429 	case KVM_SET_LAPIC: {
3430 		r = -EINVAL;
3431 		if (!lapic_in_kernel(vcpu))
3432 			goto out;
3433 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3434 		if (IS_ERR(u.lapic))
3435 			return PTR_ERR(u.lapic);
3436 
3437 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3438 		break;
3439 	}
3440 	case KVM_INTERRUPT: {
3441 		struct kvm_interrupt irq;
3442 
3443 		r = -EFAULT;
3444 		if (copy_from_user(&irq, argp, sizeof irq))
3445 			goto out;
3446 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3447 		break;
3448 	}
3449 	case KVM_NMI: {
3450 		r = kvm_vcpu_ioctl_nmi(vcpu);
3451 		break;
3452 	}
3453 	case KVM_SMI: {
3454 		r = kvm_vcpu_ioctl_smi(vcpu);
3455 		break;
3456 	}
3457 	case KVM_SET_CPUID: {
3458 		struct kvm_cpuid __user *cpuid_arg = argp;
3459 		struct kvm_cpuid cpuid;
3460 
3461 		r = -EFAULT;
3462 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3463 			goto out;
3464 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3465 		break;
3466 	}
3467 	case KVM_SET_CPUID2: {
3468 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3469 		struct kvm_cpuid2 cpuid;
3470 
3471 		r = -EFAULT;
3472 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3473 			goto out;
3474 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3475 					      cpuid_arg->entries);
3476 		break;
3477 	}
3478 	case KVM_GET_CPUID2: {
3479 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3480 		struct kvm_cpuid2 cpuid;
3481 
3482 		r = -EFAULT;
3483 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3484 			goto out;
3485 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3486 					      cpuid_arg->entries);
3487 		if (r)
3488 			goto out;
3489 		r = -EFAULT;
3490 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3491 			goto out;
3492 		r = 0;
3493 		break;
3494 	}
3495 	case KVM_GET_MSRS:
3496 		r = msr_io(vcpu, argp, do_get_msr, 1);
3497 		break;
3498 	case KVM_SET_MSRS:
3499 		r = msr_io(vcpu, argp, do_set_msr, 0);
3500 		break;
3501 	case KVM_TPR_ACCESS_REPORTING: {
3502 		struct kvm_tpr_access_ctl tac;
3503 
3504 		r = -EFAULT;
3505 		if (copy_from_user(&tac, argp, sizeof tac))
3506 			goto out;
3507 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3508 		if (r)
3509 			goto out;
3510 		r = -EFAULT;
3511 		if (copy_to_user(argp, &tac, sizeof tac))
3512 			goto out;
3513 		r = 0;
3514 		break;
3515 	};
3516 	case KVM_SET_VAPIC_ADDR: {
3517 		struct kvm_vapic_addr va;
3518 		int idx;
3519 
3520 		r = -EINVAL;
3521 		if (!lapic_in_kernel(vcpu))
3522 			goto out;
3523 		r = -EFAULT;
3524 		if (copy_from_user(&va, argp, sizeof va))
3525 			goto out;
3526 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3527 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3528 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3529 		break;
3530 	}
3531 	case KVM_X86_SETUP_MCE: {
3532 		u64 mcg_cap;
3533 
3534 		r = -EFAULT;
3535 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3536 			goto out;
3537 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3538 		break;
3539 	}
3540 	case KVM_X86_SET_MCE: {
3541 		struct kvm_x86_mce mce;
3542 
3543 		r = -EFAULT;
3544 		if (copy_from_user(&mce, argp, sizeof mce))
3545 			goto out;
3546 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3547 		break;
3548 	}
3549 	case KVM_GET_VCPU_EVENTS: {
3550 		struct kvm_vcpu_events events;
3551 
3552 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3553 
3554 		r = -EFAULT;
3555 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3556 			break;
3557 		r = 0;
3558 		break;
3559 	}
3560 	case KVM_SET_VCPU_EVENTS: {
3561 		struct kvm_vcpu_events events;
3562 
3563 		r = -EFAULT;
3564 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3565 			break;
3566 
3567 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3568 		break;
3569 	}
3570 	case KVM_GET_DEBUGREGS: {
3571 		struct kvm_debugregs dbgregs;
3572 
3573 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3574 
3575 		r = -EFAULT;
3576 		if (copy_to_user(argp, &dbgregs,
3577 				 sizeof(struct kvm_debugregs)))
3578 			break;
3579 		r = 0;
3580 		break;
3581 	}
3582 	case KVM_SET_DEBUGREGS: {
3583 		struct kvm_debugregs dbgregs;
3584 
3585 		r = -EFAULT;
3586 		if (copy_from_user(&dbgregs, argp,
3587 				   sizeof(struct kvm_debugregs)))
3588 			break;
3589 
3590 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3591 		break;
3592 	}
3593 	case KVM_GET_XSAVE: {
3594 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3595 		r = -ENOMEM;
3596 		if (!u.xsave)
3597 			break;
3598 
3599 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3600 
3601 		r = -EFAULT;
3602 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3603 			break;
3604 		r = 0;
3605 		break;
3606 	}
3607 	case KVM_SET_XSAVE: {
3608 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3609 		if (IS_ERR(u.xsave))
3610 			return PTR_ERR(u.xsave);
3611 
3612 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3613 		break;
3614 	}
3615 	case KVM_GET_XCRS: {
3616 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3617 		r = -ENOMEM;
3618 		if (!u.xcrs)
3619 			break;
3620 
3621 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3622 
3623 		r = -EFAULT;
3624 		if (copy_to_user(argp, u.xcrs,
3625 				 sizeof(struct kvm_xcrs)))
3626 			break;
3627 		r = 0;
3628 		break;
3629 	}
3630 	case KVM_SET_XCRS: {
3631 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3632 		if (IS_ERR(u.xcrs))
3633 			return PTR_ERR(u.xcrs);
3634 
3635 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3636 		break;
3637 	}
3638 	case KVM_SET_TSC_KHZ: {
3639 		u32 user_tsc_khz;
3640 
3641 		r = -EINVAL;
3642 		user_tsc_khz = (u32)arg;
3643 
3644 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3645 			goto out;
3646 
3647 		if (user_tsc_khz == 0)
3648 			user_tsc_khz = tsc_khz;
3649 
3650 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3651 			r = 0;
3652 
3653 		goto out;
3654 	}
3655 	case KVM_GET_TSC_KHZ: {
3656 		r = vcpu->arch.virtual_tsc_khz;
3657 		goto out;
3658 	}
3659 	case KVM_KVMCLOCK_CTRL: {
3660 		r = kvm_set_guest_paused(vcpu);
3661 		goto out;
3662 	}
3663 	case KVM_ENABLE_CAP: {
3664 		struct kvm_enable_cap cap;
3665 
3666 		r = -EFAULT;
3667 		if (copy_from_user(&cap, argp, sizeof(cap)))
3668 			goto out;
3669 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3670 		break;
3671 	}
3672 	default:
3673 		r = -EINVAL;
3674 	}
3675 out:
3676 	kfree(u.buffer);
3677 	return r;
3678 }
3679 
3680 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3681 {
3682 	return VM_FAULT_SIGBUS;
3683 }
3684 
3685 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3686 {
3687 	int ret;
3688 
3689 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3690 		return -EINVAL;
3691 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3692 	return ret;
3693 }
3694 
3695 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3696 					      u64 ident_addr)
3697 {
3698 	kvm->arch.ept_identity_map_addr = ident_addr;
3699 	return 0;
3700 }
3701 
3702 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3703 					  u32 kvm_nr_mmu_pages)
3704 {
3705 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3706 		return -EINVAL;
3707 
3708 	mutex_lock(&kvm->slots_lock);
3709 
3710 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3711 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3712 
3713 	mutex_unlock(&kvm->slots_lock);
3714 	return 0;
3715 }
3716 
3717 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3718 {
3719 	return kvm->arch.n_max_mmu_pages;
3720 }
3721 
3722 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3723 {
3724 	struct kvm_pic *pic = kvm->arch.vpic;
3725 	int r;
3726 
3727 	r = 0;
3728 	switch (chip->chip_id) {
3729 	case KVM_IRQCHIP_PIC_MASTER:
3730 		memcpy(&chip->chip.pic, &pic->pics[0],
3731 			sizeof(struct kvm_pic_state));
3732 		break;
3733 	case KVM_IRQCHIP_PIC_SLAVE:
3734 		memcpy(&chip->chip.pic, &pic->pics[1],
3735 			sizeof(struct kvm_pic_state));
3736 		break;
3737 	case KVM_IRQCHIP_IOAPIC:
3738 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
3739 		break;
3740 	default:
3741 		r = -EINVAL;
3742 		break;
3743 	}
3744 	return r;
3745 }
3746 
3747 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3748 {
3749 	struct kvm_pic *pic = kvm->arch.vpic;
3750 	int r;
3751 
3752 	r = 0;
3753 	switch (chip->chip_id) {
3754 	case KVM_IRQCHIP_PIC_MASTER:
3755 		spin_lock(&pic->lock);
3756 		memcpy(&pic->pics[0], &chip->chip.pic,
3757 			sizeof(struct kvm_pic_state));
3758 		spin_unlock(&pic->lock);
3759 		break;
3760 	case KVM_IRQCHIP_PIC_SLAVE:
3761 		spin_lock(&pic->lock);
3762 		memcpy(&pic->pics[1], &chip->chip.pic,
3763 			sizeof(struct kvm_pic_state));
3764 		spin_unlock(&pic->lock);
3765 		break;
3766 	case KVM_IRQCHIP_IOAPIC:
3767 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
3768 		break;
3769 	default:
3770 		r = -EINVAL;
3771 		break;
3772 	}
3773 	kvm_pic_update_irq(pic);
3774 	return r;
3775 }
3776 
3777 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3778 {
3779 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3780 
3781 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3782 
3783 	mutex_lock(&kps->lock);
3784 	memcpy(ps, &kps->channels, sizeof(*ps));
3785 	mutex_unlock(&kps->lock);
3786 	return 0;
3787 }
3788 
3789 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3790 {
3791 	int i;
3792 	struct kvm_pit *pit = kvm->arch.vpit;
3793 
3794 	mutex_lock(&pit->pit_state.lock);
3795 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3796 	for (i = 0; i < 3; i++)
3797 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3798 	mutex_unlock(&pit->pit_state.lock);
3799 	return 0;
3800 }
3801 
3802 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3803 {
3804 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3805 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3806 		sizeof(ps->channels));
3807 	ps->flags = kvm->arch.vpit->pit_state.flags;
3808 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3809 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3810 	return 0;
3811 }
3812 
3813 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3814 {
3815 	int start = 0;
3816 	int i;
3817 	u32 prev_legacy, cur_legacy;
3818 	struct kvm_pit *pit = kvm->arch.vpit;
3819 
3820 	mutex_lock(&pit->pit_state.lock);
3821 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3822 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3823 	if (!prev_legacy && cur_legacy)
3824 		start = 1;
3825 	memcpy(&pit->pit_state.channels, &ps->channels,
3826 	       sizeof(pit->pit_state.channels));
3827 	pit->pit_state.flags = ps->flags;
3828 	for (i = 0; i < 3; i++)
3829 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3830 				   start && i == 0);
3831 	mutex_unlock(&pit->pit_state.lock);
3832 	return 0;
3833 }
3834 
3835 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3836 				 struct kvm_reinject_control *control)
3837 {
3838 	struct kvm_pit *pit = kvm->arch.vpit;
3839 
3840 	if (!pit)
3841 		return -ENXIO;
3842 
3843 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
3844 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3845 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3846 	 */
3847 	mutex_lock(&pit->pit_state.lock);
3848 	kvm_pit_set_reinject(pit, control->pit_reinject);
3849 	mutex_unlock(&pit->pit_state.lock);
3850 
3851 	return 0;
3852 }
3853 
3854 /**
3855  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3856  * @kvm: kvm instance
3857  * @log: slot id and address to which we copy the log
3858  *
3859  * Steps 1-4 below provide general overview of dirty page logging. See
3860  * kvm_get_dirty_log_protect() function description for additional details.
3861  *
3862  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3863  * always flush the TLB (step 4) even if previous step failed  and the dirty
3864  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3865  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3866  * writes will be marked dirty for next log read.
3867  *
3868  *   1. Take a snapshot of the bit and clear it if needed.
3869  *   2. Write protect the corresponding page.
3870  *   3. Copy the snapshot to the userspace.
3871  *   4. Flush TLB's if needed.
3872  */
3873 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3874 {
3875 	bool is_dirty = false;
3876 	int r;
3877 
3878 	mutex_lock(&kvm->slots_lock);
3879 
3880 	/*
3881 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3882 	 */
3883 	if (kvm_x86_ops->flush_log_dirty)
3884 		kvm_x86_ops->flush_log_dirty(kvm);
3885 
3886 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3887 
3888 	/*
3889 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3890 	 * kvm_mmu_slot_remove_write_access().
3891 	 */
3892 	lockdep_assert_held(&kvm->slots_lock);
3893 	if (is_dirty)
3894 		kvm_flush_remote_tlbs(kvm);
3895 
3896 	mutex_unlock(&kvm->slots_lock);
3897 	return r;
3898 }
3899 
3900 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3901 			bool line_status)
3902 {
3903 	if (!irqchip_in_kernel(kvm))
3904 		return -ENXIO;
3905 
3906 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3907 					irq_event->irq, irq_event->level,
3908 					line_status);
3909 	return 0;
3910 }
3911 
3912 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3913 				   struct kvm_enable_cap *cap)
3914 {
3915 	int r;
3916 
3917 	if (cap->flags)
3918 		return -EINVAL;
3919 
3920 	switch (cap->cap) {
3921 	case KVM_CAP_DISABLE_QUIRKS:
3922 		kvm->arch.disabled_quirks = cap->args[0];
3923 		r = 0;
3924 		break;
3925 	case KVM_CAP_SPLIT_IRQCHIP: {
3926 		mutex_lock(&kvm->lock);
3927 		r = -EINVAL;
3928 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3929 			goto split_irqchip_unlock;
3930 		r = -EEXIST;
3931 		if (irqchip_in_kernel(kvm))
3932 			goto split_irqchip_unlock;
3933 		if (kvm->created_vcpus)
3934 			goto split_irqchip_unlock;
3935 		r = kvm_setup_empty_irq_routing(kvm);
3936 		if (r)
3937 			goto split_irqchip_unlock;
3938 		/* Pairs with irqchip_in_kernel. */
3939 		smp_wmb();
3940 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3941 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3942 		r = 0;
3943 split_irqchip_unlock:
3944 		mutex_unlock(&kvm->lock);
3945 		break;
3946 	}
3947 	case KVM_CAP_X2APIC_API:
3948 		r = -EINVAL;
3949 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3950 			break;
3951 
3952 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3953 			kvm->arch.x2apic_format = true;
3954 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3955 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
3956 
3957 		r = 0;
3958 		break;
3959 	default:
3960 		r = -EINVAL;
3961 		break;
3962 	}
3963 	return r;
3964 }
3965 
3966 long kvm_arch_vm_ioctl(struct file *filp,
3967 		       unsigned int ioctl, unsigned long arg)
3968 {
3969 	struct kvm *kvm = filp->private_data;
3970 	void __user *argp = (void __user *)arg;
3971 	int r = -ENOTTY;
3972 	/*
3973 	 * This union makes it completely explicit to gcc-3.x
3974 	 * that these two variables' stack usage should be
3975 	 * combined, not added together.
3976 	 */
3977 	union {
3978 		struct kvm_pit_state ps;
3979 		struct kvm_pit_state2 ps2;
3980 		struct kvm_pit_config pit_config;
3981 	} u;
3982 
3983 	switch (ioctl) {
3984 	case KVM_SET_TSS_ADDR:
3985 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3986 		break;
3987 	case KVM_SET_IDENTITY_MAP_ADDR: {
3988 		u64 ident_addr;
3989 
3990 		r = -EFAULT;
3991 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3992 			goto out;
3993 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3994 		break;
3995 	}
3996 	case KVM_SET_NR_MMU_PAGES:
3997 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3998 		break;
3999 	case KVM_GET_NR_MMU_PAGES:
4000 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4001 		break;
4002 	case KVM_CREATE_IRQCHIP: {
4003 		mutex_lock(&kvm->lock);
4004 
4005 		r = -EEXIST;
4006 		if (irqchip_in_kernel(kvm))
4007 			goto create_irqchip_unlock;
4008 
4009 		r = -EINVAL;
4010 		if (kvm->created_vcpus)
4011 			goto create_irqchip_unlock;
4012 
4013 		r = kvm_pic_init(kvm);
4014 		if (r)
4015 			goto create_irqchip_unlock;
4016 
4017 		r = kvm_ioapic_init(kvm);
4018 		if (r) {
4019 			kvm_pic_destroy(kvm);
4020 			goto create_irqchip_unlock;
4021 		}
4022 
4023 		r = kvm_setup_default_irq_routing(kvm);
4024 		if (r) {
4025 			kvm_ioapic_destroy(kvm);
4026 			kvm_pic_destroy(kvm);
4027 			goto create_irqchip_unlock;
4028 		}
4029 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4030 		smp_wmb();
4031 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4032 	create_irqchip_unlock:
4033 		mutex_unlock(&kvm->lock);
4034 		break;
4035 	}
4036 	case KVM_CREATE_PIT:
4037 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4038 		goto create_pit;
4039 	case KVM_CREATE_PIT2:
4040 		r = -EFAULT;
4041 		if (copy_from_user(&u.pit_config, argp,
4042 				   sizeof(struct kvm_pit_config)))
4043 			goto out;
4044 	create_pit:
4045 		mutex_lock(&kvm->lock);
4046 		r = -EEXIST;
4047 		if (kvm->arch.vpit)
4048 			goto create_pit_unlock;
4049 		r = -ENOMEM;
4050 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4051 		if (kvm->arch.vpit)
4052 			r = 0;
4053 	create_pit_unlock:
4054 		mutex_unlock(&kvm->lock);
4055 		break;
4056 	case KVM_GET_IRQCHIP: {
4057 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4058 		struct kvm_irqchip *chip;
4059 
4060 		chip = memdup_user(argp, sizeof(*chip));
4061 		if (IS_ERR(chip)) {
4062 			r = PTR_ERR(chip);
4063 			goto out;
4064 		}
4065 
4066 		r = -ENXIO;
4067 		if (!irqchip_kernel(kvm))
4068 			goto get_irqchip_out;
4069 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4070 		if (r)
4071 			goto get_irqchip_out;
4072 		r = -EFAULT;
4073 		if (copy_to_user(argp, chip, sizeof *chip))
4074 			goto get_irqchip_out;
4075 		r = 0;
4076 	get_irqchip_out:
4077 		kfree(chip);
4078 		break;
4079 	}
4080 	case KVM_SET_IRQCHIP: {
4081 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4082 		struct kvm_irqchip *chip;
4083 
4084 		chip = memdup_user(argp, sizeof(*chip));
4085 		if (IS_ERR(chip)) {
4086 			r = PTR_ERR(chip);
4087 			goto out;
4088 		}
4089 
4090 		r = -ENXIO;
4091 		if (!irqchip_kernel(kvm))
4092 			goto set_irqchip_out;
4093 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4094 		if (r)
4095 			goto set_irqchip_out;
4096 		r = 0;
4097 	set_irqchip_out:
4098 		kfree(chip);
4099 		break;
4100 	}
4101 	case KVM_GET_PIT: {
4102 		r = -EFAULT;
4103 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4104 			goto out;
4105 		r = -ENXIO;
4106 		if (!kvm->arch.vpit)
4107 			goto out;
4108 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4109 		if (r)
4110 			goto out;
4111 		r = -EFAULT;
4112 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4113 			goto out;
4114 		r = 0;
4115 		break;
4116 	}
4117 	case KVM_SET_PIT: {
4118 		r = -EFAULT;
4119 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4120 			goto out;
4121 		r = -ENXIO;
4122 		if (!kvm->arch.vpit)
4123 			goto out;
4124 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4125 		break;
4126 	}
4127 	case KVM_GET_PIT2: {
4128 		r = -ENXIO;
4129 		if (!kvm->arch.vpit)
4130 			goto out;
4131 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4132 		if (r)
4133 			goto out;
4134 		r = -EFAULT;
4135 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4136 			goto out;
4137 		r = 0;
4138 		break;
4139 	}
4140 	case KVM_SET_PIT2: {
4141 		r = -EFAULT;
4142 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4143 			goto out;
4144 		r = -ENXIO;
4145 		if (!kvm->arch.vpit)
4146 			goto out;
4147 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4148 		break;
4149 	}
4150 	case KVM_REINJECT_CONTROL: {
4151 		struct kvm_reinject_control control;
4152 		r =  -EFAULT;
4153 		if (copy_from_user(&control, argp, sizeof(control)))
4154 			goto out;
4155 		r = kvm_vm_ioctl_reinject(kvm, &control);
4156 		break;
4157 	}
4158 	case KVM_SET_BOOT_CPU_ID:
4159 		r = 0;
4160 		mutex_lock(&kvm->lock);
4161 		if (kvm->created_vcpus)
4162 			r = -EBUSY;
4163 		else
4164 			kvm->arch.bsp_vcpu_id = arg;
4165 		mutex_unlock(&kvm->lock);
4166 		break;
4167 	case KVM_XEN_HVM_CONFIG: {
4168 		r = -EFAULT;
4169 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4170 				   sizeof(struct kvm_xen_hvm_config)))
4171 			goto out;
4172 		r = -EINVAL;
4173 		if (kvm->arch.xen_hvm_config.flags)
4174 			goto out;
4175 		r = 0;
4176 		break;
4177 	}
4178 	case KVM_SET_CLOCK: {
4179 		struct kvm_clock_data user_ns;
4180 		u64 now_ns;
4181 
4182 		r = -EFAULT;
4183 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4184 			goto out;
4185 
4186 		r = -EINVAL;
4187 		if (user_ns.flags)
4188 			goto out;
4189 
4190 		r = 0;
4191 		now_ns = get_kvmclock_ns(kvm);
4192 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4193 		kvm_gen_update_masterclock(kvm);
4194 		break;
4195 	}
4196 	case KVM_GET_CLOCK: {
4197 		struct kvm_clock_data user_ns;
4198 		u64 now_ns;
4199 
4200 		now_ns = get_kvmclock_ns(kvm);
4201 		user_ns.clock = now_ns;
4202 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4203 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4204 
4205 		r = -EFAULT;
4206 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4207 			goto out;
4208 		r = 0;
4209 		break;
4210 	}
4211 	case KVM_ENABLE_CAP: {
4212 		struct kvm_enable_cap cap;
4213 
4214 		r = -EFAULT;
4215 		if (copy_from_user(&cap, argp, sizeof(cap)))
4216 			goto out;
4217 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4218 		break;
4219 	}
4220 	default:
4221 		r = -ENOTTY;
4222 	}
4223 out:
4224 	return r;
4225 }
4226 
4227 static void kvm_init_msr_list(void)
4228 {
4229 	u32 dummy[2];
4230 	unsigned i, j;
4231 
4232 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4233 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4234 			continue;
4235 
4236 		/*
4237 		 * Even MSRs that are valid in the host may not be exposed
4238 		 * to the guests in some cases.
4239 		 */
4240 		switch (msrs_to_save[i]) {
4241 		case MSR_IA32_BNDCFGS:
4242 			if (!kvm_x86_ops->mpx_supported())
4243 				continue;
4244 			break;
4245 		case MSR_TSC_AUX:
4246 			if (!kvm_x86_ops->rdtscp_supported())
4247 				continue;
4248 			break;
4249 		default:
4250 			break;
4251 		}
4252 
4253 		if (j < i)
4254 			msrs_to_save[j] = msrs_to_save[i];
4255 		j++;
4256 	}
4257 	num_msrs_to_save = j;
4258 
4259 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4260 		switch (emulated_msrs[i]) {
4261 		case MSR_IA32_SMBASE:
4262 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4263 				continue;
4264 			break;
4265 		default:
4266 			break;
4267 		}
4268 
4269 		if (j < i)
4270 			emulated_msrs[j] = emulated_msrs[i];
4271 		j++;
4272 	}
4273 	num_emulated_msrs = j;
4274 }
4275 
4276 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4277 			   const void *v)
4278 {
4279 	int handled = 0;
4280 	int n;
4281 
4282 	do {
4283 		n = min(len, 8);
4284 		if (!(lapic_in_kernel(vcpu) &&
4285 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4286 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4287 			break;
4288 		handled += n;
4289 		addr += n;
4290 		len -= n;
4291 		v += n;
4292 	} while (len);
4293 
4294 	return handled;
4295 }
4296 
4297 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4298 {
4299 	int handled = 0;
4300 	int n;
4301 
4302 	do {
4303 		n = min(len, 8);
4304 		if (!(lapic_in_kernel(vcpu) &&
4305 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4306 					 addr, n, v))
4307 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4308 			break;
4309 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4310 		handled += n;
4311 		addr += n;
4312 		len -= n;
4313 		v += n;
4314 	} while (len);
4315 
4316 	return handled;
4317 }
4318 
4319 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4320 			struct kvm_segment *var, int seg)
4321 {
4322 	kvm_x86_ops->set_segment(vcpu, var, seg);
4323 }
4324 
4325 void kvm_get_segment(struct kvm_vcpu *vcpu,
4326 		     struct kvm_segment *var, int seg)
4327 {
4328 	kvm_x86_ops->get_segment(vcpu, var, seg);
4329 }
4330 
4331 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4332 			   struct x86_exception *exception)
4333 {
4334 	gpa_t t_gpa;
4335 
4336 	BUG_ON(!mmu_is_nested(vcpu));
4337 
4338 	/* NPT walks are always user-walks */
4339 	access |= PFERR_USER_MASK;
4340 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4341 
4342 	return t_gpa;
4343 }
4344 
4345 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4346 			      struct x86_exception *exception)
4347 {
4348 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4349 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4350 }
4351 
4352  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4353 				struct x86_exception *exception)
4354 {
4355 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4356 	access |= PFERR_FETCH_MASK;
4357 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4358 }
4359 
4360 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4361 			       struct x86_exception *exception)
4362 {
4363 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4364 	access |= PFERR_WRITE_MASK;
4365 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4366 }
4367 
4368 /* uses this to access any guest's mapped memory without checking CPL */
4369 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4370 				struct x86_exception *exception)
4371 {
4372 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4373 }
4374 
4375 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4376 				      struct kvm_vcpu *vcpu, u32 access,
4377 				      struct x86_exception *exception)
4378 {
4379 	void *data = val;
4380 	int r = X86EMUL_CONTINUE;
4381 
4382 	while (bytes) {
4383 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4384 							    exception);
4385 		unsigned offset = addr & (PAGE_SIZE-1);
4386 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4387 		int ret;
4388 
4389 		if (gpa == UNMAPPED_GVA)
4390 			return X86EMUL_PROPAGATE_FAULT;
4391 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4392 					       offset, toread);
4393 		if (ret < 0) {
4394 			r = X86EMUL_IO_NEEDED;
4395 			goto out;
4396 		}
4397 
4398 		bytes -= toread;
4399 		data += toread;
4400 		addr += toread;
4401 	}
4402 out:
4403 	return r;
4404 }
4405 
4406 /* used for instruction fetching */
4407 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4408 				gva_t addr, void *val, unsigned int bytes,
4409 				struct x86_exception *exception)
4410 {
4411 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4412 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4413 	unsigned offset;
4414 	int ret;
4415 
4416 	/* Inline kvm_read_guest_virt_helper for speed.  */
4417 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4418 						    exception);
4419 	if (unlikely(gpa == UNMAPPED_GVA))
4420 		return X86EMUL_PROPAGATE_FAULT;
4421 
4422 	offset = addr & (PAGE_SIZE-1);
4423 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4424 		bytes = (unsigned)PAGE_SIZE - offset;
4425 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4426 				       offset, bytes);
4427 	if (unlikely(ret < 0))
4428 		return X86EMUL_IO_NEEDED;
4429 
4430 	return X86EMUL_CONTINUE;
4431 }
4432 
4433 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4434 			       gva_t addr, void *val, unsigned int bytes,
4435 			       struct x86_exception *exception)
4436 {
4437 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4438 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4439 
4440 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4441 					  exception);
4442 }
4443 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4444 
4445 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4446 				      gva_t addr, void *val, unsigned int bytes,
4447 				      struct x86_exception *exception)
4448 {
4449 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4450 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4451 }
4452 
4453 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4454 		unsigned long addr, void *val, unsigned int bytes)
4455 {
4456 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4457 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4458 
4459 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4460 }
4461 
4462 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4463 				       gva_t addr, void *val,
4464 				       unsigned int bytes,
4465 				       struct x86_exception *exception)
4466 {
4467 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4468 	void *data = val;
4469 	int r = X86EMUL_CONTINUE;
4470 
4471 	while (bytes) {
4472 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4473 							     PFERR_WRITE_MASK,
4474 							     exception);
4475 		unsigned offset = addr & (PAGE_SIZE-1);
4476 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4477 		int ret;
4478 
4479 		if (gpa == UNMAPPED_GVA)
4480 			return X86EMUL_PROPAGATE_FAULT;
4481 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4482 		if (ret < 0) {
4483 			r = X86EMUL_IO_NEEDED;
4484 			goto out;
4485 		}
4486 
4487 		bytes -= towrite;
4488 		data += towrite;
4489 		addr += towrite;
4490 	}
4491 out:
4492 	return r;
4493 }
4494 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4495 
4496 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4497 			    gpa_t gpa, bool write)
4498 {
4499 	/* For APIC access vmexit */
4500 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4501 		return 1;
4502 
4503 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4504 		trace_vcpu_match_mmio(gva, gpa, write, true);
4505 		return 1;
4506 	}
4507 
4508 	return 0;
4509 }
4510 
4511 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4512 				gpa_t *gpa, struct x86_exception *exception,
4513 				bool write)
4514 {
4515 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4516 		| (write ? PFERR_WRITE_MASK : 0);
4517 
4518 	/*
4519 	 * currently PKRU is only applied to ept enabled guest so
4520 	 * there is no pkey in EPT page table for L1 guest or EPT
4521 	 * shadow page table for L2 guest.
4522 	 */
4523 	if (vcpu_match_mmio_gva(vcpu, gva)
4524 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4525 				 vcpu->arch.access, 0, access)) {
4526 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4527 					(gva & (PAGE_SIZE - 1));
4528 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4529 		return 1;
4530 	}
4531 
4532 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4533 
4534 	if (*gpa == UNMAPPED_GVA)
4535 		return -1;
4536 
4537 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4538 }
4539 
4540 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4541 			const void *val, int bytes)
4542 {
4543 	int ret;
4544 
4545 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4546 	if (ret < 0)
4547 		return 0;
4548 	kvm_page_track_write(vcpu, gpa, val, bytes);
4549 	return 1;
4550 }
4551 
4552 struct read_write_emulator_ops {
4553 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4554 				  int bytes);
4555 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4556 				  void *val, int bytes);
4557 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4558 			       int bytes, void *val);
4559 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4560 				    void *val, int bytes);
4561 	bool write;
4562 };
4563 
4564 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4565 {
4566 	if (vcpu->mmio_read_completed) {
4567 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4568 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4569 		vcpu->mmio_read_completed = 0;
4570 		return 1;
4571 	}
4572 
4573 	return 0;
4574 }
4575 
4576 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4577 			void *val, int bytes)
4578 {
4579 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4580 }
4581 
4582 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4583 			 void *val, int bytes)
4584 {
4585 	return emulator_write_phys(vcpu, gpa, val, bytes);
4586 }
4587 
4588 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4589 {
4590 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4591 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4592 }
4593 
4594 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4595 			  void *val, int bytes)
4596 {
4597 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4598 	return X86EMUL_IO_NEEDED;
4599 }
4600 
4601 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4602 			   void *val, int bytes)
4603 {
4604 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4605 
4606 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4607 	return X86EMUL_CONTINUE;
4608 }
4609 
4610 static const struct read_write_emulator_ops read_emultor = {
4611 	.read_write_prepare = read_prepare,
4612 	.read_write_emulate = read_emulate,
4613 	.read_write_mmio = vcpu_mmio_read,
4614 	.read_write_exit_mmio = read_exit_mmio,
4615 };
4616 
4617 static const struct read_write_emulator_ops write_emultor = {
4618 	.read_write_emulate = write_emulate,
4619 	.read_write_mmio = write_mmio,
4620 	.read_write_exit_mmio = write_exit_mmio,
4621 	.write = true,
4622 };
4623 
4624 static int emulator_read_write_onepage(unsigned long addr, void *val,
4625 				       unsigned int bytes,
4626 				       struct x86_exception *exception,
4627 				       struct kvm_vcpu *vcpu,
4628 				       const struct read_write_emulator_ops *ops)
4629 {
4630 	gpa_t gpa;
4631 	int handled, ret;
4632 	bool write = ops->write;
4633 	struct kvm_mmio_fragment *frag;
4634 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4635 
4636 	/*
4637 	 * If the exit was due to a NPF we may already have a GPA.
4638 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4639 	 * Note, this cannot be used on string operations since string
4640 	 * operation using rep will only have the initial GPA from the NPF
4641 	 * occurred.
4642 	 */
4643 	if (vcpu->arch.gpa_available &&
4644 	    emulator_can_use_gpa(ctxt) &&
4645 	    vcpu_is_mmio_gpa(vcpu, addr, exception->address, write) &&
4646 	    (addr & ~PAGE_MASK) == (exception->address & ~PAGE_MASK)) {
4647 		gpa = exception->address;
4648 		goto mmio;
4649 	}
4650 
4651 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4652 
4653 	if (ret < 0)
4654 		return X86EMUL_PROPAGATE_FAULT;
4655 
4656 	/* For APIC access vmexit */
4657 	if (ret)
4658 		goto mmio;
4659 
4660 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4661 		return X86EMUL_CONTINUE;
4662 
4663 mmio:
4664 	/*
4665 	 * Is this MMIO handled locally?
4666 	 */
4667 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4668 	if (handled == bytes)
4669 		return X86EMUL_CONTINUE;
4670 
4671 	gpa += handled;
4672 	bytes -= handled;
4673 	val += handled;
4674 
4675 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4676 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4677 	frag->gpa = gpa;
4678 	frag->data = val;
4679 	frag->len = bytes;
4680 	return X86EMUL_CONTINUE;
4681 }
4682 
4683 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4684 			unsigned long addr,
4685 			void *val, unsigned int bytes,
4686 			struct x86_exception *exception,
4687 			const struct read_write_emulator_ops *ops)
4688 {
4689 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4690 	gpa_t gpa;
4691 	int rc;
4692 
4693 	if (ops->read_write_prepare &&
4694 		  ops->read_write_prepare(vcpu, val, bytes))
4695 		return X86EMUL_CONTINUE;
4696 
4697 	vcpu->mmio_nr_fragments = 0;
4698 
4699 	/* Crossing a page boundary? */
4700 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4701 		int now;
4702 
4703 		now = -addr & ~PAGE_MASK;
4704 		rc = emulator_read_write_onepage(addr, val, now, exception,
4705 						 vcpu, ops);
4706 
4707 		if (rc != X86EMUL_CONTINUE)
4708 			return rc;
4709 		addr += now;
4710 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4711 			addr = (u32)addr;
4712 		val += now;
4713 		bytes -= now;
4714 	}
4715 
4716 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4717 					 vcpu, ops);
4718 	if (rc != X86EMUL_CONTINUE)
4719 		return rc;
4720 
4721 	if (!vcpu->mmio_nr_fragments)
4722 		return rc;
4723 
4724 	gpa = vcpu->mmio_fragments[0].gpa;
4725 
4726 	vcpu->mmio_needed = 1;
4727 	vcpu->mmio_cur_fragment = 0;
4728 
4729 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4730 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4731 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4732 	vcpu->run->mmio.phys_addr = gpa;
4733 
4734 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4735 }
4736 
4737 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4738 				  unsigned long addr,
4739 				  void *val,
4740 				  unsigned int bytes,
4741 				  struct x86_exception *exception)
4742 {
4743 	return emulator_read_write(ctxt, addr, val, bytes,
4744 				   exception, &read_emultor);
4745 }
4746 
4747 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4748 			    unsigned long addr,
4749 			    const void *val,
4750 			    unsigned int bytes,
4751 			    struct x86_exception *exception)
4752 {
4753 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4754 				   exception, &write_emultor);
4755 }
4756 
4757 #define CMPXCHG_TYPE(t, ptr, old, new) \
4758 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4759 
4760 #ifdef CONFIG_X86_64
4761 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4762 #else
4763 #  define CMPXCHG64(ptr, old, new) \
4764 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4765 #endif
4766 
4767 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4768 				     unsigned long addr,
4769 				     const void *old,
4770 				     const void *new,
4771 				     unsigned int bytes,
4772 				     struct x86_exception *exception)
4773 {
4774 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4775 	gpa_t gpa;
4776 	struct page *page;
4777 	char *kaddr;
4778 	bool exchanged;
4779 
4780 	/* guests cmpxchg8b have to be emulated atomically */
4781 	if (bytes > 8 || (bytes & (bytes - 1)))
4782 		goto emul_write;
4783 
4784 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4785 
4786 	if (gpa == UNMAPPED_GVA ||
4787 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4788 		goto emul_write;
4789 
4790 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4791 		goto emul_write;
4792 
4793 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4794 	if (is_error_page(page))
4795 		goto emul_write;
4796 
4797 	kaddr = kmap_atomic(page);
4798 	kaddr += offset_in_page(gpa);
4799 	switch (bytes) {
4800 	case 1:
4801 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4802 		break;
4803 	case 2:
4804 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4805 		break;
4806 	case 4:
4807 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4808 		break;
4809 	case 8:
4810 		exchanged = CMPXCHG64(kaddr, old, new);
4811 		break;
4812 	default:
4813 		BUG();
4814 	}
4815 	kunmap_atomic(kaddr);
4816 	kvm_release_page_dirty(page);
4817 
4818 	if (!exchanged)
4819 		return X86EMUL_CMPXCHG_FAILED;
4820 
4821 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4822 	kvm_page_track_write(vcpu, gpa, new, bytes);
4823 
4824 	return X86EMUL_CONTINUE;
4825 
4826 emul_write:
4827 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4828 
4829 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4830 }
4831 
4832 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4833 {
4834 	int r = 0, i;
4835 
4836 	for (i = 0; i < vcpu->arch.pio.count; i++) {
4837 		if (vcpu->arch.pio.in)
4838 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4839 					    vcpu->arch.pio.size, pd);
4840 		else
4841 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4842 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
4843 					     pd);
4844 		if (r)
4845 			break;
4846 		pd += vcpu->arch.pio.size;
4847 	}
4848 	return r;
4849 }
4850 
4851 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4852 			       unsigned short port, void *val,
4853 			       unsigned int count, bool in)
4854 {
4855 	vcpu->arch.pio.port = port;
4856 	vcpu->arch.pio.in = in;
4857 	vcpu->arch.pio.count  = count;
4858 	vcpu->arch.pio.size = size;
4859 
4860 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4861 		vcpu->arch.pio.count = 0;
4862 		return 1;
4863 	}
4864 
4865 	vcpu->run->exit_reason = KVM_EXIT_IO;
4866 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4867 	vcpu->run->io.size = size;
4868 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4869 	vcpu->run->io.count = count;
4870 	vcpu->run->io.port = port;
4871 
4872 	return 0;
4873 }
4874 
4875 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4876 				    int size, unsigned short port, void *val,
4877 				    unsigned int count)
4878 {
4879 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4880 	int ret;
4881 
4882 	if (vcpu->arch.pio.count)
4883 		goto data_avail;
4884 
4885 	memset(vcpu->arch.pio_data, 0, size * count);
4886 
4887 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4888 	if (ret) {
4889 data_avail:
4890 		memcpy(val, vcpu->arch.pio_data, size * count);
4891 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4892 		vcpu->arch.pio.count = 0;
4893 		return 1;
4894 	}
4895 
4896 	return 0;
4897 }
4898 
4899 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4900 				     int size, unsigned short port,
4901 				     const void *val, unsigned int count)
4902 {
4903 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4904 
4905 	memcpy(vcpu->arch.pio_data, val, size * count);
4906 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4907 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4908 }
4909 
4910 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4911 {
4912 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4913 }
4914 
4915 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4916 {
4917 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4918 }
4919 
4920 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4921 {
4922 	if (!need_emulate_wbinvd(vcpu))
4923 		return X86EMUL_CONTINUE;
4924 
4925 	if (kvm_x86_ops->has_wbinvd_exit()) {
4926 		int cpu = get_cpu();
4927 
4928 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4929 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4930 				wbinvd_ipi, NULL, 1);
4931 		put_cpu();
4932 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4933 	} else
4934 		wbinvd();
4935 	return X86EMUL_CONTINUE;
4936 }
4937 
4938 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4939 {
4940 	kvm_emulate_wbinvd_noskip(vcpu);
4941 	return kvm_skip_emulated_instruction(vcpu);
4942 }
4943 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4944 
4945 
4946 
4947 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4948 {
4949 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4950 }
4951 
4952 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4953 			   unsigned long *dest)
4954 {
4955 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4956 }
4957 
4958 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4959 			   unsigned long value)
4960 {
4961 
4962 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4963 }
4964 
4965 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4966 {
4967 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4968 }
4969 
4970 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4971 {
4972 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4973 	unsigned long value;
4974 
4975 	switch (cr) {
4976 	case 0:
4977 		value = kvm_read_cr0(vcpu);
4978 		break;
4979 	case 2:
4980 		value = vcpu->arch.cr2;
4981 		break;
4982 	case 3:
4983 		value = kvm_read_cr3(vcpu);
4984 		break;
4985 	case 4:
4986 		value = kvm_read_cr4(vcpu);
4987 		break;
4988 	case 8:
4989 		value = kvm_get_cr8(vcpu);
4990 		break;
4991 	default:
4992 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4993 		return 0;
4994 	}
4995 
4996 	return value;
4997 }
4998 
4999 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5000 {
5001 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5002 	int res = 0;
5003 
5004 	switch (cr) {
5005 	case 0:
5006 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5007 		break;
5008 	case 2:
5009 		vcpu->arch.cr2 = val;
5010 		break;
5011 	case 3:
5012 		res = kvm_set_cr3(vcpu, val);
5013 		break;
5014 	case 4:
5015 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5016 		break;
5017 	case 8:
5018 		res = kvm_set_cr8(vcpu, val);
5019 		break;
5020 	default:
5021 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5022 		res = -1;
5023 	}
5024 
5025 	return res;
5026 }
5027 
5028 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5029 {
5030 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5031 }
5032 
5033 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5034 {
5035 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5036 }
5037 
5038 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5039 {
5040 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5041 }
5042 
5043 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5044 {
5045 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5046 }
5047 
5048 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5049 {
5050 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5051 }
5052 
5053 static unsigned long emulator_get_cached_segment_base(
5054 	struct x86_emulate_ctxt *ctxt, int seg)
5055 {
5056 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5057 }
5058 
5059 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5060 				 struct desc_struct *desc, u32 *base3,
5061 				 int seg)
5062 {
5063 	struct kvm_segment var;
5064 
5065 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5066 	*selector = var.selector;
5067 
5068 	if (var.unusable) {
5069 		memset(desc, 0, sizeof(*desc));
5070 		if (base3)
5071 			*base3 = 0;
5072 		return false;
5073 	}
5074 
5075 	if (var.g)
5076 		var.limit >>= 12;
5077 	set_desc_limit(desc, var.limit);
5078 	set_desc_base(desc, (unsigned long)var.base);
5079 #ifdef CONFIG_X86_64
5080 	if (base3)
5081 		*base3 = var.base >> 32;
5082 #endif
5083 	desc->type = var.type;
5084 	desc->s = var.s;
5085 	desc->dpl = var.dpl;
5086 	desc->p = var.present;
5087 	desc->avl = var.avl;
5088 	desc->l = var.l;
5089 	desc->d = var.db;
5090 	desc->g = var.g;
5091 
5092 	return true;
5093 }
5094 
5095 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5096 				 struct desc_struct *desc, u32 base3,
5097 				 int seg)
5098 {
5099 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5100 	struct kvm_segment var;
5101 
5102 	var.selector = selector;
5103 	var.base = get_desc_base(desc);
5104 #ifdef CONFIG_X86_64
5105 	var.base |= ((u64)base3) << 32;
5106 #endif
5107 	var.limit = get_desc_limit(desc);
5108 	if (desc->g)
5109 		var.limit = (var.limit << 12) | 0xfff;
5110 	var.type = desc->type;
5111 	var.dpl = desc->dpl;
5112 	var.db = desc->d;
5113 	var.s = desc->s;
5114 	var.l = desc->l;
5115 	var.g = desc->g;
5116 	var.avl = desc->avl;
5117 	var.present = desc->p;
5118 	var.unusable = !var.present;
5119 	var.padding = 0;
5120 
5121 	kvm_set_segment(vcpu, &var, seg);
5122 	return;
5123 }
5124 
5125 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5126 			    u32 msr_index, u64 *pdata)
5127 {
5128 	struct msr_data msr;
5129 	int r;
5130 
5131 	msr.index = msr_index;
5132 	msr.host_initiated = false;
5133 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5134 	if (r)
5135 		return r;
5136 
5137 	*pdata = msr.data;
5138 	return 0;
5139 }
5140 
5141 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5142 			    u32 msr_index, u64 data)
5143 {
5144 	struct msr_data msr;
5145 
5146 	msr.data = data;
5147 	msr.index = msr_index;
5148 	msr.host_initiated = false;
5149 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5150 }
5151 
5152 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5153 {
5154 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5155 
5156 	return vcpu->arch.smbase;
5157 }
5158 
5159 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5160 {
5161 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5162 
5163 	vcpu->arch.smbase = smbase;
5164 }
5165 
5166 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5167 			      u32 pmc)
5168 {
5169 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5170 }
5171 
5172 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5173 			     u32 pmc, u64 *pdata)
5174 {
5175 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5176 }
5177 
5178 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5179 {
5180 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5181 }
5182 
5183 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5184 {
5185 	preempt_disable();
5186 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5187 }
5188 
5189 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5190 {
5191 	preempt_enable();
5192 }
5193 
5194 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5195 			      struct x86_instruction_info *info,
5196 			      enum x86_intercept_stage stage)
5197 {
5198 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5199 }
5200 
5201 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5202 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5203 {
5204 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5205 }
5206 
5207 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5208 {
5209 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5210 }
5211 
5212 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5213 {
5214 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5215 }
5216 
5217 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5218 {
5219 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5220 }
5221 
5222 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5223 {
5224 	return emul_to_vcpu(ctxt)->arch.hflags;
5225 }
5226 
5227 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5228 {
5229 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5230 }
5231 
5232 static const struct x86_emulate_ops emulate_ops = {
5233 	.read_gpr            = emulator_read_gpr,
5234 	.write_gpr           = emulator_write_gpr,
5235 	.read_std            = kvm_read_guest_virt_system,
5236 	.write_std           = kvm_write_guest_virt_system,
5237 	.read_phys           = kvm_read_guest_phys_system,
5238 	.fetch               = kvm_fetch_guest_virt,
5239 	.read_emulated       = emulator_read_emulated,
5240 	.write_emulated      = emulator_write_emulated,
5241 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5242 	.invlpg              = emulator_invlpg,
5243 	.pio_in_emulated     = emulator_pio_in_emulated,
5244 	.pio_out_emulated    = emulator_pio_out_emulated,
5245 	.get_segment         = emulator_get_segment,
5246 	.set_segment         = emulator_set_segment,
5247 	.get_cached_segment_base = emulator_get_cached_segment_base,
5248 	.get_gdt             = emulator_get_gdt,
5249 	.get_idt	     = emulator_get_idt,
5250 	.set_gdt             = emulator_set_gdt,
5251 	.set_idt	     = emulator_set_idt,
5252 	.get_cr              = emulator_get_cr,
5253 	.set_cr              = emulator_set_cr,
5254 	.cpl                 = emulator_get_cpl,
5255 	.get_dr              = emulator_get_dr,
5256 	.set_dr              = emulator_set_dr,
5257 	.get_smbase          = emulator_get_smbase,
5258 	.set_smbase          = emulator_set_smbase,
5259 	.set_msr             = emulator_set_msr,
5260 	.get_msr             = emulator_get_msr,
5261 	.check_pmc	     = emulator_check_pmc,
5262 	.read_pmc            = emulator_read_pmc,
5263 	.halt                = emulator_halt,
5264 	.wbinvd              = emulator_wbinvd,
5265 	.fix_hypercall       = emulator_fix_hypercall,
5266 	.get_fpu             = emulator_get_fpu,
5267 	.put_fpu             = emulator_put_fpu,
5268 	.intercept           = emulator_intercept,
5269 	.get_cpuid           = emulator_get_cpuid,
5270 	.set_nmi_mask        = emulator_set_nmi_mask,
5271 	.get_hflags          = emulator_get_hflags,
5272 	.set_hflags          = emulator_set_hflags,
5273 };
5274 
5275 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5276 {
5277 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5278 	/*
5279 	 * an sti; sti; sequence only disable interrupts for the first
5280 	 * instruction. So, if the last instruction, be it emulated or
5281 	 * not, left the system with the INT_STI flag enabled, it
5282 	 * means that the last instruction is an sti. We should not
5283 	 * leave the flag on in this case. The same goes for mov ss
5284 	 */
5285 	if (int_shadow & mask)
5286 		mask = 0;
5287 	if (unlikely(int_shadow || mask)) {
5288 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5289 		if (!mask)
5290 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5291 	}
5292 }
5293 
5294 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5295 {
5296 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5297 	if (ctxt->exception.vector == PF_VECTOR)
5298 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5299 
5300 	if (ctxt->exception.error_code_valid)
5301 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5302 				      ctxt->exception.error_code);
5303 	else
5304 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5305 	return false;
5306 }
5307 
5308 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5309 {
5310 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5311 	int cs_db, cs_l;
5312 
5313 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5314 
5315 	ctxt->eflags = kvm_get_rflags(vcpu);
5316 	ctxt->eip = kvm_rip_read(vcpu);
5317 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5318 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5319 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5320 		     cs_db				? X86EMUL_MODE_PROT32 :
5321 							  X86EMUL_MODE_PROT16;
5322 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5323 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5324 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5325 
5326 	init_decode_cache(ctxt);
5327 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5328 }
5329 
5330 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5331 {
5332 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5333 	int ret;
5334 
5335 	init_emulate_ctxt(vcpu);
5336 
5337 	ctxt->op_bytes = 2;
5338 	ctxt->ad_bytes = 2;
5339 	ctxt->_eip = ctxt->eip + inc_eip;
5340 	ret = emulate_int_real(ctxt, irq);
5341 
5342 	if (ret != X86EMUL_CONTINUE)
5343 		return EMULATE_FAIL;
5344 
5345 	ctxt->eip = ctxt->_eip;
5346 	kvm_rip_write(vcpu, ctxt->eip);
5347 	kvm_set_rflags(vcpu, ctxt->eflags);
5348 
5349 	if (irq == NMI_VECTOR)
5350 		vcpu->arch.nmi_pending = 0;
5351 	else
5352 		vcpu->arch.interrupt.pending = false;
5353 
5354 	return EMULATE_DONE;
5355 }
5356 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5357 
5358 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5359 {
5360 	int r = EMULATE_DONE;
5361 
5362 	++vcpu->stat.insn_emulation_fail;
5363 	trace_kvm_emulate_insn_failed(vcpu);
5364 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5365 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5366 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5367 		vcpu->run->internal.ndata = 0;
5368 		r = EMULATE_FAIL;
5369 	}
5370 	kvm_queue_exception(vcpu, UD_VECTOR);
5371 
5372 	return r;
5373 }
5374 
5375 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5376 				  bool write_fault_to_shadow_pgtable,
5377 				  int emulation_type)
5378 {
5379 	gpa_t gpa = cr2;
5380 	kvm_pfn_t pfn;
5381 
5382 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5383 		return false;
5384 
5385 	if (!vcpu->arch.mmu.direct_map) {
5386 		/*
5387 		 * Write permission should be allowed since only
5388 		 * write access need to be emulated.
5389 		 */
5390 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5391 
5392 		/*
5393 		 * If the mapping is invalid in guest, let cpu retry
5394 		 * it to generate fault.
5395 		 */
5396 		if (gpa == UNMAPPED_GVA)
5397 			return true;
5398 	}
5399 
5400 	/*
5401 	 * Do not retry the unhandleable instruction if it faults on the
5402 	 * readonly host memory, otherwise it will goto a infinite loop:
5403 	 * retry instruction -> write #PF -> emulation fail -> retry
5404 	 * instruction -> ...
5405 	 */
5406 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5407 
5408 	/*
5409 	 * If the instruction failed on the error pfn, it can not be fixed,
5410 	 * report the error to userspace.
5411 	 */
5412 	if (is_error_noslot_pfn(pfn))
5413 		return false;
5414 
5415 	kvm_release_pfn_clean(pfn);
5416 
5417 	/* The instructions are well-emulated on direct mmu. */
5418 	if (vcpu->arch.mmu.direct_map) {
5419 		unsigned int indirect_shadow_pages;
5420 
5421 		spin_lock(&vcpu->kvm->mmu_lock);
5422 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5423 		spin_unlock(&vcpu->kvm->mmu_lock);
5424 
5425 		if (indirect_shadow_pages)
5426 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5427 
5428 		return true;
5429 	}
5430 
5431 	/*
5432 	 * if emulation was due to access to shadowed page table
5433 	 * and it failed try to unshadow page and re-enter the
5434 	 * guest to let CPU execute the instruction.
5435 	 */
5436 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5437 
5438 	/*
5439 	 * If the access faults on its page table, it can not
5440 	 * be fixed by unprotecting shadow page and it should
5441 	 * be reported to userspace.
5442 	 */
5443 	return !write_fault_to_shadow_pgtable;
5444 }
5445 
5446 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5447 			      unsigned long cr2,  int emulation_type)
5448 {
5449 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5450 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5451 
5452 	last_retry_eip = vcpu->arch.last_retry_eip;
5453 	last_retry_addr = vcpu->arch.last_retry_addr;
5454 
5455 	/*
5456 	 * If the emulation is caused by #PF and it is non-page_table
5457 	 * writing instruction, it means the VM-EXIT is caused by shadow
5458 	 * page protected, we can zap the shadow page and retry this
5459 	 * instruction directly.
5460 	 *
5461 	 * Note: if the guest uses a non-page-table modifying instruction
5462 	 * on the PDE that points to the instruction, then we will unmap
5463 	 * the instruction and go to an infinite loop. So, we cache the
5464 	 * last retried eip and the last fault address, if we meet the eip
5465 	 * and the address again, we can break out of the potential infinite
5466 	 * loop.
5467 	 */
5468 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5469 
5470 	if (!(emulation_type & EMULTYPE_RETRY))
5471 		return false;
5472 
5473 	if (x86_page_table_writing_insn(ctxt))
5474 		return false;
5475 
5476 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5477 		return false;
5478 
5479 	vcpu->arch.last_retry_eip = ctxt->eip;
5480 	vcpu->arch.last_retry_addr = cr2;
5481 
5482 	if (!vcpu->arch.mmu.direct_map)
5483 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5484 
5485 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5486 
5487 	return true;
5488 }
5489 
5490 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5491 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5492 
5493 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5494 {
5495 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5496 		/* This is a good place to trace that we are exiting SMM.  */
5497 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5498 
5499 		/* Process a latched INIT or SMI, if any.  */
5500 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5501 	}
5502 
5503 	kvm_mmu_reset_context(vcpu);
5504 }
5505 
5506 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5507 {
5508 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5509 
5510 	vcpu->arch.hflags = emul_flags;
5511 
5512 	if (changed & HF_SMM_MASK)
5513 		kvm_smm_changed(vcpu);
5514 }
5515 
5516 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5517 				unsigned long *db)
5518 {
5519 	u32 dr6 = 0;
5520 	int i;
5521 	u32 enable, rwlen;
5522 
5523 	enable = dr7;
5524 	rwlen = dr7 >> 16;
5525 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5526 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5527 			dr6 |= (1 << i);
5528 	return dr6;
5529 }
5530 
5531 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5532 {
5533 	struct kvm_run *kvm_run = vcpu->run;
5534 
5535 	/*
5536 	 * rflags is the old, "raw" value of the flags.  The new value has
5537 	 * not been saved yet.
5538 	 *
5539 	 * This is correct even for TF set by the guest, because "the
5540 	 * processor will not generate this exception after the instruction
5541 	 * that sets the TF flag".
5542 	 */
5543 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5544 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5545 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5546 						  DR6_RTM;
5547 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5548 			kvm_run->debug.arch.exception = DB_VECTOR;
5549 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5550 			*r = EMULATE_USER_EXIT;
5551 		} else {
5552 			/*
5553 			 * "Certain debug exceptions may clear bit 0-3.  The
5554 			 * remaining contents of the DR6 register are never
5555 			 * cleared by the processor".
5556 			 */
5557 			vcpu->arch.dr6 &= ~15;
5558 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5559 			kvm_queue_exception(vcpu, DB_VECTOR);
5560 		}
5561 	}
5562 }
5563 
5564 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5565 {
5566 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5567 	int r = EMULATE_DONE;
5568 
5569 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5570 	kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5571 	return r == EMULATE_DONE;
5572 }
5573 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5574 
5575 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5576 {
5577 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5578 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5579 		struct kvm_run *kvm_run = vcpu->run;
5580 		unsigned long eip = kvm_get_linear_rip(vcpu);
5581 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5582 					   vcpu->arch.guest_debug_dr7,
5583 					   vcpu->arch.eff_db);
5584 
5585 		if (dr6 != 0) {
5586 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5587 			kvm_run->debug.arch.pc = eip;
5588 			kvm_run->debug.arch.exception = DB_VECTOR;
5589 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5590 			*r = EMULATE_USER_EXIT;
5591 			return true;
5592 		}
5593 	}
5594 
5595 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5596 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5597 		unsigned long eip = kvm_get_linear_rip(vcpu);
5598 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5599 					   vcpu->arch.dr7,
5600 					   vcpu->arch.db);
5601 
5602 		if (dr6 != 0) {
5603 			vcpu->arch.dr6 &= ~15;
5604 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5605 			kvm_queue_exception(vcpu, DB_VECTOR);
5606 			*r = EMULATE_DONE;
5607 			return true;
5608 		}
5609 	}
5610 
5611 	return false;
5612 }
5613 
5614 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5615 			    unsigned long cr2,
5616 			    int emulation_type,
5617 			    void *insn,
5618 			    int insn_len)
5619 {
5620 	int r;
5621 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5622 	bool writeback = true;
5623 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5624 
5625 	/*
5626 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5627 	 * never reused.
5628 	 */
5629 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5630 	kvm_clear_exception_queue(vcpu);
5631 
5632 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5633 		init_emulate_ctxt(vcpu);
5634 
5635 		/*
5636 		 * We will reenter on the same instruction since
5637 		 * we do not set complete_userspace_io.  This does not
5638 		 * handle watchpoints yet, those would be handled in
5639 		 * the emulate_ops.
5640 		 */
5641 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5642 			return r;
5643 
5644 		ctxt->interruptibility = 0;
5645 		ctxt->have_exception = false;
5646 		ctxt->exception.vector = -1;
5647 		ctxt->perm_ok = false;
5648 
5649 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5650 
5651 		r = x86_decode_insn(ctxt, insn, insn_len);
5652 
5653 		trace_kvm_emulate_insn_start(vcpu);
5654 		++vcpu->stat.insn_emulation;
5655 		if (r != EMULATION_OK)  {
5656 			if (emulation_type & EMULTYPE_TRAP_UD)
5657 				return EMULATE_FAIL;
5658 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5659 						emulation_type))
5660 				return EMULATE_DONE;
5661 			if (emulation_type & EMULTYPE_SKIP)
5662 				return EMULATE_FAIL;
5663 			return handle_emulation_failure(vcpu);
5664 		}
5665 	}
5666 
5667 	if (emulation_type & EMULTYPE_SKIP) {
5668 		kvm_rip_write(vcpu, ctxt->_eip);
5669 		if (ctxt->eflags & X86_EFLAGS_RF)
5670 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5671 		return EMULATE_DONE;
5672 	}
5673 
5674 	if (retry_instruction(ctxt, cr2, emulation_type))
5675 		return EMULATE_DONE;
5676 
5677 	/* this is needed for vmware backdoor interface to work since it
5678 	   changes registers values  during IO operation */
5679 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5680 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5681 		emulator_invalidate_register_cache(ctxt);
5682 	}
5683 
5684 restart:
5685 	/* Save the faulting GPA (cr2) in the address field */
5686 	ctxt->exception.address = cr2;
5687 
5688 	r = x86_emulate_insn(ctxt);
5689 
5690 	if (r == EMULATION_INTERCEPTED)
5691 		return EMULATE_DONE;
5692 
5693 	if (r == EMULATION_FAILED) {
5694 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5695 					emulation_type))
5696 			return EMULATE_DONE;
5697 
5698 		return handle_emulation_failure(vcpu);
5699 	}
5700 
5701 	if (ctxt->have_exception) {
5702 		r = EMULATE_DONE;
5703 		if (inject_emulated_exception(vcpu))
5704 			return r;
5705 	} else if (vcpu->arch.pio.count) {
5706 		if (!vcpu->arch.pio.in) {
5707 			/* FIXME: return into emulator if single-stepping.  */
5708 			vcpu->arch.pio.count = 0;
5709 		} else {
5710 			writeback = false;
5711 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5712 		}
5713 		r = EMULATE_USER_EXIT;
5714 	} else if (vcpu->mmio_needed) {
5715 		if (!vcpu->mmio_is_write)
5716 			writeback = false;
5717 		r = EMULATE_USER_EXIT;
5718 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5719 	} else if (r == EMULATION_RESTART)
5720 		goto restart;
5721 	else
5722 		r = EMULATE_DONE;
5723 
5724 	if (writeback) {
5725 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5726 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5727 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5728 		kvm_rip_write(vcpu, ctxt->eip);
5729 		if (r == EMULATE_DONE)
5730 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5731 		if (!ctxt->have_exception ||
5732 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5733 			__kvm_set_rflags(vcpu, ctxt->eflags);
5734 
5735 		/*
5736 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5737 		 * do nothing, and it will be requested again as soon as
5738 		 * the shadow expires.  But we still need to check here,
5739 		 * because POPF has no interrupt shadow.
5740 		 */
5741 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5742 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5743 	} else
5744 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5745 
5746 	return r;
5747 }
5748 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5749 
5750 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5751 {
5752 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5753 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5754 					    size, port, &val, 1);
5755 	/* do not return to emulator after return from userspace */
5756 	vcpu->arch.pio.count = 0;
5757 	return ret;
5758 }
5759 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5760 
5761 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5762 {
5763 	unsigned long val;
5764 
5765 	/* We should only ever be called with arch.pio.count equal to 1 */
5766 	BUG_ON(vcpu->arch.pio.count != 1);
5767 
5768 	/* For size less than 4 we merge, else we zero extend */
5769 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5770 					: 0;
5771 
5772 	/*
5773 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5774 	 * the copy and tracing
5775 	 */
5776 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5777 				 vcpu->arch.pio.port, &val, 1);
5778 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5779 
5780 	return 1;
5781 }
5782 
5783 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5784 {
5785 	unsigned long val;
5786 	int ret;
5787 
5788 	/* For size less than 4 we merge, else we zero extend */
5789 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5790 
5791 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5792 				       &val, 1);
5793 	if (ret) {
5794 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5795 		return ret;
5796 	}
5797 
5798 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5799 
5800 	return 0;
5801 }
5802 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5803 
5804 static int kvmclock_cpu_down_prep(unsigned int cpu)
5805 {
5806 	__this_cpu_write(cpu_tsc_khz, 0);
5807 	return 0;
5808 }
5809 
5810 static void tsc_khz_changed(void *data)
5811 {
5812 	struct cpufreq_freqs *freq = data;
5813 	unsigned long khz = 0;
5814 
5815 	if (data)
5816 		khz = freq->new;
5817 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5818 		khz = cpufreq_quick_get(raw_smp_processor_id());
5819 	if (!khz)
5820 		khz = tsc_khz;
5821 	__this_cpu_write(cpu_tsc_khz, khz);
5822 }
5823 
5824 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5825 				     void *data)
5826 {
5827 	struct cpufreq_freqs *freq = data;
5828 	struct kvm *kvm;
5829 	struct kvm_vcpu *vcpu;
5830 	int i, send_ipi = 0;
5831 
5832 	/*
5833 	 * We allow guests to temporarily run on slowing clocks,
5834 	 * provided we notify them after, or to run on accelerating
5835 	 * clocks, provided we notify them before.  Thus time never
5836 	 * goes backwards.
5837 	 *
5838 	 * However, we have a problem.  We can't atomically update
5839 	 * the frequency of a given CPU from this function; it is
5840 	 * merely a notifier, which can be called from any CPU.
5841 	 * Changing the TSC frequency at arbitrary points in time
5842 	 * requires a recomputation of local variables related to
5843 	 * the TSC for each VCPU.  We must flag these local variables
5844 	 * to be updated and be sure the update takes place with the
5845 	 * new frequency before any guests proceed.
5846 	 *
5847 	 * Unfortunately, the combination of hotplug CPU and frequency
5848 	 * change creates an intractable locking scenario; the order
5849 	 * of when these callouts happen is undefined with respect to
5850 	 * CPU hotplug, and they can race with each other.  As such,
5851 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5852 	 * undefined; you can actually have a CPU frequency change take
5853 	 * place in between the computation of X and the setting of the
5854 	 * variable.  To protect against this problem, all updates of
5855 	 * the per_cpu tsc_khz variable are done in an interrupt
5856 	 * protected IPI, and all callers wishing to update the value
5857 	 * must wait for a synchronous IPI to complete (which is trivial
5858 	 * if the caller is on the CPU already).  This establishes the
5859 	 * necessary total order on variable updates.
5860 	 *
5861 	 * Note that because a guest time update may take place
5862 	 * anytime after the setting of the VCPU's request bit, the
5863 	 * correct TSC value must be set before the request.  However,
5864 	 * to ensure the update actually makes it to any guest which
5865 	 * starts running in hardware virtualization between the set
5866 	 * and the acquisition of the spinlock, we must also ping the
5867 	 * CPU after setting the request bit.
5868 	 *
5869 	 */
5870 
5871 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5872 		return 0;
5873 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5874 		return 0;
5875 
5876 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5877 
5878 	spin_lock(&kvm_lock);
5879 	list_for_each_entry(kvm, &vm_list, vm_list) {
5880 		kvm_for_each_vcpu(i, vcpu, kvm) {
5881 			if (vcpu->cpu != freq->cpu)
5882 				continue;
5883 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5884 			if (vcpu->cpu != smp_processor_id())
5885 				send_ipi = 1;
5886 		}
5887 	}
5888 	spin_unlock(&kvm_lock);
5889 
5890 	if (freq->old < freq->new && send_ipi) {
5891 		/*
5892 		 * We upscale the frequency.  Must make the guest
5893 		 * doesn't see old kvmclock values while running with
5894 		 * the new frequency, otherwise we risk the guest sees
5895 		 * time go backwards.
5896 		 *
5897 		 * In case we update the frequency for another cpu
5898 		 * (which might be in guest context) send an interrupt
5899 		 * to kick the cpu out of guest context.  Next time
5900 		 * guest context is entered kvmclock will be updated,
5901 		 * so the guest will not see stale values.
5902 		 */
5903 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5904 	}
5905 	return 0;
5906 }
5907 
5908 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5909 	.notifier_call  = kvmclock_cpufreq_notifier
5910 };
5911 
5912 static int kvmclock_cpu_online(unsigned int cpu)
5913 {
5914 	tsc_khz_changed(NULL);
5915 	return 0;
5916 }
5917 
5918 static void kvm_timer_init(void)
5919 {
5920 	max_tsc_khz = tsc_khz;
5921 
5922 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5923 #ifdef CONFIG_CPU_FREQ
5924 		struct cpufreq_policy policy;
5925 		int cpu;
5926 
5927 		memset(&policy, 0, sizeof(policy));
5928 		cpu = get_cpu();
5929 		cpufreq_get_policy(&policy, cpu);
5930 		if (policy.cpuinfo.max_freq)
5931 			max_tsc_khz = policy.cpuinfo.max_freq;
5932 		put_cpu();
5933 #endif
5934 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5935 					  CPUFREQ_TRANSITION_NOTIFIER);
5936 	}
5937 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5938 
5939 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5940 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
5941 }
5942 
5943 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5944 
5945 int kvm_is_in_guest(void)
5946 {
5947 	return __this_cpu_read(current_vcpu) != NULL;
5948 }
5949 
5950 static int kvm_is_user_mode(void)
5951 {
5952 	int user_mode = 3;
5953 
5954 	if (__this_cpu_read(current_vcpu))
5955 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5956 
5957 	return user_mode != 0;
5958 }
5959 
5960 static unsigned long kvm_get_guest_ip(void)
5961 {
5962 	unsigned long ip = 0;
5963 
5964 	if (__this_cpu_read(current_vcpu))
5965 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5966 
5967 	return ip;
5968 }
5969 
5970 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5971 	.is_in_guest		= kvm_is_in_guest,
5972 	.is_user_mode		= kvm_is_user_mode,
5973 	.get_guest_ip		= kvm_get_guest_ip,
5974 };
5975 
5976 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5977 {
5978 	__this_cpu_write(current_vcpu, vcpu);
5979 }
5980 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5981 
5982 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5983 {
5984 	__this_cpu_write(current_vcpu, NULL);
5985 }
5986 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5987 
5988 static void kvm_set_mmio_spte_mask(void)
5989 {
5990 	u64 mask;
5991 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5992 
5993 	/*
5994 	 * Set the reserved bits and the present bit of an paging-structure
5995 	 * entry to generate page fault with PFER.RSV = 1.
5996 	 */
5997 	 /* Mask the reserved physical address bits. */
5998 	mask = rsvd_bits(maxphyaddr, 51);
5999 
6000 	/* Set the present bit. */
6001 	mask |= 1ull;
6002 
6003 #ifdef CONFIG_X86_64
6004 	/*
6005 	 * If reserved bit is not supported, clear the present bit to disable
6006 	 * mmio page fault.
6007 	 */
6008 	if (maxphyaddr == 52)
6009 		mask &= ~1ull;
6010 #endif
6011 
6012 	kvm_mmu_set_mmio_spte_mask(mask);
6013 }
6014 
6015 #ifdef CONFIG_X86_64
6016 static void pvclock_gtod_update_fn(struct work_struct *work)
6017 {
6018 	struct kvm *kvm;
6019 
6020 	struct kvm_vcpu *vcpu;
6021 	int i;
6022 
6023 	spin_lock(&kvm_lock);
6024 	list_for_each_entry(kvm, &vm_list, vm_list)
6025 		kvm_for_each_vcpu(i, vcpu, kvm)
6026 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6027 	atomic_set(&kvm_guest_has_master_clock, 0);
6028 	spin_unlock(&kvm_lock);
6029 }
6030 
6031 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6032 
6033 /*
6034  * Notification about pvclock gtod data update.
6035  */
6036 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6037 			       void *priv)
6038 {
6039 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6040 	struct timekeeper *tk = priv;
6041 
6042 	update_pvclock_gtod(tk);
6043 
6044 	/* disable master clock if host does not trust, or does not
6045 	 * use, TSC clocksource
6046 	 */
6047 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6048 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6049 		queue_work(system_long_wq, &pvclock_gtod_work);
6050 
6051 	return 0;
6052 }
6053 
6054 static struct notifier_block pvclock_gtod_notifier = {
6055 	.notifier_call = pvclock_gtod_notify,
6056 };
6057 #endif
6058 
6059 int kvm_arch_init(void *opaque)
6060 {
6061 	int r;
6062 	struct kvm_x86_ops *ops = opaque;
6063 
6064 	if (kvm_x86_ops) {
6065 		printk(KERN_ERR "kvm: already loaded the other module\n");
6066 		r = -EEXIST;
6067 		goto out;
6068 	}
6069 
6070 	if (!ops->cpu_has_kvm_support()) {
6071 		printk(KERN_ERR "kvm: no hardware support\n");
6072 		r = -EOPNOTSUPP;
6073 		goto out;
6074 	}
6075 	if (ops->disabled_by_bios()) {
6076 		printk(KERN_ERR "kvm: disabled by bios\n");
6077 		r = -EOPNOTSUPP;
6078 		goto out;
6079 	}
6080 
6081 	r = -ENOMEM;
6082 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6083 	if (!shared_msrs) {
6084 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6085 		goto out;
6086 	}
6087 
6088 	r = kvm_mmu_module_init();
6089 	if (r)
6090 		goto out_free_percpu;
6091 
6092 	kvm_set_mmio_spte_mask();
6093 
6094 	kvm_x86_ops = ops;
6095 
6096 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6097 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6098 			PT_PRESENT_MASK, 0);
6099 	kvm_timer_init();
6100 
6101 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6102 
6103 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6104 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6105 
6106 	kvm_lapic_init();
6107 #ifdef CONFIG_X86_64
6108 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6109 #endif
6110 
6111 	return 0;
6112 
6113 out_free_percpu:
6114 	free_percpu(shared_msrs);
6115 out:
6116 	return r;
6117 }
6118 
6119 void kvm_arch_exit(void)
6120 {
6121 	kvm_lapic_exit();
6122 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6123 
6124 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6125 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6126 					    CPUFREQ_TRANSITION_NOTIFIER);
6127 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6128 #ifdef CONFIG_X86_64
6129 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6130 #endif
6131 	kvm_x86_ops = NULL;
6132 	kvm_mmu_module_exit();
6133 	free_percpu(shared_msrs);
6134 }
6135 
6136 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6137 {
6138 	++vcpu->stat.halt_exits;
6139 	if (lapic_in_kernel(vcpu)) {
6140 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6141 		return 1;
6142 	} else {
6143 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6144 		return 0;
6145 	}
6146 }
6147 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6148 
6149 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6150 {
6151 	int ret = kvm_skip_emulated_instruction(vcpu);
6152 	/*
6153 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6154 	 * KVM_EXIT_DEBUG here.
6155 	 */
6156 	return kvm_vcpu_halt(vcpu) && ret;
6157 }
6158 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6159 
6160 #ifdef CONFIG_X86_64
6161 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6162 			        unsigned long clock_type)
6163 {
6164 	struct kvm_clock_pairing clock_pairing;
6165 	struct timespec ts;
6166 	u64 cycle;
6167 	int ret;
6168 
6169 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6170 		return -KVM_EOPNOTSUPP;
6171 
6172 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6173 		return -KVM_EOPNOTSUPP;
6174 
6175 	clock_pairing.sec = ts.tv_sec;
6176 	clock_pairing.nsec = ts.tv_nsec;
6177 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6178 	clock_pairing.flags = 0;
6179 
6180 	ret = 0;
6181 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6182 			    sizeof(struct kvm_clock_pairing)))
6183 		ret = -KVM_EFAULT;
6184 
6185 	return ret;
6186 }
6187 #endif
6188 
6189 /*
6190  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6191  *
6192  * @apicid - apicid of vcpu to be kicked.
6193  */
6194 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6195 {
6196 	struct kvm_lapic_irq lapic_irq;
6197 
6198 	lapic_irq.shorthand = 0;
6199 	lapic_irq.dest_mode = 0;
6200 	lapic_irq.dest_id = apicid;
6201 	lapic_irq.msi_redir_hint = false;
6202 
6203 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6204 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6205 }
6206 
6207 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6208 {
6209 	vcpu->arch.apicv_active = false;
6210 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6211 }
6212 
6213 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6214 {
6215 	unsigned long nr, a0, a1, a2, a3, ret;
6216 	int op_64_bit, r;
6217 
6218 	r = kvm_skip_emulated_instruction(vcpu);
6219 
6220 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6221 		return kvm_hv_hypercall(vcpu);
6222 
6223 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6224 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6225 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6226 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6227 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6228 
6229 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6230 
6231 	op_64_bit = is_64_bit_mode(vcpu);
6232 	if (!op_64_bit) {
6233 		nr &= 0xFFFFFFFF;
6234 		a0 &= 0xFFFFFFFF;
6235 		a1 &= 0xFFFFFFFF;
6236 		a2 &= 0xFFFFFFFF;
6237 		a3 &= 0xFFFFFFFF;
6238 	}
6239 
6240 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6241 		ret = -KVM_EPERM;
6242 		goto out;
6243 	}
6244 
6245 	switch (nr) {
6246 	case KVM_HC_VAPIC_POLL_IRQ:
6247 		ret = 0;
6248 		break;
6249 	case KVM_HC_KICK_CPU:
6250 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6251 		ret = 0;
6252 		break;
6253 #ifdef CONFIG_X86_64
6254 	case KVM_HC_CLOCK_PAIRING:
6255 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6256 		break;
6257 #endif
6258 	default:
6259 		ret = -KVM_ENOSYS;
6260 		break;
6261 	}
6262 out:
6263 	if (!op_64_bit)
6264 		ret = (u32)ret;
6265 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6266 	++vcpu->stat.hypercalls;
6267 	return r;
6268 }
6269 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6270 
6271 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6272 {
6273 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6274 	char instruction[3];
6275 	unsigned long rip = kvm_rip_read(vcpu);
6276 
6277 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6278 
6279 	return emulator_write_emulated(ctxt, rip, instruction, 3,
6280 		&ctxt->exception);
6281 }
6282 
6283 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6284 {
6285 	return vcpu->run->request_interrupt_window &&
6286 		likely(!pic_in_kernel(vcpu->kvm));
6287 }
6288 
6289 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6290 {
6291 	struct kvm_run *kvm_run = vcpu->run;
6292 
6293 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6294 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6295 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6296 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6297 	kvm_run->ready_for_interrupt_injection =
6298 		pic_in_kernel(vcpu->kvm) ||
6299 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6300 }
6301 
6302 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6303 {
6304 	int max_irr, tpr;
6305 
6306 	if (!kvm_x86_ops->update_cr8_intercept)
6307 		return;
6308 
6309 	if (!lapic_in_kernel(vcpu))
6310 		return;
6311 
6312 	if (vcpu->arch.apicv_active)
6313 		return;
6314 
6315 	if (!vcpu->arch.apic->vapic_addr)
6316 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6317 	else
6318 		max_irr = -1;
6319 
6320 	if (max_irr != -1)
6321 		max_irr >>= 4;
6322 
6323 	tpr = kvm_lapic_get_cr8(vcpu);
6324 
6325 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6326 }
6327 
6328 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6329 {
6330 	int r;
6331 
6332 	/* try to reinject previous events if any */
6333 	if (vcpu->arch.exception.pending) {
6334 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6335 					vcpu->arch.exception.has_error_code,
6336 					vcpu->arch.exception.error_code);
6337 
6338 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6339 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6340 					     X86_EFLAGS_RF);
6341 
6342 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6343 		    (vcpu->arch.dr7 & DR7_GD)) {
6344 			vcpu->arch.dr7 &= ~DR7_GD;
6345 			kvm_update_dr7(vcpu);
6346 		}
6347 
6348 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6349 					  vcpu->arch.exception.has_error_code,
6350 					  vcpu->arch.exception.error_code,
6351 					  vcpu->arch.exception.reinject);
6352 		return 0;
6353 	}
6354 
6355 	if (vcpu->arch.nmi_injected) {
6356 		kvm_x86_ops->set_nmi(vcpu);
6357 		return 0;
6358 	}
6359 
6360 	if (vcpu->arch.interrupt.pending) {
6361 		kvm_x86_ops->set_irq(vcpu);
6362 		return 0;
6363 	}
6364 
6365 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6366 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6367 		if (r != 0)
6368 			return r;
6369 	}
6370 
6371 	/* try to inject new event if pending */
6372 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6373 		vcpu->arch.smi_pending = false;
6374 		enter_smm(vcpu);
6375 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6376 		--vcpu->arch.nmi_pending;
6377 		vcpu->arch.nmi_injected = true;
6378 		kvm_x86_ops->set_nmi(vcpu);
6379 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6380 		/*
6381 		 * Because interrupts can be injected asynchronously, we are
6382 		 * calling check_nested_events again here to avoid a race condition.
6383 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6384 		 * proposal and current concerns.  Perhaps we should be setting
6385 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6386 		 */
6387 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6388 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6389 			if (r != 0)
6390 				return r;
6391 		}
6392 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6393 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6394 					    false);
6395 			kvm_x86_ops->set_irq(vcpu);
6396 		}
6397 	}
6398 
6399 	return 0;
6400 }
6401 
6402 static void process_nmi(struct kvm_vcpu *vcpu)
6403 {
6404 	unsigned limit = 2;
6405 
6406 	/*
6407 	 * x86 is limited to one NMI running, and one NMI pending after it.
6408 	 * If an NMI is already in progress, limit further NMIs to just one.
6409 	 * Otherwise, allow two (and we'll inject the first one immediately).
6410 	 */
6411 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6412 		limit = 1;
6413 
6414 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6415 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6416 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6417 }
6418 
6419 #define put_smstate(type, buf, offset, val)			  \
6420 	*(type *)((buf) + (offset) - 0x7e00) = val
6421 
6422 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6423 {
6424 	u32 flags = 0;
6425 	flags |= seg->g       << 23;
6426 	flags |= seg->db      << 22;
6427 	flags |= seg->l       << 21;
6428 	flags |= seg->avl     << 20;
6429 	flags |= seg->present << 15;
6430 	flags |= seg->dpl     << 13;
6431 	flags |= seg->s       << 12;
6432 	flags |= seg->type    << 8;
6433 	return flags;
6434 }
6435 
6436 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6437 {
6438 	struct kvm_segment seg;
6439 	int offset;
6440 
6441 	kvm_get_segment(vcpu, &seg, n);
6442 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6443 
6444 	if (n < 3)
6445 		offset = 0x7f84 + n * 12;
6446 	else
6447 		offset = 0x7f2c + (n - 3) * 12;
6448 
6449 	put_smstate(u32, buf, offset + 8, seg.base);
6450 	put_smstate(u32, buf, offset + 4, seg.limit);
6451 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6452 }
6453 
6454 #ifdef CONFIG_X86_64
6455 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6456 {
6457 	struct kvm_segment seg;
6458 	int offset;
6459 	u16 flags;
6460 
6461 	kvm_get_segment(vcpu, &seg, n);
6462 	offset = 0x7e00 + n * 16;
6463 
6464 	flags = enter_smm_get_segment_flags(&seg) >> 8;
6465 	put_smstate(u16, buf, offset, seg.selector);
6466 	put_smstate(u16, buf, offset + 2, flags);
6467 	put_smstate(u32, buf, offset + 4, seg.limit);
6468 	put_smstate(u64, buf, offset + 8, seg.base);
6469 }
6470 #endif
6471 
6472 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6473 {
6474 	struct desc_ptr dt;
6475 	struct kvm_segment seg;
6476 	unsigned long val;
6477 	int i;
6478 
6479 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6480 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6481 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6482 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6483 
6484 	for (i = 0; i < 8; i++)
6485 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6486 
6487 	kvm_get_dr(vcpu, 6, &val);
6488 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6489 	kvm_get_dr(vcpu, 7, &val);
6490 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6491 
6492 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6493 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6494 	put_smstate(u32, buf, 0x7f64, seg.base);
6495 	put_smstate(u32, buf, 0x7f60, seg.limit);
6496 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6497 
6498 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6499 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6500 	put_smstate(u32, buf, 0x7f80, seg.base);
6501 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6502 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6503 
6504 	kvm_x86_ops->get_gdt(vcpu, &dt);
6505 	put_smstate(u32, buf, 0x7f74, dt.address);
6506 	put_smstate(u32, buf, 0x7f70, dt.size);
6507 
6508 	kvm_x86_ops->get_idt(vcpu, &dt);
6509 	put_smstate(u32, buf, 0x7f58, dt.address);
6510 	put_smstate(u32, buf, 0x7f54, dt.size);
6511 
6512 	for (i = 0; i < 6; i++)
6513 		enter_smm_save_seg_32(vcpu, buf, i);
6514 
6515 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6516 
6517 	/* revision id */
6518 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6519 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6520 }
6521 
6522 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6523 {
6524 #ifdef CONFIG_X86_64
6525 	struct desc_ptr dt;
6526 	struct kvm_segment seg;
6527 	unsigned long val;
6528 	int i;
6529 
6530 	for (i = 0; i < 16; i++)
6531 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6532 
6533 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6534 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6535 
6536 	kvm_get_dr(vcpu, 6, &val);
6537 	put_smstate(u64, buf, 0x7f68, val);
6538 	kvm_get_dr(vcpu, 7, &val);
6539 	put_smstate(u64, buf, 0x7f60, val);
6540 
6541 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6542 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6543 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6544 
6545 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6546 
6547 	/* revision id */
6548 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6549 
6550 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6551 
6552 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6553 	put_smstate(u16, buf, 0x7e90, seg.selector);
6554 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6555 	put_smstate(u32, buf, 0x7e94, seg.limit);
6556 	put_smstate(u64, buf, 0x7e98, seg.base);
6557 
6558 	kvm_x86_ops->get_idt(vcpu, &dt);
6559 	put_smstate(u32, buf, 0x7e84, dt.size);
6560 	put_smstate(u64, buf, 0x7e88, dt.address);
6561 
6562 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6563 	put_smstate(u16, buf, 0x7e70, seg.selector);
6564 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6565 	put_smstate(u32, buf, 0x7e74, seg.limit);
6566 	put_smstate(u64, buf, 0x7e78, seg.base);
6567 
6568 	kvm_x86_ops->get_gdt(vcpu, &dt);
6569 	put_smstate(u32, buf, 0x7e64, dt.size);
6570 	put_smstate(u64, buf, 0x7e68, dt.address);
6571 
6572 	for (i = 0; i < 6; i++)
6573 		enter_smm_save_seg_64(vcpu, buf, i);
6574 #else
6575 	WARN_ON_ONCE(1);
6576 #endif
6577 }
6578 
6579 static void enter_smm(struct kvm_vcpu *vcpu)
6580 {
6581 	struct kvm_segment cs, ds;
6582 	struct desc_ptr dt;
6583 	char buf[512];
6584 	u32 cr0;
6585 
6586 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6587 	vcpu->arch.hflags |= HF_SMM_MASK;
6588 	memset(buf, 0, 512);
6589 	if (guest_cpuid_has_longmode(vcpu))
6590 		enter_smm_save_state_64(vcpu, buf);
6591 	else
6592 		enter_smm_save_state_32(vcpu, buf);
6593 
6594 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6595 
6596 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6597 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6598 	else
6599 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6600 
6601 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6602 	kvm_rip_write(vcpu, 0x8000);
6603 
6604 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6605 	kvm_x86_ops->set_cr0(vcpu, cr0);
6606 	vcpu->arch.cr0 = cr0;
6607 
6608 	kvm_x86_ops->set_cr4(vcpu, 0);
6609 
6610 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6611 	dt.address = dt.size = 0;
6612 	kvm_x86_ops->set_idt(vcpu, &dt);
6613 
6614 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6615 
6616 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6617 	cs.base = vcpu->arch.smbase;
6618 
6619 	ds.selector = 0;
6620 	ds.base = 0;
6621 
6622 	cs.limit    = ds.limit = 0xffffffff;
6623 	cs.type     = ds.type = 0x3;
6624 	cs.dpl      = ds.dpl = 0;
6625 	cs.db       = ds.db = 0;
6626 	cs.s        = ds.s = 1;
6627 	cs.l        = ds.l = 0;
6628 	cs.g        = ds.g = 1;
6629 	cs.avl      = ds.avl = 0;
6630 	cs.present  = ds.present = 1;
6631 	cs.unusable = ds.unusable = 0;
6632 	cs.padding  = ds.padding = 0;
6633 
6634 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6635 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6636 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6637 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6638 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6639 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6640 
6641 	if (guest_cpuid_has_longmode(vcpu))
6642 		kvm_x86_ops->set_efer(vcpu, 0);
6643 
6644 	kvm_update_cpuid(vcpu);
6645 	kvm_mmu_reset_context(vcpu);
6646 }
6647 
6648 static void process_smi(struct kvm_vcpu *vcpu)
6649 {
6650 	vcpu->arch.smi_pending = true;
6651 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6652 }
6653 
6654 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6655 {
6656 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6657 }
6658 
6659 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6660 {
6661 	u64 eoi_exit_bitmap[4];
6662 
6663 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6664 		return;
6665 
6666 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6667 
6668 	if (irqchip_split(vcpu->kvm))
6669 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6670 	else {
6671 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6672 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6673 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6674 	}
6675 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6676 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
6677 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6678 }
6679 
6680 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6681 {
6682 	++vcpu->stat.tlb_flush;
6683 	kvm_x86_ops->tlb_flush(vcpu);
6684 }
6685 
6686 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6687 {
6688 	struct page *page = NULL;
6689 
6690 	if (!lapic_in_kernel(vcpu))
6691 		return;
6692 
6693 	if (!kvm_x86_ops->set_apic_access_page_addr)
6694 		return;
6695 
6696 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6697 	if (is_error_page(page))
6698 		return;
6699 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6700 
6701 	/*
6702 	 * Do not pin apic access page in memory, the MMU notifier
6703 	 * will call us again if it is migrated or swapped out.
6704 	 */
6705 	put_page(page);
6706 }
6707 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6708 
6709 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6710 					   unsigned long address)
6711 {
6712 	/*
6713 	 * The physical address of apic access page is stored in the VMCS.
6714 	 * Update it when it becomes invalid.
6715 	 */
6716 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6717 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6718 }
6719 
6720 /*
6721  * Returns 1 to let vcpu_run() continue the guest execution loop without
6722  * exiting to the userspace.  Otherwise, the value will be returned to the
6723  * userspace.
6724  */
6725 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6726 {
6727 	int r;
6728 	bool req_int_win =
6729 		dm_request_for_irq_injection(vcpu) &&
6730 		kvm_cpu_accept_dm_intr(vcpu);
6731 
6732 	bool req_immediate_exit = false;
6733 
6734 	if (vcpu->requests) {
6735 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6736 			kvm_mmu_unload(vcpu);
6737 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6738 			__kvm_migrate_timers(vcpu);
6739 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6740 			kvm_gen_update_masterclock(vcpu->kvm);
6741 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6742 			kvm_gen_kvmclock_update(vcpu);
6743 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6744 			r = kvm_guest_time_update(vcpu);
6745 			if (unlikely(r))
6746 				goto out;
6747 		}
6748 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6749 			kvm_mmu_sync_roots(vcpu);
6750 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6751 			kvm_vcpu_flush_tlb(vcpu);
6752 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6753 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6754 			r = 0;
6755 			goto out;
6756 		}
6757 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6758 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6759 			r = 0;
6760 			goto out;
6761 		}
6762 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6763 			/* Page is swapped out. Do synthetic halt */
6764 			vcpu->arch.apf.halted = true;
6765 			r = 1;
6766 			goto out;
6767 		}
6768 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6769 			record_steal_time(vcpu);
6770 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6771 			process_smi(vcpu);
6772 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6773 			process_nmi(vcpu);
6774 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6775 			kvm_pmu_handle_event(vcpu);
6776 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6777 			kvm_pmu_deliver_pmi(vcpu);
6778 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6779 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6780 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
6781 				     vcpu->arch.ioapic_handled_vectors)) {
6782 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6783 				vcpu->run->eoi.vector =
6784 						vcpu->arch.pending_ioapic_eoi;
6785 				r = 0;
6786 				goto out;
6787 			}
6788 		}
6789 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6790 			vcpu_scan_ioapic(vcpu);
6791 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6792 			kvm_vcpu_reload_apic_access_page(vcpu);
6793 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6794 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6795 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6796 			r = 0;
6797 			goto out;
6798 		}
6799 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6800 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6801 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6802 			r = 0;
6803 			goto out;
6804 		}
6805 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6806 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6807 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6808 			r = 0;
6809 			goto out;
6810 		}
6811 
6812 		/*
6813 		 * KVM_REQ_HV_STIMER has to be processed after
6814 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6815 		 * depend on the guest clock being up-to-date
6816 		 */
6817 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6818 			kvm_hv_process_stimers(vcpu);
6819 	}
6820 
6821 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6822 		++vcpu->stat.req_event;
6823 		kvm_apic_accept_events(vcpu);
6824 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6825 			r = 1;
6826 			goto out;
6827 		}
6828 
6829 		if (inject_pending_event(vcpu, req_int_win) != 0)
6830 			req_immediate_exit = true;
6831 		else {
6832 			/* Enable NMI/IRQ window open exits if needed.
6833 			 *
6834 			 * SMIs have two cases: 1) they can be nested, and
6835 			 * then there is nothing to do here because RSM will
6836 			 * cause a vmexit anyway; 2) or the SMI can be pending
6837 			 * because inject_pending_event has completed the
6838 			 * injection of an IRQ or NMI from the previous vmexit,
6839 			 * and then we request an immediate exit to inject the SMI.
6840 			 */
6841 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
6842 				req_immediate_exit = true;
6843 			if (vcpu->arch.nmi_pending)
6844 				kvm_x86_ops->enable_nmi_window(vcpu);
6845 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6846 				kvm_x86_ops->enable_irq_window(vcpu);
6847 		}
6848 
6849 		if (kvm_lapic_enabled(vcpu)) {
6850 			update_cr8_intercept(vcpu);
6851 			kvm_lapic_sync_to_vapic(vcpu);
6852 		}
6853 	}
6854 
6855 	r = kvm_mmu_reload(vcpu);
6856 	if (unlikely(r)) {
6857 		goto cancel_injection;
6858 	}
6859 
6860 	preempt_disable();
6861 
6862 	kvm_x86_ops->prepare_guest_switch(vcpu);
6863 	kvm_load_guest_fpu(vcpu);
6864 
6865 	/*
6866 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
6867 	 * IPI are then delayed after guest entry, which ensures that they
6868 	 * result in virtual interrupt delivery.
6869 	 */
6870 	local_irq_disable();
6871 	vcpu->mode = IN_GUEST_MODE;
6872 
6873 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6874 
6875 	/*
6876 	 * 1) We should set ->mode before checking ->requests.  Please see
6877 	 * the comment in kvm_vcpu_exiting_guest_mode().
6878 	 *
6879 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
6880 	 * pairs with the memory barrier implicit in pi_test_and_set_on
6881 	 * (see vmx_deliver_posted_interrupt).
6882 	 *
6883 	 * 3) This also orders the write to mode from any reads to the page
6884 	 * tables done while the VCPU is running.  Please see the comment
6885 	 * in kvm_flush_remote_tlbs.
6886 	 */
6887 	smp_mb__after_srcu_read_unlock();
6888 
6889 	/*
6890 	 * This handles the case where a posted interrupt was
6891 	 * notified with kvm_vcpu_kick.
6892 	 */
6893 	if (kvm_lapic_enabled(vcpu)) {
6894 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6895 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6896 	}
6897 
6898 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6899 	    || need_resched() || signal_pending(current)) {
6900 		vcpu->mode = OUTSIDE_GUEST_MODE;
6901 		smp_wmb();
6902 		local_irq_enable();
6903 		preempt_enable();
6904 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6905 		r = 1;
6906 		goto cancel_injection;
6907 	}
6908 
6909 	kvm_load_guest_xcr0(vcpu);
6910 
6911 	if (req_immediate_exit) {
6912 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6913 		smp_send_reschedule(vcpu->cpu);
6914 	}
6915 
6916 	trace_kvm_entry(vcpu->vcpu_id);
6917 	wait_lapic_expire(vcpu);
6918 	guest_enter_irqoff();
6919 
6920 	if (unlikely(vcpu->arch.switch_db_regs)) {
6921 		set_debugreg(0, 7);
6922 		set_debugreg(vcpu->arch.eff_db[0], 0);
6923 		set_debugreg(vcpu->arch.eff_db[1], 1);
6924 		set_debugreg(vcpu->arch.eff_db[2], 2);
6925 		set_debugreg(vcpu->arch.eff_db[3], 3);
6926 		set_debugreg(vcpu->arch.dr6, 6);
6927 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6928 	}
6929 
6930 	kvm_x86_ops->run(vcpu);
6931 
6932 	/*
6933 	 * Do this here before restoring debug registers on the host.  And
6934 	 * since we do this before handling the vmexit, a DR access vmexit
6935 	 * can (a) read the correct value of the debug registers, (b) set
6936 	 * KVM_DEBUGREG_WONT_EXIT again.
6937 	 */
6938 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6939 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6940 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6941 		kvm_update_dr0123(vcpu);
6942 		kvm_update_dr6(vcpu);
6943 		kvm_update_dr7(vcpu);
6944 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6945 	}
6946 
6947 	/*
6948 	 * If the guest has used debug registers, at least dr7
6949 	 * will be disabled while returning to the host.
6950 	 * If we don't have active breakpoints in the host, we don't
6951 	 * care about the messed up debug address registers. But if
6952 	 * we have some of them active, restore the old state.
6953 	 */
6954 	if (hw_breakpoint_active())
6955 		hw_breakpoint_restore();
6956 
6957 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6958 
6959 	vcpu->mode = OUTSIDE_GUEST_MODE;
6960 	smp_wmb();
6961 
6962 	kvm_put_guest_xcr0(vcpu);
6963 
6964 	kvm_x86_ops->handle_external_intr(vcpu);
6965 
6966 	++vcpu->stat.exits;
6967 
6968 	guest_exit_irqoff();
6969 
6970 	local_irq_enable();
6971 	preempt_enable();
6972 
6973 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6974 
6975 	/*
6976 	 * Profile KVM exit RIPs:
6977 	 */
6978 	if (unlikely(prof_on == KVM_PROFILING)) {
6979 		unsigned long rip = kvm_rip_read(vcpu);
6980 		profile_hit(KVM_PROFILING, (void *)rip);
6981 	}
6982 
6983 	if (unlikely(vcpu->arch.tsc_always_catchup))
6984 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6985 
6986 	if (vcpu->arch.apic_attention)
6987 		kvm_lapic_sync_from_vapic(vcpu);
6988 
6989 	r = kvm_x86_ops->handle_exit(vcpu);
6990 	return r;
6991 
6992 cancel_injection:
6993 	kvm_x86_ops->cancel_injection(vcpu);
6994 	if (unlikely(vcpu->arch.apic_attention))
6995 		kvm_lapic_sync_from_vapic(vcpu);
6996 out:
6997 	return r;
6998 }
6999 
7000 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7001 {
7002 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7003 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7004 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7005 		kvm_vcpu_block(vcpu);
7006 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7007 
7008 		if (kvm_x86_ops->post_block)
7009 			kvm_x86_ops->post_block(vcpu);
7010 
7011 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7012 			return 1;
7013 	}
7014 
7015 	kvm_apic_accept_events(vcpu);
7016 	switch(vcpu->arch.mp_state) {
7017 	case KVM_MP_STATE_HALTED:
7018 		vcpu->arch.pv.pv_unhalted = false;
7019 		vcpu->arch.mp_state =
7020 			KVM_MP_STATE_RUNNABLE;
7021 	case KVM_MP_STATE_RUNNABLE:
7022 		vcpu->arch.apf.halted = false;
7023 		break;
7024 	case KVM_MP_STATE_INIT_RECEIVED:
7025 		break;
7026 	default:
7027 		return -EINTR;
7028 		break;
7029 	}
7030 	return 1;
7031 }
7032 
7033 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7034 {
7035 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7036 		kvm_x86_ops->check_nested_events(vcpu, false);
7037 
7038 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7039 		!vcpu->arch.apf.halted);
7040 }
7041 
7042 static int vcpu_run(struct kvm_vcpu *vcpu)
7043 {
7044 	int r;
7045 	struct kvm *kvm = vcpu->kvm;
7046 
7047 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7048 
7049 	for (;;) {
7050 		if (kvm_vcpu_running(vcpu)) {
7051 			r = vcpu_enter_guest(vcpu);
7052 		} else {
7053 			r = vcpu_block(kvm, vcpu);
7054 		}
7055 
7056 		if (r <= 0)
7057 			break;
7058 
7059 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7060 		if (kvm_cpu_has_pending_timer(vcpu))
7061 			kvm_inject_pending_timer_irqs(vcpu);
7062 
7063 		if (dm_request_for_irq_injection(vcpu) &&
7064 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7065 			r = 0;
7066 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7067 			++vcpu->stat.request_irq_exits;
7068 			break;
7069 		}
7070 
7071 		kvm_check_async_pf_completion(vcpu);
7072 
7073 		if (signal_pending(current)) {
7074 			r = -EINTR;
7075 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7076 			++vcpu->stat.signal_exits;
7077 			break;
7078 		}
7079 		if (need_resched()) {
7080 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7081 			cond_resched();
7082 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7083 		}
7084 	}
7085 
7086 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7087 
7088 	return r;
7089 }
7090 
7091 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7092 {
7093 	int r;
7094 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7095 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7096 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7097 	if (r != EMULATE_DONE)
7098 		return 0;
7099 	return 1;
7100 }
7101 
7102 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7103 {
7104 	BUG_ON(!vcpu->arch.pio.count);
7105 
7106 	return complete_emulated_io(vcpu);
7107 }
7108 
7109 /*
7110  * Implements the following, as a state machine:
7111  *
7112  * read:
7113  *   for each fragment
7114  *     for each mmio piece in the fragment
7115  *       write gpa, len
7116  *       exit
7117  *       copy data
7118  *   execute insn
7119  *
7120  * write:
7121  *   for each fragment
7122  *     for each mmio piece in the fragment
7123  *       write gpa, len
7124  *       copy data
7125  *       exit
7126  */
7127 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7128 {
7129 	struct kvm_run *run = vcpu->run;
7130 	struct kvm_mmio_fragment *frag;
7131 	unsigned len;
7132 
7133 	BUG_ON(!vcpu->mmio_needed);
7134 
7135 	/* Complete previous fragment */
7136 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7137 	len = min(8u, frag->len);
7138 	if (!vcpu->mmio_is_write)
7139 		memcpy(frag->data, run->mmio.data, len);
7140 
7141 	if (frag->len <= 8) {
7142 		/* Switch to the next fragment. */
7143 		frag++;
7144 		vcpu->mmio_cur_fragment++;
7145 	} else {
7146 		/* Go forward to the next mmio piece. */
7147 		frag->data += len;
7148 		frag->gpa += len;
7149 		frag->len -= len;
7150 	}
7151 
7152 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7153 		vcpu->mmio_needed = 0;
7154 
7155 		/* FIXME: return into emulator if single-stepping.  */
7156 		if (vcpu->mmio_is_write)
7157 			return 1;
7158 		vcpu->mmio_read_completed = 1;
7159 		return complete_emulated_io(vcpu);
7160 	}
7161 
7162 	run->exit_reason = KVM_EXIT_MMIO;
7163 	run->mmio.phys_addr = frag->gpa;
7164 	if (vcpu->mmio_is_write)
7165 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7166 	run->mmio.len = min(8u, frag->len);
7167 	run->mmio.is_write = vcpu->mmio_is_write;
7168 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7169 	return 0;
7170 }
7171 
7172 
7173 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7174 {
7175 	struct fpu *fpu = &current->thread.fpu;
7176 	int r;
7177 	sigset_t sigsaved;
7178 
7179 	fpu__activate_curr(fpu);
7180 
7181 	if (vcpu->sigset_active)
7182 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7183 
7184 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7185 		kvm_vcpu_block(vcpu);
7186 		kvm_apic_accept_events(vcpu);
7187 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7188 		r = -EAGAIN;
7189 		goto out;
7190 	}
7191 
7192 	/* re-sync apic's tpr */
7193 	if (!lapic_in_kernel(vcpu)) {
7194 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7195 			r = -EINVAL;
7196 			goto out;
7197 		}
7198 	}
7199 
7200 	if (unlikely(vcpu->arch.complete_userspace_io)) {
7201 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7202 		vcpu->arch.complete_userspace_io = NULL;
7203 		r = cui(vcpu);
7204 		if (r <= 0)
7205 			goto out;
7206 	} else
7207 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7208 
7209 	if (kvm_run->immediate_exit)
7210 		r = -EINTR;
7211 	else
7212 		r = vcpu_run(vcpu);
7213 
7214 out:
7215 	post_kvm_run_save(vcpu);
7216 	if (vcpu->sigset_active)
7217 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7218 
7219 	return r;
7220 }
7221 
7222 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7223 {
7224 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7225 		/*
7226 		 * We are here if userspace calls get_regs() in the middle of
7227 		 * instruction emulation. Registers state needs to be copied
7228 		 * back from emulation context to vcpu. Userspace shouldn't do
7229 		 * that usually, but some bad designed PV devices (vmware
7230 		 * backdoor interface) need this to work
7231 		 */
7232 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7233 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7234 	}
7235 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7236 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7237 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7238 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7239 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7240 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7241 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7242 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7243 #ifdef CONFIG_X86_64
7244 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7245 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7246 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7247 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7248 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7249 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7250 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7251 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7252 #endif
7253 
7254 	regs->rip = kvm_rip_read(vcpu);
7255 	regs->rflags = kvm_get_rflags(vcpu);
7256 
7257 	return 0;
7258 }
7259 
7260 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7261 {
7262 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7263 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7264 
7265 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7266 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7267 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7268 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7269 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7270 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7271 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7272 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7273 #ifdef CONFIG_X86_64
7274 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7275 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7276 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7277 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7278 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7279 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7280 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7281 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7282 #endif
7283 
7284 	kvm_rip_write(vcpu, regs->rip);
7285 	kvm_set_rflags(vcpu, regs->rflags);
7286 
7287 	vcpu->arch.exception.pending = false;
7288 
7289 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7290 
7291 	return 0;
7292 }
7293 
7294 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7295 {
7296 	struct kvm_segment cs;
7297 
7298 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7299 	*db = cs.db;
7300 	*l = cs.l;
7301 }
7302 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7303 
7304 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7305 				  struct kvm_sregs *sregs)
7306 {
7307 	struct desc_ptr dt;
7308 
7309 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7310 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7311 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7312 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7313 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7314 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7315 
7316 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7317 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7318 
7319 	kvm_x86_ops->get_idt(vcpu, &dt);
7320 	sregs->idt.limit = dt.size;
7321 	sregs->idt.base = dt.address;
7322 	kvm_x86_ops->get_gdt(vcpu, &dt);
7323 	sregs->gdt.limit = dt.size;
7324 	sregs->gdt.base = dt.address;
7325 
7326 	sregs->cr0 = kvm_read_cr0(vcpu);
7327 	sregs->cr2 = vcpu->arch.cr2;
7328 	sregs->cr3 = kvm_read_cr3(vcpu);
7329 	sregs->cr4 = kvm_read_cr4(vcpu);
7330 	sregs->cr8 = kvm_get_cr8(vcpu);
7331 	sregs->efer = vcpu->arch.efer;
7332 	sregs->apic_base = kvm_get_apic_base(vcpu);
7333 
7334 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7335 
7336 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7337 		set_bit(vcpu->arch.interrupt.nr,
7338 			(unsigned long *)sregs->interrupt_bitmap);
7339 
7340 	return 0;
7341 }
7342 
7343 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7344 				    struct kvm_mp_state *mp_state)
7345 {
7346 	kvm_apic_accept_events(vcpu);
7347 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7348 					vcpu->arch.pv.pv_unhalted)
7349 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7350 	else
7351 		mp_state->mp_state = vcpu->arch.mp_state;
7352 
7353 	return 0;
7354 }
7355 
7356 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7357 				    struct kvm_mp_state *mp_state)
7358 {
7359 	if (!lapic_in_kernel(vcpu) &&
7360 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7361 		return -EINVAL;
7362 
7363 	/* INITs are latched while in SMM */
7364 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7365 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7366 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7367 		return -EINVAL;
7368 
7369 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7370 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7371 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7372 	} else
7373 		vcpu->arch.mp_state = mp_state->mp_state;
7374 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7375 	return 0;
7376 }
7377 
7378 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7379 		    int reason, bool has_error_code, u32 error_code)
7380 {
7381 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7382 	int ret;
7383 
7384 	init_emulate_ctxt(vcpu);
7385 
7386 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7387 				   has_error_code, error_code);
7388 
7389 	if (ret)
7390 		return EMULATE_FAIL;
7391 
7392 	kvm_rip_write(vcpu, ctxt->eip);
7393 	kvm_set_rflags(vcpu, ctxt->eflags);
7394 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7395 	return EMULATE_DONE;
7396 }
7397 EXPORT_SYMBOL_GPL(kvm_task_switch);
7398 
7399 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7400 				  struct kvm_sregs *sregs)
7401 {
7402 	struct msr_data apic_base_msr;
7403 	int mmu_reset_needed = 0;
7404 	int pending_vec, max_bits, idx;
7405 	struct desc_ptr dt;
7406 
7407 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7408 		return -EINVAL;
7409 
7410 	dt.size = sregs->idt.limit;
7411 	dt.address = sregs->idt.base;
7412 	kvm_x86_ops->set_idt(vcpu, &dt);
7413 	dt.size = sregs->gdt.limit;
7414 	dt.address = sregs->gdt.base;
7415 	kvm_x86_ops->set_gdt(vcpu, &dt);
7416 
7417 	vcpu->arch.cr2 = sregs->cr2;
7418 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7419 	vcpu->arch.cr3 = sregs->cr3;
7420 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7421 
7422 	kvm_set_cr8(vcpu, sregs->cr8);
7423 
7424 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7425 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7426 	apic_base_msr.data = sregs->apic_base;
7427 	apic_base_msr.host_initiated = true;
7428 	kvm_set_apic_base(vcpu, &apic_base_msr);
7429 
7430 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7431 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7432 	vcpu->arch.cr0 = sregs->cr0;
7433 
7434 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7435 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7436 	if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7437 		kvm_update_cpuid(vcpu);
7438 
7439 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7440 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7441 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7442 		mmu_reset_needed = 1;
7443 	}
7444 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7445 
7446 	if (mmu_reset_needed)
7447 		kvm_mmu_reset_context(vcpu);
7448 
7449 	max_bits = KVM_NR_INTERRUPTS;
7450 	pending_vec = find_first_bit(
7451 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7452 	if (pending_vec < max_bits) {
7453 		kvm_queue_interrupt(vcpu, pending_vec, false);
7454 		pr_debug("Set back pending irq %d\n", pending_vec);
7455 	}
7456 
7457 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7458 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7459 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7460 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7461 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7462 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7463 
7464 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7465 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7466 
7467 	update_cr8_intercept(vcpu);
7468 
7469 	/* Older userspace won't unhalt the vcpu on reset. */
7470 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7471 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7472 	    !is_protmode(vcpu))
7473 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7474 
7475 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7476 
7477 	return 0;
7478 }
7479 
7480 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7481 					struct kvm_guest_debug *dbg)
7482 {
7483 	unsigned long rflags;
7484 	int i, r;
7485 
7486 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7487 		r = -EBUSY;
7488 		if (vcpu->arch.exception.pending)
7489 			goto out;
7490 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7491 			kvm_queue_exception(vcpu, DB_VECTOR);
7492 		else
7493 			kvm_queue_exception(vcpu, BP_VECTOR);
7494 	}
7495 
7496 	/*
7497 	 * Read rflags as long as potentially injected trace flags are still
7498 	 * filtered out.
7499 	 */
7500 	rflags = kvm_get_rflags(vcpu);
7501 
7502 	vcpu->guest_debug = dbg->control;
7503 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7504 		vcpu->guest_debug = 0;
7505 
7506 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7507 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7508 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7509 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7510 	} else {
7511 		for (i = 0; i < KVM_NR_DB_REGS; i++)
7512 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7513 	}
7514 	kvm_update_dr7(vcpu);
7515 
7516 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7517 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7518 			get_segment_base(vcpu, VCPU_SREG_CS);
7519 
7520 	/*
7521 	 * Trigger an rflags update that will inject or remove the trace
7522 	 * flags.
7523 	 */
7524 	kvm_set_rflags(vcpu, rflags);
7525 
7526 	kvm_x86_ops->update_bp_intercept(vcpu);
7527 
7528 	r = 0;
7529 
7530 out:
7531 
7532 	return r;
7533 }
7534 
7535 /*
7536  * Translate a guest virtual address to a guest physical address.
7537  */
7538 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7539 				    struct kvm_translation *tr)
7540 {
7541 	unsigned long vaddr = tr->linear_address;
7542 	gpa_t gpa;
7543 	int idx;
7544 
7545 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7546 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7547 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7548 	tr->physical_address = gpa;
7549 	tr->valid = gpa != UNMAPPED_GVA;
7550 	tr->writeable = 1;
7551 	tr->usermode = 0;
7552 
7553 	return 0;
7554 }
7555 
7556 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7557 {
7558 	struct fxregs_state *fxsave =
7559 			&vcpu->arch.guest_fpu.state.fxsave;
7560 
7561 	memcpy(fpu->fpr, fxsave->st_space, 128);
7562 	fpu->fcw = fxsave->cwd;
7563 	fpu->fsw = fxsave->swd;
7564 	fpu->ftwx = fxsave->twd;
7565 	fpu->last_opcode = fxsave->fop;
7566 	fpu->last_ip = fxsave->rip;
7567 	fpu->last_dp = fxsave->rdp;
7568 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7569 
7570 	return 0;
7571 }
7572 
7573 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7574 {
7575 	struct fxregs_state *fxsave =
7576 			&vcpu->arch.guest_fpu.state.fxsave;
7577 
7578 	memcpy(fxsave->st_space, fpu->fpr, 128);
7579 	fxsave->cwd = fpu->fcw;
7580 	fxsave->swd = fpu->fsw;
7581 	fxsave->twd = fpu->ftwx;
7582 	fxsave->fop = fpu->last_opcode;
7583 	fxsave->rip = fpu->last_ip;
7584 	fxsave->rdp = fpu->last_dp;
7585 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7586 
7587 	return 0;
7588 }
7589 
7590 static void fx_init(struct kvm_vcpu *vcpu)
7591 {
7592 	fpstate_init(&vcpu->arch.guest_fpu.state);
7593 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7594 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7595 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7596 
7597 	/*
7598 	 * Ensure guest xcr0 is valid for loading
7599 	 */
7600 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7601 
7602 	vcpu->arch.cr0 |= X86_CR0_ET;
7603 }
7604 
7605 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7606 {
7607 	if (vcpu->guest_fpu_loaded)
7608 		return;
7609 
7610 	/*
7611 	 * Restore all possible states in the guest,
7612 	 * and assume host would use all available bits.
7613 	 * Guest xcr0 would be loaded later.
7614 	 */
7615 	vcpu->guest_fpu_loaded = 1;
7616 	__kernel_fpu_begin();
7617 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7618 	trace_kvm_fpu(1);
7619 }
7620 
7621 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7622 {
7623 	if (!vcpu->guest_fpu_loaded)
7624 		return;
7625 
7626 	vcpu->guest_fpu_loaded = 0;
7627 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7628 	__kernel_fpu_end();
7629 	++vcpu->stat.fpu_reload;
7630 	trace_kvm_fpu(0);
7631 }
7632 
7633 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7634 {
7635 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7636 
7637 	kvmclock_reset(vcpu);
7638 
7639 	kvm_x86_ops->vcpu_free(vcpu);
7640 	free_cpumask_var(wbinvd_dirty_mask);
7641 }
7642 
7643 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7644 						unsigned int id)
7645 {
7646 	struct kvm_vcpu *vcpu;
7647 
7648 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7649 		printk_once(KERN_WARNING
7650 		"kvm: SMP vm created on host with unstable TSC; "
7651 		"guest TSC will not be reliable\n");
7652 
7653 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7654 
7655 	return vcpu;
7656 }
7657 
7658 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7659 {
7660 	int r;
7661 
7662 	kvm_vcpu_mtrr_init(vcpu);
7663 	r = vcpu_load(vcpu);
7664 	if (r)
7665 		return r;
7666 	kvm_vcpu_reset(vcpu, false);
7667 	kvm_mmu_setup(vcpu);
7668 	vcpu_put(vcpu);
7669 	return r;
7670 }
7671 
7672 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7673 {
7674 	struct msr_data msr;
7675 	struct kvm *kvm = vcpu->kvm;
7676 
7677 	if (vcpu_load(vcpu))
7678 		return;
7679 	msr.data = 0x0;
7680 	msr.index = MSR_IA32_TSC;
7681 	msr.host_initiated = true;
7682 	kvm_write_tsc(vcpu, &msr);
7683 	vcpu_put(vcpu);
7684 
7685 	if (!kvmclock_periodic_sync)
7686 		return;
7687 
7688 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7689 					KVMCLOCK_SYNC_PERIOD);
7690 }
7691 
7692 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7693 {
7694 	int r;
7695 	vcpu->arch.apf.msr_val = 0;
7696 
7697 	r = vcpu_load(vcpu);
7698 	BUG_ON(r);
7699 	kvm_mmu_unload(vcpu);
7700 	vcpu_put(vcpu);
7701 
7702 	kvm_x86_ops->vcpu_free(vcpu);
7703 }
7704 
7705 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7706 {
7707 	vcpu->arch.hflags = 0;
7708 
7709 	vcpu->arch.smi_pending = 0;
7710 	atomic_set(&vcpu->arch.nmi_queued, 0);
7711 	vcpu->arch.nmi_pending = 0;
7712 	vcpu->arch.nmi_injected = false;
7713 	kvm_clear_interrupt_queue(vcpu);
7714 	kvm_clear_exception_queue(vcpu);
7715 
7716 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7717 	kvm_update_dr0123(vcpu);
7718 	vcpu->arch.dr6 = DR6_INIT;
7719 	kvm_update_dr6(vcpu);
7720 	vcpu->arch.dr7 = DR7_FIXED_1;
7721 	kvm_update_dr7(vcpu);
7722 
7723 	vcpu->arch.cr2 = 0;
7724 
7725 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7726 	vcpu->arch.apf.msr_val = 0;
7727 	vcpu->arch.st.msr_val = 0;
7728 
7729 	kvmclock_reset(vcpu);
7730 
7731 	kvm_clear_async_pf_completion_queue(vcpu);
7732 	kvm_async_pf_hash_reset(vcpu);
7733 	vcpu->arch.apf.halted = false;
7734 
7735 	if (!init_event) {
7736 		kvm_pmu_reset(vcpu);
7737 		vcpu->arch.smbase = 0x30000;
7738 
7739 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7740 		vcpu->arch.msr_misc_features_enables = 0;
7741 	}
7742 
7743 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7744 	vcpu->arch.regs_avail = ~0;
7745 	vcpu->arch.regs_dirty = ~0;
7746 
7747 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7748 }
7749 
7750 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7751 {
7752 	struct kvm_segment cs;
7753 
7754 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7755 	cs.selector = vector << 8;
7756 	cs.base = vector << 12;
7757 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7758 	kvm_rip_write(vcpu, 0);
7759 }
7760 
7761 int kvm_arch_hardware_enable(void)
7762 {
7763 	struct kvm *kvm;
7764 	struct kvm_vcpu *vcpu;
7765 	int i;
7766 	int ret;
7767 	u64 local_tsc;
7768 	u64 max_tsc = 0;
7769 	bool stable, backwards_tsc = false;
7770 
7771 	kvm_shared_msr_cpu_online();
7772 	ret = kvm_x86_ops->hardware_enable();
7773 	if (ret != 0)
7774 		return ret;
7775 
7776 	local_tsc = rdtsc();
7777 	stable = !check_tsc_unstable();
7778 	list_for_each_entry(kvm, &vm_list, vm_list) {
7779 		kvm_for_each_vcpu(i, vcpu, kvm) {
7780 			if (!stable && vcpu->cpu == smp_processor_id())
7781 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7782 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7783 				backwards_tsc = true;
7784 				if (vcpu->arch.last_host_tsc > max_tsc)
7785 					max_tsc = vcpu->arch.last_host_tsc;
7786 			}
7787 		}
7788 	}
7789 
7790 	/*
7791 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7792 	 * platforms that reset TSC during suspend or hibernate actions, but
7793 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7794 	 * detect that condition here, which happens early in CPU bringup,
7795 	 * before any KVM threads can be running.  Unfortunately, we can't
7796 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7797 	 * enough into CPU bringup that we know how much real time has actually
7798 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7799 	 * variables that haven't been updated yet.
7800 	 *
7801 	 * So we simply find the maximum observed TSC above, then record the
7802 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7803 	 * the adjustment will be applied.  Note that we accumulate
7804 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7805 	 * gets a chance to run again.  In the event that no KVM threads get a
7806 	 * chance to run, we will miss the entire elapsed period, as we'll have
7807 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7808 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7809 	 * uniform across all VCPUs (not to mention the scenario is extremely
7810 	 * unlikely). It is possible that a second hibernate recovery happens
7811 	 * much faster than a first, causing the observed TSC here to be
7812 	 * smaller; this would require additional padding adjustment, which is
7813 	 * why we set last_host_tsc to the local tsc observed here.
7814 	 *
7815 	 * N.B. - this code below runs only on platforms with reliable TSC,
7816 	 * as that is the only way backwards_tsc is set above.  Also note
7817 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7818 	 * have the same delta_cyc adjustment applied if backwards_tsc
7819 	 * is detected.  Note further, this adjustment is only done once,
7820 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7821 	 * called multiple times (one for each physical CPU bringup).
7822 	 *
7823 	 * Platforms with unreliable TSCs don't have to deal with this, they
7824 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7825 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7826 	 * guarantee that they stay in perfect synchronization.
7827 	 */
7828 	if (backwards_tsc) {
7829 		u64 delta_cyc = max_tsc - local_tsc;
7830 		backwards_tsc_observed = true;
7831 		list_for_each_entry(kvm, &vm_list, vm_list) {
7832 			kvm_for_each_vcpu(i, vcpu, kvm) {
7833 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7834 				vcpu->arch.last_host_tsc = local_tsc;
7835 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7836 			}
7837 
7838 			/*
7839 			 * We have to disable TSC offset matching.. if you were
7840 			 * booting a VM while issuing an S4 host suspend....
7841 			 * you may have some problem.  Solving this issue is
7842 			 * left as an exercise to the reader.
7843 			 */
7844 			kvm->arch.last_tsc_nsec = 0;
7845 			kvm->arch.last_tsc_write = 0;
7846 		}
7847 
7848 	}
7849 	return 0;
7850 }
7851 
7852 void kvm_arch_hardware_disable(void)
7853 {
7854 	kvm_x86_ops->hardware_disable();
7855 	drop_user_return_notifiers();
7856 }
7857 
7858 int kvm_arch_hardware_setup(void)
7859 {
7860 	int r;
7861 
7862 	r = kvm_x86_ops->hardware_setup();
7863 	if (r != 0)
7864 		return r;
7865 
7866 	if (kvm_has_tsc_control) {
7867 		/*
7868 		 * Make sure the user can only configure tsc_khz values that
7869 		 * fit into a signed integer.
7870 		 * A min value is not calculated needed because it will always
7871 		 * be 1 on all machines.
7872 		 */
7873 		u64 max = min(0x7fffffffULL,
7874 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7875 		kvm_max_guest_tsc_khz = max;
7876 
7877 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7878 	}
7879 
7880 	kvm_init_msr_list();
7881 	return 0;
7882 }
7883 
7884 void kvm_arch_hardware_unsetup(void)
7885 {
7886 	kvm_x86_ops->hardware_unsetup();
7887 }
7888 
7889 void kvm_arch_check_processor_compat(void *rtn)
7890 {
7891 	kvm_x86_ops->check_processor_compatibility(rtn);
7892 }
7893 
7894 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7895 {
7896 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7897 }
7898 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7899 
7900 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7901 {
7902 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7903 }
7904 
7905 struct static_key kvm_no_apic_vcpu __read_mostly;
7906 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7907 
7908 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7909 {
7910 	struct page *page;
7911 	struct kvm *kvm;
7912 	int r;
7913 
7914 	BUG_ON(vcpu->kvm == NULL);
7915 	kvm = vcpu->kvm;
7916 
7917 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7918 	vcpu->arch.pv.pv_unhalted = false;
7919 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7920 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7921 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7922 	else
7923 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7924 
7925 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7926 	if (!page) {
7927 		r = -ENOMEM;
7928 		goto fail;
7929 	}
7930 	vcpu->arch.pio_data = page_address(page);
7931 
7932 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7933 
7934 	r = kvm_mmu_create(vcpu);
7935 	if (r < 0)
7936 		goto fail_free_pio_data;
7937 
7938 	if (irqchip_in_kernel(kvm)) {
7939 		r = kvm_create_lapic(vcpu);
7940 		if (r < 0)
7941 			goto fail_mmu_destroy;
7942 	} else
7943 		static_key_slow_inc(&kvm_no_apic_vcpu);
7944 
7945 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7946 				       GFP_KERNEL);
7947 	if (!vcpu->arch.mce_banks) {
7948 		r = -ENOMEM;
7949 		goto fail_free_lapic;
7950 	}
7951 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7952 
7953 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7954 		r = -ENOMEM;
7955 		goto fail_free_mce_banks;
7956 	}
7957 
7958 	fx_init(vcpu);
7959 
7960 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7961 	vcpu->arch.pv_time_enabled = false;
7962 
7963 	vcpu->arch.guest_supported_xcr0 = 0;
7964 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7965 
7966 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7967 
7968 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7969 
7970 	kvm_async_pf_hash_reset(vcpu);
7971 	kvm_pmu_init(vcpu);
7972 
7973 	vcpu->arch.pending_external_vector = -1;
7974 
7975 	kvm_hv_vcpu_init(vcpu);
7976 
7977 	return 0;
7978 
7979 fail_free_mce_banks:
7980 	kfree(vcpu->arch.mce_banks);
7981 fail_free_lapic:
7982 	kvm_free_lapic(vcpu);
7983 fail_mmu_destroy:
7984 	kvm_mmu_destroy(vcpu);
7985 fail_free_pio_data:
7986 	free_page((unsigned long)vcpu->arch.pio_data);
7987 fail:
7988 	return r;
7989 }
7990 
7991 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7992 {
7993 	int idx;
7994 
7995 	kvm_hv_vcpu_uninit(vcpu);
7996 	kvm_pmu_destroy(vcpu);
7997 	kfree(vcpu->arch.mce_banks);
7998 	kvm_free_lapic(vcpu);
7999 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8000 	kvm_mmu_destroy(vcpu);
8001 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8002 	free_page((unsigned long)vcpu->arch.pio_data);
8003 	if (!lapic_in_kernel(vcpu))
8004 		static_key_slow_dec(&kvm_no_apic_vcpu);
8005 }
8006 
8007 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8008 {
8009 	kvm_x86_ops->sched_in(vcpu, cpu);
8010 }
8011 
8012 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8013 {
8014 	if (type)
8015 		return -EINVAL;
8016 
8017 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8018 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8019 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8020 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8021 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8022 
8023 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8024 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8025 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8026 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8027 		&kvm->arch.irq_sources_bitmap);
8028 
8029 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8030 	mutex_init(&kvm->arch.apic_map_lock);
8031 	mutex_init(&kvm->arch.hyperv.hv_lock);
8032 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8033 
8034 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8035 	pvclock_update_vm_gtod_copy(kvm);
8036 
8037 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8038 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8039 
8040 	kvm_page_track_init(kvm);
8041 	kvm_mmu_init_vm(kvm);
8042 
8043 	if (kvm_x86_ops->vm_init)
8044 		return kvm_x86_ops->vm_init(kvm);
8045 
8046 	return 0;
8047 }
8048 
8049 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8050 {
8051 	int r;
8052 	r = vcpu_load(vcpu);
8053 	BUG_ON(r);
8054 	kvm_mmu_unload(vcpu);
8055 	vcpu_put(vcpu);
8056 }
8057 
8058 static void kvm_free_vcpus(struct kvm *kvm)
8059 {
8060 	unsigned int i;
8061 	struct kvm_vcpu *vcpu;
8062 
8063 	/*
8064 	 * Unpin any mmu pages first.
8065 	 */
8066 	kvm_for_each_vcpu(i, vcpu, kvm) {
8067 		kvm_clear_async_pf_completion_queue(vcpu);
8068 		kvm_unload_vcpu_mmu(vcpu);
8069 	}
8070 	kvm_for_each_vcpu(i, vcpu, kvm)
8071 		kvm_arch_vcpu_free(vcpu);
8072 
8073 	mutex_lock(&kvm->lock);
8074 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8075 		kvm->vcpus[i] = NULL;
8076 
8077 	atomic_set(&kvm->online_vcpus, 0);
8078 	mutex_unlock(&kvm->lock);
8079 }
8080 
8081 void kvm_arch_sync_events(struct kvm *kvm)
8082 {
8083 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8084 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8085 	kvm_free_pit(kvm);
8086 }
8087 
8088 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8089 {
8090 	int i, r;
8091 	unsigned long hva;
8092 	struct kvm_memslots *slots = kvm_memslots(kvm);
8093 	struct kvm_memory_slot *slot, old;
8094 
8095 	/* Called with kvm->slots_lock held.  */
8096 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8097 		return -EINVAL;
8098 
8099 	slot = id_to_memslot(slots, id);
8100 	if (size) {
8101 		if (slot->npages)
8102 			return -EEXIST;
8103 
8104 		/*
8105 		 * MAP_SHARED to prevent internal slot pages from being moved
8106 		 * by fork()/COW.
8107 		 */
8108 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8109 			      MAP_SHARED | MAP_ANONYMOUS, 0);
8110 		if (IS_ERR((void *)hva))
8111 			return PTR_ERR((void *)hva);
8112 	} else {
8113 		if (!slot->npages)
8114 			return 0;
8115 
8116 		hva = 0;
8117 	}
8118 
8119 	old = *slot;
8120 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8121 		struct kvm_userspace_memory_region m;
8122 
8123 		m.slot = id | (i << 16);
8124 		m.flags = 0;
8125 		m.guest_phys_addr = gpa;
8126 		m.userspace_addr = hva;
8127 		m.memory_size = size;
8128 		r = __kvm_set_memory_region(kvm, &m);
8129 		if (r < 0)
8130 			return r;
8131 	}
8132 
8133 	if (!size) {
8134 		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8135 		WARN_ON(r < 0);
8136 	}
8137 
8138 	return 0;
8139 }
8140 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8141 
8142 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8143 {
8144 	int r;
8145 
8146 	mutex_lock(&kvm->slots_lock);
8147 	r = __x86_set_memory_region(kvm, id, gpa, size);
8148 	mutex_unlock(&kvm->slots_lock);
8149 
8150 	return r;
8151 }
8152 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8153 
8154 void kvm_arch_destroy_vm(struct kvm *kvm)
8155 {
8156 	if (current->mm == kvm->mm) {
8157 		/*
8158 		 * Free memory regions allocated on behalf of userspace,
8159 		 * unless the the memory map has changed due to process exit
8160 		 * or fd copying.
8161 		 */
8162 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8163 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8164 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8165 	}
8166 	if (kvm_x86_ops->vm_destroy)
8167 		kvm_x86_ops->vm_destroy(kvm);
8168 	kvm_pic_destroy(kvm);
8169 	kvm_ioapic_destroy(kvm);
8170 	kvm_free_vcpus(kvm);
8171 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8172 	kvm_mmu_uninit_vm(kvm);
8173 	kvm_page_track_cleanup(kvm);
8174 }
8175 
8176 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8177 			   struct kvm_memory_slot *dont)
8178 {
8179 	int i;
8180 
8181 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8182 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8183 			kvfree(free->arch.rmap[i]);
8184 			free->arch.rmap[i] = NULL;
8185 		}
8186 		if (i == 0)
8187 			continue;
8188 
8189 		if (!dont || free->arch.lpage_info[i - 1] !=
8190 			     dont->arch.lpage_info[i - 1]) {
8191 			kvfree(free->arch.lpage_info[i - 1]);
8192 			free->arch.lpage_info[i - 1] = NULL;
8193 		}
8194 	}
8195 
8196 	kvm_page_track_free_memslot(free, dont);
8197 }
8198 
8199 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8200 			    unsigned long npages)
8201 {
8202 	int i;
8203 
8204 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8205 		struct kvm_lpage_info *linfo;
8206 		unsigned long ugfn;
8207 		int lpages;
8208 		int level = i + 1;
8209 
8210 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
8211 				      slot->base_gfn, level) + 1;
8212 
8213 		slot->arch.rmap[i] =
8214 			kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8215 		if (!slot->arch.rmap[i])
8216 			goto out_free;
8217 		if (i == 0)
8218 			continue;
8219 
8220 		linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8221 		if (!linfo)
8222 			goto out_free;
8223 
8224 		slot->arch.lpage_info[i - 1] = linfo;
8225 
8226 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8227 			linfo[0].disallow_lpage = 1;
8228 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8229 			linfo[lpages - 1].disallow_lpage = 1;
8230 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
8231 		/*
8232 		 * If the gfn and userspace address are not aligned wrt each
8233 		 * other, or if explicitly asked to, disable large page
8234 		 * support for this slot
8235 		 */
8236 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8237 		    !kvm_largepages_enabled()) {
8238 			unsigned long j;
8239 
8240 			for (j = 0; j < lpages; ++j)
8241 				linfo[j].disallow_lpage = 1;
8242 		}
8243 	}
8244 
8245 	if (kvm_page_track_create_memslot(slot, npages))
8246 		goto out_free;
8247 
8248 	return 0;
8249 
8250 out_free:
8251 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8252 		kvfree(slot->arch.rmap[i]);
8253 		slot->arch.rmap[i] = NULL;
8254 		if (i == 0)
8255 			continue;
8256 
8257 		kvfree(slot->arch.lpage_info[i - 1]);
8258 		slot->arch.lpage_info[i - 1] = NULL;
8259 	}
8260 	return -ENOMEM;
8261 }
8262 
8263 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8264 {
8265 	/*
8266 	 * memslots->generation has been incremented.
8267 	 * mmio generation may have reached its maximum value.
8268 	 */
8269 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8270 }
8271 
8272 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8273 				struct kvm_memory_slot *memslot,
8274 				const struct kvm_userspace_memory_region *mem,
8275 				enum kvm_mr_change change)
8276 {
8277 	return 0;
8278 }
8279 
8280 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8281 				     struct kvm_memory_slot *new)
8282 {
8283 	/* Still write protect RO slot */
8284 	if (new->flags & KVM_MEM_READONLY) {
8285 		kvm_mmu_slot_remove_write_access(kvm, new);
8286 		return;
8287 	}
8288 
8289 	/*
8290 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
8291 	 *
8292 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
8293 	 *
8294 	 *  - KVM_MR_CREATE with dirty logging is disabled
8295 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8296 	 *
8297 	 * The reason is, in case of PML, we need to set D-bit for any slots
8298 	 * with dirty logging disabled in order to eliminate unnecessary GPA
8299 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
8300 	 * guarantees leaving PML enabled during guest's lifetime won't have
8301 	 * any additonal overhead from PML when guest is running with dirty
8302 	 * logging disabled for memory slots.
8303 	 *
8304 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8305 	 * to dirty logging mode.
8306 	 *
8307 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8308 	 *
8309 	 * In case of write protect:
8310 	 *
8311 	 * Write protect all pages for dirty logging.
8312 	 *
8313 	 * All the sptes including the large sptes which point to this
8314 	 * slot are set to readonly. We can not create any new large
8315 	 * spte on this slot until the end of the logging.
8316 	 *
8317 	 * See the comments in fast_page_fault().
8318 	 */
8319 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8320 		if (kvm_x86_ops->slot_enable_log_dirty)
8321 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8322 		else
8323 			kvm_mmu_slot_remove_write_access(kvm, new);
8324 	} else {
8325 		if (kvm_x86_ops->slot_disable_log_dirty)
8326 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8327 	}
8328 }
8329 
8330 void kvm_arch_commit_memory_region(struct kvm *kvm,
8331 				const struct kvm_userspace_memory_region *mem,
8332 				const struct kvm_memory_slot *old,
8333 				const struct kvm_memory_slot *new,
8334 				enum kvm_mr_change change)
8335 {
8336 	int nr_mmu_pages = 0;
8337 
8338 	if (!kvm->arch.n_requested_mmu_pages)
8339 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8340 
8341 	if (nr_mmu_pages)
8342 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8343 
8344 	/*
8345 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
8346 	 * sptes have to be split.  If live migration is successful, the guest
8347 	 * in the source machine will be destroyed and large sptes will be
8348 	 * created in the destination. However, if the guest continues to run
8349 	 * in the source machine (for example if live migration fails), small
8350 	 * sptes will remain around and cause bad performance.
8351 	 *
8352 	 * Scan sptes if dirty logging has been stopped, dropping those
8353 	 * which can be collapsed into a single large-page spte.  Later
8354 	 * page faults will create the large-page sptes.
8355 	 */
8356 	if ((change != KVM_MR_DELETE) &&
8357 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8358 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8359 		kvm_mmu_zap_collapsible_sptes(kvm, new);
8360 
8361 	/*
8362 	 * Set up write protection and/or dirty logging for the new slot.
8363 	 *
8364 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8365 	 * been zapped so no dirty logging staff is needed for old slot. For
8366 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8367 	 * new and it's also covered when dealing with the new slot.
8368 	 *
8369 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
8370 	 */
8371 	if (change != KVM_MR_DELETE)
8372 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8373 }
8374 
8375 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8376 {
8377 	kvm_mmu_invalidate_zap_all_pages(kvm);
8378 }
8379 
8380 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8381 				   struct kvm_memory_slot *slot)
8382 {
8383 	kvm_page_track_flush_slot(kvm, slot);
8384 }
8385 
8386 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8387 {
8388 	if (!list_empty_careful(&vcpu->async_pf.done))
8389 		return true;
8390 
8391 	if (kvm_apic_has_events(vcpu))
8392 		return true;
8393 
8394 	if (vcpu->arch.pv.pv_unhalted)
8395 		return true;
8396 
8397 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8398 	    (vcpu->arch.nmi_pending &&
8399 	     kvm_x86_ops->nmi_allowed(vcpu)))
8400 		return true;
8401 
8402 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8403 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
8404 		return true;
8405 
8406 	if (kvm_arch_interrupt_allowed(vcpu) &&
8407 	    kvm_cpu_has_interrupt(vcpu))
8408 		return true;
8409 
8410 	if (kvm_hv_has_stimer_pending(vcpu))
8411 		return true;
8412 
8413 	return false;
8414 }
8415 
8416 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8417 {
8418 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8419 }
8420 
8421 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8422 {
8423 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8424 }
8425 
8426 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8427 {
8428 	return kvm_x86_ops->interrupt_allowed(vcpu);
8429 }
8430 
8431 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8432 {
8433 	if (is_64_bit_mode(vcpu))
8434 		return kvm_rip_read(vcpu);
8435 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8436 		     kvm_rip_read(vcpu));
8437 }
8438 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8439 
8440 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8441 {
8442 	return kvm_get_linear_rip(vcpu) == linear_rip;
8443 }
8444 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8445 
8446 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8447 {
8448 	unsigned long rflags;
8449 
8450 	rflags = kvm_x86_ops->get_rflags(vcpu);
8451 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8452 		rflags &= ~X86_EFLAGS_TF;
8453 	return rflags;
8454 }
8455 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8456 
8457 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8458 {
8459 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8460 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8461 		rflags |= X86_EFLAGS_TF;
8462 	kvm_x86_ops->set_rflags(vcpu, rflags);
8463 }
8464 
8465 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8466 {
8467 	__kvm_set_rflags(vcpu, rflags);
8468 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8469 }
8470 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8471 
8472 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8473 {
8474 	int r;
8475 
8476 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8477 	      work->wakeup_all)
8478 		return;
8479 
8480 	r = kvm_mmu_reload(vcpu);
8481 	if (unlikely(r))
8482 		return;
8483 
8484 	if (!vcpu->arch.mmu.direct_map &&
8485 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8486 		return;
8487 
8488 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8489 }
8490 
8491 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8492 {
8493 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8494 }
8495 
8496 static inline u32 kvm_async_pf_next_probe(u32 key)
8497 {
8498 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8499 }
8500 
8501 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8502 {
8503 	u32 key = kvm_async_pf_hash_fn(gfn);
8504 
8505 	while (vcpu->arch.apf.gfns[key] != ~0)
8506 		key = kvm_async_pf_next_probe(key);
8507 
8508 	vcpu->arch.apf.gfns[key] = gfn;
8509 }
8510 
8511 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8512 {
8513 	int i;
8514 	u32 key = kvm_async_pf_hash_fn(gfn);
8515 
8516 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8517 		     (vcpu->arch.apf.gfns[key] != gfn &&
8518 		      vcpu->arch.apf.gfns[key] != ~0); i++)
8519 		key = kvm_async_pf_next_probe(key);
8520 
8521 	return key;
8522 }
8523 
8524 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8525 {
8526 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8527 }
8528 
8529 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8530 {
8531 	u32 i, j, k;
8532 
8533 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8534 	while (true) {
8535 		vcpu->arch.apf.gfns[i] = ~0;
8536 		do {
8537 			j = kvm_async_pf_next_probe(j);
8538 			if (vcpu->arch.apf.gfns[j] == ~0)
8539 				return;
8540 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8541 			/*
8542 			 * k lies cyclically in ]i,j]
8543 			 * |    i.k.j |
8544 			 * |....j i.k.| or  |.k..j i...|
8545 			 */
8546 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8547 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8548 		i = j;
8549 	}
8550 }
8551 
8552 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8553 {
8554 
8555 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8556 				      sizeof(val));
8557 }
8558 
8559 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8560 				     struct kvm_async_pf *work)
8561 {
8562 	struct x86_exception fault;
8563 
8564 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8565 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8566 
8567 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8568 	    (vcpu->arch.apf.send_user_only &&
8569 	     kvm_x86_ops->get_cpl(vcpu) == 0))
8570 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8571 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8572 		fault.vector = PF_VECTOR;
8573 		fault.error_code_valid = true;
8574 		fault.error_code = 0;
8575 		fault.nested_page_fault = false;
8576 		fault.address = work->arch.token;
8577 		kvm_inject_page_fault(vcpu, &fault);
8578 	}
8579 }
8580 
8581 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8582 				 struct kvm_async_pf *work)
8583 {
8584 	struct x86_exception fault;
8585 
8586 	if (work->wakeup_all)
8587 		work->arch.token = ~0; /* broadcast wakeup */
8588 	else
8589 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8590 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8591 
8592 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8593 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8594 		fault.vector = PF_VECTOR;
8595 		fault.error_code_valid = true;
8596 		fault.error_code = 0;
8597 		fault.nested_page_fault = false;
8598 		fault.address = work->arch.token;
8599 		kvm_inject_page_fault(vcpu, &fault);
8600 	}
8601 	vcpu->arch.apf.halted = false;
8602 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8603 }
8604 
8605 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8606 {
8607 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8608 		return true;
8609 	else
8610 		return kvm_can_do_async_pf(vcpu);
8611 }
8612 
8613 void kvm_arch_start_assignment(struct kvm *kvm)
8614 {
8615 	atomic_inc(&kvm->arch.assigned_device_count);
8616 }
8617 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8618 
8619 void kvm_arch_end_assignment(struct kvm *kvm)
8620 {
8621 	atomic_dec(&kvm->arch.assigned_device_count);
8622 }
8623 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8624 
8625 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8626 {
8627 	return atomic_read(&kvm->arch.assigned_device_count);
8628 }
8629 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8630 
8631 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8632 {
8633 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8634 }
8635 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8636 
8637 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8638 {
8639 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8640 }
8641 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8642 
8643 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8644 {
8645 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8646 }
8647 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8648 
8649 bool kvm_arch_has_irq_bypass(void)
8650 {
8651 	return kvm_x86_ops->update_pi_irte != NULL;
8652 }
8653 
8654 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8655 				      struct irq_bypass_producer *prod)
8656 {
8657 	struct kvm_kernel_irqfd *irqfd =
8658 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8659 
8660 	irqfd->producer = prod;
8661 
8662 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8663 					   prod->irq, irqfd->gsi, 1);
8664 }
8665 
8666 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8667 				      struct irq_bypass_producer *prod)
8668 {
8669 	int ret;
8670 	struct kvm_kernel_irqfd *irqfd =
8671 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8672 
8673 	WARN_ON(irqfd->producer != prod);
8674 	irqfd->producer = NULL;
8675 
8676 	/*
8677 	 * When producer of consumer is unregistered, we change back to
8678 	 * remapped mode, so we can re-use the current implementation
8679 	 * when the irq is masked/disabled or the consumer side (KVM
8680 	 * int this case doesn't want to receive the interrupts.
8681 	*/
8682 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8683 	if (ret)
8684 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8685 		       " fails: %d\n", irqfd->consumer.token, ret);
8686 }
8687 
8688 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8689 				   uint32_t guest_irq, bool set)
8690 {
8691 	if (!kvm_x86_ops->update_pi_irte)
8692 		return -EINVAL;
8693 
8694 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8695 }
8696 
8697 bool kvm_vector_hashing_enabled(void)
8698 {
8699 	return vector_hashing;
8700 }
8701 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8702 
8703 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8704 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8705 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8706 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8707 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8708 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8709 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8710 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8711 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8712 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8713 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8714 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8715 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8716 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8717 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8718 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8719 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8720 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8721 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
8722