xref: /openbmc/linux/arch/x86/kvm/x86.c (revision d4295e12)
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 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 
73 #define CREATE_TRACE_POINTS
74 #include "trace.h"
75 
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
80 
81 #define emul_to_vcpu(ctxt) \
82 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83 
84 /* EFER defaults:
85  * - enable syscall per default because its emulated by KVM
86  * - enable LME and LMA per default on 64 bit KVM
87  */
88 #ifdef CONFIG_X86_64
89 static
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
91 #else
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93 #endif
94 
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
97 
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
100 
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
107 
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
110 
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
113 
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
116 
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
119 
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
122 
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32  __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64  __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
133 
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
137 
138 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
139 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
142 
143 static bool __read_mostly vector_hashing = true;
144 module_param(vector_hashing, bool, S_IRUGO);
145 
146 bool __read_mostly enable_vmware_backdoor = false;
147 module_param(enable_vmware_backdoor, bool, S_IRUGO);
148 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
149 
150 static bool __read_mostly force_emulation_prefix = false;
151 module_param(force_emulation_prefix, bool, S_IRUGO);
152 
153 #define KVM_NR_SHARED_MSRS 16
154 
155 struct kvm_shared_msrs_global {
156 	int nr;
157 	u32 msrs[KVM_NR_SHARED_MSRS];
158 };
159 
160 struct kvm_shared_msrs {
161 	struct user_return_notifier urn;
162 	bool registered;
163 	struct kvm_shared_msr_values {
164 		u64 host;
165 		u64 curr;
166 	} values[KVM_NR_SHARED_MSRS];
167 };
168 
169 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
170 static struct kvm_shared_msrs __percpu *shared_msrs;
171 
172 struct kvm_stats_debugfs_item debugfs_entries[] = {
173 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
174 	{ "pf_guest", VCPU_STAT(pf_guest) },
175 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
176 	{ "invlpg", VCPU_STAT(invlpg) },
177 	{ "exits", VCPU_STAT(exits) },
178 	{ "io_exits", VCPU_STAT(io_exits) },
179 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
180 	{ "signal_exits", VCPU_STAT(signal_exits) },
181 	{ "irq_window", VCPU_STAT(irq_window_exits) },
182 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
183 	{ "halt_exits", VCPU_STAT(halt_exits) },
184 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
185 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
186 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
187 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
188 	{ "hypercalls", VCPU_STAT(hypercalls) },
189 	{ "request_irq", VCPU_STAT(request_irq_exits) },
190 	{ "irq_exits", VCPU_STAT(irq_exits) },
191 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
192 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
193 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
194 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
195 	{ "irq_injections", VCPU_STAT(irq_injections) },
196 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
197 	{ "req_event", VCPU_STAT(req_event) },
198 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
199 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
200 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
201 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
202 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
203 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
204 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
205 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
206 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
207 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
208 	{ "largepages", VM_STAT(lpages) },
209 	{ "max_mmu_page_hash_collisions",
210 		VM_STAT(max_mmu_page_hash_collisions) },
211 	{ NULL }
212 };
213 
214 u64 __read_mostly host_xcr0;
215 
216 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
217 
218 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
219 {
220 	int i;
221 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
222 		vcpu->arch.apf.gfns[i] = ~0;
223 }
224 
225 static void kvm_on_user_return(struct user_return_notifier *urn)
226 {
227 	unsigned slot;
228 	struct kvm_shared_msrs *locals
229 		= container_of(urn, struct kvm_shared_msrs, urn);
230 	struct kvm_shared_msr_values *values;
231 	unsigned long flags;
232 
233 	/*
234 	 * Disabling irqs at this point since the following code could be
235 	 * interrupted and executed through kvm_arch_hardware_disable()
236 	 */
237 	local_irq_save(flags);
238 	if (locals->registered) {
239 		locals->registered = false;
240 		user_return_notifier_unregister(urn);
241 	}
242 	local_irq_restore(flags);
243 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
244 		values = &locals->values[slot];
245 		if (values->host != values->curr) {
246 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
247 			values->curr = values->host;
248 		}
249 	}
250 }
251 
252 static void shared_msr_update(unsigned slot, u32 msr)
253 {
254 	u64 value;
255 	unsigned int cpu = smp_processor_id();
256 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
257 
258 	/* only read, and nobody should modify it at this time,
259 	 * so don't need lock */
260 	if (slot >= shared_msrs_global.nr) {
261 		printk(KERN_ERR "kvm: invalid MSR slot!");
262 		return;
263 	}
264 	rdmsrl_safe(msr, &value);
265 	smsr->values[slot].host = value;
266 	smsr->values[slot].curr = value;
267 }
268 
269 void kvm_define_shared_msr(unsigned slot, u32 msr)
270 {
271 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
272 	shared_msrs_global.msrs[slot] = msr;
273 	if (slot >= shared_msrs_global.nr)
274 		shared_msrs_global.nr = slot + 1;
275 }
276 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
277 
278 static void kvm_shared_msr_cpu_online(void)
279 {
280 	unsigned i;
281 
282 	for (i = 0; i < shared_msrs_global.nr; ++i)
283 		shared_msr_update(i, shared_msrs_global.msrs[i]);
284 }
285 
286 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
287 {
288 	unsigned int cpu = smp_processor_id();
289 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
290 	int err;
291 
292 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
293 		return 0;
294 	smsr->values[slot].curr = value;
295 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
296 	if (err)
297 		return 1;
298 
299 	if (!smsr->registered) {
300 		smsr->urn.on_user_return = kvm_on_user_return;
301 		user_return_notifier_register(&smsr->urn);
302 		smsr->registered = true;
303 	}
304 	return 0;
305 }
306 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
307 
308 static void drop_user_return_notifiers(void)
309 {
310 	unsigned int cpu = smp_processor_id();
311 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
312 
313 	if (smsr->registered)
314 		kvm_on_user_return(&smsr->urn);
315 }
316 
317 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
318 {
319 	return vcpu->arch.apic_base;
320 }
321 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
322 
323 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
324 {
325 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
326 }
327 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
328 
329 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
330 {
331 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
332 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
333 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
334 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
335 
336 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
337 		return 1;
338 	if (!msr_info->host_initiated) {
339 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
340 			return 1;
341 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
342 			return 1;
343 	}
344 
345 	kvm_lapic_set_base(vcpu, msr_info->data);
346 	return 0;
347 }
348 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
349 
350 asmlinkage __visible void kvm_spurious_fault(void)
351 {
352 	/* Fault while not rebooting.  We want the trace. */
353 	BUG();
354 }
355 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
356 
357 #define EXCPT_BENIGN		0
358 #define EXCPT_CONTRIBUTORY	1
359 #define EXCPT_PF		2
360 
361 static int exception_class(int vector)
362 {
363 	switch (vector) {
364 	case PF_VECTOR:
365 		return EXCPT_PF;
366 	case DE_VECTOR:
367 	case TS_VECTOR:
368 	case NP_VECTOR:
369 	case SS_VECTOR:
370 	case GP_VECTOR:
371 		return EXCPT_CONTRIBUTORY;
372 	default:
373 		break;
374 	}
375 	return EXCPT_BENIGN;
376 }
377 
378 #define EXCPT_FAULT		0
379 #define EXCPT_TRAP		1
380 #define EXCPT_ABORT		2
381 #define EXCPT_INTERRUPT		3
382 
383 static int exception_type(int vector)
384 {
385 	unsigned int mask;
386 
387 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
388 		return EXCPT_INTERRUPT;
389 
390 	mask = 1 << vector;
391 
392 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
393 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
394 		return EXCPT_TRAP;
395 
396 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
397 		return EXCPT_ABORT;
398 
399 	/* Reserved exceptions will result in fault */
400 	return EXCPT_FAULT;
401 }
402 
403 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
404 {
405 	unsigned nr = vcpu->arch.exception.nr;
406 	bool has_payload = vcpu->arch.exception.has_payload;
407 	unsigned long payload = vcpu->arch.exception.payload;
408 
409 	if (!has_payload)
410 		return;
411 
412 	switch (nr) {
413 	case DB_VECTOR:
414 		/*
415 		 * "Certain debug exceptions may clear bit 0-3.  The
416 		 * remaining contents of the DR6 register are never
417 		 * cleared by the processor".
418 		 */
419 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
420 		/*
421 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
422 		 */
423 		vcpu->arch.dr6 |= DR6_RTM;
424 		vcpu->arch.dr6 |= payload;
425 		/*
426 		 * Bit 16 should be set in the payload whenever the #DB
427 		 * exception should clear DR6.RTM. This makes the payload
428 		 * compatible with the pending debug exceptions under VMX.
429 		 * Though not currently documented in the SDM, this also
430 		 * makes the payload compatible with the exit qualification
431 		 * for #DB exceptions under VMX.
432 		 */
433 		vcpu->arch.dr6 ^= payload & DR6_RTM;
434 		break;
435 	case PF_VECTOR:
436 		vcpu->arch.cr2 = payload;
437 		break;
438 	}
439 
440 	vcpu->arch.exception.has_payload = false;
441 	vcpu->arch.exception.payload = 0;
442 }
443 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
444 
445 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
446 		unsigned nr, bool has_error, u32 error_code,
447 	        bool has_payload, unsigned long payload, bool reinject)
448 {
449 	u32 prev_nr;
450 	int class1, class2;
451 
452 	kvm_make_request(KVM_REQ_EVENT, vcpu);
453 
454 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
455 	queue:
456 		if (has_error && !is_protmode(vcpu))
457 			has_error = false;
458 		if (reinject) {
459 			/*
460 			 * On vmentry, vcpu->arch.exception.pending is only
461 			 * true if an event injection was blocked by
462 			 * nested_run_pending.  In that case, however,
463 			 * vcpu_enter_guest requests an immediate exit,
464 			 * and the guest shouldn't proceed far enough to
465 			 * need reinjection.
466 			 */
467 			WARN_ON_ONCE(vcpu->arch.exception.pending);
468 			vcpu->arch.exception.injected = true;
469 			if (WARN_ON_ONCE(has_payload)) {
470 				/*
471 				 * A reinjected event has already
472 				 * delivered its payload.
473 				 */
474 				has_payload = false;
475 				payload = 0;
476 			}
477 		} else {
478 			vcpu->arch.exception.pending = true;
479 			vcpu->arch.exception.injected = false;
480 		}
481 		vcpu->arch.exception.has_error_code = has_error;
482 		vcpu->arch.exception.nr = nr;
483 		vcpu->arch.exception.error_code = error_code;
484 		vcpu->arch.exception.has_payload = has_payload;
485 		vcpu->arch.exception.payload = payload;
486 		/*
487 		 * In guest mode, payload delivery should be deferred,
488 		 * so that the L1 hypervisor can intercept #PF before
489 		 * CR2 is modified (or intercept #DB before DR6 is
490 		 * modified under nVMX).  However, for ABI
491 		 * compatibility with KVM_GET_VCPU_EVENTS and
492 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
493 		 * delivery unless userspace has enabled this
494 		 * functionality via the per-VM capability,
495 		 * KVM_CAP_EXCEPTION_PAYLOAD.
496 		 */
497 		if (!vcpu->kvm->arch.exception_payload_enabled ||
498 		    !is_guest_mode(vcpu))
499 			kvm_deliver_exception_payload(vcpu);
500 		return;
501 	}
502 
503 	/* to check exception */
504 	prev_nr = vcpu->arch.exception.nr;
505 	if (prev_nr == DF_VECTOR) {
506 		/* triple fault -> shutdown */
507 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
508 		return;
509 	}
510 	class1 = exception_class(prev_nr);
511 	class2 = exception_class(nr);
512 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
513 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
514 		/*
515 		 * Generate double fault per SDM Table 5-5.  Set
516 		 * exception.pending = true so that the double fault
517 		 * can trigger a nested vmexit.
518 		 */
519 		vcpu->arch.exception.pending = true;
520 		vcpu->arch.exception.injected = false;
521 		vcpu->arch.exception.has_error_code = true;
522 		vcpu->arch.exception.nr = DF_VECTOR;
523 		vcpu->arch.exception.error_code = 0;
524 		vcpu->arch.exception.has_payload = false;
525 		vcpu->arch.exception.payload = 0;
526 	} else
527 		/* replace previous exception with a new one in a hope
528 		   that instruction re-execution will regenerate lost
529 		   exception */
530 		goto queue;
531 }
532 
533 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
534 {
535 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
536 }
537 EXPORT_SYMBOL_GPL(kvm_queue_exception);
538 
539 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
540 {
541 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
542 }
543 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
544 
545 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
546 				  unsigned long payload)
547 {
548 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
549 }
550 
551 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
552 				    u32 error_code, unsigned long payload)
553 {
554 	kvm_multiple_exception(vcpu, nr, true, error_code,
555 			       true, payload, false);
556 }
557 
558 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
559 {
560 	if (err)
561 		kvm_inject_gp(vcpu, 0);
562 	else
563 		return kvm_skip_emulated_instruction(vcpu);
564 
565 	return 1;
566 }
567 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
568 
569 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
570 {
571 	++vcpu->stat.pf_guest;
572 	vcpu->arch.exception.nested_apf =
573 		is_guest_mode(vcpu) && fault->async_page_fault;
574 	if (vcpu->arch.exception.nested_apf) {
575 		vcpu->arch.apf.nested_apf_token = fault->address;
576 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
577 	} else {
578 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
579 					fault->address);
580 	}
581 }
582 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
583 
584 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
585 {
586 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
587 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
588 	else
589 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
590 
591 	return fault->nested_page_fault;
592 }
593 
594 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
595 {
596 	atomic_inc(&vcpu->arch.nmi_queued);
597 	kvm_make_request(KVM_REQ_NMI, vcpu);
598 }
599 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
600 
601 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
602 {
603 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
604 }
605 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
606 
607 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
608 {
609 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
610 }
611 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
612 
613 /*
614  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
615  * a #GP and return false.
616  */
617 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
618 {
619 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
620 		return true;
621 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
622 	return false;
623 }
624 EXPORT_SYMBOL_GPL(kvm_require_cpl);
625 
626 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
627 {
628 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
629 		return true;
630 
631 	kvm_queue_exception(vcpu, UD_VECTOR);
632 	return false;
633 }
634 EXPORT_SYMBOL_GPL(kvm_require_dr);
635 
636 /*
637  * This function will be used to read from the physical memory of the currently
638  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
639  * can read from guest physical or from the guest's guest physical memory.
640  */
641 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
642 			    gfn_t ngfn, void *data, int offset, int len,
643 			    u32 access)
644 {
645 	struct x86_exception exception;
646 	gfn_t real_gfn;
647 	gpa_t ngpa;
648 
649 	ngpa     = gfn_to_gpa(ngfn);
650 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
651 	if (real_gfn == UNMAPPED_GVA)
652 		return -EFAULT;
653 
654 	real_gfn = gpa_to_gfn(real_gfn);
655 
656 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
657 }
658 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
659 
660 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
661 			       void *data, int offset, int len, u32 access)
662 {
663 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
664 				       data, offset, len, access);
665 }
666 
667 /*
668  * Load the pae pdptrs.  Return true is they are all valid.
669  */
670 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
671 {
672 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
673 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
674 	int i;
675 	int ret;
676 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
677 
678 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
679 				      offset * sizeof(u64), sizeof(pdpte),
680 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
681 	if (ret < 0) {
682 		ret = 0;
683 		goto out;
684 	}
685 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
686 		if ((pdpte[i] & PT_PRESENT_MASK) &&
687 		    (pdpte[i] &
688 		     vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
689 			ret = 0;
690 			goto out;
691 		}
692 	}
693 	ret = 1;
694 
695 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
696 	__set_bit(VCPU_EXREG_PDPTR,
697 		  (unsigned long *)&vcpu->arch.regs_avail);
698 	__set_bit(VCPU_EXREG_PDPTR,
699 		  (unsigned long *)&vcpu->arch.regs_dirty);
700 out:
701 
702 	return ret;
703 }
704 EXPORT_SYMBOL_GPL(load_pdptrs);
705 
706 bool pdptrs_changed(struct kvm_vcpu *vcpu)
707 {
708 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
709 	bool changed = true;
710 	int offset;
711 	gfn_t gfn;
712 	int r;
713 
714 	if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
715 		return false;
716 
717 	if (!test_bit(VCPU_EXREG_PDPTR,
718 		      (unsigned long *)&vcpu->arch.regs_avail))
719 		return true;
720 
721 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
722 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
723 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
724 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
725 	if (r < 0)
726 		goto out;
727 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
728 out:
729 
730 	return changed;
731 }
732 EXPORT_SYMBOL_GPL(pdptrs_changed);
733 
734 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
735 {
736 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
737 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
738 
739 	cr0 |= X86_CR0_ET;
740 
741 #ifdef CONFIG_X86_64
742 	if (cr0 & 0xffffffff00000000UL)
743 		return 1;
744 #endif
745 
746 	cr0 &= ~CR0_RESERVED_BITS;
747 
748 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
749 		return 1;
750 
751 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
752 		return 1;
753 
754 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
755 #ifdef CONFIG_X86_64
756 		if ((vcpu->arch.efer & EFER_LME)) {
757 			int cs_db, cs_l;
758 
759 			if (!is_pae(vcpu))
760 				return 1;
761 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
762 			if (cs_l)
763 				return 1;
764 		} else
765 #endif
766 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
767 						 kvm_read_cr3(vcpu)))
768 			return 1;
769 	}
770 
771 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
772 		return 1;
773 
774 	kvm_x86_ops->set_cr0(vcpu, cr0);
775 
776 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
777 		kvm_clear_async_pf_completion_queue(vcpu);
778 		kvm_async_pf_hash_reset(vcpu);
779 	}
780 
781 	if ((cr0 ^ old_cr0) & update_bits)
782 		kvm_mmu_reset_context(vcpu);
783 
784 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
785 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
786 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
787 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
788 
789 	return 0;
790 }
791 EXPORT_SYMBOL_GPL(kvm_set_cr0);
792 
793 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
794 {
795 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
796 }
797 EXPORT_SYMBOL_GPL(kvm_lmsw);
798 
799 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
800 {
801 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
802 			!vcpu->guest_xcr0_loaded) {
803 		/* kvm_set_xcr() also depends on this */
804 		if (vcpu->arch.xcr0 != host_xcr0)
805 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
806 		vcpu->guest_xcr0_loaded = 1;
807 	}
808 }
809 
810 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
811 {
812 	if (vcpu->guest_xcr0_loaded) {
813 		if (vcpu->arch.xcr0 != host_xcr0)
814 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
815 		vcpu->guest_xcr0_loaded = 0;
816 	}
817 }
818 
819 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
820 {
821 	u64 xcr0 = xcr;
822 	u64 old_xcr0 = vcpu->arch.xcr0;
823 	u64 valid_bits;
824 
825 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
826 	if (index != XCR_XFEATURE_ENABLED_MASK)
827 		return 1;
828 	if (!(xcr0 & XFEATURE_MASK_FP))
829 		return 1;
830 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
831 		return 1;
832 
833 	/*
834 	 * Do not allow the guest to set bits that we do not support
835 	 * saving.  However, xcr0 bit 0 is always set, even if the
836 	 * emulated CPU does not support XSAVE (see fx_init).
837 	 */
838 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
839 	if (xcr0 & ~valid_bits)
840 		return 1;
841 
842 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
843 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
844 		return 1;
845 
846 	if (xcr0 & XFEATURE_MASK_AVX512) {
847 		if (!(xcr0 & XFEATURE_MASK_YMM))
848 			return 1;
849 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
850 			return 1;
851 	}
852 	vcpu->arch.xcr0 = xcr0;
853 
854 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
855 		kvm_update_cpuid(vcpu);
856 	return 0;
857 }
858 
859 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
860 {
861 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
862 	    __kvm_set_xcr(vcpu, index, xcr)) {
863 		kvm_inject_gp(vcpu, 0);
864 		return 1;
865 	}
866 	return 0;
867 }
868 EXPORT_SYMBOL_GPL(kvm_set_xcr);
869 
870 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
871 {
872 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
873 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
874 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
875 
876 	if (cr4 & CR4_RESERVED_BITS)
877 		return 1;
878 
879 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
880 		return 1;
881 
882 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
883 		return 1;
884 
885 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
886 		return 1;
887 
888 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
889 		return 1;
890 
891 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
892 		return 1;
893 
894 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
895 		return 1;
896 
897 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
898 		return 1;
899 
900 	if (is_long_mode(vcpu)) {
901 		if (!(cr4 & X86_CR4_PAE))
902 			return 1;
903 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
904 		   && ((cr4 ^ old_cr4) & pdptr_bits)
905 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
906 				   kvm_read_cr3(vcpu)))
907 		return 1;
908 
909 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
910 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
911 			return 1;
912 
913 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
914 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
915 			return 1;
916 	}
917 
918 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
919 		return 1;
920 
921 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
922 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
923 		kvm_mmu_reset_context(vcpu);
924 
925 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
926 		kvm_update_cpuid(vcpu);
927 
928 	return 0;
929 }
930 EXPORT_SYMBOL_GPL(kvm_set_cr4);
931 
932 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
933 {
934 	bool skip_tlb_flush = false;
935 #ifdef CONFIG_X86_64
936 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
937 
938 	if (pcid_enabled) {
939 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
940 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
941 	}
942 #endif
943 
944 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
945 		if (!skip_tlb_flush) {
946 			kvm_mmu_sync_roots(vcpu);
947 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
948 		}
949 		return 0;
950 	}
951 
952 	if (is_long_mode(vcpu) &&
953 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
954 		return 1;
955 	else if (is_pae(vcpu) && is_paging(vcpu) &&
956 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
957 		return 1;
958 
959 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
960 	vcpu->arch.cr3 = cr3;
961 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
962 
963 	return 0;
964 }
965 EXPORT_SYMBOL_GPL(kvm_set_cr3);
966 
967 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
968 {
969 	if (cr8 & CR8_RESERVED_BITS)
970 		return 1;
971 	if (lapic_in_kernel(vcpu))
972 		kvm_lapic_set_tpr(vcpu, cr8);
973 	else
974 		vcpu->arch.cr8 = cr8;
975 	return 0;
976 }
977 EXPORT_SYMBOL_GPL(kvm_set_cr8);
978 
979 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
980 {
981 	if (lapic_in_kernel(vcpu))
982 		return kvm_lapic_get_cr8(vcpu);
983 	else
984 		return vcpu->arch.cr8;
985 }
986 EXPORT_SYMBOL_GPL(kvm_get_cr8);
987 
988 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
989 {
990 	int i;
991 
992 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
993 		for (i = 0; i < KVM_NR_DB_REGS; i++)
994 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
995 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
996 	}
997 }
998 
999 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1000 {
1001 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1002 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1003 }
1004 
1005 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1006 {
1007 	unsigned long dr7;
1008 
1009 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1010 		dr7 = vcpu->arch.guest_debug_dr7;
1011 	else
1012 		dr7 = vcpu->arch.dr7;
1013 	kvm_x86_ops->set_dr7(vcpu, dr7);
1014 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1015 	if (dr7 & DR7_BP_EN_MASK)
1016 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1017 }
1018 
1019 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1020 {
1021 	u64 fixed = DR6_FIXED_1;
1022 
1023 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1024 		fixed |= DR6_RTM;
1025 	return fixed;
1026 }
1027 
1028 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1029 {
1030 	switch (dr) {
1031 	case 0 ... 3:
1032 		vcpu->arch.db[dr] = val;
1033 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1034 			vcpu->arch.eff_db[dr] = val;
1035 		break;
1036 	case 4:
1037 		/* fall through */
1038 	case 6:
1039 		if (val & 0xffffffff00000000ULL)
1040 			return -1; /* #GP */
1041 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1042 		kvm_update_dr6(vcpu);
1043 		break;
1044 	case 5:
1045 		/* fall through */
1046 	default: /* 7 */
1047 		if (val & 0xffffffff00000000ULL)
1048 			return -1; /* #GP */
1049 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1050 		kvm_update_dr7(vcpu);
1051 		break;
1052 	}
1053 
1054 	return 0;
1055 }
1056 
1057 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1058 {
1059 	if (__kvm_set_dr(vcpu, dr, val)) {
1060 		kvm_inject_gp(vcpu, 0);
1061 		return 1;
1062 	}
1063 	return 0;
1064 }
1065 EXPORT_SYMBOL_GPL(kvm_set_dr);
1066 
1067 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1068 {
1069 	switch (dr) {
1070 	case 0 ... 3:
1071 		*val = vcpu->arch.db[dr];
1072 		break;
1073 	case 4:
1074 		/* fall through */
1075 	case 6:
1076 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1077 			*val = vcpu->arch.dr6;
1078 		else
1079 			*val = kvm_x86_ops->get_dr6(vcpu);
1080 		break;
1081 	case 5:
1082 		/* fall through */
1083 	default: /* 7 */
1084 		*val = vcpu->arch.dr7;
1085 		break;
1086 	}
1087 	return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_get_dr);
1090 
1091 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1092 {
1093 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1094 	u64 data;
1095 	int err;
1096 
1097 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1098 	if (err)
1099 		return err;
1100 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1101 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1102 	return err;
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1105 
1106 /*
1107  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1108  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1109  *
1110  * This list is modified at module load time to reflect the
1111  * capabilities of the host cpu. This capabilities test skips MSRs that are
1112  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1113  * may depend on host virtualization features rather than host cpu features.
1114  */
1115 
1116 static u32 msrs_to_save[] = {
1117 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1118 	MSR_STAR,
1119 #ifdef CONFIG_X86_64
1120 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1121 #endif
1122 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1123 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1124 	MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1125 };
1126 
1127 static unsigned num_msrs_to_save;
1128 
1129 static u32 emulated_msrs[] = {
1130 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1131 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1132 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1133 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1134 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1135 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1136 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1137 	HV_X64_MSR_RESET,
1138 	HV_X64_MSR_VP_INDEX,
1139 	HV_X64_MSR_VP_RUNTIME,
1140 	HV_X64_MSR_SCONTROL,
1141 	HV_X64_MSR_STIMER0_CONFIG,
1142 	HV_X64_MSR_VP_ASSIST_PAGE,
1143 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1144 	HV_X64_MSR_TSC_EMULATION_STATUS,
1145 
1146 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1147 	MSR_KVM_PV_EOI_EN,
1148 
1149 	MSR_IA32_TSC_ADJUST,
1150 	MSR_IA32_TSCDEADLINE,
1151 	MSR_IA32_MISC_ENABLE,
1152 	MSR_IA32_MCG_STATUS,
1153 	MSR_IA32_MCG_CTL,
1154 	MSR_IA32_MCG_EXT_CTL,
1155 	MSR_IA32_SMBASE,
1156 	MSR_SMI_COUNT,
1157 	MSR_PLATFORM_INFO,
1158 	MSR_MISC_FEATURES_ENABLES,
1159 	MSR_AMD64_VIRT_SPEC_CTRL,
1160 };
1161 
1162 static unsigned num_emulated_msrs;
1163 
1164 /*
1165  * List of msr numbers which are used to expose MSR-based features that
1166  * can be used by a hypervisor to validate requested CPU features.
1167  */
1168 static u32 msr_based_features[] = {
1169 	MSR_IA32_VMX_BASIC,
1170 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1171 	MSR_IA32_VMX_PINBASED_CTLS,
1172 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1173 	MSR_IA32_VMX_PROCBASED_CTLS,
1174 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1175 	MSR_IA32_VMX_EXIT_CTLS,
1176 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1177 	MSR_IA32_VMX_ENTRY_CTLS,
1178 	MSR_IA32_VMX_MISC,
1179 	MSR_IA32_VMX_CR0_FIXED0,
1180 	MSR_IA32_VMX_CR0_FIXED1,
1181 	MSR_IA32_VMX_CR4_FIXED0,
1182 	MSR_IA32_VMX_CR4_FIXED1,
1183 	MSR_IA32_VMX_VMCS_ENUM,
1184 	MSR_IA32_VMX_PROCBASED_CTLS2,
1185 	MSR_IA32_VMX_EPT_VPID_CAP,
1186 	MSR_IA32_VMX_VMFUNC,
1187 
1188 	MSR_F10H_DECFG,
1189 	MSR_IA32_UCODE_REV,
1190 	MSR_IA32_ARCH_CAPABILITIES,
1191 };
1192 
1193 static unsigned int num_msr_based_features;
1194 
1195 u64 kvm_get_arch_capabilities(void)
1196 {
1197 	u64 data;
1198 
1199 	rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1200 
1201 	/*
1202 	 * If we're doing cache flushes (either "always" or "cond")
1203 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1204 	 * If an outer hypervisor is doing the cache flush for us
1205 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1206 	 * capability to the guest too, and if EPT is disabled we're not
1207 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1208 	 * require a nested hypervisor to do a flush of its own.
1209 	 */
1210 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1211 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1212 
1213 	return data;
1214 }
1215 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1216 
1217 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1218 {
1219 	switch (msr->index) {
1220 	case MSR_IA32_ARCH_CAPABILITIES:
1221 		msr->data = kvm_get_arch_capabilities();
1222 		break;
1223 	case MSR_IA32_UCODE_REV:
1224 		rdmsrl_safe(msr->index, &msr->data);
1225 		break;
1226 	default:
1227 		if (kvm_x86_ops->get_msr_feature(msr))
1228 			return 1;
1229 	}
1230 	return 0;
1231 }
1232 
1233 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1234 {
1235 	struct kvm_msr_entry msr;
1236 	int r;
1237 
1238 	msr.index = index;
1239 	r = kvm_get_msr_feature(&msr);
1240 	if (r)
1241 		return r;
1242 
1243 	*data = msr.data;
1244 
1245 	return 0;
1246 }
1247 
1248 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1249 {
1250 	if (efer & efer_reserved_bits)
1251 		return false;
1252 
1253 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1254 			return false;
1255 
1256 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1257 			return false;
1258 
1259 	return true;
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1262 
1263 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1264 {
1265 	u64 old_efer = vcpu->arch.efer;
1266 
1267 	if (!kvm_valid_efer(vcpu, efer))
1268 		return 1;
1269 
1270 	if (is_paging(vcpu)
1271 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1272 		return 1;
1273 
1274 	efer &= ~EFER_LMA;
1275 	efer |= vcpu->arch.efer & EFER_LMA;
1276 
1277 	kvm_x86_ops->set_efer(vcpu, efer);
1278 
1279 	/* Update reserved bits */
1280 	if ((efer ^ old_efer) & EFER_NX)
1281 		kvm_mmu_reset_context(vcpu);
1282 
1283 	return 0;
1284 }
1285 
1286 void kvm_enable_efer_bits(u64 mask)
1287 {
1288        efer_reserved_bits &= ~mask;
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1291 
1292 /*
1293  * Writes msr value into into the appropriate "register".
1294  * Returns 0 on success, non-0 otherwise.
1295  * Assumes vcpu_load() was already called.
1296  */
1297 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1298 {
1299 	switch (msr->index) {
1300 	case MSR_FS_BASE:
1301 	case MSR_GS_BASE:
1302 	case MSR_KERNEL_GS_BASE:
1303 	case MSR_CSTAR:
1304 	case MSR_LSTAR:
1305 		if (is_noncanonical_address(msr->data, vcpu))
1306 			return 1;
1307 		break;
1308 	case MSR_IA32_SYSENTER_EIP:
1309 	case MSR_IA32_SYSENTER_ESP:
1310 		/*
1311 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1312 		 * non-canonical address is written on Intel but not on
1313 		 * AMD (which ignores the top 32-bits, because it does
1314 		 * not implement 64-bit SYSENTER).
1315 		 *
1316 		 * 64-bit code should hence be able to write a non-canonical
1317 		 * value on AMD.  Making the address canonical ensures that
1318 		 * vmentry does not fail on Intel after writing a non-canonical
1319 		 * value, and that something deterministic happens if the guest
1320 		 * invokes 64-bit SYSENTER.
1321 		 */
1322 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1323 	}
1324 	return kvm_x86_ops->set_msr(vcpu, msr);
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_set_msr);
1327 
1328 /*
1329  * Adapt set_msr() to msr_io()'s calling convention
1330  */
1331 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1332 {
1333 	struct msr_data msr;
1334 	int r;
1335 
1336 	msr.index = index;
1337 	msr.host_initiated = true;
1338 	r = kvm_get_msr(vcpu, &msr);
1339 	if (r)
1340 		return r;
1341 
1342 	*data = msr.data;
1343 	return 0;
1344 }
1345 
1346 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1347 {
1348 	struct msr_data msr;
1349 
1350 	msr.data = *data;
1351 	msr.index = index;
1352 	msr.host_initiated = true;
1353 	return kvm_set_msr(vcpu, &msr);
1354 }
1355 
1356 #ifdef CONFIG_X86_64
1357 struct pvclock_gtod_data {
1358 	seqcount_t	seq;
1359 
1360 	struct { /* extract of a clocksource struct */
1361 		int vclock_mode;
1362 		u64	cycle_last;
1363 		u64	mask;
1364 		u32	mult;
1365 		u32	shift;
1366 	} clock;
1367 
1368 	u64		boot_ns;
1369 	u64		nsec_base;
1370 	u64		wall_time_sec;
1371 };
1372 
1373 static struct pvclock_gtod_data pvclock_gtod_data;
1374 
1375 static void update_pvclock_gtod(struct timekeeper *tk)
1376 {
1377 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1378 	u64 boot_ns;
1379 
1380 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1381 
1382 	write_seqcount_begin(&vdata->seq);
1383 
1384 	/* copy pvclock gtod data */
1385 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1386 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1387 	vdata->clock.mask		= tk->tkr_mono.mask;
1388 	vdata->clock.mult		= tk->tkr_mono.mult;
1389 	vdata->clock.shift		= tk->tkr_mono.shift;
1390 
1391 	vdata->boot_ns			= boot_ns;
1392 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1393 
1394 	vdata->wall_time_sec            = tk->xtime_sec;
1395 
1396 	write_seqcount_end(&vdata->seq);
1397 }
1398 #endif
1399 
1400 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1401 {
1402 	/*
1403 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1404 	 * vcpu_enter_guest.  This function is only called from
1405 	 * the physical CPU that is running vcpu.
1406 	 */
1407 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1408 }
1409 
1410 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1411 {
1412 	int version;
1413 	int r;
1414 	struct pvclock_wall_clock wc;
1415 	struct timespec64 boot;
1416 
1417 	if (!wall_clock)
1418 		return;
1419 
1420 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1421 	if (r)
1422 		return;
1423 
1424 	if (version & 1)
1425 		++version;  /* first time write, random junk */
1426 
1427 	++version;
1428 
1429 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1430 		return;
1431 
1432 	/*
1433 	 * The guest calculates current wall clock time by adding
1434 	 * system time (updated by kvm_guest_time_update below) to the
1435 	 * wall clock specified here.  guest system time equals host
1436 	 * system time for us, thus we must fill in host boot time here.
1437 	 */
1438 	getboottime64(&boot);
1439 
1440 	if (kvm->arch.kvmclock_offset) {
1441 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1442 		boot = timespec64_sub(boot, ts);
1443 	}
1444 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1445 	wc.nsec = boot.tv_nsec;
1446 	wc.version = version;
1447 
1448 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1449 
1450 	version++;
1451 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1452 }
1453 
1454 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1455 {
1456 	do_shl32_div32(dividend, divisor);
1457 	return dividend;
1458 }
1459 
1460 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1461 			       s8 *pshift, u32 *pmultiplier)
1462 {
1463 	uint64_t scaled64;
1464 	int32_t  shift = 0;
1465 	uint64_t tps64;
1466 	uint32_t tps32;
1467 
1468 	tps64 = base_hz;
1469 	scaled64 = scaled_hz;
1470 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1471 		tps64 >>= 1;
1472 		shift--;
1473 	}
1474 
1475 	tps32 = (uint32_t)tps64;
1476 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1477 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1478 			scaled64 >>= 1;
1479 		else
1480 			tps32 <<= 1;
1481 		shift++;
1482 	}
1483 
1484 	*pshift = shift;
1485 	*pmultiplier = div_frac(scaled64, tps32);
1486 
1487 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1488 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1489 }
1490 
1491 #ifdef CONFIG_X86_64
1492 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1493 #endif
1494 
1495 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1496 static unsigned long max_tsc_khz;
1497 
1498 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1499 {
1500 	u64 v = (u64)khz * (1000000 + ppm);
1501 	do_div(v, 1000000);
1502 	return v;
1503 }
1504 
1505 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1506 {
1507 	u64 ratio;
1508 
1509 	/* Guest TSC same frequency as host TSC? */
1510 	if (!scale) {
1511 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1512 		return 0;
1513 	}
1514 
1515 	/* TSC scaling supported? */
1516 	if (!kvm_has_tsc_control) {
1517 		if (user_tsc_khz > tsc_khz) {
1518 			vcpu->arch.tsc_catchup = 1;
1519 			vcpu->arch.tsc_always_catchup = 1;
1520 			return 0;
1521 		} else {
1522 			WARN(1, "user requested TSC rate below hardware speed\n");
1523 			return -1;
1524 		}
1525 	}
1526 
1527 	/* TSC scaling required  - calculate ratio */
1528 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1529 				user_tsc_khz, tsc_khz);
1530 
1531 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1532 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1533 			  user_tsc_khz);
1534 		return -1;
1535 	}
1536 
1537 	vcpu->arch.tsc_scaling_ratio = ratio;
1538 	return 0;
1539 }
1540 
1541 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1542 {
1543 	u32 thresh_lo, thresh_hi;
1544 	int use_scaling = 0;
1545 
1546 	/* tsc_khz can be zero if TSC calibration fails */
1547 	if (user_tsc_khz == 0) {
1548 		/* set tsc_scaling_ratio to a safe value */
1549 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1550 		return -1;
1551 	}
1552 
1553 	/* Compute a scale to convert nanoseconds in TSC cycles */
1554 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1555 			   &vcpu->arch.virtual_tsc_shift,
1556 			   &vcpu->arch.virtual_tsc_mult);
1557 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1558 
1559 	/*
1560 	 * Compute the variation in TSC rate which is acceptable
1561 	 * within the range of tolerance and decide if the
1562 	 * rate being applied is within that bounds of the hardware
1563 	 * rate.  If so, no scaling or compensation need be done.
1564 	 */
1565 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1566 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1567 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1568 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1569 		use_scaling = 1;
1570 	}
1571 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1572 }
1573 
1574 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1575 {
1576 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1577 				      vcpu->arch.virtual_tsc_mult,
1578 				      vcpu->arch.virtual_tsc_shift);
1579 	tsc += vcpu->arch.this_tsc_write;
1580 	return tsc;
1581 }
1582 
1583 static inline int gtod_is_based_on_tsc(int mode)
1584 {
1585 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1586 }
1587 
1588 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1589 {
1590 #ifdef CONFIG_X86_64
1591 	bool vcpus_matched;
1592 	struct kvm_arch *ka = &vcpu->kvm->arch;
1593 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1594 
1595 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1596 			 atomic_read(&vcpu->kvm->online_vcpus));
1597 
1598 	/*
1599 	 * Once the masterclock is enabled, always perform request in
1600 	 * order to update it.
1601 	 *
1602 	 * In order to enable masterclock, the host clocksource must be TSC
1603 	 * and the vcpus need to have matched TSCs.  When that happens,
1604 	 * perform request to enable masterclock.
1605 	 */
1606 	if (ka->use_master_clock ||
1607 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1608 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1609 
1610 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1611 			    atomic_read(&vcpu->kvm->online_vcpus),
1612 		            ka->use_master_clock, gtod->clock.vclock_mode);
1613 #endif
1614 }
1615 
1616 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1617 {
1618 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1619 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1620 }
1621 
1622 /*
1623  * Multiply tsc by a fixed point number represented by ratio.
1624  *
1625  * The most significant 64-N bits (mult) of ratio represent the
1626  * integral part of the fixed point number; the remaining N bits
1627  * (frac) represent the fractional part, ie. ratio represents a fixed
1628  * point number (mult + frac * 2^(-N)).
1629  *
1630  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1631  */
1632 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1633 {
1634 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1635 }
1636 
1637 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1638 {
1639 	u64 _tsc = tsc;
1640 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1641 
1642 	if (ratio != kvm_default_tsc_scaling_ratio)
1643 		_tsc = __scale_tsc(ratio, tsc);
1644 
1645 	return _tsc;
1646 }
1647 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1648 
1649 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1650 {
1651 	u64 tsc;
1652 
1653 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1654 
1655 	return target_tsc - tsc;
1656 }
1657 
1658 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1659 {
1660 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1661 
1662 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1665 
1666 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1667 {
1668 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1669 	vcpu->arch.tsc_offset = offset;
1670 }
1671 
1672 static inline bool kvm_check_tsc_unstable(void)
1673 {
1674 #ifdef CONFIG_X86_64
1675 	/*
1676 	 * TSC is marked unstable when we're running on Hyper-V,
1677 	 * 'TSC page' clocksource is good.
1678 	 */
1679 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1680 		return false;
1681 #endif
1682 	return check_tsc_unstable();
1683 }
1684 
1685 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1686 {
1687 	struct kvm *kvm = vcpu->kvm;
1688 	u64 offset, ns, elapsed;
1689 	unsigned long flags;
1690 	bool matched;
1691 	bool already_matched;
1692 	u64 data = msr->data;
1693 	bool synchronizing = false;
1694 
1695 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1696 	offset = kvm_compute_tsc_offset(vcpu, data);
1697 	ns = ktime_get_boot_ns();
1698 	elapsed = ns - kvm->arch.last_tsc_nsec;
1699 
1700 	if (vcpu->arch.virtual_tsc_khz) {
1701 		if (data == 0 && msr->host_initiated) {
1702 			/*
1703 			 * detection of vcpu initialization -- need to sync
1704 			 * with other vCPUs. This particularly helps to keep
1705 			 * kvm_clock stable after CPU hotplug
1706 			 */
1707 			synchronizing = true;
1708 		} else {
1709 			u64 tsc_exp = kvm->arch.last_tsc_write +
1710 						nsec_to_cycles(vcpu, elapsed);
1711 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1712 			/*
1713 			 * Special case: TSC write with a small delta (1 second)
1714 			 * of virtual cycle time against real time is
1715 			 * interpreted as an attempt to synchronize the CPU.
1716 			 */
1717 			synchronizing = data < tsc_exp + tsc_hz &&
1718 					data + tsc_hz > tsc_exp;
1719 		}
1720 	}
1721 
1722 	/*
1723 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1724 	 * TSC, we add elapsed time in this computation.  We could let the
1725 	 * compensation code attempt to catch up if we fall behind, but
1726 	 * it's better to try to match offsets from the beginning.
1727          */
1728 	if (synchronizing &&
1729 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1730 		if (!kvm_check_tsc_unstable()) {
1731 			offset = kvm->arch.cur_tsc_offset;
1732 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1733 		} else {
1734 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1735 			data += delta;
1736 			offset = kvm_compute_tsc_offset(vcpu, data);
1737 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1738 		}
1739 		matched = true;
1740 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1741 	} else {
1742 		/*
1743 		 * We split periods of matched TSC writes into generations.
1744 		 * For each generation, we track the original measured
1745 		 * nanosecond time, offset, and write, so if TSCs are in
1746 		 * sync, we can match exact offset, and if not, we can match
1747 		 * exact software computation in compute_guest_tsc()
1748 		 *
1749 		 * These values are tracked in kvm->arch.cur_xxx variables.
1750 		 */
1751 		kvm->arch.cur_tsc_generation++;
1752 		kvm->arch.cur_tsc_nsec = ns;
1753 		kvm->arch.cur_tsc_write = data;
1754 		kvm->arch.cur_tsc_offset = offset;
1755 		matched = false;
1756 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1757 			 kvm->arch.cur_tsc_generation, data);
1758 	}
1759 
1760 	/*
1761 	 * We also track th most recent recorded KHZ, write and time to
1762 	 * allow the matching interval to be extended at each write.
1763 	 */
1764 	kvm->arch.last_tsc_nsec = ns;
1765 	kvm->arch.last_tsc_write = data;
1766 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1767 
1768 	vcpu->arch.last_guest_tsc = data;
1769 
1770 	/* Keep track of which generation this VCPU has synchronized to */
1771 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1772 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1773 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1774 
1775 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1776 		update_ia32_tsc_adjust_msr(vcpu, offset);
1777 
1778 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1779 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1780 
1781 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1782 	if (!matched) {
1783 		kvm->arch.nr_vcpus_matched_tsc = 0;
1784 	} else if (!already_matched) {
1785 		kvm->arch.nr_vcpus_matched_tsc++;
1786 	}
1787 
1788 	kvm_track_tsc_matching(vcpu);
1789 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1790 }
1791 
1792 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1793 
1794 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1795 					   s64 adjustment)
1796 {
1797 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1798 }
1799 
1800 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1801 {
1802 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1803 		WARN_ON(adjustment < 0);
1804 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1805 	adjust_tsc_offset_guest(vcpu, adjustment);
1806 }
1807 
1808 #ifdef CONFIG_X86_64
1809 
1810 static u64 read_tsc(void)
1811 {
1812 	u64 ret = (u64)rdtsc_ordered();
1813 	u64 last = pvclock_gtod_data.clock.cycle_last;
1814 
1815 	if (likely(ret >= last))
1816 		return ret;
1817 
1818 	/*
1819 	 * GCC likes to generate cmov here, but this branch is extremely
1820 	 * predictable (it's just a function of time and the likely is
1821 	 * very likely) and there's a data dependence, so force GCC
1822 	 * to generate a branch instead.  I don't barrier() because
1823 	 * we don't actually need a barrier, and if this function
1824 	 * ever gets inlined it will generate worse code.
1825 	 */
1826 	asm volatile ("");
1827 	return last;
1828 }
1829 
1830 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1831 {
1832 	long v;
1833 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1834 	u64 tsc_pg_val;
1835 
1836 	switch (gtod->clock.vclock_mode) {
1837 	case VCLOCK_HVCLOCK:
1838 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1839 						  tsc_timestamp);
1840 		if (tsc_pg_val != U64_MAX) {
1841 			/* TSC page valid */
1842 			*mode = VCLOCK_HVCLOCK;
1843 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1844 				gtod->clock.mask;
1845 		} else {
1846 			/* TSC page invalid */
1847 			*mode = VCLOCK_NONE;
1848 		}
1849 		break;
1850 	case VCLOCK_TSC:
1851 		*mode = VCLOCK_TSC;
1852 		*tsc_timestamp = read_tsc();
1853 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1854 			gtod->clock.mask;
1855 		break;
1856 	default:
1857 		*mode = VCLOCK_NONE;
1858 	}
1859 
1860 	if (*mode == VCLOCK_NONE)
1861 		*tsc_timestamp = v = 0;
1862 
1863 	return v * gtod->clock.mult;
1864 }
1865 
1866 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1867 {
1868 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1869 	unsigned long seq;
1870 	int mode;
1871 	u64 ns;
1872 
1873 	do {
1874 		seq = read_seqcount_begin(&gtod->seq);
1875 		ns = gtod->nsec_base;
1876 		ns += vgettsc(tsc_timestamp, &mode);
1877 		ns >>= gtod->clock.shift;
1878 		ns += gtod->boot_ns;
1879 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1880 	*t = ns;
1881 
1882 	return mode;
1883 }
1884 
1885 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1886 {
1887 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1888 	unsigned long seq;
1889 	int mode;
1890 	u64 ns;
1891 
1892 	do {
1893 		seq = read_seqcount_begin(&gtod->seq);
1894 		ts->tv_sec = gtod->wall_time_sec;
1895 		ns = gtod->nsec_base;
1896 		ns += vgettsc(tsc_timestamp, &mode);
1897 		ns >>= gtod->clock.shift;
1898 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1899 
1900 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1901 	ts->tv_nsec = ns;
1902 
1903 	return mode;
1904 }
1905 
1906 /* returns true if host is using TSC based clocksource */
1907 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1908 {
1909 	/* checked again under seqlock below */
1910 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1911 		return false;
1912 
1913 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1914 						      tsc_timestamp));
1915 }
1916 
1917 /* returns true if host is using TSC based clocksource */
1918 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1919 					   u64 *tsc_timestamp)
1920 {
1921 	/* checked again under seqlock below */
1922 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1923 		return false;
1924 
1925 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1926 }
1927 #endif
1928 
1929 /*
1930  *
1931  * Assuming a stable TSC across physical CPUS, and a stable TSC
1932  * across virtual CPUs, the following condition is possible.
1933  * Each numbered line represents an event visible to both
1934  * CPUs at the next numbered event.
1935  *
1936  * "timespecX" represents host monotonic time. "tscX" represents
1937  * RDTSC value.
1938  *
1939  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1940  *
1941  * 1.  read timespec0,tsc0
1942  * 2.					| timespec1 = timespec0 + N
1943  * 					| tsc1 = tsc0 + M
1944  * 3. transition to guest		| transition to guest
1945  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1946  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1947  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1948  *
1949  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1950  *
1951  * 	- ret0 < ret1
1952  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1953  *		...
1954  *	- 0 < N - M => M < N
1955  *
1956  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1957  * always the case (the difference between two distinct xtime instances
1958  * might be smaller then the difference between corresponding TSC reads,
1959  * when updating guest vcpus pvclock areas).
1960  *
1961  * To avoid that problem, do not allow visibility of distinct
1962  * system_timestamp/tsc_timestamp values simultaneously: use a master
1963  * copy of host monotonic time values. Update that master copy
1964  * in lockstep.
1965  *
1966  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1967  *
1968  */
1969 
1970 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1971 {
1972 #ifdef CONFIG_X86_64
1973 	struct kvm_arch *ka = &kvm->arch;
1974 	int vclock_mode;
1975 	bool host_tsc_clocksource, vcpus_matched;
1976 
1977 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1978 			atomic_read(&kvm->online_vcpus));
1979 
1980 	/*
1981 	 * If the host uses TSC clock, then passthrough TSC as stable
1982 	 * to the guest.
1983 	 */
1984 	host_tsc_clocksource = kvm_get_time_and_clockread(
1985 					&ka->master_kernel_ns,
1986 					&ka->master_cycle_now);
1987 
1988 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1989 				&& !ka->backwards_tsc_observed
1990 				&& !ka->boot_vcpu_runs_old_kvmclock;
1991 
1992 	if (ka->use_master_clock)
1993 		atomic_set(&kvm_guest_has_master_clock, 1);
1994 
1995 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1996 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1997 					vcpus_matched);
1998 #endif
1999 }
2000 
2001 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2002 {
2003 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2004 }
2005 
2006 static void kvm_gen_update_masterclock(struct kvm *kvm)
2007 {
2008 #ifdef CONFIG_X86_64
2009 	int i;
2010 	struct kvm_vcpu *vcpu;
2011 	struct kvm_arch *ka = &kvm->arch;
2012 
2013 	spin_lock(&ka->pvclock_gtod_sync_lock);
2014 	kvm_make_mclock_inprogress_request(kvm);
2015 	/* no guest entries from this point */
2016 	pvclock_update_vm_gtod_copy(kvm);
2017 
2018 	kvm_for_each_vcpu(i, vcpu, kvm)
2019 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2020 
2021 	/* guest entries allowed */
2022 	kvm_for_each_vcpu(i, vcpu, kvm)
2023 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2024 
2025 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2026 #endif
2027 }
2028 
2029 u64 get_kvmclock_ns(struct kvm *kvm)
2030 {
2031 	struct kvm_arch *ka = &kvm->arch;
2032 	struct pvclock_vcpu_time_info hv_clock;
2033 	u64 ret;
2034 
2035 	spin_lock(&ka->pvclock_gtod_sync_lock);
2036 	if (!ka->use_master_clock) {
2037 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2038 		return ktime_get_boot_ns() + ka->kvmclock_offset;
2039 	}
2040 
2041 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2042 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2043 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2044 
2045 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2046 	get_cpu();
2047 
2048 	if (__this_cpu_read(cpu_tsc_khz)) {
2049 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2050 				   &hv_clock.tsc_shift,
2051 				   &hv_clock.tsc_to_system_mul);
2052 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2053 	} else
2054 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2055 
2056 	put_cpu();
2057 
2058 	return ret;
2059 }
2060 
2061 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2062 {
2063 	struct kvm_vcpu_arch *vcpu = &v->arch;
2064 	struct pvclock_vcpu_time_info guest_hv_clock;
2065 
2066 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2067 		&guest_hv_clock, sizeof(guest_hv_clock))))
2068 		return;
2069 
2070 	/* This VCPU is paused, but it's legal for a guest to read another
2071 	 * VCPU's kvmclock, so we really have to follow the specification where
2072 	 * it says that version is odd if data is being modified, and even after
2073 	 * it is consistent.
2074 	 *
2075 	 * Version field updates must be kept separate.  This is because
2076 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2077 	 * writes within a string instruction are weakly ordered.  So there
2078 	 * are three writes overall.
2079 	 *
2080 	 * As a small optimization, only write the version field in the first
2081 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2082 	 * version field is the first in the struct.
2083 	 */
2084 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2085 
2086 	if (guest_hv_clock.version & 1)
2087 		++guest_hv_clock.version;  /* first time write, random junk */
2088 
2089 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2090 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2091 				&vcpu->hv_clock,
2092 				sizeof(vcpu->hv_clock.version));
2093 
2094 	smp_wmb();
2095 
2096 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2097 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2098 
2099 	if (vcpu->pvclock_set_guest_stopped_request) {
2100 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2101 		vcpu->pvclock_set_guest_stopped_request = false;
2102 	}
2103 
2104 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2105 
2106 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2107 				&vcpu->hv_clock,
2108 				sizeof(vcpu->hv_clock));
2109 
2110 	smp_wmb();
2111 
2112 	vcpu->hv_clock.version++;
2113 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2114 				&vcpu->hv_clock,
2115 				sizeof(vcpu->hv_clock.version));
2116 }
2117 
2118 static int kvm_guest_time_update(struct kvm_vcpu *v)
2119 {
2120 	unsigned long flags, tgt_tsc_khz;
2121 	struct kvm_vcpu_arch *vcpu = &v->arch;
2122 	struct kvm_arch *ka = &v->kvm->arch;
2123 	s64 kernel_ns;
2124 	u64 tsc_timestamp, host_tsc;
2125 	u8 pvclock_flags;
2126 	bool use_master_clock;
2127 
2128 	kernel_ns = 0;
2129 	host_tsc = 0;
2130 
2131 	/*
2132 	 * If the host uses TSC clock, then passthrough TSC as stable
2133 	 * to the guest.
2134 	 */
2135 	spin_lock(&ka->pvclock_gtod_sync_lock);
2136 	use_master_clock = ka->use_master_clock;
2137 	if (use_master_clock) {
2138 		host_tsc = ka->master_cycle_now;
2139 		kernel_ns = ka->master_kernel_ns;
2140 	}
2141 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2142 
2143 	/* Keep irq disabled to prevent changes to the clock */
2144 	local_irq_save(flags);
2145 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2146 	if (unlikely(tgt_tsc_khz == 0)) {
2147 		local_irq_restore(flags);
2148 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2149 		return 1;
2150 	}
2151 	if (!use_master_clock) {
2152 		host_tsc = rdtsc();
2153 		kernel_ns = ktime_get_boot_ns();
2154 	}
2155 
2156 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2157 
2158 	/*
2159 	 * We may have to catch up the TSC to match elapsed wall clock
2160 	 * time for two reasons, even if kvmclock is used.
2161 	 *   1) CPU could have been running below the maximum TSC rate
2162 	 *   2) Broken TSC compensation resets the base at each VCPU
2163 	 *      entry to avoid unknown leaps of TSC even when running
2164 	 *      again on the same CPU.  This may cause apparent elapsed
2165 	 *      time to disappear, and the guest to stand still or run
2166 	 *	very slowly.
2167 	 */
2168 	if (vcpu->tsc_catchup) {
2169 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2170 		if (tsc > tsc_timestamp) {
2171 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2172 			tsc_timestamp = tsc;
2173 		}
2174 	}
2175 
2176 	local_irq_restore(flags);
2177 
2178 	/* With all the info we got, fill in the values */
2179 
2180 	if (kvm_has_tsc_control)
2181 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2182 
2183 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2184 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2185 				   &vcpu->hv_clock.tsc_shift,
2186 				   &vcpu->hv_clock.tsc_to_system_mul);
2187 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2188 	}
2189 
2190 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2191 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2192 	vcpu->last_guest_tsc = tsc_timestamp;
2193 
2194 	/* If the host uses TSC clocksource, then it is stable */
2195 	pvclock_flags = 0;
2196 	if (use_master_clock)
2197 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2198 
2199 	vcpu->hv_clock.flags = pvclock_flags;
2200 
2201 	if (vcpu->pv_time_enabled)
2202 		kvm_setup_pvclock_page(v);
2203 	if (v == kvm_get_vcpu(v->kvm, 0))
2204 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2205 	return 0;
2206 }
2207 
2208 /*
2209  * kvmclock updates which are isolated to a given vcpu, such as
2210  * vcpu->cpu migration, should not allow system_timestamp from
2211  * the rest of the vcpus to remain static. Otherwise ntp frequency
2212  * correction applies to one vcpu's system_timestamp but not
2213  * the others.
2214  *
2215  * So in those cases, request a kvmclock update for all vcpus.
2216  * We need to rate-limit these requests though, as they can
2217  * considerably slow guests that have a large number of vcpus.
2218  * The time for a remote vcpu to update its kvmclock is bound
2219  * by the delay we use to rate-limit the updates.
2220  */
2221 
2222 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2223 
2224 static void kvmclock_update_fn(struct work_struct *work)
2225 {
2226 	int i;
2227 	struct delayed_work *dwork = to_delayed_work(work);
2228 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2229 					   kvmclock_update_work);
2230 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2231 	struct kvm_vcpu *vcpu;
2232 
2233 	kvm_for_each_vcpu(i, vcpu, kvm) {
2234 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2235 		kvm_vcpu_kick(vcpu);
2236 	}
2237 }
2238 
2239 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2240 {
2241 	struct kvm *kvm = v->kvm;
2242 
2243 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2244 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2245 					KVMCLOCK_UPDATE_DELAY);
2246 }
2247 
2248 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2249 
2250 static void kvmclock_sync_fn(struct work_struct *work)
2251 {
2252 	struct delayed_work *dwork = to_delayed_work(work);
2253 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2254 					   kvmclock_sync_work);
2255 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2256 
2257 	if (!kvmclock_periodic_sync)
2258 		return;
2259 
2260 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2261 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2262 					KVMCLOCK_SYNC_PERIOD);
2263 }
2264 
2265 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2266 {
2267 	u64 mcg_cap = vcpu->arch.mcg_cap;
2268 	unsigned bank_num = mcg_cap & 0xff;
2269 	u32 msr = msr_info->index;
2270 	u64 data = msr_info->data;
2271 
2272 	switch (msr) {
2273 	case MSR_IA32_MCG_STATUS:
2274 		vcpu->arch.mcg_status = data;
2275 		break;
2276 	case MSR_IA32_MCG_CTL:
2277 		if (!(mcg_cap & MCG_CTL_P) &&
2278 		    (data || !msr_info->host_initiated))
2279 			return 1;
2280 		if (data != 0 && data != ~(u64)0)
2281 			return 1;
2282 		vcpu->arch.mcg_ctl = data;
2283 		break;
2284 	default:
2285 		if (msr >= MSR_IA32_MC0_CTL &&
2286 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2287 			u32 offset = msr - MSR_IA32_MC0_CTL;
2288 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2289 			 * some Linux kernels though clear bit 10 in bank 4 to
2290 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2291 			 * this to avoid an uncatched #GP in the guest
2292 			 */
2293 			if ((offset & 0x3) == 0 &&
2294 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2295 				return -1;
2296 			if (!msr_info->host_initiated &&
2297 				(offset & 0x3) == 1 && data != 0)
2298 				return -1;
2299 			vcpu->arch.mce_banks[offset] = data;
2300 			break;
2301 		}
2302 		return 1;
2303 	}
2304 	return 0;
2305 }
2306 
2307 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2308 {
2309 	struct kvm *kvm = vcpu->kvm;
2310 	int lm = is_long_mode(vcpu);
2311 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2312 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2313 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2314 		: kvm->arch.xen_hvm_config.blob_size_32;
2315 	u32 page_num = data & ~PAGE_MASK;
2316 	u64 page_addr = data & PAGE_MASK;
2317 	u8 *page;
2318 	int r;
2319 
2320 	r = -E2BIG;
2321 	if (page_num >= blob_size)
2322 		goto out;
2323 	r = -ENOMEM;
2324 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2325 	if (IS_ERR(page)) {
2326 		r = PTR_ERR(page);
2327 		goto out;
2328 	}
2329 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2330 		goto out_free;
2331 	r = 0;
2332 out_free:
2333 	kfree(page);
2334 out:
2335 	return r;
2336 }
2337 
2338 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2339 {
2340 	gpa_t gpa = data & ~0x3f;
2341 
2342 	/* Bits 3:5 are reserved, Should be zero */
2343 	if (data & 0x38)
2344 		return 1;
2345 
2346 	vcpu->arch.apf.msr_val = data;
2347 
2348 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2349 		kvm_clear_async_pf_completion_queue(vcpu);
2350 		kvm_async_pf_hash_reset(vcpu);
2351 		return 0;
2352 	}
2353 
2354 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2355 					sizeof(u32)))
2356 		return 1;
2357 
2358 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2359 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2360 	kvm_async_pf_wakeup_all(vcpu);
2361 	return 0;
2362 }
2363 
2364 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2365 {
2366 	vcpu->arch.pv_time_enabled = false;
2367 }
2368 
2369 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2370 {
2371 	++vcpu->stat.tlb_flush;
2372 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2373 }
2374 
2375 static void record_steal_time(struct kvm_vcpu *vcpu)
2376 {
2377 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2378 		return;
2379 
2380 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2381 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2382 		return;
2383 
2384 	/*
2385 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2386 	 * expensive IPIs.
2387 	 */
2388 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2389 		kvm_vcpu_flush_tlb(vcpu, false);
2390 
2391 	if (vcpu->arch.st.steal.version & 1)
2392 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2393 
2394 	vcpu->arch.st.steal.version += 1;
2395 
2396 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2397 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2398 
2399 	smp_wmb();
2400 
2401 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2402 		vcpu->arch.st.last_steal;
2403 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2404 
2405 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2406 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2407 
2408 	smp_wmb();
2409 
2410 	vcpu->arch.st.steal.version += 1;
2411 
2412 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2413 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2414 }
2415 
2416 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2417 {
2418 	bool pr = false;
2419 	u32 msr = msr_info->index;
2420 	u64 data = msr_info->data;
2421 
2422 	switch (msr) {
2423 	case MSR_AMD64_NB_CFG:
2424 	case MSR_IA32_UCODE_WRITE:
2425 	case MSR_VM_HSAVE_PA:
2426 	case MSR_AMD64_PATCH_LOADER:
2427 	case MSR_AMD64_BU_CFG2:
2428 	case MSR_AMD64_DC_CFG:
2429 		break;
2430 
2431 	case MSR_IA32_UCODE_REV:
2432 		if (msr_info->host_initiated)
2433 			vcpu->arch.microcode_version = data;
2434 		break;
2435 	case MSR_EFER:
2436 		return set_efer(vcpu, data);
2437 	case MSR_K7_HWCR:
2438 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2439 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2440 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2441 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2442 		if (data != 0) {
2443 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2444 				    data);
2445 			return 1;
2446 		}
2447 		break;
2448 	case MSR_FAM10H_MMIO_CONF_BASE:
2449 		if (data != 0) {
2450 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2451 				    "0x%llx\n", data);
2452 			return 1;
2453 		}
2454 		break;
2455 	case MSR_IA32_DEBUGCTLMSR:
2456 		if (!data) {
2457 			/* We support the non-activated case already */
2458 			break;
2459 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2460 			/* Values other than LBR and BTF are vendor-specific,
2461 			   thus reserved and should throw a #GP */
2462 			return 1;
2463 		}
2464 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2465 			    __func__, data);
2466 		break;
2467 	case 0x200 ... 0x2ff:
2468 		return kvm_mtrr_set_msr(vcpu, msr, data);
2469 	case MSR_IA32_APICBASE:
2470 		return kvm_set_apic_base(vcpu, msr_info);
2471 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2472 		return kvm_x2apic_msr_write(vcpu, msr, data);
2473 	case MSR_IA32_TSCDEADLINE:
2474 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2475 		break;
2476 	case MSR_IA32_TSC_ADJUST:
2477 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2478 			if (!msr_info->host_initiated) {
2479 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2480 				adjust_tsc_offset_guest(vcpu, adj);
2481 			}
2482 			vcpu->arch.ia32_tsc_adjust_msr = data;
2483 		}
2484 		break;
2485 	case MSR_IA32_MISC_ENABLE:
2486 		vcpu->arch.ia32_misc_enable_msr = data;
2487 		break;
2488 	case MSR_IA32_SMBASE:
2489 		if (!msr_info->host_initiated)
2490 			return 1;
2491 		vcpu->arch.smbase = data;
2492 		break;
2493 	case MSR_IA32_TSC:
2494 		kvm_write_tsc(vcpu, msr_info);
2495 		break;
2496 	case MSR_SMI_COUNT:
2497 		if (!msr_info->host_initiated)
2498 			return 1;
2499 		vcpu->arch.smi_count = data;
2500 		break;
2501 	case MSR_KVM_WALL_CLOCK_NEW:
2502 	case MSR_KVM_WALL_CLOCK:
2503 		vcpu->kvm->arch.wall_clock = data;
2504 		kvm_write_wall_clock(vcpu->kvm, data);
2505 		break;
2506 	case MSR_KVM_SYSTEM_TIME_NEW:
2507 	case MSR_KVM_SYSTEM_TIME: {
2508 		struct kvm_arch *ka = &vcpu->kvm->arch;
2509 
2510 		kvmclock_reset(vcpu);
2511 
2512 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2513 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2514 
2515 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2516 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2517 
2518 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2519 		}
2520 
2521 		vcpu->arch.time = data;
2522 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2523 
2524 		/* we verify if the enable bit is set... */
2525 		if (!(data & 1))
2526 			break;
2527 
2528 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2529 		     &vcpu->arch.pv_time, data & ~1ULL,
2530 		     sizeof(struct pvclock_vcpu_time_info)))
2531 			vcpu->arch.pv_time_enabled = false;
2532 		else
2533 			vcpu->arch.pv_time_enabled = true;
2534 
2535 		break;
2536 	}
2537 	case MSR_KVM_ASYNC_PF_EN:
2538 		if (kvm_pv_enable_async_pf(vcpu, data))
2539 			return 1;
2540 		break;
2541 	case MSR_KVM_STEAL_TIME:
2542 
2543 		if (unlikely(!sched_info_on()))
2544 			return 1;
2545 
2546 		if (data & KVM_STEAL_RESERVED_MASK)
2547 			return 1;
2548 
2549 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2550 						data & KVM_STEAL_VALID_BITS,
2551 						sizeof(struct kvm_steal_time)))
2552 			return 1;
2553 
2554 		vcpu->arch.st.msr_val = data;
2555 
2556 		if (!(data & KVM_MSR_ENABLED))
2557 			break;
2558 
2559 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2560 
2561 		break;
2562 	case MSR_KVM_PV_EOI_EN:
2563 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2564 			return 1;
2565 		break;
2566 
2567 	case MSR_IA32_MCG_CTL:
2568 	case MSR_IA32_MCG_STATUS:
2569 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2570 		return set_msr_mce(vcpu, msr_info);
2571 
2572 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2573 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2574 		pr = true; /* fall through */
2575 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2576 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2577 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2578 			return kvm_pmu_set_msr(vcpu, msr_info);
2579 
2580 		if (pr || data != 0)
2581 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2582 				    "0x%x data 0x%llx\n", msr, data);
2583 		break;
2584 	case MSR_K7_CLK_CTL:
2585 		/*
2586 		 * Ignore all writes to this no longer documented MSR.
2587 		 * Writes are only relevant for old K7 processors,
2588 		 * all pre-dating SVM, but a recommended workaround from
2589 		 * AMD for these chips. It is possible to specify the
2590 		 * affected processor models on the command line, hence
2591 		 * the need to ignore the workaround.
2592 		 */
2593 		break;
2594 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2595 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2596 	case HV_X64_MSR_CRASH_CTL:
2597 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2598 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2599 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2600 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2601 		return kvm_hv_set_msr_common(vcpu, msr, data,
2602 					     msr_info->host_initiated);
2603 	case MSR_IA32_BBL_CR_CTL3:
2604 		/* Drop writes to this legacy MSR -- see rdmsr
2605 		 * counterpart for further detail.
2606 		 */
2607 		if (report_ignored_msrs)
2608 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2609 				msr, data);
2610 		break;
2611 	case MSR_AMD64_OSVW_ID_LENGTH:
2612 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2613 			return 1;
2614 		vcpu->arch.osvw.length = data;
2615 		break;
2616 	case MSR_AMD64_OSVW_STATUS:
2617 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2618 			return 1;
2619 		vcpu->arch.osvw.status = data;
2620 		break;
2621 	case MSR_PLATFORM_INFO:
2622 		if (!msr_info->host_initiated ||
2623 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2624 		     cpuid_fault_enabled(vcpu)))
2625 			return 1;
2626 		vcpu->arch.msr_platform_info = data;
2627 		break;
2628 	case MSR_MISC_FEATURES_ENABLES:
2629 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2630 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2631 		     !supports_cpuid_fault(vcpu)))
2632 			return 1;
2633 		vcpu->arch.msr_misc_features_enables = data;
2634 		break;
2635 	default:
2636 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2637 			return xen_hvm_config(vcpu, data);
2638 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2639 			return kvm_pmu_set_msr(vcpu, msr_info);
2640 		if (!ignore_msrs) {
2641 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2642 				    msr, data);
2643 			return 1;
2644 		} else {
2645 			if (report_ignored_msrs)
2646 				vcpu_unimpl(vcpu,
2647 					"ignored wrmsr: 0x%x data 0x%llx\n",
2648 					msr, data);
2649 			break;
2650 		}
2651 	}
2652 	return 0;
2653 }
2654 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2655 
2656 
2657 /*
2658  * Reads an msr value (of 'msr_index') into 'pdata'.
2659  * Returns 0 on success, non-0 otherwise.
2660  * Assumes vcpu_load() was already called.
2661  */
2662 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2663 {
2664 	return kvm_x86_ops->get_msr(vcpu, msr);
2665 }
2666 EXPORT_SYMBOL_GPL(kvm_get_msr);
2667 
2668 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2669 {
2670 	u64 data;
2671 	u64 mcg_cap = vcpu->arch.mcg_cap;
2672 	unsigned bank_num = mcg_cap & 0xff;
2673 
2674 	switch (msr) {
2675 	case MSR_IA32_P5_MC_ADDR:
2676 	case MSR_IA32_P5_MC_TYPE:
2677 		data = 0;
2678 		break;
2679 	case MSR_IA32_MCG_CAP:
2680 		data = vcpu->arch.mcg_cap;
2681 		break;
2682 	case MSR_IA32_MCG_CTL:
2683 		if (!(mcg_cap & MCG_CTL_P) && !host)
2684 			return 1;
2685 		data = vcpu->arch.mcg_ctl;
2686 		break;
2687 	case MSR_IA32_MCG_STATUS:
2688 		data = vcpu->arch.mcg_status;
2689 		break;
2690 	default:
2691 		if (msr >= MSR_IA32_MC0_CTL &&
2692 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2693 			u32 offset = msr - MSR_IA32_MC0_CTL;
2694 			data = vcpu->arch.mce_banks[offset];
2695 			break;
2696 		}
2697 		return 1;
2698 	}
2699 	*pdata = data;
2700 	return 0;
2701 }
2702 
2703 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2704 {
2705 	switch (msr_info->index) {
2706 	case MSR_IA32_PLATFORM_ID:
2707 	case MSR_IA32_EBL_CR_POWERON:
2708 	case MSR_IA32_DEBUGCTLMSR:
2709 	case MSR_IA32_LASTBRANCHFROMIP:
2710 	case MSR_IA32_LASTBRANCHTOIP:
2711 	case MSR_IA32_LASTINTFROMIP:
2712 	case MSR_IA32_LASTINTTOIP:
2713 	case MSR_K8_SYSCFG:
2714 	case MSR_K8_TSEG_ADDR:
2715 	case MSR_K8_TSEG_MASK:
2716 	case MSR_K7_HWCR:
2717 	case MSR_VM_HSAVE_PA:
2718 	case MSR_K8_INT_PENDING_MSG:
2719 	case MSR_AMD64_NB_CFG:
2720 	case MSR_FAM10H_MMIO_CONF_BASE:
2721 	case MSR_AMD64_BU_CFG2:
2722 	case MSR_IA32_PERF_CTL:
2723 	case MSR_AMD64_DC_CFG:
2724 		msr_info->data = 0;
2725 		break;
2726 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2727 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2728 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2729 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2730 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2731 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2732 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2733 		msr_info->data = 0;
2734 		break;
2735 	case MSR_IA32_UCODE_REV:
2736 		msr_info->data = vcpu->arch.microcode_version;
2737 		break;
2738 	case MSR_IA32_TSC:
2739 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2740 		break;
2741 	case MSR_MTRRcap:
2742 	case 0x200 ... 0x2ff:
2743 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2744 	case 0xcd: /* fsb frequency */
2745 		msr_info->data = 3;
2746 		break;
2747 		/*
2748 		 * MSR_EBC_FREQUENCY_ID
2749 		 * Conservative value valid for even the basic CPU models.
2750 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2751 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2752 		 * and 266MHz for model 3, or 4. Set Core Clock
2753 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2754 		 * 31:24) even though these are only valid for CPU
2755 		 * models > 2, however guests may end up dividing or
2756 		 * multiplying by zero otherwise.
2757 		 */
2758 	case MSR_EBC_FREQUENCY_ID:
2759 		msr_info->data = 1 << 24;
2760 		break;
2761 	case MSR_IA32_APICBASE:
2762 		msr_info->data = kvm_get_apic_base(vcpu);
2763 		break;
2764 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2765 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2766 		break;
2767 	case MSR_IA32_TSCDEADLINE:
2768 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2769 		break;
2770 	case MSR_IA32_TSC_ADJUST:
2771 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2772 		break;
2773 	case MSR_IA32_MISC_ENABLE:
2774 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2775 		break;
2776 	case MSR_IA32_SMBASE:
2777 		if (!msr_info->host_initiated)
2778 			return 1;
2779 		msr_info->data = vcpu->arch.smbase;
2780 		break;
2781 	case MSR_SMI_COUNT:
2782 		msr_info->data = vcpu->arch.smi_count;
2783 		break;
2784 	case MSR_IA32_PERF_STATUS:
2785 		/* TSC increment by tick */
2786 		msr_info->data = 1000ULL;
2787 		/* CPU multiplier */
2788 		msr_info->data |= (((uint64_t)4ULL) << 40);
2789 		break;
2790 	case MSR_EFER:
2791 		msr_info->data = vcpu->arch.efer;
2792 		break;
2793 	case MSR_KVM_WALL_CLOCK:
2794 	case MSR_KVM_WALL_CLOCK_NEW:
2795 		msr_info->data = vcpu->kvm->arch.wall_clock;
2796 		break;
2797 	case MSR_KVM_SYSTEM_TIME:
2798 	case MSR_KVM_SYSTEM_TIME_NEW:
2799 		msr_info->data = vcpu->arch.time;
2800 		break;
2801 	case MSR_KVM_ASYNC_PF_EN:
2802 		msr_info->data = vcpu->arch.apf.msr_val;
2803 		break;
2804 	case MSR_KVM_STEAL_TIME:
2805 		msr_info->data = vcpu->arch.st.msr_val;
2806 		break;
2807 	case MSR_KVM_PV_EOI_EN:
2808 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2809 		break;
2810 	case MSR_IA32_P5_MC_ADDR:
2811 	case MSR_IA32_P5_MC_TYPE:
2812 	case MSR_IA32_MCG_CAP:
2813 	case MSR_IA32_MCG_CTL:
2814 	case MSR_IA32_MCG_STATUS:
2815 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2816 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2817 				   msr_info->host_initiated);
2818 	case MSR_K7_CLK_CTL:
2819 		/*
2820 		 * Provide expected ramp-up count for K7. All other
2821 		 * are set to zero, indicating minimum divisors for
2822 		 * every field.
2823 		 *
2824 		 * This prevents guest kernels on AMD host with CPU
2825 		 * type 6, model 8 and higher from exploding due to
2826 		 * the rdmsr failing.
2827 		 */
2828 		msr_info->data = 0x20000000;
2829 		break;
2830 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2831 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2832 	case HV_X64_MSR_CRASH_CTL:
2833 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2834 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2835 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2836 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2837 		return kvm_hv_get_msr_common(vcpu,
2838 					     msr_info->index, &msr_info->data,
2839 					     msr_info->host_initiated);
2840 		break;
2841 	case MSR_IA32_BBL_CR_CTL3:
2842 		/* This legacy MSR exists but isn't fully documented in current
2843 		 * silicon.  It is however accessed by winxp in very narrow
2844 		 * scenarios where it sets bit #19, itself documented as
2845 		 * a "reserved" bit.  Best effort attempt to source coherent
2846 		 * read data here should the balance of the register be
2847 		 * interpreted by the guest:
2848 		 *
2849 		 * L2 cache control register 3: 64GB range, 256KB size,
2850 		 * enabled, latency 0x1, configured
2851 		 */
2852 		msr_info->data = 0xbe702111;
2853 		break;
2854 	case MSR_AMD64_OSVW_ID_LENGTH:
2855 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2856 			return 1;
2857 		msr_info->data = vcpu->arch.osvw.length;
2858 		break;
2859 	case MSR_AMD64_OSVW_STATUS:
2860 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2861 			return 1;
2862 		msr_info->data = vcpu->arch.osvw.status;
2863 		break;
2864 	case MSR_PLATFORM_INFO:
2865 		if (!msr_info->host_initiated &&
2866 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2867 			return 1;
2868 		msr_info->data = vcpu->arch.msr_platform_info;
2869 		break;
2870 	case MSR_MISC_FEATURES_ENABLES:
2871 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2872 		break;
2873 	default:
2874 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2875 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2876 		if (!ignore_msrs) {
2877 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2878 					       msr_info->index);
2879 			return 1;
2880 		} else {
2881 			if (report_ignored_msrs)
2882 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2883 					msr_info->index);
2884 			msr_info->data = 0;
2885 		}
2886 		break;
2887 	}
2888 	return 0;
2889 }
2890 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2891 
2892 /*
2893  * Read or write a bunch of msrs. All parameters are kernel addresses.
2894  *
2895  * @return number of msrs set successfully.
2896  */
2897 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2898 		    struct kvm_msr_entry *entries,
2899 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2900 				  unsigned index, u64 *data))
2901 {
2902 	int i;
2903 
2904 	for (i = 0; i < msrs->nmsrs; ++i)
2905 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2906 			break;
2907 
2908 	return i;
2909 }
2910 
2911 /*
2912  * Read or write a bunch of msrs. Parameters are user addresses.
2913  *
2914  * @return number of msrs set successfully.
2915  */
2916 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2917 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2918 				unsigned index, u64 *data),
2919 		  int writeback)
2920 {
2921 	struct kvm_msrs msrs;
2922 	struct kvm_msr_entry *entries;
2923 	int r, n;
2924 	unsigned size;
2925 
2926 	r = -EFAULT;
2927 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2928 		goto out;
2929 
2930 	r = -E2BIG;
2931 	if (msrs.nmsrs >= MAX_IO_MSRS)
2932 		goto out;
2933 
2934 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2935 	entries = memdup_user(user_msrs->entries, size);
2936 	if (IS_ERR(entries)) {
2937 		r = PTR_ERR(entries);
2938 		goto out;
2939 	}
2940 
2941 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2942 	if (r < 0)
2943 		goto out_free;
2944 
2945 	r = -EFAULT;
2946 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2947 		goto out_free;
2948 
2949 	r = n;
2950 
2951 out_free:
2952 	kfree(entries);
2953 out:
2954 	return r;
2955 }
2956 
2957 static inline bool kvm_can_mwait_in_guest(void)
2958 {
2959 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2960 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2961 		boot_cpu_has(X86_FEATURE_ARAT);
2962 }
2963 
2964 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2965 {
2966 	int r = 0;
2967 
2968 	switch (ext) {
2969 	case KVM_CAP_IRQCHIP:
2970 	case KVM_CAP_HLT:
2971 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2972 	case KVM_CAP_SET_TSS_ADDR:
2973 	case KVM_CAP_EXT_CPUID:
2974 	case KVM_CAP_EXT_EMUL_CPUID:
2975 	case KVM_CAP_CLOCKSOURCE:
2976 	case KVM_CAP_PIT:
2977 	case KVM_CAP_NOP_IO_DELAY:
2978 	case KVM_CAP_MP_STATE:
2979 	case KVM_CAP_SYNC_MMU:
2980 	case KVM_CAP_USER_NMI:
2981 	case KVM_CAP_REINJECT_CONTROL:
2982 	case KVM_CAP_IRQ_INJECT_STATUS:
2983 	case KVM_CAP_IOEVENTFD:
2984 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2985 	case KVM_CAP_PIT2:
2986 	case KVM_CAP_PIT_STATE2:
2987 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2988 	case KVM_CAP_XEN_HVM:
2989 	case KVM_CAP_VCPU_EVENTS:
2990 	case KVM_CAP_HYPERV:
2991 	case KVM_CAP_HYPERV_VAPIC:
2992 	case KVM_CAP_HYPERV_SPIN:
2993 	case KVM_CAP_HYPERV_SYNIC:
2994 	case KVM_CAP_HYPERV_SYNIC2:
2995 	case KVM_CAP_HYPERV_VP_INDEX:
2996 	case KVM_CAP_HYPERV_EVENTFD:
2997 	case KVM_CAP_HYPERV_TLBFLUSH:
2998 	case KVM_CAP_HYPERV_SEND_IPI:
2999 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3000 	case KVM_CAP_PCI_SEGMENT:
3001 	case KVM_CAP_DEBUGREGS:
3002 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3003 	case KVM_CAP_XSAVE:
3004 	case KVM_CAP_ASYNC_PF:
3005 	case KVM_CAP_GET_TSC_KHZ:
3006 	case KVM_CAP_KVMCLOCK_CTRL:
3007 	case KVM_CAP_READONLY_MEM:
3008 	case KVM_CAP_HYPERV_TIME:
3009 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3010 	case KVM_CAP_TSC_DEADLINE_TIMER:
3011 	case KVM_CAP_ENABLE_CAP_VM:
3012 	case KVM_CAP_DISABLE_QUIRKS:
3013 	case KVM_CAP_SET_BOOT_CPU_ID:
3014  	case KVM_CAP_SPLIT_IRQCHIP:
3015 	case KVM_CAP_IMMEDIATE_EXIT:
3016 	case KVM_CAP_GET_MSR_FEATURES:
3017 	case KVM_CAP_MSR_PLATFORM_INFO:
3018 	case KVM_CAP_EXCEPTION_PAYLOAD:
3019 		r = 1;
3020 		break;
3021 	case KVM_CAP_SYNC_REGS:
3022 		r = KVM_SYNC_X86_VALID_FIELDS;
3023 		break;
3024 	case KVM_CAP_ADJUST_CLOCK:
3025 		r = KVM_CLOCK_TSC_STABLE;
3026 		break;
3027 	case KVM_CAP_X86_DISABLE_EXITS:
3028 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3029 		if(kvm_can_mwait_in_guest())
3030 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3031 		break;
3032 	case KVM_CAP_X86_SMM:
3033 		/* SMBASE is usually relocated above 1M on modern chipsets,
3034 		 * and SMM handlers might indeed rely on 4G segment limits,
3035 		 * so do not report SMM to be available if real mode is
3036 		 * emulated via vm86 mode.  Still, do not go to great lengths
3037 		 * to avoid userspace's usage of the feature, because it is a
3038 		 * fringe case that is not enabled except via specific settings
3039 		 * of the module parameters.
3040 		 */
3041 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3042 		break;
3043 	case KVM_CAP_VAPIC:
3044 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3045 		break;
3046 	case KVM_CAP_NR_VCPUS:
3047 		r = KVM_SOFT_MAX_VCPUS;
3048 		break;
3049 	case KVM_CAP_MAX_VCPUS:
3050 		r = KVM_MAX_VCPUS;
3051 		break;
3052 	case KVM_CAP_NR_MEMSLOTS:
3053 		r = KVM_USER_MEM_SLOTS;
3054 		break;
3055 	case KVM_CAP_PV_MMU:	/* obsolete */
3056 		r = 0;
3057 		break;
3058 	case KVM_CAP_MCE:
3059 		r = KVM_MAX_MCE_BANKS;
3060 		break;
3061 	case KVM_CAP_XCRS:
3062 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3063 		break;
3064 	case KVM_CAP_TSC_CONTROL:
3065 		r = kvm_has_tsc_control;
3066 		break;
3067 	case KVM_CAP_X2APIC_API:
3068 		r = KVM_X2APIC_API_VALID_FLAGS;
3069 		break;
3070 	case KVM_CAP_NESTED_STATE:
3071 		r = kvm_x86_ops->get_nested_state ?
3072 			kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3073 		break;
3074 	default:
3075 		break;
3076 	}
3077 	return r;
3078 
3079 }
3080 
3081 long kvm_arch_dev_ioctl(struct file *filp,
3082 			unsigned int ioctl, unsigned long arg)
3083 {
3084 	void __user *argp = (void __user *)arg;
3085 	long r;
3086 
3087 	switch (ioctl) {
3088 	case KVM_GET_MSR_INDEX_LIST: {
3089 		struct kvm_msr_list __user *user_msr_list = argp;
3090 		struct kvm_msr_list msr_list;
3091 		unsigned n;
3092 
3093 		r = -EFAULT;
3094 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3095 			goto out;
3096 		n = msr_list.nmsrs;
3097 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3098 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3099 			goto out;
3100 		r = -E2BIG;
3101 		if (n < msr_list.nmsrs)
3102 			goto out;
3103 		r = -EFAULT;
3104 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3105 				 num_msrs_to_save * sizeof(u32)))
3106 			goto out;
3107 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3108 				 &emulated_msrs,
3109 				 num_emulated_msrs * sizeof(u32)))
3110 			goto out;
3111 		r = 0;
3112 		break;
3113 	}
3114 	case KVM_GET_SUPPORTED_CPUID:
3115 	case KVM_GET_EMULATED_CPUID: {
3116 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3117 		struct kvm_cpuid2 cpuid;
3118 
3119 		r = -EFAULT;
3120 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3121 			goto out;
3122 
3123 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3124 					    ioctl);
3125 		if (r)
3126 			goto out;
3127 
3128 		r = -EFAULT;
3129 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3130 			goto out;
3131 		r = 0;
3132 		break;
3133 	}
3134 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3135 		r = -EFAULT;
3136 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3137 				 sizeof(kvm_mce_cap_supported)))
3138 			goto out;
3139 		r = 0;
3140 		break;
3141 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3142 		struct kvm_msr_list __user *user_msr_list = argp;
3143 		struct kvm_msr_list msr_list;
3144 		unsigned int n;
3145 
3146 		r = -EFAULT;
3147 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3148 			goto out;
3149 		n = msr_list.nmsrs;
3150 		msr_list.nmsrs = num_msr_based_features;
3151 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3152 			goto out;
3153 		r = -E2BIG;
3154 		if (n < msr_list.nmsrs)
3155 			goto out;
3156 		r = -EFAULT;
3157 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3158 				 num_msr_based_features * sizeof(u32)))
3159 			goto out;
3160 		r = 0;
3161 		break;
3162 	}
3163 	case KVM_GET_MSRS:
3164 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3165 		break;
3166 	}
3167 	default:
3168 		r = -EINVAL;
3169 	}
3170 out:
3171 	return r;
3172 }
3173 
3174 static void wbinvd_ipi(void *garbage)
3175 {
3176 	wbinvd();
3177 }
3178 
3179 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3180 {
3181 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3182 }
3183 
3184 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3185 {
3186 	/* Address WBINVD may be executed by guest */
3187 	if (need_emulate_wbinvd(vcpu)) {
3188 		if (kvm_x86_ops->has_wbinvd_exit())
3189 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3190 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3191 			smp_call_function_single(vcpu->cpu,
3192 					wbinvd_ipi, NULL, 1);
3193 	}
3194 
3195 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3196 
3197 	/* Apply any externally detected TSC adjustments (due to suspend) */
3198 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3199 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3200 		vcpu->arch.tsc_offset_adjustment = 0;
3201 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3202 	}
3203 
3204 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3205 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3206 				rdtsc() - vcpu->arch.last_host_tsc;
3207 		if (tsc_delta < 0)
3208 			mark_tsc_unstable("KVM discovered backwards TSC");
3209 
3210 		if (kvm_check_tsc_unstable()) {
3211 			u64 offset = kvm_compute_tsc_offset(vcpu,
3212 						vcpu->arch.last_guest_tsc);
3213 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3214 			vcpu->arch.tsc_catchup = 1;
3215 		}
3216 
3217 		if (kvm_lapic_hv_timer_in_use(vcpu))
3218 			kvm_lapic_restart_hv_timer(vcpu);
3219 
3220 		/*
3221 		 * On a host with synchronized TSC, there is no need to update
3222 		 * kvmclock on vcpu->cpu migration
3223 		 */
3224 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3225 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3226 		if (vcpu->cpu != cpu)
3227 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3228 		vcpu->cpu = cpu;
3229 	}
3230 
3231 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3232 }
3233 
3234 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3235 {
3236 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3237 		return;
3238 
3239 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3240 
3241 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3242 			&vcpu->arch.st.steal.preempted,
3243 			offsetof(struct kvm_steal_time, preempted),
3244 			sizeof(vcpu->arch.st.steal.preempted));
3245 }
3246 
3247 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3248 {
3249 	int idx;
3250 
3251 	if (vcpu->preempted)
3252 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3253 
3254 	/*
3255 	 * Disable page faults because we're in atomic context here.
3256 	 * kvm_write_guest_offset_cached() would call might_fault()
3257 	 * that relies on pagefault_disable() to tell if there's a
3258 	 * bug. NOTE: the write to guest memory may not go through if
3259 	 * during postcopy live migration or if there's heavy guest
3260 	 * paging.
3261 	 */
3262 	pagefault_disable();
3263 	/*
3264 	 * kvm_memslots() will be called by
3265 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3266 	 */
3267 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3268 	kvm_steal_time_set_preempted(vcpu);
3269 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3270 	pagefault_enable();
3271 	kvm_x86_ops->vcpu_put(vcpu);
3272 	vcpu->arch.last_host_tsc = rdtsc();
3273 	/*
3274 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3275 	 * on every vmexit, but if not, we might have a stale dr6 from the
3276 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3277 	 */
3278 	set_debugreg(0, 6);
3279 }
3280 
3281 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3282 				    struct kvm_lapic_state *s)
3283 {
3284 	if (vcpu->arch.apicv_active)
3285 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3286 
3287 	return kvm_apic_get_state(vcpu, s);
3288 }
3289 
3290 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3291 				    struct kvm_lapic_state *s)
3292 {
3293 	int r;
3294 
3295 	r = kvm_apic_set_state(vcpu, s);
3296 	if (r)
3297 		return r;
3298 	update_cr8_intercept(vcpu);
3299 
3300 	return 0;
3301 }
3302 
3303 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3304 {
3305 	return (!lapic_in_kernel(vcpu) ||
3306 		kvm_apic_accept_pic_intr(vcpu));
3307 }
3308 
3309 /*
3310  * if userspace requested an interrupt window, check that the
3311  * interrupt window is open.
3312  *
3313  * No need to exit to userspace if we already have an interrupt queued.
3314  */
3315 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3316 {
3317 	return kvm_arch_interrupt_allowed(vcpu) &&
3318 		!kvm_cpu_has_interrupt(vcpu) &&
3319 		!kvm_event_needs_reinjection(vcpu) &&
3320 		kvm_cpu_accept_dm_intr(vcpu);
3321 }
3322 
3323 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3324 				    struct kvm_interrupt *irq)
3325 {
3326 	if (irq->irq >= KVM_NR_INTERRUPTS)
3327 		return -EINVAL;
3328 
3329 	if (!irqchip_in_kernel(vcpu->kvm)) {
3330 		kvm_queue_interrupt(vcpu, irq->irq, false);
3331 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3332 		return 0;
3333 	}
3334 
3335 	/*
3336 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3337 	 * fail for in-kernel 8259.
3338 	 */
3339 	if (pic_in_kernel(vcpu->kvm))
3340 		return -ENXIO;
3341 
3342 	if (vcpu->arch.pending_external_vector != -1)
3343 		return -EEXIST;
3344 
3345 	vcpu->arch.pending_external_vector = irq->irq;
3346 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3347 	return 0;
3348 }
3349 
3350 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3351 {
3352 	kvm_inject_nmi(vcpu);
3353 
3354 	return 0;
3355 }
3356 
3357 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3358 {
3359 	kvm_make_request(KVM_REQ_SMI, vcpu);
3360 
3361 	return 0;
3362 }
3363 
3364 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3365 					   struct kvm_tpr_access_ctl *tac)
3366 {
3367 	if (tac->flags)
3368 		return -EINVAL;
3369 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3370 	return 0;
3371 }
3372 
3373 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3374 					u64 mcg_cap)
3375 {
3376 	int r;
3377 	unsigned bank_num = mcg_cap & 0xff, bank;
3378 
3379 	r = -EINVAL;
3380 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3381 		goto out;
3382 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3383 		goto out;
3384 	r = 0;
3385 	vcpu->arch.mcg_cap = mcg_cap;
3386 	/* Init IA32_MCG_CTL to all 1s */
3387 	if (mcg_cap & MCG_CTL_P)
3388 		vcpu->arch.mcg_ctl = ~(u64)0;
3389 	/* Init IA32_MCi_CTL to all 1s */
3390 	for (bank = 0; bank < bank_num; bank++)
3391 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3392 
3393 	if (kvm_x86_ops->setup_mce)
3394 		kvm_x86_ops->setup_mce(vcpu);
3395 out:
3396 	return r;
3397 }
3398 
3399 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3400 				      struct kvm_x86_mce *mce)
3401 {
3402 	u64 mcg_cap = vcpu->arch.mcg_cap;
3403 	unsigned bank_num = mcg_cap & 0xff;
3404 	u64 *banks = vcpu->arch.mce_banks;
3405 
3406 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3407 		return -EINVAL;
3408 	/*
3409 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3410 	 * reporting is disabled
3411 	 */
3412 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3413 	    vcpu->arch.mcg_ctl != ~(u64)0)
3414 		return 0;
3415 	banks += 4 * mce->bank;
3416 	/*
3417 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3418 	 * reporting is disabled for the bank
3419 	 */
3420 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3421 		return 0;
3422 	if (mce->status & MCI_STATUS_UC) {
3423 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3424 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3425 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3426 			return 0;
3427 		}
3428 		if (banks[1] & MCI_STATUS_VAL)
3429 			mce->status |= MCI_STATUS_OVER;
3430 		banks[2] = mce->addr;
3431 		banks[3] = mce->misc;
3432 		vcpu->arch.mcg_status = mce->mcg_status;
3433 		banks[1] = mce->status;
3434 		kvm_queue_exception(vcpu, MC_VECTOR);
3435 	} else if (!(banks[1] & MCI_STATUS_VAL)
3436 		   || !(banks[1] & MCI_STATUS_UC)) {
3437 		if (banks[1] & MCI_STATUS_VAL)
3438 			mce->status |= MCI_STATUS_OVER;
3439 		banks[2] = mce->addr;
3440 		banks[3] = mce->misc;
3441 		banks[1] = mce->status;
3442 	} else
3443 		banks[1] |= MCI_STATUS_OVER;
3444 	return 0;
3445 }
3446 
3447 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3448 					       struct kvm_vcpu_events *events)
3449 {
3450 	process_nmi(vcpu);
3451 
3452 	/*
3453 	 * The API doesn't provide the instruction length for software
3454 	 * exceptions, so don't report them. As long as the guest RIP
3455 	 * isn't advanced, we should expect to encounter the exception
3456 	 * again.
3457 	 */
3458 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3459 		events->exception.injected = 0;
3460 		events->exception.pending = 0;
3461 	} else {
3462 		events->exception.injected = vcpu->arch.exception.injected;
3463 		events->exception.pending = vcpu->arch.exception.pending;
3464 		/*
3465 		 * For ABI compatibility, deliberately conflate
3466 		 * pending and injected exceptions when
3467 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3468 		 */
3469 		if (!vcpu->kvm->arch.exception_payload_enabled)
3470 			events->exception.injected |=
3471 				vcpu->arch.exception.pending;
3472 	}
3473 	events->exception.nr = vcpu->arch.exception.nr;
3474 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3475 	events->exception.error_code = vcpu->arch.exception.error_code;
3476 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3477 	events->exception_payload = vcpu->arch.exception.payload;
3478 
3479 	events->interrupt.injected =
3480 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3481 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3482 	events->interrupt.soft = 0;
3483 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3484 
3485 	events->nmi.injected = vcpu->arch.nmi_injected;
3486 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3487 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3488 	events->nmi.pad = 0;
3489 
3490 	events->sipi_vector = 0; /* never valid when reporting to user space */
3491 
3492 	events->smi.smm = is_smm(vcpu);
3493 	events->smi.pending = vcpu->arch.smi_pending;
3494 	events->smi.smm_inside_nmi =
3495 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3496 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3497 
3498 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3499 			 | KVM_VCPUEVENT_VALID_SHADOW
3500 			 | KVM_VCPUEVENT_VALID_SMM);
3501 	if (vcpu->kvm->arch.exception_payload_enabled)
3502 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3503 
3504 	memset(&events->reserved, 0, sizeof(events->reserved));
3505 }
3506 
3507 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3508 
3509 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3510 					      struct kvm_vcpu_events *events)
3511 {
3512 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3513 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3514 			      | KVM_VCPUEVENT_VALID_SHADOW
3515 			      | KVM_VCPUEVENT_VALID_SMM
3516 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3517 		return -EINVAL;
3518 
3519 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3520 		if (!vcpu->kvm->arch.exception_payload_enabled)
3521 			return -EINVAL;
3522 		if (events->exception.pending)
3523 			events->exception.injected = 0;
3524 		else
3525 			events->exception_has_payload = 0;
3526 	} else {
3527 		events->exception.pending = 0;
3528 		events->exception_has_payload = 0;
3529 	}
3530 
3531 	if ((events->exception.injected || events->exception.pending) &&
3532 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3533 		return -EINVAL;
3534 
3535 	/* INITs are latched while in SMM */
3536 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3537 	    (events->smi.smm || events->smi.pending) &&
3538 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3539 		return -EINVAL;
3540 
3541 	process_nmi(vcpu);
3542 	vcpu->arch.exception.injected = events->exception.injected;
3543 	vcpu->arch.exception.pending = events->exception.pending;
3544 	vcpu->arch.exception.nr = events->exception.nr;
3545 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3546 	vcpu->arch.exception.error_code = events->exception.error_code;
3547 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3548 	vcpu->arch.exception.payload = events->exception_payload;
3549 
3550 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3551 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3552 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3553 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3554 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3555 						  events->interrupt.shadow);
3556 
3557 	vcpu->arch.nmi_injected = events->nmi.injected;
3558 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3559 		vcpu->arch.nmi_pending = events->nmi.pending;
3560 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3561 
3562 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3563 	    lapic_in_kernel(vcpu))
3564 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3565 
3566 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3567 		u32 hflags = vcpu->arch.hflags;
3568 		if (events->smi.smm)
3569 			hflags |= HF_SMM_MASK;
3570 		else
3571 			hflags &= ~HF_SMM_MASK;
3572 		kvm_set_hflags(vcpu, hflags);
3573 
3574 		vcpu->arch.smi_pending = events->smi.pending;
3575 
3576 		if (events->smi.smm) {
3577 			if (events->smi.smm_inside_nmi)
3578 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3579 			else
3580 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3581 			if (lapic_in_kernel(vcpu)) {
3582 				if (events->smi.latched_init)
3583 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3584 				else
3585 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3586 			}
3587 		}
3588 	}
3589 
3590 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3591 
3592 	return 0;
3593 }
3594 
3595 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3596 					     struct kvm_debugregs *dbgregs)
3597 {
3598 	unsigned long val;
3599 
3600 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3601 	kvm_get_dr(vcpu, 6, &val);
3602 	dbgregs->dr6 = val;
3603 	dbgregs->dr7 = vcpu->arch.dr7;
3604 	dbgregs->flags = 0;
3605 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3606 }
3607 
3608 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3609 					    struct kvm_debugregs *dbgregs)
3610 {
3611 	if (dbgregs->flags)
3612 		return -EINVAL;
3613 
3614 	if (dbgregs->dr6 & ~0xffffffffull)
3615 		return -EINVAL;
3616 	if (dbgregs->dr7 & ~0xffffffffull)
3617 		return -EINVAL;
3618 
3619 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3620 	kvm_update_dr0123(vcpu);
3621 	vcpu->arch.dr6 = dbgregs->dr6;
3622 	kvm_update_dr6(vcpu);
3623 	vcpu->arch.dr7 = dbgregs->dr7;
3624 	kvm_update_dr7(vcpu);
3625 
3626 	return 0;
3627 }
3628 
3629 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3630 
3631 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3632 {
3633 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3634 	u64 xstate_bv = xsave->header.xfeatures;
3635 	u64 valid;
3636 
3637 	/*
3638 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3639 	 * leaves 0 and 1 in the loop below.
3640 	 */
3641 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3642 
3643 	/* Set XSTATE_BV */
3644 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3645 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3646 
3647 	/*
3648 	 * Copy each region from the possibly compacted offset to the
3649 	 * non-compacted offset.
3650 	 */
3651 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3652 	while (valid) {
3653 		u64 feature = valid & -valid;
3654 		int index = fls64(feature) - 1;
3655 		void *src = get_xsave_addr(xsave, feature);
3656 
3657 		if (src) {
3658 			u32 size, offset, ecx, edx;
3659 			cpuid_count(XSTATE_CPUID, index,
3660 				    &size, &offset, &ecx, &edx);
3661 			if (feature == XFEATURE_MASK_PKRU)
3662 				memcpy(dest + offset, &vcpu->arch.pkru,
3663 				       sizeof(vcpu->arch.pkru));
3664 			else
3665 				memcpy(dest + offset, src, size);
3666 
3667 		}
3668 
3669 		valid -= feature;
3670 	}
3671 }
3672 
3673 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3674 {
3675 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3676 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3677 	u64 valid;
3678 
3679 	/*
3680 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3681 	 * leaves 0 and 1 in the loop below.
3682 	 */
3683 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3684 
3685 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3686 	xsave->header.xfeatures = xstate_bv;
3687 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3688 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3689 
3690 	/*
3691 	 * Copy each region from the non-compacted offset to the
3692 	 * possibly compacted offset.
3693 	 */
3694 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3695 	while (valid) {
3696 		u64 feature = valid & -valid;
3697 		int index = fls64(feature) - 1;
3698 		void *dest = get_xsave_addr(xsave, feature);
3699 
3700 		if (dest) {
3701 			u32 size, offset, ecx, edx;
3702 			cpuid_count(XSTATE_CPUID, index,
3703 				    &size, &offset, &ecx, &edx);
3704 			if (feature == XFEATURE_MASK_PKRU)
3705 				memcpy(&vcpu->arch.pkru, src + offset,
3706 				       sizeof(vcpu->arch.pkru));
3707 			else
3708 				memcpy(dest, src + offset, size);
3709 		}
3710 
3711 		valid -= feature;
3712 	}
3713 }
3714 
3715 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3716 					 struct kvm_xsave *guest_xsave)
3717 {
3718 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3719 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3720 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3721 	} else {
3722 		memcpy(guest_xsave->region,
3723 			&vcpu->arch.guest_fpu.state.fxsave,
3724 			sizeof(struct fxregs_state));
3725 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3726 			XFEATURE_MASK_FPSSE;
3727 	}
3728 }
3729 
3730 #define XSAVE_MXCSR_OFFSET 24
3731 
3732 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3733 					struct kvm_xsave *guest_xsave)
3734 {
3735 	u64 xstate_bv =
3736 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3737 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3738 
3739 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3740 		/*
3741 		 * Here we allow setting states that are not present in
3742 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3743 		 * with old userspace.
3744 		 */
3745 		if (xstate_bv & ~kvm_supported_xcr0() ||
3746 			mxcsr & ~mxcsr_feature_mask)
3747 			return -EINVAL;
3748 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3749 	} else {
3750 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3751 			mxcsr & ~mxcsr_feature_mask)
3752 			return -EINVAL;
3753 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3754 			guest_xsave->region, sizeof(struct fxregs_state));
3755 	}
3756 	return 0;
3757 }
3758 
3759 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3760 					struct kvm_xcrs *guest_xcrs)
3761 {
3762 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3763 		guest_xcrs->nr_xcrs = 0;
3764 		return;
3765 	}
3766 
3767 	guest_xcrs->nr_xcrs = 1;
3768 	guest_xcrs->flags = 0;
3769 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3770 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3771 }
3772 
3773 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3774 				       struct kvm_xcrs *guest_xcrs)
3775 {
3776 	int i, r = 0;
3777 
3778 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3779 		return -EINVAL;
3780 
3781 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3782 		return -EINVAL;
3783 
3784 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3785 		/* Only support XCR0 currently */
3786 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3787 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3788 				guest_xcrs->xcrs[i].value);
3789 			break;
3790 		}
3791 	if (r)
3792 		r = -EINVAL;
3793 	return r;
3794 }
3795 
3796 /*
3797  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3798  * stopped by the hypervisor.  This function will be called from the host only.
3799  * EINVAL is returned when the host attempts to set the flag for a guest that
3800  * does not support pv clocks.
3801  */
3802 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3803 {
3804 	if (!vcpu->arch.pv_time_enabled)
3805 		return -EINVAL;
3806 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3807 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3808 	return 0;
3809 }
3810 
3811 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3812 				     struct kvm_enable_cap *cap)
3813 {
3814 	int r;
3815 	uint16_t vmcs_version;
3816 	void __user *user_ptr;
3817 
3818 	if (cap->flags)
3819 		return -EINVAL;
3820 
3821 	switch (cap->cap) {
3822 	case KVM_CAP_HYPERV_SYNIC2:
3823 		if (cap->args[0])
3824 			return -EINVAL;
3825 	case KVM_CAP_HYPERV_SYNIC:
3826 		if (!irqchip_in_kernel(vcpu->kvm))
3827 			return -EINVAL;
3828 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3829 					     KVM_CAP_HYPERV_SYNIC2);
3830 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3831 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3832 		if (!r) {
3833 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
3834 			if (copy_to_user(user_ptr, &vmcs_version,
3835 					 sizeof(vmcs_version)))
3836 				r = -EFAULT;
3837 		}
3838 		return r;
3839 
3840 	default:
3841 		return -EINVAL;
3842 	}
3843 }
3844 
3845 long kvm_arch_vcpu_ioctl(struct file *filp,
3846 			 unsigned int ioctl, unsigned long arg)
3847 {
3848 	struct kvm_vcpu *vcpu = filp->private_data;
3849 	void __user *argp = (void __user *)arg;
3850 	int r;
3851 	union {
3852 		struct kvm_lapic_state *lapic;
3853 		struct kvm_xsave *xsave;
3854 		struct kvm_xcrs *xcrs;
3855 		void *buffer;
3856 	} u;
3857 
3858 	vcpu_load(vcpu);
3859 
3860 	u.buffer = NULL;
3861 	switch (ioctl) {
3862 	case KVM_GET_LAPIC: {
3863 		r = -EINVAL;
3864 		if (!lapic_in_kernel(vcpu))
3865 			goto out;
3866 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3867 
3868 		r = -ENOMEM;
3869 		if (!u.lapic)
3870 			goto out;
3871 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3872 		if (r)
3873 			goto out;
3874 		r = -EFAULT;
3875 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3876 			goto out;
3877 		r = 0;
3878 		break;
3879 	}
3880 	case KVM_SET_LAPIC: {
3881 		r = -EINVAL;
3882 		if (!lapic_in_kernel(vcpu))
3883 			goto out;
3884 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3885 		if (IS_ERR(u.lapic)) {
3886 			r = PTR_ERR(u.lapic);
3887 			goto out_nofree;
3888 		}
3889 
3890 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3891 		break;
3892 	}
3893 	case KVM_INTERRUPT: {
3894 		struct kvm_interrupt irq;
3895 
3896 		r = -EFAULT;
3897 		if (copy_from_user(&irq, argp, sizeof(irq)))
3898 			goto out;
3899 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3900 		break;
3901 	}
3902 	case KVM_NMI: {
3903 		r = kvm_vcpu_ioctl_nmi(vcpu);
3904 		break;
3905 	}
3906 	case KVM_SMI: {
3907 		r = kvm_vcpu_ioctl_smi(vcpu);
3908 		break;
3909 	}
3910 	case KVM_SET_CPUID: {
3911 		struct kvm_cpuid __user *cpuid_arg = argp;
3912 		struct kvm_cpuid cpuid;
3913 
3914 		r = -EFAULT;
3915 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3916 			goto out;
3917 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3918 		break;
3919 	}
3920 	case KVM_SET_CPUID2: {
3921 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3922 		struct kvm_cpuid2 cpuid;
3923 
3924 		r = -EFAULT;
3925 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3926 			goto out;
3927 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3928 					      cpuid_arg->entries);
3929 		break;
3930 	}
3931 	case KVM_GET_CPUID2: {
3932 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3933 		struct kvm_cpuid2 cpuid;
3934 
3935 		r = -EFAULT;
3936 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3937 			goto out;
3938 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3939 					      cpuid_arg->entries);
3940 		if (r)
3941 			goto out;
3942 		r = -EFAULT;
3943 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3944 			goto out;
3945 		r = 0;
3946 		break;
3947 	}
3948 	case KVM_GET_MSRS: {
3949 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3950 		r = msr_io(vcpu, argp, do_get_msr, 1);
3951 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3952 		break;
3953 	}
3954 	case KVM_SET_MSRS: {
3955 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3956 		r = msr_io(vcpu, argp, do_set_msr, 0);
3957 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3958 		break;
3959 	}
3960 	case KVM_TPR_ACCESS_REPORTING: {
3961 		struct kvm_tpr_access_ctl tac;
3962 
3963 		r = -EFAULT;
3964 		if (copy_from_user(&tac, argp, sizeof(tac)))
3965 			goto out;
3966 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3967 		if (r)
3968 			goto out;
3969 		r = -EFAULT;
3970 		if (copy_to_user(argp, &tac, sizeof(tac)))
3971 			goto out;
3972 		r = 0;
3973 		break;
3974 	};
3975 	case KVM_SET_VAPIC_ADDR: {
3976 		struct kvm_vapic_addr va;
3977 		int idx;
3978 
3979 		r = -EINVAL;
3980 		if (!lapic_in_kernel(vcpu))
3981 			goto out;
3982 		r = -EFAULT;
3983 		if (copy_from_user(&va, argp, sizeof(va)))
3984 			goto out;
3985 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3986 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3987 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3988 		break;
3989 	}
3990 	case KVM_X86_SETUP_MCE: {
3991 		u64 mcg_cap;
3992 
3993 		r = -EFAULT;
3994 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
3995 			goto out;
3996 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3997 		break;
3998 	}
3999 	case KVM_X86_SET_MCE: {
4000 		struct kvm_x86_mce mce;
4001 
4002 		r = -EFAULT;
4003 		if (copy_from_user(&mce, argp, sizeof(mce)))
4004 			goto out;
4005 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4006 		break;
4007 	}
4008 	case KVM_GET_VCPU_EVENTS: {
4009 		struct kvm_vcpu_events events;
4010 
4011 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4012 
4013 		r = -EFAULT;
4014 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4015 			break;
4016 		r = 0;
4017 		break;
4018 	}
4019 	case KVM_SET_VCPU_EVENTS: {
4020 		struct kvm_vcpu_events events;
4021 
4022 		r = -EFAULT;
4023 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4024 			break;
4025 
4026 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4027 		break;
4028 	}
4029 	case KVM_GET_DEBUGREGS: {
4030 		struct kvm_debugregs dbgregs;
4031 
4032 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4033 
4034 		r = -EFAULT;
4035 		if (copy_to_user(argp, &dbgregs,
4036 				 sizeof(struct kvm_debugregs)))
4037 			break;
4038 		r = 0;
4039 		break;
4040 	}
4041 	case KVM_SET_DEBUGREGS: {
4042 		struct kvm_debugregs dbgregs;
4043 
4044 		r = -EFAULT;
4045 		if (copy_from_user(&dbgregs, argp,
4046 				   sizeof(struct kvm_debugregs)))
4047 			break;
4048 
4049 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4050 		break;
4051 	}
4052 	case KVM_GET_XSAVE: {
4053 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
4054 		r = -ENOMEM;
4055 		if (!u.xsave)
4056 			break;
4057 
4058 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4059 
4060 		r = -EFAULT;
4061 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4062 			break;
4063 		r = 0;
4064 		break;
4065 	}
4066 	case KVM_SET_XSAVE: {
4067 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4068 		if (IS_ERR(u.xsave)) {
4069 			r = PTR_ERR(u.xsave);
4070 			goto out_nofree;
4071 		}
4072 
4073 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4074 		break;
4075 	}
4076 	case KVM_GET_XCRS: {
4077 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
4078 		r = -ENOMEM;
4079 		if (!u.xcrs)
4080 			break;
4081 
4082 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4083 
4084 		r = -EFAULT;
4085 		if (copy_to_user(argp, u.xcrs,
4086 				 sizeof(struct kvm_xcrs)))
4087 			break;
4088 		r = 0;
4089 		break;
4090 	}
4091 	case KVM_SET_XCRS: {
4092 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4093 		if (IS_ERR(u.xcrs)) {
4094 			r = PTR_ERR(u.xcrs);
4095 			goto out_nofree;
4096 		}
4097 
4098 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4099 		break;
4100 	}
4101 	case KVM_SET_TSC_KHZ: {
4102 		u32 user_tsc_khz;
4103 
4104 		r = -EINVAL;
4105 		user_tsc_khz = (u32)arg;
4106 
4107 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4108 			goto out;
4109 
4110 		if (user_tsc_khz == 0)
4111 			user_tsc_khz = tsc_khz;
4112 
4113 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4114 			r = 0;
4115 
4116 		goto out;
4117 	}
4118 	case KVM_GET_TSC_KHZ: {
4119 		r = vcpu->arch.virtual_tsc_khz;
4120 		goto out;
4121 	}
4122 	case KVM_KVMCLOCK_CTRL: {
4123 		r = kvm_set_guest_paused(vcpu);
4124 		goto out;
4125 	}
4126 	case KVM_ENABLE_CAP: {
4127 		struct kvm_enable_cap cap;
4128 
4129 		r = -EFAULT;
4130 		if (copy_from_user(&cap, argp, sizeof(cap)))
4131 			goto out;
4132 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4133 		break;
4134 	}
4135 	case KVM_GET_NESTED_STATE: {
4136 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4137 		u32 user_data_size;
4138 
4139 		r = -EINVAL;
4140 		if (!kvm_x86_ops->get_nested_state)
4141 			break;
4142 
4143 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4144 		r = -EFAULT;
4145 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4146 			break;
4147 
4148 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4149 						  user_data_size);
4150 		if (r < 0)
4151 			break;
4152 
4153 		if (r > user_data_size) {
4154 			if (put_user(r, &user_kvm_nested_state->size))
4155 				r = -EFAULT;
4156 			else
4157 				r = -E2BIG;
4158 			break;
4159 		}
4160 
4161 		r = 0;
4162 		break;
4163 	}
4164 	case KVM_SET_NESTED_STATE: {
4165 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4166 		struct kvm_nested_state kvm_state;
4167 
4168 		r = -EINVAL;
4169 		if (!kvm_x86_ops->set_nested_state)
4170 			break;
4171 
4172 		r = -EFAULT;
4173 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4174 			break;
4175 
4176 		r = -EINVAL;
4177 		if (kvm_state.size < sizeof(kvm_state))
4178 			break;
4179 
4180 		if (kvm_state.flags &
4181 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4182 		      | KVM_STATE_NESTED_EVMCS))
4183 			break;
4184 
4185 		/* nested_run_pending implies guest_mode.  */
4186 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4187 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4188 			break;
4189 
4190 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4191 		break;
4192 	}
4193 	default:
4194 		r = -EINVAL;
4195 	}
4196 out:
4197 	kfree(u.buffer);
4198 out_nofree:
4199 	vcpu_put(vcpu);
4200 	return r;
4201 }
4202 
4203 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4204 {
4205 	return VM_FAULT_SIGBUS;
4206 }
4207 
4208 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4209 {
4210 	int ret;
4211 
4212 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4213 		return -EINVAL;
4214 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4215 	return ret;
4216 }
4217 
4218 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4219 					      u64 ident_addr)
4220 {
4221 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4222 }
4223 
4224 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4225 					  u32 kvm_nr_mmu_pages)
4226 {
4227 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4228 		return -EINVAL;
4229 
4230 	mutex_lock(&kvm->slots_lock);
4231 
4232 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4233 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4234 
4235 	mutex_unlock(&kvm->slots_lock);
4236 	return 0;
4237 }
4238 
4239 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4240 {
4241 	return kvm->arch.n_max_mmu_pages;
4242 }
4243 
4244 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4245 {
4246 	struct kvm_pic *pic = kvm->arch.vpic;
4247 	int r;
4248 
4249 	r = 0;
4250 	switch (chip->chip_id) {
4251 	case KVM_IRQCHIP_PIC_MASTER:
4252 		memcpy(&chip->chip.pic, &pic->pics[0],
4253 			sizeof(struct kvm_pic_state));
4254 		break;
4255 	case KVM_IRQCHIP_PIC_SLAVE:
4256 		memcpy(&chip->chip.pic, &pic->pics[1],
4257 			sizeof(struct kvm_pic_state));
4258 		break;
4259 	case KVM_IRQCHIP_IOAPIC:
4260 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4261 		break;
4262 	default:
4263 		r = -EINVAL;
4264 		break;
4265 	}
4266 	return r;
4267 }
4268 
4269 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4270 {
4271 	struct kvm_pic *pic = kvm->arch.vpic;
4272 	int r;
4273 
4274 	r = 0;
4275 	switch (chip->chip_id) {
4276 	case KVM_IRQCHIP_PIC_MASTER:
4277 		spin_lock(&pic->lock);
4278 		memcpy(&pic->pics[0], &chip->chip.pic,
4279 			sizeof(struct kvm_pic_state));
4280 		spin_unlock(&pic->lock);
4281 		break;
4282 	case KVM_IRQCHIP_PIC_SLAVE:
4283 		spin_lock(&pic->lock);
4284 		memcpy(&pic->pics[1], &chip->chip.pic,
4285 			sizeof(struct kvm_pic_state));
4286 		spin_unlock(&pic->lock);
4287 		break;
4288 	case KVM_IRQCHIP_IOAPIC:
4289 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4290 		break;
4291 	default:
4292 		r = -EINVAL;
4293 		break;
4294 	}
4295 	kvm_pic_update_irq(pic);
4296 	return r;
4297 }
4298 
4299 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4300 {
4301 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4302 
4303 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4304 
4305 	mutex_lock(&kps->lock);
4306 	memcpy(ps, &kps->channels, sizeof(*ps));
4307 	mutex_unlock(&kps->lock);
4308 	return 0;
4309 }
4310 
4311 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4312 {
4313 	int i;
4314 	struct kvm_pit *pit = kvm->arch.vpit;
4315 
4316 	mutex_lock(&pit->pit_state.lock);
4317 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4318 	for (i = 0; i < 3; i++)
4319 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4320 	mutex_unlock(&pit->pit_state.lock);
4321 	return 0;
4322 }
4323 
4324 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4325 {
4326 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4327 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4328 		sizeof(ps->channels));
4329 	ps->flags = kvm->arch.vpit->pit_state.flags;
4330 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4331 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4332 	return 0;
4333 }
4334 
4335 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4336 {
4337 	int start = 0;
4338 	int i;
4339 	u32 prev_legacy, cur_legacy;
4340 	struct kvm_pit *pit = kvm->arch.vpit;
4341 
4342 	mutex_lock(&pit->pit_state.lock);
4343 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4344 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4345 	if (!prev_legacy && cur_legacy)
4346 		start = 1;
4347 	memcpy(&pit->pit_state.channels, &ps->channels,
4348 	       sizeof(pit->pit_state.channels));
4349 	pit->pit_state.flags = ps->flags;
4350 	for (i = 0; i < 3; i++)
4351 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4352 				   start && i == 0);
4353 	mutex_unlock(&pit->pit_state.lock);
4354 	return 0;
4355 }
4356 
4357 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4358 				 struct kvm_reinject_control *control)
4359 {
4360 	struct kvm_pit *pit = kvm->arch.vpit;
4361 
4362 	if (!pit)
4363 		return -ENXIO;
4364 
4365 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4366 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4367 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4368 	 */
4369 	mutex_lock(&pit->pit_state.lock);
4370 	kvm_pit_set_reinject(pit, control->pit_reinject);
4371 	mutex_unlock(&pit->pit_state.lock);
4372 
4373 	return 0;
4374 }
4375 
4376 /**
4377  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4378  * @kvm: kvm instance
4379  * @log: slot id and address to which we copy the log
4380  *
4381  * Steps 1-4 below provide general overview of dirty page logging. See
4382  * kvm_get_dirty_log_protect() function description for additional details.
4383  *
4384  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4385  * always flush the TLB (step 4) even if previous step failed  and the dirty
4386  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4387  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4388  * writes will be marked dirty for next log read.
4389  *
4390  *   1. Take a snapshot of the bit and clear it if needed.
4391  *   2. Write protect the corresponding page.
4392  *   3. Copy the snapshot to the userspace.
4393  *   4. Flush TLB's if needed.
4394  */
4395 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4396 {
4397 	bool is_dirty = false;
4398 	int r;
4399 
4400 	mutex_lock(&kvm->slots_lock);
4401 
4402 	/*
4403 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4404 	 */
4405 	if (kvm_x86_ops->flush_log_dirty)
4406 		kvm_x86_ops->flush_log_dirty(kvm);
4407 
4408 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4409 
4410 	/*
4411 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4412 	 * kvm_mmu_slot_remove_write_access().
4413 	 */
4414 	lockdep_assert_held(&kvm->slots_lock);
4415 	if (is_dirty)
4416 		kvm_flush_remote_tlbs(kvm);
4417 
4418 	mutex_unlock(&kvm->slots_lock);
4419 	return r;
4420 }
4421 
4422 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4423 			bool line_status)
4424 {
4425 	if (!irqchip_in_kernel(kvm))
4426 		return -ENXIO;
4427 
4428 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4429 					irq_event->irq, irq_event->level,
4430 					line_status);
4431 	return 0;
4432 }
4433 
4434 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4435 				   struct kvm_enable_cap *cap)
4436 {
4437 	int r;
4438 
4439 	if (cap->flags)
4440 		return -EINVAL;
4441 
4442 	switch (cap->cap) {
4443 	case KVM_CAP_DISABLE_QUIRKS:
4444 		kvm->arch.disabled_quirks = cap->args[0];
4445 		r = 0;
4446 		break;
4447 	case KVM_CAP_SPLIT_IRQCHIP: {
4448 		mutex_lock(&kvm->lock);
4449 		r = -EINVAL;
4450 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4451 			goto split_irqchip_unlock;
4452 		r = -EEXIST;
4453 		if (irqchip_in_kernel(kvm))
4454 			goto split_irqchip_unlock;
4455 		if (kvm->created_vcpus)
4456 			goto split_irqchip_unlock;
4457 		r = kvm_setup_empty_irq_routing(kvm);
4458 		if (r)
4459 			goto split_irqchip_unlock;
4460 		/* Pairs with irqchip_in_kernel. */
4461 		smp_wmb();
4462 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4463 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4464 		r = 0;
4465 split_irqchip_unlock:
4466 		mutex_unlock(&kvm->lock);
4467 		break;
4468 	}
4469 	case KVM_CAP_X2APIC_API:
4470 		r = -EINVAL;
4471 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4472 			break;
4473 
4474 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4475 			kvm->arch.x2apic_format = true;
4476 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4477 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4478 
4479 		r = 0;
4480 		break;
4481 	case KVM_CAP_X86_DISABLE_EXITS:
4482 		r = -EINVAL;
4483 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4484 			break;
4485 
4486 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4487 			kvm_can_mwait_in_guest())
4488 			kvm->arch.mwait_in_guest = true;
4489 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4490 			kvm->arch.hlt_in_guest = true;
4491 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4492 			kvm->arch.pause_in_guest = true;
4493 		r = 0;
4494 		break;
4495 	case KVM_CAP_MSR_PLATFORM_INFO:
4496 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4497 		r = 0;
4498 		break;
4499 	case KVM_CAP_EXCEPTION_PAYLOAD:
4500 		kvm->arch.exception_payload_enabled = cap->args[0];
4501 		r = 0;
4502 		break;
4503 	default:
4504 		r = -EINVAL;
4505 		break;
4506 	}
4507 	return r;
4508 }
4509 
4510 long kvm_arch_vm_ioctl(struct file *filp,
4511 		       unsigned int ioctl, unsigned long arg)
4512 {
4513 	struct kvm *kvm = filp->private_data;
4514 	void __user *argp = (void __user *)arg;
4515 	int r = -ENOTTY;
4516 	/*
4517 	 * This union makes it completely explicit to gcc-3.x
4518 	 * that these two variables' stack usage should be
4519 	 * combined, not added together.
4520 	 */
4521 	union {
4522 		struct kvm_pit_state ps;
4523 		struct kvm_pit_state2 ps2;
4524 		struct kvm_pit_config pit_config;
4525 	} u;
4526 
4527 	switch (ioctl) {
4528 	case KVM_SET_TSS_ADDR:
4529 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4530 		break;
4531 	case KVM_SET_IDENTITY_MAP_ADDR: {
4532 		u64 ident_addr;
4533 
4534 		mutex_lock(&kvm->lock);
4535 		r = -EINVAL;
4536 		if (kvm->created_vcpus)
4537 			goto set_identity_unlock;
4538 		r = -EFAULT;
4539 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4540 			goto set_identity_unlock;
4541 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4542 set_identity_unlock:
4543 		mutex_unlock(&kvm->lock);
4544 		break;
4545 	}
4546 	case KVM_SET_NR_MMU_PAGES:
4547 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4548 		break;
4549 	case KVM_GET_NR_MMU_PAGES:
4550 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4551 		break;
4552 	case KVM_CREATE_IRQCHIP: {
4553 		mutex_lock(&kvm->lock);
4554 
4555 		r = -EEXIST;
4556 		if (irqchip_in_kernel(kvm))
4557 			goto create_irqchip_unlock;
4558 
4559 		r = -EINVAL;
4560 		if (kvm->created_vcpus)
4561 			goto create_irqchip_unlock;
4562 
4563 		r = kvm_pic_init(kvm);
4564 		if (r)
4565 			goto create_irqchip_unlock;
4566 
4567 		r = kvm_ioapic_init(kvm);
4568 		if (r) {
4569 			kvm_pic_destroy(kvm);
4570 			goto create_irqchip_unlock;
4571 		}
4572 
4573 		r = kvm_setup_default_irq_routing(kvm);
4574 		if (r) {
4575 			kvm_ioapic_destroy(kvm);
4576 			kvm_pic_destroy(kvm);
4577 			goto create_irqchip_unlock;
4578 		}
4579 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4580 		smp_wmb();
4581 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4582 	create_irqchip_unlock:
4583 		mutex_unlock(&kvm->lock);
4584 		break;
4585 	}
4586 	case KVM_CREATE_PIT:
4587 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4588 		goto create_pit;
4589 	case KVM_CREATE_PIT2:
4590 		r = -EFAULT;
4591 		if (copy_from_user(&u.pit_config, argp,
4592 				   sizeof(struct kvm_pit_config)))
4593 			goto out;
4594 	create_pit:
4595 		mutex_lock(&kvm->lock);
4596 		r = -EEXIST;
4597 		if (kvm->arch.vpit)
4598 			goto create_pit_unlock;
4599 		r = -ENOMEM;
4600 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4601 		if (kvm->arch.vpit)
4602 			r = 0;
4603 	create_pit_unlock:
4604 		mutex_unlock(&kvm->lock);
4605 		break;
4606 	case KVM_GET_IRQCHIP: {
4607 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4608 		struct kvm_irqchip *chip;
4609 
4610 		chip = memdup_user(argp, sizeof(*chip));
4611 		if (IS_ERR(chip)) {
4612 			r = PTR_ERR(chip);
4613 			goto out;
4614 		}
4615 
4616 		r = -ENXIO;
4617 		if (!irqchip_kernel(kvm))
4618 			goto get_irqchip_out;
4619 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4620 		if (r)
4621 			goto get_irqchip_out;
4622 		r = -EFAULT;
4623 		if (copy_to_user(argp, chip, sizeof(*chip)))
4624 			goto get_irqchip_out;
4625 		r = 0;
4626 	get_irqchip_out:
4627 		kfree(chip);
4628 		break;
4629 	}
4630 	case KVM_SET_IRQCHIP: {
4631 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4632 		struct kvm_irqchip *chip;
4633 
4634 		chip = memdup_user(argp, sizeof(*chip));
4635 		if (IS_ERR(chip)) {
4636 			r = PTR_ERR(chip);
4637 			goto out;
4638 		}
4639 
4640 		r = -ENXIO;
4641 		if (!irqchip_kernel(kvm))
4642 			goto set_irqchip_out;
4643 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4644 		if (r)
4645 			goto set_irqchip_out;
4646 		r = 0;
4647 	set_irqchip_out:
4648 		kfree(chip);
4649 		break;
4650 	}
4651 	case KVM_GET_PIT: {
4652 		r = -EFAULT;
4653 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4654 			goto out;
4655 		r = -ENXIO;
4656 		if (!kvm->arch.vpit)
4657 			goto out;
4658 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4659 		if (r)
4660 			goto out;
4661 		r = -EFAULT;
4662 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4663 			goto out;
4664 		r = 0;
4665 		break;
4666 	}
4667 	case KVM_SET_PIT: {
4668 		r = -EFAULT;
4669 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4670 			goto out;
4671 		r = -ENXIO;
4672 		if (!kvm->arch.vpit)
4673 			goto out;
4674 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4675 		break;
4676 	}
4677 	case KVM_GET_PIT2: {
4678 		r = -ENXIO;
4679 		if (!kvm->arch.vpit)
4680 			goto out;
4681 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4682 		if (r)
4683 			goto out;
4684 		r = -EFAULT;
4685 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4686 			goto out;
4687 		r = 0;
4688 		break;
4689 	}
4690 	case KVM_SET_PIT2: {
4691 		r = -EFAULT;
4692 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4693 			goto out;
4694 		r = -ENXIO;
4695 		if (!kvm->arch.vpit)
4696 			goto out;
4697 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4698 		break;
4699 	}
4700 	case KVM_REINJECT_CONTROL: {
4701 		struct kvm_reinject_control control;
4702 		r =  -EFAULT;
4703 		if (copy_from_user(&control, argp, sizeof(control)))
4704 			goto out;
4705 		r = kvm_vm_ioctl_reinject(kvm, &control);
4706 		break;
4707 	}
4708 	case KVM_SET_BOOT_CPU_ID:
4709 		r = 0;
4710 		mutex_lock(&kvm->lock);
4711 		if (kvm->created_vcpus)
4712 			r = -EBUSY;
4713 		else
4714 			kvm->arch.bsp_vcpu_id = arg;
4715 		mutex_unlock(&kvm->lock);
4716 		break;
4717 	case KVM_XEN_HVM_CONFIG: {
4718 		struct kvm_xen_hvm_config xhc;
4719 		r = -EFAULT;
4720 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4721 			goto out;
4722 		r = -EINVAL;
4723 		if (xhc.flags)
4724 			goto out;
4725 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4726 		r = 0;
4727 		break;
4728 	}
4729 	case KVM_SET_CLOCK: {
4730 		struct kvm_clock_data user_ns;
4731 		u64 now_ns;
4732 
4733 		r = -EFAULT;
4734 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4735 			goto out;
4736 
4737 		r = -EINVAL;
4738 		if (user_ns.flags)
4739 			goto out;
4740 
4741 		r = 0;
4742 		/*
4743 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4744 		 * kvm_gen_update_masterclock() can be cut down to locked
4745 		 * pvclock_update_vm_gtod_copy().
4746 		 */
4747 		kvm_gen_update_masterclock(kvm);
4748 		now_ns = get_kvmclock_ns(kvm);
4749 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4750 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4751 		break;
4752 	}
4753 	case KVM_GET_CLOCK: {
4754 		struct kvm_clock_data user_ns;
4755 		u64 now_ns;
4756 
4757 		now_ns = get_kvmclock_ns(kvm);
4758 		user_ns.clock = now_ns;
4759 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4760 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4761 
4762 		r = -EFAULT;
4763 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4764 			goto out;
4765 		r = 0;
4766 		break;
4767 	}
4768 	case KVM_ENABLE_CAP: {
4769 		struct kvm_enable_cap cap;
4770 
4771 		r = -EFAULT;
4772 		if (copy_from_user(&cap, argp, sizeof(cap)))
4773 			goto out;
4774 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4775 		break;
4776 	}
4777 	case KVM_MEMORY_ENCRYPT_OP: {
4778 		r = -ENOTTY;
4779 		if (kvm_x86_ops->mem_enc_op)
4780 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4781 		break;
4782 	}
4783 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4784 		struct kvm_enc_region region;
4785 
4786 		r = -EFAULT;
4787 		if (copy_from_user(&region, argp, sizeof(region)))
4788 			goto out;
4789 
4790 		r = -ENOTTY;
4791 		if (kvm_x86_ops->mem_enc_reg_region)
4792 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4793 		break;
4794 	}
4795 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4796 		struct kvm_enc_region region;
4797 
4798 		r = -EFAULT;
4799 		if (copy_from_user(&region, argp, sizeof(region)))
4800 			goto out;
4801 
4802 		r = -ENOTTY;
4803 		if (kvm_x86_ops->mem_enc_unreg_region)
4804 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4805 		break;
4806 	}
4807 	case KVM_HYPERV_EVENTFD: {
4808 		struct kvm_hyperv_eventfd hvevfd;
4809 
4810 		r = -EFAULT;
4811 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4812 			goto out;
4813 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4814 		break;
4815 	}
4816 	default:
4817 		r = -ENOTTY;
4818 	}
4819 out:
4820 	return r;
4821 }
4822 
4823 static void kvm_init_msr_list(void)
4824 {
4825 	u32 dummy[2];
4826 	unsigned i, j;
4827 
4828 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4829 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4830 			continue;
4831 
4832 		/*
4833 		 * Even MSRs that are valid in the host may not be exposed
4834 		 * to the guests in some cases.
4835 		 */
4836 		switch (msrs_to_save[i]) {
4837 		case MSR_IA32_BNDCFGS:
4838 			if (!kvm_mpx_supported())
4839 				continue;
4840 			break;
4841 		case MSR_TSC_AUX:
4842 			if (!kvm_x86_ops->rdtscp_supported())
4843 				continue;
4844 			break;
4845 		default:
4846 			break;
4847 		}
4848 
4849 		if (j < i)
4850 			msrs_to_save[j] = msrs_to_save[i];
4851 		j++;
4852 	}
4853 	num_msrs_to_save = j;
4854 
4855 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4856 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4857 			continue;
4858 
4859 		if (j < i)
4860 			emulated_msrs[j] = emulated_msrs[i];
4861 		j++;
4862 	}
4863 	num_emulated_msrs = j;
4864 
4865 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4866 		struct kvm_msr_entry msr;
4867 
4868 		msr.index = msr_based_features[i];
4869 		if (kvm_get_msr_feature(&msr))
4870 			continue;
4871 
4872 		if (j < i)
4873 			msr_based_features[j] = msr_based_features[i];
4874 		j++;
4875 	}
4876 	num_msr_based_features = j;
4877 }
4878 
4879 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4880 			   const void *v)
4881 {
4882 	int handled = 0;
4883 	int n;
4884 
4885 	do {
4886 		n = min(len, 8);
4887 		if (!(lapic_in_kernel(vcpu) &&
4888 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4889 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4890 			break;
4891 		handled += n;
4892 		addr += n;
4893 		len -= n;
4894 		v += n;
4895 	} while (len);
4896 
4897 	return handled;
4898 }
4899 
4900 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4901 {
4902 	int handled = 0;
4903 	int n;
4904 
4905 	do {
4906 		n = min(len, 8);
4907 		if (!(lapic_in_kernel(vcpu) &&
4908 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4909 					 addr, n, v))
4910 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4911 			break;
4912 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4913 		handled += n;
4914 		addr += n;
4915 		len -= n;
4916 		v += n;
4917 	} while (len);
4918 
4919 	return handled;
4920 }
4921 
4922 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4923 			struct kvm_segment *var, int seg)
4924 {
4925 	kvm_x86_ops->set_segment(vcpu, var, seg);
4926 }
4927 
4928 void kvm_get_segment(struct kvm_vcpu *vcpu,
4929 		     struct kvm_segment *var, int seg)
4930 {
4931 	kvm_x86_ops->get_segment(vcpu, var, seg);
4932 }
4933 
4934 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4935 			   struct x86_exception *exception)
4936 {
4937 	gpa_t t_gpa;
4938 
4939 	BUG_ON(!mmu_is_nested(vcpu));
4940 
4941 	/* NPT walks are always user-walks */
4942 	access |= PFERR_USER_MASK;
4943 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
4944 
4945 	return t_gpa;
4946 }
4947 
4948 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4949 			      struct x86_exception *exception)
4950 {
4951 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4952 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4953 }
4954 
4955  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4956 				struct x86_exception *exception)
4957 {
4958 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4959 	access |= PFERR_FETCH_MASK;
4960 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4961 }
4962 
4963 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4964 			       struct x86_exception *exception)
4965 {
4966 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4967 	access |= PFERR_WRITE_MASK;
4968 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4969 }
4970 
4971 /* uses this to access any guest's mapped memory without checking CPL */
4972 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4973 				struct x86_exception *exception)
4974 {
4975 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4976 }
4977 
4978 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4979 				      struct kvm_vcpu *vcpu, u32 access,
4980 				      struct x86_exception *exception)
4981 {
4982 	void *data = val;
4983 	int r = X86EMUL_CONTINUE;
4984 
4985 	while (bytes) {
4986 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4987 							    exception);
4988 		unsigned offset = addr & (PAGE_SIZE-1);
4989 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4990 		int ret;
4991 
4992 		if (gpa == UNMAPPED_GVA)
4993 			return X86EMUL_PROPAGATE_FAULT;
4994 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4995 					       offset, toread);
4996 		if (ret < 0) {
4997 			r = X86EMUL_IO_NEEDED;
4998 			goto out;
4999 		}
5000 
5001 		bytes -= toread;
5002 		data += toread;
5003 		addr += toread;
5004 	}
5005 out:
5006 	return r;
5007 }
5008 
5009 /* used for instruction fetching */
5010 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5011 				gva_t addr, void *val, unsigned int bytes,
5012 				struct x86_exception *exception)
5013 {
5014 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5015 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5016 	unsigned offset;
5017 	int ret;
5018 
5019 	/* Inline kvm_read_guest_virt_helper for speed.  */
5020 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5021 						    exception);
5022 	if (unlikely(gpa == UNMAPPED_GVA))
5023 		return X86EMUL_PROPAGATE_FAULT;
5024 
5025 	offset = addr & (PAGE_SIZE-1);
5026 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5027 		bytes = (unsigned)PAGE_SIZE - offset;
5028 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5029 				       offset, bytes);
5030 	if (unlikely(ret < 0))
5031 		return X86EMUL_IO_NEEDED;
5032 
5033 	return X86EMUL_CONTINUE;
5034 }
5035 
5036 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5037 			       gva_t addr, void *val, unsigned int bytes,
5038 			       struct x86_exception *exception)
5039 {
5040 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5041 
5042 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5043 					  exception);
5044 }
5045 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5046 
5047 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5048 			     gva_t addr, void *val, unsigned int bytes,
5049 			     struct x86_exception *exception, bool system)
5050 {
5051 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5052 	u32 access = 0;
5053 
5054 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5055 		access |= PFERR_USER_MASK;
5056 
5057 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5058 }
5059 
5060 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5061 		unsigned long addr, void *val, unsigned int bytes)
5062 {
5063 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5064 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5065 
5066 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5067 }
5068 
5069 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5070 				      struct kvm_vcpu *vcpu, u32 access,
5071 				      struct x86_exception *exception)
5072 {
5073 	void *data = val;
5074 	int r = X86EMUL_CONTINUE;
5075 
5076 	while (bytes) {
5077 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5078 							     access,
5079 							     exception);
5080 		unsigned offset = addr & (PAGE_SIZE-1);
5081 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5082 		int ret;
5083 
5084 		if (gpa == UNMAPPED_GVA)
5085 			return X86EMUL_PROPAGATE_FAULT;
5086 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5087 		if (ret < 0) {
5088 			r = X86EMUL_IO_NEEDED;
5089 			goto out;
5090 		}
5091 
5092 		bytes -= towrite;
5093 		data += towrite;
5094 		addr += towrite;
5095 	}
5096 out:
5097 	return r;
5098 }
5099 
5100 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5101 			      unsigned int bytes, struct x86_exception *exception,
5102 			      bool system)
5103 {
5104 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5105 	u32 access = PFERR_WRITE_MASK;
5106 
5107 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5108 		access |= PFERR_USER_MASK;
5109 
5110 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5111 					   access, exception);
5112 }
5113 
5114 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5115 				unsigned int bytes, struct x86_exception *exception)
5116 {
5117 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5118 	vcpu->arch.l1tf_flush_l1d = true;
5119 
5120 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5121 					   PFERR_WRITE_MASK, exception);
5122 }
5123 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5124 
5125 int handle_ud(struct kvm_vcpu *vcpu)
5126 {
5127 	int emul_type = EMULTYPE_TRAP_UD;
5128 	enum emulation_result er;
5129 	char sig[5]; /* ud2; .ascii "kvm" */
5130 	struct x86_exception e;
5131 
5132 	if (force_emulation_prefix &&
5133 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5134 				sig, sizeof(sig), &e) == 0 &&
5135 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5136 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5137 		emul_type = 0;
5138 	}
5139 
5140 	er = kvm_emulate_instruction(vcpu, emul_type);
5141 	if (er == EMULATE_USER_EXIT)
5142 		return 0;
5143 	if (er != EMULATE_DONE)
5144 		kvm_queue_exception(vcpu, UD_VECTOR);
5145 	return 1;
5146 }
5147 EXPORT_SYMBOL_GPL(handle_ud);
5148 
5149 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5150 			    gpa_t gpa, bool write)
5151 {
5152 	/* For APIC access vmexit */
5153 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5154 		return 1;
5155 
5156 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5157 		trace_vcpu_match_mmio(gva, gpa, write, true);
5158 		return 1;
5159 	}
5160 
5161 	return 0;
5162 }
5163 
5164 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5165 				gpa_t *gpa, struct x86_exception *exception,
5166 				bool write)
5167 {
5168 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5169 		| (write ? PFERR_WRITE_MASK : 0);
5170 
5171 	/*
5172 	 * currently PKRU is only applied to ept enabled guest so
5173 	 * there is no pkey in EPT page table for L1 guest or EPT
5174 	 * shadow page table for L2 guest.
5175 	 */
5176 	if (vcpu_match_mmio_gva(vcpu, gva)
5177 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5178 				 vcpu->arch.access, 0, access)) {
5179 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5180 					(gva & (PAGE_SIZE - 1));
5181 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5182 		return 1;
5183 	}
5184 
5185 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5186 
5187 	if (*gpa == UNMAPPED_GVA)
5188 		return -1;
5189 
5190 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5191 }
5192 
5193 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5194 			const void *val, int bytes)
5195 {
5196 	int ret;
5197 
5198 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5199 	if (ret < 0)
5200 		return 0;
5201 	kvm_page_track_write(vcpu, gpa, val, bytes);
5202 	return 1;
5203 }
5204 
5205 struct read_write_emulator_ops {
5206 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5207 				  int bytes);
5208 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5209 				  void *val, int bytes);
5210 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5211 			       int bytes, void *val);
5212 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5213 				    void *val, int bytes);
5214 	bool write;
5215 };
5216 
5217 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5218 {
5219 	if (vcpu->mmio_read_completed) {
5220 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5221 			       vcpu->mmio_fragments[0].gpa, val);
5222 		vcpu->mmio_read_completed = 0;
5223 		return 1;
5224 	}
5225 
5226 	return 0;
5227 }
5228 
5229 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5230 			void *val, int bytes)
5231 {
5232 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5233 }
5234 
5235 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5236 			 void *val, int bytes)
5237 {
5238 	return emulator_write_phys(vcpu, gpa, val, bytes);
5239 }
5240 
5241 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5242 {
5243 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5244 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5245 }
5246 
5247 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5248 			  void *val, int bytes)
5249 {
5250 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5251 	return X86EMUL_IO_NEEDED;
5252 }
5253 
5254 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5255 			   void *val, int bytes)
5256 {
5257 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5258 
5259 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5260 	return X86EMUL_CONTINUE;
5261 }
5262 
5263 static const struct read_write_emulator_ops read_emultor = {
5264 	.read_write_prepare = read_prepare,
5265 	.read_write_emulate = read_emulate,
5266 	.read_write_mmio = vcpu_mmio_read,
5267 	.read_write_exit_mmio = read_exit_mmio,
5268 };
5269 
5270 static const struct read_write_emulator_ops write_emultor = {
5271 	.read_write_emulate = write_emulate,
5272 	.read_write_mmio = write_mmio,
5273 	.read_write_exit_mmio = write_exit_mmio,
5274 	.write = true,
5275 };
5276 
5277 static int emulator_read_write_onepage(unsigned long addr, void *val,
5278 				       unsigned int bytes,
5279 				       struct x86_exception *exception,
5280 				       struct kvm_vcpu *vcpu,
5281 				       const struct read_write_emulator_ops *ops)
5282 {
5283 	gpa_t gpa;
5284 	int handled, ret;
5285 	bool write = ops->write;
5286 	struct kvm_mmio_fragment *frag;
5287 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5288 
5289 	/*
5290 	 * If the exit was due to a NPF we may already have a GPA.
5291 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5292 	 * Note, this cannot be used on string operations since string
5293 	 * operation using rep will only have the initial GPA from the NPF
5294 	 * occurred.
5295 	 */
5296 	if (vcpu->arch.gpa_available &&
5297 	    emulator_can_use_gpa(ctxt) &&
5298 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5299 		gpa = vcpu->arch.gpa_val;
5300 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5301 	} else {
5302 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5303 		if (ret < 0)
5304 			return X86EMUL_PROPAGATE_FAULT;
5305 	}
5306 
5307 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5308 		return X86EMUL_CONTINUE;
5309 
5310 	/*
5311 	 * Is this MMIO handled locally?
5312 	 */
5313 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5314 	if (handled == bytes)
5315 		return X86EMUL_CONTINUE;
5316 
5317 	gpa += handled;
5318 	bytes -= handled;
5319 	val += handled;
5320 
5321 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5322 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5323 	frag->gpa = gpa;
5324 	frag->data = val;
5325 	frag->len = bytes;
5326 	return X86EMUL_CONTINUE;
5327 }
5328 
5329 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5330 			unsigned long addr,
5331 			void *val, unsigned int bytes,
5332 			struct x86_exception *exception,
5333 			const struct read_write_emulator_ops *ops)
5334 {
5335 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5336 	gpa_t gpa;
5337 	int rc;
5338 
5339 	if (ops->read_write_prepare &&
5340 		  ops->read_write_prepare(vcpu, val, bytes))
5341 		return X86EMUL_CONTINUE;
5342 
5343 	vcpu->mmio_nr_fragments = 0;
5344 
5345 	/* Crossing a page boundary? */
5346 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5347 		int now;
5348 
5349 		now = -addr & ~PAGE_MASK;
5350 		rc = emulator_read_write_onepage(addr, val, now, exception,
5351 						 vcpu, ops);
5352 
5353 		if (rc != X86EMUL_CONTINUE)
5354 			return rc;
5355 		addr += now;
5356 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5357 			addr = (u32)addr;
5358 		val += now;
5359 		bytes -= now;
5360 	}
5361 
5362 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5363 					 vcpu, ops);
5364 	if (rc != X86EMUL_CONTINUE)
5365 		return rc;
5366 
5367 	if (!vcpu->mmio_nr_fragments)
5368 		return rc;
5369 
5370 	gpa = vcpu->mmio_fragments[0].gpa;
5371 
5372 	vcpu->mmio_needed = 1;
5373 	vcpu->mmio_cur_fragment = 0;
5374 
5375 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5376 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5377 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5378 	vcpu->run->mmio.phys_addr = gpa;
5379 
5380 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5381 }
5382 
5383 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5384 				  unsigned long addr,
5385 				  void *val,
5386 				  unsigned int bytes,
5387 				  struct x86_exception *exception)
5388 {
5389 	return emulator_read_write(ctxt, addr, val, bytes,
5390 				   exception, &read_emultor);
5391 }
5392 
5393 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5394 			    unsigned long addr,
5395 			    const void *val,
5396 			    unsigned int bytes,
5397 			    struct x86_exception *exception)
5398 {
5399 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5400 				   exception, &write_emultor);
5401 }
5402 
5403 #define CMPXCHG_TYPE(t, ptr, old, new) \
5404 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5405 
5406 #ifdef CONFIG_X86_64
5407 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5408 #else
5409 #  define CMPXCHG64(ptr, old, new) \
5410 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5411 #endif
5412 
5413 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5414 				     unsigned long addr,
5415 				     const void *old,
5416 				     const void *new,
5417 				     unsigned int bytes,
5418 				     struct x86_exception *exception)
5419 {
5420 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5421 	gpa_t gpa;
5422 	struct page *page;
5423 	char *kaddr;
5424 	bool exchanged;
5425 
5426 	/* guests cmpxchg8b have to be emulated atomically */
5427 	if (bytes > 8 || (bytes & (bytes - 1)))
5428 		goto emul_write;
5429 
5430 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5431 
5432 	if (gpa == UNMAPPED_GVA ||
5433 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5434 		goto emul_write;
5435 
5436 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5437 		goto emul_write;
5438 
5439 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5440 	if (is_error_page(page))
5441 		goto emul_write;
5442 
5443 	kaddr = kmap_atomic(page);
5444 	kaddr += offset_in_page(gpa);
5445 	switch (bytes) {
5446 	case 1:
5447 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5448 		break;
5449 	case 2:
5450 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5451 		break;
5452 	case 4:
5453 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5454 		break;
5455 	case 8:
5456 		exchanged = CMPXCHG64(kaddr, old, new);
5457 		break;
5458 	default:
5459 		BUG();
5460 	}
5461 	kunmap_atomic(kaddr);
5462 	kvm_release_page_dirty(page);
5463 
5464 	if (!exchanged)
5465 		return X86EMUL_CMPXCHG_FAILED;
5466 
5467 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5468 	kvm_page_track_write(vcpu, gpa, new, bytes);
5469 
5470 	return X86EMUL_CONTINUE;
5471 
5472 emul_write:
5473 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5474 
5475 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5476 }
5477 
5478 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5479 {
5480 	int r = 0, i;
5481 
5482 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5483 		if (vcpu->arch.pio.in)
5484 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5485 					    vcpu->arch.pio.size, pd);
5486 		else
5487 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5488 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5489 					     pd);
5490 		if (r)
5491 			break;
5492 		pd += vcpu->arch.pio.size;
5493 	}
5494 	return r;
5495 }
5496 
5497 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5498 			       unsigned short port, void *val,
5499 			       unsigned int count, bool in)
5500 {
5501 	vcpu->arch.pio.port = port;
5502 	vcpu->arch.pio.in = in;
5503 	vcpu->arch.pio.count  = count;
5504 	vcpu->arch.pio.size = size;
5505 
5506 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5507 		vcpu->arch.pio.count = 0;
5508 		return 1;
5509 	}
5510 
5511 	vcpu->run->exit_reason = KVM_EXIT_IO;
5512 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5513 	vcpu->run->io.size = size;
5514 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5515 	vcpu->run->io.count = count;
5516 	vcpu->run->io.port = port;
5517 
5518 	return 0;
5519 }
5520 
5521 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5522 				    int size, unsigned short port, void *val,
5523 				    unsigned int count)
5524 {
5525 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5526 	int ret;
5527 
5528 	if (vcpu->arch.pio.count)
5529 		goto data_avail;
5530 
5531 	memset(vcpu->arch.pio_data, 0, size * count);
5532 
5533 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5534 	if (ret) {
5535 data_avail:
5536 		memcpy(val, vcpu->arch.pio_data, size * count);
5537 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5538 		vcpu->arch.pio.count = 0;
5539 		return 1;
5540 	}
5541 
5542 	return 0;
5543 }
5544 
5545 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5546 				     int size, unsigned short port,
5547 				     const void *val, unsigned int count)
5548 {
5549 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5550 
5551 	memcpy(vcpu->arch.pio_data, val, size * count);
5552 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5553 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5554 }
5555 
5556 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5557 {
5558 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5559 }
5560 
5561 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5562 {
5563 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5564 }
5565 
5566 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5567 {
5568 	if (!need_emulate_wbinvd(vcpu))
5569 		return X86EMUL_CONTINUE;
5570 
5571 	if (kvm_x86_ops->has_wbinvd_exit()) {
5572 		int cpu = get_cpu();
5573 
5574 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5575 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5576 				wbinvd_ipi, NULL, 1);
5577 		put_cpu();
5578 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5579 	} else
5580 		wbinvd();
5581 	return X86EMUL_CONTINUE;
5582 }
5583 
5584 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5585 {
5586 	kvm_emulate_wbinvd_noskip(vcpu);
5587 	return kvm_skip_emulated_instruction(vcpu);
5588 }
5589 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5590 
5591 
5592 
5593 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5594 {
5595 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5596 }
5597 
5598 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5599 			   unsigned long *dest)
5600 {
5601 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5602 }
5603 
5604 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5605 			   unsigned long value)
5606 {
5607 
5608 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5609 }
5610 
5611 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5612 {
5613 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5614 }
5615 
5616 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5617 {
5618 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5619 	unsigned long value;
5620 
5621 	switch (cr) {
5622 	case 0:
5623 		value = kvm_read_cr0(vcpu);
5624 		break;
5625 	case 2:
5626 		value = vcpu->arch.cr2;
5627 		break;
5628 	case 3:
5629 		value = kvm_read_cr3(vcpu);
5630 		break;
5631 	case 4:
5632 		value = kvm_read_cr4(vcpu);
5633 		break;
5634 	case 8:
5635 		value = kvm_get_cr8(vcpu);
5636 		break;
5637 	default:
5638 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5639 		return 0;
5640 	}
5641 
5642 	return value;
5643 }
5644 
5645 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5646 {
5647 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5648 	int res = 0;
5649 
5650 	switch (cr) {
5651 	case 0:
5652 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5653 		break;
5654 	case 2:
5655 		vcpu->arch.cr2 = val;
5656 		break;
5657 	case 3:
5658 		res = kvm_set_cr3(vcpu, val);
5659 		break;
5660 	case 4:
5661 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5662 		break;
5663 	case 8:
5664 		res = kvm_set_cr8(vcpu, val);
5665 		break;
5666 	default:
5667 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5668 		res = -1;
5669 	}
5670 
5671 	return res;
5672 }
5673 
5674 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5675 {
5676 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5677 }
5678 
5679 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5680 {
5681 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5682 }
5683 
5684 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5685 {
5686 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5687 }
5688 
5689 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5690 {
5691 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5692 }
5693 
5694 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5695 {
5696 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5697 }
5698 
5699 static unsigned long emulator_get_cached_segment_base(
5700 	struct x86_emulate_ctxt *ctxt, int seg)
5701 {
5702 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5703 }
5704 
5705 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5706 				 struct desc_struct *desc, u32 *base3,
5707 				 int seg)
5708 {
5709 	struct kvm_segment var;
5710 
5711 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5712 	*selector = var.selector;
5713 
5714 	if (var.unusable) {
5715 		memset(desc, 0, sizeof(*desc));
5716 		if (base3)
5717 			*base3 = 0;
5718 		return false;
5719 	}
5720 
5721 	if (var.g)
5722 		var.limit >>= 12;
5723 	set_desc_limit(desc, var.limit);
5724 	set_desc_base(desc, (unsigned long)var.base);
5725 #ifdef CONFIG_X86_64
5726 	if (base3)
5727 		*base3 = var.base >> 32;
5728 #endif
5729 	desc->type = var.type;
5730 	desc->s = var.s;
5731 	desc->dpl = var.dpl;
5732 	desc->p = var.present;
5733 	desc->avl = var.avl;
5734 	desc->l = var.l;
5735 	desc->d = var.db;
5736 	desc->g = var.g;
5737 
5738 	return true;
5739 }
5740 
5741 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5742 				 struct desc_struct *desc, u32 base3,
5743 				 int seg)
5744 {
5745 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5746 	struct kvm_segment var;
5747 
5748 	var.selector = selector;
5749 	var.base = get_desc_base(desc);
5750 #ifdef CONFIG_X86_64
5751 	var.base |= ((u64)base3) << 32;
5752 #endif
5753 	var.limit = get_desc_limit(desc);
5754 	if (desc->g)
5755 		var.limit = (var.limit << 12) | 0xfff;
5756 	var.type = desc->type;
5757 	var.dpl = desc->dpl;
5758 	var.db = desc->d;
5759 	var.s = desc->s;
5760 	var.l = desc->l;
5761 	var.g = desc->g;
5762 	var.avl = desc->avl;
5763 	var.present = desc->p;
5764 	var.unusable = !var.present;
5765 	var.padding = 0;
5766 
5767 	kvm_set_segment(vcpu, &var, seg);
5768 	return;
5769 }
5770 
5771 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5772 			    u32 msr_index, u64 *pdata)
5773 {
5774 	struct msr_data msr;
5775 	int r;
5776 
5777 	msr.index = msr_index;
5778 	msr.host_initiated = false;
5779 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5780 	if (r)
5781 		return r;
5782 
5783 	*pdata = msr.data;
5784 	return 0;
5785 }
5786 
5787 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5788 			    u32 msr_index, u64 data)
5789 {
5790 	struct msr_data msr;
5791 
5792 	msr.data = data;
5793 	msr.index = msr_index;
5794 	msr.host_initiated = false;
5795 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5796 }
5797 
5798 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5799 {
5800 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5801 
5802 	return vcpu->arch.smbase;
5803 }
5804 
5805 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5806 {
5807 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5808 
5809 	vcpu->arch.smbase = smbase;
5810 }
5811 
5812 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5813 			      u32 pmc)
5814 {
5815 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5816 }
5817 
5818 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5819 			     u32 pmc, u64 *pdata)
5820 {
5821 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5822 }
5823 
5824 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5825 {
5826 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5827 }
5828 
5829 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5830 			      struct x86_instruction_info *info,
5831 			      enum x86_intercept_stage stage)
5832 {
5833 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5834 }
5835 
5836 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5837 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5838 {
5839 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5840 }
5841 
5842 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5843 {
5844 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5845 }
5846 
5847 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5848 {
5849 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5850 }
5851 
5852 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5853 {
5854 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5855 }
5856 
5857 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5858 {
5859 	return emul_to_vcpu(ctxt)->arch.hflags;
5860 }
5861 
5862 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5863 {
5864 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5865 }
5866 
5867 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5868 {
5869 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5870 }
5871 
5872 static const struct x86_emulate_ops emulate_ops = {
5873 	.read_gpr            = emulator_read_gpr,
5874 	.write_gpr           = emulator_write_gpr,
5875 	.read_std            = emulator_read_std,
5876 	.write_std           = emulator_write_std,
5877 	.read_phys           = kvm_read_guest_phys_system,
5878 	.fetch               = kvm_fetch_guest_virt,
5879 	.read_emulated       = emulator_read_emulated,
5880 	.write_emulated      = emulator_write_emulated,
5881 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5882 	.invlpg              = emulator_invlpg,
5883 	.pio_in_emulated     = emulator_pio_in_emulated,
5884 	.pio_out_emulated    = emulator_pio_out_emulated,
5885 	.get_segment         = emulator_get_segment,
5886 	.set_segment         = emulator_set_segment,
5887 	.get_cached_segment_base = emulator_get_cached_segment_base,
5888 	.get_gdt             = emulator_get_gdt,
5889 	.get_idt	     = emulator_get_idt,
5890 	.set_gdt             = emulator_set_gdt,
5891 	.set_idt	     = emulator_set_idt,
5892 	.get_cr              = emulator_get_cr,
5893 	.set_cr              = emulator_set_cr,
5894 	.cpl                 = emulator_get_cpl,
5895 	.get_dr              = emulator_get_dr,
5896 	.set_dr              = emulator_set_dr,
5897 	.get_smbase          = emulator_get_smbase,
5898 	.set_smbase          = emulator_set_smbase,
5899 	.set_msr             = emulator_set_msr,
5900 	.get_msr             = emulator_get_msr,
5901 	.check_pmc	     = emulator_check_pmc,
5902 	.read_pmc            = emulator_read_pmc,
5903 	.halt                = emulator_halt,
5904 	.wbinvd              = emulator_wbinvd,
5905 	.fix_hypercall       = emulator_fix_hypercall,
5906 	.intercept           = emulator_intercept,
5907 	.get_cpuid           = emulator_get_cpuid,
5908 	.set_nmi_mask        = emulator_set_nmi_mask,
5909 	.get_hflags          = emulator_get_hflags,
5910 	.set_hflags          = emulator_set_hflags,
5911 	.pre_leave_smm       = emulator_pre_leave_smm,
5912 };
5913 
5914 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5915 {
5916 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5917 	/*
5918 	 * an sti; sti; sequence only disable interrupts for the first
5919 	 * instruction. So, if the last instruction, be it emulated or
5920 	 * not, left the system with the INT_STI flag enabled, it
5921 	 * means that the last instruction is an sti. We should not
5922 	 * leave the flag on in this case. The same goes for mov ss
5923 	 */
5924 	if (int_shadow & mask)
5925 		mask = 0;
5926 	if (unlikely(int_shadow || mask)) {
5927 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5928 		if (!mask)
5929 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5930 	}
5931 }
5932 
5933 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5934 {
5935 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5936 	if (ctxt->exception.vector == PF_VECTOR)
5937 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5938 
5939 	if (ctxt->exception.error_code_valid)
5940 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5941 				      ctxt->exception.error_code);
5942 	else
5943 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5944 	return false;
5945 }
5946 
5947 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5948 {
5949 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5950 	int cs_db, cs_l;
5951 
5952 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5953 
5954 	ctxt->eflags = kvm_get_rflags(vcpu);
5955 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5956 
5957 	ctxt->eip = kvm_rip_read(vcpu);
5958 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5959 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5960 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5961 		     cs_db				? X86EMUL_MODE_PROT32 :
5962 							  X86EMUL_MODE_PROT16;
5963 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5964 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5965 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5966 
5967 	init_decode_cache(ctxt);
5968 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5969 }
5970 
5971 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5972 {
5973 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5974 	int ret;
5975 
5976 	init_emulate_ctxt(vcpu);
5977 
5978 	ctxt->op_bytes = 2;
5979 	ctxt->ad_bytes = 2;
5980 	ctxt->_eip = ctxt->eip + inc_eip;
5981 	ret = emulate_int_real(ctxt, irq);
5982 
5983 	if (ret != X86EMUL_CONTINUE)
5984 		return EMULATE_FAIL;
5985 
5986 	ctxt->eip = ctxt->_eip;
5987 	kvm_rip_write(vcpu, ctxt->eip);
5988 	kvm_set_rflags(vcpu, ctxt->eflags);
5989 
5990 	return EMULATE_DONE;
5991 }
5992 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5993 
5994 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5995 {
5996 	int r = EMULATE_DONE;
5997 
5998 	++vcpu->stat.insn_emulation_fail;
5999 	trace_kvm_emulate_insn_failed(vcpu);
6000 
6001 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6002 		return EMULATE_FAIL;
6003 
6004 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6005 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6006 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6007 		vcpu->run->internal.ndata = 0;
6008 		r = EMULATE_USER_EXIT;
6009 	}
6010 
6011 	kvm_queue_exception(vcpu, UD_VECTOR);
6012 
6013 	return r;
6014 }
6015 
6016 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6017 				  bool write_fault_to_shadow_pgtable,
6018 				  int emulation_type)
6019 {
6020 	gpa_t gpa = cr2;
6021 	kvm_pfn_t pfn;
6022 
6023 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6024 		return false;
6025 
6026 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6027 		return false;
6028 
6029 	if (!vcpu->arch.mmu->direct_map) {
6030 		/*
6031 		 * Write permission should be allowed since only
6032 		 * write access need to be emulated.
6033 		 */
6034 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6035 
6036 		/*
6037 		 * If the mapping is invalid in guest, let cpu retry
6038 		 * it to generate fault.
6039 		 */
6040 		if (gpa == UNMAPPED_GVA)
6041 			return true;
6042 	}
6043 
6044 	/*
6045 	 * Do not retry the unhandleable instruction if it faults on the
6046 	 * readonly host memory, otherwise it will goto a infinite loop:
6047 	 * retry instruction -> write #PF -> emulation fail -> retry
6048 	 * instruction -> ...
6049 	 */
6050 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6051 
6052 	/*
6053 	 * If the instruction failed on the error pfn, it can not be fixed,
6054 	 * report the error to userspace.
6055 	 */
6056 	if (is_error_noslot_pfn(pfn))
6057 		return false;
6058 
6059 	kvm_release_pfn_clean(pfn);
6060 
6061 	/* The instructions are well-emulated on direct mmu. */
6062 	if (vcpu->arch.mmu->direct_map) {
6063 		unsigned int indirect_shadow_pages;
6064 
6065 		spin_lock(&vcpu->kvm->mmu_lock);
6066 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6067 		spin_unlock(&vcpu->kvm->mmu_lock);
6068 
6069 		if (indirect_shadow_pages)
6070 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6071 
6072 		return true;
6073 	}
6074 
6075 	/*
6076 	 * if emulation was due to access to shadowed page table
6077 	 * and it failed try to unshadow page and re-enter the
6078 	 * guest to let CPU execute the instruction.
6079 	 */
6080 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6081 
6082 	/*
6083 	 * If the access faults on its page table, it can not
6084 	 * be fixed by unprotecting shadow page and it should
6085 	 * be reported to userspace.
6086 	 */
6087 	return !write_fault_to_shadow_pgtable;
6088 }
6089 
6090 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6091 			      unsigned long cr2,  int emulation_type)
6092 {
6093 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6094 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6095 
6096 	last_retry_eip = vcpu->arch.last_retry_eip;
6097 	last_retry_addr = vcpu->arch.last_retry_addr;
6098 
6099 	/*
6100 	 * If the emulation is caused by #PF and it is non-page_table
6101 	 * writing instruction, it means the VM-EXIT is caused by shadow
6102 	 * page protected, we can zap the shadow page and retry this
6103 	 * instruction directly.
6104 	 *
6105 	 * Note: if the guest uses a non-page-table modifying instruction
6106 	 * on the PDE that points to the instruction, then we will unmap
6107 	 * the instruction and go to an infinite loop. So, we cache the
6108 	 * last retried eip and the last fault address, if we meet the eip
6109 	 * and the address again, we can break out of the potential infinite
6110 	 * loop.
6111 	 */
6112 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6113 
6114 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6115 		return false;
6116 
6117 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6118 		return false;
6119 
6120 	if (x86_page_table_writing_insn(ctxt))
6121 		return false;
6122 
6123 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6124 		return false;
6125 
6126 	vcpu->arch.last_retry_eip = ctxt->eip;
6127 	vcpu->arch.last_retry_addr = cr2;
6128 
6129 	if (!vcpu->arch.mmu->direct_map)
6130 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6131 
6132 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6133 
6134 	return true;
6135 }
6136 
6137 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6138 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6139 
6140 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6141 {
6142 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6143 		/* This is a good place to trace that we are exiting SMM.  */
6144 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6145 
6146 		/* Process a latched INIT or SMI, if any.  */
6147 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6148 	}
6149 
6150 	kvm_mmu_reset_context(vcpu);
6151 }
6152 
6153 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6154 {
6155 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
6156 
6157 	vcpu->arch.hflags = emul_flags;
6158 
6159 	if (changed & HF_SMM_MASK)
6160 		kvm_smm_changed(vcpu);
6161 }
6162 
6163 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6164 				unsigned long *db)
6165 {
6166 	u32 dr6 = 0;
6167 	int i;
6168 	u32 enable, rwlen;
6169 
6170 	enable = dr7;
6171 	rwlen = dr7 >> 16;
6172 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6173 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6174 			dr6 |= (1 << i);
6175 	return dr6;
6176 }
6177 
6178 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6179 {
6180 	struct kvm_run *kvm_run = vcpu->run;
6181 
6182 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6183 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6184 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6185 		kvm_run->debug.arch.exception = DB_VECTOR;
6186 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6187 		*r = EMULATE_USER_EXIT;
6188 	} else {
6189 		kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6190 	}
6191 }
6192 
6193 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6194 {
6195 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6196 	int r = EMULATE_DONE;
6197 
6198 	kvm_x86_ops->skip_emulated_instruction(vcpu);
6199 
6200 	/*
6201 	 * rflags is the old, "raw" value of the flags.  The new value has
6202 	 * not been saved yet.
6203 	 *
6204 	 * This is correct even for TF set by the guest, because "the
6205 	 * processor will not generate this exception after the instruction
6206 	 * that sets the TF flag".
6207 	 */
6208 	if (unlikely(rflags & X86_EFLAGS_TF))
6209 		kvm_vcpu_do_singlestep(vcpu, &r);
6210 	return r == EMULATE_DONE;
6211 }
6212 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6213 
6214 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6215 {
6216 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6217 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6218 		struct kvm_run *kvm_run = vcpu->run;
6219 		unsigned long eip = kvm_get_linear_rip(vcpu);
6220 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6221 					   vcpu->arch.guest_debug_dr7,
6222 					   vcpu->arch.eff_db);
6223 
6224 		if (dr6 != 0) {
6225 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6226 			kvm_run->debug.arch.pc = eip;
6227 			kvm_run->debug.arch.exception = DB_VECTOR;
6228 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6229 			*r = EMULATE_USER_EXIT;
6230 			return true;
6231 		}
6232 	}
6233 
6234 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6235 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6236 		unsigned long eip = kvm_get_linear_rip(vcpu);
6237 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6238 					   vcpu->arch.dr7,
6239 					   vcpu->arch.db);
6240 
6241 		if (dr6 != 0) {
6242 			vcpu->arch.dr6 &= ~15;
6243 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6244 			kvm_queue_exception(vcpu, DB_VECTOR);
6245 			*r = EMULATE_DONE;
6246 			return true;
6247 		}
6248 	}
6249 
6250 	return false;
6251 }
6252 
6253 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6254 {
6255 	switch (ctxt->opcode_len) {
6256 	case 1:
6257 		switch (ctxt->b) {
6258 		case 0xe4:	/* IN */
6259 		case 0xe5:
6260 		case 0xec:
6261 		case 0xed:
6262 		case 0xe6:	/* OUT */
6263 		case 0xe7:
6264 		case 0xee:
6265 		case 0xef:
6266 		case 0x6c:	/* INS */
6267 		case 0x6d:
6268 		case 0x6e:	/* OUTS */
6269 		case 0x6f:
6270 			return true;
6271 		}
6272 		break;
6273 	case 2:
6274 		switch (ctxt->b) {
6275 		case 0x33:	/* RDPMC */
6276 			return true;
6277 		}
6278 		break;
6279 	}
6280 
6281 	return false;
6282 }
6283 
6284 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6285 			    unsigned long cr2,
6286 			    int emulation_type,
6287 			    void *insn,
6288 			    int insn_len)
6289 {
6290 	int r;
6291 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6292 	bool writeback = true;
6293 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6294 
6295 	vcpu->arch.l1tf_flush_l1d = true;
6296 
6297 	/*
6298 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6299 	 * never reused.
6300 	 */
6301 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6302 	kvm_clear_exception_queue(vcpu);
6303 
6304 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6305 		init_emulate_ctxt(vcpu);
6306 
6307 		/*
6308 		 * We will reenter on the same instruction since
6309 		 * we do not set complete_userspace_io.  This does not
6310 		 * handle watchpoints yet, those would be handled in
6311 		 * the emulate_ops.
6312 		 */
6313 		if (!(emulation_type & EMULTYPE_SKIP) &&
6314 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6315 			return r;
6316 
6317 		ctxt->interruptibility = 0;
6318 		ctxt->have_exception = false;
6319 		ctxt->exception.vector = -1;
6320 		ctxt->perm_ok = false;
6321 
6322 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6323 
6324 		r = x86_decode_insn(ctxt, insn, insn_len);
6325 
6326 		trace_kvm_emulate_insn_start(vcpu);
6327 		++vcpu->stat.insn_emulation;
6328 		if (r != EMULATION_OK)  {
6329 			if (emulation_type & EMULTYPE_TRAP_UD)
6330 				return EMULATE_FAIL;
6331 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6332 						emulation_type))
6333 				return EMULATE_DONE;
6334 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6335 				return EMULATE_DONE;
6336 			if (emulation_type & EMULTYPE_SKIP)
6337 				return EMULATE_FAIL;
6338 			return handle_emulation_failure(vcpu, emulation_type);
6339 		}
6340 	}
6341 
6342 	if ((emulation_type & EMULTYPE_VMWARE) &&
6343 	    !is_vmware_backdoor_opcode(ctxt))
6344 		return EMULATE_FAIL;
6345 
6346 	if (emulation_type & EMULTYPE_SKIP) {
6347 		kvm_rip_write(vcpu, ctxt->_eip);
6348 		if (ctxt->eflags & X86_EFLAGS_RF)
6349 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6350 		return EMULATE_DONE;
6351 	}
6352 
6353 	if (retry_instruction(ctxt, cr2, emulation_type))
6354 		return EMULATE_DONE;
6355 
6356 	/* this is needed for vmware backdoor interface to work since it
6357 	   changes registers values  during IO operation */
6358 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6359 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6360 		emulator_invalidate_register_cache(ctxt);
6361 	}
6362 
6363 restart:
6364 	/* Save the faulting GPA (cr2) in the address field */
6365 	ctxt->exception.address = cr2;
6366 
6367 	r = x86_emulate_insn(ctxt);
6368 
6369 	if (r == EMULATION_INTERCEPTED)
6370 		return EMULATE_DONE;
6371 
6372 	if (r == EMULATION_FAILED) {
6373 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6374 					emulation_type))
6375 			return EMULATE_DONE;
6376 
6377 		return handle_emulation_failure(vcpu, emulation_type);
6378 	}
6379 
6380 	if (ctxt->have_exception) {
6381 		r = EMULATE_DONE;
6382 		if (inject_emulated_exception(vcpu))
6383 			return r;
6384 	} else if (vcpu->arch.pio.count) {
6385 		if (!vcpu->arch.pio.in) {
6386 			/* FIXME: return into emulator if single-stepping.  */
6387 			vcpu->arch.pio.count = 0;
6388 		} else {
6389 			writeback = false;
6390 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6391 		}
6392 		r = EMULATE_USER_EXIT;
6393 	} else if (vcpu->mmio_needed) {
6394 		if (!vcpu->mmio_is_write)
6395 			writeback = false;
6396 		r = EMULATE_USER_EXIT;
6397 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6398 	} else if (r == EMULATION_RESTART)
6399 		goto restart;
6400 	else
6401 		r = EMULATE_DONE;
6402 
6403 	if (writeback) {
6404 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6405 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6406 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6407 		kvm_rip_write(vcpu, ctxt->eip);
6408 		if (r == EMULATE_DONE &&
6409 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6410 			kvm_vcpu_do_singlestep(vcpu, &r);
6411 		if (!ctxt->have_exception ||
6412 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6413 			__kvm_set_rflags(vcpu, ctxt->eflags);
6414 
6415 		/*
6416 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6417 		 * do nothing, and it will be requested again as soon as
6418 		 * the shadow expires.  But we still need to check here,
6419 		 * because POPF has no interrupt shadow.
6420 		 */
6421 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6422 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6423 	} else
6424 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6425 
6426 	return r;
6427 }
6428 
6429 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6430 {
6431 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6432 }
6433 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6434 
6435 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6436 					void *insn, int insn_len)
6437 {
6438 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6439 }
6440 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6441 
6442 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6443 			    unsigned short port)
6444 {
6445 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6446 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6447 					    size, port, &val, 1);
6448 	/* do not return to emulator after return from userspace */
6449 	vcpu->arch.pio.count = 0;
6450 	return ret;
6451 }
6452 
6453 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6454 {
6455 	unsigned long val;
6456 
6457 	/* We should only ever be called with arch.pio.count equal to 1 */
6458 	BUG_ON(vcpu->arch.pio.count != 1);
6459 
6460 	/* For size less than 4 we merge, else we zero extend */
6461 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6462 					: 0;
6463 
6464 	/*
6465 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6466 	 * the copy and tracing
6467 	 */
6468 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6469 				 vcpu->arch.pio.port, &val, 1);
6470 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6471 
6472 	return 1;
6473 }
6474 
6475 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6476 			   unsigned short port)
6477 {
6478 	unsigned long val;
6479 	int ret;
6480 
6481 	/* For size less than 4 we merge, else we zero extend */
6482 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6483 
6484 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6485 				       &val, 1);
6486 	if (ret) {
6487 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6488 		return ret;
6489 	}
6490 
6491 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6492 
6493 	return 0;
6494 }
6495 
6496 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6497 {
6498 	int ret = kvm_skip_emulated_instruction(vcpu);
6499 
6500 	/*
6501 	 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6502 	 * KVM_EXIT_DEBUG here.
6503 	 */
6504 	if (in)
6505 		return kvm_fast_pio_in(vcpu, size, port) && ret;
6506 	else
6507 		return kvm_fast_pio_out(vcpu, size, port) && ret;
6508 }
6509 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6510 
6511 static int kvmclock_cpu_down_prep(unsigned int cpu)
6512 {
6513 	__this_cpu_write(cpu_tsc_khz, 0);
6514 	return 0;
6515 }
6516 
6517 static void tsc_khz_changed(void *data)
6518 {
6519 	struct cpufreq_freqs *freq = data;
6520 	unsigned long khz = 0;
6521 
6522 	if (data)
6523 		khz = freq->new;
6524 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6525 		khz = cpufreq_quick_get(raw_smp_processor_id());
6526 	if (!khz)
6527 		khz = tsc_khz;
6528 	__this_cpu_write(cpu_tsc_khz, khz);
6529 }
6530 
6531 #ifdef CONFIG_X86_64
6532 static void kvm_hyperv_tsc_notifier(void)
6533 {
6534 	struct kvm *kvm;
6535 	struct kvm_vcpu *vcpu;
6536 	int cpu;
6537 
6538 	spin_lock(&kvm_lock);
6539 	list_for_each_entry(kvm, &vm_list, vm_list)
6540 		kvm_make_mclock_inprogress_request(kvm);
6541 
6542 	hyperv_stop_tsc_emulation();
6543 
6544 	/* TSC frequency always matches when on Hyper-V */
6545 	for_each_present_cpu(cpu)
6546 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6547 	kvm_max_guest_tsc_khz = tsc_khz;
6548 
6549 	list_for_each_entry(kvm, &vm_list, vm_list) {
6550 		struct kvm_arch *ka = &kvm->arch;
6551 
6552 		spin_lock(&ka->pvclock_gtod_sync_lock);
6553 
6554 		pvclock_update_vm_gtod_copy(kvm);
6555 
6556 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6557 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6558 
6559 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6560 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6561 
6562 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6563 	}
6564 	spin_unlock(&kvm_lock);
6565 }
6566 #endif
6567 
6568 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6569 				     void *data)
6570 {
6571 	struct cpufreq_freqs *freq = data;
6572 	struct kvm *kvm;
6573 	struct kvm_vcpu *vcpu;
6574 	int i, send_ipi = 0;
6575 
6576 	/*
6577 	 * We allow guests to temporarily run on slowing clocks,
6578 	 * provided we notify them after, or to run on accelerating
6579 	 * clocks, provided we notify them before.  Thus time never
6580 	 * goes backwards.
6581 	 *
6582 	 * However, we have a problem.  We can't atomically update
6583 	 * the frequency of a given CPU from this function; it is
6584 	 * merely a notifier, which can be called from any CPU.
6585 	 * Changing the TSC frequency at arbitrary points in time
6586 	 * requires a recomputation of local variables related to
6587 	 * the TSC for each VCPU.  We must flag these local variables
6588 	 * to be updated and be sure the update takes place with the
6589 	 * new frequency before any guests proceed.
6590 	 *
6591 	 * Unfortunately, the combination of hotplug CPU and frequency
6592 	 * change creates an intractable locking scenario; the order
6593 	 * of when these callouts happen is undefined with respect to
6594 	 * CPU hotplug, and they can race with each other.  As such,
6595 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6596 	 * undefined; you can actually have a CPU frequency change take
6597 	 * place in between the computation of X and the setting of the
6598 	 * variable.  To protect against this problem, all updates of
6599 	 * the per_cpu tsc_khz variable are done in an interrupt
6600 	 * protected IPI, and all callers wishing to update the value
6601 	 * must wait for a synchronous IPI to complete (which is trivial
6602 	 * if the caller is on the CPU already).  This establishes the
6603 	 * necessary total order on variable updates.
6604 	 *
6605 	 * Note that because a guest time update may take place
6606 	 * anytime after the setting of the VCPU's request bit, the
6607 	 * correct TSC value must be set before the request.  However,
6608 	 * to ensure the update actually makes it to any guest which
6609 	 * starts running in hardware virtualization between the set
6610 	 * and the acquisition of the spinlock, we must also ping the
6611 	 * CPU after setting the request bit.
6612 	 *
6613 	 */
6614 
6615 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6616 		return 0;
6617 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6618 		return 0;
6619 
6620 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6621 
6622 	spin_lock(&kvm_lock);
6623 	list_for_each_entry(kvm, &vm_list, vm_list) {
6624 		kvm_for_each_vcpu(i, vcpu, kvm) {
6625 			if (vcpu->cpu != freq->cpu)
6626 				continue;
6627 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6628 			if (vcpu->cpu != smp_processor_id())
6629 				send_ipi = 1;
6630 		}
6631 	}
6632 	spin_unlock(&kvm_lock);
6633 
6634 	if (freq->old < freq->new && send_ipi) {
6635 		/*
6636 		 * We upscale the frequency.  Must make the guest
6637 		 * doesn't see old kvmclock values while running with
6638 		 * the new frequency, otherwise we risk the guest sees
6639 		 * time go backwards.
6640 		 *
6641 		 * In case we update the frequency for another cpu
6642 		 * (which might be in guest context) send an interrupt
6643 		 * to kick the cpu out of guest context.  Next time
6644 		 * guest context is entered kvmclock will be updated,
6645 		 * so the guest will not see stale values.
6646 		 */
6647 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6648 	}
6649 	return 0;
6650 }
6651 
6652 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6653 	.notifier_call  = kvmclock_cpufreq_notifier
6654 };
6655 
6656 static int kvmclock_cpu_online(unsigned int cpu)
6657 {
6658 	tsc_khz_changed(NULL);
6659 	return 0;
6660 }
6661 
6662 static void kvm_timer_init(void)
6663 {
6664 	max_tsc_khz = tsc_khz;
6665 
6666 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6667 #ifdef CONFIG_CPU_FREQ
6668 		struct cpufreq_policy policy;
6669 		int cpu;
6670 
6671 		memset(&policy, 0, sizeof(policy));
6672 		cpu = get_cpu();
6673 		cpufreq_get_policy(&policy, cpu);
6674 		if (policy.cpuinfo.max_freq)
6675 			max_tsc_khz = policy.cpuinfo.max_freq;
6676 		put_cpu();
6677 #endif
6678 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6679 					  CPUFREQ_TRANSITION_NOTIFIER);
6680 	}
6681 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6682 
6683 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6684 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6685 }
6686 
6687 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6688 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6689 
6690 int kvm_is_in_guest(void)
6691 {
6692 	return __this_cpu_read(current_vcpu) != NULL;
6693 }
6694 
6695 static int kvm_is_user_mode(void)
6696 {
6697 	int user_mode = 3;
6698 
6699 	if (__this_cpu_read(current_vcpu))
6700 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6701 
6702 	return user_mode != 0;
6703 }
6704 
6705 static unsigned long kvm_get_guest_ip(void)
6706 {
6707 	unsigned long ip = 0;
6708 
6709 	if (__this_cpu_read(current_vcpu))
6710 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6711 
6712 	return ip;
6713 }
6714 
6715 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6716 	.is_in_guest		= kvm_is_in_guest,
6717 	.is_user_mode		= kvm_is_user_mode,
6718 	.get_guest_ip		= kvm_get_guest_ip,
6719 };
6720 
6721 static void kvm_set_mmio_spte_mask(void)
6722 {
6723 	u64 mask;
6724 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6725 
6726 	/*
6727 	 * Set the reserved bits and the present bit of an paging-structure
6728 	 * entry to generate page fault with PFER.RSV = 1.
6729 	 */
6730 
6731 	/*
6732 	 * Mask the uppermost physical address bit, which would be reserved as
6733 	 * long as the supported physical address width is less than 52.
6734 	 */
6735 	mask = 1ull << 51;
6736 
6737 	/* Set the present bit. */
6738 	mask |= 1ull;
6739 
6740 	/*
6741 	 * If reserved bit is not supported, clear the present bit to disable
6742 	 * mmio page fault.
6743 	 */
6744 	if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6745 		mask &= ~1ull;
6746 
6747 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6748 }
6749 
6750 #ifdef CONFIG_X86_64
6751 static void pvclock_gtod_update_fn(struct work_struct *work)
6752 {
6753 	struct kvm *kvm;
6754 
6755 	struct kvm_vcpu *vcpu;
6756 	int i;
6757 
6758 	spin_lock(&kvm_lock);
6759 	list_for_each_entry(kvm, &vm_list, vm_list)
6760 		kvm_for_each_vcpu(i, vcpu, kvm)
6761 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6762 	atomic_set(&kvm_guest_has_master_clock, 0);
6763 	spin_unlock(&kvm_lock);
6764 }
6765 
6766 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6767 
6768 /*
6769  * Notification about pvclock gtod data update.
6770  */
6771 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6772 			       void *priv)
6773 {
6774 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6775 	struct timekeeper *tk = priv;
6776 
6777 	update_pvclock_gtod(tk);
6778 
6779 	/* disable master clock if host does not trust, or does not
6780 	 * use, TSC based clocksource.
6781 	 */
6782 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6783 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6784 		queue_work(system_long_wq, &pvclock_gtod_work);
6785 
6786 	return 0;
6787 }
6788 
6789 static struct notifier_block pvclock_gtod_notifier = {
6790 	.notifier_call = pvclock_gtod_notify,
6791 };
6792 #endif
6793 
6794 int kvm_arch_init(void *opaque)
6795 {
6796 	int r;
6797 	struct kvm_x86_ops *ops = opaque;
6798 
6799 	if (kvm_x86_ops) {
6800 		printk(KERN_ERR "kvm: already loaded the other module\n");
6801 		r = -EEXIST;
6802 		goto out;
6803 	}
6804 
6805 	if (!ops->cpu_has_kvm_support()) {
6806 		printk(KERN_ERR "kvm: no hardware support\n");
6807 		r = -EOPNOTSUPP;
6808 		goto out;
6809 	}
6810 	if (ops->disabled_by_bios()) {
6811 		printk(KERN_ERR "kvm: disabled by bios\n");
6812 		r = -EOPNOTSUPP;
6813 		goto out;
6814 	}
6815 
6816 	r = -ENOMEM;
6817 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6818 	if (!shared_msrs) {
6819 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6820 		goto out;
6821 	}
6822 
6823 	r = kvm_mmu_module_init();
6824 	if (r)
6825 		goto out_free_percpu;
6826 
6827 	kvm_set_mmio_spte_mask();
6828 
6829 	kvm_x86_ops = ops;
6830 
6831 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6832 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6833 			PT_PRESENT_MASK, 0, sme_me_mask);
6834 	kvm_timer_init();
6835 
6836 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6837 
6838 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6839 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6840 
6841 	kvm_lapic_init();
6842 #ifdef CONFIG_X86_64
6843 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6844 
6845 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6846 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6847 #endif
6848 
6849 	return 0;
6850 
6851 out_free_percpu:
6852 	free_percpu(shared_msrs);
6853 out:
6854 	return r;
6855 }
6856 
6857 void kvm_arch_exit(void)
6858 {
6859 #ifdef CONFIG_X86_64
6860 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6861 		clear_hv_tscchange_cb();
6862 #endif
6863 	kvm_lapic_exit();
6864 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6865 
6866 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6867 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6868 					    CPUFREQ_TRANSITION_NOTIFIER);
6869 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6870 #ifdef CONFIG_X86_64
6871 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6872 #endif
6873 	kvm_x86_ops = NULL;
6874 	kvm_mmu_module_exit();
6875 	free_percpu(shared_msrs);
6876 }
6877 
6878 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6879 {
6880 	++vcpu->stat.halt_exits;
6881 	if (lapic_in_kernel(vcpu)) {
6882 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6883 		return 1;
6884 	} else {
6885 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6886 		return 0;
6887 	}
6888 }
6889 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6890 
6891 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6892 {
6893 	int ret = kvm_skip_emulated_instruction(vcpu);
6894 	/*
6895 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6896 	 * KVM_EXIT_DEBUG here.
6897 	 */
6898 	return kvm_vcpu_halt(vcpu) && ret;
6899 }
6900 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6901 
6902 #ifdef CONFIG_X86_64
6903 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6904 			        unsigned long clock_type)
6905 {
6906 	struct kvm_clock_pairing clock_pairing;
6907 	struct timespec64 ts;
6908 	u64 cycle;
6909 	int ret;
6910 
6911 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6912 		return -KVM_EOPNOTSUPP;
6913 
6914 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6915 		return -KVM_EOPNOTSUPP;
6916 
6917 	clock_pairing.sec = ts.tv_sec;
6918 	clock_pairing.nsec = ts.tv_nsec;
6919 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6920 	clock_pairing.flags = 0;
6921 
6922 	ret = 0;
6923 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6924 			    sizeof(struct kvm_clock_pairing)))
6925 		ret = -KVM_EFAULT;
6926 
6927 	return ret;
6928 }
6929 #endif
6930 
6931 /*
6932  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6933  *
6934  * @apicid - apicid of vcpu to be kicked.
6935  */
6936 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6937 {
6938 	struct kvm_lapic_irq lapic_irq;
6939 
6940 	lapic_irq.shorthand = 0;
6941 	lapic_irq.dest_mode = 0;
6942 	lapic_irq.level = 0;
6943 	lapic_irq.dest_id = apicid;
6944 	lapic_irq.msi_redir_hint = false;
6945 
6946 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6947 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6948 }
6949 
6950 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6951 {
6952 	vcpu->arch.apicv_active = false;
6953 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6954 }
6955 
6956 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6957 {
6958 	unsigned long nr, a0, a1, a2, a3, ret;
6959 	int op_64_bit;
6960 
6961 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6962 		return kvm_hv_hypercall(vcpu);
6963 
6964 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6965 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6966 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6967 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6968 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6969 
6970 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6971 
6972 	op_64_bit = is_64_bit_mode(vcpu);
6973 	if (!op_64_bit) {
6974 		nr &= 0xFFFFFFFF;
6975 		a0 &= 0xFFFFFFFF;
6976 		a1 &= 0xFFFFFFFF;
6977 		a2 &= 0xFFFFFFFF;
6978 		a3 &= 0xFFFFFFFF;
6979 	}
6980 
6981 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6982 		ret = -KVM_EPERM;
6983 		goto out;
6984 	}
6985 
6986 	switch (nr) {
6987 	case KVM_HC_VAPIC_POLL_IRQ:
6988 		ret = 0;
6989 		break;
6990 	case KVM_HC_KICK_CPU:
6991 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6992 		ret = 0;
6993 		break;
6994 #ifdef CONFIG_X86_64
6995 	case KVM_HC_CLOCK_PAIRING:
6996 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6997 		break;
6998 	case KVM_HC_SEND_IPI:
6999 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7000 		break;
7001 #endif
7002 	default:
7003 		ret = -KVM_ENOSYS;
7004 		break;
7005 	}
7006 out:
7007 	if (!op_64_bit)
7008 		ret = (u32)ret;
7009 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7010 
7011 	++vcpu->stat.hypercalls;
7012 	return kvm_skip_emulated_instruction(vcpu);
7013 }
7014 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7015 
7016 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7017 {
7018 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7019 	char instruction[3];
7020 	unsigned long rip = kvm_rip_read(vcpu);
7021 
7022 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7023 
7024 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7025 		&ctxt->exception);
7026 }
7027 
7028 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7029 {
7030 	return vcpu->run->request_interrupt_window &&
7031 		likely(!pic_in_kernel(vcpu->kvm));
7032 }
7033 
7034 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7035 {
7036 	struct kvm_run *kvm_run = vcpu->run;
7037 
7038 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7039 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7040 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7041 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7042 	kvm_run->ready_for_interrupt_injection =
7043 		pic_in_kernel(vcpu->kvm) ||
7044 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7045 }
7046 
7047 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7048 {
7049 	int max_irr, tpr;
7050 
7051 	if (!kvm_x86_ops->update_cr8_intercept)
7052 		return;
7053 
7054 	if (!lapic_in_kernel(vcpu))
7055 		return;
7056 
7057 	if (vcpu->arch.apicv_active)
7058 		return;
7059 
7060 	if (!vcpu->arch.apic->vapic_addr)
7061 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7062 	else
7063 		max_irr = -1;
7064 
7065 	if (max_irr != -1)
7066 		max_irr >>= 4;
7067 
7068 	tpr = kvm_lapic_get_cr8(vcpu);
7069 
7070 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7071 }
7072 
7073 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7074 {
7075 	int r;
7076 
7077 	/* try to reinject previous events if any */
7078 
7079 	if (vcpu->arch.exception.injected)
7080 		kvm_x86_ops->queue_exception(vcpu);
7081 	/*
7082 	 * Do not inject an NMI or interrupt if there is a pending
7083 	 * exception.  Exceptions and interrupts are recognized at
7084 	 * instruction boundaries, i.e. the start of an instruction.
7085 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7086 	 * NMIs and interrupts, i.e. traps are recognized before an
7087 	 * NMI/interrupt that's pending on the same instruction.
7088 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7089 	 * priority, but are only generated (pended) during instruction
7090 	 * execution, i.e. a pending fault-like exception means the
7091 	 * fault occurred on the *previous* instruction and must be
7092 	 * serviced prior to recognizing any new events in order to
7093 	 * fully complete the previous instruction.
7094 	 */
7095 	else if (!vcpu->arch.exception.pending) {
7096 		if (vcpu->arch.nmi_injected)
7097 			kvm_x86_ops->set_nmi(vcpu);
7098 		else if (vcpu->arch.interrupt.injected)
7099 			kvm_x86_ops->set_irq(vcpu);
7100 	}
7101 
7102 	/*
7103 	 * Call check_nested_events() even if we reinjected a previous event
7104 	 * in order for caller to determine if it should require immediate-exit
7105 	 * from L2 to L1 due to pending L1 events which require exit
7106 	 * from L2 to L1.
7107 	 */
7108 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7109 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7110 		if (r != 0)
7111 			return r;
7112 	}
7113 
7114 	/* try to inject new event if pending */
7115 	if (vcpu->arch.exception.pending) {
7116 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7117 					vcpu->arch.exception.has_error_code,
7118 					vcpu->arch.exception.error_code);
7119 
7120 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7121 		vcpu->arch.exception.pending = false;
7122 		vcpu->arch.exception.injected = true;
7123 
7124 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7125 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7126 					     X86_EFLAGS_RF);
7127 
7128 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7129 			/*
7130 			 * This code assumes that nSVM doesn't use
7131 			 * check_nested_events(). If it does, the
7132 			 * DR6/DR7 changes should happen before L1
7133 			 * gets a #VMEXIT for an intercepted #DB in
7134 			 * L2.  (Under VMX, on the other hand, the
7135 			 * DR6/DR7 changes should not happen in the
7136 			 * event of a VM-exit to L1 for an intercepted
7137 			 * #DB in L2.)
7138 			 */
7139 			kvm_deliver_exception_payload(vcpu);
7140 			if (vcpu->arch.dr7 & DR7_GD) {
7141 				vcpu->arch.dr7 &= ~DR7_GD;
7142 				kvm_update_dr7(vcpu);
7143 			}
7144 		}
7145 
7146 		kvm_x86_ops->queue_exception(vcpu);
7147 	}
7148 
7149 	/* Don't consider new event if we re-injected an event */
7150 	if (kvm_event_needs_reinjection(vcpu))
7151 		return 0;
7152 
7153 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7154 	    kvm_x86_ops->smi_allowed(vcpu)) {
7155 		vcpu->arch.smi_pending = false;
7156 		++vcpu->arch.smi_count;
7157 		enter_smm(vcpu);
7158 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7159 		--vcpu->arch.nmi_pending;
7160 		vcpu->arch.nmi_injected = true;
7161 		kvm_x86_ops->set_nmi(vcpu);
7162 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7163 		/*
7164 		 * Because interrupts can be injected asynchronously, we are
7165 		 * calling check_nested_events again here to avoid a race condition.
7166 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7167 		 * proposal and current concerns.  Perhaps we should be setting
7168 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7169 		 */
7170 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7171 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7172 			if (r != 0)
7173 				return r;
7174 		}
7175 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7176 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7177 					    false);
7178 			kvm_x86_ops->set_irq(vcpu);
7179 		}
7180 	}
7181 
7182 	return 0;
7183 }
7184 
7185 static void process_nmi(struct kvm_vcpu *vcpu)
7186 {
7187 	unsigned limit = 2;
7188 
7189 	/*
7190 	 * x86 is limited to one NMI running, and one NMI pending after it.
7191 	 * If an NMI is already in progress, limit further NMIs to just one.
7192 	 * Otherwise, allow two (and we'll inject the first one immediately).
7193 	 */
7194 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7195 		limit = 1;
7196 
7197 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7198 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7199 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7200 }
7201 
7202 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7203 {
7204 	u32 flags = 0;
7205 	flags |= seg->g       << 23;
7206 	flags |= seg->db      << 22;
7207 	flags |= seg->l       << 21;
7208 	flags |= seg->avl     << 20;
7209 	flags |= seg->present << 15;
7210 	flags |= seg->dpl     << 13;
7211 	flags |= seg->s       << 12;
7212 	flags |= seg->type    << 8;
7213 	return flags;
7214 }
7215 
7216 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7217 {
7218 	struct kvm_segment seg;
7219 	int offset;
7220 
7221 	kvm_get_segment(vcpu, &seg, n);
7222 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7223 
7224 	if (n < 3)
7225 		offset = 0x7f84 + n * 12;
7226 	else
7227 		offset = 0x7f2c + (n - 3) * 12;
7228 
7229 	put_smstate(u32, buf, offset + 8, seg.base);
7230 	put_smstate(u32, buf, offset + 4, seg.limit);
7231 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7232 }
7233 
7234 #ifdef CONFIG_X86_64
7235 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7236 {
7237 	struct kvm_segment seg;
7238 	int offset;
7239 	u16 flags;
7240 
7241 	kvm_get_segment(vcpu, &seg, n);
7242 	offset = 0x7e00 + n * 16;
7243 
7244 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7245 	put_smstate(u16, buf, offset, seg.selector);
7246 	put_smstate(u16, buf, offset + 2, flags);
7247 	put_smstate(u32, buf, offset + 4, seg.limit);
7248 	put_smstate(u64, buf, offset + 8, seg.base);
7249 }
7250 #endif
7251 
7252 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7253 {
7254 	struct desc_ptr dt;
7255 	struct kvm_segment seg;
7256 	unsigned long val;
7257 	int i;
7258 
7259 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7260 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7261 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7262 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7263 
7264 	for (i = 0; i < 8; i++)
7265 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7266 
7267 	kvm_get_dr(vcpu, 6, &val);
7268 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7269 	kvm_get_dr(vcpu, 7, &val);
7270 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7271 
7272 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7273 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7274 	put_smstate(u32, buf, 0x7f64, seg.base);
7275 	put_smstate(u32, buf, 0x7f60, seg.limit);
7276 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7277 
7278 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7279 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7280 	put_smstate(u32, buf, 0x7f80, seg.base);
7281 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7282 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7283 
7284 	kvm_x86_ops->get_gdt(vcpu, &dt);
7285 	put_smstate(u32, buf, 0x7f74, dt.address);
7286 	put_smstate(u32, buf, 0x7f70, dt.size);
7287 
7288 	kvm_x86_ops->get_idt(vcpu, &dt);
7289 	put_smstate(u32, buf, 0x7f58, dt.address);
7290 	put_smstate(u32, buf, 0x7f54, dt.size);
7291 
7292 	for (i = 0; i < 6; i++)
7293 		enter_smm_save_seg_32(vcpu, buf, i);
7294 
7295 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7296 
7297 	/* revision id */
7298 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7299 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7300 }
7301 
7302 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7303 {
7304 #ifdef CONFIG_X86_64
7305 	struct desc_ptr dt;
7306 	struct kvm_segment seg;
7307 	unsigned long val;
7308 	int i;
7309 
7310 	for (i = 0; i < 16; i++)
7311 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7312 
7313 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7314 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7315 
7316 	kvm_get_dr(vcpu, 6, &val);
7317 	put_smstate(u64, buf, 0x7f68, val);
7318 	kvm_get_dr(vcpu, 7, &val);
7319 	put_smstate(u64, buf, 0x7f60, val);
7320 
7321 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7322 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7323 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7324 
7325 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7326 
7327 	/* revision id */
7328 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7329 
7330 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7331 
7332 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7333 	put_smstate(u16, buf, 0x7e90, seg.selector);
7334 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7335 	put_smstate(u32, buf, 0x7e94, seg.limit);
7336 	put_smstate(u64, buf, 0x7e98, seg.base);
7337 
7338 	kvm_x86_ops->get_idt(vcpu, &dt);
7339 	put_smstate(u32, buf, 0x7e84, dt.size);
7340 	put_smstate(u64, buf, 0x7e88, dt.address);
7341 
7342 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7343 	put_smstate(u16, buf, 0x7e70, seg.selector);
7344 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7345 	put_smstate(u32, buf, 0x7e74, seg.limit);
7346 	put_smstate(u64, buf, 0x7e78, seg.base);
7347 
7348 	kvm_x86_ops->get_gdt(vcpu, &dt);
7349 	put_smstate(u32, buf, 0x7e64, dt.size);
7350 	put_smstate(u64, buf, 0x7e68, dt.address);
7351 
7352 	for (i = 0; i < 6; i++)
7353 		enter_smm_save_seg_64(vcpu, buf, i);
7354 #else
7355 	WARN_ON_ONCE(1);
7356 #endif
7357 }
7358 
7359 static void enter_smm(struct kvm_vcpu *vcpu)
7360 {
7361 	struct kvm_segment cs, ds;
7362 	struct desc_ptr dt;
7363 	char buf[512];
7364 	u32 cr0;
7365 
7366 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7367 	memset(buf, 0, 512);
7368 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7369 		enter_smm_save_state_64(vcpu, buf);
7370 	else
7371 		enter_smm_save_state_32(vcpu, buf);
7372 
7373 	/*
7374 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7375 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7376 	 * the SMM state-save area.
7377 	 */
7378 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7379 
7380 	vcpu->arch.hflags |= HF_SMM_MASK;
7381 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7382 
7383 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7384 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7385 	else
7386 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7387 
7388 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7389 	kvm_rip_write(vcpu, 0x8000);
7390 
7391 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7392 	kvm_x86_ops->set_cr0(vcpu, cr0);
7393 	vcpu->arch.cr0 = cr0;
7394 
7395 	kvm_x86_ops->set_cr4(vcpu, 0);
7396 
7397 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7398 	dt.address = dt.size = 0;
7399 	kvm_x86_ops->set_idt(vcpu, &dt);
7400 
7401 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7402 
7403 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7404 	cs.base = vcpu->arch.smbase;
7405 
7406 	ds.selector = 0;
7407 	ds.base = 0;
7408 
7409 	cs.limit    = ds.limit = 0xffffffff;
7410 	cs.type     = ds.type = 0x3;
7411 	cs.dpl      = ds.dpl = 0;
7412 	cs.db       = ds.db = 0;
7413 	cs.s        = ds.s = 1;
7414 	cs.l        = ds.l = 0;
7415 	cs.g        = ds.g = 1;
7416 	cs.avl      = ds.avl = 0;
7417 	cs.present  = ds.present = 1;
7418 	cs.unusable = ds.unusable = 0;
7419 	cs.padding  = ds.padding = 0;
7420 
7421 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7422 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7423 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7424 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7425 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7426 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7427 
7428 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7429 		kvm_x86_ops->set_efer(vcpu, 0);
7430 
7431 	kvm_update_cpuid(vcpu);
7432 	kvm_mmu_reset_context(vcpu);
7433 }
7434 
7435 static void process_smi(struct kvm_vcpu *vcpu)
7436 {
7437 	vcpu->arch.smi_pending = true;
7438 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7439 }
7440 
7441 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7442 {
7443 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7444 }
7445 
7446 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7447 {
7448 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7449 		return;
7450 
7451 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7452 
7453 	if (irqchip_split(vcpu->kvm))
7454 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7455 	else {
7456 		if (vcpu->arch.apicv_active)
7457 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7458 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7459 	}
7460 
7461 	if (is_guest_mode(vcpu))
7462 		vcpu->arch.load_eoi_exitmap_pending = true;
7463 	else
7464 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7465 }
7466 
7467 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7468 {
7469 	u64 eoi_exit_bitmap[4];
7470 
7471 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7472 		return;
7473 
7474 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7475 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7476 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7477 }
7478 
7479 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7480 		unsigned long start, unsigned long end,
7481 		bool blockable)
7482 {
7483 	unsigned long apic_address;
7484 
7485 	/*
7486 	 * The physical address of apic access page is stored in the VMCS.
7487 	 * Update it when it becomes invalid.
7488 	 */
7489 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7490 	if (start <= apic_address && apic_address < end)
7491 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7492 
7493 	return 0;
7494 }
7495 
7496 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7497 {
7498 	struct page *page = NULL;
7499 
7500 	if (!lapic_in_kernel(vcpu))
7501 		return;
7502 
7503 	if (!kvm_x86_ops->set_apic_access_page_addr)
7504 		return;
7505 
7506 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7507 	if (is_error_page(page))
7508 		return;
7509 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7510 
7511 	/*
7512 	 * Do not pin apic access page in memory, the MMU notifier
7513 	 * will call us again if it is migrated or swapped out.
7514 	 */
7515 	put_page(page);
7516 }
7517 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7518 
7519 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7520 {
7521 	smp_send_reschedule(vcpu->cpu);
7522 }
7523 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7524 
7525 /*
7526  * Returns 1 to let vcpu_run() continue the guest execution loop without
7527  * exiting to the userspace.  Otherwise, the value will be returned to the
7528  * userspace.
7529  */
7530 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7531 {
7532 	int r;
7533 	bool req_int_win =
7534 		dm_request_for_irq_injection(vcpu) &&
7535 		kvm_cpu_accept_dm_intr(vcpu);
7536 
7537 	bool req_immediate_exit = false;
7538 
7539 	if (kvm_request_pending(vcpu)) {
7540 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7541 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7542 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7543 			kvm_mmu_unload(vcpu);
7544 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7545 			__kvm_migrate_timers(vcpu);
7546 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7547 			kvm_gen_update_masterclock(vcpu->kvm);
7548 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7549 			kvm_gen_kvmclock_update(vcpu);
7550 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7551 			r = kvm_guest_time_update(vcpu);
7552 			if (unlikely(r))
7553 				goto out;
7554 		}
7555 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7556 			kvm_mmu_sync_roots(vcpu);
7557 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7558 			kvm_mmu_load_cr3(vcpu);
7559 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7560 			kvm_vcpu_flush_tlb(vcpu, true);
7561 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7562 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7563 			r = 0;
7564 			goto out;
7565 		}
7566 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7567 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7568 			vcpu->mmio_needed = 0;
7569 			r = 0;
7570 			goto out;
7571 		}
7572 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7573 			/* Page is swapped out. Do synthetic halt */
7574 			vcpu->arch.apf.halted = true;
7575 			r = 1;
7576 			goto out;
7577 		}
7578 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7579 			record_steal_time(vcpu);
7580 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7581 			process_smi(vcpu);
7582 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7583 			process_nmi(vcpu);
7584 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7585 			kvm_pmu_handle_event(vcpu);
7586 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7587 			kvm_pmu_deliver_pmi(vcpu);
7588 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7589 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7590 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7591 				     vcpu->arch.ioapic_handled_vectors)) {
7592 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7593 				vcpu->run->eoi.vector =
7594 						vcpu->arch.pending_ioapic_eoi;
7595 				r = 0;
7596 				goto out;
7597 			}
7598 		}
7599 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7600 			vcpu_scan_ioapic(vcpu);
7601 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7602 			vcpu_load_eoi_exitmap(vcpu);
7603 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7604 			kvm_vcpu_reload_apic_access_page(vcpu);
7605 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7606 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7607 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7608 			r = 0;
7609 			goto out;
7610 		}
7611 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7612 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7613 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7614 			r = 0;
7615 			goto out;
7616 		}
7617 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7618 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7619 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7620 			r = 0;
7621 			goto out;
7622 		}
7623 
7624 		/*
7625 		 * KVM_REQ_HV_STIMER has to be processed after
7626 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7627 		 * depend on the guest clock being up-to-date
7628 		 */
7629 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7630 			kvm_hv_process_stimers(vcpu);
7631 	}
7632 
7633 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7634 		++vcpu->stat.req_event;
7635 		kvm_apic_accept_events(vcpu);
7636 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7637 			r = 1;
7638 			goto out;
7639 		}
7640 
7641 		if (inject_pending_event(vcpu, req_int_win) != 0)
7642 			req_immediate_exit = true;
7643 		else {
7644 			/* Enable SMI/NMI/IRQ window open exits if needed.
7645 			 *
7646 			 * SMIs have three cases:
7647 			 * 1) They can be nested, and then there is nothing to
7648 			 *    do here because RSM will cause a vmexit anyway.
7649 			 * 2) There is an ISA-specific reason why SMI cannot be
7650 			 *    injected, and the moment when this changes can be
7651 			 *    intercepted.
7652 			 * 3) Or the SMI can be pending because
7653 			 *    inject_pending_event has completed the injection
7654 			 *    of an IRQ or NMI from the previous vmexit, and
7655 			 *    then we request an immediate exit to inject the
7656 			 *    SMI.
7657 			 */
7658 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7659 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7660 					req_immediate_exit = true;
7661 			if (vcpu->arch.nmi_pending)
7662 				kvm_x86_ops->enable_nmi_window(vcpu);
7663 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7664 				kvm_x86_ops->enable_irq_window(vcpu);
7665 			WARN_ON(vcpu->arch.exception.pending);
7666 		}
7667 
7668 		if (kvm_lapic_enabled(vcpu)) {
7669 			update_cr8_intercept(vcpu);
7670 			kvm_lapic_sync_to_vapic(vcpu);
7671 		}
7672 	}
7673 
7674 	r = kvm_mmu_reload(vcpu);
7675 	if (unlikely(r)) {
7676 		goto cancel_injection;
7677 	}
7678 
7679 	preempt_disable();
7680 
7681 	kvm_x86_ops->prepare_guest_switch(vcpu);
7682 
7683 	/*
7684 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7685 	 * IPI are then delayed after guest entry, which ensures that they
7686 	 * result in virtual interrupt delivery.
7687 	 */
7688 	local_irq_disable();
7689 	vcpu->mode = IN_GUEST_MODE;
7690 
7691 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7692 
7693 	/*
7694 	 * 1) We should set ->mode before checking ->requests.  Please see
7695 	 * the comment in kvm_vcpu_exiting_guest_mode().
7696 	 *
7697 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
7698 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7699 	 * (see vmx_deliver_posted_interrupt).
7700 	 *
7701 	 * 3) This also orders the write to mode from any reads to the page
7702 	 * tables done while the VCPU is running.  Please see the comment
7703 	 * in kvm_flush_remote_tlbs.
7704 	 */
7705 	smp_mb__after_srcu_read_unlock();
7706 
7707 	/*
7708 	 * This handles the case where a posted interrupt was
7709 	 * notified with kvm_vcpu_kick.
7710 	 */
7711 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7712 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7713 
7714 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7715 	    || need_resched() || signal_pending(current)) {
7716 		vcpu->mode = OUTSIDE_GUEST_MODE;
7717 		smp_wmb();
7718 		local_irq_enable();
7719 		preempt_enable();
7720 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7721 		r = 1;
7722 		goto cancel_injection;
7723 	}
7724 
7725 	kvm_load_guest_xcr0(vcpu);
7726 
7727 	if (req_immediate_exit) {
7728 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7729 		kvm_x86_ops->request_immediate_exit(vcpu);
7730 	}
7731 
7732 	trace_kvm_entry(vcpu->vcpu_id);
7733 	if (lapic_timer_advance_ns)
7734 		wait_lapic_expire(vcpu);
7735 	guest_enter_irqoff();
7736 
7737 	if (unlikely(vcpu->arch.switch_db_regs)) {
7738 		set_debugreg(0, 7);
7739 		set_debugreg(vcpu->arch.eff_db[0], 0);
7740 		set_debugreg(vcpu->arch.eff_db[1], 1);
7741 		set_debugreg(vcpu->arch.eff_db[2], 2);
7742 		set_debugreg(vcpu->arch.eff_db[3], 3);
7743 		set_debugreg(vcpu->arch.dr6, 6);
7744 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7745 	}
7746 
7747 	kvm_x86_ops->run(vcpu);
7748 
7749 	/*
7750 	 * Do this here before restoring debug registers on the host.  And
7751 	 * since we do this before handling the vmexit, a DR access vmexit
7752 	 * can (a) read the correct value of the debug registers, (b) set
7753 	 * KVM_DEBUGREG_WONT_EXIT again.
7754 	 */
7755 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7756 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7757 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7758 		kvm_update_dr0123(vcpu);
7759 		kvm_update_dr6(vcpu);
7760 		kvm_update_dr7(vcpu);
7761 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7762 	}
7763 
7764 	/*
7765 	 * If the guest has used debug registers, at least dr7
7766 	 * will be disabled while returning to the host.
7767 	 * If we don't have active breakpoints in the host, we don't
7768 	 * care about the messed up debug address registers. But if
7769 	 * we have some of them active, restore the old state.
7770 	 */
7771 	if (hw_breakpoint_active())
7772 		hw_breakpoint_restore();
7773 
7774 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7775 
7776 	vcpu->mode = OUTSIDE_GUEST_MODE;
7777 	smp_wmb();
7778 
7779 	kvm_put_guest_xcr0(vcpu);
7780 
7781 	kvm_before_interrupt(vcpu);
7782 	kvm_x86_ops->handle_external_intr(vcpu);
7783 	kvm_after_interrupt(vcpu);
7784 
7785 	++vcpu->stat.exits;
7786 
7787 	guest_exit_irqoff();
7788 
7789 	local_irq_enable();
7790 	preempt_enable();
7791 
7792 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7793 
7794 	/*
7795 	 * Profile KVM exit RIPs:
7796 	 */
7797 	if (unlikely(prof_on == KVM_PROFILING)) {
7798 		unsigned long rip = kvm_rip_read(vcpu);
7799 		profile_hit(KVM_PROFILING, (void *)rip);
7800 	}
7801 
7802 	if (unlikely(vcpu->arch.tsc_always_catchup))
7803 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7804 
7805 	if (vcpu->arch.apic_attention)
7806 		kvm_lapic_sync_from_vapic(vcpu);
7807 
7808 	vcpu->arch.gpa_available = false;
7809 	r = kvm_x86_ops->handle_exit(vcpu);
7810 	return r;
7811 
7812 cancel_injection:
7813 	kvm_x86_ops->cancel_injection(vcpu);
7814 	if (unlikely(vcpu->arch.apic_attention))
7815 		kvm_lapic_sync_from_vapic(vcpu);
7816 out:
7817 	return r;
7818 }
7819 
7820 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7821 {
7822 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7823 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7824 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7825 		kvm_vcpu_block(vcpu);
7826 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7827 
7828 		if (kvm_x86_ops->post_block)
7829 			kvm_x86_ops->post_block(vcpu);
7830 
7831 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7832 			return 1;
7833 	}
7834 
7835 	kvm_apic_accept_events(vcpu);
7836 	switch(vcpu->arch.mp_state) {
7837 	case KVM_MP_STATE_HALTED:
7838 		vcpu->arch.pv.pv_unhalted = false;
7839 		vcpu->arch.mp_state =
7840 			KVM_MP_STATE_RUNNABLE;
7841 	case KVM_MP_STATE_RUNNABLE:
7842 		vcpu->arch.apf.halted = false;
7843 		break;
7844 	case KVM_MP_STATE_INIT_RECEIVED:
7845 		break;
7846 	default:
7847 		return -EINTR;
7848 		break;
7849 	}
7850 	return 1;
7851 }
7852 
7853 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7854 {
7855 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7856 		kvm_x86_ops->check_nested_events(vcpu, false);
7857 
7858 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7859 		!vcpu->arch.apf.halted);
7860 }
7861 
7862 static int vcpu_run(struct kvm_vcpu *vcpu)
7863 {
7864 	int r;
7865 	struct kvm *kvm = vcpu->kvm;
7866 
7867 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7868 	vcpu->arch.l1tf_flush_l1d = true;
7869 
7870 	for (;;) {
7871 		if (kvm_vcpu_running(vcpu)) {
7872 			r = vcpu_enter_guest(vcpu);
7873 		} else {
7874 			r = vcpu_block(kvm, vcpu);
7875 		}
7876 
7877 		if (r <= 0)
7878 			break;
7879 
7880 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7881 		if (kvm_cpu_has_pending_timer(vcpu))
7882 			kvm_inject_pending_timer_irqs(vcpu);
7883 
7884 		if (dm_request_for_irq_injection(vcpu) &&
7885 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7886 			r = 0;
7887 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7888 			++vcpu->stat.request_irq_exits;
7889 			break;
7890 		}
7891 
7892 		kvm_check_async_pf_completion(vcpu);
7893 
7894 		if (signal_pending(current)) {
7895 			r = -EINTR;
7896 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7897 			++vcpu->stat.signal_exits;
7898 			break;
7899 		}
7900 		if (need_resched()) {
7901 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7902 			cond_resched();
7903 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7904 		}
7905 	}
7906 
7907 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7908 
7909 	return r;
7910 }
7911 
7912 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7913 {
7914 	int r;
7915 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7916 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7917 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7918 	if (r != EMULATE_DONE)
7919 		return 0;
7920 	return 1;
7921 }
7922 
7923 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7924 {
7925 	BUG_ON(!vcpu->arch.pio.count);
7926 
7927 	return complete_emulated_io(vcpu);
7928 }
7929 
7930 /*
7931  * Implements the following, as a state machine:
7932  *
7933  * read:
7934  *   for each fragment
7935  *     for each mmio piece in the fragment
7936  *       write gpa, len
7937  *       exit
7938  *       copy data
7939  *   execute insn
7940  *
7941  * write:
7942  *   for each fragment
7943  *     for each mmio piece in the fragment
7944  *       write gpa, len
7945  *       copy data
7946  *       exit
7947  */
7948 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7949 {
7950 	struct kvm_run *run = vcpu->run;
7951 	struct kvm_mmio_fragment *frag;
7952 	unsigned len;
7953 
7954 	BUG_ON(!vcpu->mmio_needed);
7955 
7956 	/* Complete previous fragment */
7957 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7958 	len = min(8u, frag->len);
7959 	if (!vcpu->mmio_is_write)
7960 		memcpy(frag->data, run->mmio.data, len);
7961 
7962 	if (frag->len <= 8) {
7963 		/* Switch to the next fragment. */
7964 		frag++;
7965 		vcpu->mmio_cur_fragment++;
7966 	} else {
7967 		/* Go forward to the next mmio piece. */
7968 		frag->data += len;
7969 		frag->gpa += len;
7970 		frag->len -= len;
7971 	}
7972 
7973 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7974 		vcpu->mmio_needed = 0;
7975 
7976 		/* FIXME: return into emulator if single-stepping.  */
7977 		if (vcpu->mmio_is_write)
7978 			return 1;
7979 		vcpu->mmio_read_completed = 1;
7980 		return complete_emulated_io(vcpu);
7981 	}
7982 
7983 	run->exit_reason = KVM_EXIT_MMIO;
7984 	run->mmio.phys_addr = frag->gpa;
7985 	if (vcpu->mmio_is_write)
7986 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7987 	run->mmio.len = min(8u, frag->len);
7988 	run->mmio.is_write = vcpu->mmio_is_write;
7989 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7990 	return 0;
7991 }
7992 
7993 /* Swap (qemu) user FPU context for the guest FPU context. */
7994 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7995 {
7996 	preempt_disable();
7997 	copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
7998 	/* PKRU is separately restored in kvm_x86_ops->run.  */
7999 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8000 				~XFEATURE_MASK_PKRU);
8001 	preempt_enable();
8002 	trace_kvm_fpu(1);
8003 }
8004 
8005 /* When vcpu_run ends, restore user space FPU context. */
8006 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8007 {
8008 	preempt_disable();
8009 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8010 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8011 	preempt_enable();
8012 	++vcpu->stat.fpu_reload;
8013 	trace_kvm_fpu(0);
8014 }
8015 
8016 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8017 {
8018 	int r;
8019 
8020 	vcpu_load(vcpu);
8021 	kvm_sigset_activate(vcpu);
8022 	kvm_load_guest_fpu(vcpu);
8023 
8024 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8025 		if (kvm_run->immediate_exit) {
8026 			r = -EINTR;
8027 			goto out;
8028 		}
8029 		kvm_vcpu_block(vcpu);
8030 		kvm_apic_accept_events(vcpu);
8031 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8032 		r = -EAGAIN;
8033 		if (signal_pending(current)) {
8034 			r = -EINTR;
8035 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8036 			++vcpu->stat.signal_exits;
8037 		}
8038 		goto out;
8039 	}
8040 
8041 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8042 		r = -EINVAL;
8043 		goto out;
8044 	}
8045 
8046 	if (vcpu->run->kvm_dirty_regs) {
8047 		r = sync_regs(vcpu);
8048 		if (r != 0)
8049 			goto out;
8050 	}
8051 
8052 	/* re-sync apic's tpr */
8053 	if (!lapic_in_kernel(vcpu)) {
8054 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8055 			r = -EINVAL;
8056 			goto out;
8057 		}
8058 	}
8059 
8060 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8061 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8062 		vcpu->arch.complete_userspace_io = NULL;
8063 		r = cui(vcpu);
8064 		if (r <= 0)
8065 			goto out;
8066 	} else
8067 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8068 
8069 	if (kvm_run->immediate_exit)
8070 		r = -EINTR;
8071 	else
8072 		r = vcpu_run(vcpu);
8073 
8074 out:
8075 	kvm_put_guest_fpu(vcpu);
8076 	if (vcpu->run->kvm_valid_regs)
8077 		store_regs(vcpu);
8078 	post_kvm_run_save(vcpu);
8079 	kvm_sigset_deactivate(vcpu);
8080 
8081 	vcpu_put(vcpu);
8082 	return r;
8083 }
8084 
8085 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8086 {
8087 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8088 		/*
8089 		 * We are here if userspace calls get_regs() in the middle of
8090 		 * instruction emulation. Registers state needs to be copied
8091 		 * back from emulation context to vcpu. Userspace shouldn't do
8092 		 * that usually, but some bad designed PV devices (vmware
8093 		 * backdoor interface) need this to work
8094 		 */
8095 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8096 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8097 	}
8098 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8099 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8100 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8101 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8102 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8103 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8104 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8105 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8106 #ifdef CONFIG_X86_64
8107 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8108 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8109 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8110 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8111 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8112 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8113 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8114 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8115 #endif
8116 
8117 	regs->rip = kvm_rip_read(vcpu);
8118 	regs->rflags = kvm_get_rflags(vcpu);
8119 }
8120 
8121 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8122 {
8123 	vcpu_load(vcpu);
8124 	__get_regs(vcpu, regs);
8125 	vcpu_put(vcpu);
8126 	return 0;
8127 }
8128 
8129 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8130 {
8131 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8132 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8133 
8134 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8135 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8136 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8137 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8138 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8139 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8140 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8141 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8142 #ifdef CONFIG_X86_64
8143 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8144 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8145 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8146 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8147 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8148 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8149 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8150 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8151 #endif
8152 
8153 	kvm_rip_write(vcpu, regs->rip);
8154 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8155 
8156 	vcpu->arch.exception.pending = false;
8157 
8158 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8159 }
8160 
8161 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8162 {
8163 	vcpu_load(vcpu);
8164 	__set_regs(vcpu, regs);
8165 	vcpu_put(vcpu);
8166 	return 0;
8167 }
8168 
8169 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8170 {
8171 	struct kvm_segment cs;
8172 
8173 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8174 	*db = cs.db;
8175 	*l = cs.l;
8176 }
8177 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8178 
8179 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8180 {
8181 	struct desc_ptr dt;
8182 
8183 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8184 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8185 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8186 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8187 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8188 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8189 
8190 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8191 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8192 
8193 	kvm_x86_ops->get_idt(vcpu, &dt);
8194 	sregs->idt.limit = dt.size;
8195 	sregs->idt.base = dt.address;
8196 	kvm_x86_ops->get_gdt(vcpu, &dt);
8197 	sregs->gdt.limit = dt.size;
8198 	sregs->gdt.base = dt.address;
8199 
8200 	sregs->cr0 = kvm_read_cr0(vcpu);
8201 	sregs->cr2 = vcpu->arch.cr2;
8202 	sregs->cr3 = kvm_read_cr3(vcpu);
8203 	sregs->cr4 = kvm_read_cr4(vcpu);
8204 	sregs->cr8 = kvm_get_cr8(vcpu);
8205 	sregs->efer = vcpu->arch.efer;
8206 	sregs->apic_base = kvm_get_apic_base(vcpu);
8207 
8208 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8209 
8210 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8211 		set_bit(vcpu->arch.interrupt.nr,
8212 			(unsigned long *)sregs->interrupt_bitmap);
8213 }
8214 
8215 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8216 				  struct kvm_sregs *sregs)
8217 {
8218 	vcpu_load(vcpu);
8219 	__get_sregs(vcpu, sregs);
8220 	vcpu_put(vcpu);
8221 	return 0;
8222 }
8223 
8224 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8225 				    struct kvm_mp_state *mp_state)
8226 {
8227 	vcpu_load(vcpu);
8228 
8229 	kvm_apic_accept_events(vcpu);
8230 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8231 					vcpu->arch.pv.pv_unhalted)
8232 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8233 	else
8234 		mp_state->mp_state = vcpu->arch.mp_state;
8235 
8236 	vcpu_put(vcpu);
8237 	return 0;
8238 }
8239 
8240 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8241 				    struct kvm_mp_state *mp_state)
8242 {
8243 	int ret = -EINVAL;
8244 
8245 	vcpu_load(vcpu);
8246 
8247 	if (!lapic_in_kernel(vcpu) &&
8248 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8249 		goto out;
8250 
8251 	/* INITs are latched while in SMM */
8252 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8253 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8254 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8255 		goto out;
8256 
8257 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8258 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8259 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8260 	} else
8261 		vcpu->arch.mp_state = mp_state->mp_state;
8262 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8263 
8264 	ret = 0;
8265 out:
8266 	vcpu_put(vcpu);
8267 	return ret;
8268 }
8269 
8270 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8271 		    int reason, bool has_error_code, u32 error_code)
8272 {
8273 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8274 	int ret;
8275 
8276 	init_emulate_ctxt(vcpu);
8277 
8278 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8279 				   has_error_code, error_code);
8280 
8281 	if (ret)
8282 		return EMULATE_FAIL;
8283 
8284 	kvm_rip_write(vcpu, ctxt->eip);
8285 	kvm_set_rflags(vcpu, ctxt->eflags);
8286 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8287 	return EMULATE_DONE;
8288 }
8289 EXPORT_SYMBOL_GPL(kvm_task_switch);
8290 
8291 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8292 {
8293 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8294 			(sregs->cr4 & X86_CR4_OSXSAVE))
8295 		return  -EINVAL;
8296 
8297 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8298 		/*
8299 		 * When EFER.LME and CR0.PG are set, the processor is in
8300 		 * 64-bit mode (though maybe in a 32-bit code segment).
8301 		 * CR4.PAE and EFER.LMA must be set.
8302 		 */
8303 		if (!(sregs->cr4 & X86_CR4_PAE)
8304 		    || !(sregs->efer & EFER_LMA))
8305 			return -EINVAL;
8306 	} else {
8307 		/*
8308 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8309 		 * segment cannot be 64-bit.
8310 		 */
8311 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8312 			return -EINVAL;
8313 	}
8314 
8315 	return 0;
8316 }
8317 
8318 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8319 {
8320 	struct msr_data apic_base_msr;
8321 	int mmu_reset_needed = 0;
8322 	int cpuid_update_needed = 0;
8323 	int pending_vec, max_bits, idx;
8324 	struct desc_ptr dt;
8325 	int ret = -EINVAL;
8326 
8327 	if (kvm_valid_sregs(vcpu, sregs))
8328 		goto out;
8329 
8330 	apic_base_msr.data = sregs->apic_base;
8331 	apic_base_msr.host_initiated = true;
8332 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8333 		goto out;
8334 
8335 	dt.size = sregs->idt.limit;
8336 	dt.address = sregs->idt.base;
8337 	kvm_x86_ops->set_idt(vcpu, &dt);
8338 	dt.size = sregs->gdt.limit;
8339 	dt.address = sregs->gdt.base;
8340 	kvm_x86_ops->set_gdt(vcpu, &dt);
8341 
8342 	vcpu->arch.cr2 = sregs->cr2;
8343 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8344 	vcpu->arch.cr3 = sregs->cr3;
8345 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8346 
8347 	kvm_set_cr8(vcpu, sregs->cr8);
8348 
8349 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8350 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8351 
8352 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8353 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8354 	vcpu->arch.cr0 = sregs->cr0;
8355 
8356 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8357 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8358 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8359 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8360 	if (cpuid_update_needed)
8361 		kvm_update_cpuid(vcpu);
8362 
8363 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8364 	if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8365 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8366 		mmu_reset_needed = 1;
8367 	}
8368 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8369 
8370 	if (mmu_reset_needed)
8371 		kvm_mmu_reset_context(vcpu);
8372 
8373 	max_bits = KVM_NR_INTERRUPTS;
8374 	pending_vec = find_first_bit(
8375 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8376 	if (pending_vec < max_bits) {
8377 		kvm_queue_interrupt(vcpu, pending_vec, false);
8378 		pr_debug("Set back pending irq %d\n", pending_vec);
8379 	}
8380 
8381 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8382 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8383 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8384 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8385 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8386 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8387 
8388 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8389 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8390 
8391 	update_cr8_intercept(vcpu);
8392 
8393 	/* Older userspace won't unhalt the vcpu on reset. */
8394 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8395 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8396 	    !is_protmode(vcpu))
8397 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8398 
8399 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8400 
8401 	ret = 0;
8402 out:
8403 	return ret;
8404 }
8405 
8406 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8407 				  struct kvm_sregs *sregs)
8408 {
8409 	int ret;
8410 
8411 	vcpu_load(vcpu);
8412 	ret = __set_sregs(vcpu, sregs);
8413 	vcpu_put(vcpu);
8414 	return ret;
8415 }
8416 
8417 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8418 					struct kvm_guest_debug *dbg)
8419 {
8420 	unsigned long rflags;
8421 	int i, r;
8422 
8423 	vcpu_load(vcpu);
8424 
8425 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8426 		r = -EBUSY;
8427 		if (vcpu->arch.exception.pending)
8428 			goto out;
8429 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8430 			kvm_queue_exception(vcpu, DB_VECTOR);
8431 		else
8432 			kvm_queue_exception(vcpu, BP_VECTOR);
8433 	}
8434 
8435 	/*
8436 	 * Read rflags as long as potentially injected trace flags are still
8437 	 * filtered out.
8438 	 */
8439 	rflags = kvm_get_rflags(vcpu);
8440 
8441 	vcpu->guest_debug = dbg->control;
8442 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8443 		vcpu->guest_debug = 0;
8444 
8445 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8446 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8447 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8448 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8449 	} else {
8450 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8451 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8452 	}
8453 	kvm_update_dr7(vcpu);
8454 
8455 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8456 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8457 			get_segment_base(vcpu, VCPU_SREG_CS);
8458 
8459 	/*
8460 	 * Trigger an rflags update that will inject or remove the trace
8461 	 * flags.
8462 	 */
8463 	kvm_set_rflags(vcpu, rflags);
8464 
8465 	kvm_x86_ops->update_bp_intercept(vcpu);
8466 
8467 	r = 0;
8468 
8469 out:
8470 	vcpu_put(vcpu);
8471 	return r;
8472 }
8473 
8474 /*
8475  * Translate a guest virtual address to a guest physical address.
8476  */
8477 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8478 				    struct kvm_translation *tr)
8479 {
8480 	unsigned long vaddr = tr->linear_address;
8481 	gpa_t gpa;
8482 	int idx;
8483 
8484 	vcpu_load(vcpu);
8485 
8486 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8487 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8488 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8489 	tr->physical_address = gpa;
8490 	tr->valid = gpa != UNMAPPED_GVA;
8491 	tr->writeable = 1;
8492 	tr->usermode = 0;
8493 
8494 	vcpu_put(vcpu);
8495 	return 0;
8496 }
8497 
8498 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8499 {
8500 	struct fxregs_state *fxsave;
8501 
8502 	vcpu_load(vcpu);
8503 
8504 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8505 	memcpy(fpu->fpr, fxsave->st_space, 128);
8506 	fpu->fcw = fxsave->cwd;
8507 	fpu->fsw = fxsave->swd;
8508 	fpu->ftwx = fxsave->twd;
8509 	fpu->last_opcode = fxsave->fop;
8510 	fpu->last_ip = fxsave->rip;
8511 	fpu->last_dp = fxsave->rdp;
8512 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8513 
8514 	vcpu_put(vcpu);
8515 	return 0;
8516 }
8517 
8518 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8519 {
8520 	struct fxregs_state *fxsave;
8521 
8522 	vcpu_load(vcpu);
8523 
8524 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8525 
8526 	memcpy(fxsave->st_space, fpu->fpr, 128);
8527 	fxsave->cwd = fpu->fcw;
8528 	fxsave->swd = fpu->fsw;
8529 	fxsave->twd = fpu->ftwx;
8530 	fxsave->fop = fpu->last_opcode;
8531 	fxsave->rip = fpu->last_ip;
8532 	fxsave->rdp = fpu->last_dp;
8533 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8534 
8535 	vcpu_put(vcpu);
8536 	return 0;
8537 }
8538 
8539 static void store_regs(struct kvm_vcpu *vcpu)
8540 {
8541 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8542 
8543 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8544 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8545 
8546 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8547 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8548 
8549 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8550 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8551 				vcpu, &vcpu->run->s.regs.events);
8552 }
8553 
8554 static int sync_regs(struct kvm_vcpu *vcpu)
8555 {
8556 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8557 		return -EINVAL;
8558 
8559 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8560 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8561 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8562 	}
8563 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8564 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8565 			return -EINVAL;
8566 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8567 	}
8568 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8569 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8570 				vcpu, &vcpu->run->s.regs.events))
8571 			return -EINVAL;
8572 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8573 	}
8574 
8575 	return 0;
8576 }
8577 
8578 static void fx_init(struct kvm_vcpu *vcpu)
8579 {
8580 	fpstate_init(&vcpu->arch.guest_fpu.state);
8581 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8582 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8583 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8584 
8585 	/*
8586 	 * Ensure guest xcr0 is valid for loading
8587 	 */
8588 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8589 
8590 	vcpu->arch.cr0 |= X86_CR0_ET;
8591 }
8592 
8593 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8594 {
8595 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8596 
8597 	kvmclock_reset(vcpu);
8598 
8599 	kvm_x86_ops->vcpu_free(vcpu);
8600 	free_cpumask_var(wbinvd_dirty_mask);
8601 }
8602 
8603 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8604 						unsigned int id)
8605 {
8606 	struct kvm_vcpu *vcpu;
8607 
8608 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8609 		printk_once(KERN_WARNING
8610 		"kvm: SMP vm created on host with unstable TSC; "
8611 		"guest TSC will not be reliable\n");
8612 
8613 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8614 
8615 	return vcpu;
8616 }
8617 
8618 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8619 {
8620 	kvm_vcpu_mtrr_init(vcpu);
8621 	vcpu_load(vcpu);
8622 	kvm_vcpu_reset(vcpu, false);
8623 	kvm_init_mmu(vcpu, false);
8624 	vcpu_put(vcpu);
8625 	return 0;
8626 }
8627 
8628 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8629 {
8630 	struct msr_data msr;
8631 	struct kvm *kvm = vcpu->kvm;
8632 
8633 	kvm_hv_vcpu_postcreate(vcpu);
8634 
8635 	if (mutex_lock_killable(&vcpu->mutex))
8636 		return;
8637 	vcpu_load(vcpu);
8638 	msr.data = 0x0;
8639 	msr.index = MSR_IA32_TSC;
8640 	msr.host_initiated = true;
8641 	kvm_write_tsc(vcpu, &msr);
8642 	vcpu_put(vcpu);
8643 	mutex_unlock(&vcpu->mutex);
8644 
8645 	if (!kvmclock_periodic_sync)
8646 		return;
8647 
8648 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8649 					KVMCLOCK_SYNC_PERIOD);
8650 }
8651 
8652 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8653 {
8654 	vcpu->arch.apf.msr_val = 0;
8655 
8656 	vcpu_load(vcpu);
8657 	kvm_mmu_unload(vcpu);
8658 	vcpu_put(vcpu);
8659 
8660 	kvm_x86_ops->vcpu_free(vcpu);
8661 }
8662 
8663 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8664 {
8665 	kvm_lapic_reset(vcpu, init_event);
8666 
8667 	vcpu->arch.hflags = 0;
8668 
8669 	vcpu->arch.smi_pending = 0;
8670 	vcpu->arch.smi_count = 0;
8671 	atomic_set(&vcpu->arch.nmi_queued, 0);
8672 	vcpu->arch.nmi_pending = 0;
8673 	vcpu->arch.nmi_injected = false;
8674 	kvm_clear_interrupt_queue(vcpu);
8675 	kvm_clear_exception_queue(vcpu);
8676 	vcpu->arch.exception.pending = false;
8677 
8678 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8679 	kvm_update_dr0123(vcpu);
8680 	vcpu->arch.dr6 = DR6_INIT;
8681 	kvm_update_dr6(vcpu);
8682 	vcpu->arch.dr7 = DR7_FIXED_1;
8683 	kvm_update_dr7(vcpu);
8684 
8685 	vcpu->arch.cr2 = 0;
8686 
8687 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8688 	vcpu->arch.apf.msr_val = 0;
8689 	vcpu->arch.st.msr_val = 0;
8690 
8691 	kvmclock_reset(vcpu);
8692 
8693 	kvm_clear_async_pf_completion_queue(vcpu);
8694 	kvm_async_pf_hash_reset(vcpu);
8695 	vcpu->arch.apf.halted = false;
8696 
8697 	if (kvm_mpx_supported()) {
8698 		void *mpx_state_buffer;
8699 
8700 		/*
8701 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8702 		 * called with loaded FPU and does not let userspace fix the state.
8703 		 */
8704 		if (init_event)
8705 			kvm_put_guest_fpu(vcpu);
8706 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8707 					XFEATURE_MASK_BNDREGS);
8708 		if (mpx_state_buffer)
8709 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8710 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8711 					XFEATURE_MASK_BNDCSR);
8712 		if (mpx_state_buffer)
8713 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8714 		if (init_event)
8715 			kvm_load_guest_fpu(vcpu);
8716 	}
8717 
8718 	if (!init_event) {
8719 		kvm_pmu_reset(vcpu);
8720 		vcpu->arch.smbase = 0x30000;
8721 
8722 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8723 		vcpu->arch.msr_misc_features_enables = 0;
8724 
8725 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8726 	}
8727 
8728 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8729 	vcpu->arch.regs_avail = ~0;
8730 	vcpu->arch.regs_dirty = ~0;
8731 
8732 	vcpu->arch.ia32_xss = 0;
8733 
8734 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8735 }
8736 
8737 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8738 {
8739 	struct kvm_segment cs;
8740 
8741 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8742 	cs.selector = vector << 8;
8743 	cs.base = vector << 12;
8744 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8745 	kvm_rip_write(vcpu, 0);
8746 }
8747 
8748 int kvm_arch_hardware_enable(void)
8749 {
8750 	struct kvm *kvm;
8751 	struct kvm_vcpu *vcpu;
8752 	int i;
8753 	int ret;
8754 	u64 local_tsc;
8755 	u64 max_tsc = 0;
8756 	bool stable, backwards_tsc = false;
8757 
8758 	kvm_shared_msr_cpu_online();
8759 	ret = kvm_x86_ops->hardware_enable();
8760 	if (ret != 0)
8761 		return ret;
8762 
8763 	local_tsc = rdtsc();
8764 	stable = !kvm_check_tsc_unstable();
8765 	list_for_each_entry(kvm, &vm_list, vm_list) {
8766 		kvm_for_each_vcpu(i, vcpu, kvm) {
8767 			if (!stable && vcpu->cpu == smp_processor_id())
8768 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8769 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8770 				backwards_tsc = true;
8771 				if (vcpu->arch.last_host_tsc > max_tsc)
8772 					max_tsc = vcpu->arch.last_host_tsc;
8773 			}
8774 		}
8775 	}
8776 
8777 	/*
8778 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8779 	 * platforms that reset TSC during suspend or hibernate actions, but
8780 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8781 	 * detect that condition here, which happens early in CPU bringup,
8782 	 * before any KVM threads can be running.  Unfortunately, we can't
8783 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8784 	 * enough into CPU bringup that we know how much real time has actually
8785 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8786 	 * variables that haven't been updated yet.
8787 	 *
8788 	 * So we simply find the maximum observed TSC above, then record the
8789 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8790 	 * the adjustment will be applied.  Note that we accumulate
8791 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8792 	 * gets a chance to run again.  In the event that no KVM threads get a
8793 	 * chance to run, we will miss the entire elapsed period, as we'll have
8794 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8795 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8796 	 * uniform across all VCPUs (not to mention the scenario is extremely
8797 	 * unlikely). It is possible that a second hibernate recovery happens
8798 	 * much faster than a first, causing the observed TSC here to be
8799 	 * smaller; this would require additional padding adjustment, which is
8800 	 * why we set last_host_tsc to the local tsc observed here.
8801 	 *
8802 	 * N.B. - this code below runs only on platforms with reliable TSC,
8803 	 * as that is the only way backwards_tsc is set above.  Also note
8804 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8805 	 * have the same delta_cyc adjustment applied if backwards_tsc
8806 	 * is detected.  Note further, this adjustment is only done once,
8807 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8808 	 * called multiple times (one for each physical CPU bringup).
8809 	 *
8810 	 * Platforms with unreliable TSCs don't have to deal with this, they
8811 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8812 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8813 	 * guarantee that they stay in perfect synchronization.
8814 	 */
8815 	if (backwards_tsc) {
8816 		u64 delta_cyc = max_tsc - local_tsc;
8817 		list_for_each_entry(kvm, &vm_list, vm_list) {
8818 			kvm->arch.backwards_tsc_observed = true;
8819 			kvm_for_each_vcpu(i, vcpu, kvm) {
8820 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8821 				vcpu->arch.last_host_tsc = local_tsc;
8822 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8823 			}
8824 
8825 			/*
8826 			 * We have to disable TSC offset matching.. if you were
8827 			 * booting a VM while issuing an S4 host suspend....
8828 			 * you may have some problem.  Solving this issue is
8829 			 * left as an exercise to the reader.
8830 			 */
8831 			kvm->arch.last_tsc_nsec = 0;
8832 			kvm->arch.last_tsc_write = 0;
8833 		}
8834 
8835 	}
8836 	return 0;
8837 }
8838 
8839 void kvm_arch_hardware_disable(void)
8840 {
8841 	kvm_x86_ops->hardware_disable();
8842 	drop_user_return_notifiers();
8843 }
8844 
8845 int kvm_arch_hardware_setup(void)
8846 {
8847 	int r;
8848 
8849 	r = kvm_x86_ops->hardware_setup();
8850 	if (r != 0)
8851 		return r;
8852 
8853 	if (kvm_has_tsc_control) {
8854 		/*
8855 		 * Make sure the user can only configure tsc_khz values that
8856 		 * fit into a signed integer.
8857 		 * A min value is not calculated because it will always
8858 		 * be 1 on all machines.
8859 		 */
8860 		u64 max = min(0x7fffffffULL,
8861 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8862 		kvm_max_guest_tsc_khz = max;
8863 
8864 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8865 	}
8866 
8867 	kvm_init_msr_list();
8868 	return 0;
8869 }
8870 
8871 void kvm_arch_hardware_unsetup(void)
8872 {
8873 	kvm_x86_ops->hardware_unsetup();
8874 }
8875 
8876 void kvm_arch_check_processor_compat(void *rtn)
8877 {
8878 	kvm_x86_ops->check_processor_compatibility(rtn);
8879 }
8880 
8881 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8882 {
8883 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8884 }
8885 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8886 
8887 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8888 {
8889 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8890 }
8891 
8892 struct static_key kvm_no_apic_vcpu __read_mostly;
8893 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8894 
8895 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8896 {
8897 	struct page *page;
8898 	int r;
8899 
8900 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8901 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8902 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8903 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8904 	else
8905 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8906 
8907 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8908 	if (!page) {
8909 		r = -ENOMEM;
8910 		goto fail;
8911 	}
8912 	vcpu->arch.pio_data = page_address(page);
8913 
8914 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
8915 
8916 	r = kvm_mmu_create(vcpu);
8917 	if (r < 0)
8918 		goto fail_free_pio_data;
8919 
8920 	if (irqchip_in_kernel(vcpu->kvm)) {
8921 		r = kvm_create_lapic(vcpu);
8922 		if (r < 0)
8923 			goto fail_mmu_destroy;
8924 	} else
8925 		static_key_slow_inc(&kvm_no_apic_vcpu);
8926 
8927 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8928 				       GFP_KERNEL);
8929 	if (!vcpu->arch.mce_banks) {
8930 		r = -ENOMEM;
8931 		goto fail_free_lapic;
8932 	}
8933 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8934 
8935 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8936 		r = -ENOMEM;
8937 		goto fail_free_mce_banks;
8938 	}
8939 
8940 	fx_init(vcpu);
8941 
8942 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8943 
8944 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8945 
8946 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8947 
8948 	kvm_async_pf_hash_reset(vcpu);
8949 	kvm_pmu_init(vcpu);
8950 
8951 	vcpu->arch.pending_external_vector = -1;
8952 	vcpu->arch.preempted_in_kernel = false;
8953 
8954 	kvm_hv_vcpu_init(vcpu);
8955 
8956 	return 0;
8957 
8958 fail_free_mce_banks:
8959 	kfree(vcpu->arch.mce_banks);
8960 fail_free_lapic:
8961 	kvm_free_lapic(vcpu);
8962 fail_mmu_destroy:
8963 	kvm_mmu_destroy(vcpu);
8964 fail_free_pio_data:
8965 	free_page((unsigned long)vcpu->arch.pio_data);
8966 fail:
8967 	return r;
8968 }
8969 
8970 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8971 {
8972 	int idx;
8973 
8974 	kvm_hv_vcpu_uninit(vcpu);
8975 	kvm_pmu_destroy(vcpu);
8976 	kfree(vcpu->arch.mce_banks);
8977 	kvm_free_lapic(vcpu);
8978 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8979 	kvm_mmu_destroy(vcpu);
8980 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8981 	free_page((unsigned long)vcpu->arch.pio_data);
8982 	if (!lapic_in_kernel(vcpu))
8983 		static_key_slow_dec(&kvm_no_apic_vcpu);
8984 }
8985 
8986 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8987 {
8988 	vcpu->arch.l1tf_flush_l1d = true;
8989 	kvm_x86_ops->sched_in(vcpu, cpu);
8990 }
8991 
8992 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8993 {
8994 	if (type)
8995 		return -EINVAL;
8996 
8997 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8998 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8999 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9000 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9001 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9002 
9003 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9004 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9005 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9006 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9007 		&kvm->arch.irq_sources_bitmap);
9008 
9009 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9010 	mutex_init(&kvm->arch.apic_map_lock);
9011 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9012 
9013 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9014 	pvclock_update_vm_gtod_copy(kvm);
9015 
9016 	kvm->arch.guest_can_read_msr_platform_info = true;
9017 
9018 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9019 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9020 
9021 	kvm_hv_init_vm(kvm);
9022 	kvm_page_track_init(kvm);
9023 	kvm_mmu_init_vm(kvm);
9024 
9025 	if (kvm_x86_ops->vm_init)
9026 		return kvm_x86_ops->vm_init(kvm);
9027 
9028 	return 0;
9029 }
9030 
9031 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9032 {
9033 	vcpu_load(vcpu);
9034 	kvm_mmu_unload(vcpu);
9035 	vcpu_put(vcpu);
9036 }
9037 
9038 static void kvm_free_vcpus(struct kvm *kvm)
9039 {
9040 	unsigned int i;
9041 	struct kvm_vcpu *vcpu;
9042 
9043 	/*
9044 	 * Unpin any mmu pages first.
9045 	 */
9046 	kvm_for_each_vcpu(i, vcpu, kvm) {
9047 		kvm_clear_async_pf_completion_queue(vcpu);
9048 		kvm_unload_vcpu_mmu(vcpu);
9049 	}
9050 	kvm_for_each_vcpu(i, vcpu, kvm)
9051 		kvm_arch_vcpu_free(vcpu);
9052 
9053 	mutex_lock(&kvm->lock);
9054 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9055 		kvm->vcpus[i] = NULL;
9056 
9057 	atomic_set(&kvm->online_vcpus, 0);
9058 	mutex_unlock(&kvm->lock);
9059 }
9060 
9061 void kvm_arch_sync_events(struct kvm *kvm)
9062 {
9063 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9064 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9065 	kvm_free_pit(kvm);
9066 }
9067 
9068 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9069 {
9070 	int i, r;
9071 	unsigned long hva;
9072 	struct kvm_memslots *slots = kvm_memslots(kvm);
9073 	struct kvm_memory_slot *slot, old;
9074 
9075 	/* Called with kvm->slots_lock held.  */
9076 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9077 		return -EINVAL;
9078 
9079 	slot = id_to_memslot(slots, id);
9080 	if (size) {
9081 		if (slot->npages)
9082 			return -EEXIST;
9083 
9084 		/*
9085 		 * MAP_SHARED to prevent internal slot pages from being moved
9086 		 * by fork()/COW.
9087 		 */
9088 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9089 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9090 		if (IS_ERR((void *)hva))
9091 			return PTR_ERR((void *)hva);
9092 	} else {
9093 		if (!slot->npages)
9094 			return 0;
9095 
9096 		hva = 0;
9097 	}
9098 
9099 	old = *slot;
9100 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9101 		struct kvm_userspace_memory_region m;
9102 
9103 		m.slot = id | (i << 16);
9104 		m.flags = 0;
9105 		m.guest_phys_addr = gpa;
9106 		m.userspace_addr = hva;
9107 		m.memory_size = size;
9108 		r = __kvm_set_memory_region(kvm, &m);
9109 		if (r < 0)
9110 			return r;
9111 	}
9112 
9113 	if (!size)
9114 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9115 
9116 	return 0;
9117 }
9118 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9119 
9120 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9121 {
9122 	int r;
9123 
9124 	mutex_lock(&kvm->slots_lock);
9125 	r = __x86_set_memory_region(kvm, id, gpa, size);
9126 	mutex_unlock(&kvm->slots_lock);
9127 
9128 	return r;
9129 }
9130 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9131 
9132 void kvm_arch_destroy_vm(struct kvm *kvm)
9133 {
9134 	if (current->mm == kvm->mm) {
9135 		/*
9136 		 * Free memory regions allocated on behalf of userspace,
9137 		 * unless the the memory map has changed due to process exit
9138 		 * or fd copying.
9139 		 */
9140 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9141 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9142 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9143 	}
9144 	if (kvm_x86_ops->vm_destroy)
9145 		kvm_x86_ops->vm_destroy(kvm);
9146 	kvm_pic_destroy(kvm);
9147 	kvm_ioapic_destroy(kvm);
9148 	kvm_free_vcpus(kvm);
9149 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9150 	kvm_mmu_uninit_vm(kvm);
9151 	kvm_page_track_cleanup(kvm);
9152 	kvm_hv_destroy_vm(kvm);
9153 }
9154 
9155 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9156 			   struct kvm_memory_slot *dont)
9157 {
9158 	int i;
9159 
9160 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9161 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9162 			kvfree(free->arch.rmap[i]);
9163 			free->arch.rmap[i] = NULL;
9164 		}
9165 		if (i == 0)
9166 			continue;
9167 
9168 		if (!dont || free->arch.lpage_info[i - 1] !=
9169 			     dont->arch.lpage_info[i - 1]) {
9170 			kvfree(free->arch.lpage_info[i - 1]);
9171 			free->arch.lpage_info[i - 1] = NULL;
9172 		}
9173 	}
9174 
9175 	kvm_page_track_free_memslot(free, dont);
9176 }
9177 
9178 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9179 			    unsigned long npages)
9180 {
9181 	int i;
9182 
9183 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9184 		struct kvm_lpage_info *linfo;
9185 		unsigned long ugfn;
9186 		int lpages;
9187 		int level = i + 1;
9188 
9189 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9190 				      slot->base_gfn, level) + 1;
9191 
9192 		slot->arch.rmap[i] =
9193 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9194 				 GFP_KERNEL);
9195 		if (!slot->arch.rmap[i])
9196 			goto out_free;
9197 		if (i == 0)
9198 			continue;
9199 
9200 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9201 		if (!linfo)
9202 			goto out_free;
9203 
9204 		slot->arch.lpage_info[i - 1] = linfo;
9205 
9206 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9207 			linfo[0].disallow_lpage = 1;
9208 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9209 			linfo[lpages - 1].disallow_lpage = 1;
9210 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9211 		/*
9212 		 * If the gfn and userspace address are not aligned wrt each
9213 		 * other, or if explicitly asked to, disable large page
9214 		 * support for this slot
9215 		 */
9216 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9217 		    !kvm_largepages_enabled()) {
9218 			unsigned long j;
9219 
9220 			for (j = 0; j < lpages; ++j)
9221 				linfo[j].disallow_lpage = 1;
9222 		}
9223 	}
9224 
9225 	if (kvm_page_track_create_memslot(slot, npages))
9226 		goto out_free;
9227 
9228 	return 0;
9229 
9230 out_free:
9231 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9232 		kvfree(slot->arch.rmap[i]);
9233 		slot->arch.rmap[i] = NULL;
9234 		if (i == 0)
9235 			continue;
9236 
9237 		kvfree(slot->arch.lpage_info[i - 1]);
9238 		slot->arch.lpage_info[i - 1] = NULL;
9239 	}
9240 	return -ENOMEM;
9241 }
9242 
9243 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
9244 {
9245 	/*
9246 	 * memslots->generation has been incremented.
9247 	 * mmio generation may have reached its maximum value.
9248 	 */
9249 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
9250 }
9251 
9252 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9253 				struct kvm_memory_slot *memslot,
9254 				const struct kvm_userspace_memory_region *mem,
9255 				enum kvm_mr_change change)
9256 {
9257 	return 0;
9258 }
9259 
9260 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9261 				     struct kvm_memory_slot *new)
9262 {
9263 	/* Still write protect RO slot */
9264 	if (new->flags & KVM_MEM_READONLY) {
9265 		kvm_mmu_slot_remove_write_access(kvm, new);
9266 		return;
9267 	}
9268 
9269 	/*
9270 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9271 	 *
9272 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9273 	 *
9274 	 *  - KVM_MR_CREATE with dirty logging is disabled
9275 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9276 	 *
9277 	 * The reason is, in case of PML, we need to set D-bit for any slots
9278 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9279 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9280 	 * guarantees leaving PML enabled during guest's lifetime won't have
9281 	 * any additonal overhead from PML when guest is running with dirty
9282 	 * logging disabled for memory slots.
9283 	 *
9284 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9285 	 * to dirty logging mode.
9286 	 *
9287 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9288 	 *
9289 	 * In case of write protect:
9290 	 *
9291 	 * Write protect all pages for dirty logging.
9292 	 *
9293 	 * All the sptes including the large sptes which point to this
9294 	 * slot are set to readonly. We can not create any new large
9295 	 * spte on this slot until the end of the logging.
9296 	 *
9297 	 * See the comments in fast_page_fault().
9298 	 */
9299 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9300 		if (kvm_x86_ops->slot_enable_log_dirty)
9301 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9302 		else
9303 			kvm_mmu_slot_remove_write_access(kvm, new);
9304 	} else {
9305 		if (kvm_x86_ops->slot_disable_log_dirty)
9306 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9307 	}
9308 }
9309 
9310 void kvm_arch_commit_memory_region(struct kvm *kvm,
9311 				const struct kvm_userspace_memory_region *mem,
9312 				const struct kvm_memory_slot *old,
9313 				const struct kvm_memory_slot *new,
9314 				enum kvm_mr_change change)
9315 {
9316 	int nr_mmu_pages = 0;
9317 
9318 	if (!kvm->arch.n_requested_mmu_pages)
9319 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9320 
9321 	if (nr_mmu_pages)
9322 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9323 
9324 	/*
9325 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9326 	 * sptes have to be split.  If live migration is successful, the guest
9327 	 * in the source machine will be destroyed and large sptes will be
9328 	 * created in the destination. However, if the guest continues to run
9329 	 * in the source machine (for example if live migration fails), small
9330 	 * sptes will remain around and cause bad performance.
9331 	 *
9332 	 * Scan sptes if dirty logging has been stopped, dropping those
9333 	 * which can be collapsed into a single large-page spte.  Later
9334 	 * page faults will create the large-page sptes.
9335 	 */
9336 	if ((change != KVM_MR_DELETE) &&
9337 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9338 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9339 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9340 
9341 	/*
9342 	 * Set up write protection and/or dirty logging for the new slot.
9343 	 *
9344 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9345 	 * been zapped so no dirty logging staff is needed for old slot. For
9346 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9347 	 * new and it's also covered when dealing with the new slot.
9348 	 *
9349 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9350 	 */
9351 	if (change != KVM_MR_DELETE)
9352 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9353 }
9354 
9355 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9356 {
9357 	kvm_mmu_invalidate_zap_all_pages(kvm);
9358 }
9359 
9360 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9361 				   struct kvm_memory_slot *slot)
9362 {
9363 	kvm_page_track_flush_slot(kvm, slot);
9364 }
9365 
9366 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9367 {
9368 	return (is_guest_mode(vcpu) &&
9369 			kvm_x86_ops->guest_apic_has_interrupt &&
9370 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9371 }
9372 
9373 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9374 {
9375 	if (!list_empty_careful(&vcpu->async_pf.done))
9376 		return true;
9377 
9378 	if (kvm_apic_has_events(vcpu))
9379 		return true;
9380 
9381 	if (vcpu->arch.pv.pv_unhalted)
9382 		return true;
9383 
9384 	if (vcpu->arch.exception.pending)
9385 		return true;
9386 
9387 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9388 	    (vcpu->arch.nmi_pending &&
9389 	     kvm_x86_ops->nmi_allowed(vcpu)))
9390 		return true;
9391 
9392 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9393 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9394 		return true;
9395 
9396 	if (kvm_arch_interrupt_allowed(vcpu) &&
9397 	    (kvm_cpu_has_interrupt(vcpu) ||
9398 	    kvm_guest_apic_has_interrupt(vcpu)))
9399 		return true;
9400 
9401 	if (kvm_hv_has_stimer_pending(vcpu))
9402 		return true;
9403 
9404 	return false;
9405 }
9406 
9407 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9408 {
9409 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9410 }
9411 
9412 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9413 {
9414 	return vcpu->arch.preempted_in_kernel;
9415 }
9416 
9417 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9418 {
9419 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9420 }
9421 
9422 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9423 {
9424 	return kvm_x86_ops->interrupt_allowed(vcpu);
9425 }
9426 
9427 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9428 {
9429 	if (is_64_bit_mode(vcpu))
9430 		return kvm_rip_read(vcpu);
9431 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9432 		     kvm_rip_read(vcpu));
9433 }
9434 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9435 
9436 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9437 {
9438 	return kvm_get_linear_rip(vcpu) == linear_rip;
9439 }
9440 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9441 
9442 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9443 {
9444 	unsigned long rflags;
9445 
9446 	rflags = kvm_x86_ops->get_rflags(vcpu);
9447 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9448 		rflags &= ~X86_EFLAGS_TF;
9449 	return rflags;
9450 }
9451 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9452 
9453 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9454 {
9455 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9456 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9457 		rflags |= X86_EFLAGS_TF;
9458 	kvm_x86_ops->set_rflags(vcpu, rflags);
9459 }
9460 
9461 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9462 {
9463 	__kvm_set_rflags(vcpu, rflags);
9464 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9465 }
9466 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9467 
9468 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9469 {
9470 	int r;
9471 
9472 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9473 	      work->wakeup_all)
9474 		return;
9475 
9476 	r = kvm_mmu_reload(vcpu);
9477 	if (unlikely(r))
9478 		return;
9479 
9480 	if (!vcpu->arch.mmu->direct_map &&
9481 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9482 		return;
9483 
9484 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9485 }
9486 
9487 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9488 {
9489 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9490 }
9491 
9492 static inline u32 kvm_async_pf_next_probe(u32 key)
9493 {
9494 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9495 }
9496 
9497 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9498 {
9499 	u32 key = kvm_async_pf_hash_fn(gfn);
9500 
9501 	while (vcpu->arch.apf.gfns[key] != ~0)
9502 		key = kvm_async_pf_next_probe(key);
9503 
9504 	vcpu->arch.apf.gfns[key] = gfn;
9505 }
9506 
9507 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9508 {
9509 	int i;
9510 	u32 key = kvm_async_pf_hash_fn(gfn);
9511 
9512 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9513 		     (vcpu->arch.apf.gfns[key] != gfn &&
9514 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9515 		key = kvm_async_pf_next_probe(key);
9516 
9517 	return key;
9518 }
9519 
9520 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9521 {
9522 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9523 }
9524 
9525 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9526 {
9527 	u32 i, j, k;
9528 
9529 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9530 	while (true) {
9531 		vcpu->arch.apf.gfns[i] = ~0;
9532 		do {
9533 			j = kvm_async_pf_next_probe(j);
9534 			if (vcpu->arch.apf.gfns[j] == ~0)
9535 				return;
9536 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9537 			/*
9538 			 * k lies cyclically in ]i,j]
9539 			 * |    i.k.j |
9540 			 * |....j i.k.| or  |.k..j i...|
9541 			 */
9542 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9543 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9544 		i = j;
9545 	}
9546 }
9547 
9548 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9549 {
9550 
9551 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9552 				      sizeof(val));
9553 }
9554 
9555 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9556 {
9557 
9558 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9559 				      sizeof(u32));
9560 }
9561 
9562 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9563 				     struct kvm_async_pf *work)
9564 {
9565 	struct x86_exception fault;
9566 
9567 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9568 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9569 
9570 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9571 	    (vcpu->arch.apf.send_user_only &&
9572 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9573 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9574 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9575 		fault.vector = PF_VECTOR;
9576 		fault.error_code_valid = true;
9577 		fault.error_code = 0;
9578 		fault.nested_page_fault = false;
9579 		fault.address = work->arch.token;
9580 		fault.async_page_fault = true;
9581 		kvm_inject_page_fault(vcpu, &fault);
9582 	}
9583 }
9584 
9585 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9586 				 struct kvm_async_pf *work)
9587 {
9588 	struct x86_exception fault;
9589 	u32 val;
9590 
9591 	if (work->wakeup_all)
9592 		work->arch.token = ~0; /* broadcast wakeup */
9593 	else
9594 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9595 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9596 
9597 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9598 	    !apf_get_user(vcpu, &val)) {
9599 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9600 		    vcpu->arch.exception.pending &&
9601 		    vcpu->arch.exception.nr == PF_VECTOR &&
9602 		    !apf_put_user(vcpu, 0)) {
9603 			vcpu->arch.exception.injected = false;
9604 			vcpu->arch.exception.pending = false;
9605 			vcpu->arch.exception.nr = 0;
9606 			vcpu->arch.exception.has_error_code = false;
9607 			vcpu->arch.exception.error_code = 0;
9608 			vcpu->arch.exception.has_payload = false;
9609 			vcpu->arch.exception.payload = 0;
9610 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9611 			fault.vector = PF_VECTOR;
9612 			fault.error_code_valid = true;
9613 			fault.error_code = 0;
9614 			fault.nested_page_fault = false;
9615 			fault.address = work->arch.token;
9616 			fault.async_page_fault = true;
9617 			kvm_inject_page_fault(vcpu, &fault);
9618 		}
9619 	}
9620 	vcpu->arch.apf.halted = false;
9621 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9622 }
9623 
9624 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9625 {
9626 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9627 		return true;
9628 	else
9629 		return kvm_can_do_async_pf(vcpu);
9630 }
9631 
9632 void kvm_arch_start_assignment(struct kvm *kvm)
9633 {
9634 	atomic_inc(&kvm->arch.assigned_device_count);
9635 }
9636 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9637 
9638 void kvm_arch_end_assignment(struct kvm *kvm)
9639 {
9640 	atomic_dec(&kvm->arch.assigned_device_count);
9641 }
9642 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9643 
9644 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9645 {
9646 	return atomic_read(&kvm->arch.assigned_device_count);
9647 }
9648 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9649 
9650 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9651 {
9652 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9653 }
9654 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9655 
9656 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9657 {
9658 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9659 }
9660 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9661 
9662 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9663 {
9664 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9665 }
9666 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9667 
9668 bool kvm_arch_has_irq_bypass(void)
9669 {
9670 	return kvm_x86_ops->update_pi_irte != NULL;
9671 }
9672 
9673 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9674 				      struct irq_bypass_producer *prod)
9675 {
9676 	struct kvm_kernel_irqfd *irqfd =
9677 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9678 
9679 	irqfd->producer = prod;
9680 
9681 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9682 					   prod->irq, irqfd->gsi, 1);
9683 }
9684 
9685 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9686 				      struct irq_bypass_producer *prod)
9687 {
9688 	int ret;
9689 	struct kvm_kernel_irqfd *irqfd =
9690 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9691 
9692 	WARN_ON(irqfd->producer != prod);
9693 	irqfd->producer = NULL;
9694 
9695 	/*
9696 	 * When producer of consumer is unregistered, we change back to
9697 	 * remapped mode, so we can re-use the current implementation
9698 	 * when the irq is masked/disabled or the consumer side (KVM
9699 	 * int this case doesn't want to receive the interrupts.
9700 	*/
9701 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9702 	if (ret)
9703 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9704 		       " fails: %d\n", irqfd->consumer.token, ret);
9705 }
9706 
9707 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9708 				   uint32_t guest_irq, bool set)
9709 {
9710 	if (!kvm_x86_ops->update_pi_irte)
9711 		return -EINVAL;
9712 
9713 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9714 }
9715 
9716 bool kvm_vector_hashing_enabled(void)
9717 {
9718 	return vector_hashing;
9719 }
9720 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9721 
9722 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9723 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9724 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9725 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9726 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9727 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9728 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9729 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9730 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9731 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9732 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9733 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9738 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9739 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9740 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9741