xref: /openbmc/linux/arch/x86/kvm/x86.c (revision fb960bd2)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
58 
59 #include <trace/events/kvm.h>
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 
71 #define CREATE_TRACE_POINTS
72 #include "trace.h"
73 
74 #define MAX_IO_MSRS 256
75 #define KVM_MAX_MCE_BANKS 32
76 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
77 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
78 
79 #define emul_to_vcpu(ctxt) \
80 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
81 
82 /* EFER defaults:
83  * - enable syscall per default because its emulated by KVM
84  * - enable LME and LMA per default on 64 bit KVM
85  */
86 #ifdef CONFIG_X86_64
87 static
88 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
89 #else
90 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
91 #endif
92 
93 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
95 
96 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
98 
99 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
100 static void process_nmi(struct kvm_vcpu *vcpu);
101 static void enter_smm(struct kvm_vcpu *vcpu);
102 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
103 
104 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
105 EXPORT_SYMBOL_GPL(kvm_x86_ops);
106 
107 static bool __read_mostly ignore_msrs = 0;
108 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
109 
110 static bool __read_mostly report_ignored_msrs = true;
111 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
112 
113 unsigned int min_timer_period_us = 500;
114 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
115 
116 static bool __read_mostly kvmclock_periodic_sync = true;
117 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
118 
119 bool __read_mostly kvm_has_tsc_control;
120 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
121 u32  __read_mostly kvm_max_guest_tsc_khz;
122 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
123 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
124 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
125 u64  __read_mostly kvm_max_tsc_scaling_ratio;
126 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
127 u64 __read_mostly kvm_default_tsc_scaling_ratio;
128 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
129 
130 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
131 static u32 __read_mostly tsc_tolerance_ppm = 250;
132 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
133 
134 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
135 unsigned int __read_mostly lapic_timer_advance_ns = 0;
136 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
137 
138 static bool __read_mostly vector_hashing = true;
139 module_param(vector_hashing, bool, S_IRUGO);
140 
141 #define KVM_NR_SHARED_MSRS 16
142 
143 struct kvm_shared_msrs_global {
144 	int nr;
145 	u32 msrs[KVM_NR_SHARED_MSRS];
146 };
147 
148 struct kvm_shared_msrs {
149 	struct user_return_notifier urn;
150 	bool registered;
151 	struct kvm_shared_msr_values {
152 		u64 host;
153 		u64 curr;
154 	} values[KVM_NR_SHARED_MSRS];
155 };
156 
157 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
158 static struct kvm_shared_msrs __percpu *shared_msrs;
159 
160 struct kvm_stats_debugfs_item debugfs_entries[] = {
161 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
162 	{ "pf_guest", VCPU_STAT(pf_guest) },
163 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
164 	{ "invlpg", VCPU_STAT(invlpg) },
165 	{ "exits", VCPU_STAT(exits) },
166 	{ "io_exits", VCPU_STAT(io_exits) },
167 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
168 	{ "signal_exits", VCPU_STAT(signal_exits) },
169 	{ "irq_window", VCPU_STAT(irq_window_exits) },
170 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
171 	{ "halt_exits", VCPU_STAT(halt_exits) },
172 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
173 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
174 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
175 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
176 	{ "hypercalls", VCPU_STAT(hypercalls) },
177 	{ "request_irq", VCPU_STAT(request_irq_exits) },
178 	{ "irq_exits", VCPU_STAT(irq_exits) },
179 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
180 	{ "efer_reload", VCPU_STAT(efer_reload) },
181 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
182 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
183 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
184 	{ "irq_injections", VCPU_STAT(irq_injections) },
185 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
186 	{ "req_event", VCPU_STAT(req_event) },
187 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
188 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
189 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
190 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
191 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
192 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
193 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
194 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
195 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
196 	{ "largepages", VM_STAT(lpages) },
197 	{ "max_mmu_page_hash_collisions",
198 		VM_STAT(max_mmu_page_hash_collisions) },
199 	{ NULL }
200 };
201 
202 u64 __read_mostly host_xcr0;
203 
204 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
205 
206 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
207 {
208 	int i;
209 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
210 		vcpu->arch.apf.gfns[i] = ~0;
211 }
212 
213 static void kvm_on_user_return(struct user_return_notifier *urn)
214 {
215 	unsigned slot;
216 	struct kvm_shared_msrs *locals
217 		= container_of(urn, struct kvm_shared_msrs, urn);
218 	struct kvm_shared_msr_values *values;
219 	unsigned long flags;
220 
221 	/*
222 	 * Disabling irqs at this point since the following code could be
223 	 * interrupted and executed through kvm_arch_hardware_disable()
224 	 */
225 	local_irq_save(flags);
226 	if (locals->registered) {
227 		locals->registered = false;
228 		user_return_notifier_unregister(urn);
229 	}
230 	local_irq_restore(flags);
231 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
232 		values = &locals->values[slot];
233 		if (values->host != values->curr) {
234 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
235 			values->curr = values->host;
236 		}
237 	}
238 }
239 
240 static void shared_msr_update(unsigned slot, u32 msr)
241 {
242 	u64 value;
243 	unsigned int cpu = smp_processor_id();
244 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
245 
246 	/* only read, and nobody should modify it at this time,
247 	 * so don't need lock */
248 	if (slot >= shared_msrs_global.nr) {
249 		printk(KERN_ERR "kvm: invalid MSR slot!");
250 		return;
251 	}
252 	rdmsrl_safe(msr, &value);
253 	smsr->values[slot].host = value;
254 	smsr->values[slot].curr = value;
255 }
256 
257 void kvm_define_shared_msr(unsigned slot, u32 msr)
258 {
259 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
260 	shared_msrs_global.msrs[slot] = msr;
261 	if (slot >= shared_msrs_global.nr)
262 		shared_msrs_global.nr = slot + 1;
263 }
264 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
265 
266 static void kvm_shared_msr_cpu_online(void)
267 {
268 	unsigned i;
269 
270 	for (i = 0; i < shared_msrs_global.nr; ++i)
271 		shared_msr_update(i, shared_msrs_global.msrs[i]);
272 }
273 
274 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
275 {
276 	unsigned int cpu = smp_processor_id();
277 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
278 	int err;
279 
280 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
281 		return 0;
282 	smsr->values[slot].curr = value;
283 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
284 	if (err)
285 		return 1;
286 
287 	if (!smsr->registered) {
288 		smsr->urn.on_user_return = kvm_on_user_return;
289 		user_return_notifier_register(&smsr->urn);
290 		smsr->registered = true;
291 	}
292 	return 0;
293 }
294 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
295 
296 static void drop_user_return_notifiers(void)
297 {
298 	unsigned int cpu = smp_processor_id();
299 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
300 
301 	if (smsr->registered)
302 		kvm_on_user_return(&smsr->urn);
303 }
304 
305 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
306 {
307 	return vcpu->arch.apic_base;
308 }
309 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
310 
311 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
312 {
313 	u64 old_state = vcpu->arch.apic_base &
314 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 	u64 new_state = msr_info->data &
316 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
317 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
318 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
319 
320 	if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
321 		return 1;
322 	if (!msr_info->host_initiated &&
323 	    ((new_state == MSR_IA32_APICBASE_ENABLE &&
324 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
325 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
326 	      old_state == 0)))
327 		return 1;
328 
329 	kvm_lapic_set_base(vcpu, msr_info->data);
330 	return 0;
331 }
332 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
333 
334 asmlinkage __visible void kvm_spurious_fault(void)
335 {
336 	/* Fault while not rebooting.  We want the trace. */
337 	BUG();
338 }
339 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
340 
341 #define EXCPT_BENIGN		0
342 #define EXCPT_CONTRIBUTORY	1
343 #define EXCPT_PF		2
344 
345 static int exception_class(int vector)
346 {
347 	switch (vector) {
348 	case PF_VECTOR:
349 		return EXCPT_PF;
350 	case DE_VECTOR:
351 	case TS_VECTOR:
352 	case NP_VECTOR:
353 	case SS_VECTOR:
354 	case GP_VECTOR:
355 		return EXCPT_CONTRIBUTORY;
356 	default:
357 		break;
358 	}
359 	return EXCPT_BENIGN;
360 }
361 
362 #define EXCPT_FAULT		0
363 #define EXCPT_TRAP		1
364 #define EXCPT_ABORT		2
365 #define EXCPT_INTERRUPT		3
366 
367 static int exception_type(int vector)
368 {
369 	unsigned int mask;
370 
371 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
372 		return EXCPT_INTERRUPT;
373 
374 	mask = 1 << vector;
375 
376 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
377 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
378 		return EXCPT_TRAP;
379 
380 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
381 		return EXCPT_ABORT;
382 
383 	/* Reserved exceptions will result in fault */
384 	return EXCPT_FAULT;
385 }
386 
387 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
388 		unsigned nr, bool has_error, u32 error_code,
389 		bool reinject)
390 {
391 	u32 prev_nr;
392 	int class1, class2;
393 
394 	kvm_make_request(KVM_REQ_EVENT, vcpu);
395 
396 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
397 	queue:
398 		if (has_error && !is_protmode(vcpu))
399 			has_error = false;
400 		if (reinject) {
401 			/*
402 			 * On vmentry, vcpu->arch.exception.pending is only
403 			 * true if an event injection was blocked by
404 			 * nested_run_pending.  In that case, however,
405 			 * vcpu_enter_guest requests an immediate exit,
406 			 * and the guest shouldn't proceed far enough to
407 			 * need reinjection.
408 			 */
409 			WARN_ON_ONCE(vcpu->arch.exception.pending);
410 			vcpu->arch.exception.injected = true;
411 		} else {
412 			vcpu->arch.exception.pending = true;
413 			vcpu->arch.exception.injected = false;
414 		}
415 		vcpu->arch.exception.has_error_code = has_error;
416 		vcpu->arch.exception.nr = nr;
417 		vcpu->arch.exception.error_code = error_code;
418 		return;
419 	}
420 
421 	/* to check exception */
422 	prev_nr = vcpu->arch.exception.nr;
423 	if (prev_nr == DF_VECTOR) {
424 		/* triple fault -> shutdown */
425 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
426 		return;
427 	}
428 	class1 = exception_class(prev_nr);
429 	class2 = exception_class(nr);
430 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
431 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
432 		/*
433 		 * Generate double fault per SDM Table 5-5.  Set
434 		 * exception.pending = true so that the double fault
435 		 * can trigger a nested vmexit.
436 		 */
437 		vcpu->arch.exception.pending = true;
438 		vcpu->arch.exception.injected = false;
439 		vcpu->arch.exception.has_error_code = true;
440 		vcpu->arch.exception.nr = DF_VECTOR;
441 		vcpu->arch.exception.error_code = 0;
442 	} else
443 		/* replace previous exception with a new one in a hope
444 		   that instruction re-execution will regenerate lost
445 		   exception */
446 		goto queue;
447 }
448 
449 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
450 {
451 	kvm_multiple_exception(vcpu, nr, false, 0, false);
452 }
453 EXPORT_SYMBOL_GPL(kvm_queue_exception);
454 
455 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
456 {
457 	kvm_multiple_exception(vcpu, nr, false, 0, true);
458 }
459 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
460 
461 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
462 {
463 	if (err)
464 		kvm_inject_gp(vcpu, 0);
465 	else
466 		return kvm_skip_emulated_instruction(vcpu);
467 
468 	return 1;
469 }
470 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
471 
472 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
473 {
474 	++vcpu->stat.pf_guest;
475 	vcpu->arch.exception.nested_apf =
476 		is_guest_mode(vcpu) && fault->async_page_fault;
477 	if (vcpu->arch.exception.nested_apf)
478 		vcpu->arch.apf.nested_apf_token = fault->address;
479 	else
480 		vcpu->arch.cr2 = fault->address;
481 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
482 }
483 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
484 
485 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
486 {
487 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
488 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
489 	else
490 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
491 
492 	return fault->nested_page_fault;
493 }
494 
495 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
496 {
497 	atomic_inc(&vcpu->arch.nmi_queued);
498 	kvm_make_request(KVM_REQ_NMI, vcpu);
499 }
500 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
501 
502 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
503 {
504 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
505 }
506 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
507 
508 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
509 {
510 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
511 }
512 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
513 
514 /*
515  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
516  * a #GP and return false.
517  */
518 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
519 {
520 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
521 		return true;
522 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
523 	return false;
524 }
525 EXPORT_SYMBOL_GPL(kvm_require_cpl);
526 
527 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
528 {
529 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
530 		return true;
531 
532 	kvm_queue_exception(vcpu, UD_VECTOR);
533 	return false;
534 }
535 EXPORT_SYMBOL_GPL(kvm_require_dr);
536 
537 /*
538  * This function will be used to read from the physical memory of the currently
539  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
540  * can read from guest physical or from the guest's guest physical memory.
541  */
542 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
543 			    gfn_t ngfn, void *data, int offset, int len,
544 			    u32 access)
545 {
546 	struct x86_exception exception;
547 	gfn_t real_gfn;
548 	gpa_t ngpa;
549 
550 	ngpa     = gfn_to_gpa(ngfn);
551 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
552 	if (real_gfn == UNMAPPED_GVA)
553 		return -EFAULT;
554 
555 	real_gfn = gpa_to_gfn(real_gfn);
556 
557 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
558 }
559 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
560 
561 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
562 			       void *data, int offset, int len, u32 access)
563 {
564 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
565 				       data, offset, len, access);
566 }
567 
568 /*
569  * Load the pae pdptrs.  Return true is they are all valid.
570  */
571 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
572 {
573 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
574 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
575 	int i;
576 	int ret;
577 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
578 
579 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
580 				      offset * sizeof(u64), sizeof(pdpte),
581 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
582 	if (ret < 0) {
583 		ret = 0;
584 		goto out;
585 	}
586 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
587 		if ((pdpte[i] & PT_PRESENT_MASK) &&
588 		    (pdpte[i] &
589 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
590 			ret = 0;
591 			goto out;
592 		}
593 	}
594 	ret = 1;
595 
596 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
597 	__set_bit(VCPU_EXREG_PDPTR,
598 		  (unsigned long *)&vcpu->arch.regs_avail);
599 	__set_bit(VCPU_EXREG_PDPTR,
600 		  (unsigned long *)&vcpu->arch.regs_dirty);
601 out:
602 
603 	return ret;
604 }
605 EXPORT_SYMBOL_GPL(load_pdptrs);
606 
607 bool pdptrs_changed(struct kvm_vcpu *vcpu)
608 {
609 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
610 	bool changed = true;
611 	int offset;
612 	gfn_t gfn;
613 	int r;
614 
615 	if (is_long_mode(vcpu) || !is_pae(vcpu))
616 		return false;
617 
618 	if (!test_bit(VCPU_EXREG_PDPTR,
619 		      (unsigned long *)&vcpu->arch.regs_avail))
620 		return true;
621 
622 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
623 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
624 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
625 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
626 	if (r < 0)
627 		goto out;
628 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
629 out:
630 
631 	return changed;
632 }
633 EXPORT_SYMBOL_GPL(pdptrs_changed);
634 
635 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
636 {
637 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
638 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
639 
640 	cr0 |= X86_CR0_ET;
641 
642 #ifdef CONFIG_X86_64
643 	if (cr0 & 0xffffffff00000000UL)
644 		return 1;
645 #endif
646 
647 	cr0 &= ~CR0_RESERVED_BITS;
648 
649 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
650 		return 1;
651 
652 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
653 		return 1;
654 
655 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
656 #ifdef CONFIG_X86_64
657 		if ((vcpu->arch.efer & EFER_LME)) {
658 			int cs_db, cs_l;
659 
660 			if (!is_pae(vcpu))
661 				return 1;
662 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
663 			if (cs_l)
664 				return 1;
665 		} else
666 #endif
667 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
668 						 kvm_read_cr3(vcpu)))
669 			return 1;
670 	}
671 
672 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
673 		return 1;
674 
675 	kvm_x86_ops->set_cr0(vcpu, cr0);
676 
677 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
678 		kvm_clear_async_pf_completion_queue(vcpu);
679 		kvm_async_pf_hash_reset(vcpu);
680 	}
681 
682 	if ((cr0 ^ old_cr0) & update_bits)
683 		kvm_mmu_reset_context(vcpu);
684 
685 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
686 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
687 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
688 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
689 
690 	return 0;
691 }
692 EXPORT_SYMBOL_GPL(kvm_set_cr0);
693 
694 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
695 {
696 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
697 }
698 EXPORT_SYMBOL_GPL(kvm_lmsw);
699 
700 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
701 {
702 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
703 			!vcpu->guest_xcr0_loaded) {
704 		/* kvm_set_xcr() also depends on this */
705 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
706 		vcpu->guest_xcr0_loaded = 1;
707 	}
708 }
709 
710 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
711 {
712 	if (vcpu->guest_xcr0_loaded) {
713 		if (vcpu->arch.xcr0 != host_xcr0)
714 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
715 		vcpu->guest_xcr0_loaded = 0;
716 	}
717 }
718 
719 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
720 {
721 	u64 xcr0 = xcr;
722 	u64 old_xcr0 = vcpu->arch.xcr0;
723 	u64 valid_bits;
724 
725 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
726 	if (index != XCR_XFEATURE_ENABLED_MASK)
727 		return 1;
728 	if (!(xcr0 & XFEATURE_MASK_FP))
729 		return 1;
730 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
731 		return 1;
732 
733 	/*
734 	 * Do not allow the guest to set bits that we do not support
735 	 * saving.  However, xcr0 bit 0 is always set, even if the
736 	 * emulated CPU does not support XSAVE (see fx_init).
737 	 */
738 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
739 	if (xcr0 & ~valid_bits)
740 		return 1;
741 
742 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
743 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
744 		return 1;
745 
746 	if (xcr0 & XFEATURE_MASK_AVX512) {
747 		if (!(xcr0 & XFEATURE_MASK_YMM))
748 			return 1;
749 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
750 			return 1;
751 	}
752 	vcpu->arch.xcr0 = xcr0;
753 
754 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
755 		kvm_update_cpuid(vcpu);
756 	return 0;
757 }
758 
759 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
760 {
761 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
762 	    __kvm_set_xcr(vcpu, index, xcr)) {
763 		kvm_inject_gp(vcpu, 0);
764 		return 1;
765 	}
766 	return 0;
767 }
768 EXPORT_SYMBOL_GPL(kvm_set_xcr);
769 
770 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
771 {
772 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
773 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
774 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
775 
776 	if (cr4 & CR4_RESERVED_BITS)
777 		return 1;
778 
779 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
780 		return 1;
781 
782 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
783 		return 1;
784 
785 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
786 		return 1;
787 
788 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
789 		return 1;
790 
791 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
792 		return 1;
793 
794 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
795 		return 1;
796 
797 	if (is_long_mode(vcpu)) {
798 		if (!(cr4 & X86_CR4_PAE))
799 			return 1;
800 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
801 		   && ((cr4 ^ old_cr4) & pdptr_bits)
802 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
803 				   kvm_read_cr3(vcpu)))
804 		return 1;
805 
806 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
807 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
808 			return 1;
809 
810 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
811 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
812 			return 1;
813 	}
814 
815 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
816 		return 1;
817 
818 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
819 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
820 		kvm_mmu_reset_context(vcpu);
821 
822 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
823 		kvm_update_cpuid(vcpu);
824 
825 	return 0;
826 }
827 EXPORT_SYMBOL_GPL(kvm_set_cr4);
828 
829 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
830 {
831 #ifdef CONFIG_X86_64
832 	cr3 &= ~CR3_PCID_INVD;
833 #endif
834 
835 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
836 		kvm_mmu_sync_roots(vcpu);
837 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
838 		return 0;
839 	}
840 
841 	if (is_long_mode(vcpu) &&
842 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
843 		return 1;
844 	else if (is_pae(vcpu) && is_paging(vcpu) &&
845 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
846 		return 1;
847 
848 	vcpu->arch.cr3 = cr3;
849 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
850 	kvm_mmu_new_cr3(vcpu);
851 	return 0;
852 }
853 EXPORT_SYMBOL_GPL(kvm_set_cr3);
854 
855 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
856 {
857 	if (cr8 & CR8_RESERVED_BITS)
858 		return 1;
859 	if (lapic_in_kernel(vcpu))
860 		kvm_lapic_set_tpr(vcpu, cr8);
861 	else
862 		vcpu->arch.cr8 = cr8;
863 	return 0;
864 }
865 EXPORT_SYMBOL_GPL(kvm_set_cr8);
866 
867 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
868 {
869 	if (lapic_in_kernel(vcpu))
870 		return kvm_lapic_get_cr8(vcpu);
871 	else
872 		return vcpu->arch.cr8;
873 }
874 EXPORT_SYMBOL_GPL(kvm_get_cr8);
875 
876 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
877 {
878 	int i;
879 
880 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
881 		for (i = 0; i < KVM_NR_DB_REGS; i++)
882 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
883 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
884 	}
885 }
886 
887 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
888 {
889 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
890 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
891 }
892 
893 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
894 {
895 	unsigned long dr7;
896 
897 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
898 		dr7 = vcpu->arch.guest_debug_dr7;
899 	else
900 		dr7 = vcpu->arch.dr7;
901 	kvm_x86_ops->set_dr7(vcpu, dr7);
902 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
903 	if (dr7 & DR7_BP_EN_MASK)
904 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
905 }
906 
907 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
908 {
909 	u64 fixed = DR6_FIXED_1;
910 
911 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
912 		fixed |= DR6_RTM;
913 	return fixed;
914 }
915 
916 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
917 {
918 	switch (dr) {
919 	case 0 ... 3:
920 		vcpu->arch.db[dr] = val;
921 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
922 			vcpu->arch.eff_db[dr] = val;
923 		break;
924 	case 4:
925 		/* fall through */
926 	case 6:
927 		if (val & 0xffffffff00000000ULL)
928 			return -1; /* #GP */
929 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
930 		kvm_update_dr6(vcpu);
931 		break;
932 	case 5:
933 		/* fall through */
934 	default: /* 7 */
935 		if (val & 0xffffffff00000000ULL)
936 			return -1; /* #GP */
937 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
938 		kvm_update_dr7(vcpu);
939 		break;
940 	}
941 
942 	return 0;
943 }
944 
945 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
946 {
947 	if (__kvm_set_dr(vcpu, dr, val)) {
948 		kvm_inject_gp(vcpu, 0);
949 		return 1;
950 	}
951 	return 0;
952 }
953 EXPORT_SYMBOL_GPL(kvm_set_dr);
954 
955 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
956 {
957 	switch (dr) {
958 	case 0 ... 3:
959 		*val = vcpu->arch.db[dr];
960 		break;
961 	case 4:
962 		/* fall through */
963 	case 6:
964 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
965 			*val = vcpu->arch.dr6;
966 		else
967 			*val = kvm_x86_ops->get_dr6(vcpu);
968 		break;
969 	case 5:
970 		/* fall through */
971 	default: /* 7 */
972 		*val = vcpu->arch.dr7;
973 		break;
974 	}
975 	return 0;
976 }
977 EXPORT_SYMBOL_GPL(kvm_get_dr);
978 
979 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
980 {
981 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
982 	u64 data;
983 	int err;
984 
985 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
986 	if (err)
987 		return err;
988 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
989 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
990 	return err;
991 }
992 EXPORT_SYMBOL_GPL(kvm_rdpmc);
993 
994 /*
995  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
996  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
997  *
998  * This list is modified at module load time to reflect the
999  * capabilities of the host cpu. This capabilities test skips MSRs that are
1000  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1001  * may depend on host virtualization features rather than host cpu features.
1002  */
1003 
1004 static u32 msrs_to_save[] = {
1005 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1006 	MSR_STAR,
1007 #ifdef CONFIG_X86_64
1008 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1009 #endif
1010 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1011 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1012 };
1013 
1014 static unsigned num_msrs_to_save;
1015 
1016 static u32 emulated_msrs[] = {
1017 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1018 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1019 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1020 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1021 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1022 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1023 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1024 	HV_X64_MSR_RESET,
1025 	HV_X64_MSR_VP_INDEX,
1026 	HV_X64_MSR_VP_RUNTIME,
1027 	HV_X64_MSR_SCONTROL,
1028 	HV_X64_MSR_STIMER0_CONFIG,
1029 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1030 	MSR_KVM_PV_EOI_EN,
1031 
1032 	MSR_IA32_TSC_ADJUST,
1033 	MSR_IA32_TSCDEADLINE,
1034 	MSR_IA32_MISC_ENABLE,
1035 	MSR_IA32_MCG_STATUS,
1036 	MSR_IA32_MCG_CTL,
1037 	MSR_IA32_MCG_EXT_CTL,
1038 	MSR_IA32_SMBASE,
1039 	MSR_PLATFORM_INFO,
1040 	MSR_MISC_FEATURES_ENABLES,
1041 };
1042 
1043 static unsigned num_emulated_msrs;
1044 
1045 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1046 {
1047 	if (efer & efer_reserved_bits)
1048 		return false;
1049 
1050 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1051 			return false;
1052 
1053 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1054 			return false;
1055 
1056 	return true;
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1059 
1060 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1061 {
1062 	u64 old_efer = vcpu->arch.efer;
1063 
1064 	if (!kvm_valid_efer(vcpu, efer))
1065 		return 1;
1066 
1067 	if (is_paging(vcpu)
1068 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1069 		return 1;
1070 
1071 	efer &= ~EFER_LMA;
1072 	efer |= vcpu->arch.efer & EFER_LMA;
1073 
1074 	kvm_x86_ops->set_efer(vcpu, efer);
1075 
1076 	/* Update reserved bits */
1077 	if ((efer ^ old_efer) & EFER_NX)
1078 		kvm_mmu_reset_context(vcpu);
1079 
1080 	return 0;
1081 }
1082 
1083 void kvm_enable_efer_bits(u64 mask)
1084 {
1085        efer_reserved_bits &= ~mask;
1086 }
1087 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1088 
1089 /*
1090  * Writes msr value into into the appropriate "register".
1091  * Returns 0 on success, non-0 otherwise.
1092  * Assumes vcpu_load() was already called.
1093  */
1094 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1095 {
1096 	switch (msr->index) {
1097 	case MSR_FS_BASE:
1098 	case MSR_GS_BASE:
1099 	case MSR_KERNEL_GS_BASE:
1100 	case MSR_CSTAR:
1101 	case MSR_LSTAR:
1102 		if (is_noncanonical_address(msr->data, vcpu))
1103 			return 1;
1104 		break;
1105 	case MSR_IA32_SYSENTER_EIP:
1106 	case MSR_IA32_SYSENTER_ESP:
1107 		/*
1108 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1109 		 * non-canonical address is written on Intel but not on
1110 		 * AMD (which ignores the top 32-bits, because it does
1111 		 * not implement 64-bit SYSENTER).
1112 		 *
1113 		 * 64-bit code should hence be able to write a non-canonical
1114 		 * value on AMD.  Making the address canonical ensures that
1115 		 * vmentry does not fail on Intel after writing a non-canonical
1116 		 * value, and that something deterministic happens if the guest
1117 		 * invokes 64-bit SYSENTER.
1118 		 */
1119 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1120 	}
1121 	return kvm_x86_ops->set_msr(vcpu, msr);
1122 }
1123 EXPORT_SYMBOL_GPL(kvm_set_msr);
1124 
1125 /*
1126  * Adapt set_msr() to msr_io()'s calling convention
1127  */
1128 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1129 {
1130 	struct msr_data msr;
1131 	int r;
1132 
1133 	msr.index = index;
1134 	msr.host_initiated = true;
1135 	r = kvm_get_msr(vcpu, &msr);
1136 	if (r)
1137 		return r;
1138 
1139 	*data = msr.data;
1140 	return 0;
1141 }
1142 
1143 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1144 {
1145 	struct msr_data msr;
1146 
1147 	msr.data = *data;
1148 	msr.index = index;
1149 	msr.host_initiated = true;
1150 	return kvm_set_msr(vcpu, &msr);
1151 }
1152 
1153 #ifdef CONFIG_X86_64
1154 struct pvclock_gtod_data {
1155 	seqcount_t	seq;
1156 
1157 	struct { /* extract of a clocksource struct */
1158 		int vclock_mode;
1159 		u64	cycle_last;
1160 		u64	mask;
1161 		u32	mult;
1162 		u32	shift;
1163 	} clock;
1164 
1165 	u64		boot_ns;
1166 	u64		nsec_base;
1167 	u64		wall_time_sec;
1168 };
1169 
1170 static struct pvclock_gtod_data pvclock_gtod_data;
1171 
1172 static void update_pvclock_gtod(struct timekeeper *tk)
1173 {
1174 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1175 	u64 boot_ns;
1176 
1177 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1178 
1179 	write_seqcount_begin(&vdata->seq);
1180 
1181 	/* copy pvclock gtod data */
1182 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1183 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1184 	vdata->clock.mask		= tk->tkr_mono.mask;
1185 	vdata->clock.mult		= tk->tkr_mono.mult;
1186 	vdata->clock.shift		= tk->tkr_mono.shift;
1187 
1188 	vdata->boot_ns			= boot_ns;
1189 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1190 
1191 	vdata->wall_time_sec            = tk->xtime_sec;
1192 
1193 	write_seqcount_end(&vdata->seq);
1194 }
1195 #endif
1196 
1197 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1198 {
1199 	/*
1200 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1201 	 * vcpu_enter_guest.  This function is only called from
1202 	 * the physical CPU that is running vcpu.
1203 	 */
1204 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1205 }
1206 
1207 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1208 {
1209 	int version;
1210 	int r;
1211 	struct pvclock_wall_clock wc;
1212 	struct timespec64 boot;
1213 
1214 	if (!wall_clock)
1215 		return;
1216 
1217 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1218 	if (r)
1219 		return;
1220 
1221 	if (version & 1)
1222 		++version;  /* first time write, random junk */
1223 
1224 	++version;
1225 
1226 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1227 		return;
1228 
1229 	/*
1230 	 * The guest calculates current wall clock time by adding
1231 	 * system time (updated by kvm_guest_time_update below) to the
1232 	 * wall clock specified here.  guest system time equals host
1233 	 * system time for us, thus we must fill in host boot time here.
1234 	 */
1235 	getboottime64(&boot);
1236 
1237 	if (kvm->arch.kvmclock_offset) {
1238 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1239 		boot = timespec64_sub(boot, ts);
1240 	}
1241 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1242 	wc.nsec = boot.tv_nsec;
1243 	wc.version = version;
1244 
1245 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1246 
1247 	version++;
1248 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1249 }
1250 
1251 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1252 {
1253 	do_shl32_div32(dividend, divisor);
1254 	return dividend;
1255 }
1256 
1257 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1258 			       s8 *pshift, u32 *pmultiplier)
1259 {
1260 	uint64_t scaled64;
1261 	int32_t  shift = 0;
1262 	uint64_t tps64;
1263 	uint32_t tps32;
1264 
1265 	tps64 = base_hz;
1266 	scaled64 = scaled_hz;
1267 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1268 		tps64 >>= 1;
1269 		shift--;
1270 	}
1271 
1272 	tps32 = (uint32_t)tps64;
1273 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1274 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1275 			scaled64 >>= 1;
1276 		else
1277 			tps32 <<= 1;
1278 		shift++;
1279 	}
1280 
1281 	*pshift = shift;
1282 	*pmultiplier = div_frac(scaled64, tps32);
1283 
1284 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1285 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1286 }
1287 
1288 #ifdef CONFIG_X86_64
1289 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1290 #endif
1291 
1292 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1293 static unsigned long max_tsc_khz;
1294 
1295 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1296 {
1297 	u64 v = (u64)khz * (1000000 + ppm);
1298 	do_div(v, 1000000);
1299 	return v;
1300 }
1301 
1302 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1303 {
1304 	u64 ratio;
1305 
1306 	/* Guest TSC same frequency as host TSC? */
1307 	if (!scale) {
1308 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1309 		return 0;
1310 	}
1311 
1312 	/* TSC scaling supported? */
1313 	if (!kvm_has_tsc_control) {
1314 		if (user_tsc_khz > tsc_khz) {
1315 			vcpu->arch.tsc_catchup = 1;
1316 			vcpu->arch.tsc_always_catchup = 1;
1317 			return 0;
1318 		} else {
1319 			WARN(1, "user requested TSC rate below hardware speed\n");
1320 			return -1;
1321 		}
1322 	}
1323 
1324 	/* TSC scaling required  - calculate ratio */
1325 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1326 				user_tsc_khz, tsc_khz);
1327 
1328 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1329 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1330 			  user_tsc_khz);
1331 		return -1;
1332 	}
1333 
1334 	vcpu->arch.tsc_scaling_ratio = ratio;
1335 	return 0;
1336 }
1337 
1338 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1339 {
1340 	u32 thresh_lo, thresh_hi;
1341 	int use_scaling = 0;
1342 
1343 	/* tsc_khz can be zero if TSC calibration fails */
1344 	if (user_tsc_khz == 0) {
1345 		/* set tsc_scaling_ratio to a safe value */
1346 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1347 		return -1;
1348 	}
1349 
1350 	/* Compute a scale to convert nanoseconds in TSC cycles */
1351 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1352 			   &vcpu->arch.virtual_tsc_shift,
1353 			   &vcpu->arch.virtual_tsc_mult);
1354 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1355 
1356 	/*
1357 	 * Compute the variation in TSC rate which is acceptable
1358 	 * within the range of tolerance and decide if the
1359 	 * rate being applied is within that bounds of the hardware
1360 	 * rate.  If so, no scaling or compensation need be done.
1361 	 */
1362 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1363 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1364 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1365 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1366 		use_scaling = 1;
1367 	}
1368 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1369 }
1370 
1371 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1372 {
1373 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1374 				      vcpu->arch.virtual_tsc_mult,
1375 				      vcpu->arch.virtual_tsc_shift);
1376 	tsc += vcpu->arch.this_tsc_write;
1377 	return tsc;
1378 }
1379 
1380 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1381 {
1382 #ifdef CONFIG_X86_64
1383 	bool vcpus_matched;
1384 	struct kvm_arch *ka = &vcpu->kvm->arch;
1385 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1386 
1387 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1388 			 atomic_read(&vcpu->kvm->online_vcpus));
1389 
1390 	/*
1391 	 * Once the masterclock is enabled, always perform request in
1392 	 * order to update it.
1393 	 *
1394 	 * In order to enable masterclock, the host clocksource must be TSC
1395 	 * and the vcpus need to have matched TSCs.  When that happens,
1396 	 * perform request to enable masterclock.
1397 	 */
1398 	if (ka->use_master_clock ||
1399 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1400 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1401 
1402 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1403 			    atomic_read(&vcpu->kvm->online_vcpus),
1404 		            ka->use_master_clock, gtod->clock.vclock_mode);
1405 #endif
1406 }
1407 
1408 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1409 {
1410 	u64 curr_offset = vcpu->arch.tsc_offset;
1411 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1412 }
1413 
1414 /*
1415  * Multiply tsc by a fixed point number represented by ratio.
1416  *
1417  * The most significant 64-N bits (mult) of ratio represent the
1418  * integral part of the fixed point number; the remaining N bits
1419  * (frac) represent the fractional part, ie. ratio represents a fixed
1420  * point number (mult + frac * 2^(-N)).
1421  *
1422  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1423  */
1424 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1425 {
1426 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1427 }
1428 
1429 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1430 {
1431 	u64 _tsc = tsc;
1432 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1433 
1434 	if (ratio != kvm_default_tsc_scaling_ratio)
1435 		_tsc = __scale_tsc(ratio, tsc);
1436 
1437 	return _tsc;
1438 }
1439 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1440 
1441 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1442 {
1443 	u64 tsc;
1444 
1445 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1446 
1447 	return target_tsc - tsc;
1448 }
1449 
1450 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1451 {
1452 	return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1453 }
1454 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1455 
1456 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1457 {
1458 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1459 	vcpu->arch.tsc_offset = offset;
1460 }
1461 
1462 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1463 {
1464 	struct kvm *kvm = vcpu->kvm;
1465 	u64 offset, ns, elapsed;
1466 	unsigned long flags;
1467 	bool matched;
1468 	bool already_matched;
1469 	u64 data = msr->data;
1470 	bool synchronizing = false;
1471 
1472 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1473 	offset = kvm_compute_tsc_offset(vcpu, data);
1474 	ns = ktime_get_boot_ns();
1475 	elapsed = ns - kvm->arch.last_tsc_nsec;
1476 
1477 	if (vcpu->arch.virtual_tsc_khz) {
1478 		if (data == 0 && msr->host_initiated) {
1479 			/*
1480 			 * detection of vcpu initialization -- need to sync
1481 			 * with other vCPUs. This particularly helps to keep
1482 			 * kvm_clock stable after CPU hotplug
1483 			 */
1484 			synchronizing = true;
1485 		} else {
1486 			u64 tsc_exp = kvm->arch.last_tsc_write +
1487 						nsec_to_cycles(vcpu, elapsed);
1488 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1489 			/*
1490 			 * Special case: TSC write with a small delta (1 second)
1491 			 * of virtual cycle time against real time is
1492 			 * interpreted as an attempt to synchronize the CPU.
1493 			 */
1494 			synchronizing = data < tsc_exp + tsc_hz &&
1495 					data + tsc_hz > tsc_exp;
1496 		}
1497 	}
1498 
1499 	/*
1500 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1501 	 * TSC, we add elapsed time in this computation.  We could let the
1502 	 * compensation code attempt to catch up if we fall behind, but
1503 	 * it's better to try to match offsets from the beginning.
1504          */
1505 	if (synchronizing &&
1506 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1507 		if (!check_tsc_unstable()) {
1508 			offset = kvm->arch.cur_tsc_offset;
1509 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1510 		} else {
1511 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1512 			data += delta;
1513 			offset = kvm_compute_tsc_offset(vcpu, data);
1514 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1515 		}
1516 		matched = true;
1517 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1518 	} else {
1519 		/*
1520 		 * We split periods of matched TSC writes into generations.
1521 		 * For each generation, we track the original measured
1522 		 * nanosecond time, offset, and write, so if TSCs are in
1523 		 * sync, we can match exact offset, and if not, we can match
1524 		 * exact software computation in compute_guest_tsc()
1525 		 *
1526 		 * These values are tracked in kvm->arch.cur_xxx variables.
1527 		 */
1528 		kvm->arch.cur_tsc_generation++;
1529 		kvm->arch.cur_tsc_nsec = ns;
1530 		kvm->arch.cur_tsc_write = data;
1531 		kvm->arch.cur_tsc_offset = offset;
1532 		matched = false;
1533 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1534 			 kvm->arch.cur_tsc_generation, data);
1535 	}
1536 
1537 	/*
1538 	 * We also track th most recent recorded KHZ, write and time to
1539 	 * allow the matching interval to be extended at each write.
1540 	 */
1541 	kvm->arch.last_tsc_nsec = ns;
1542 	kvm->arch.last_tsc_write = data;
1543 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1544 
1545 	vcpu->arch.last_guest_tsc = data;
1546 
1547 	/* Keep track of which generation this VCPU has synchronized to */
1548 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1549 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1550 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1551 
1552 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1553 		update_ia32_tsc_adjust_msr(vcpu, offset);
1554 
1555 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1556 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1557 
1558 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1559 	if (!matched) {
1560 		kvm->arch.nr_vcpus_matched_tsc = 0;
1561 	} else if (!already_matched) {
1562 		kvm->arch.nr_vcpus_matched_tsc++;
1563 	}
1564 
1565 	kvm_track_tsc_matching(vcpu);
1566 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1567 }
1568 
1569 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1570 
1571 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1572 					   s64 adjustment)
1573 {
1574 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1575 }
1576 
1577 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1578 {
1579 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1580 		WARN_ON(adjustment < 0);
1581 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1582 	adjust_tsc_offset_guest(vcpu, adjustment);
1583 }
1584 
1585 #ifdef CONFIG_X86_64
1586 
1587 static u64 read_tsc(void)
1588 {
1589 	u64 ret = (u64)rdtsc_ordered();
1590 	u64 last = pvclock_gtod_data.clock.cycle_last;
1591 
1592 	if (likely(ret >= last))
1593 		return ret;
1594 
1595 	/*
1596 	 * GCC likes to generate cmov here, but this branch is extremely
1597 	 * predictable (it's just a function of time and the likely is
1598 	 * very likely) and there's a data dependence, so force GCC
1599 	 * to generate a branch instead.  I don't barrier() because
1600 	 * we don't actually need a barrier, and if this function
1601 	 * ever gets inlined it will generate worse code.
1602 	 */
1603 	asm volatile ("");
1604 	return last;
1605 }
1606 
1607 static inline u64 vgettsc(u64 *cycle_now)
1608 {
1609 	long v;
1610 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1611 
1612 	*cycle_now = read_tsc();
1613 
1614 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1615 	return v * gtod->clock.mult;
1616 }
1617 
1618 static int do_monotonic_boot(s64 *t, 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 		ns = gtod->nsec_base;
1629 		ns += vgettsc(cycle_now);
1630 		ns >>= gtod->clock.shift;
1631 		ns += gtod->boot_ns;
1632 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1633 	*t = ns;
1634 
1635 	return mode;
1636 }
1637 
1638 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1639 {
1640 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1641 	unsigned long seq;
1642 	int mode;
1643 	u64 ns;
1644 
1645 	do {
1646 		seq = read_seqcount_begin(&gtod->seq);
1647 		mode = gtod->clock.vclock_mode;
1648 		ts->tv_sec = gtod->wall_time_sec;
1649 		ns = gtod->nsec_base;
1650 		ns += vgettsc(cycle_now);
1651 		ns >>= gtod->clock.shift;
1652 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1653 
1654 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1655 	ts->tv_nsec = ns;
1656 
1657 	return mode;
1658 }
1659 
1660 /* returns true if host is using tsc clocksource */
1661 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1662 {
1663 	/* checked again under seqlock below */
1664 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1665 		return false;
1666 
1667 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1668 }
1669 
1670 /* returns true if host is using tsc clocksource */
1671 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1672 					   u64 *cycle_now)
1673 {
1674 	/* checked again under seqlock below */
1675 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1676 		return false;
1677 
1678 	return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1679 }
1680 #endif
1681 
1682 /*
1683  *
1684  * Assuming a stable TSC across physical CPUS, and a stable TSC
1685  * across virtual CPUs, the following condition is possible.
1686  * Each numbered line represents an event visible to both
1687  * CPUs at the next numbered event.
1688  *
1689  * "timespecX" represents host monotonic time. "tscX" represents
1690  * RDTSC value.
1691  *
1692  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1693  *
1694  * 1.  read timespec0,tsc0
1695  * 2.					| timespec1 = timespec0 + N
1696  * 					| tsc1 = tsc0 + M
1697  * 3. transition to guest		| transition to guest
1698  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1699  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1700  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1701  *
1702  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1703  *
1704  * 	- ret0 < ret1
1705  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1706  *		...
1707  *	- 0 < N - M => M < N
1708  *
1709  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1710  * always the case (the difference between two distinct xtime instances
1711  * might be smaller then the difference between corresponding TSC reads,
1712  * when updating guest vcpus pvclock areas).
1713  *
1714  * To avoid that problem, do not allow visibility of distinct
1715  * system_timestamp/tsc_timestamp values simultaneously: use a master
1716  * copy of host monotonic time values. Update that master copy
1717  * in lockstep.
1718  *
1719  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1720  *
1721  */
1722 
1723 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1724 {
1725 #ifdef CONFIG_X86_64
1726 	struct kvm_arch *ka = &kvm->arch;
1727 	int vclock_mode;
1728 	bool host_tsc_clocksource, vcpus_matched;
1729 
1730 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1731 			atomic_read(&kvm->online_vcpus));
1732 
1733 	/*
1734 	 * If the host uses TSC clock, then passthrough TSC as stable
1735 	 * to the guest.
1736 	 */
1737 	host_tsc_clocksource = kvm_get_time_and_clockread(
1738 					&ka->master_kernel_ns,
1739 					&ka->master_cycle_now);
1740 
1741 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1742 				&& !ka->backwards_tsc_observed
1743 				&& !ka->boot_vcpu_runs_old_kvmclock;
1744 
1745 	if (ka->use_master_clock)
1746 		atomic_set(&kvm_guest_has_master_clock, 1);
1747 
1748 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1749 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1750 					vcpus_matched);
1751 #endif
1752 }
1753 
1754 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1755 {
1756 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1757 }
1758 
1759 static void kvm_gen_update_masterclock(struct kvm *kvm)
1760 {
1761 #ifdef CONFIG_X86_64
1762 	int i;
1763 	struct kvm_vcpu *vcpu;
1764 	struct kvm_arch *ka = &kvm->arch;
1765 
1766 	spin_lock(&ka->pvclock_gtod_sync_lock);
1767 	kvm_make_mclock_inprogress_request(kvm);
1768 	/* no guest entries from this point */
1769 	pvclock_update_vm_gtod_copy(kvm);
1770 
1771 	kvm_for_each_vcpu(i, vcpu, kvm)
1772 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1773 
1774 	/* guest entries allowed */
1775 	kvm_for_each_vcpu(i, vcpu, kvm)
1776 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1777 
1778 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1779 #endif
1780 }
1781 
1782 u64 get_kvmclock_ns(struct kvm *kvm)
1783 {
1784 	struct kvm_arch *ka = &kvm->arch;
1785 	struct pvclock_vcpu_time_info hv_clock;
1786 	u64 ret;
1787 
1788 	spin_lock(&ka->pvclock_gtod_sync_lock);
1789 	if (!ka->use_master_clock) {
1790 		spin_unlock(&ka->pvclock_gtod_sync_lock);
1791 		return ktime_get_boot_ns() + ka->kvmclock_offset;
1792 	}
1793 
1794 	hv_clock.tsc_timestamp = ka->master_cycle_now;
1795 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1796 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1797 
1798 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
1799 	get_cpu();
1800 
1801 	if (__this_cpu_read(cpu_tsc_khz)) {
1802 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1803 				   &hv_clock.tsc_shift,
1804 				   &hv_clock.tsc_to_system_mul);
1805 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1806 	} else
1807 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1808 
1809 	put_cpu();
1810 
1811 	return ret;
1812 }
1813 
1814 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1815 {
1816 	struct kvm_vcpu_arch *vcpu = &v->arch;
1817 	struct pvclock_vcpu_time_info guest_hv_clock;
1818 
1819 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1820 		&guest_hv_clock, sizeof(guest_hv_clock))))
1821 		return;
1822 
1823 	/* This VCPU is paused, but it's legal for a guest to read another
1824 	 * VCPU's kvmclock, so we really have to follow the specification where
1825 	 * it says that version is odd if data is being modified, and even after
1826 	 * it is consistent.
1827 	 *
1828 	 * Version field updates must be kept separate.  This is because
1829 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1830 	 * writes within a string instruction are weakly ordered.  So there
1831 	 * are three writes overall.
1832 	 *
1833 	 * As a small optimization, only write the version field in the first
1834 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1835 	 * version field is the first in the struct.
1836 	 */
1837 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1838 
1839 	if (guest_hv_clock.version & 1)
1840 		++guest_hv_clock.version;  /* first time write, random junk */
1841 
1842 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1843 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1844 				&vcpu->hv_clock,
1845 				sizeof(vcpu->hv_clock.version));
1846 
1847 	smp_wmb();
1848 
1849 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1850 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1851 
1852 	if (vcpu->pvclock_set_guest_stopped_request) {
1853 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1854 		vcpu->pvclock_set_guest_stopped_request = false;
1855 	}
1856 
1857 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1858 
1859 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1860 				&vcpu->hv_clock,
1861 				sizeof(vcpu->hv_clock));
1862 
1863 	smp_wmb();
1864 
1865 	vcpu->hv_clock.version++;
1866 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1867 				&vcpu->hv_clock,
1868 				sizeof(vcpu->hv_clock.version));
1869 }
1870 
1871 static int kvm_guest_time_update(struct kvm_vcpu *v)
1872 {
1873 	unsigned long flags, tgt_tsc_khz;
1874 	struct kvm_vcpu_arch *vcpu = &v->arch;
1875 	struct kvm_arch *ka = &v->kvm->arch;
1876 	s64 kernel_ns;
1877 	u64 tsc_timestamp, host_tsc;
1878 	u8 pvclock_flags;
1879 	bool use_master_clock;
1880 
1881 	kernel_ns = 0;
1882 	host_tsc = 0;
1883 
1884 	/*
1885 	 * If the host uses TSC clock, then passthrough TSC as stable
1886 	 * to the guest.
1887 	 */
1888 	spin_lock(&ka->pvclock_gtod_sync_lock);
1889 	use_master_clock = ka->use_master_clock;
1890 	if (use_master_clock) {
1891 		host_tsc = ka->master_cycle_now;
1892 		kernel_ns = ka->master_kernel_ns;
1893 	}
1894 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1895 
1896 	/* Keep irq disabled to prevent changes to the clock */
1897 	local_irq_save(flags);
1898 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1899 	if (unlikely(tgt_tsc_khz == 0)) {
1900 		local_irq_restore(flags);
1901 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1902 		return 1;
1903 	}
1904 	if (!use_master_clock) {
1905 		host_tsc = rdtsc();
1906 		kernel_ns = ktime_get_boot_ns();
1907 	}
1908 
1909 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1910 
1911 	/*
1912 	 * We may have to catch up the TSC to match elapsed wall clock
1913 	 * time for two reasons, even if kvmclock is used.
1914 	 *   1) CPU could have been running below the maximum TSC rate
1915 	 *   2) Broken TSC compensation resets the base at each VCPU
1916 	 *      entry to avoid unknown leaps of TSC even when running
1917 	 *      again on the same CPU.  This may cause apparent elapsed
1918 	 *      time to disappear, and the guest to stand still or run
1919 	 *	very slowly.
1920 	 */
1921 	if (vcpu->tsc_catchup) {
1922 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1923 		if (tsc > tsc_timestamp) {
1924 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1925 			tsc_timestamp = tsc;
1926 		}
1927 	}
1928 
1929 	local_irq_restore(flags);
1930 
1931 	/* With all the info we got, fill in the values */
1932 
1933 	if (kvm_has_tsc_control)
1934 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1935 
1936 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1937 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1938 				   &vcpu->hv_clock.tsc_shift,
1939 				   &vcpu->hv_clock.tsc_to_system_mul);
1940 		vcpu->hw_tsc_khz = tgt_tsc_khz;
1941 	}
1942 
1943 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1944 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1945 	vcpu->last_guest_tsc = tsc_timestamp;
1946 
1947 	/* If the host uses TSC clocksource, then it is stable */
1948 	pvclock_flags = 0;
1949 	if (use_master_clock)
1950 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1951 
1952 	vcpu->hv_clock.flags = pvclock_flags;
1953 
1954 	if (vcpu->pv_time_enabled)
1955 		kvm_setup_pvclock_page(v);
1956 	if (v == kvm_get_vcpu(v->kvm, 0))
1957 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1958 	return 0;
1959 }
1960 
1961 /*
1962  * kvmclock updates which are isolated to a given vcpu, such as
1963  * vcpu->cpu migration, should not allow system_timestamp from
1964  * the rest of the vcpus to remain static. Otherwise ntp frequency
1965  * correction applies to one vcpu's system_timestamp but not
1966  * the others.
1967  *
1968  * So in those cases, request a kvmclock update for all vcpus.
1969  * We need to rate-limit these requests though, as they can
1970  * considerably slow guests that have a large number of vcpus.
1971  * The time for a remote vcpu to update its kvmclock is bound
1972  * by the delay we use to rate-limit the updates.
1973  */
1974 
1975 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1976 
1977 static void kvmclock_update_fn(struct work_struct *work)
1978 {
1979 	int i;
1980 	struct delayed_work *dwork = to_delayed_work(work);
1981 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1982 					   kvmclock_update_work);
1983 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1984 	struct kvm_vcpu *vcpu;
1985 
1986 	kvm_for_each_vcpu(i, vcpu, kvm) {
1987 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1988 		kvm_vcpu_kick(vcpu);
1989 	}
1990 }
1991 
1992 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1993 {
1994 	struct kvm *kvm = v->kvm;
1995 
1996 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1997 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1998 					KVMCLOCK_UPDATE_DELAY);
1999 }
2000 
2001 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2002 
2003 static void kvmclock_sync_fn(struct work_struct *work)
2004 {
2005 	struct delayed_work *dwork = to_delayed_work(work);
2006 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2007 					   kvmclock_sync_work);
2008 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2009 
2010 	if (!kvmclock_periodic_sync)
2011 		return;
2012 
2013 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2014 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2015 					KVMCLOCK_SYNC_PERIOD);
2016 }
2017 
2018 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2019 {
2020 	u64 mcg_cap = vcpu->arch.mcg_cap;
2021 	unsigned bank_num = mcg_cap & 0xff;
2022 	u32 msr = msr_info->index;
2023 	u64 data = msr_info->data;
2024 
2025 	switch (msr) {
2026 	case MSR_IA32_MCG_STATUS:
2027 		vcpu->arch.mcg_status = data;
2028 		break;
2029 	case MSR_IA32_MCG_CTL:
2030 		if (!(mcg_cap & MCG_CTL_P))
2031 			return 1;
2032 		if (data != 0 && data != ~(u64)0)
2033 			return -1;
2034 		vcpu->arch.mcg_ctl = data;
2035 		break;
2036 	default:
2037 		if (msr >= MSR_IA32_MC0_CTL &&
2038 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2039 			u32 offset = msr - MSR_IA32_MC0_CTL;
2040 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2041 			 * some Linux kernels though clear bit 10 in bank 4 to
2042 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2043 			 * this to avoid an uncatched #GP in the guest
2044 			 */
2045 			if ((offset & 0x3) == 0 &&
2046 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2047 				return -1;
2048 			if (!msr_info->host_initiated &&
2049 				(offset & 0x3) == 1 && data != 0)
2050 				return -1;
2051 			vcpu->arch.mce_banks[offset] = data;
2052 			break;
2053 		}
2054 		return 1;
2055 	}
2056 	return 0;
2057 }
2058 
2059 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2060 {
2061 	struct kvm *kvm = vcpu->kvm;
2062 	int lm = is_long_mode(vcpu);
2063 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2064 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2065 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2066 		: kvm->arch.xen_hvm_config.blob_size_32;
2067 	u32 page_num = data & ~PAGE_MASK;
2068 	u64 page_addr = data & PAGE_MASK;
2069 	u8 *page;
2070 	int r;
2071 
2072 	r = -E2BIG;
2073 	if (page_num >= blob_size)
2074 		goto out;
2075 	r = -ENOMEM;
2076 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2077 	if (IS_ERR(page)) {
2078 		r = PTR_ERR(page);
2079 		goto out;
2080 	}
2081 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2082 		goto out_free;
2083 	r = 0;
2084 out_free:
2085 	kfree(page);
2086 out:
2087 	return r;
2088 }
2089 
2090 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2091 {
2092 	gpa_t gpa = data & ~0x3f;
2093 
2094 	/* Bits 3:5 are reserved, Should be zero */
2095 	if (data & 0x38)
2096 		return 1;
2097 
2098 	vcpu->arch.apf.msr_val = data;
2099 
2100 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2101 		kvm_clear_async_pf_completion_queue(vcpu);
2102 		kvm_async_pf_hash_reset(vcpu);
2103 		return 0;
2104 	}
2105 
2106 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2107 					sizeof(u32)))
2108 		return 1;
2109 
2110 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2111 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2112 	kvm_async_pf_wakeup_all(vcpu);
2113 	return 0;
2114 }
2115 
2116 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2117 {
2118 	vcpu->arch.pv_time_enabled = false;
2119 }
2120 
2121 static void record_steal_time(struct kvm_vcpu *vcpu)
2122 {
2123 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2124 		return;
2125 
2126 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2127 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2128 		return;
2129 
2130 	vcpu->arch.st.steal.preempted = 0;
2131 
2132 	if (vcpu->arch.st.steal.version & 1)
2133 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2134 
2135 	vcpu->arch.st.steal.version += 1;
2136 
2137 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2138 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2139 
2140 	smp_wmb();
2141 
2142 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2143 		vcpu->arch.st.last_steal;
2144 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2145 
2146 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2147 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2148 
2149 	smp_wmb();
2150 
2151 	vcpu->arch.st.steal.version += 1;
2152 
2153 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2154 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2155 }
2156 
2157 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2158 {
2159 	bool pr = false;
2160 	u32 msr = msr_info->index;
2161 	u64 data = msr_info->data;
2162 
2163 	switch (msr) {
2164 	case MSR_AMD64_NB_CFG:
2165 	case MSR_IA32_UCODE_REV:
2166 	case MSR_IA32_UCODE_WRITE:
2167 	case MSR_VM_HSAVE_PA:
2168 	case MSR_AMD64_PATCH_LOADER:
2169 	case MSR_AMD64_BU_CFG2:
2170 	case MSR_AMD64_DC_CFG:
2171 		break;
2172 
2173 	case MSR_EFER:
2174 		return set_efer(vcpu, data);
2175 	case MSR_K7_HWCR:
2176 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2177 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2178 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2179 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2180 		if (data != 0) {
2181 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2182 				    data);
2183 			return 1;
2184 		}
2185 		break;
2186 	case MSR_FAM10H_MMIO_CONF_BASE:
2187 		if (data != 0) {
2188 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2189 				    "0x%llx\n", data);
2190 			return 1;
2191 		}
2192 		break;
2193 	case MSR_IA32_DEBUGCTLMSR:
2194 		if (!data) {
2195 			/* We support the non-activated case already */
2196 			break;
2197 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2198 			/* Values other than LBR and BTF are vendor-specific,
2199 			   thus reserved and should throw a #GP */
2200 			return 1;
2201 		}
2202 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2203 			    __func__, data);
2204 		break;
2205 	case 0x200 ... 0x2ff:
2206 		return kvm_mtrr_set_msr(vcpu, msr, data);
2207 	case MSR_IA32_APICBASE:
2208 		return kvm_set_apic_base(vcpu, msr_info);
2209 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2210 		return kvm_x2apic_msr_write(vcpu, msr, data);
2211 	case MSR_IA32_TSCDEADLINE:
2212 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2213 		break;
2214 	case MSR_IA32_TSC_ADJUST:
2215 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2216 			if (!msr_info->host_initiated) {
2217 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2218 				adjust_tsc_offset_guest(vcpu, adj);
2219 			}
2220 			vcpu->arch.ia32_tsc_adjust_msr = data;
2221 		}
2222 		break;
2223 	case MSR_IA32_MISC_ENABLE:
2224 		vcpu->arch.ia32_misc_enable_msr = data;
2225 		break;
2226 	case MSR_IA32_SMBASE:
2227 		if (!msr_info->host_initiated)
2228 			return 1;
2229 		vcpu->arch.smbase = data;
2230 		break;
2231 	case MSR_KVM_WALL_CLOCK_NEW:
2232 	case MSR_KVM_WALL_CLOCK:
2233 		vcpu->kvm->arch.wall_clock = data;
2234 		kvm_write_wall_clock(vcpu->kvm, data);
2235 		break;
2236 	case MSR_KVM_SYSTEM_TIME_NEW:
2237 	case MSR_KVM_SYSTEM_TIME: {
2238 		struct kvm_arch *ka = &vcpu->kvm->arch;
2239 
2240 		kvmclock_reset(vcpu);
2241 
2242 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2243 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2244 
2245 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2246 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2247 
2248 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2249 		}
2250 
2251 		vcpu->arch.time = data;
2252 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2253 
2254 		/* we verify if the enable bit is set... */
2255 		if (!(data & 1))
2256 			break;
2257 
2258 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2259 		     &vcpu->arch.pv_time, data & ~1ULL,
2260 		     sizeof(struct pvclock_vcpu_time_info)))
2261 			vcpu->arch.pv_time_enabled = false;
2262 		else
2263 			vcpu->arch.pv_time_enabled = true;
2264 
2265 		break;
2266 	}
2267 	case MSR_KVM_ASYNC_PF_EN:
2268 		if (kvm_pv_enable_async_pf(vcpu, data))
2269 			return 1;
2270 		break;
2271 	case MSR_KVM_STEAL_TIME:
2272 
2273 		if (unlikely(!sched_info_on()))
2274 			return 1;
2275 
2276 		if (data & KVM_STEAL_RESERVED_MASK)
2277 			return 1;
2278 
2279 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2280 						data & KVM_STEAL_VALID_BITS,
2281 						sizeof(struct kvm_steal_time)))
2282 			return 1;
2283 
2284 		vcpu->arch.st.msr_val = data;
2285 
2286 		if (!(data & KVM_MSR_ENABLED))
2287 			break;
2288 
2289 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2290 
2291 		break;
2292 	case MSR_KVM_PV_EOI_EN:
2293 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2294 			return 1;
2295 		break;
2296 
2297 	case MSR_IA32_MCG_CTL:
2298 	case MSR_IA32_MCG_STATUS:
2299 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2300 		return set_msr_mce(vcpu, msr_info);
2301 
2302 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2303 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2304 		pr = true; /* fall through */
2305 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2306 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2307 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2308 			return kvm_pmu_set_msr(vcpu, msr_info);
2309 
2310 		if (pr || data != 0)
2311 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2312 				    "0x%x data 0x%llx\n", msr, data);
2313 		break;
2314 	case MSR_K7_CLK_CTL:
2315 		/*
2316 		 * Ignore all writes to this no longer documented MSR.
2317 		 * Writes are only relevant for old K7 processors,
2318 		 * all pre-dating SVM, but a recommended workaround from
2319 		 * AMD for these chips. It is possible to specify the
2320 		 * affected processor models on the command line, hence
2321 		 * the need to ignore the workaround.
2322 		 */
2323 		break;
2324 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2325 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2326 	case HV_X64_MSR_CRASH_CTL:
2327 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2328 		return kvm_hv_set_msr_common(vcpu, msr, data,
2329 					     msr_info->host_initiated);
2330 	case MSR_IA32_BBL_CR_CTL3:
2331 		/* Drop writes to this legacy MSR -- see rdmsr
2332 		 * counterpart for further detail.
2333 		 */
2334 		if (report_ignored_msrs)
2335 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2336 				msr, data);
2337 		break;
2338 	case MSR_AMD64_OSVW_ID_LENGTH:
2339 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2340 			return 1;
2341 		vcpu->arch.osvw.length = data;
2342 		break;
2343 	case MSR_AMD64_OSVW_STATUS:
2344 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2345 			return 1;
2346 		vcpu->arch.osvw.status = data;
2347 		break;
2348 	case MSR_PLATFORM_INFO:
2349 		if (!msr_info->host_initiated ||
2350 		    data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2351 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2352 		     cpuid_fault_enabled(vcpu)))
2353 			return 1;
2354 		vcpu->arch.msr_platform_info = data;
2355 		break;
2356 	case MSR_MISC_FEATURES_ENABLES:
2357 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2358 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2359 		     !supports_cpuid_fault(vcpu)))
2360 			return 1;
2361 		vcpu->arch.msr_misc_features_enables = data;
2362 		break;
2363 	default:
2364 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2365 			return xen_hvm_config(vcpu, data);
2366 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2367 			return kvm_pmu_set_msr(vcpu, msr_info);
2368 		if (!ignore_msrs) {
2369 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2370 				    msr, data);
2371 			return 1;
2372 		} else {
2373 			if (report_ignored_msrs)
2374 				vcpu_unimpl(vcpu,
2375 					"ignored wrmsr: 0x%x data 0x%llx\n",
2376 					msr, data);
2377 			break;
2378 		}
2379 	}
2380 	return 0;
2381 }
2382 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2383 
2384 
2385 /*
2386  * Reads an msr value (of 'msr_index') into 'pdata'.
2387  * Returns 0 on success, non-0 otherwise.
2388  * Assumes vcpu_load() was already called.
2389  */
2390 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2391 {
2392 	return kvm_x86_ops->get_msr(vcpu, msr);
2393 }
2394 EXPORT_SYMBOL_GPL(kvm_get_msr);
2395 
2396 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2397 {
2398 	u64 data;
2399 	u64 mcg_cap = vcpu->arch.mcg_cap;
2400 	unsigned bank_num = mcg_cap & 0xff;
2401 
2402 	switch (msr) {
2403 	case MSR_IA32_P5_MC_ADDR:
2404 	case MSR_IA32_P5_MC_TYPE:
2405 		data = 0;
2406 		break;
2407 	case MSR_IA32_MCG_CAP:
2408 		data = vcpu->arch.mcg_cap;
2409 		break;
2410 	case MSR_IA32_MCG_CTL:
2411 		if (!(mcg_cap & MCG_CTL_P))
2412 			return 1;
2413 		data = vcpu->arch.mcg_ctl;
2414 		break;
2415 	case MSR_IA32_MCG_STATUS:
2416 		data = vcpu->arch.mcg_status;
2417 		break;
2418 	default:
2419 		if (msr >= MSR_IA32_MC0_CTL &&
2420 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2421 			u32 offset = msr - MSR_IA32_MC0_CTL;
2422 			data = vcpu->arch.mce_banks[offset];
2423 			break;
2424 		}
2425 		return 1;
2426 	}
2427 	*pdata = data;
2428 	return 0;
2429 }
2430 
2431 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2432 {
2433 	switch (msr_info->index) {
2434 	case MSR_IA32_PLATFORM_ID:
2435 	case MSR_IA32_EBL_CR_POWERON:
2436 	case MSR_IA32_DEBUGCTLMSR:
2437 	case MSR_IA32_LASTBRANCHFROMIP:
2438 	case MSR_IA32_LASTBRANCHTOIP:
2439 	case MSR_IA32_LASTINTFROMIP:
2440 	case MSR_IA32_LASTINTTOIP:
2441 	case MSR_K8_SYSCFG:
2442 	case MSR_K8_TSEG_ADDR:
2443 	case MSR_K8_TSEG_MASK:
2444 	case MSR_K7_HWCR:
2445 	case MSR_VM_HSAVE_PA:
2446 	case MSR_K8_INT_PENDING_MSG:
2447 	case MSR_AMD64_NB_CFG:
2448 	case MSR_FAM10H_MMIO_CONF_BASE:
2449 	case MSR_AMD64_BU_CFG2:
2450 	case MSR_IA32_PERF_CTL:
2451 	case MSR_AMD64_DC_CFG:
2452 		msr_info->data = 0;
2453 		break;
2454 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2455 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2456 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2457 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2458 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2459 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2460 		msr_info->data = 0;
2461 		break;
2462 	case MSR_IA32_UCODE_REV:
2463 		msr_info->data = 0x100000000ULL;
2464 		break;
2465 	case MSR_MTRRcap:
2466 	case 0x200 ... 0x2ff:
2467 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2468 	case 0xcd: /* fsb frequency */
2469 		msr_info->data = 3;
2470 		break;
2471 		/*
2472 		 * MSR_EBC_FREQUENCY_ID
2473 		 * Conservative value valid for even the basic CPU models.
2474 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2475 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2476 		 * and 266MHz for model 3, or 4. Set Core Clock
2477 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2478 		 * 31:24) even though these are only valid for CPU
2479 		 * models > 2, however guests may end up dividing or
2480 		 * multiplying by zero otherwise.
2481 		 */
2482 	case MSR_EBC_FREQUENCY_ID:
2483 		msr_info->data = 1 << 24;
2484 		break;
2485 	case MSR_IA32_APICBASE:
2486 		msr_info->data = kvm_get_apic_base(vcpu);
2487 		break;
2488 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2489 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2490 		break;
2491 	case MSR_IA32_TSCDEADLINE:
2492 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2493 		break;
2494 	case MSR_IA32_TSC_ADJUST:
2495 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2496 		break;
2497 	case MSR_IA32_MISC_ENABLE:
2498 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2499 		break;
2500 	case MSR_IA32_SMBASE:
2501 		if (!msr_info->host_initiated)
2502 			return 1;
2503 		msr_info->data = vcpu->arch.smbase;
2504 		break;
2505 	case MSR_IA32_PERF_STATUS:
2506 		/* TSC increment by tick */
2507 		msr_info->data = 1000ULL;
2508 		/* CPU multiplier */
2509 		msr_info->data |= (((uint64_t)4ULL) << 40);
2510 		break;
2511 	case MSR_EFER:
2512 		msr_info->data = vcpu->arch.efer;
2513 		break;
2514 	case MSR_KVM_WALL_CLOCK:
2515 	case MSR_KVM_WALL_CLOCK_NEW:
2516 		msr_info->data = vcpu->kvm->arch.wall_clock;
2517 		break;
2518 	case MSR_KVM_SYSTEM_TIME:
2519 	case MSR_KVM_SYSTEM_TIME_NEW:
2520 		msr_info->data = vcpu->arch.time;
2521 		break;
2522 	case MSR_KVM_ASYNC_PF_EN:
2523 		msr_info->data = vcpu->arch.apf.msr_val;
2524 		break;
2525 	case MSR_KVM_STEAL_TIME:
2526 		msr_info->data = vcpu->arch.st.msr_val;
2527 		break;
2528 	case MSR_KVM_PV_EOI_EN:
2529 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2530 		break;
2531 	case MSR_IA32_P5_MC_ADDR:
2532 	case MSR_IA32_P5_MC_TYPE:
2533 	case MSR_IA32_MCG_CAP:
2534 	case MSR_IA32_MCG_CTL:
2535 	case MSR_IA32_MCG_STATUS:
2536 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2537 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2538 	case MSR_K7_CLK_CTL:
2539 		/*
2540 		 * Provide expected ramp-up count for K7. All other
2541 		 * are set to zero, indicating minimum divisors for
2542 		 * every field.
2543 		 *
2544 		 * This prevents guest kernels on AMD host with CPU
2545 		 * type 6, model 8 and higher from exploding due to
2546 		 * the rdmsr failing.
2547 		 */
2548 		msr_info->data = 0x20000000;
2549 		break;
2550 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2551 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2552 	case HV_X64_MSR_CRASH_CTL:
2553 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2554 		return kvm_hv_get_msr_common(vcpu,
2555 					     msr_info->index, &msr_info->data);
2556 		break;
2557 	case MSR_IA32_BBL_CR_CTL3:
2558 		/* This legacy MSR exists but isn't fully documented in current
2559 		 * silicon.  It is however accessed by winxp in very narrow
2560 		 * scenarios where it sets bit #19, itself documented as
2561 		 * a "reserved" bit.  Best effort attempt to source coherent
2562 		 * read data here should the balance of the register be
2563 		 * interpreted by the guest:
2564 		 *
2565 		 * L2 cache control register 3: 64GB range, 256KB size,
2566 		 * enabled, latency 0x1, configured
2567 		 */
2568 		msr_info->data = 0xbe702111;
2569 		break;
2570 	case MSR_AMD64_OSVW_ID_LENGTH:
2571 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2572 			return 1;
2573 		msr_info->data = vcpu->arch.osvw.length;
2574 		break;
2575 	case MSR_AMD64_OSVW_STATUS:
2576 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2577 			return 1;
2578 		msr_info->data = vcpu->arch.osvw.status;
2579 		break;
2580 	case MSR_PLATFORM_INFO:
2581 		msr_info->data = vcpu->arch.msr_platform_info;
2582 		break;
2583 	case MSR_MISC_FEATURES_ENABLES:
2584 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2585 		break;
2586 	default:
2587 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2588 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2589 		if (!ignore_msrs) {
2590 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2591 					       msr_info->index);
2592 			return 1;
2593 		} else {
2594 			if (report_ignored_msrs)
2595 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2596 					msr_info->index);
2597 			msr_info->data = 0;
2598 		}
2599 		break;
2600 	}
2601 	return 0;
2602 }
2603 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2604 
2605 /*
2606  * Read or write a bunch of msrs. All parameters are kernel addresses.
2607  *
2608  * @return number of msrs set successfully.
2609  */
2610 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2611 		    struct kvm_msr_entry *entries,
2612 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2613 				  unsigned index, u64 *data))
2614 {
2615 	int i, idx;
2616 
2617 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2618 	for (i = 0; i < msrs->nmsrs; ++i)
2619 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2620 			break;
2621 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2622 
2623 	return i;
2624 }
2625 
2626 /*
2627  * Read or write a bunch of msrs. Parameters are user addresses.
2628  *
2629  * @return number of msrs set successfully.
2630  */
2631 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2632 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2633 				unsigned index, u64 *data),
2634 		  int writeback)
2635 {
2636 	struct kvm_msrs msrs;
2637 	struct kvm_msr_entry *entries;
2638 	int r, n;
2639 	unsigned size;
2640 
2641 	r = -EFAULT;
2642 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2643 		goto out;
2644 
2645 	r = -E2BIG;
2646 	if (msrs.nmsrs >= MAX_IO_MSRS)
2647 		goto out;
2648 
2649 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2650 	entries = memdup_user(user_msrs->entries, size);
2651 	if (IS_ERR(entries)) {
2652 		r = PTR_ERR(entries);
2653 		goto out;
2654 	}
2655 
2656 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2657 	if (r < 0)
2658 		goto out_free;
2659 
2660 	r = -EFAULT;
2661 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2662 		goto out_free;
2663 
2664 	r = n;
2665 
2666 out_free:
2667 	kfree(entries);
2668 out:
2669 	return r;
2670 }
2671 
2672 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2673 {
2674 	int r;
2675 
2676 	switch (ext) {
2677 	case KVM_CAP_IRQCHIP:
2678 	case KVM_CAP_HLT:
2679 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2680 	case KVM_CAP_SET_TSS_ADDR:
2681 	case KVM_CAP_EXT_CPUID:
2682 	case KVM_CAP_EXT_EMUL_CPUID:
2683 	case KVM_CAP_CLOCKSOURCE:
2684 	case KVM_CAP_PIT:
2685 	case KVM_CAP_NOP_IO_DELAY:
2686 	case KVM_CAP_MP_STATE:
2687 	case KVM_CAP_SYNC_MMU:
2688 	case KVM_CAP_USER_NMI:
2689 	case KVM_CAP_REINJECT_CONTROL:
2690 	case KVM_CAP_IRQ_INJECT_STATUS:
2691 	case KVM_CAP_IOEVENTFD:
2692 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2693 	case KVM_CAP_PIT2:
2694 	case KVM_CAP_PIT_STATE2:
2695 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2696 	case KVM_CAP_XEN_HVM:
2697 	case KVM_CAP_VCPU_EVENTS:
2698 	case KVM_CAP_HYPERV:
2699 	case KVM_CAP_HYPERV_VAPIC:
2700 	case KVM_CAP_HYPERV_SPIN:
2701 	case KVM_CAP_HYPERV_SYNIC:
2702 	case KVM_CAP_HYPERV_SYNIC2:
2703 	case KVM_CAP_HYPERV_VP_INDEX:
2704 	case KVM_CAP_PCI_SEGMENT:
2705 	case KVM_CAP_DEBUGREGS:
2706 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2707 	case KVM_CAP_XSAVE:
2708 	case KVM_CAP_ASYNC_PF:
2709 	case KVM_CAP_GET_TSC_KHZ:
2710 	case KVM_CAP_KVMCLOCK_CTRL:
2711 	case KVM_CAP_READONLY_MEM:
2712 	case KVM_CAP_HYPERV_TIME:
2713 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2714 	case KVM_CAP_TSC_DEADLINE_TIMER:
2715 	case KVM_CAP_ENABLE_CAP_VM:
2716 	case KVM_CAP_DISABLE_QUIRKS:
2717 	case KVM_CAP_SET_BOOT_CPU_ID:
2718  	case KVM_CAP_SPLIT_IRQCHIP:
2719 	case KVM_CAP_IMMEDIATE_EXIT:
2720 		r = 1;
2721 		break;
2722 	case KVM_CAP_ADJUST_CLOCK:
2723 		r = KVM_CLOCK_TSC_STABLE;
2724 		break;
2725 	case KVM_CAP_X86_GUEST_MWAIT:
2726 		r = kvm_mwait_in_guest();
2727 		break;
2728 	case KVM_CAP_X86_SMM:
2729 		/* SMBASE is usually relocated above 1M on modern chipsets,
2730 		 * and SMM handlers might indeed rely on 4G segment limits,
2731 		 * so do not report SMM to be available if real mode is
2732 		 * emulated via vm86 mode.  Still, do not go to great lengths
2733 		 * to avoid userspace's usage of the feature, because it is a
2734 		 * fringe case that is not enabled except via specific settings
2735 		 * of the module parameters.
2736 		 */
2737 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2738 		break;
2739 	case KVM_CAP_VAPIC:
2740 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2741 		break;
2742 	case KVM_CAP_NR_VCPUS:
2743 		r = KVM_SOFT_MAX_VCPUS;
2744 		break;
2745 	case KVM_CAP_MAX_VCPUS:
2746 		r = KVM_MAX_VCPUS;
2747 		break;
2748 	case KVM_CAP_NR_MEMSLOTS:
2749 		r = KVM_USER_MEM_SLOTS;
2750 		break;
2751 	case KVM_CAP_PV_MMU:	/* obsolete */
2752 		r = 0;
2753 		break;
2754 	case KVM_CAP_MCE:
2755 		r = KVM_MAX_MCE_BANKS;
2756 		break;
2757 	case KVM_CAP_XCRS:
2758 		r = boot_cpu_has(X86_FEATURE_XSAVE);
2759 		break;
2760 	case KVM_CAP_TSC_CONTROL:
2761 		r = kvm_has_tsc_control;
2762 		break;
2763 	case KVM_CAP_X2APIC_API:
2764 		r = KVM_X2APIC_API_VALID_FLAGS;
2765 		break;
2766 	default:
2767 		r = 0;
2768 		break;
2769 	}
2770 	return r;
2771 
2772 }
2773 
2774 long kvm_arch_dev_ioctl(struct file *filp,
2775 			unsigned int ioctl, unsigned long arg)
2776 {
2777 	void __user *argp = (void __user *)arg;
2778 	long r;
2779 
2780 	switch (ioctl) {
2781 	case KVM_GET_MSR_INDEX_LIST: {
2782 		struct kvm_msr_list __user *user_msr_list = argp;
2783 		struct kvm_msr_list msr_list;
2784 		unsigned n;
2785 
2786 		r = -EFAULT;
2787 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2788 			goto out;
2789 		n = msr_list.nmsrs;
2790 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2791 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2792 			goto out;
2793 		r = -E2BIG;
2794 		if (n < msr_list.nmsrs)
2795 			goto out;
2796 		r = -EFAULT;
2797 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2798 				 num_msrs_to_save * sizeof(u32)))
2799 			goto out;
2800 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2801 				 &emulated_msrs,
2802 				 num_emulated_msrs * sizeof(u32)))
2803 			goto out;
2804 		r = 0;
2805 		break;
2806 	}
2807 	case KVM_GET_SUPPORTED_CPUID:
2808 	case KVM_GET_EMULATED_CPUID: {
2809 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2810 		struct kvm_cpuid2 cpuid;
2811 
2812 		r = -EFAULT;
2813 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2814 			goto out;
2815 
2816 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2817 					    ioctl);
2818 		if (r)
2819 			goto out;
2820 
2821 		r = -EFAULT;
2822 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2823 			goto out;
2824 		r = 0;
2825 		break;
2826 	}
2827 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2828 		r = -EFAULT;
2829 		if (copy_to_user(argp, &kvm_mce_cap_supported,
2830 				 sizeof(kvm_mce_cap_supported)))
2831 			goto out;
2832 		r = 0;
2833 		break;
2834 	}
2835 	default:
2836 		r = -EINVAL;
2837 	}
2838 out:
2839 	return r;
2840 }
2841 
2842 static void wbinvd_ipi(void *garbage)
2843 {
2844 	wbinvd();
2845 }
2846 
2847 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2848 {
2849 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2850 }
2851 
2852 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2853 {
2854 	/* Address WBINVD may be executed by guest */
2855 	if (need_emulate_wbinvd(vcpu)) {
2856 		if (kvm_x86_ops->has_wbinvd_exit())
2857 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2858 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2859 			smp_call_function_single(vcpu->cpu,
2860 					wbinvd_ipi, NULL, 1);
2861 	}
2862 
2863 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2864 
2865 	/* Apply any externally detected TSC adjustments (due to suspend) */
2866 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2867 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2868 		vcpu->arch.tsc_offset_adjustment = 0;
2869 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2870 	}
2871 
2872 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2873 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2874 				rdtsc() - vcpu->arch.last_host_tsc;
2875 		if (tsc_delta < 0)
2876 			mark_tsc_unstable("KVM discovered backwards TSC");
2877 
2878 		if (check_tsc_unstable()) {
2879 			u64 offset = kvm_compute_tsc_offset(vcpu,
2880 						vcpu->arch.last_guest_tsc);
2881 			kvm_vcpu_write_tsc_offset(vcpu, offset);
2882 			vcpu->arch.tsc_catchup = 1;
2883 		}
2884 
2885 		if (kvm_lapic_hv_timer_in_use(vcpu))
2886 			kvm_lapic_restart_hv_timer(vcpu);
2887 
2888 		/*
2889 		 * On a host with synchronized TSC, there is no need to update
2890 		 * kvmclock on vcpu->cpu migration
2891 		 */
2892 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2893 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2894 		if (vcpu->cpu != cpu)
2895 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2896 		vcpu->cpu = cpu;
2897 	}
2898 
2899 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2900 }
2901 
2902 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2903 {
2904 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2905 		return;
2906 
2907 	vcpu->arch.st.steal.preempted = 1;
2908 
2909 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2910 			&vcpu->arch.st.steal.preempted,
2911 			offsetof(struct kvm_steal_time, preempted),
2912 			sizeof(vcpu->arch.st.steal.preempted));
2913 }
2914 
2915 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2916 {
2917 	int idx;
2918 
2919 	if (vcpu->preempted)
2920 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
2921 
2922 	/*
2923 	 * Disable page faults because we're in atomic context here.
2924 	 * kvm_write_guest_offset_cached() would call might_fault()
2925 	 * that relies on pagefault_disable() to tell if there's a
2926 	 * bug. NOTE: the write to guest memory may not go through if
2927 	 * during postcopy live migration or if there's heavy guest
2928 	 * paging.
2929 	 */
2930 	pagefault_disable();
2931 	/*
2932 	 * kvm_memslots() will be called by
2933 	 * kvm_write_guest_offset_cached() so take the srcu lock.
2934 	 */
2935 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2936 	kvm_steal_time_set_preempted(vcpu);
2937 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2938 	pagefault_enable();
2939 	kvm_x86_ops->vcpu_put(vcpu);
2940 	kvm_put_guest_fpu(vcpu);
2941 	vcpu->arch.last_host_tsc = rdtsc();
2942 }
2943 
2944 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2945 				    struct kvm_lapic_state *s)
2946 {
2947 	if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2948 		kvm_x86_ops->sync_pir_to_irr(vcpu);
2949 
2950 	return kvm_apic_get_state(vcpu, s);
2951 }
2952 
2953 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2954 				    struct kvm_lapic_state *s)
2955 {
2956 	int r;
2957 
2958 	r = kvm_apic_set_state(vcpu, s);
2959 	if (r)
2960 		return r;
2961 	update_cr8_intercept(vcpu);
2962 
2963 	return 0;
2964 }
2965 
2966 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2967 {
2968 	return (!lapic_in_kernel(vcpu) ||
2969 		kvm_apic_accept_pic_intr(vcpu));
2970 }
2971 
2972 /*
2973  * if userspace requested an interrupt window, check that the
2974  * interrupt window is open.
2975  *
2976  * No need to exit to userspace if we already have an interrupt queued.
2977  */
2978 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2979 {
2980 	return kvm_arch_interrupt_allowed(vcpu) &&
2981 		!kvm_cpu_has_interrupt(vcpu) &&
2982 		!kvm_event_needs_reinjection(vcpu) &&
2983 		kvm_cpu_accept_dm_intr(vcpu);
2984 }
2985 
2986 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2987 				    struct kvm_interrupt *irq)
2988 {
2989 	if (irq->irq >= KVM_NR_INTERRUPTS)
2990 		return -EINVAL;
2991 
2992 	if (!irqchip_in_kernel(vcpu->kvm)) {
2993 		kvm_queue_interrupt(vcpu, irq->irq, false);
2994 		kvm_make_request(KVM_REQ_EVENT, vcpu);
2995 		return 0;
2996 	}
2997 
2998 	/*
2999 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3000 	 * fail for in-kernel 8259.
3001 	 */
3002 	if (pic_in_kernel(vcpu->kvm))
3003 		return -ENXIO;
3004 
3005 	if (vcpu->arch.pending_external_vector != -1)
3006 		return -EEXIST;
3007 
3008 	vcpu->arch.pending_external_vector = irq->irq;
3009 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3010 	return 0;
3011 }
3012 
3013 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3014 {
3015 	kvm_inject_nmi(vcpu);
3016 
3017 	return 0;
3018 }
3019 
3020 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3021 {
3022 	kvm_make_request(KVM_REQ_SMI, vcpu);
3023 
3024 	return 0;
3025 }
3026 
3027 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3028 					   struct kvm_tpr_access_ctl *tac)
3029 {
3030 	if (tac->flags)
3031 		return -EINVAL;
3032 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3033 	return 0;
3034 }
3035 
3036 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3037 					u64 mcg_cap)
3038 {
3039 	int r;
3040 	unsigned bank_num = mcg_cap & 0xff, bank;
3041 
3042 	r = -EINVAL;
3043 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3044 		goto out;
3045 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3046 		goto out;
3047 	r = 0;
3048 	vcpu->arch.mcg_cap = mcg_cap;
3049 	/* Init IA32_MCG_CTL to all 1s */
3050 	if (mcg_cap & MCG_CTL_P)
3051 		vcpu->arch.mcg_ctl = ~(u64)0;
3052 	/* Init IA32_MCi_CTL to all 1s */
3053 	for (bank = 0; bank < bank_num; bank++)
3054 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3055 
3056 	if (kvm_x86_ops->setup_mce)
3057 		kvm_x86_ops->setup_mce(vcpu);
3058 out:
3059 	return r;
3060 }
3061 
3062 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3063 				      struct kvm_x86_mce *mce)
3064 {
3065 	u64 mcg_cap = vcpu->arch.mcg_cap;
3066 	unsigned bank_num = mcg_cap & 0xff;
3067 	u64 *banks = vcpu->arch.mce_banks;
3068 
3069 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3070 		return -EINVAL;
3071 	/*
3072 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3073 	 * reporting is disabled
3074 	 */
3075 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3076 	    vcpu->arch.mcg_ctl != ~(u64)0)
3077 		return 0;
3078 	banks += 4 * mce->bank;
3079 	/*
3080 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3081 	 * reporting is disabled for the bank
3082 	 */
3083 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3084 		return 0;
3085 	if (mce->status & MCI_STATUS_UC) {
3086 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3087 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3088 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3089 			return 0;
3090 		}
3091 		if (banks[1] & MCI_STATUS_VAL)
3092 			mce->status |= MCI_STATUS_OVER;
3093 		banks[2] = mce->addr;
3094 		banks[3] = mce->misc;
3095 		vcpu->arch.mcg_status = mce->mcg_status;
3096 		banks[1] = mce->status;
3097 		kvm_queue_exception(vcpu, MC_VECTOR);
3098 	} else if (!(banks[1] & MCI_STATUS_VAL)
3099 		   || !(banks[1] & MCI_STATUS_UC)) {
3100 		if (banks[1] & MCI_STATUS_VAL)
3101 			mce->status |= MCI_STATUS_OVER;
3102 		banks[2] = mce->addr;
3103 		banks[3] = mce->misc;
3104 		banks[1] = mce->status;
3105 	} else
3106 		banks[1] |= MCI_STATUS_OVER;
3107 	return 0;
3108 }
3109 
3110 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3111 					       struct kvm_vcpu_events *events)
3112 {
3113 	process_nmi(vcpu);
3114 	/*
3115 	 * FIXME: pass injected and pending separately.  This is only
3116 	 * needed for nested virtualization, whose state cannot be
3117 	 * migrated yet.  For now we can combine them.
3118 	 */
3119 	events->exception.injected =
3120 		(vcpu->arch.exception.pending ||
3121 		 vcpu->arch.exception.injected) &&
3122 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3123 	events->exception.nr = vcpu->arch.exception.nr;
3124 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3125 	events->exception.pad = 0;
3126 	events->exception.error_code = vcpu->arch.exception.error_code;
3127 
3128 	events->interrupt.injected =
3129 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3130 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3131 	events->interrupt.soft = 0;
3132 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3133 
3134 	events->nmi.injected = vcpu->arch.nmi_injected;
3135 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3136 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3137 	events->nmi.pad = 0;
3138 
3139 	events->sipi_vector = 0; /* never valid when reporting to user space */
3140 
3141 	events->smi.smm = is_smm(vcpu);
3142 	events->smi.pending = vcpu->arch.smi_pending;
3143 	events->smi.smm_inside_nmi =
3144 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3145 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3146 
3147 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3148 			 | KVM_VCPUEVENT_VALID_SHADOW
3149 			 | KVM_VCPUEVENT_VALID_SMM);
3150 	memset(&events->reserved, 0, sizeof(events->reserved));
3151 }
3152 
3153 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3154 
3155 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3156 					      struct kvm_vcpu_events *events)
3157 {
3158 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3159 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3160 			      | KVM_VCPUEVENT_VALID_SHADOW
3161 			      | KVM_VCPUEVENT_VALID_SMM))
3162 		return -EINVAL;
3163 
3164 	if (events->exception.injected &&
3165 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3166 	     is_guest_mode(vcpu)))
3167 		return -EINVAL;
3168 
3169 	/* INITs are latched while in SMM */
3170 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3171 	    (events->smi.smm || events->smi.pending) &&
3172 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3173 		return -EINVAL;
3174 
3175 	process_nmi(vcpu);
3176 	vcpu->arch.exception.injected = false;
3177 	vcpu->arch.exception.pending = events->exception.injected;
3178 	vcpu->arch.exception.nr = events->exception.nr;
3179 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3180 	vcpu->arch.exception.error_code = events->exception.error_code;
3181 
3182 	vcpu->arch.interrupt.pending = events->interrupt.injected;
3183 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3184 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3185 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3186 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3187 						  events->interrupt.shadow);
3188 
3189 	vcpu->arch.nmi_injected = events->nmi.injected;
3190 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3191 		vcpu->arch.nmi_pending = events->nmi.pending;
3192 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3193 
3194 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3195 	    lapic_in_kernel(vcpu))
3196 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3197 
3198 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3199 		u32 hflags = vcpu->arch.hflags;
3200 		if (events->smi.smm)
3201 			hflags |= HF_SMM_MASK;
3202 		else
3203 			hflags &= ~HF_SMM_MASK;
3204 		kvm_set_hflags(vcpu, hflags);
3205 
3206 		vcpu->arch.smi_pending = events->smi.pending;
3207 
3208 		if (events->smi.smm) {
3209 			if (events->smi.smm_inside_nmi)
3210 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3211 			else
3212 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3213 			if (lapic_in_kernel(vcpu)) {
3214 				if (events->smi.latched_init)
3215 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3216 				else
3217 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3218 			}
3219 		}
3220 	}
3221 
3222 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3223 
3224 	return 0;
3225 }
3226 
3227 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3228 					     struct kvm_debugregs *dbgregs)
3229 {
3230 	unsigned long val;
3231 
3232 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3233 	kvm_get_dr(vcpu, 6, &val);
3234 	dbgregs->dr6 = val;
3235 	dbgregs->dr7 = vcpu->arch.dr7;
3236 	dbgregs->flags = 0;
3237 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3238 }
3239 
3240 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3241 					    struct kvm_debugregs *dbgregs)
3242 {
3243 	if (dbgregs->flags)
3244 		return -EINVAL;
3245 
3246 	if (dbgregs->dr6 & ~0xffffffffull)
3247 		return -EINVAL;
3248 	if (dbgregs->dr7 & ~0xffffffffull)
3249 		return -EINVAL;
3250 
3251 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3252 	kvm_update_dr0123(vcpu);
3253 	vcpu->arch.dr6 = dbgregs->dr6;
3254 	kvm_update_dr6(vcpu);
3255 	vcpu->arch.dr7 = dbgregs->dr7;
3256 	kvm_update_dr7(vcpu);
3257 
3258 	return 0;
3259 }
3260 
3261 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3262 
3263 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3264 {
3265 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3266 	u64 xstate_bv = xsave->header.xfeatures;
3267 	u64 valid;
3268 
3269 	/*
3270 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3271 	 * leaves 0 and 1 in the loop below.
3272 	 */
3273 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3274 
3275 	/* Set XSTATE_BV */
3276 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3277 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3278 
3279 	/*
3280 	 * Copy each region from the possibly compacted offset to the
3281 	 * non-compacted offset.
3282 	 */
3283 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3284 	while (valid) {
3285 		u64 feature = valid & -valid;
3286 		int index = fls64(feature) - 1;
3287 		void *src = get_xsave_addr(xsave, feature);
3288 
3289 		if (src) {
3290 			u32 size, offset, ecx, edx;
3291 			cpuid_count(XSTATE_CPUID, index,
3292 				    &size, &offset, &ecx, &edx);
3293 			if (feature == XFEATURE_MASK_PKRU)
3294 				memcpy(dest + offset, &vcpu->arch.pkru,
3295 				       sizeof(vcpu->arch.pkru));
3296 			else
3297 				memcpy(dest + offset, src, size);
3298 
3299 		}
3300 
3301 		valid -= feature;
3302 	}
3303 }
3304 
3305 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3306 {
3307 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3308 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3309 	u64 valid;
3310 
3311 	/*
3312 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3313 	 * leaves 0 and 1 in the loop below.
3314 	 */
3315 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3316 
3317 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3318 	xsave->header.xfeatures = xstate_bv;
3319 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3320 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3321 
3322 	/*
3323 	 * Copy each region from the non-compacted offset to the
3324 	 * possibly compacted offset.
3325 	 */
3326 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3327 	while (valid) {
3328 		u64 feature = valid & -valid;
3329 		int index = fls64(feature) - 1;
3330 		void *dest = get_xsave_addr(xsave, feature);
3331 
3332 		if (dest) {
3333 			u32 size, offset, ecx, edx;
3334 			cpuid_count(XSTATE_CPUID, index,
3335 				    &size, &offset, &ecx, &edx);
3336 			if (feature == XFEATURE_MASK_PKRU)
3337 				memcpy(&vcpu->arch.pkru, src + offset,
3338 				       sizeof(vcpu->arch.pkru));
3339 			else
3340 				memcpy(dest, src + offset, size);
3341 		}
3342 
3343 		valid -= feature;
3344 	}
3345 }
3346 
3347 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3348 					 struct kvm_xsave *guest_xsave)
3349 {
3350 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3351 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3352 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3353 	} else {
3354 		memcpy(guest_xsave->region,
3355 			&vcpu->arch.guest_fpu.state.fxsave,
3356 			sizeof(struct fxregs_state));
3357 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3358 			XFEATURE_MASK_FPSSE;
3359 	}
3360 }
3361 
3362 #define XSAVE_MXCSR_OFFSET 24
3363 
3364 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3365 					struct kvm_xsave *guest_xsave)
3366 {
3367 	u64 xstate_bv =
3368 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3369 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3370 
3371 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3372 		/*
3373 		 * Here we allow setting states that are not present in
3374 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3375 		 * with old userspace.
3376 		 */
3377 		if (xstate_bv & ~kvm_supported_xcr0() ||
3378 			mxcsr & ~mxcsr_feature_mask)
3379 			return -EINVAL;
3380 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3381 	} else {
3382 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3383 			mxcsr & ~mxcsr_feature_mask)
3384 			return -EINVAL;
3385 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3386 			guest_xsave->region, sizeof(struct fxregs_state));
3387 	}
3388 	return 0;
3389 }
3390 
3391 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3392 					struct kvm_xcrs *guest_xcrs)
3393 {
3394 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3395 		guest_xcrs->nr_xcrs = 0;
3396 		return;
3397 	}
3398 
3399 	guest_xcrs->nr_xcrs = 1;
3400 	guest_xcrs->flags = 0;
3401 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3402 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3403 }
3404 
3405 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3406 				       struct kvm_xcrs *guest_xcrs)
3407 {
3408 	int i, r = 0;
3409 
3410 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3411 		return -EINVAL;
3412 
3413 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3414 		return -EINVAL;
3415 
3416 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3417 		/* Only support XCR0 currently */
3418 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3419 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3420 				guest_xcrs->xcrs[i].value);
3421 			break;
3422 		}
3423 	if (r)
3424 		r = -EINVAL;
3425 	return r;
3426 }
3427 
3428 /*
3429  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3430  * stopped by the hypervisor.  This function will be called from the host only.
3431  * EINVAL is returned when the host attempts to set the flag for a guest that
3432  * does not support pv clocks.
3433  */
3434 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3435 {
3436 	if (!vcpu->arch.pv_time_enabled)
3437 		return -EINVAL;
3438 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3439 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3440 	return 0;
3441 }
3442 
3443 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3444 				     struct kvm_enable_cap *cap)
3445 {
3446 	if (cap->flags)
3447 		return -EINVAL;
3448 
3449 	switch (cap->cap) {
3450 	case KVM_CAP_HYPERV_SYNIC2:
3451 		if (cap->args[0])
3452 			return -EINVAL;
3453 	case KVM_CAP_HYPERV_SYNIC:
3454 		if (!irqchip_in_kernel(vcpu->kvm))
3455 			return -EINVAL;
3456 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3457 					     KVM_CAP_HYPERV_SYNIC2);
3458 	default:
3459 		return -EINVAL;
3460 	}
3461 }
3462 
3463 long kvm_arch_vcpu_ioctl(struct file *filp,
3464 			 unsigned int ioctl, unsigned long arg)
3465 {
3466 	struct kvm_vcpu *vcpu = filp->private_data;
3467 	void __user *argp = (void __user *)arg;
3468 	int r;
3469 	union {
3470 		struct kvm_lapic_state *lapic;
3471 		struct kvm_xsave *xsave;
3472 		struct kvm_xcrs *xcrs;
3473 		void *buffer;
3474 	} u;
3475 
3476 	u.buffer = NULL;
3477 	switch (ioctl) {
3478 	case KVM_GET_LAPIC: {
3479 		r = -EINVAL;
3480 		if (!lapic_in_kernel(vcpu))
3481 			goto out;
3482 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3483 
3484 		r = -ENOMEM;
3485 		if (!u.lapic)
3486 			goto out;
3487 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3488 		if (r)
3489 			goto out;
3490 		r = -EFAULT;
3491 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3492 			goto out;
3493 		r = 0;
3494 		break;
3495 	}
3496 	case KVM_SET_LAPIC: {
3497 		r = -EINVAL;
3498 		if (!lapic_in_kernel(vcpu))
3499 			goto out;
3500 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3501 		if (IS_ERR(u.lapic))
3502 			return PTR_ERR(u.lapic);
3503 
3504 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3505 		break;
3506 	}
3507 	case KVM_INTERRUPT: {
3508 		struct kvm_interrupt irq;
3509 
3510 		r = -EFAULT;
3511 		if (copy_from_user(&irq, argp, sizeof irq))
3512 			goto out;
3513 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3514 		break;
3515 	}
3516 	case KVM_NMI: {
3517 		r = kvm_vcpu_ioctl_nmi(vcpu);
3518 		break;
3519 	}
3520 	case KVM_SMI: {
3521 		r = kvm_vcpu_ioctl_smi(vcpu);
3522 		break;
3523 	}
3524 	case KVM_SET_CPUID: {
3525 		struct kvm_cpuid __user *cpuid_arg = argp;
3526 		struct kvm_cpuid cpuid;
3527 
3528 		r = -EFAULT;
3529 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3530 			goto out;
3531 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3532 		break;
3533 	}
3534 	case KVM_SET_CPUID2: {
3535 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3536 		struct kvm_cpuid2 cpuid;
3537 
3538 		r = -EFAULT;
3539 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3540 			goto out;
3541 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3542 					      cpuid_arg->entries);
3543 		break;
3544 	}
3545 	case KVM_GET_CPUID2: {
3546 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3547 		struct kvm_cpuid2 cpuid;
3548 
3549 		r = -EFAULT;
3550 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3551 			goto out;
3552 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3553 					      cpuid_arg->entries);
3554 		if (r)
3555 			goto out;
3556 		r = -EFAULT;
3557 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3558 			goto out;
3559 		r = 0;
3560 		break;
3561 	}
3562 	case KVM_GET_MSRS:
3563 		r = msr_io(vcpu, argp, do_get_msr, 1);
3564 		break;
3565 	case KVM_SET_MSRS:
3566 		r = msr_io(vcpu, argp, do_set_msr, 0);
3567 		break;
3568 	case KVM_TPR_ACCESS_REPORTING: {
3569 		struct kvm_tpr_access_ctl tac;
3570 
3571 		r = -EFAULT;
3572 		if (copy_from_user(&tac, argp, sizeof tac))
3573 			goto out;
3574 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3575 		if (r)
3576 			goto out;
3577 		r = -EFAULT;
3578 		if (copy_to_user(argp, &tac, sizeof tac))
3579 			goto out;
3580 		r = 0;
3581 		break;
3582 	};
3583 	case KVM_SET_VAPIC_ADDR: {
3584 		struct kvm_vapic_addr va;
3585 		int idx;
3586 
3587 		r = -EINVAL;
3588 		if (!lapic_in_kernel(vcpu))
3589 			goto out;
3590 		r = -EFAULT;
3591 		if (copy_from_user(&va, argp, sizeof va))
3592 			goto out;
3593 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3594 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3595 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3596 		break;
3597 	}
3598 	case KVM_X86_SETUP_MCE: {
3599 		u64 mcg_cap;
3600 
3601 		r = -EFAULT;
3602 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3603 			goto out;
3604 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3605 		break;
3606 	}
3607 	case KVM_X86_SET_MCE: {
3608 		struct kvm_x86_mce mce;
3609 
3610 		r = -EFAULT;
3611 		if (copy_from_user(&mce, argp, sizeof mce))
3612 			goto out;
3613 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3614 		break;
3615 	}
3616 	case KVM_GET_VCPU_EVENTS: {
3617 		struct kvm_vcpu_events events;
3618 
3619 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3620 
3621 		r = -EFAULT;
3622 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3623 			break;
3624 		r = 0;
3625 		break;
3626 	}
3627 	case KVM_SET_VCPU_EVENTS: {
3628 		struct kvm_vcpu_events events;
3629 
3630 		r = -EFAULT;
3631 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3632 			break;
3633 
3634 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3635 		break;
3636 	}
3637 	case KVM_GET_DEBUGREGS: {
3638 		struct kvm_debugregs dbgregs;
3639 
3640 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3641 
3642 		r = -EFAULT;
3643 		if (copy_to_user(argp, &dbgregs,
3644 				 sizeof(struct kvm_debugregs)))
3645 			break;
3646 		r = 0;
3647 		break;
3648 	}
3649 	case KVM_SET_DEBUGREGS: {
3650 		struct kvm_debugregs dbgregs;
3651 
3652 		r = -EFAULT;
3653 		if (copy_from_user(&dbgregs, argp,
3654 				   sizeof(struct kvm_debugregs)))
3655 			break;
3656 
3657 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3658 		break;
3659 	}
3660 	case KVM_GET_XSAVE: {
3661 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3662 		r = -ENOMEM;
3663 		if (!u.xsave)
3664 			break;
3665 
3666 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3667 
3668 		r = -EFAULT;
3669 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3670 			break;
3671 		r = 0;
3672 		break;
3673 	}
3674 	case KVM_SET_XSAVE: {
3675 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3676 		if (IS_ERR(u.xsave))
3677 			return PTR_ERR(u.xsave);
3678 
3679 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3680 		break;
3681 	}
3682 	case KVM_GET_XCRS: {
3683 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3684 		r = -ENOMEM;
3685 		if (!u.xcrs)
3686 			break;
3687 
3688 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3689 
3690 		r = -EFAULT;
3691 		if (copy_to_user(argp, u.xcrs,
3692 				 sizeof(struct kvm_xcrs)))
3693 			break;
3694 		r = 0;
3695 		break;
3696 	}
3697 	case KVM_SET_XCRS: {
3698 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3699 		if (IS_ERR(u.xcrs))
3700 			return PTR_ERR(u.xcrs);
3701 
3702 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3703 		break;
3704 	}
3705 	case KVM_SET_TSC_KHZ: {
3706 		u32 user_tsc_khz;
3707 
3708 		r = -EINVAL;
3709 		user_tsc_khz = (u32)arg;
3710 
3711 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3712 			goto out;
3713 
3714 		if (user_tsc_khz == 0)
3715 			user_tsc_khz = tsc_khz;
3716 
3717 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3718 			r = 0;
3719 
3720 		goto out;
3721 	}
3722 	case KVM_GET_TSC_KHZ: {
3723 		r = vcpu->arch.virtual_tsc_khz;
3724 		goto out;
3725 	}
3726 	case KVM_KVMCLOCK_CTRL: {
3727 		r = kvm_set_guest_paused(vcpu);
3728 		goto out;
3729 	}
3730 	case KVM_ENABLE_CAP: {
3731 		struct kvm_enable_cap cap;
3732 
3733 		r = -EFAULT;
3734 		if (copy_from_user(&cap, argp, sizeof(cap)))
3735 			goto out;
3736 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3737 		break;
3738 	}
3739 	default:
3740 		r = -EINVAL;
3741 	}
3742 out:
3743 	kfree(u.buffer);
3744 	return r;
3745 }
3746 
3747 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3748 {
3749 	return VM_FAULT_SIGBUS;
3750 }
3751 
3752 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3753 {
3754 	int ret;
3755 
3756 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3757 		return -EINVAL;
3758 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3759 	return ret;
3760 }
3761 
3762 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3763 					      u64 ident_addr)
3764 {
3765 	kvm->arch.ept_identity_map_addr = ident_addr;
3766 	return 0;
3767 }
3768 
3769 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3770 					  u32 kvm_nr_mmu_pages)
3771 {
3772 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3773 		return -EINVAL;
3774 
3775 	mutex_lock(&kvm->slots_lock);
3776 
3777 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3778 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3779 
3780 	mutex_unlock(&kvm->slots_lock);
3781 	return 0;
3782 }
3783 
3784 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3785 {
3786 	return kvm->arch.n_max_mmu_pages;
3787 }
3788 
3789 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3790 {
3791 	struct kvm_pic *pic = kvm->arch.vpic;
3792 	int r;
3793 
3794 	r = 0;
3795 	switch (chip->chip_id) {
3796 	case KVM_IRQCHIP_PIC_MASTER:
3797 		memcpy(&chip->chip.pic, &pic->pics[0],
3798 			sizeof(struct kvm_pic_state));
3799 		break;
3800 	case KVM_IRQCHIP_PIC_SLAVE:
3801 		memcpy(&chip->chip.pic, &pic->pics[1],
3802 			sizeof(struct kvm_pic_state));
3803 		break;
3804 	case KVM_IRQCHIP_IOAPIC:
3805 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
3806 		break;
3807 	default:
3808 		r = -EINVAL;
3809 		break;
3810 	}
3811 	return r;
3812 }
3813 
3814 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3815 {
3816 	struct kvm_pic *pic = kvm->arch.vpic;
3817 	int r;
3818 
3819 	r = 0;
3820 	switch (chip->chip_id) {
3821 	case KVM_IRQCHIP_PIC_MASTER:
3822 		spin_lock(&pic->lock);
3823 		memcpy(&pic->pics[0], &chip->chip.pic,
3824 			sizeof(struct kvm_pic_state));
3825 		spin_unlock(&pic->lock);
3826 		break;
3827 	case KVM_IRQCHIP_PIC_SLAVE:
3828 		spin_lock(&pic->lock);
3829 		memcpy(&pic->pics[1], &chip->chip.pic,
3830 			sizeof(struct kvm_pic_state));
3831 		spin_unlock(&pic->lock);
3832 		break;
3833 	case KVM_IRQCHIP_IOAPIC:
3834 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
3835 		break;
3836 	default:
3837 		r = -EINVAL;
3838 		break;
3839 	}
3840 	kvm_pic_update_irq(pic);
3841 	return r;
3842 }
3843 
3844 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3845 {
3846 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3847 
3848 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3849 
3850 	mutex_lock(&kps->lock);
3851 	memcpy(ps, &kps->channels, sizeof(*ps));
3852 	mutex_unlock(&kps->lock);
3853 	return 0;
3854 }
3855 
3856 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3857 {
3858 	int i;
3859 	struct kvm_pit *pit = kvm->arch.vpit;
3860 
3861 	mutex_lock(&pit->pit_state.lock);
3862 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3863 	for (i = 0; i < 3; i++)
3864 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3865 	mutex_unlock(&pit->pit_state.lock);
3866 	return 0;
3867 }
3868 
3869 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3870 {
3871 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3872 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3873 		sizeof(ps->channels));
3874 	ps->flags = kvm->arch.vpit->pit_state.flags;
3875 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3876 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3877 	return 0;
3878 }
3879 
3880 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3881 {
3882 	int start = 0;
3883 	int i;
3884 	u32 prev_legacy, cur_legacy;
3885 	struct kvm_pit *pit = kvm->arch.vpit;
3886 
3887 	mutex_lock(&pit->pit_state.lock);
3888 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3889 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3890 	if (!prev_legacy && cur_legacy)
3891 		start = 1;
3892 	memcpy(&pit->pit_state.channels, &ps->channels,
3893 	       sizeof(pit->pit_state.channels));
3894 	pit->pit_state.flags = ps->flags;
3895 	for (i = 0; i < 3; i++)
3896 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3897 				   start && i == 0);
3898 	mutex_unlock(&pit->pit_state.lock);
3899 	return 0;
3900 }
3901 
3902 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3903 				 struct kvm_reinject_control *control)
3904 {
3905 	struct kvm_pit *pit = kvm->arch.vpit;
3906 
3907 	if (!pit)
3908 		return -ENXIO;
3909 
3910 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
3911 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3912 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3913 	 */
3914 	mutex_lock(&pit->pit_state.lock);
3915 	kvm_pit_set_reinject(pit, control->pit_reinject);
3916 	mutex_unlock(&pit->pit_state.lock);
3917 
3918 	return 0;
3919 }
3920 
3921 /**
3922  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3923  * @kvm: kvm instance
3924  * @log: slot id and address to which we copy the log
3925  *
3926  * Steps 1-4 below provide general overview of dirty page logging. See
3927  * kvm_get_dirty_log_protect() function description for additional details.
3928  *
3929  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3930  * always flush the TLB (step 4) even if previous step failed  and the dirty
3931  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3932  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3933  * writes will be marked dirty for next log read.
3934  *
3935  *   1. Take a snapshot of the bit and clear it if needed.
3936  *   2. Write protect the corresponding page.
3937  *   3. Copy the snapshot to the userspace.
3938  *   4. Flush TLB's if needed.
3939  */
3940 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3941 {
3942 	bool is_dirty = false;
3943 	int r;
3944 
3945 	mutex_lock(&kvm->slots_lock);
3946 
3947 	/*
3948 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3949 	 */
3950 	if (kvm_x86_ops->flush_log_dirty)
3951 		kvm_x86_ops->flush_log_dirty(kvm);
3952 
3953 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3954 
3955 	/*
3956 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3957 	 * kvm_mmu_slot_remove_write_access().
3958 	 */
3959 	lockdep_assert_held(&kvm->slots_lock);
3960 	if (is_dirty)
3961 		kvm_flush_remote_tlbs(kvm);
3962 
3963 	mutex_unlock(&kvm->slots_lock);
3964 	return r;
3965 }
3966 
3967 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3968 			bool line_status)
3969 {
3970 	if (!irqchip_in_kernel(kvm))
3971 		return -ENXIO;
3972 
3973 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3974 					irq_event->irq, irq_event->level,
3975 					line_status);
3976 	return 0;
3977 }
3978 
3979 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3980 				   struct kvm_enable_cap *cap)
3981 {
3982 	int r;
3983 
3984 	if (cap->flags)
3985 		return -EINVAL;
3986 
3987 	switch (cap->cap) {
3988 	case KVM_CAP_DISABLE_QUIRKS:
3989 		kvm->arch.disabled_quirks = cap->args[0];
3990 		r = 0;
3991 		break;
3992 	case KVM_CAP_SPLIT_IRQCHIP: {
3993 		mutex_lock(&kvm->lock);
3994 		r = -EINVAL;
3995 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3996 			goto split_irqchip_unlock;
3997 		r = -EEXIST;
3998 		if (irqchip_in_kernel(kvm))
3999 			goto split_irqchip_unlock;
4000 		if (kvm->created_vcpus)
4001 			goto split_irqchip_unlock;
4002 		r = kvm_setup_empty_irq_routing(kvm);
4003 		if (r)
4004 			goto split_irqchip_unlock;
4005 		/* Pairs with irqchip_in_kernel. */
4006 		smp_wmb();
4007 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4008 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4009 		r = 0;
4010 split_irqchip_unlock:
4011 		mutex_unlock(&kvm->lock);
4012 		break;
4013 	}
4014 	case KVM_CAP_X2APIC_API:
4015 		r = -EINVAL;
4016 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4017 			break;
4018 
4019 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4020 			kvm->arch.x2apic_format = true;
4021 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4022 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4023 
4024 		r = 0;
4025 		break;
4026 	default:
4027 		r = -EINVAL;
4028 		break;
4029 	}
4030 	return r;
4031 }
4032 
4033 long kvm_arch_vm_ioctl(struct file *filp,
4034 		       unsigned int ioctl, unsigned long arg)
4035 {
4036 	struct kvm *kvm = filp->private_data;
4037 	void __user *argp = (void __user *)arg;
4038 	int r = -ENOTTY;
4039 	/*
4040 	 * This union makes it completely explicit to gcc-3.x
4041 	 * that these two variables' stack usage should be
4042 	 * combined, not added together.
4043 	 */
4044 	union {
4045 		struct kvm_pit_state ps;
4046 		struct kvm_pit_state2 ps2;
4047 		struct kvm_pit_config pit_config;
4048 	} u;
4049 
4050 	switch (ioctl) {
4051 	case KVM_SET_TSS_ADDR:
4052 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4053 		break;
4054 	case KVM_SET_IDENTITY_MAP_ADDR: {
4055 		u64 ident_addr;
4056 
4057 		mutex_lock(&kvm->lock);
4058 		r = -EINVAL;
4059 		if (kvm->created_vcpus)
4060 			goto set_identity_unlock;
4061 		r = -EFAULT;
4062 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4063 			goto set_identity_unlock;
4064 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4065 set_identity_unlock:
4066 		mutex_unlock(&kvm->lock);
4067 		break;
4068 	}
4069 	case KVM_SET_NR_MMU_PAGES:
4070 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4071 		break;
4072 	case KVM_GET_NR_MMU_PAGES:
4073 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4074 		break;
4075 	case KVM_CREATE_IRQCHIP: {
4076 		mutex_lock(&kvm->lock);
4077 
4078 		r = -EEXIST;
4079 		if (irqchip_in_kernel(kvm))
4080 			goto create_irqchip_unlock;
4081 
4082 		r = -EINVAL;
4083 		if (kvm->created_vcpus)
4084 			goto create_irqchip_unlock;
4085 
4086 		r = kvm_pic_init(kvm);
4087 		if (r)
4088 			goto create_irqchip_unlock;
4089 
4090 		r = kvm_ioapic_init(kvm);
4091 		if (r) {
4092 			kvm_pic_destroy(kvm);
4093 			goto create_irqchip_unlock;
4094 		}
4095 
4096 		r = kvm_setup_default_irq_routing(kvm);
4097 		if (r) {
4098 			kvm_ioapic_destroy(kvm);
4099 			kvm_pic_destroy(kvm);
4100 			goto create_irqchip_unlock;
4101 		}
4102 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4103 		smp_wmb();
4104 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4105 	create_irqchip_unlock:
4106 		mutex_unlock(&kvm->lock);
4107 		break;
4108 	}
4109 	case KVM_CREATE_PIT:
4110 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4111 		goto create_pit;
4112 	case KVM_CREATE_PIT2:
4113 		r = -EFAULT;
4114 		if (copy_from_user(&u.pit_config, argp,
4115 				   sizeof(struct kvm_pit_config)))
4116 			goto out;
4117 	create_pit:
4118 		mutex_lock(&kvm->lock);
4119 		r = -EEXIST;
4120 		if (kvm->arch.vpit)
4121 			goto create_pit_unlock;
4122 		r = -ENOMEM;
4123 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4124 		if (kvm->arch.vpit)
4125 			r = 0;
4126 	create_pit_unlock:
4127 		mutex_unlock(&kvm->lock);
4128 		break;
4129 	case KVM_GET_IRQCHIP: {
4130 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4131 		struct kvm_irqchip *chip;
4132 
4133 		chip = memdup_user(argp, sizeof(*chip));
4134 		if (IS_ERR(chip)) {
4135 			r = PTR_ERR(chip);
4136 			goto out;
4137 		}
4138 
4139 		r = -ENXIO;
4140 		if (!irqchip_kernel(kvm))
4141 			goto get_irqchip_out;
4142 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4143 		if (r)
4144 			goto get_irqchip_out;
4145 		r = -EFAULT;
4146 		if (copy_to_user(argp, chip, sizeof *chip))
4147 			goto get_irqchip_out;
4148 		r = 0;
4149 	get_irqchip_out:
4150 		kfree(chip);
4151 		break;
4152 	}
4153 	case KVM_SET_IRQCHIP: {
4154 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4155 		struct kvm_irqchip *chip;
4156 
4157 		chip = memdup_user(argp, sizeof(*chip));
4158 		if (IS_ERR(chip)) {
4159 			r = PTR_ERR(chip);
4160 			goto out;
4161 		}
4162 
4163 		r = -ENXIO;
4164 		if (!irqchip_kernel(kvm))
4165 			goto set_irqchip_out;
4166 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4167 		if (r)
4168 			goto set_irqchip_out;
4169 		r = 0;
4170 	set_irqchip_out:
4171 		kfree(chip);
4172 		break;
4173 	}
4174 	case KVM_GET_PIT: {
4175 		r = -EFAULT;
4176 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4177 			goto out;
4178 		r = -ENXIO;
4179 		if (!kvm->arch.vpit)
4180 			goto out;
4181 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4182 		if (r)
4183 			goto out;
4184 		r = -EFAULT;
4185 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4186 			goto out;
4187 		r = 0;
4188 		break;
4189 	}
4190 	case KVM_SET_PIT: {
4191 		r = -EFAULT;
4192 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4193 			goto out;
4194 		r = -ENXIO;
4195 		if (!kvm->arch.vpit)
4196 			goto out;
4197 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4198 		break;
4199 	}
4200 	case KVM_GET_PIT2: {
4201 		r = -ENXIO;
4202 		if (!kvm->arch.vpit)
4203 			goto out;
4204 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4205 		if (r)
4206 			goto out;
4207 		r = -EFAULT;
4208 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4209 			goto out;
4210 		r = 0;
4211 		break;
4212 	}
4213 	case KVM_SET_PIT2: {
4214 		r = -EFAULT;
4215 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4216 			goto out;
4217 		r = -ENXIO;
4218 		if (!kvm->arch.vpit)
4219 			goto out;
4220 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4221 		break;
4222 	}
4223 	case KVM_REINJECT_CONTROL: {
4224 		struct kvm_reinject_control control;
4225 		r =  -EFAULT;
4226 		if (copy_from_user(&control, argp, sizeof(control)))
4227 			goto out;
4228 		r = kvm_vm_ioctl_reinject(kvm, &control);
4229 		break;
4230 	}
4231 	case KVM_SET_BOOT_CPU_ID:
4232 		r = 0;
4233 		mutex_lock(&kvm->lock);
4234 		if (kvm->created_vcpus)
4235 			r = -EBUSY;
4236 		else
4237 			kvm->arch.bsp_vcpu_id = arg;
4238 		mutex_unlock(&kvm->lock);
4239 		break;
4240 	case KVM_XEN_HVM_CONFIG: {
4241 		r = -EFAULT;
4242 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4243 				   sizeof(struct kvm_xen_hvm_config)))
4244 			goto out;
4245 		r = -EINVAL;
4246 		if (kvm->arch.xen_hvm_config.flags)
4247 			goto out;
4248 		r = 0;
4249 		break;
4250 	}
4251 	case KVM_SET_CLOCK: {
4252 		struct kvm_clock_data user_ns;
4253 		u64 now_ns;
4254 
4255 		r = -EFAULT;
4256 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4257 			goto out;
4258 
4259 		r = -EINVAL;
4260 		if (user_ns.flags)
4261 			goto out;
4262 
4263 		r = 0;
4264 		/*
4265 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4266 		 * kvm_gen_update_masterclock() can be cut down to locked
4267 		 * pvclock_update_vm_gtod_copy().
4268 		 */
4269 		kvm_gen_update_masterclock(kvm);
4270 		now_ns = get_kvmclock_ns(kvm);
4271 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4272 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4273 		break;
4274 	}
4275 	case KVM_GET_CLOCK: {
4276 		struct kvm_clock_data user_ns;
4277 		u64 now_ns;
4278 
4279 		now_ns = get_kvmclock_ns(kvm);
4280 		user_ns.clock = now_ns;
4281 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4282 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4283 
4284 		r = -EFAULT;
4285 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4286 			goto out;
4287 		r = 0;
4288 		break;
4289 	}
4290 	case KVM_ENABLE_CAP: {
4291 		struct kvm_enable_cap cap;
4292 
4293 		r = -EFAULT;
4294 		if (copy_from_user(&cap, argp, sizeof(cap)))
4295 			goto out;
4296 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4297 		break;
4298 	}
4299 	default:
4300 		r = -ENOTTY;
4301 	}
4302 out:
4303 	return r;
4304 }
4305 
4306 static void kvm_init_msr_list(void)
4307 {
4308 	u32 dummy[2];
4309 	unsigned i, j;
4310 
4311 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4312 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4313 			continue;
4314 
4315 		/*
4316 		 * Even MSRs that are valid in the host may not be exposed
4317 		 * to the guests in some cases.
4318 		 */
4319 		switch (msrs_to_save[i]) {
4320 		case MSR_IA32_BNDCFGS:
4321 			if (!kvm_x86_ops->mpx_supported())
4322 				continue;
4323 			break;
4324 		case MSR_TSC_AUX:
4325 			if (!kvm_x86_ops->rdtscp_supported())
4326 				continue;
4327 			break;
4328 		default:
4329 			break;
4330 		}
4331 
4332 		if (j < i)
4333 			msrs_to_save[j] = msrs_to_save[i];
4334 		j++;
4335 	}
4336 	num_msrs_to_save = j;
4337 
4338 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4339 		switch (emulated_msrs[i]) {
4340 		case MSR_IA32_SMBASE:
4341 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4342 				continue;
4343 			break;
4344 		default:
4345 			break;
4346 		}
4347 
4348 		if (j < i)
4349 			emulated_msrs[j] = emulated_msrs[i];
4350 		j++;
4351 	}
4352 	num_emulated_msrs = j;
4353 }
4354 
4355 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4356 			   const void *v)
4357 {
4358 	int handled = 0;
4359 	int n;
4360 
4361 	do {
4362 		n = min(len, 8);
4363 		if (!(lapic_in_kernel(vcpu) &&
4364 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4365 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4366 			break;
4367 		handled += n;
4368 		addr += n;
4369 		len -= n;
4370 		v += n;
4371 	} while (len);
4372 
4373 	return handled;
4374 }
4375 
4376 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4377 {
4378 	int handled = 0;
4379 	int n;
4380 
4381 	do {
4382 		n = min(len, 8);
4383 		if (!(lapic_in_kernel(vcpu) &&
4384 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4385 					 addr, n, v))
4386 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4387 			break;
4388 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4389 		handled += n;
4390 		addr += n;
4391 		len -= n;
4392 		v += n;
4393 	} while (len);
4394 
4395 	return handled;
4396 }
4397 
4398 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4399 			struct kvm_segment *var, int seg)
4400 {
4401 	kvm_x86_ops->set_segment(vcpu, var, seg);
4402 }
4403 
4404 void kvm_get_segment(struct kvm_vcpu *vcpu,
4405 		     struct kvm_segment *var, int seg)
4406 {
4407 	kvm_x86_ops->get_segment(vcpu, var, seg);
4408 }
4409 
4410 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4411 			   struct x86_exception *exception)
4412 {
4413 	gpa_t t_gpa;
4414 
4415 	BUG_ON(!mmu_is_nested(vcpu));
4416 
4417 	/* NPT walks are always user-walks */
4418 	access |= PFERR_USER_MASK;
4419 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4420 
4421 	return t_gpa;
4422 }
4423 
4424 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4425 			      struct x86_exception *exception)
4426 {
4427 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4428 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4429 }
4430 
4431  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4432 				struct x86_exception *exception)
4433 {
4434 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4435 	access |= PFERR_FETCH_MASK;
4436 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4437 }
4438 
4439 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4440 			       struct x86_exception *exception)
4441 {
4442 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4443 	access |= PFERR_WRITE_MASK;
4444 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4445 }
4446 
4447 /* uses this to access any guest's mapped memory without checking CPL */
4448 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4449 				struct x86_exception *exception)
4450 {
4451 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4452 }
4453 
4454 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4455 				      struct kvm_vcpu *vcpu, u32 access,
4456 				      struct x86_exception *exception)
4457 {
4458 	void *data = val;
4459 	int r = X86EMUL_CONTINUE;
4460 
4461 	while (bytes) {
4462 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4463 							    exception);
4464 		unsigned offset = addr & (PAGE_SIZE-1);
4465 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4466 		int ret;
4467 
4468 		if (gpa == UNMAPPED_GVA)
4469 			return X86EMUL_PROPAGATE_FAULT;
4470 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4471 					       offset, toread);
4472 		if (ret < 0) {
4473 			r = X86EMUL_IO_NEEDED;
4474 			goto out;
4475 		}
4476 
4477 		bytes -= toread;
4478 		data += toread;
4479 		addr += toread;
4480 	}
4481 out:
4482 	return r;
4483 }
4484 
4485 /* used for instruction fetching */
4486 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4487 				gva_t addr, void *val, unsigned int bytes,
4488 				struct x86_exception *exception)
4489 {
4490 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4491 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4492 	unsigned offset;
4493 	int ret;
4494 
4495 	/* Inline kvm_read_guest_virt_helper for speed.  */
4496 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4497 						    exception);
4498 	if (unlikely(gpa == UNMAPPED_GVA))
4499 		return X86EMUL_PROPAGATE_FAULT;
4500 
4501 	offset = addr & (PAGE_SIZE-1);
4502 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4503 		bytes = (unsigned)PAGE_SIZE - offset;
4504 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4505 				       offset, bytes);
4506 	if (unlikely(ret < 0))
4507 		return X86EMUL_IO_NEEDED;
4508 
4509 	return X86EMUL_CONTINUE;
4510 }
4511 
4512 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4513 			       gva_t addr, void *val, unsigned int bytes,
4514 			       struct x86_exception *exception)
4515 {
4516 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4517 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4518 
4519 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4520 					  exception);
4521 }
4522 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4523 
4524 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4525 				      gva_t addr, void *val, unsigned int bytes,
4526 				      struct x86_exception *exception)
4527 {
4528 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4529 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4530 }
4531 
4532 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4533 		unsigned long addr, void *val, unsigned int bytes)
4534 {
4535 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4536 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4537 
4538 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4539 }
4540 
4541 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4542 				       gva_t addr, void *val,
4543 				       unsigned int bytes,
4544 				       struct x86_exception *exception)
4545 {
4546 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4547 	void *data = val;
4548 	int r = X86EMUL_CONTINUE;
4549 
4550 	while (bytes) {
4551 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4552 							     PFERR_WRITE_MASK,
4553 							     exception);
4554 		unsigned offset = addr & (PAGE_SIZE-1);
4555 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4556 		int ret;
4557 
4558 		if (gpa == UNMAPPED_GVA)
4559 			return X86EMUL_PROPAGATE_FAULT;
4560 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4561 		if (ret < 0) {
4562 			r = X86EMUL_IO_NEEDED;
4563 			goto out;
4564 		}
4565 
4566 		bytes -= towrite;
4567 		data += towrite;
4568 		addr += towrite;
4569 	}
4570 out:
4571 	return r;
4572 }
4573 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4574 
4575 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4576 			    gpa_t gpa, bool write)
4577 {
4578 	/* For APIC access vmexit */
4579 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4580 		return 1;
4581 
4582 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4583 		trace_vcpu_match_mmio(gva, gpa, write, true);
4584 		return 1;
4585 	}
4586 
4587 	return 0;
4588 }
4589 
4590 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4591 				gpa_t *gpa, struct x86_exception *exception,
4592 				bool write)
4593 {
4594 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4595 		| (write ? PFERR_WRITE_MASK : 0);
4596 
4597 	/*
4598 	 * currently PKRU is only applied to ept enabled guest so
4599 	 * there is no pkey in EPT page table for L1 guest or EPT
4600 	 * shadow page table for L2 guest.
4601 	 */
4602 	if (vcpu_match_mmio_gva(vcpu, gva)
4603 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4604 				 vcpu->arch.access, 0, access)) {
4605 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4606 					(gva & (PAGE_SIZE - 1));
4607 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4608 		return 1;
4609 	}
4610 
4611 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4612 
4613 	if (*gpa == UNMAPPED_GVA)
4614 		return -1;
4615 
4616 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4617 }
4618 
4619 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4620 			const void *val, int bytes)
4621 {
4622 	int ret;
4623 
4624 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4625 	if (ret < 0)
4626 		return 0;
4627 	kvm_page_track_write(vcpu, gpa, val, bytes);
4628 	return 1;
4629 }
4630 
4631 struct read_write_emulator_ops {
4632 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4633 				  int bytes);
4634 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4635 				  void *val, int bytes);
4636 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4637 			       int bytes, void *val);
4638 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4639 				    void *val, int bytes);
4640 	bool write;
4641 };
4642 
4643 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4644 {
4645 	if (vcpu->mmio_read_completed) {
4646 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4647 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4648 		vcpu->mmio_read_completed = 0;
4649 		return 1;
4650 	}
4651 
4652 	return 0;
4653 }
4654 
4655 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4656 			void *val, int bytes)
4657 {
4658 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4659 }
4660 
4661 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4662 			 void *val, int bytes)
4663 {
4664 	return emulator_write_phys(vcpu, gpa, val, bytes);
4665 }
4666 
4667 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4668 {
4669 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4670 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4671 }
4672 
4673 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4674 			  void *val, int bytes)
4675 {
4676 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4677 	return X86EMUL_IO_NEEDED;
4678 }
4679 
4680 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4681 			   void *val, int bytes)
4682 {
4683 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4684 
4685 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4686 	return X86EMUL_CONTINUE;
4687 }
4688 
4689 static const struct read_write_emulator_ops read_emultor = {
4690 	.read_write_prepare = read_prepare,
4691 	.read_write_emulate = read_emulate,
4692 	.read_write_mmio = vcpu_mmio_read,
4693 	.read_write_exit_mmio = read_exit_mmio,
4694 };
4695 
4696 static const struct read_write_emulator_ops write_emultor = {
4697 	.read_write_emulate = write_emulate,
4698 	.read_write_mmio = write_mmio,
4699 	.read_write_exit_mmio = write_exit_mmio,
4700 	.write = true,
4701 };
4702 
4703 static int emulator_read_write_onepage(unsigned long addr, void *val,
4704 				       unsigned int bytes,
4705 				       struct x86_exception *exception,
4706 				       struct kvm_vcpu *vcpu,
4707 				       const struct read_write_emulator_ops *ops)
4708 {
4709 	gpa_t gpa;
4710 	int handled, ret;
4711 	bool write = ops->write;
4712 	struct kvm_mmio_fragment *frag;
4713 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4714 
4715 	/*
4716 	 * If the exit was due to a NPF we may already have a GPA.
4717 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4718 	 * Note, this cannot be used on string operations since string
4719 	 * operation using rep will only have the initial GPA from the NPF
4720 	 * occurred.
4721 	 */
4722 	if (vcpu->arch.gpa_available &&
4723 	    emulator_can_use_gpa(ctxt) &&
4724 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4725 		gpa = vcpu->arch.gpa_val;
4726 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4727 	} else {
4728 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4729 		if (ret < 0)
4730 			return X86EMUL_PROPAGATE_FAULT;
4731 	}
4732 
4733 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
4734 		return X86EMUL_CONTINUE;
4735 
4736 	/*
4737 	 * Is this MMIO handled locally?
4738 	 */
4739 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4740 	if (handled == bytes)
4741 		return X86EMUL_CONTINUE;
4742 
4743 	gpa += handled;
4744 	bytes -= handled;
4745 	val += handled;
4746 
4747 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4748 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4749 	frag->gpa = gpa;
4750 	frag->data = val;
4751 	frag->len = bytes;
4752 	return X86EMUL_CONTINUE;
4753 }
4754 
4755 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4756 			unsigned long addr,
4757 			void *val, unsigned int bytes,
4758 			struct x86_exception *exception,
4759 			const struct read_write_emulator_ops *ops)
4760 {
4761 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4762 	gpa_t gpa;
4763 	int rc;
4764 
4765 	if (ops->read_write_prepare &&
4766 		  ops->read_write_prepare(vcpu, val, bytes))
4767 		return X86EMUL_CONTINUE;
4768 
4769 	vcpu->mmio_nr_fragments = 0;
4770 
4771 	/* Crossing a page boundary? */
4772 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4773 		int now;
4774 
4775 		now = -addr & ~PAGE_MASK;
4776 		rc = emulator_read_write_onepage(addr, val, now, exception,
4777 						 vcpu, ops);
4778 
4779 		if (rc != X86EMUL_CONTINUE)
4780 			return rc;
4781 		addr += now;
4782 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4783 			addr = (u32)addr;
4784 		val += now;
4785 		bytes -= now;
4786 	}
4787 
4788 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4789 					 vcpu, ops);
4790 	if (rc != X86EMUL_CONTINUE)
4791 		return rc;
4792 
4793 	if (!vcpu->mmio_nr_fragments)
4794 		return rc;
4795 
4796 	gpa = vcpu->mmio_fragments[0].gpa;
4797 
4798 	vcpu->mmio_needed = 1;
4799 	vcpu->mmio_cur_fragment = 0;
4800 
4801 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4802 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4803 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4804 	vcpu->run->mmio.phys_addr = gpa;
4805 
4806 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4807 }
4808 
4809 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4810 				  unsigned long addr,
4811 				  void *val,
4812 				  unsigned int bytes,
4813 				  struct x86_exception *exception)
4814 {
4815 	return emulator_read_write(ctxt, addr, val, bytes,
4816 				   exception, &read_emultor);
4817 }
4818 
4819 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4820 			    unsigned long addr,
4821 			    const void *val,
4822 			    unsigned int bytes,
4823 			    struct x86_exception *exception)
4824 {
4825 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4826 				   exception, &write_emultor);
4827 }
4828 
4829 #define CMPXCHG_TYPE(t, ptr, old, new) \
4830 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4831 
4832 #ifdef CONFIG_X86_64
4833 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4834 #else
4835 #  define CMPXCHG64(ptr, old, new) \
4836 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4837 #endif
4838 
4839 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4840 				     unsigned long addr,
4841 				     const void *old,
4842 				     const void *new,
4843 				     unsigned int bytes,
4844 				     struct x86_exception *exception)
4845 {
4846 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4847 	gpa_t gpa;
4848 	struct page *page;
4849 	char *kaddr;
4850 	bool exchanged;
4851 
4852 	/* guests cmpxchg8b have to be emulated atomically */
4853 	if (bytes > 8 || (bytes & (bytes - 1)))
4854 		goto emul_write;
4855 
4856 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4857 
4858 	if (gpa == UNMAPPED_GVA ||
4859 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4860 		goto emul_write;
4861 
4862 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4863 		goto emul_write;
4864 
4865 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4866 	if (is_error_page(page))
4867 		goto emul_write;
4868 
4869 	kaddr = kmap_atomic(page);
4870 	kaddr += offset_in_page(gpa);
4871 	switch (bytes) {
4872 	case 1:
4873 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4874 		break;
4875 	case 2:
4876 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4877 		break;
4878 	case 4:
4879 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4880 		break;
4881 	case 8:
4882 		exchanged = CMPXCHG64(kaddr, old, new);
4883 		break;
4884 	default:
4885 		BUG();
4886 	}
4887 	kunmap_atomic(kaddr);
4888 	kvm_release_page_dirty(page);
4889 
4890 	if (!exchanged)
4891 		return X86EMUL_CMPXCHG_FAILED;
4892 
4893 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4894 	kvm_page_track_write(vcpu, gpa, new, bytes);
4895 
4896 	return X86EMUL_CONTINUE;
4897 
4898 emul_write:
4899 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4900 
4901 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4902 }
4903 
4904 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4905 {
4906 	int r = 0, i;
4907 
4908 	for (i = 0; i < vcpu->arch.pio.count; i++) {
4909 		if (vcpu->arch.pio.in)
4910 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4911 					    vcpu->arch.pio.size, pd);
4912 		else
4913 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4914 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
4915 					     pd);
4916 		if (r)
4917 			break;
4918 		pd += vcpu->arch.pio.size;
4919 	}
4920 	return r;
4921 }
4922 
4923 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4924 			       unsigned short port, void *val,
4925 			       unsigned int count, bool in)
4926 {
4927 	vcpu->arch.pio.port = port;
4928 	vcpu->arch.pio.in = in;
4929 	vcpu->arch.pio.count  = count;
4930 	vcpu->arch.pio.size = size;
4931 
4932 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4933 		vcpu->arch.pio.count = 0;
4934 		return 1;
4935 	}
4936 
4937 	vcpu->run->exit_reason = KVM_EXIT_IO;
4938 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4939 	vcpu->run->io.size = size;
4940 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4941 	vcpu->run->io.count = count;
4942 	vcpu->run->io.port = port;
4943 
4944 	return 0;
4945 }
4946 
4947 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4948 				    int size, unsigned short port, void *val,
4949 				    unsigned int count)
4950 {
4951 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4952 	int ret;
4953 
4954 	if (vcpu->arch.pio.count)
4955 		goto data_avail;
4956 
4957 	memset(vcpu->arch.pio_data, 0, size * count);
4958 
4959 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4960 	if (ret) {
4961 data_avail:
4962 		memcpy(val, vcpu->arch.pio_data, size * count);
4963 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4964 		vcpu->arch.pio.count = 0;
4965 		return 1;
4966 	}
4967 
4968 	return 0;
4969 }
4970 
4971 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4972 				     int size, unsigned short port,
4973 				     const void *val, unsigned int count)
4974 {
4975 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4976 
4977 	memcpy(vcpu->arch.pio_data, val, size * count);
4978 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4979 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4980 }
4981 
4982 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4983 {
4984 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4985 }
4986 
4987 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4988 {
4989 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4990 }
4991 
4992 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4993 {
4994 	if (!need_emulate_wbinvd(vcpu))
4995 		return X86EMUL_CONTINUE;
4996 
4997 	if (kvm_x86_ops->has_wbinvd_exit()) {
4998 		int cpu = get_cpu();
4999 
5000 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5001 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5002 				wbinvd_ipi, NULL, 1);
5003 		put_cpu();
5004 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5005 	} else
5006 		wbinvd();
5007 	return X86EMUL_CONTINUE;
5008 }
5009 
5010 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5011 {
5012 	kvm_emulate_wbinvd_noskip(vcpu);
5013 	return kvm_skip_emulated_instruction(vcpu);
5014 }
5015 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5016 
5017 
5018 
5019 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5020 {
5021 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5022 }
5023 
5024 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5025 			   unsigned long *dest)
5026 {
5027 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5028 }
5029 
5030 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5031 			   unsigned long value)
5032 {
5033 
5034 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5035 }
5036 
5037 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5038 {
5039 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5040 }
5041 
5042 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5043 {
5044 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5045 	unsigned long value;
5046 
5047 	switch (cr) {
5048 	case 0:
5049 		value = kvm_read_cr0(vcpu);
5050 		break;
5051 	case 2:
5052 		value = vcpu->arch.cr2;
5053 		break;
5054 	case 3:
5055 		value = kvm_read_cr3(vcpu);
5056 		break;
5057 	case 4:
5058 		value = kvm_read_cr4(vcpu);
5059 		break;
5060 	case 8:
5061 		value = kvm_get_cr8(vcpu);
5062 		break;
5063 	default:
5064 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5065 		return 0;
5066 	}
5067 
5068 	return value;
5069 }
5070 
5071 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5072 {
5073 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5074 	int res = 0;
5075 
5076 	switch (cr) {
5077 	case 0:
5078 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5079 		break;
5080 	case 2:
5081 		vcpu->arch.cr2 = val;
5082 		break;
5083 	case 3:
5084 		res = kvm_set_cr3(vcpu, val);
5085 		break;
5086 	case 4:
5087 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5088 		break;
5089 	case 8:
5090 		res = kvm_set_cr8(vcpu, val);
5091 		break;
5092 	default:
5093 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5094 		res = -1;
5095 	}
5096 
5097 	return res;
5098 }
5099 
5100 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5101 {
5102 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5103 }
5104 
5105 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5106 {
5107 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5108 }
5109 
5110 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5111 {
5112 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5113 }
5114 
5115 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5116 {
5117 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5118 }
5119 
5120 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5121 {
5122 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5123 }
5124 
5125 static unsigned long emulator_get_cached_segment_base(
5126 	struct x86_emulate_ctxt *ctxt, int seg)
5127 {
5128 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5129 }
5130 
5131 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5132 				 struct desc_struct *desc, u32 *base3,
5133 				 int seg)
5134 {
5135 	struct kvm_segment var;
5136 
5137 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5138 	*selector = var.selector;
5139 
5140 	if (var.unusable) {
5141 		memset(desc, 0, sizeof(*desc));
5142 		if (base3)
5143 			*base3 = 0;
5144 		return false;
5145 	}
5146 
5147 	if (var.g)
5148 		var.limit >>= 12;
5149 	set_desc_limit(desc, var.limit);
5150 	set_desc_base(desc, (unsigned long)var.base);
5151 #ifdef CONFIG_X86_64
5152 	if (base3)
5153 		*base3 = var.base >> 32;
5154 #endif
5155 	desc->type = var.type;
5156 	desc->s = var.s;
5157 	desc->dpl = var.dpl;
5158 	desc->p = var.present;
5159 	desc->avl = var.avl;
5160 	desc->l = var.l;
5161 	desc->d = var.db;
5162 	desc->g = var.g;
5163 
5164 	return true;
5165 }
5166 
5167 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5168 				 struct desc_struct *desc, u32 base3,
5169 				 int seg)
5170 {
5171 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5172 	struct kvm_segment var;
5173 
5174 	var.selector = selector;
5175 	var.base = get_desc_base(desc);
5176 #ifdef CONFIG_X86_64
5177 	var.base |= ((u64)base3) << 32;
5178 #endif
5179 	var.limit = get_desc_limit(desc);
5180 	if (desc->g)
5181 		var.limit = (var.limit << 12) | 0xfff;
5182 	var.type = desc->type;
5183 	var.dpl = desc->dpl;
5184 	var.db = desc->d;
5185 	var.s = desc->s;
5186 	var.l = desc->l;
5187 	var.g = desc->g;
5188 	var.avl = desc->avl;
5189 	var.present = desc->p;
5190 	var.unusable = !var.present;
5191 	var.padding = 0;
5192 
5193 	kvm_set_segment(vcpu, &var, seg);
5194 	return;
5195 }
5196 
5197 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5198 			    u32 msr_index, u64 *pdata)
5199 {
5200 	struct msr_data msr;
5201 	int r;
5202 
5203 	msr.index = msr_index;
5204 	msr.host_initiated = false;
5205 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5206 	if (r)
5207 		return r;
5208 
5209 	*pdata = msr.data;
5210 	return 0;
5211 }
5212 
5213 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5214 			    u32 msr_index, u64 data)
5215 {
5216 	struct msr_data msr;
5217 
5218 	msr.data = data;
5219 	msr.index = msr_index;
5220 	msr.host_initiated = false;
5221 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5222 }
5223 
5224 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5225 {
5226 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5227 
5228 	return vcpu->arch.smbase;
5229 }
5230 
5231 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5232 {
5233 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5234 
5235 	vcpu->arch.smbase = smbase;
5236 }
5237 
5238 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5239 			      u32 pmc)
5240 {
5241 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5242 }
5243 
5244 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5245 			     u32 pmc, u64 *pdata)
5246 {
5247 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5248 }
5249 
5250 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5251 {
5252 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5253 }
5254 
5255 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5256 {
5257 	preempt_disable();
5258 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5259 }
5260 
5261 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5262 {
5263 	preempt_enable();
5264 }
5265 
5266 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5267 			      struct x86_instruction_info *info,
5268 			      enum x86_intercept_stage stage)
5269 {
5270 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5271 }
5272 
5273 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5274 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5275 {
5276 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5277 }
5278 
5279 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5280 {
5281 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5282 }
5283 
5284 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5285 {
5286 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5287 }
5288 
5289 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5290 {
5291 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5292 }
5293 
5294 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5295 {
5296 	return emul_to_vcpu(ctxt)->arch.hflags;
5297 }
5298 
5299 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5300 {
5301 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5302 }
5303 
5304 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5305 {
5306 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5307 }
5308 
5309 static const struct x86_emulate_ops emulate_ops = {
5310 	.read_gpr            = emulator_read_gpr,
5311 	.write_gpr           = emulator_write_gpr,
5312 	.read_std            = kvm_read_guest_virt_system,
5313 	.write_std           = kvm_write_guest_virt_system,
5314 	.read_phys           = kvm_read_guest_phys_system,
5315 	.fetch               = kvm_fetch_guest_virt,
5316 	.read_emulated       = emulator_read_emulated,
5317 	.write_emulated      = emulator_write_emulated,
5318 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5319 	.invlpg              = emulator_invlpg,
5320 	.pio_in_emulated     = emulator_pio_in_emulated,
5321 	.pio_out_emulated    = emulator_pio_out_emulated,
5322 	.get_segment         = emulator_get_segment,
5323 	.set_segment         = emulator_set_segment,
5324 	.get_cached_segment_base = emulator_get_cached_segment_base,
5325 	.get_gdt             = emulator_get_gdt,
5326 	.get_idt	     = emulator_get_idt,
5327 	.set_gdt             = emulator_set_gdt,
5328 	.set_idt	     = emulator_set_idt,
5329 	.get_cr              = emulator_get_cr,
5330 	.set_cr              = emulator_set_cr,
5331 	.cpl                 = emulator_get_cpl,
5332 	.get_dr              = emulator_get_dr,
5333 	.set_dr              = emulator_set_dr,
5334 	.get_smbase          = emulator_get_smbase,
5335 	.set_smbase          = emulator_set_smbase,
5336 	.set_msr             = emulator_set_msr,
5337 	.get_msr             = emulator_get_msr,
5338 	.check_pmc	     = emulator_check_pmc,
5339 	.read_pmc            = emulator_read_pmc,
5340 	.halt                = emulator_halt,
5341 	.wbinvd              = emulator_wbinvd,
5342 	.fix_hypercall       = emulator_fix_hypercall,
5343 	.get_fpu             = emulator_get_fpu,
5344 	.put_fpu             = emulator_put_fpu,
5345 	.intercept           = emulator_intercept,
5346 	.get_cpuid           = emulator_get_cpuid,
5347 	.set_nmi_mask        = emulator_set_nmi_mask,
5348 	.get_hflags          = emulator_get_hflags,
5349 	.set_hflags          = emulator_set_hflags,
5350 	.pre_leave_smm       = emulator_pre_leave_smm,
5351 };
5352 
5353 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5354 {
5355 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5356 	/*
5357 	 * an sti; sti; sequence only disable interrupts for the first
5358 	 * instruction. So, if the last instruction, be it emulated or
5359 	 * not, left the system with the INT_STI flag enabled, it
5360 	 * means that the last instruction is an sti. We should not
5361 	 * leave the flag on in this case. The same goes for mov ss
5362 	 */
5363 	if (int_shadow & mask)
5364 		mask = 0;
5365 	if (unlikely(int_shadow || mask)) {
5366 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5367 		if (!mask)
5368 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5369 	}
5370 }
5371 
5372 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5373 {
5374 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5375 	if (ctxt->exception.vector == PF_VECTOR)
5376 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5377 
5378 	if (ctxt->exception.error_code_valid)
5379 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5380 				      ctxt->exception.error_code);
5381 	else
5382 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5383 	return false;
5384 }
5385 
5386 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5387 {
5388 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5389 	int cs_db, cs_l;
5390 
5391 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5392 
5393 	ctxt->eflags = kvm_get_rflags(vcpu);
5394 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5395 
5396 	ctxt->eip = kvm_rip_read(vcpu);
5397 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5398 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5399 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5400 		     cs_db				? X86EMUL_MODE_PROT32 :
5401 							  X86EMUL_MODE_PROT16;
5402 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5403 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5404 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5405 
5406 	init_decode_cache(ctxt);
5407 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5408 }
5409 
5410 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5411 {
5412 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5413 	int ret;
5414 
5415 	init_emulate_ctxt(vcpu);
5416 
5417 	ctxt->op_bytes = 2;
5418 	ctxt->ad_bytes = 2;
5419 	ctxt->_eip = ctxt->eip + inc_eip;
5420 	ret = emulate_int_real(ctxt, irq);
5421 
5422 	if (ret != X86EMUL_CONTINUE)
5423 		return EMULATE_FAIL;
5424 
5425 	ctxt->eip = ctxt->_eip;
5426 	kvm_rip_write(vcpu, ctxt->eip);
5427 	kvm_set_rflags(vcpu, ctxt->eflags);
5428 
5429 	if (irq == NMI_VECTOR)
5430 		vcpu->arch.nmi_pending = 0;
5431 	else
5432 		vcpu->arch.interrupt.pending = false;
5433 
5434 	return EMULATE_DONE;
5435 }
5436 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5437 
5438 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5439 {
5440 	int r = EMULATE_DONE;
5441 
5442 	++vcpu->stat.insn_emulation_fail;
5443 	trace_kvm_emulate_insn_failed(vcpu);
5444 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5445 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5446 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5447 		vcpu->run->internal.ndata = 0;
5448 		r = EMULATE_USER_EXIT;
5449 	}
5450 	kvm_queue_exception(vcpu, UD_VECTOR);
5451 
5452 	return r;
5453 }
5454 
5455 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5456 				  bool write_fault_to_shadow_pgtable,
5457 				  int emulation_type)
5458 {
5459 	gpa_t gpa = cr2;
5460 	kvm_pfn_t pfn;
5461 
5462 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5463 		return false;
5464 
5465 	if (!vcpu->arch.mmu.direct_map) {
5466 		/*
5467 		 * Write permission should be allowed since only
5468 		 * write access need to be emulated.
5469 		 */
5470 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5471 
5472 		/*
5473 		 * If the mapping is invalid in guest, let cpu retry
5474 		 * it to generate fault.
5475 		 */
5476 		if (gpa == UNMAPPED_GVA)
5477 			return true;
5478 	}
5479 
5480 	/*
5481 	 * Do not retry the unhandleable instruction if it faults on the
5482 	 * readonly host memory, otherwise it will goto a infinite loop:
5483 	 * retry instruction -> write #PF -> emulation fail -> retry
5484 	 * instruction -> ...
5485 	 */
5486 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5487 
5488 	/*
5489 	 * If the instruction failed on the error pfn, it can not be fixed,
5490 	 * report the error to userspace.
5491 	 */
5492 	if (is_error_noslot_pfn(pfn))
5493 		return false;
5494 
5495 	kvm_release_pfn_clean(pfn);
5496 
5497 	/* The instructions are well-emulated on direct mmu. */
5498 	if (vcpu->arch.mmu.direct_map) {
5499 		unsigned int indirect_shadow_pages;
5500 
5501 		spin_lock(&vcpu->kvm->mmu_lock);
5502 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5503 		spin_unlock(&vcpu->kvm->mmu_lock);
5504 
5505 		if (indirect_shadow_pages)
5506 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5507 
5508 		return true;
5509 	}
5510 
5511 	/*
5512 	 * if emulation was due to access to shadowed page table
5513 	 * and it failed try to unshadow page and re-enter the
5514 	 * guest to let CPU execute the instruction.
5515 	 */
5516 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5517 
5518 	/*
5519 	 * If the access faults on its page table, it can not
5520 	 * be fixed by unprotecting shadow page and it should
5521 	 * be reported to userspace.
5522 	 */
5523 	return !write_fault_to_shadow_pgtable;
5524 }
5525 
5526 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5527 			      unsigned long cr2,  int emulation_type)
5528 {
5529 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5530 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5531 
5532 	last_retry_eip = vcpu->arch.last_retry_eip;
5533 	last_retry_addr = vcpu->arch.last_retry_addr;
5534 
5535 	/*
5536 	 * If the emulation is caused by #PF and it is non-page_table
5537 	 * writing instruction, it means the VM-EXIT is caused by shadow
5538 	 * page protected, we can zap the shadow page and retry this
5539 	 * instruction directly.
5540 	 *
5541 	 * Note: if the guest uses a non-page-table modifying instruction
5542 	 * on the PDE that points to the instruction, then we will unmap
5543 	 * the instruction and go to an infinite loop. So, we cache the
5544 	 * last retried eip and the last fault address, if we meet the eip
5545 	 * and the address again, we can break out of the potential infinite
5546 	 * loop.
5547 	 */
5548 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5549 
5550 	if (!(emulation_type & EMULTYPE_RETRY))
5551 		return false;
5552 
5553 	if (x86_page_table_writing_insn(ctxt))
5554 		return false;
5555 
5556 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5557 		return false;
5558 
5559 	vcpu->arch.last_retry_eip = ctxt->eip;
5560 	vcpu->arch.last_retry_addr = cr2;
5561 
5562 	if (!vcpu->arch.mmu.direct_map)
5563 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5564 
5565 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5566 
5567 	return true;
5568 }
5569 
5570 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5571 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5572 
5573 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5574 {
5575 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5576 		/* This is a good place to trace that we are exiting SMM.  */
5577 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5578 
5579 		/* Process a latched INIT or SMI, if any.  */
5580 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5581 	}
5582 
5583 	kvm_mmu_reset_context(vcpu);
5584 }
5585 
5586 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5587 {
5588 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5589 
5590 	vcpu->arch.hflags = emul_flags;
5591 
5592 	if (changed & HF_SMM_MASK)
5593 		kvm_smm_changed(vcpu);
5594 }
5595 
5596 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5597 				unsigned long *db)
5598 {
5599 	u32 dr6 = 0;
5600 	int i;
5601 	u32 enable, rwlen;
5602 
5603 	enable = dr7;
5604 	rwlen = dr7 >> 16;
5605 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5606 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5607 			dr6 |= (1 << i);
5608 	return dr6;
5609 }
5610 
5611 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5612 {
5613 	struct kvm_run *kvm_run = vcpu->run;
5614 
5615 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5616 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5617 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5618 		kvm_run->debug.arch.exception = DB_VECTOR;
5619 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5620 		*r = EMULATE_USER_EXIT;
5621 	} else {
5622 		/*
5623 		 * "Certain debug exceptions may clear bit 0-3.  The
5624 		 * remaining contents of the DR6 register are never
5625 		 * cleared by the processor".
5626 		 */
5627 		vcpu->arch.dr6 &= ~15;
5628 		vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5629 		kvm_queue_exception(vcpu, DB_VECTOR);
5630 	}
5631 }
5632 
5633 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5634 {
5635 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5636 	int r = EMULATE_DONE;
5637 
5638 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5639 
5640 	/*
5641 	 * rflags is the old, "raw" value of the flags.  The new value has
5642 	 * not been saved yet.
5643 	 *
5644 	 * This is correct even for TF set by the guest, because "the
5645 	 * processor will not generate this exception after the instruction
5646 	 * that sets the TF flag".
5647 	 */
5648 	if (unlikely(rflags & X86_EFLAGS_TF))
5649 		kvm_vcpu_do_singlestep(vcpu, &r);
5650 	return r == EMULATE_DONE;
5651 }
5652 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5653 
5654 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5655 {
5656 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5657 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5658 		struct kvm_run *kvm_run = vcpu->run;
5659 		unsigned long eip = kvm_get_linear_rip(vcpu);
5660 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5661 					   vcpu->arch.guest_debug_dr7,
5662 					   vcpu->arch.eff_db);
5663 
5664 		if (dr6 != 0) {
5665 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5666 			kvm_run->debug.arch.pc = eip;
5667 			kvm_run->debug.arch.exception = DB_VECTOR;
5668 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5669 			*r = EMULATE_USER_EXIT;
5670 			return true;
5671 		}
5672 	}
5673 
5674 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5675 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5676 		unsigned long eip = kvm_get_linear_rip(vcpu);
5677 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5678 					   vcpu->arch.dr7,
5679 					   vcpu->arch.db);
5680 
5681 		if (dr6 != 0) {
5682 			vcpu->arch.dr6 &= ~15;
5683 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5684 			kvm_queue_exception(vcpu, DB_VECTOR);
5685 			*r = EMULATE_DONE;
5686 			return true;
5687 		}
5688 	}
5689 
5690 	return false;
5691 }
5692 
5693 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5694 			    unsigned long cr2,
5695 			    int emulation_type,
5696 			    void *insn,
5697 			    int insn_len)
5698 {
5699 	int r;
5700 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5701 	bool writeback = true;
5702 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5703 
5704 	/*
5705 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5706 	 * never reused.
5707 	 */
5708 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5709 	kvm_clear_exception_queue(vcpu);
5710 
5711 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5712 		init_emulate_ctxt(vcpu);
5713 
5714 		/*
5715 		 * We will reenter on the same instruction since
5716 		 * we do not set complete_userspace_io.  This does not
5717 		 * handle watchpoints yet, those would be handled in
5718 		 * the emulate_ops.
5719 		 */
5720 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5721 			return r;
5722 
5723 		ctxt->interruptibility = 0;
5724 		ctxt->have_exception = false;
5725 		ctxt->exception.vector = -1;
5726 		ctxt->perm_ok = false;
5727 
5728 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5729 
5730 		r = x86_decode_insn(ctxt, insn, insn_len);
5731 
5732 		trace_kvm_emulate_insn_start(vcpu);
5733 		++vcpu->stat.insn_emulation;
5734 		if (r != EMULATION_OK)  {
5735 			if (emulation_type & EMULTYPE_TRAP_UD)
5736 				return EMULATE_FAIL;
5737 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5738 						emulation_type))
5739 				return EMULATE_DONE;
5740 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
5741 				return EMULATE_DONE;
5742 			if (emulation_type & EMULTYPE_SKIP)
5743 				return EMULATE_FAIL;
5744 			return handle_emulation_failure(vcpu);
5745 		}
5746 	}
5747 
5748 	if (emulation_type & EMULTYPE_SKIP) {
5749 		kvm_rip_write(vcpu, ctxt->_eip);
5750 		if (ctxt->eflags & X86_EFLAGS_RF)
5751 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5752 		return EMULATE_DONE;
5753 	}
5754 
5755 	if (retry_instruction(ctxt, cr2, emulation_type))
5756 		return EMULATE_DONE;
5757 
5758 	/* this is needed for vmware backdoor interface to work since it
5759 	   changes registers values  during IO operation */
5760 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5761 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5762 		emulator_invalidate_register_cache(ctxt);
5763 	}
5764 
5765 restart:
5766 	/* Save the faulting GPA (cr2) in the address field */
5767 	ctxt->exception.address = cr2;
5768 
5769 	r = x86_emulate_insn(ctxt);
5770 
5771 	if (r == EMULATION_INTERCEPTED)
5772 		return EMULATE_DONE;
5773 
5774 	if (r == EMULATION_FAILED) {
5775 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5776 					emulation_type))
5777 			return EMULATE_DONE;
5778 
5779 		return handle_emulation_failure(vcpu);
5780 	}
5781 
5782 	if (ctxt->have_exception) {
5783 		r = EMULATE_DONE;
5784 		if (inject_emulated_exception(vcpu))
5785 			return r;
5786 	} else if (vcpu->arch.pio.count) {
5787 		if (!vcpu->arch.pio.in) {
5788 			/* FIXME: return into emulator if single-stepping.  */
5789 			vcpu->arch.pio.count = 0;
5790 		} else {
5791 			writeback = false;
5792 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5793 		}
5794 		r = EMULATE_USER_EXIT;
5795 	} else if (vcpu->mmio_needed) {
5796 		if (!vcpu->mmio_is_write)
5797 			writeback = false;
5798 		r = EMULATE_USER_EXIT;
5799 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5800 	} else if (r == EMULATION_RESTART)
5801 		goto restart;
5802 	else
5803 		r = EMULATE_DONE;
5804 
5805 	if (writeback) {
5806 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5807 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5808 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5809 		kvm_rip_write(vcpu, ctxt->eip);
5810 		if (r == EMULATE_DONE &&
5811 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5812 			kvm_vcpu_do_singlestep(vcpu, &r);
5813 		if (!ctxt->have_exception ||
5814 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5815 			__kvm_set_rflags(vcpu, ctxt->eflags);
5816 
5817 		/*
5818 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5819 		 * do nothing, and it will be requested again as soon as
5820 		 * the shadow expires.  But we still need to check here,
5821 		 * because POPF has no interrupt shadow.
5822 		 */
5823 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5824 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5825 	} else
5826 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5827 
5828 	return r;
5829 }
5830 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5831 
5832 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5833 {
5834 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5835 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5836 					    size, port, &val, 1);
5837 	/* do not return to emulator after return from userspace */
5838 	vcpu->arch.pio.count = 0;
5839 	return ret;
5840 }
5841 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5842 
5843 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5844 {
5845 	unsigned long val;
5846 
5847 	/* We should only ever be called with arch.pio.count equal to 1 */
5848 	BUG_ON(vcpu->arch.pio.count != 1);
5849 
5850 	/* For size less than 4 we merge, else we zero extend */
5851 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5852 					: 0;
5853 
5854 	/*
5855 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5856 	 * the copy and tracing
5857 	 */
5858 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5859 				 vcpu->arch.pio.port, &val, 1);
5860 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5861 
5862 	return 1;
5863 }
5864 
5865 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5866 {
5867 	unsigned long val;
5868 	int ret;
5869 
5870 	/* For size less than 4 we merge, else we zero extend */
5871 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5872 
5873 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5874 				       &val, 1);
5875 	if (ret) {
5876 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5877 		return ret;
5878 	}
5879 
5880 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5881 
5882 	return 0;
5883 }
5884 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5885 
5886 static int kvmclock_cpu_down_prep(unsigned int cpu)
5887 {
5888 	__this_cpu_write(cpu_tsc_khz, 0);
5889 	return 0;
5890 }
5891 
5892 static void tsc_khz_changed(void *data)
5893 {
5894 	struct cpufreq_freqs *freq = data;
5895 	unsigned long khz = 0;
5896 
5897 	if (data)
5898 		khz = freq->new;
5899 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5900 		khz = cpufreq_quick_get(raw_smp_processor_id());
5901 	if (!khz)
5902 		khz = tsc_khz;
5903 	__this_cpu_write(cpu_tsc_khz, khz);
5904 }
5905 
5906 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5907 				     void *data)
5908 {
5909 	struct cpufreq_freqs *freq = data;
5910 	struct kvm *kvm;
5911 	struct kvm_vcpu *vcpu;
5912 	int i, send_ipi = 0;
5913 
5914 	/*
5915 	 * We allow guests to temporarily run on slowing clocks,
5916 	 * provided we notify them after, or to run on accelerating
5917 	 * clocks, provided we notify them before.  Thus time never
5918 	 * goes backwards.
5919 	 *
5920 	 * However, we have a problem.  We can't atomically update
5921 	 * the frequency of a given CPU from this function; it is
5922 	 * merely a notifier, which can be called from any CPU.
5923 	 * Changing the TSC frequency at arbitrary points in time
5924 	 * requires a recomputation of local variables related to
5925 	 * the TSC for each VCPU.  We must flag these local variables
5926 	 * to be updated and be sure the update takes place with the
5927 	 * new frequency before any guests proceed.
5928 	 *
5929 	 * Unfortunately, the combination of hotplug CPU and frequency
5930 	 * change creates an intractable locking scenario; the order
5931 	 * of when these callouts happen is undefined with respect to
5932 	 * CPU hotplug, and they can race with each other.  As such,
5933 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5934 	 * undefined; you can actually have a CPU frequency change take
5935 	 * place in between the computation of X and the setting of the
5936 	 * variable.  To protect against this problem, all updates of
5937 	 * the per_cpu tsc_khz variable are done in an interrupt
5938 	 * protected IPI, and all callers wishing to update the value
5939 	 * must wait for a synchronous IPI to complete (which is trivial
5940 	 * if the caller is on the CPU already).  This establishes the
5941 	 * necessary total order on variable updates.
5942 	 *
5943 	 * Note that because a guest time update may take place
5944 	 * anytime after the setting of the VCPU's request bit, the
5945 	 * correct TSC value must be set before the request.  However,
5946 	 * to ensure the update actually makes it to any guest which
5947 	 * starts running in hardware virtualization between the set
5948 	 * and the acquisition of the spinlock, we must also ping the
5949 	 * CPU after setting the request bit.
5950 	 *
5951 	 */
5952 
5953 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5954 		return 0;
5955 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5956 		return 0;
5957 
5958 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5959 
5960 	spin_lock(&kvm_lock);
5961 	list_for_each_entry(kvm, &vm_list, vm_list) {
5962 		kvm_for_each_vcpu(i, vcpu, kvm) {
5963 			if (vcpu->cpu != freq->cpu)
5964 				continue;
5965 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5966 			if (vcpu->cpu != smp_processor_id())
5967 				send_ipi = 1;
5968 		}
5969 	}
5970 	spin_unlock(&kvm_lock);
5971 
5972 	if (freq->old < freq->new && send_ipi) {
5973 		/*
5974 		 * We upscale the frequency.  Must make the guest
5975 		 * doesn't see old kvmclock values while running with
5976 		 * the new frequency, otherwise we risk the guest sees
5977 		 * time go backwards.
5978 		 *
5979 		 * In case we update the frequency for another cpu
5980 		 * (which might be in guest context) send an interrupt
5981 		 * to kick the cpu out of guest context.  Next time
5982 		 * guest context is entered kvmclock will be updated,
5983 		 * so the guest will not see stale values.
5984 		 */
5985 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5986 	}
5987 	return 0;
5988 }
5989 
5990 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5991 	.notifier_call  = kvmclock_cpufreq_notifier
5992 };
5993 
5994 static int kvmclock_cpu_online(unsigned int cpu)
5995 {
5996 	tsc_khz_changed(NULL);
5997 	return 0;
5998 }
5999 
6000 static void kvm_timer_init(void)
6001 {
6002 	max_tsc_khz = tsc_khz;
6003 
6004 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6005 #ifdef CONFIG_CPU_FREQ
6006 		struct cpufreq_policy policy;
6007 		int cpu;
6008 
6009 		memset(&policy, 0, sizeof(policy));
6010 		cpu = get_cpu();
6011 		cpufreq_get_policy(&policy, cpu);
6012 		if (policy.cpuinfo.max_freq)
6013 			max_tsc_khz = policy.cpuinfo.max_freq;
6014 		put_cpu();
6015 #endif
6016 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6017 					  CPUFREQ_TRANSITION_NOTIFIER);
6018 	}
6019 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6020 
6021 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6022 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6023 }
6024 
6025 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6026 
6027 int kvm_is_in_guest(void)
6028 {
6029 	return __this_cpu_read(current_vcpu) != NULL;
6030 }
6031 
6032 static int kvm_is_user_mode(void)
6033 {
6034 	int user_mode = 3;
6035 
6036 	if (__this_cpu_read(current_vcpu))
6037 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6038 
6039 	return user_mode != 0;
6040 }
6041 
6042 static unsigned long kvm_get_guest_ip(void)
6043 {
6044 	unsigned long ip = 0;
6045 
6046 	if (__this_cpu_read(current_vcpu))
6047 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6048 
6049 	return ip;
6050 }
6051 
6052 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6053 	.is_in_guest		= kvm_is_in_guest,
6054 	.is_user_mode		= kvm_is_user_mode,
6055 	.get_guest_ip		= kvm_get_guest_ip,
6056 };
6057 
6058 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6059 {
6060 	__this_cpu_write(current_vcpu, vcpu);
6061 }
6062 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6063 
6064 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6065 {
6066 	__this_cpu_write(current_vcpu, NULL);
6067 }
6068 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6069 
6070 static void kvm_set_mmio_spte_mask(void)
6071 {
6072 	u64 mask;
6073 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6074 
6075 	/*
6076 	 * Set the reserved bits and the present bit of an paging-structure
6077 	 * entry to generate page fault with PFER.RSV = 1.
6078 	 */
6079 	 /* Mask the reserved physical address bits. */
6080 	mask = rsvd_bits(maxphyaddr, 51);
6081 
6082 	/* Set the present bit. */
6083 	mask |= 1ull;
6084 
6085 #ifdef CONFIG_X86_64
6086 	/*
6087 	 * If reserved bit is not supported, clear the present bit to disable
6088 	 * mmio page fault.
6089 	 */
6090 	if (maxphyaddr == 52)
6091 		mask &= ~1ull;
6092 #endif
6093 
6094 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6095 }
6096 
6097 #ifdef CONFIG_X86_64
6098 static void pvclock_gtod_update_fn(struct work_struct *work)
6099 {
6100 	struct kvm *kvm;
6101 
6102 	struct kvm_vcpu *vcpu;
6103 	int i;
6104 
6105 	spin_lock(&kvm_lock);
6106 	list_for_each_entry(kvm, &vm_list, vm_list)
6107 		kvm_for_each_vcpu(i, vcpu, kvm)
6108 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6109 	atomic_set(&kvm_guest_has_master_clock, 0);
6110 	spin_unlock(&kvm_lock);
6111 }
6112 
6113 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6114 
6115 /*
6116  * Notification about pvclock gtod data update.
6117  */
6118 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6119 			       void *priv)
6120 {
6121 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6122 	struct timekeeper *tk = priv;
6123 
6124 	update_pvclock_gtod(tk);
6125 
6126 	/* disable master clock if host does not trust, or does not
6127 	 * use, TSC clocksource
6128 	 */
6129 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6130 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6131 		queue_work(system_long_wq, &pvclock_gtod_work);
6132 
6133 	return 0;
6134 }
6135 
6136 static struct notifier_block pvclock_gtod_notifier = {
6137 	.notifier_call = pvclock_gtod_notify,
6138 };
6139 #endif
6140 
6141 int kvm_arch_init(void *opaque)
6142 {
6143 	int r;
6144 	struct kvm_x86_ops *ops = opaque;
6145 
6146 	if (kvm_x86_ops) {
6147 		printk(KERN_ERR "kvm: already loaded the other module\n");
6148 		r = -EEXIST;
6149 		goto out;
6150 	}
6151 
6152 	if (!ops->cpu_has_kvm_support()) {
6153 		printk(KERN_ERR "kvm: no hardware support\n");
6154 		r = -EOPNOTSUPP;
6155 		goto out;
6156 	}
6157 	if (ops->disabled_by_bios()) {
6158 		printk(KERN_ERR "kvm: disabled by bios\n");
6159 		r = -EOPNOTSUPP;
6160 		goto out;
6161 	}
6162 
6163 	r = -ENOMEM;
6164 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6165 	if (!shared_msrs) {
6166 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6167 		goto out;
6168 	}
6169 
6170 	r = kvm_mmu_module_init();
6171 	if (r)
6172 		goto out_free_percpu;
6173 
6174 	kvm_set_mmio_spte_mask();
6175 
6176 	kvm_x86_ops = ops;
6177 
6178 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6179 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6180 			PT_PRESENT_MASK, 0, sme_me_mask);
6181 	kvm_timer_init();
6182 
6183 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6184 
6185 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6186 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6187 
6188 	kvm_lapic_init();
6189 #ifdef CONFIG_X86_64
6190 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6191 #endif
6192 
6193 	return 0;
6194 
6195 out_free_percpu:
6196 	free_percpu(shared_msrs);
6197 out:
6198 	return r;
6199 }
6200 
6201 void kvm_arch_exit(void)
6202 {
6203 	kvm_lapic_exit();
6204 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6205 
6206 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6207 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6208 					    CPUFREQ_TRANSITION_NOTIFIER);
6209 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6210 #ifdef CONFIG_X86_64
6211 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6212 #endif
6213 	kvm_x86_ops = NULL;
6214 	kvm_mmu_module_exit();
6215 	free_percpu(shared_msrs);
6216 }
6217 
6218 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6219 {
6220 	++vcpu->stat.halt_exits;
6221 	if (lapic_in_kernel(vcpu)) {
6222 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6223 		return 1;
6224 	} else {
6225 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6226 		return 0;
6227 	}
6228 }
6229 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6230 
6231 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6232 {
6233 	int ret = kvm_skip_emulated_instruction(vcpu);
6234 	/*
6235 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6236 	 * KVM_EXIT_DEBUG here.
6237 	 */
6238 	return kvm_vcpu_halt(vcpu) && ret;
6239 }
6240 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6241 
6242 #ifdef CONFIG_X86_64
6243 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6244 			        unsigned long clock_type)
6245 {
6246 	struct kvm_clock_pairing clock_pairing;
6247 	struct timespec ts;
6248 	u64 cycle;
6249 	int ret;
6250 
6251 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6252 		return -KVM_EOPNOTSUPP;
6253 
6254 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6255 		return -KVM_EOPNOTSUPP;
6256 
6257 	clock_pairing.sec = ts.tv_sec;
6258 	clock_pairing.nsec = ts.tv_nsec;
6259 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6260 	clock_pairing.flags = 0;
6261 
6262 	ret = 0;
6263 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6264 			    sizeof(struct kvm_clock_pairing)))
6265 		ret = -KVM_EFAULT;
6266 
6267 	return ret;
6268 }
6269 #endif
6270 
6271 /*
6272  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6273  *
6274  * @apicid - apicid of vcpu to be kicked.
6275  */
6276 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6277 {
6278 	struct kvm_lapic_irq lapic_irq;
6279 
6280 	lapic_irq.shorthand = 0;
6281 	lapic_irq.dest_mode = 0;
6282 	lapic_irq.level = 0;
6283 	lapic_irq.dest_id = apicid;
6284 	lapic_irq.msi_redir_hint = false;
6285 
6286 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6287 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6288 }
6289 
6290 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6291 {
6292 	vcpu->arch.apicv_active = false;
6293 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6294 }
6295 
6296 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6297 {
6298 	unsigned long nr, a0, a1, a2, a3, ret;
6299 	int op_64_bit, r;
6300 
6301 	r = kvm_skip_emulated_instruction(vcpu);
6302 
6303 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6304 		return kvm_hv_hypercall(vcpu);
6305 
6306 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6307 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6308 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6309 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6310 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6311 
6312 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6313 
6314 	op_64_bit = is_64_bit_mode(vcpu);
6315 	if (!op_64_bit) {
6316 		nr &= 0xFFFFFFFF;
6317 		a0 &= 0xFFFFFFFF;
6318 		a1 &= 0xFFFFFFFF;
6319 		a2 &= 0xFFFFFFFF;
6320 		a3 &= 0xFFFFFFFF;
6321 	}
6322 
6323 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6324 		ret = -KVM_EPERM;
6325 		goto out;
6326 	}
6327 
6328 	switch (nr) {
6329 	case KVM_HC_VAPIC_POLL_IRQ:
6330 		ret = 0;
6331 		break;
6332 	case KVM_HC_KICK_CPU:
6333 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6334 		ret = 0;
6335 		break;
6336 #ifdef CONFIG_X86_64
6337 	case KVM_HC_CLOCK_PAIRING:
6338 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6339 		break;
6340 #endif
6341 	default:
6342 		ret = -KVM_ENOSYS;
6343 		break;
6344 	}
6345 out:
6346 	if (!op_64_bit)
6347 		ret = (u32)ret;
6348 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6349 	++vcpu->stat.hypercalls;
6350 	return r;
6351 }
6352 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6353 
6354 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6355 {
6356 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6357 	char instruction[3];
6358 	unsigned long rip = kvm_rip_read(vcpu);
6359 
6360 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6361 
6362 	return emulator_write_emulated(ctxt, rip, instruction, 3,
6363 		&ctxt->exception);
6364 }
6365 
6366 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6367 {
6368 	return vcpu->run->request_interrupt_window &&
6369 		likely(!pic_in_kernel(vcpu->kvm));
6370 }
6371 
6372 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6373 {
6374 	struct kvm_run *kvm_run = vcpu->run;
6375 
6376 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6377 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6378 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6379 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6380 	kvm_run->ready_for_interrupt_injection =
6381 		pic_in_kernel(vcpu->kvm) ||
6382 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6383 }
6384 
6385 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6386 {
6387 	int max_irr, tpr;
6388 
6389 	if (!kvm_x86_ops->update_cr8_intercept)
6390 		return;
6391 
6392 	if (!lapic_in_kernel(vcpu))
6393 		return;
6394 
6395 	if (vcpu->arch.apicv_active)
6396 		return;
6397 
6398 	if (!vcpu->arch.apic->vapic_addr)
6399 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6400 	else
6401 		max_irr = -1;
6402 
6403 	if (max_irr != -1)
6404 		max_irr >>= 4;
6405 
6406 	tpr = kvm_lapic_get_cr8(vcpu);
6407 
6408 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6409 }
6410 
6411 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6412 {
6413 	int r;
6414 
6415 	/* try to reinject previous events if any */
6416 	if (vcpu->arch.exception.injected) {
6417 		kvm_x86_ops->queue_exception(vcpu);
6418 		return 0;
6419 	}
6420 
6421 	/*
6422 	 * Exceptions must be injected immediately, or the exception
6423 	 * frame will have the address of the NMI or interrupt handler.
6424 	 */
6425 	if (!vcpu->arch.exception.pending) {
6426 		if (vcpu->arch.nmi_injected) {
6427 			kvm_x86_ops->set_nmi(vcpu);
6428 			return 0;
6429 		}
6430 
6431 		if (vcpu->arch.interrupt.pending) {
6432 			kvm_x86_ops->set_irq(vcpu);
6433 			return 0;
6434 		}
6435 	}
6436 
6437 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6438 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6439 		if (r != 0)
6440 			return r;
6441 	}
6442 
6443 	/* try to inject new event if pending */
6444 	if (vcpu->arch.exception.pending) {
6445 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6446 					vcpu->arch.exception.has_error_code,
6447 					vcpu->arch.exception.error_code);
6448 
6449 		vcpu->arch.exception.pending = false;
6450 		vcpu->arch.exception.injected = true;
6451 
6452 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6453 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6454 					     X86_EFLAGS_RF);
6455 
6456 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6457 		    (vcpu->arch.dr7 & DR7_GD)) {
6458 			vcpu->arch.dr7 &= ~DR7_GD;
6459 			kvm_update_dr7(vcpu);
6460 		}
6461 
6462 		kvm_x86_ops->queue_exception(vcpu);
6463 	} else if (vcpu->arch.smi_pending && !is_smm(vcpu) && kvm_x86_ops->smi_allowed(vcpu)) {
6464 		vcpu->arch.smi_pending = false;
6465 		enter_smm(vcpu);
6466 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6467 		--vcpu->arch.nmi_pending;
6468 		vcpu->arch.nmi_injected = true;
6469 		kvm_x86_ops->set_nmi(vcpu);
6470 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6471 		/*
6472 		 * Because interrupts can be injected asynchronously, we are
6473 		 * calling check_nested_events again here to avoid a race condition.
6474 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6475 		 * proposal and current concerns.  Perhaps we should be setting
6476 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6477 		 */
6478 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6479 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6480 			if (r != 0)
6481 				return r;
6482 		}
6483 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6484 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6485 					    false);
6486 			kvm_x86_ops->set_irq(vcpu);
6487 		}
6488 	}
6489 
6490 	return 0;
6491 }
6492 
6493 static void process_nmi(struct kvm_vcpu *vcpu)
6494 {
6495 	unsigned limit = 2;
6496 
6497 	/*
6498 	 * x86 is limited to one NMI running, and one NMI pending after it.
6499 	 * If an NMI is already in progress, limit further NMIs to just one.
6500 	 * Otherwise, allow two (and we'll inject the first one immediately).
6501 	 */
6502 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6503 		limit = 1;
6504 
6505 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6506 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6507 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6508 }
6509 
6510 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6511 {
6512 	u32 flags = 0;
6513 	flags |= seg->g       << 23;
6514 	flags |= seg->db      << 22;
6515 	flags |= seg->l       << 21;
6516 	flags |= seg->avl     << 20;
6517 	flags |= seg->present << 15;
6518 	flags |= seg->dpl     << 13;
6519 	flags |= seg->s       << 12;
6520 	flags |= seg->type    << 8;
6521 	return flags;
6522 }
6523 
6524 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6525 {
6526 	struct kvm_segment seg;
6527 	int offset;
6528 
6529 	kvm_get_segment(vcpu, &seg, n);
6530 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6531 
6532 	if (n < 3)
6533 		offset = 0x7f84 + n * 12;
6534 	else
6535 		offset = 0x7f2c + (n - 3) * 12;
6536 
6537 	put_smstate(u32, buf, offset + 8, seg.base);
6538 	put_smstate(u32, buf, offset + 4, seg.limit);
6539 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6540 }
6541 
6542 #ifdef CONFIG_X86_64
6543 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6544 {
6545 	struct kvm_segment seg;
6546 	int offset;
6547 	u16 flags;
6548 
6549 	kvm_get_segment(vcpu, &seg, n);
6550 	offset = 0x7e00 + n * 16;
6551 
6552 	flags = enter_smm_get_segment_flags(&seg) >> 8;
6553 	put_smstate(u16, buf, offset, seg.selector);
6554 	put_smstate(u16, buf, offset + 2, flags);
6555 	put_smstate(u32, buf, offset + 4, seg.limit);
6556 	put_smstate(u64, buf, offset + 8, seg.base);
6557 }
6558 #endif
6559 
6560 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6561 {
6562 	struct desc_ptr dt;
6563 	struct kvm_segment seg;
6564 	unsigned long val;
6565 	int i;
6566 
6567 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6568 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6569 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6570 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6571 
6572 	for (i = 0; i < 8; i++)
6573 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6574 
6575 	kvm_get_dr(vcpu, 6, &val);
6576 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6577 	kvm_get_dr(vcpu, 7, &val);
6578 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6579 
6580 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6581 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6582 	put_smstate(u32, buf, 0x7f64, seg.base);
6583 	put_smstate(u32, buf, 0x7f60, seg.limit);
6584 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6585 
6586 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6587 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6588 	put_smstate(u32, buf, 0x7f80, seg.base);
6589 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6590 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6591 
6592 	kvm_x86_ops->get_gdt(vcpu, &dt);
6593 	put_smstate(u32, buf, 0x7f74, dt.address);
6594 	put_smstate(u32, buf, 0x7f70, dt.size);
6595 
6596 	kvm_x86_ops->get_idt(vcpu, &dt);
6597 	put_smstate(u32, buf, 0x7f58, dt.address);
6598 	put_smstate(u32, buf, 0x7f54, dt.size);
6599 
6600 	for (i = 0; i < 6; i++)
6601 		enter_smm_save_seg_32(vcpu, buf, i);
6602 
6603 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6604 
6605 	/* revision id */
6606 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6607 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6608 }
6609 
6610 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6611 {
6612 #ifdef CONFIG_X86_64
6613 	struct desc_ptr dt;
6614 	struct kvm_segment seg;
6615 	unsigned long val;
6616 	int i;
6617 
6618 	for (i = 0; i < 16; i++)
6619 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6620 
6621 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6622 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6623 
6624 	kvm_get_dr(vcpu, 6, &val);
6625 	put_smstate(u64, buf, 0x7f68, val);
6626 	kvm_get_dr(vcpu, 7, &val);
6627 	put_smstate(u64, buf, 0x7f60, val);
6628 
6629 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6630 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6631 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6632 
6633 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6634 
6635 	/* revision id */
6636 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6637 
6638 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6639 
6640 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6641 	put_smstate(u16, buf, 0x7e90, seg.selector);
6642 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6643 	put_smstate(u32, buf, 0x7e94, seg.limit);
6644 	put_smstate(u64, buf, 0x7e98, seg.base);
6645 
6646 	kvm_x86_ops->get_idt(vcpu, &dt);
6647 	put_smstate(u32, buf, 0x7e84, dt.size);
6648 	put_smstate(u64, buf, 0x7e88, dt.address);
6649 
6650 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6651 	put_smstate(u16, buf, 0x7e70, seg.selector);
6652 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6653 	put_smstate(u32, buf, 0x7e74, seg.limit);
6654 	put_smstate(u64, buf, 0x7e78, seg.base);
6655 
6656 	kvm_x86_ops->get_gdt(vcpu, &dt);
6657 	put_smstate(u32, buf, 0x7e64, dt.size);
6658 	put_smstate(u64, buf, 0x7e68, dt.address);
6659 
6660 	for (i = 0; i < 6; i++)
6661 		enter_smm_save_seg_64(vcpu, buf, i);
6662 #else
6663 	WARN_ON_ONCE(1);
6664 #endif
6665 }
6666 
6667 static void enter_smm(struct kvm_vcpu *vcpu)
6668 {
6669 	struct kvm_segment cs, ds;
6670 	struct desc_ptr dt;
6671 	char buf[512];
6672 	u32 cr0;
6673 
6674 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6675 	memset(buf, 0, 512);
6676 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6677 		enter_smm_save_state_64(vcpu, buf);
6678 	else
6679 		enter_smm_save_state_32(vcpu, buf);
6680 
6681 	/*
6682 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
6683 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
6684 	 * the SMM state-save area.
6685 	 */
6686 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
6687 
6688 	vcpu->arch.hflags |= HF_SMM_MASK;
6689 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6690 
6691 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6692 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6693 	else
6694 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6695 
6696 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6697 	kvm_rip_write(vcpu, 0x8000);
6698 
6699 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6700 	kvm_x86_ops->set_cr0(vcpu, cr0);
6701 	vcpu->arch.cr0 = cr0;
6702 
6703 	kvm_x86_ops->set_cr4(vcpu, 0);
6704 
6705 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6706 	dt.address = dt.size = 0;
6707 	kvm_x86_ops->set_idt(vcpu, &dt);
6708 
6709 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6710 
6711 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6712 	cs.base = vcpu->arch.smbase;
6713 
6714 	ds.selector = 0;
6715 	ds.base = 0;
6716 
6717 	cs.limit    = ds.limit = 0xffffffff;
6718 	cs.type     = ds.type = 0x3;
6719 	cs.dpl      = ds.dpl = 0;
6720 	cs.db       = ds.db = 0;
6721 	cs.s        = ds.s = 1;
6722 	cs.l        = ds.l = 0;
6723 	cs.g        = ds.g = 1;
6724 	cs.avl      = ds.avl = 0;
6725 	cs.present  = ds.present = 1;
6726 	cs.unusable = ds.unusable = 0;
6727 	cs.padding  = ds.padding = 0;
6728 
6729 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6730 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6731 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6732 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6733 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6734 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6735 
6736 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6737 		kvm_x86_ops->set_efer(vcpu, 0);
6738 
6739 	kvm_update_cpuid(vcpu);
6740 	kvm_mmu_reset_context(vcpu);
6741 }
6742 
6743 static void process_smi(struct kvm_vcpu *vcpu)
6744 {
6745 	vcpu->arch.smi_pending = true;
6746 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6747 }
6748 
6749 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6750 {
6751 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6752 }
6753 
6754 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6755 {
6756 	u64 eoi_exit_bitmap[4];
6757 
6758 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6759 		return;
6760 
6761 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6762 
6763 	if (irqchip_split(vcpu->kvm))
6764 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6765 	else {
6766 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6767 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6768 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6769 	}
6770 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6771 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
6772 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6773 }
6774 
6775 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6776 {
6777 	++vcpu->stat.tlb_flush;
6778 	kvm_x86_ops->tlb_flush(vcpu);
6779 }
6780 
6781 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6782 {
6783 	struct page *page = NULL;
6784 
6785 	if (!lapic_in_kernel(vcpu))
6786 		return;
6787 
6788 	if (!kvm_x86_ops->set_apic_access_page_addr)
6789 		return;
6790 
6791 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6792 	if (is_error_page(page))
6793 		return;
6794 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6795 
6796 	/*
6797 	 * Do not pin apic access page in memory, the MMU notifier
6798 	 * will call us again if it is migrated or swapped out.
6799 	 */
6800 	put_page(page);
6801 }
6802 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6803 
6804 /*
6805  * Returns 1 to let vcpu_run() continue the guest execution loop without
6806  * exiting to the userspace.  Otherwise, the value will be returned to the
6807  * userspace.
6808  */
6809 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6810 {
6811 	int r;
6812 	bool req_int_win =
6813 		dm_request_for_irq_injection(vcpu) &&
6814 		kvm_cpu_accept_dm_intr(vcpu);
6815 
6816 	bool req_immediate_exit = false;
6817 
6818 	if (kvm_request_pending(vcpu)) {
6819 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6820 			kvm_mmu_unload(vcpu);
6821 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6822 			__kvm_migrate_timers(vcpu);
6823 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6824 			kvm_gen_update_masterclock(vcpu->kvm);
6825 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6826 			kvm_gen_kvmclock_update(vcpu);
6827 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6828 			r = kvm_guest_time_update(vcpu);
6829 			if (unlikely(r))
6830 				goto out;
6831 		}
6832 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6833 			kvm_mmu_sync_roots(vcpu);
6834 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6835 			kvm_vcpu_flush_tlb(vcpu);
6836 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6837 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6838 			r = 0;
6839 			goto out;
6840 		}
6841 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6842 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6843 			vcpu->mmio_needed = 0;
6844 			r = 0;
6845 			goto out;
6846 		}
6847 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6848 			/* Page is swapped out. Do synthetic halt */
6849 			vcpu->arch.apf.halted = true;
6850 			r = 1;
6851 			goto out;
6852 		}
6853 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6854 			record_steal_time(vcpu);
6855 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6856 			process_smi(vcpu);
6857 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6858 			process_nmi(vcpu);
6859 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6860 			kvm_pmu_handle_event(vcpu);
6861 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6862 			kvm_pmu_deliver_pmi(vcpu);
6863 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6864 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6865 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
6866 				     vcpu->arch.ioapic_handled_vectors)) {
6867 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6868 				vcpu->run->eoi.vector =
6869 						vcpu->arch.pending_ioapic_eoi;
6870 				r = 0;
6871 				goto out;
6872 			}
6873 		}
6874 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6875 			vcpu_scan_ioapic(vcpu);
6876 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6877 			kvm_vcpu_reload_apic_access_page(vcpu);
6878 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6879 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6880 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6881 			r = 0;
6882 			goto out;
6883 		}
6884 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6885 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6886 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6887 			r = 0;
6888 			goto out;
6889 		}
6890 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6891 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6892 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6893 			r = 0;
6894 			goto out;
6895 		}
6896 
6897 		/*
6898 		 * KVM_REQ_HV_STIMER has to be processed after
6899 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6900 		 * depend on the guest clock being up-to-date
6901 		 */
6902 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6903 			kvm_hv_process_stimers(vcpu);
6904 	}
6905 
6906 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6907 		++vcpu->stat.req_event;
6908 		kvm_apic_accept_events(vcpu);
6909 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6910 			r = 1;
6911 			goto out;
6912 		}
6913 
6914 		if (inject_pending_event(vcpu, req_int_win) != 0)
6915 			req_immediate_exit = true;
6916 		else {
6917 			/* Enable SMI/NMI/IRQ window open exits if needed.
6918 			 *
6919 			 * SMIs have three cases:
6920 			 * 1) They can be nested, and then there is nothing to
6921 			 *    do here because RSM will cause a vmexit anyway.
6922 			 * 2) There is an ISA-specific reason why SMI cannot be
6923 			 *    injected, and the moment when this changes can be
6924 			 *    intercepted.
6925 			 * 3) Or the SMI can be pending because
6926 			 *    inject_pending_event has completed the injection
6927 			 *    of an IRQ or NMI from the previous vmexit, and
6928 			 *    then we request an immediate exit to inject the
6929 			 *    SMI.
6930 			 */
6931 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
6932 				if (!kvm_x86_ops->enable_smi_window(vcpu))
6933 					req_immediate_exit = true;
6934 			if (vcpu->arch.nmi_pending)
6935 				kvm_x86_ops->enable_nmi_window(vcpu);
6936 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6937 				kvm_x86_ops->enable_irq_window(vcpu);
6938 			WARN_ON(vcpu->arch.exception.pending);
6939 		}
6940 
6941 		if (kvm_lapic_enabled(vcpu)) {
6942 			update_cr8_intercept(vcpu);
6943 			kvm_lapic_sync_to_vapic(vcpu);
6944 		}
6945 	}
6946 
6947 	r = kvm_mmu_reload(vcpu);
6948 	if (unlikely(r)) {
6949 		goto cancel_injection;
6950 	}
6951 
6952 	preempt_disable();
6953 
6954 	kvm_x86_ops->prepare_guest_switch(vcpu);
6955 	kvm_load_guest_fpu(vcpu);
6956 
6957 	/*
6958 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
6959 	 * IPI are then delayed after guest entry, which ensures that they
6960 	 * result in virtual interrupt delivery.
6961 	 */
6962 	local_irq_disable();
6963 	vcpu->mode = IN_GUEST_MODE;
6964 
6965 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6966 
6967 	/*
6968 	 * 1) We should set ->mode before checking ->requests.  Please see
6969 	 * the comment in kvm_vcpu_exiting_guest_mode().
6970 	 *
6971 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
6972 	 * pairs with the memory barrier implicit in pi_test_and_set_on
6973 	 * (see vmx_deliver_posted_interrupt).
6974 	 *
6975 	 * 3) This also orders the write to mode from any reads to the page
6976 	 * tables done while the VCPU is running.  Please see the comment
6977 	 * in kvm_flush_remote_tlbs.
6978 	 */
6979 	smp_mb__after_srcu_read_unlock();
6980 
6981 	/*
6982 	 * This handles the case where a posted interrupt was
6983 	 * notified with kvm_vcpu_kick.
6984 	 */
6985 	if (kvm_lapic_enabled(vcpu)) {
6986 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6987 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6988 	}
6989 
6990 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
6991 	    || need_resched() || signal_pending(current)) {
6992 		vcpu->mode = OUTSIDE_GUEST_MODE;
6993 		smp_wmb();
6994 		local_irq_enable();
6995 		preempt_enable();
6996 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6997 		r = 1;
6998 		goto cancel_injection;
6999 	}
7000 
7001 	kvm_load_guest_xcr0(vcpu);
7002 
7003 	if (req_immediate_exit) {
7004 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7005 		smp_send_reschedule(vcpu->cpu);
7006 	}
7007 
7008 	trace_kvm_entry(vcpu->vcpu_id);
7009 	wait_lapic_expire(vcpu);
7010 	guest_enter_irqoff();
7011 
7012 	if (unlikely(vcpu->arch.switch_db_regs)) {
7013 		set_debugreg(0, 7);
7014 		set_debugreg(vcpu->arch.eff_db[0], 0);
7015 		set_debugreg(vcpu->arch.eff_db[1], 1);
7016 		set_debugreg(vcpu->arch.eff_db[2], 2);
7017 		set_debugreg(vcpu->arch.eff_db[3], 3);
7018 		set_debugreg(vcpu->arch.dr6, 6);
7019 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7020 	}
7021 
7022 	kvm_x86_ops->run(vcpu);
7023 
7024 	/*
7025 	 * Do this here before restoring debug registers on the host.  And
7026 	 * since we do this before handling the vmexit, a DR access vmexit
7027 	 * can (a) read the correct value of the debug registers, (b) set
7028 	 * KVM_DEBUGREG_WONT_EXIT again.
7029 	 */
7030 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7031 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7032 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7033 		kvm_update_dr0123(vcpu);
7034 		kvm_update_dr6(vcpu);
7035 		kvm_update_dr7(vcpu);
7036 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7037 	}
7038 
7039 	/*
7040 	 * If the guest has used debug registers, at least dr7
7041 	 * will be disabled while returning to the host.
7042 	 * If we don't have active breakpoints in the host, we don't
7043 	 * care about the messed up debug address registers. But if
7044 	 * we have some of them active, restore the old state.
7045 	 */
7046 	if (hw_breakpoint_active())
7047 		hw_breakpoint_restore();
7048 
7049 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7050 
7051 	vcpu->mode = OUTSIDE_GUEST_MODE;
7052 	smp_wmb();
7053 
7054 	kvm_put_guest_xcr0(vcpu);
7055 
7056 	kvm_x86_ops->handle_external_intr(vcpu);
7057 
7058 	++vcpu->stat.exits;
7059 
7060 	guest_exit_irqoff();
7061 
7062 	local_irq_enable();
7063 	preempt_enable();
7064 
7065 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7066 
7067 	/*
7068 	 * Profile KVM exit RIPs:
7069 	 */
7070 	if (unlikely(prof_on == KVM_PROFILING)) {
7071 		unsigned long rip = kvm_rip_read(vcpu);
7072 		profile_hit(KVM_PROFILING, (void *)rip);
7073 	}
7074 
7075 	if (unlikely(vcpu->arch.tsc_always_catchup))
7076 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7077 
7078 	if (vcpu->arch.apic_attention)
7079 		kvm_lapic_sync_from_vapic(vcpu);
7080 
7081 	vcpu->arch.gpa_available = false;
7082 	r = kvm_x86_ops->handle_exit(vcpu);
7083 	return r;
7084 
7085 cancel_injection:
7086 	kvm_x86_ops->cancel_injection(vcpu);
7087 	if (unlikely(vcpu->arch.apic_attention))
7088 		kvm_lapic_sync_from_vapic(vcpu);
7089 out:
7090 	return r;
7091 }
7092 
7093 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7094 {
7095 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7096 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7097 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7098 		kvm_vcpu_block(vcpu);
7099 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7100 
7101 		if (kvm_x86_ops->post_block)
7102 			kvm_x86_ops->post_block(vcpu);
7103 
7104 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7105 			return 1;
7106 	}
7107 
7108 	kvm_apic_accept_events(vcpu);
7109 	switch(vcpu->arch.mp_state) {
7110 	case KVM_MP_STATE_HALTED:
7111 		vcpu->arch.pv.pv_unhalted = false;
7112 		vcpu->arch.mp_state =
7113 			KVM_MP_STATE_RUNNABLE;
7114 	case KVM_MP_STATE_RUNNABLE:
7115 		vcpu->arch.apf.halted = false;
7116 		break;
7117 	case KVM_MP_STATE_INIT_RECEIVED:
7118 		break;
7119 	default:
7120 		return -EINTR;
7121 		break;
7122 	}
7123 	return 1;
7124 }
7125 
7126 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7127 {
7128 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7129 		kvm_x86_ops->check_nested_events(vcpu, false);
7130 
7131 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7132 		!vcpu->arch.apf.halted);
7133 }
7134 
7135 static int vcpu_run(struct kvm_vcpu *vcpu)
7136 {
7137 	int r;
7138 	struct kvm *kvm = vcpu->kvm;
7139 
7140 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7141 
7142 	for (;;) {
7143 		if (kvm_vcpu_running(vcpu)) {
7144 			r = vcpu_enter_guest(vcpu);
7145 		} else {
7146 			r = vcpu_block(kvm, vcpu);
7147 		}
7148 
7149 		if (r <= 0)
7150 			break;
7151 
7152 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7153 		if (kvm_cpu_has_pending_timer(vcpu))
7154 			kvm_inject_pending_timer_irqs(vcpu);
7155 
7156 		if (dm_request_for_irq_injection(vcpu) &&
7157 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7158 			r = 0;
7159 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7160 			++vcpu->stat.request_irq_exits;
7161 			break;
7162 		}
7163 
7164 		kvm_check_async_pf_completion(vcpu);
7165 
7166 		if (signal_pending(current)) {
7167 			r = -EINTR;
7168 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7169 			++vcpu->stat.signal_exits;
7170 			break;
7171 		}
7172 		if (need_resched()) {
7173 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7174 			cond_resched();
7175 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7176 		}
7177 	}
7178 
7179 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7180 
7181 	return r;
7182 }
7183 
7184 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7185 {
7186 	int r;
7187 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7188 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7189 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7190 	if (r != EMULATE_DONE)
7191 		return 0;
7192 	return 1;
7193 }
7194 
7195 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7196 {
7197 	BUG_ON(!vcpu->arch.pio.count);
7198 
7199 	return complete_emulated_io(vcpu);
7200 }
7201 
7202 /*
7203  * Implements the following, as a state machine:
7204  *
7205  * read:
7206  *   for each fragment
7207  *     for each mmio piece in the fragment
7208  *       write gpa, len
7209  *       exit
7210  *       copy data
7211  *   execute insn
7212  *
7213  * write:
7214  *   for each fragment
7215  *     for each mmio piece in the fragment
7216  *       write gpa, len
7217  *       copy data
7218  *       exit
7219  */
7220 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7221 {
7222 	struct kvm_run *run = vcpu->run;
7223 	struct kvm_mmio_fragment *frag;
7224 	unsigned len;
7225 
7226 	BUG_ON(!vcpu->mmio_needed);
7227 
7228 	/* Complete previous fragment */
7229 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7230 	len = min(8u, frag->len);
7231 	if (!vcpu->mmio_is_write)
7232 		memcpy(frag->data, run->mmio.data, len);
7233 
7234 	if (frag->len <= 8) {
7235 		/* Switch to the next fragment. */
7236 		frag++;
7237 		vcpu->mmio_cur_fragment++;
7238 	} else {
7239 		/* Go forward to the next mmio piece. */
7240 		frag->data += len;
7241 		frag->gpa += len;
7242 		frag->len -= len;
7243 	}
7244 
7245 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7246 		vcpu->mmio_needed = 0;
7247 
7248 		/* FIXME: return into emulator if single-stepping.  */
7249 		if (vcpu->mmio_is_write)
7250 			return 1;
7251 		vcpu->mmio_read_completed = 1;
7252 		return complete_emulated_io(vcpu);
7253 	}
7254 
7255 	run->exit_reason = KVM_EXIT_MMIO;
7256 	run->mmio.phys_addr = frag->gpa;
7257 	if (vcpu->mmio_is_write)
7258 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7259 	run->mmio.len = min(8u, frag->len);
7260 	run->mmio.is_write = vcpu->mmio_is_write;
7261 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7262 	return 0;
7263 }
7264 
7265 
7266 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7267 {
7268 	struct fpu *fpu = &current->thread.fpu;
7269 	int r;
7270 
7271 	fpu__initialize(fpu);
7272 
7273 	kvm_sigset_activate(vcpu);
7274 
7275 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7276 		if (kvm_run->immediate_exit) {
7277 			r = -EINTR;
7278 			goto out;
7279 		}
7280 		kvm_vcpu_block(vcpu);
7281 		kvm_apic_accept_events(vcpu);
7282 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7283 		r = -EAGAIN;
7284 		if (signal_pending(current)) {
7285 			r = -EINTR;
7286 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7287 			++vcpu->stat.signal_exits;
7288 		}
7289 		goto out;
7290 	}
7291 
7292 	/* re-sync apic's tpr */
7293 	if (!lapic_in_kernel(vcpu)) {
7294 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7295 			r = -EINVAL;
7296 			goto out;
7297 		}
7298 	}
7299 
7300 	if (unlikely(vcpu->arch.complete_userspace_io)) {
7301 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7302 		vcpu->arch.complete_userspace_io = NULL;
7303 		r = cui(vcpu);
7304 		if (r <= 0)
7305 			goto out;
7306 	} else
7307 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7308 
7309 	if (kvm_run->immediate_exit)
7310 		r = -EINTR;
7311 	else
7312 		r = vcpu_run(vcpu);
7313 
7314 out:
7315 	post_kvm_run_save(vcpu);
7316 	kvm_sigset_deactivate(vcpu);
7317 
7318 	return r;
7319 }
7320 
7321 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7322 {
7323 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7324 		/*
7325 		 * We are here if userspace calls get_regs() in the middle of
7326 		 * instruction emulation. Registers state needs to be copied
7327 		 * back from emulation context to vcpu. Userspace shouldn't do
7328 		 * that usually, but some bad designed PV devices (vmware
7329 		 * backdoor interface) need this to work
7330 		 */
7331 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7332 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7333 	}
7334 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7335 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7336 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7337 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7338 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7339 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7340 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7341 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7342 #ifdef CONFIG_X86_64
7343 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7344 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7345 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7346 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7347 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7348 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7349 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7350 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7351 #endif
7352 
7353 	regs->rip = kvm_rip_read(vcpu);
7354 	regs->rflags = kvm_get_rflags(vcpu);
7355 
7356 	return 0;
7357 }
7358 
7359 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7360 {
7361 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7362 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7363 
7364 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7365 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7366 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7367 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7368 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7369 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7370 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7371 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7372 #ifdef CONFIG_X86_64
7373 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7374 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7375 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7376 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7377 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7378 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7379 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7380 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7381 #endif
7382 
7383 	kvm_rip_write(vcpu, regs->rip);
7384 	kvm_set_rflags(vcpu, regs->rflags);
7385 
7386 	vcpu->arch.exception.pending = false;
7387 
7388 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7389 
7390 	return 0;
7391 }
7392 
7393 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7394 {
7395 	struct kvm_segment cs;
7396 
7397 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7398 	*db = cs.db;
7399 	*l = cs.l;
7400 }
7401 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7402 
7403 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7404 				  struct kvm_sregs *sregs)
7405 {
7406 	struct desc_ptr dt;
7407 
7408 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7409 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7410 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7411 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7412 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7413 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7414 
7415 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7416 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7417 
7418 	kvm_x86_ops->get_idt(vcpu, &dt);
7419 	sregs->idt.limit = dt.size;
7420 	sregs->idt.base = dt.address;
7421 	kvm_x86_ops->get_gdt(vcpu, &dt);
7422 	sregs->gdt.limit = dt.size;
7423 	sregs->gdt.base = dt.address;
7424 
7425 	sregs->cr0 = kvm_read_cr0(vcpu);
7426 	sregs->cr2 = vcpu->arch.cr2;
7427 	sregs->cr3 = kvm_read_cr3(vcpu);
7428 	sregs->cr4 = kvm_read_cr4(vcpu);
7429 	sregs->cr8 = kvm_get_cr8(vcpu);
7430 	sregs->efer = vcpu->arch.efer;
7431 	sregs->apic_base = kvm_get_apic_base(vcpu);
7432 
7433 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7434 
7435 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7436 		set_bit(vcpu->arch.interrupt.nr,
7437 			(unsigned long *)sregs->interrupt_bitmap);
7438 
7439 	return 0;
7440 }
7441 
7442 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7443 				    struct kvm_mp_state *mp_state)
7444 {
7445 	kvm_apic_accept_events(vcpu);
7446 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7447 					vcpu->arch.pv.pv_unhalted)
7448 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7449 	else
7450 		mp_state->mp_state = vcpu->arch.mp_state;
7451 
7452 	return 0;
7453 }
7454 
7455 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7456 				    struct kvm_mp_state *mp_state)
7457 {
7458 	if (!lapic_in_kernel(vcpu) &&
7459 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7460 		return -EINVAL;
7461 
7462 	/* INITs are latched while in SMM */
7463 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7464 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7465 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7466 		return -EINVAL;
7467 
7468 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7469 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7470 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7471 	} else
7472 		vcpu->arch.mp_state = mp_state->mp_state;
7473 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7474 	return 0;
7475 }
7476 
7477 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7478 		    int reason, bool has_error_code, u32 error_code)
7479 {
7480 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7481 	int ret;
7482 
7483 	init_emulate_ctxt(vcpu);
7484 
7485 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7486 				   has_error_code, error_code);
7487 
7488 	if (ret)
7489 		return EMULATE_FAIL;
7490 
7491 	kvm_rip_write(vcpu, ctxt->eip);
7492 	kvm_set_rflags(vcpu, ctxt->eflags);
7493 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7494 	return EMULATE_DONE;
7495 }
7496 EXPORT_SYMBOL_GPL(kvm_task_switch);
7497 
7498 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7499 				  struct kvm_sregs *sregs)
7500 {
7501 	struct msr_data apic_base_msr;
7502 	int mmu_reset_needed = 0;
7503 	int pending_vec, max_bits, idx;
7504 	struct desc_ptr dt;
7505 
7506 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7507 			(sregs->cr4 & X86_CR4_OSXSAVE))
7508 		return -EINVAL;
7509 
7510 	apic_base_msr.data = sregs->apic_base;
7511 	apic_base_msr.host_initiated = true;
7512 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
7513 		return -EINVAL;
7514 
7515 	dt.size = sregs->idt.limit;
7516 	dt.address = sregs->idt.base;
7517 	kvm_x86_ops->set_idt(vcpu, &dt);
7518 	dt.size = sregs->gdt.limit;
7519 	dt.address = sregs->gdt.base;
7520 	kvm_x86_ops->set_gdt(vcpu, &dt);
7521 
7522 	vcpu->arch.cr2 = sregs->cr2;
7523 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7524 	vcpu->arch.cr3 = sregs->cr3;
7525 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7526 
7527 	kvm_set_cr8(vcpu, sregs->cr8);
7528 
7529 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7530 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7531 
7532 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7533 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7534 	vcpu->arch.cr0 = sregs->cr0;
7535 
7536 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7537 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7538 	if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7539 		kvm_update_cpuid(vcpu);
7540 
7541 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7542 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7543 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7544 		mmu_reset_needed = 1;
7545 	}
7546 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7547 
7548 	if (mmu_reset_needed)
7549 		kvm_mmu_reset_context(vcpu);
7550 
7551 	max_bits = KVM_NR_INTERRUPTS;
7552 	pending_vec = find_first_bit(
7553 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7554 	if (pending_vec < max_bits) {
7555 		kvm_queue_interrupt(vcpu, pending_vec, false);
7556 		pr_debug("Set back pending irq %d\n", pending_vec);
7557 	}
7558 
7559 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7560 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7561 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7562 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7563 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7564 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7565 
7566 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7567 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7568 
7569 	update_cr8_intercept(vcpu);
7570 
7571 	/* Older userspace won't unhalt the vcpu on reset. */
7572 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7573 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7574 	    !is_protmode(vcpu))
7575 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7576 
7577 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7578 
7579 	return 0;
7580 }
7581 
7582 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7583 					struct kvm_guest_debug *dbg)
7584 {
7585 	unsigned long rflags;
7586 	int i, r;
7587 
7588 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7589 		r = -EBUSY;
7590 		if (vcpu->arch.exception.pending)
7591 			goto out;
7592 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7593 			kvm_queue_exception(vcpu, DB_VECTOR);
7594 		else
7595 			kvm_queue_exception(vcpu, BP_VECTOR);
7596 	}
7597 
7598 	/*
7599 	 * Read rflags as long as potentially injected trace flags are still
7600 	 * filtered out.
7601 	 */
7602 	rflags = kvm_get_rflags(vcpu);
7603 
7604 	vcpu->guest_debug = dbg->control;
7605 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7606 		vcpu->guest_debug = 0;
7607 
7608 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7609 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7610 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7611 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7612 	} else {
7613 		for (i = 0; i < KVM_NR_DB_REGS; i++)
7614 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7615 	}
7616 	kvm_update_dr7(vcpu);
7617 
7618 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7619 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7620 			get_segment_base(vcpu, VCPU_SREG_CS);
7621 
7622 	/*
7623 	 * Trigger an rflags update that will inject or remove the trace
7624 	 * flags.
7625 	 */
7626 	kvm_set_rflags(vcpu, rflags);
7627 
7628 	kvm_x86_ops->update_bp_intercept(vcpu);
7629 
7630 	r = 0;
7631 
7632 out:
7633 
7634 	return r;
7635 }
7636 
7637 /*
7638  * Translate a guest virtual address to a guest physical address.
7639  */
7640 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7641 				    struct kvm_translation *tr)
7642 {
7643 	unsigned long vaddr = tr->linear_address;
7644 	gpa_t gpa;
7645 	int idx;
7646 
7647 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7648 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7649 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7650 	tr->physical_address = gpa;
7651 	tr->valid = gpa != UNMAPPED_GVA;
7652 	tr->writeable = 1;
7653 	tr->usermode = 0;
7654 
7655 	return 0;
7656 }
7657 
7658 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7659 {
7660 	struct fxregs_state *fxsave =
7661 			&vcpu->arch.guest_fpu.state.fxsave;
7662 
7663 	memcpy(fpu->fpr, fxsave->st_space, 128);
7664 	fpu->fcw = fxsave->cwd;
7665 	fpu->fsw = fxsave->swd;
7666 	fpu->ftwx = fxsave->twd;
7667 	fpu->last_opcode = fxsave->fop;
7668 	fpu->last_ip = fxsave->rip;
7669 	fpu->last_dp = fxsave->rdp;
7670 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7671 
7672 	return 0;
7673 }
7674 
7675 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7676 {
7677 	struct fxregs_state *fxsave =
7678 			&vcpu->arch.guest_fpu.state.fxsave;
7679 
7680 	memcpy(fxsave->st_space, fpu->fpr, 128);
7681 	fxsave->cwd = fpu->fcw;
7682 	fxsave->swd = fpu->fsw;
7683 	fxsave->twd = fpu->ftwx;
7684 	fxsave->fop = fpu->last_opcode;
7685 	fxsave->rip = fpu->last_ip;
7686 	fxsave->rdp = fpu->last_dp;
7687 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7688 
7689 	return 0;
7690 }
7691 
7692 static void fx_init(struct kvm_vcpu *vcpu)
7693 {
7694 	fpstate_init(&vcpu->arch.guest_fpu.state);
7695 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7696 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7697 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7698 
7699 	/*
7700 	 * Ensure guest xcr0 is valid for loading
7701 	 */
7702 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7703 
7704 	vcpu->arch.cr0 |= X86_CR0_ET;
7705 }
7706 
7707 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7708 {
7709 	if (vcpu->guest_fpu_loaded)
7710 		return;
7711 
7712 	/*
7713 	 * Restore all possible states in the guest,
7714 	 * and assume host would use all available bits.
7715 	 * Guest xcr0 would be loaded later.
7716 	 */
7717 	vcpu->guest_fpu_loaded = 1;
7718 	__kernel_fpu_begin();
7719 	/* PKRU is separately restored in kvm_x86_ops->run.  */
7720 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7721 				~XFEATURE_MASK_PKRU);
7722 	trace_kvm_fpu(1);
7723 }
7724 
7725 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7726 {
7727 	if (!vcpu->guest_fpu_loaded)
7728 		return;
7729 
7730 	vcpu->guest_fpu_loaded = 0;
7731 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7732 	__kernel_fpu_end();
7733 	++vcpu->stat.fpu_reload;
7734 	trace_kvm_fpu(0);
7735 }
7736 
7737 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7738 {
7739 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7740 
7741 	kvmclock_reset(vcpu);
7742 
7743 	kvm_x86_ops->vcpu_free(vcpu);
7744 	free_cpumask_var(wbinvd_dirty_mask);
7745 }
7746 
7747 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7748 						unsigned int id)
7749 {
7750 	struct kvm_vcpu *vcpu;
7751 
7752 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7753 		printk_once(KERN_WARNING
7754 		"kvm: SMP vm created on host with unstable TSC; "
7755 		"guest TSC will not be reliable\n");
7756 
7757 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7758 
7759 	return vcpu;
7760 }
7761 
7762 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7763 {
7764 	int r;
7765 
7766 	kvm_vcpu_mtrr_init(vcpu);
7767 	r = vcpu_load(vcpu);
7768 	if (r)
7769 		return r;
7770 	kvm_vcpu_reset(vcpu, false);
7771 	kvm_mmu_setup(vcpu);
7772 	vcpu_put(vcpu);
7773 	return r;
7774 }
7775 
7776 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7777 {
7778 	struct msr_data msr;
7779 	struct kvm *kvm = vcpu->kvm;
7780 
7781 	kvm_hv_vcpu_postcreate(vcpu);
7782 
7783 	if (vcpu_load(vcpu))
7784 		return;
7785 	msr.data = 0x0;
7786 	msr.index = MSR_IA32_TSC;
7787 	msr.host_initiated = true;
7788 	kvm_write_tsc(vcpu, &msr);
7789 	vcpu_put(vcpu);
7790 
7791 	if (!kvmclock_periodic_sync)
7792 		return;
7793 
7794 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7795 					KVMCLOCK_SYNC_PERIOD);
7796 }
7797 
7798 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7799 {
7800 	int r;
7801 	vcpu->arch.apf.msr_val = 0;
7802 
7803 	r = vcpu_load(vcpu);
7804 	BUG_ON(r);
7805 	kvm_mmu_unload(vcpu);
7806 	vcpu_put(vcpu);
7807 
7808 	kvm_x86_ops->vcpu_free(vcpu);
7809 }
7810 
7811 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7812 {
7813 	vcpu->arch.hflags = 0;
7814 
7815 	vcpu->arch.smi_pending = 0;
7816 	atomic_set(&vcpu->arch.nmi_queued, 0);
7817 	vcpu->arch.nmi_pending = 0;
7818 	vcpu->arch.nmi_injected = false;
7819 	kvm_clear_interrupt_queue(vcpu);
7820 	kvm_clear_exception_queue(vcpu);
7821 	vcpu->arch.exception.pending = false;
7822 
7823 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7824 	kvm_update_dr0123(vcpu);
7825 	vcpu->arch.dr6 = DR6_INIT;
7826 	kvm_update_dr6(vcpu);
7827 	vcpu->arch.dr7 = DR7_FIXED_1;
7828 	kvm_update_dr7(vcpu);
7829 
7830 	vcpu->arch.cr2 = 0;
7831 
7832 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7833 	vcpu->arch.apf.msr_val = 0;
7834 	vcpu->arch.st.msr_val = 0;
7835 
7836 	kvmclock_reset(vcpu);
7837 
7838 	kvm_clear_async_pf_completion_queue(vcpu);
7839 	kvm_async_pf_hash_reset(vcpu);
7840 	vcpu->arch.apf.halted = false;
7841 
7842 	if (kvm_mpx_supported()) {
7843 		void *mpx_state_buffer;
7844 
7845 		/*
7846 		 * To avoid have the INIT path from kvm_apic_has_events() that be
7847 		 * called with loaded FPU and does not let userspace fix the state.
7848 		 */
7849 		kvm_put_guest_fpu(vcpu);
7850 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7851 					XFEATURE_MASK_BNDREGS);
7852 		if (mpx_state_buffer)
7853 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
7854 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7855 					XFEATURE_MASK_BNDCSR);
7856 		if (mpx_state_buffer)
7857 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
7858 	}
7859 
7860 	if (!init_event) {
7861 		kvm_pmu_reset(vcpu);
7862 		vcpu->arch.smbase = 0x30000;
7863 
7864 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7865 		vcpu->arch.msr_misc_features_enables = 0;
7866 
7867 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7868 	}
7869 
7870 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7871 	vcpu->arch.regs_avail = ~0;
7872 	vcpu->arch.regs_dirty = ~0;
7873 
7874 	vcpu->arch.ia32_xss = 0;
7875 
7876 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7877 }
7878 
7879 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7880 {
7881 	struct kvm_segment cs;
7882 
7883 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7884 	cs.selector = vector << 8;
7885 	cs.base = vector << 12;
7886 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7887 	kvm_rip_write(vcpu, 0);
7888 }
7889 
7890 int kvm_arch_hardware_enable(void)
7891 {
7892 	struct kvm *kvm;
7893 	struct kvm_vcpu *vcpu;
7894 	int i;
7895 	int ret;
7896 	u64 local_tsc;
7897 	u64 max_tsc = 0;
7898 	bool stable, backwards_tsc = false;
7899 
7900 	kvm_shared_msr_cpu_online();
7901 	ret = kvm_x86_ops->hardware_enable();
7902 	if (ret != 0)
7903 		return ret;
7904 
7905 	local_tsc = rdtsc();
7906 	stable = !check_tsc_unstable();
7907 	list_for_each_entry(kvm, &vm_list, vm_list) {
7908 		kvm_for_each_vcpu(i, vcpu, kvm) {
7909 			if (!stable && vcpu->cpu == smp_processor_id())
7910 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7911 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7912 				backwards_tsc = true;
7913 				if (vcpu->arch.last_host_tsc > max_tsc)
7914 					max_tsc = vcpu->arch.last_host_tsc;
7915 			}
7916 		}
7917 	}
7918 
7919 	/*
7920 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7921 	 * platforms that reset TSC during suspend or hibernate actions, but
7922 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7923 	 * detect that condition here, which happens early in CPU bringup,
7924 	 * before any KVM threads can be running.  Unfortunately, we can't
7925 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7926 	 * enough into CPU bringup that we know how much real time has actually
7927 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7928 	 * variables that haven't been updated yet.
7929 	 *
7930 	 * So we simply find the maximum observed TSC above, then record the
7931 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7932 	 * the adjustment will be applied.  Note that we accumulate
7933 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7934 	 * gets a chance to run again.  In the event that no KVM threads get a
7935 	 * chance to run, we will miss the entire elapsed period, as we'll have
7936 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7937 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7938 	 * uniform across all VCPUs (not to mention the scenario is extremely
7939 	 * unlikely). It is possible that a second hibernate recovery happens
7940 	 * much faster than a first, causing the observed TSC here to be
7941 	 * smaller; this would require additional padding adjustment, which is
7942 	 * why we set last_host_tsc to the local tsc observed here.
7943 	 *
7944 	 * N.B. - this code below runs only on platforms with reliable TSC,
7945 	 * as that is the only way backwards_tsc is set above.  Also note
7946 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7947 	 * have the same delta_cyc adjustment applied if backwards_tsc
7948 	 * is detected.  Note further, this adjustment is only done once,
7949 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7950 	 * called multiple times (one for each physical CPU bringup).
7951 	 *
7952 	 * Platforms with unreliable TSCs don't have to deal with this, they
7953 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7954 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7955 	 * guarantee that they stay in perfect synchronization.
7956 	 */
7957 	if (backwards_tsc) {
7958 		u64 delta_cyc = max_tsc - local_tsc;
7959 		list_for_each_entry(kvm, &vm_list, vm_list) {
7960 			kvm->arch.backwards_tsc_observed = true;
7961 			kvm_for_each_vcpu(i, vcpu, kvm) {
7962 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7963 				vcpu->arch.last_host_tsc = local_tsc;
7964 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7965 			}
7966 
7967 			/*
7968 			 * We have to disable TSC offset matching.. if you were
7969 			 * booting a VM while issuing an S4 host suspend....
7970 			 * you may have some problem.  Solving this issue is
7971 			 * left as an exercise to the reader.
7972 			 */
7973 			kvm->arch.last_tsc_nsec = 0;
7974 			kvm->arch.last_tsc_write = 0;
7975 		}
7976 
7977 	}
7978 	return 0;
7979 }
7980 
7981 void kvm_arch_hardware_disable(void)
7982 {
7983 	kvm_x86_ops->hardware_disable();
7984 	drop_user_return_notifiers();
7985 }
7986 
7987 int kvm_arch_hardware_setup(void)
7988 {
7989 	int r;
7990 
7991 	r = kvm_x86_ops->hardware_setup();
7992 	if (r != 0)
7993 		return r;
7994 
7995 	if (kvm_has_tsc_control) {
7996 		/*
7997 		 * Make sure the user can only configure tsc_khz values that
7998 		 * fit into a signed integer.
7999 		 * A min value is not calculated needed because it will always
8000 		 * be 1 on all machines.
8001 		 */
8002 		u64 max = min(0x7fffffffULL,
8003 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8004 		kvm_max_guest_tsc_khz = max;
8005 
8006 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8007 	}
8008 
8009 	kvm_init_msr_list();
8010 	return 0;
8011 }
8012 
8013 void kvm_arch_hardware_unsetup(void)
8014 {
8015 	kvm_x86_ops->hardware_unsetup();
8016 }
8017 
8018 void kvm_arch_check_processor_compat(void *rtn)
8019 {
8020 	kvm_x86_ops->check_processor_compatibility(rtn);
8021 }
8022 
8023 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8024 {
8025 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8026 }
8027 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8028 
8029 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8030 {
8031 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8032 }
8033 
8034 struct static_key kvm_no_apic_vcpu __read_mostly;
8035 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8036 
8037 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8038 {
8039 	struct page *page;
8040 	int r;
8041 
8042 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8043 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8044 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8045 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8046 	else
8047 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8048 
8049 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8050 	if (!page) {
8051 		r = -ENOMEM;
8052 		goto fail;
8053 	}
8054 	vcpu->arch.pio_data = page_address(page);
8055 
8056 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
8057 
8058 	r = kvm_mmu_create(vcpu);
8059 	if (r < 0)
8060 		goto fail_free_pio_data;
8061 
8062 	if (irqchip_in_kernel(vcpu->kvm)) {
8063 		r = kvm_create_lapic(vcpu);
8064 		if (r < 0)
8065 			goto fail_mmu_destroy;
8066 	} else
8067 		static_key_slow_inc(&kvm_no_apic_vcpu);
8068 
8069 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8070 				       GFP_KERNEL);
8071 	if (!vcpu->arch.mce_banks) {
8072 		r = -ENOMEM;
8073 		goto fail_free_lapic;
8074 	}
8075 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8076 
8077 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8078 		r = -ENOMEM;
8079 		goto fail_free_mce_banks;
8080 	}
8081 
8082 	fx_init(vcpu);
8083 
8084 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8085 
8086 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8087 
8088 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8089 
8090 	kvm_async_pf_hash_reset(vcpu);
8091 	kvm_pmu_init(vcpu);
8092 
8093 	vcpu->arch.pending_external_vector = -1;
8094 	vcpu->arch.preempted_in_kernel = false;
8095 
8096 	kvm_hv_vcpu_init(vcpu);
8097 
8098 	return 0;
8099 
8100 fail_free_mce_banks:
8101 	kfree(vcpu->arch.mce_banks);
8102 fail_free_lapic:
8103 	kvm_free_lapic(vcpu);
8104 fail_mmu_destroy:
8105 	kvm_mmu_destroy(vcpu);
8106 fail_free_pio_data:
8107 	free_page((unsigned long)vcpu->arch.pio_data);
8108 fail:
8109 	return r;
8110 }
8111 
8112 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8113 {
8114 	int idx;
8115 
8116 	kvm_hv_vcpu_uninit(vcpu);
8117 	kvm_pmu_destroy(vcpu);
8118 	kfree(vcpu->arch.mce_banks);
8119 	kvm_free_lapic(vcpu);
8120 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8121 	kvm_mmu_destroy(vcpu);
8122 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8123 	free_page((unsigned long)vcpu->arch.pio_data);
8124 	if (!lapic_in_kernel(vcpu))
8125 		static_key_slow_dec(&kvm_no_apic_vcpu);
8126 }
8127 
8128 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8129 {
8130 	kvm_x86_ops->sched_in(vcpu, cpu);
8131 }
8132 
8133 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8134 {
8135 	if (type)
8136 		return -EINVAL;
8137 
8138 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8139 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8140 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8141 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8142 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8143 
8144 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8145 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8146 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8147 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8148 		&kvm->arch.irq_sources_bitmap);
8149 
8150 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8151 	mutex_init(&kvm->arch.apic_map_lock);
8152 	mutex_init(&kvm->arch.hyperv.hv_lock);
8153 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8154 
8155 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8156 	pvclock_update_vm_gtod_copy(kvm);
8157 
8158 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8159 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8160 
8161 	kvm_page_track_init(kvm);
8162 	kvm_mmu_init_vm(kvm);
8163 
8164 	if (kvm_x86_ops->vm_init)
8165 		return kvm_x86_ops->vm_init(kvm);
8166 
8167 	return 0;
8168 }
8169 
8170 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8171 {
8172 	int r;
8173 	r = vcpu_load(vcpu);
8174 	BUG_ON(r);
8175 	kvm_mmu_unload(vcpu);
8176 	vcpu_put(vcpu);
8177 }
8178 
8179 static void kvm_free_vcpus(struct kvm *kvm)
8180 {
8181 	unsigned int i;
8182 	struct kvm_vcpu *vcpu;
8183 
8184 	/*
8185 	 * Unpin any mmu pages first.
8186 	 */
8187 	kvm_for_each_vcpu(i, vcpu, kvm) {
8188 		kvm_clear_async_pf_completion_queue(vcpu);
8189 		kvm_unload_vcpu_mmu(vcpu);
8190 	}
8191 	kvm_for_each_vcpu(i, vcpu, kvm)
8192 		kvm_arch_vcpu_free(vcpu);
8193 
8194 	mutex_lock(&kvm->lock);
8195 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8196 		kvm->vcpus[i] = NULL;
8197 
8198 	atomic_set(&kvm->online_vcpus, 0);
8199 	mutex_unlock(&kvm->lock);
8200 }
8201 
8202 void kvm_arch_sync_events(struct kvm *kvm)
8203 {
8204 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8205 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8206 	kvm_free_pit(kvm);
8207 }
8208 
8209 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8210 {
8211 	int i, r;
8212 	unsigned long hva;
8213 	struct kvm_memslots *slots = kvm_memslots(kvm);
8214 	struct kvm_memory_slot *slot, old;
8215 
8216 	/* Called with kvm->slots_lock held.  */
8217 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8218 		return -EINVAL;
8219 
8220 	slot = id_to_memslot(slots, id);
8221 	if (size) {
8222 		if (slot->npages)
8223 			return -EEXIST;
8224 
8225 		/*
8226 		 * MAP_SHARED to prevent internal slot pages from being moved
8227 		 * by fork()/COW.
8228 		 */
8229 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8230 			      MAP_SHARED | MAP_ANONYMOUS, 0);
8231 		if (IS_ERR((void *)hva))
8232 			return PTR_ERR((void *)hva);
8233 	} else {
8234 		if (!slot->npages)
8235 			return 0;
8236 
8237 		hva = 0;
8238 	}
8239 
8240 	old = *slot;
8241 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8242 		struct kvm_userspace_memory_region m;
8243 
8244 		m.slot = id | (i << 16);
8245 		m.flags = 0;
8246 		m.guest_phys_addr = gpa;
8247 		m.userspace_addr = hva;
8248 		m.memory_size = size;
8249 		r = __kvm_set_memory_region(kvm, &m);
8250 		if (r < 0)
8251 			return r;
8252 	}
8253 
8254 	if (!size) {
8255 		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8256 		WARN_ON(r < 0);
8257 	}
8258 
8259 	return 0;
8260 }
8261 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8262 
8263 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8264 {
8265 	int r;
8266 
8267 	mutex_lock(&kvm->slots_lock);
8268 	r = __x86_set_memory_region(kvm, id, gpa, size);
8269 	mutex_unlock(&kvm->slots_lock);
8270 
8271 	return r;
8272 }
8273 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8274 
8275 void kvm_arch_destroy_vm(struct kvm *kvm)
8276 {
8277 	if (current->mm == kvm->mm) {
8278 		/*
8279 		 * Free memory regions allocated on behalf of userspace,
8280 		 * unless the the memory map has changed due to process exit
8281 		 * or fd copying.
8282 		 */
8283 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8284 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8285 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8286 	}
8287 	if (kvm_x86_ops->vm_destroy)
8288 		kvm_x86_ops->vm_destroy(kvm);
8289 	kvm_pic_destroy(kvm);
8290 	kvm_ioapic_destroy(kvm);
8291 	kvm_free_vcpus(kvm);
8292 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8293 	kvm_mmu_uninit_vm(kvm);
8294 	kvm_page_track_cleanup(kvm);
8295 }
8296 
8297 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8298 			   struct kvm_memory_slot *dont)
8299 {
8300 	int i;
8301 
8302 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8303 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8304 			kvfree(free->arch.rmap[i]);
8305 			free->arch.rmap[i] = NULL;
8306 		}
8307 		if (i == 0)
8308 			continue;
8309 
8310 		if (!dont || free->arch.lpage_info[i - 1] !=
8311 			     dont->arch.lpage_info[i - 1]) {
8312 			kvfree(free->arch.lpage_info[i - 1]);
8313 			free->arch.lpage_info[i - 1] = NULL;
8314 		}
8315 	}
8316 
8317 	kvm_page_track_free_memslot(free, dont);
8318 }
8319 
8320 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8321 			    unsigned long npages)
8322 {
8323 	int i;
8324 
8325 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8326 		struct kvm_lpage_info *linfo;
8327 		unsigned long ugfn;
8328 		int lpages;
8329 		int level = i + 1;
8330 
8331 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
8332 				      slot->base_gfn, level) + 1;
8333 
8334 		slot->arch.rmap[i] =
8335 			kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8336 		if (!slot->arch.rmap[i])
8337 			goto out_free;
8338 		if (i == 0)
8339 			continue;
8340 
8341 		linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8342 		if (!linfo)
8343 			goto out_free;
8344 
8345 		slot->arch.lpage_info[i - 1] = linfo;
8346 
8347 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8348 			linfo[0].disallow_lpage = 1;
8349 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8350 			linfo[lpages - 1].disallow_lpage = 1;
8351 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
8352 		/*
8353 		 * If the gfn and userspace address are not aligned wrt each
8354 		 * other, or if explicitly asked to, disable large page
8355 		 * support for this slot
8356 		 */
8357 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8358 		    !kvm_largepages_enabled()) {
8359 			unsigned long j;
8360 
8361 			for (j = 0; j < lpages; ++j)
8362 				linfo[j].disallow_lpage = 1;
8363 		}
8364 	}
8365 
8366 	if (kvm_page_track_create_memslot(slot, npages))
8367 		goto out_free;
8368 
8369 	return 0;
8370 
8371 out_free:
8372 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8373 		kvfree(slot->arch.rmap[i]);
8374 		slot->arch.rmap[i] = NULL;
8375 		if (i == 0)
8376 			continue;
8377 
8378 		kvfree(slot->arch.lpage_info[i - 1]);
8379 		slot->arch.lpage_info[i - 1] = NULL;
8380 	}
8381 	return -ENOMEM;
8382 }
8383 
8384 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8385 {
8386 	/*
8387 	 * memslots->generation has been incremented.
8388 	 * mmio generation may have reached its maximum value.
8389 	 */
8390 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8391 }
8392 
8393 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8394 				struct kvm_memory_slot *memslot,
8395 				const struct kvm_userspace_memory_region *mem,
8396 				enum kvm_mr_change change)
8397 {
8398 	return 0;
8399 }
8400 
8401 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8402 				     struct kvm_memory_slot *new)
8403 {
8404 	/* Still write protect RO slot */
8405 	if (new->flags & KVM_MEM_READONLY) {
8406 		kvm_mmu_slot_remove_write_access(kvm, new);
8407 		return;
8408 	}
8409 
8410 	/*
8411 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
8412 	 *
8413 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
8414 	 *
8415 	 *  - KVM_MR_CREATE with dirty logging is disabled
8416 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8417 	 *
8418 	 * The reason is, in case of PML, we need to set D-bit for any slots
8419 	 * with dirty logging disabled in order to eliminate unnecessary GPA
8420 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
8421 	 * guarantees leaving PML enabled during guest's lifetime won't have
8422 	 * any additonal overhead from PML when guest is running with dirty
8423 	 * logging disabled for memory slots.
8424 	 *
8425 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8426 	 * to dirty logging mode.
8427 	 *
8428 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8429 	 *
8430 	 * In case of write protect:
8431 	 *
8432 	 * Write protect all pages for dirty logging.
8433 	 *
8434 	 * All the sptes including the large sptes which point to this
8435 	 * slot are set to readonly. We can not create any new large
8436 	 * spte on this slot until the end of the logging.
8437 	 *
8438 	 * See the comments in fast_page_fault().
8439 	 */
8440 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8441 		if (kvm_x86_ops->slot_enable_log_dirty)
8442 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8443 		else
8444 			kvm_mmu_slot_remove_write_access(kvm, new);
8445 	} else {
8446 		if (kvm_x86_ops->slot_disable_log_dirty)
8447 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8448 	}
8449 }
8450 
8451 void kvm_arch_commit_memory_region(struct kvm *kvm,
8452 				const struct kvm_userspace_memory_region *mem,
8453 				const struct kvm_memory_slot *old,
8454 				const struct kvm_memory_slot *new,
8455 				enum kvm_mr_change change)
8456 {
8457 	int nr_mmu_pages = 0;
8458 
8459 	if (!kvm->arch.n_requested_mmu_pages)
8460 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8461 
8462 	if (nr_mmu_pages)
8463 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8464 
8465 	/*
8466 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
8467 	 * sptes have to be split.  If live migration is successful, the guest
8468 	 * in the source machine will be destroyed and large sptes will be
8469 	 * created in the destination. However, if the guest continues to run
8470 	 * in the source machine (for example if live migration fails), small
8471 	 * sptes will remain around and cause bad performance.
8472 	 *
8473 	 * Scan sptes if dirty logging has been stopped, dropping those
8474 	 * which can be collapsed into a single large-page spte.  Later
8475 	 * page faults will create the large-page sptes.
8476 	 */
8477 	if ((change != KVM_MR_DELETE) &&
8478 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8479 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8480 		kvm_mmu_zap_collapsible_sptes(kvm, new);
8481 
8482 	/*
8483 	 * Set up write protection and/or dirty logging for the new slot.
8484 	 *
8485 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8486 	 * been zapped so no dirty logging staff is needed for old slot. For
8487 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8488 	 * new and it's also covered when dealing with the new slot.
8489 	 *
8490 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
8491 	 */
8492 	if (change != KVM_MR_DELETE)
8493 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8494 }
8495 
8496 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8497 {
8498 	kvm_mmu_invalidate_zap_all_pages(kvm);
8499 }
8500 
8501 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8502 				   struct kvm_memory_slot *slot)
8503 {
8504 	kvm_page_track_flush_slot(kvm, slot);
8505 }
8506 
8507 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8508 {
8509 	if (!list_empty_careful(&vcpu->async_pf.done))
8510 		return true;
8511 
8512 	if (kvm_apic_has_events(vcpu))
8513 		return true;
8514 
8515 	if (vcpu->arch.pv.pv_unhalted)
8516 		return true;
8517 
8518 	if (vcpu->arch.exception.pending)
8519 		return true;
8520 
8521 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8522 	    (vcpu->arch.nmi_pending &&
8523 	     kvm_x86_ops->nmi_allowed(vcpu)))
8524 		return true;
8525 
8526 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8527 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
8528 		return true;
8529 
8530 	if (kvm_arch_interrupt_allowed(vcpu) &&
8531 	    kvm_cpu_has_interrupt(vcpu))
8532 		return true;
8533 
8534 	if (kvm_hv_has_stimer_pending(vcpu))
8535 		return true;
8536 
8537 	return false;
8538 }
8539 
8540 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8541 {
8542 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8543 }
8544 
8545 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8546 {
8547 	return vcpu->arch.preempted_in_kernel;
8548 }
8549 
8550 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8551 {
8552 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8553 }
8554 
8555 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8556 {
8557 	return kvm_x86_ops->interrupt_allowed(vcpu);
8558 }
8559 
8560 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8561 {
8562 	if (is_64_bit_mode(vcpu))
8563 		return kvm_rip_read(vcpu);
8564 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8565 		     kvm_rip_read(vcpu));
8566 }
8567 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8568 
8569 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8570 {
8571 	return kvm_get_linear_rip(vcpu) == linear_rip;
8572 }
8573 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8574 
8575 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8576 {
8577 	unsigned long rflags;
8578 
8579 	rflags = kvm_x86_ops->get_rflags(vcpu);
8580 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8581 		rflags &= ~X86_EFLAGS_TF;
8582 	return rflags;
8583 }
8584 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8585 
8586 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8587 {
8588 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8589 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8590 		rflags |= X86_EFLAGS_TF;
8591 	kvm_x86_ops->set_rflags(vcpu, rflags);
8592 }
8593 
8594 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8595 {
8596 	__kvm_set_rflags(vcpu, rflags);
8597 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8598 }
8599 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8600 
8601 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8602 {
8603 	int r;
8604 
8605 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8606 	      work->wakeup_all)
8607 		return;
8608 
8609 	r = kvm_mmu_reload(vcpu);
8610 	if (unlikely(r))
8611 		return;
8612 
8613 	if (!vcpu->arch.mmu.direct_map &&
8614 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8615 		return;
8616 
8617 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8618 }
8619 
8620 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8621 {
8622 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8623 }
8624 
8625 static inline u32 kvm_async_pf_next_probe(u32 key)
8626 {
8627 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8628 }
8629 
8630 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8631 {
8632 	u32 key = kvm_async_pf_hash_fn(gfn);
8633 
8634 	while (vcpu->arch.apf.gfns[key] != ~0)
8635 		key = kvm_async_pf_next_probe(key);
8636 
8637 	vcpu->arch.apf.gfns[key] = gfn;
8638 }
8639 
8640 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8641 {
8642 	int i;
8643 	u32 key = kvm_async_pf_hash_fn(gfn);
8644 
8645 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8646 		     (vcpu->arch.apf.gfns[key] != gfn &&
8647 		      vcpu->arch.apf.gfns[key] != ~0); i++)
8648 		key = kvm_async_pf_next_probe(key);
8649 
8650 	return key;
8651 }
8652 
8653 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8654 {
8655 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8656 }
8657 
8658 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8659 {
8660 	u32 i, j, k;
8661 
8662 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8663 	while (true) {
8664 		vcpu->arch.apf.gfns[i] = ~0;
8665 		do {
8666 			j = kvm_async_pf_next_probe(j);
8667 			if (vcpu->arch.apf.gfns[j] == ~0)
8668 				return;
8669 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8670 			/*
8671 			 * k lies cyclically in ]i,j]
8672 			 * |    i.k.j |
8673 			 * |....j i.k.| or  |.k..j i...|
8674 			 */
8675 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8676 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8677 		i = j;
8678 	}
8679 }
8680 
8681 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8682 {
8683 
8684 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8685 				      sizeof(val));
8686 }
8687 
8688 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8689 {
8690 
8691 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8692 				      sizeof(u32));
8693 }
8694 
8695 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8696 				     struct kvm_async_pf *work)
8697 {
8698 	struct x86_exception fault;
8699 
8700 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8701 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8702 
8703 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8704 	    (vcpu->arch.apf.send_user_only &&
8705 	     kvm_x86_ops->get_cpl(vcpu) == 0))
8706 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8707 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8708 		fault.vector = PF_VECTOR;
8709 		fault.error_code_valid = true;
8710 		fault.error_code = 0;
8711 		fault.nested_page_fault = false;
8712 		fault.address = work->arch.token;
8713 		fault.async_page_fault = true;
8714 		kvm_inject_page_fault(vcpu, &fault);
8715 	}
8716 }
8717 
8718 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8719 				 struct kvm_async_pf *work)
8720 {
8721 	struct x86_exception fault;
8722 	u32 val;
8723 
8724 	if (work->wakeup_all)
8725 		work->arch.token = ~0; /* broadcast wakeup */
8726 	else
8727 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8728 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8729 
8730 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8731 	    !apf_get_user(vcpu, &val)) {
8732 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8733 		    vcpu->arch.exception.pending &&
8734 		    vcpu->arch.exception.nr == PF_VECTOR &&
8735 		    !apf_put_user(vcpu, 0)) {
8736 			vcpu->arch.exception.injected = false;
8737 			vcpu->arch.exception.pending = false;
8738 			vcpu->arch.exception.nr = 0;
8739 			vcpu->arch.exception.has_error_code = false;
8740 			vcpu->arch.exception.error_code = 0;
8741 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8742 			fault.vector = PF_VECTOR;
8743 			fault.error_code_valid = true;
8744 			fault.error_code = 0;
8745 			fault.nested_page_fault = false;
8746 			fault.address = work->arch.token;
8747 			fault.async_page_fault = true;
8748 			kvm_inject_page_fault(vcpu, &fault);
8749 		}
8750 	}
8751 	vcpu->arch.apf.halted = false;
8752 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8753 }
8754 
8755 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8756 {
8757 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8758 		return true;
8759 	else
8760 		return kvm_can_do_async_pf(vcpu);
8761 }
8762 
8763 void kvm_arch_start_assignment(struct kvm *kvm)
8764 {
8765 	atomic_inc(&kvm->arch.assigned_device_count);
8766 }
8767 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8768 
8769 void kvm_arch_end_assignment(struct kvm *kvm)
8770 {
8771 	atomic_dec(&kvm->arch.assigned_device_count);
8772 }
8773 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8774 
8775 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8776 {
8777 	return atomic_read(&kvm->arch.assigned_device_count);
8778 }
8779 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8780 
8781 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8782 {
8783 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8784 }
8785 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8786 
8787 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8788 {
8789 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8790 }
8791 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8792 
8793 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8794 {
8795 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8796 }
8797 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8798 
8799 bool kvm_arch_has_irq_bypass(void)
8800 {
8801 	return kvm_x86_ops->update_pi_irte != NULL;
8802 }
8803 
8804 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8805 				      struct irq_bypass_producer *prod)
8806 {
8807 	struct kvm_kernel_irqfd *irqfd =
8808 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8809 
8810 	irqfd->producer = prod;
8811 
8812 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8813 					   prod->irq, irqfd->gsi, 1);
8814 }
8815 
8816 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8817 				      struct irq_bypass_producer *prod)
8818 {
8819 	int ret;
8820 	struct kvm_kernel_irqfd *irqfd =
8821 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8822 
8823 	WARN_ON(irqfd->producer != prod);
8824 	irqfd->producer = NULL;
8825 
8826 	/*
8827 	 * When producer of consumer is unregistered, we change back to
8828 	 * remapped mode, so we can re-use the current implementation
8829 	 * when the irq is masked/disabled or the consumer side (KVM
8830 	 * int this case doesn't want to receive the interrupts.
8831 	*/
8832 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8833 	if (ret)
8834 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8835 		       " fails: %d\n", irqfd->consumer.token, ret);
8836 }
8837 
8838 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8839 				   uint32_t guest_irq, bool set)
8840 {
8841 	if (!kvm_x86_ops->update_pi_irte)
8842 		return -EINVAL;
8843 
8844 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8845 }
8846 
8847 bool kvm_vector_hashing_enabled(void)
8848 {
8849 	return vector_hashing;
8850 }
8851 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8852 
8853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8855 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8856 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
8872