xref: /openbmc/linux/arch/x86/kvm/x86.c (revision fb8d6c8d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "mmu.h"
22 #include "i8254.h"
23 #include "tss.h"
24 #include "kvm_cache_regs.h"
25 #include "x86.h"
26 #include "cpuid.h"
27 #include "pmu.h"
28 #include "hyperv.h"
29 
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
33 #include <linux/fs.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
56 
57 #include <trace/events/kvm.h>
58 
59 #include <asm/debugreg.h>
60 #include <asm/msr.h>
61 #include <asm/desc.h>
62 #include <asm/mce.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <clocksource/hyperv_timer.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, ## __VA_ARGS__
96 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
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 /*
139  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
140  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
141  * advancement entirely.  Any other value is used as-is and disables adaptive
142  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
143  */
144 static int __read_mostly lapic_timer_advance_ns = -1;
145 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
146 
147 static bool __read_mostly vector_hashing = true;
148 module_param(vector_hashing, bool, S_IRUGO);
149 
150 bool __read_mostly enable_vmware_backdoor = false;
151 module_param(enable_vmware_backdoor, bool, S_IRUGO);
152 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
153 
154 static bool __read_mostly force_emulation_prefix = false;
155 module_param(force_emulation_prefix, bool, S_IRUGO);
156 
157 int __read_mostly pi_inject_timer = -1;
158 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
159 
160 #define KVM_NR_SHARED_MSRS 16
161 
162 struct kvm_shared_msrs_global {
163 	int nr;
164 	u32 msrs[KVM_NR_SHARED_MSRS];
165 };
166 
167 struct kvm_shared_msrs {
168 	struct user_return_notifier urn;
169 	bool registered;
170 	struct kvm_shared_msr_values {
171 		u64 host;
172 		u64 curr;
173 	} values[KVM_NR_SHARED_MSRS];
174 };
175 
176 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
177 static struct kvm_shared_msrs __percpu *shared_msrs;
178 
179 struct kvm_stats_debugfs_item debugfs_entries[] = {
180 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
181 	{ "pf_guest", VCPU_STAT(pf_guest) },
182 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
183 	{ "invlpg", VCPU_STAT(invlpg) },
184 	{ "exits", VCPU_STAT(exits) },
185 	{ "io_exits", VCPU_STAT(io_exits) },
186 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
187 	{ "signal_exits", VCPU_STAT(signal_exits) },
188 	{ "irq_window", VCPU_STAT(irq_window_exits) },
189 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
190 	{ "halt_exits", VCPU_STAT(halt_exits) },
191 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
192 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
193 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
194 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
195 	{ "hypercalls", VCPU_STAT(hypercalls) },
196 	{ "request_irq", VCPU_STAT(request_irq_exits) },
197 	{ "irq_exits", VCPU_STAT(irq_exits) },
198 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
199 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
200 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
201 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
202 	{ "irq_injections", VCPU_STAT(irq_injections) },
203 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
204 	{ "req_event", VCPU_STAT(req_event) },
205 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
206 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
207 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
208 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
209 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
210 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
211 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
212 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
213 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
214 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
215 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
216 	{ "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
217 	{ "max_mmu_page_hash_collisions",
218 		VM_STAT(max_mmu_page_hash_collisions) },
219 	{ NULL }
220 };
221 
222 u64 __read_mostly host_xcr0;
223 
224 struct kmem_cache *x86_fpu_cache;
225 EXPORT_SYMBOL_GPL(x86_fpu_cache);
226 
227 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
228 
229 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
230 {
231 	int i;
232 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
233 		vcpu->arch.apf.gfns[i] = ~0;
234 }
235 
236 static void kvm_on_user_return(struct user_return_notifier *urn)
237 {
238 	unsigned slot;
239 	struct kvm_shared_msrs *locals
240 		= container_of(urn, struct kvm_shared_msrs, urn);
241 	struct kvm_shared_msr_values *values;
242 	unsigned long flags;
243 
244 	/*
245 	 * Disabling irqs at this point since the following code could be
246 	 * interrupted and executed through kvm_arch_hardware_disable()
247 	 */
248 	local_irq_save(flags);
249 	if (locals->registered) {
250 		locals->registered = false;
251 		user_return_notifier_unregister(urn);
252 	}
253 	local_irq_restore(flags);
254 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
255 		values = &locals->values[slot];
256 		if (values->host != values->curr) {
257 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
258 			values->curr = values->host;
259 		}
260 	}
261 }
262 
263 static void shared_msr_update(unsigned slot, u32 msr)
264 {
265 	u64 value;
266 	unsigned int cpu = smp_processor_id();
267 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
268 
269 	/* only read, and nobody should modify it at this time,
270 	 * so don't need lock */
271 	if (slot >= shared_msrs_global.nr) {
272 		printk(KERN_ERR "kvm: invalid MSR slot!");
273 		return;
274 	}
275 	rdmsrl_safe(msr, &value);
276 	smsr->values[slot].host = value;
277 	smsr->values[slot].curr = value;
278 }
279 
280 void kvm_define_shared_msr(unsigned slot, u32 msr)
281 {
282 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
283 	shared_msrs_global.msrs[slot] = msr;
284 	if (slot >= shared_msrs_global.nr)
285 		shared_msrs_global.nr = slot + 1;
286 }
287 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
288 
289 static void kvm_shared_msr_cpu_online(void)
290 {
291 	unsigned i;
292 
293 	for (i = 0; i < shared_msrs_global.nr; ++i)
294 		shared_msr_update(i, shared_msrs_global.msrs[i]);
295 }
296 
297 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
298 {
299 	unsigned int cpu = smp_processor_id();
300 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
301 	int err;
302 
303 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
304 		return 0;
305 	smsr->values[slot].curr = value;
306 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
307 	if (err)
308 		return 1;
309 
310 	if (!smsr->registered) {
311 		smsr->urn.on_user_return = kvm_on_user_return;
312 		user_return_notifier_register(&smsr->urn);
313 		smsr->registered = true;
314 	}
315 	return 0;
316 }
317 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
318 
319 static void drop_user_return_notifiers(void)
320 {
321 	unsigned int cpu = smp_processor_id();
322 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
323 
324 	if (smsr->registered)
325 		kvm_on_user_return(&smsr->urn);
326 }
327 
328 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
329 {
330 	return vcpu->arch.apic_base;
331 }
332 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
333 
334 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
335 {
336 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
337 }
338 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
339 
340 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
341 {
342 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
343 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
344 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
345 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
346 
347 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
348 		return 1;
349 	if (!msr_info->host_initiated) {
350 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
351 			return 1;
352 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
353 			return 1;
354 	}
355 
356 	kvm_lapic_set_base(vcpu, msr_info->data);
357 	return 0;
358 }
359 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
360 
361 asmlinkage __visible void kvm_spurious_fault(void)
362 {
363 	/* Fault while not rebooting.  We want the trace. */
364 	BUG_ON(!kvm_rebooting);
365 }
366 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
367 
368 #define EXCPT_BENIGN		0
369 #define EXCPT_CONTRIBUTORY	1
370 #define EXCPT_PF		2
371 
372 static int exception_class(int vector)
373 {
374 	switch (vector) {
375 	case PF_VECTOR:
376 		return EXCPT_PF;
377 	case DE_VECTOR:
378 	case TS_VECTOR:
379 	case NP_VECTOR:
380 	case SS_VECTOR:
381 	case GP_VECTOR:
382 		return EXCPT_CONTRIBUTORY;
383 	default:
384 		break;
385 	}
386 	return EXCPT_BENIGN;
387 }
388 
389 #define EXCPT_FAULT		0
390 #define EXCPT_TRAP		1
391 #define EXCPT_ABORT		2
392 #define EXCPT_INTERRUPT		3
393 
394 static int exception_type(int vector)
395 {
396 	unsigned int mask;
397 
398 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
399 		return EXCPT_INTERRUPT;
400 
401 	mask = 1 << vector;
402 
403 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
404 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
405 		return EXCPT_TRAP;
406 
407 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
408 		return EXCPT_ABORT;
409 
410 	/* Reserved exceptions will result in fault */
411 	return EXCPT_FAULT;
412 }
413 
414 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
415 {
416 	unsigned nr = vcpu->arch.exception.nr;
417 	bool has_payload = vcpu->arch.exception.has_payload;
418 	unsigned long payload = vcpu->arch.exception.payload;
419 
420 	if (!has_payload)
421 		return;
422 
423 	switch (nr) {
424 	case DB_VECTOR:
425 		/*
426 		 * "Certain debug exceptions may clear bit 0-3.  The
427 		 * remaining contents of the DR6 register are never
428 		 * cleared by the processor".
429 		 */
430 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
431 		/*
432 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
433 		 */
434 		vcpu->arch.dr6 |= DR6_RTM;
435 		vcpu->arch.dr6 |= payload;
436 		/*
437 		 * Bit 16 should be set in the payload whenever the #DB
438 		 * exception should clear DR6.RTM. This makes the payload
439 		 * compatible with the pending debug exceptions under VMX.
440 		 * Though not currently documented in the SDM, this also
441 		 * makes the payload compatible with the exit qualification
442 		 * for #DB exceptions under VMX.
443 		 */
444 		vcpu->arch.dr6 ^= payload & DR6_RTM;
445 		break;
446 	case PF_VECTOR:
447 		vcpu->arch.cr2 = payload;
448 		break;
449 	}
450 
451 	vcpu->arch.exception.has_payload = false;
452 	vcpu->arch.exception.payload = 0;
453 }
454 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
455 
456 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
457 		unsigned nr, bool has_error, u32 error_code,
458 	        bool has_payload, unsigned long payload, bool reinject)
459 {
460 	u32 prev_nr;
461 	int class1, class2;
462 
463 	kvm_make_request(KVM_REQ_EVENT, vcpu);
464 
465 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
466 	queue:
467 		if (has_error && !is_protmode(vcpu))
468 			has_error = false;
469 		if (reinject) {
470 			/*
471 			 * On vmentry, vcpu->arch.exception.pending is only
472 			 * true if an event injection was blocked by
473 			 * nested_run_pending.  In that case, however,
474 			 * vcpu_enter_guest requests an immediate exit,
475 			 * and the guest shouldn't proceed far enough to
476 			 * need reinjection.
477 			 */
478 			WARN_ON_ONCE(vcpu->arch.exception.pending);
479 			vcpu->arch.exception.injected = true;
480 			if (WARN_ON_ONCE(has_payload)) {
481 				/*
482 				 * A reinjected event has already
483 				 * delivered its payload.
484 				 */
485 				has_payload = false;
486 				payload = 0;
487 			}
488 		} else {
489 			vcpu->arch.exception.pending = true;
490 			vcpu->arch.exception.injected = false;
491 		}
492 		vcpu->arch.exception.has_error_code = has_error;
493 		vcpu->arch.exception.nr = nr;
494 		vcpu->arch.exception.error_code = error_code;
495 		vcpu->arch.exception.has_payload = has_payload;
496 		vcpu->arch.exception.payload = payload;
497 		/*
498 		 * In guest mode, payload delivery should be deferred,
499 		 * so that the L1 hypervisor can intercept #PF before
500 		 * CR2 is modified (or intercept #DB before DR6 is
501 		 * modified under nVMX).  However, for ABI
502 		 * compatibility with KVM_GET_VCPU_EVENTS and
503 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
504 		 * delivery unless userspace has enabled this
505 		 * functionality via the per-VM capability,
506 		 * KVM_CAP_EXCEPTION_PAYLOAD.
507 		 */
508 		if (!vcpu->kvm->arch.exception_payload_enabled ||
509 		    !is_guest_mode(vcpu))
510 			kvm_deliver_exception_payload(vcpu);
511 		return;
512 	}
513 
514 	/* to check exception */
515 	prev_nr = vcpu->arch.exception.nr;
516 	if (prev_nr == DF_VECTOR) {
517 		/* triple fault -> shutdown */
518 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
519 		return;
520 	}
521 	class1 = exception_class(prev_nr);
522 	class2 = exception_class(nr);
523 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
524 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
525 		/*
526 		 * Generate double fault per SDM Table 5-5.  Set
527 		 * exception.pending = true so that the double fault
528 		 * can trigger a nested vmexit.
529 		 */
530 		vcpu->arch.exception.pending = true;
531 		vcpu->arch.exception.injected = false;
532 		vcpu->arch.exception.has_error_code = true;
533 		vcpu->arch.exception.nr = DF_VECTOR;
534 		vcpu->arch.exception.error_code = 0;
535 		vcpu->arch.exception.has_payload = false;
536 		vcpu->arch.exception.payload = 0;
537 	} else
538 		/* replace previous exception with a new one in a hope
539 		   that instruction re-execution will regenerate lost
540 		   exception */
541 		goto queue;
542 }
543 
544 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
545 {
546 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
547 }
548 EXPORT_SYMBOL_GPL(kvm_queue_exception);
549 
550 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
551 {
552 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
553 }
554 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
555 
556 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
557 				  unsigned long payload)
558 {
559 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
560 }
561 
562 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
563 				    u32 error_code, unsigned long payload)
564 {
565 	kvm_multiple_exception(vcpu, nr, true, error_code,
566 			       true, payload, false);
567 }
568 
569 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
570 {
571 	if (err)
572 		kvm_inject_gp(vcpu, 0);
573 	else
574 		return kvm_skip_emulated_instruction(vcpu);
575 
576 	return 1;
577 }
578 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
579 
580 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
581 {
582 	++vcpu->stat.pf_guest;
583 	vcpu->arch.exception.nested_apf =
584 		is_guest_mode(vcpu) && fault->async_page_fault;
585 	if (vcpu->arch.exception.nested_apf) {
586 		vcpu->arch.apf.nested_apf_token = fault->address;
587 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
588 	} else {
589 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
590 					fault->address);
591 	}
592 }
593 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
594 
595 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
596 {
597 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
598 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
599 	else
600 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
601 
602 	return fault->nested_page_fault;
603 }
604 
605 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
606 {
607 	atomic_inc(&vcpu->arch.nmi_queued);
608 	kvm_make_request(KVM_REQ_NMI, vcpu);
609 }
610 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
611 
612 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
613 {
614 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
615 }
616 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
617 
618 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
619 {
620 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
621 }
622 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
623 
624 /*
625  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
626  * a #GP and return false.
627  */
628 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
629 {
630 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
631 		return true;
632 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
633 	return false;
634 }
635 EXPORT_SYMBOL_GPL(kvm_require_cpl);
636 
637 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
638 {
639 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
640 		return true;
641 
642 	kvm_queue_exception(vcpu, UD_VECTOR);
643 	return false;
644 }
645 EXPORT_SYMBOL_GPL(kvm_require_dr);
646 
647 /*
648  * This function will be used to read from the physical memory of the currently
649  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
650  * can read from guest physical or from the guest's guest physical memory.
651  */
652 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
653 			    gfn_t ngfn, void *data, int offset, int len,
654 			    u32 access)
655 {
656 	struct x86_exception exception;
657 	gfn_t real_gfn;
658 	gpa_t ngpa;
659 
660 	ngpa     = gfn_to_gpa(ngfn);
661 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
662 	if (real_gfn == UNMAPPED_GVA)
663 		return -EFAULT;
664 
665 	real_gfn = gpa_to_gfn(real_gfn);
666 
667 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
668 }
669 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
670 
671 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
672 			       void *data, int offset, int len, u32 access)
673 {
674 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
675 				       data, offset, len, access);
676 }
677 
678 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
679 {
680 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
681 	       rsvd_bits(1, 2);
682 }
683 
684 /*
685  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
686  */
687 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
688 {
689 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
690 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
691 	int i;
692 	int ret;
693 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
694 
695 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
696 				      offset * sizeof(u64), sizeof(pdpte),
697 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
698 	if (ret < 0) {
699 		ret = 0;
700 		goto out;
701 	}
702 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
703 		if ((pdpte[i] & PT_PRESENT_MASK) &&
704 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
705 			ret = 0;
706 			goto out;
707 		}
708 	}
709 	ret = 1;
710 
711 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
712 	__set_bit(VCPU_EXREG_PDPTR,
713 		  (unsigned long *)&vcpu->arch.regs_avail);
714 	__set_bit(VCPU_EXREG_PDPTR,
715 		  (unsigned long *)&vcpu->arch.regs_dirty);
716 out:
717 
718 	return ret;
719 }
720 EXPORT_SYMBOL_GPL(load_pdptrs);
721 
722 bool pdptrs_changed(struct kvm_vcpu *vcpu)
723 {
724 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
725 	bool changed = true;
726 	int offset;
727 	gfn_t gfn;
728 	int r;
729 
730 	if (!is_pae_paging(vcpu))
731 		return false;
732 
733 	if (!test_bit(VCPU_EXREG_PDPTR,
734 		      (unsigned long *)&vcpu->arch.regs_avail))
735 		return true;
736 
737 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
738 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
739 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
740 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
741 	if (r < 0)
742 		goto out;
743 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
744 out:
745 
746 	return changed;
747 }
748 EXPORT_SYMBOL_GPL(pdptrs_changed);
749 
750 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
751 {
752 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
753 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
754 
755 	cr0 |= X86_CR0_ET;
756 
757 #ifdef CONFIG_X86_64
758 	if (cr0 & 0xffffffff00000000UL)
759 		return 1;
760 #endif
761 
762 	cr0 &= ~CR0_RESERVED_BITS;
763 
764 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
765 		return 1;
766 
767 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
768 		return 1;
769 
770 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
771 #ifdef CONFIG_X86_64
772 		if ((vcpu->arch.efer & EFER_LME)) {
773 			int cs_db, cs_l;
774 
775 			if (!is_pae(vcpu))
776 				return 1;
777 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
778 			if (cs_l)
779 				return 1;
780 		} else
781 #endif
782 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
783 						 kvm_read_cr3(vcpu)))
784 			return 1;
785 	}
786 
787 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
788 		return 1;
789 
790 	kvm_x86_ops->set_cr0(vcpu, cr0);
791 
792 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
793 		kvm_clear_async_pf_completion_queue(vcpu);
794 		kvm_async_pf_hash_reset(vcpu);
795 	}
796 
797 	if ((cr0 ^ old_cr0) & update_bits)
798 		kvm_mmu_reset_context(vcpu);
799 
800 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
801 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
802 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
803 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
804 
805 	return 0;
806 }
807 EXPORT_SYMBOL_GPL(kvm_set_cr0);
808 
809 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
810 {
811 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
812 }
813 EXPORT_SYMBOL_GPL(kvm_lmsw);
814 
815 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
816 {
817 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
818 			!vcpu->guest_xcr0_loaded) {
819 		/* kvm_set_xcr() also depends on this */
820 		if (vcpu->arch.xcr0 != host_xcr0)
821 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
822 		vcpu->guest_xcr0_loaded = 1;
823 	}
824 }
825 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
826 
827 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
828 {
829 	if (vcpu->guest_xcr0_loaded) {
830 		if (vcpu->arch.xcr0 != host_xcr0)
831 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
832 		vcpu->guest_xcr0_loaded = 0;
833 	}
834 }
835 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
836 
837 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
838 {
839 	u64 xcr0 = xcr;
840 	u64 old_xcr0 = vcpu->arch.xcr0;
841 	u64 valid_bits;
842 
843 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
844 	if (index != XCR_XFEATURE_ENABLED_MASK)
845 		return 1;
846 	if (!(xcr0 & XFEATURE_MASK_FP))
847 		return 1;
848 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
849 		return 1;
850 
851 	/*
852 	 * Do not allow the guest to set bits that we do not support
853 	 * saving.  However, xcr0 bit 0 is always set, even if the
854 	 * emulated CPU does not support XSAVE (see fx_init).
855 	 */
856 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
857 	if (xcr0 & ~valid_bits)
858 		return 1;
859 
860 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
861 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
862 		return 1;
863 
864 	if (xcr0 & XFEATURE_MASK_AVX512) {
865 		if (!(xcr0 & XFEATURE_MASK_YMM))
866 			return 1;
867 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
868 			return 1;
869 	}
870 	vcpu->arch.xcr0 = xcr0;
871 
872 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
873 		kvm_update_cpuid(vcpu);
874 	return 0;
875 }
876 
877 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
878 {
879 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
880 	    __kvm_set_xcr(vcpu, index, xcr)) {
881 		kvm_inject_gp(vcpu, 0);
882 		return 1;
883 	}
884 	return 0;
885 }
886 EXPORT_SYMBOL_GPL(kvm_set_xcr);
887 
888 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
889 {
890 	if (cr4 & CR4_RESERVED_BITS)
891 		return -EINVAL;
892 
893 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
894 		return -EINVAL;
895 
896 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
897 		return -EINVAL;
898 
899 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
900 		return -EINVAL;
901 
902 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
903 		return -EINVAL;
904 
905 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
906 		return -EINVAL;
907 
908 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
909 		return -EINVAL;
910 
911 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
912 		return -EINVAL;
913 
914 	return 0;
915 }
916 
917 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
918 {
919 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
920 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
921 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
922 
923 	if (kvm_valid_cr4(vcpu, cr4))
924 		return 1;
925 
926 	if (is_long_mode(vcpu)) {
927 		if (!(cr4 & X86_CR4_PAE))
928 			return 1;
929 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
930 		   && ((cr4 ^ old_cr4) & pdptr_bits)
931 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
932 				   kvm_read_cr3(vcpu)))
933 		return 1;
934 
935 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
936 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
937 			return 1;
938 
939 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
940 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
941 			return 1;
942 	}
943 
944 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
945 		return 1;
946 
947 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
948 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
949 		kvm_mmu_reset_context(vcpu);
950 
951 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
952 		kvm_update_cpuid(vcpu);
953 
954 	return 0;
955 }
956 EXPORT_SYMBOL_GPL(kvm_set_cr4);
957 
958 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
959 {
960 	bool skip_tlb_flush = false;
961 #ifdef CONFIG_X86_64
962 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
963 
964 	if (pcid_enabled) {
965 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
966 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
967 	}
968 #endif
969 
970 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
971 		if (!skip_tlb_flush) {
972 			kvm_mmu_sync_roots(vcpu);
973 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
974 		}
975 		return 0;
976 	}
977 
978 	if (is_long_mode(vcpu) &&
979 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
980 		return 1;
981 	else if (is_pae_paging(vcpu) &&
982 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
983 		return 1;
984 
985 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
986 	vcpu->arch.cr3 = cr3;
987 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
988 
989 	return 0;
990 }
991 EXPORT_SYMBOL_GPL(kvm_set_cr3);
992 
993 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
994 {
995 	if (cr8 & CR8_RESERVED_BITS)
996 		return 1;
997 	if (lapic_in_kernel(vcpu))
998 		kvm_lapic_set_tpr(vcpu, cr8);
999 	else
1000 		vcpu->arch.cr8 = cr8;
1001 	return 0;
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1004 
1005 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1006 {
1007 	if (lapic_in_kernel(vcpu))
1008 		return kvm_lapic_get_cr8(vcpu);
1009 	else
1010 		return vcpu->arch.cr8;
1011 }
1012 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1013 
1014 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1015 {
1016 	int i;
1017 
1018 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1019 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1020 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1021 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1022 	}
1023 }
1024 
1025 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1026 {
1027 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1028 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1029 }
1030 
1031 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1032 {
1033 	unsigned long dr7;
1034 
1035 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1036 		dr7 = vcpu->arch.guest_debug_dr7;
1037 	else
1038 		dr7 = vcpu->arch.dr7;
1039 	kvm_x86_ops->set_dr7(vcpu, dr7);
1040 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1041 	if (dr7 & DR7_BP_EN_MASK)
1042 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1043 }
1044 
1045 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1046 {
1047 	u64 fixed = DR6_FIXED_1;
1048 
1049 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1050 		fixed |= DR6_RTM;
1051 	return fixed;
1052 }
1053 
1054 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1055 {
1056 	switch (dr) {
1057 	case 0 ... 3:
1058 		vcpu->arch.db[dr] = val;
1059 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1060 			vcpu->arch.eff_db[dr] = val;
1061 		break;
1062 	case 4:
1063 		/* fall through */
1064 	case 6:
1065 		if (val & 0xffffffff00000000ULL)
1066 			return -1; /* #GP */
1067 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1068 		kvm_update_dr6(vcpu);
1069 		break;
1070 	case 5:
1071 		/* fall through */
1072 	default: /* 7 */
1073 		if (val & 0xffffffff00000000ULL)
1074 			return -1; /* #GP */
1075 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1076 		kvm_update_dr7(vcpu);
1077 		break;
1078 	}
1079 
1080 	return 0;
1081 }
1082 
1083 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1084 {
1085 	if (__kvm_set_dr(vcpu, dr, val)) {
1086 		kvm_inject_gp(vcpu, 0);
1087 		return 1;
1088 	}
1089 	return 0;
1090 }
1091 EXPORT_SYMBOL_GPL(kvm_set_dr);
1092 
1093 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1094 {
1095 	switch (dr) {
1096 	case 0 ... 3:
1097 		*val = vcpu->arch.db[dr];
1098 		break;
1099 	case 4:
1100 		/* fall through */
1101 	case 6:
1102 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1103 			*val = vcpu->arch.dr6;
1104 		else
1105 			*val = kvm_x86_ops->get_dr6(vcpu);
1106 		break;
1107 	case 5:
1108 		/* fall through */
1109 	default: /* 7 */
1110 		*val = vcpu->arch.dr7;
1111 		break;
1112 	}
1113 	return 0;
1114 }
1115 EXPORT_SYMBOL_GPL(kvm_get_dr);
1116 
1117 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1118 {
1119 	u32 ecx = kvm_rcx_read(vcpu);
1120 	u64 data;
1121 	int err;
1122 
1123 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1124 	if (err)
1125 		return err;
1126 	kvm_rax_write(vcpu, (u32)data);
1127 	kvm_rdx_write(vcpu, data >> 32);
1128 	return err;
1129 }
1130 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1131 
1132 /*
1133  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1134  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1135  *
1136  * This list is modified at module load time to reflect the
1137  * capabilities of the host cpu. This capabilities test skips MSRs that are
1138  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1139  * may depend on host virtualization features rather than host cpu features.
1140  */
1141 
1142 static u32 msrs_to_save[] = {
1143 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1144 	MSR_STAR,
1145 #ifdef CONFIG_X86_64
1146 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1147 #endif
1148 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1149 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1150 	MSR_IA32_SPEC_CTRL,
1151 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1152 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1153 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1154 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1155 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1156 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1157 	MSR_IA32_UMWAIT_CONTROL,
1158 
1159 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1160 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1161 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1162 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1163 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1164 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1165 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1166 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1167 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1168 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1169 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1170 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1171 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1172 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1173 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1174 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1175 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1176 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1177 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1178 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1179 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1180 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1181 };
1182 
1183 static unsigned num_msrs_to_save;
1184 
1185 static u32 emulated_msrs[] = {
1186 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1187 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1188 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1189 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1190 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1191 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1192 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1193 	HV_X64_MSR_RESET,
1194 	HV_X64_MSR_VP_INDEX,
1195 	HV_X64_MSR_VP_RUNTIME,
1196 	HV_X64_MSR_SCONTROL,
1197 	HV_X64_MSR_STIMER0_CONFIG,
1198 	HV_X64_MSR_VP_ASSIST_PAGE,
1199 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1200 	HV_X64_MSR_TSC_EMULATION_STATUS,
1201 
1202 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1203 	MSR_KVM_PV_EOI_EN,
1204 
1205 	MSR_IA32_TSC_ADJUST,
1206 	MSR_IA32_TSCDEADLINE,
1207 	MSR_IA32_ARCH_CAPABILITIES,
1208 	MSR_IA32_MISC_ENABLE,
1209 	MSR_IA32_MCG_STATUS,
1210 	MSR_IA32_MCG_CTL,
1211 	MSR_IA32_MCG_EXT_CTL,
1212 	MSR_IA32_SMBASE,
1213 	MSR_SMI_COUNT,
1214 	MSR_PLATFORM_INFO,
1215 	MSR_MISC_FEATURES_ENABLES,
1216 	MSR_AMD64_VIRT_SPEC_CTRL,
1217 	MSR_IA32_POWER_CTL,
1218 
1219 	/*
1220 	 * The following list leaves out MSRs whose values are determined
1221 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1222 	 * We always support the "true" VMX control MSRs, even if the host
1223 	 * processor does not, so I am putting these registers here rather
1224 	 * than in msrs_to_save.
1225 	 */
1226 	MSR_IA32_VMX_BASIC,
1227 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1228 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1229 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1230 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1231 	MSR_IA32_VMX_MISC,
1232 	MSR_IA32_VMX_CR0_FIXED0,
1233 	MSR_IA32_VMX_CR4_FIXED0,
1234 	MSR_IA32_VMX_VMCS_ENUM,
1235 	MSR_IA32_VMX_PROCBASED_CTLS2,
1236 	MSR_IA32_VMX_EPT_VPID_CAP,
1237 	MSR_IA32_VMX_VMFUNC,
1238 
1239 	MSR_K7_HWCR,
1240 	MSR_KVM_POLL_CONTROL,
1241 };
1242 
1243 static unsigned num_emulated_msrs;
1244 
1245 /*
1246  * List of msr numbers which are used to expose MSR-based features that
1247  * can be used by a hypervisor to validate requested CPU features.
1248  */
1249 static u32 msr_based_features[] = {
1250 	MSR_IA32_VMX_BASIC,
1251 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1252 	MSR_IA32_VMX_PINBASED_CTLS,
1253 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1254 	MSR_IA32_VMX_PROCBASED_CTLS,
1255 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1256 	MSR_IA32_VMX_EXIT_CTLS,
1257 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1258 	MSR_IA32_VMX_ENTRY_CTLS,
1259 	MSR_IA32_VMX_MISC,
1260 	MSR_IA32_VMX_CR0_FIXED0,
1261 	MSR_IA32_VMX_CR0_FIXED1,
1262 	MSR_IA32_VMX_CR4_FIXED0,
1263 	MSR_IA32_VMX_CR4_FIXED1,
1264 	MSR_IA32_VMX_VMCS_ENUM,
1265 	MSR_IA32_VMX_PROCBASED_CTLS2,
1266 	MSR_IA32_VMX_EPT_VPID_CAP,
1267 	MSR_IA32_VMX_VMFUNC,
1268 
1269 	MSR_F10H_DECFG,
1270 	MSR_IA32_UCODE_REV,
1271 	MSR_IA32_ARCH_CAPABILITIES,
1272 };
1273 
1274 static unsigned int num_msr_based_features;
1275 
1276 static u64 kvm_get_arch_capabilities(void)
1277 {
1278 	u64 data = 0;
1279 
1280 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1281 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1282 
1283 	/*
1284 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1285 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1286 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1287 	 * L1 guests, so it need not worry about its own (L2) guests.
1288 	 */
1289 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1290 
1291 	/*
1292 	 * If we're doing cache flushes (either "always" or "cond")
1293 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1294 	 * If an outer hypervisor is doing the cache flush for us
1295 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1296 	 * capability to the guest too, and if EPT is disabled we're not
1297 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1298 	 * require a nested hypervisor to do a flush of its own.
1299 	 */
1300 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1301 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1302 
1303 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1304 		data |= ARCH_CAP_RDCL_NO;
1305 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1306 		data |= ARCH_CAP_SSB_NO;
1307 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1308 		data |= ARCH_CAP_MDS_NO;
1309 
1310 	/*
1311 	 * On TAA affected systems, export MDS_NO=0 when:
1312 	 *	- TSX is enabled on the host, i.e. X86_FEATURE_RTM=1.
1313 	 *	- Updated microcode is present. This is detected by
1314 	 *	  the presence of ARCH_CAP_TSX_CTRL_MSR and ensures
1315 	 *	  that VERW clears CPU buffers.
1316 	 *
1317 	 * When MDS_NO=0 is exported, guests deploy clear CPU buffer
1318 	 * mitigation and don't complain:
1319 	 *
1320 	 *	"Vulnerable: Clear CPU buffers attempted, no microcode"
1321 	 *
1322 	 * If TSX is disabled on the system, guests are also mitigated against
1323 	 * TAA and clear CPU buffer mitigation is not required for guests.
1324 	 */
1325 	if (boot_cpu_has_bug(X86_BUG_TAA) && boot_cpu_has(X86_FEATURE_RTM) &&
1326 	    (data & ARCH_CAP_TSX_CTRL_MSR))
1327 		data &= ~ARCH_CAP_MDS_NO;
1328 
1329 	return data;
1330 }
1331 
1332 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1333 {
1334 	switch (msr->index) {
1335 	case MSR_IA32_ARCH_CAPABILITIES:
1336 		msr->data = kvm_get_arch_capabilities();
1337 		break;
1338 	case MSR_IA32_UCODE_REV:
1339 		rdmsrl_safe(msr->index, &msr->data);
1340 		break;
1341 	default:
1342 		if (kvm_x86_ops->get_msr_feature(msr))
1343 			return 1;
1344 	}
1345 	return 0;
1346 }
1347 
1348 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1349 {
1350 	struct kvm_msr_entry msr;
1351 	int r;
1352 
1353 	msr.index = index;
1354 	r = kvm_get_msr_feature(&msr);
1355 	if (r)
1356 		return r;
1357 
1358 	*data = msr.data;
1359 
1360 	return 0;
1361 }
1362 
1363 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1364 {
1365 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1366 		return false;
1367 
1368 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1369 		return false;
1370 
1371 	if (efer & (EFER_LME | EFER_LMA) &&
1372 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1373 		return false;
1374 
1375 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1376 		return false;
1377 
1378 	return true;
1379 
1380 }
1381 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1382 {
1383 	if (efer & efer_reserved_bits)
1384 		return false;
1385 
1386 	return __kvm_valid_efer(vcpu, efer);
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1389 
1390 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1391 {
1392 	u64 old_efer = vcpu->arch.efer;
1393 	u64 efer = msr_info->data;
1394 
1395 	if (efer & efer_reserved_bits)
1396 		return 1;
1397 
1398 	if (!msr_info->host_initiated) {
1399 		if (!__kvm_valid_efer(vcpu, efer))
1400 			return 1;
1401 
1402 		if (is_paging(vcpu) &&
1403 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1404 			return 1;
1405 	}
1406 
1407 	efer &= ~EFER_LMA;
1408 	efer |= vcpu->arch.efer & EFER_LMA;
1409 
1410 	kvm_x86_ops->set_efer(vcpu, efer);
1411 
1412 	/* Update reserved bits */
1413 	if ((efer ^ old_efer) & EFER_NX)
1414 		kvm_mmu_reset_context(vcpu);
1415 
1416 	return 0;
1417 }
1418 
1419 void kvm_enable_efer_bits(u64 mask)
1420 {
1421        efer_reserved_bits &= ~mask;
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1424 
1425 /*
1426  * Write @data into the MSR specified by @index.  Select MSR specific fault
1427  * checks are bypassed if @host_initiated is %true.
1428  * Returns 0 on success, non-0 otherwise.
1429  * Assumes vcpu_load() was already called.
1430  */
1431 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1432 			 bool host_initiated)
1433 {
1434 	struct msr_data msr;
1435 
1436 	switch (index) {
1437 	case MSR_FS_BASE:
1438 	case MSR_GS_BASE:
1439 	case MSR_KERNEL_GS_BASE:
1440 	case MSR_CSTAR:
1441 	case MSR_LSTAR:
1442 		if (is_noncanonical_address(data, vcpu))
1443 			return 1;
1444 		break;
1445 	case MSR_IA32_SYSENTER_EIP:
1446 	case MSR_IA32_SYSENTER_ESP:
1447 		/*
1448 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1449 		 * non-canonical address is written on Intel but not on
1450 		 * AMD (which ignores the top 32-bits, because it does
1451 		 * not implement 64-bit SYSENTER).
1452 		 *
1453 		 * 64-bit code should hence be able to write a non-canonical
1454 		 * value on AMD.  Making the address canonical ensures that
1455 		 * vmentry does not fail on Intel after writing a non-canonical
1456 		 * value, and that something deterministic happens if the guest
1457 		 * invokes 64-bit SYSENTER.
1458 		 */
1459 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1460 	}
1461 
1462 	msr.data = data;
1463 	msr.index = index;
1464 	msr.host_initiated = host_initiated;
1465 
1466 	return kvm_x86_ops->set_msr(vcpu, &msr);
1467 }
1468 
1469 /*
1470  * Read the MSR specified by @index into @data.  Select MSR specific fault
1471  * checks are bypassed if @host_initiated is %true.
1472  * Returns 0 on success, non-0 otherwise.
1473  * Assumes vcpu_load() was already called.
1474  */
1475 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1476 			 bool host_initiated)
1477 {
1478 	struct msr_data msr;
1479 	int ret;
1480 
1481 	msr.index = index;
1482 	msr.host_initiated = host_initiated;
1483 
1484 	ret = kvm_x86_ops->get_msr(vcpu, &msr);
1485 	if (!ret)
1486 		*data = msr.data;
1487 	return ret;
1488 }
1489 
1490 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1491 {
1492 	return __kvm_get_msr(vcpu, index, data, false);
1493 }
1494 EXPORT_SYMBOL_GPL(kvm_get_msr);
1495 
1496 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1497 {
1498 	return __kvm_set_msr(vcpu, index, data, false);
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_set_msr);
1501 
1502 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1503 {
1504 	u32 ecx = kvm_rcx_read(vcpu);
1505 	u64 data;
1506 
1507 	if (kvm_get_msr(vcpu, ecx, &data)) {
1508 		trace_kvm_msr_read_ex(ecx);
1509 		kvm_inject_gp(vcpu, 0);
1510 		return 1;
1511 	}
1512 
1513 	trace_kvm_msr_read(ecx, data);
1514 
1515 	kvm_rax_write(vcpu, data & -1u);
1516 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1517 	return kvm_skip_emulated_instruction(vcpu);
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1520 
1521 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1522 {
1523 	u32 ecx = kvm_rcx_read(vcpu);
1524 	u64 data = kvm_read_edx_eax(vcpu);
1525 
1526 	if (kvm_set_msr(vcpu, ecx, data)) {
1527 		trace_kvm_msr_write_ex(ecx, data);
1528 		kvm_inject_gp(vcpu, 0);
1529 		return 1;
1530 	}
1531 
1532 	trace_kvm_msr_write(ecx, data);
1533 	return kvm_skip_emulated_instruction(vcpu);
1534 }
1535 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1536 
1537 /*
1538  * Adapt set_msr() to msr_io()'s calling convention
1539  */
1540 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1541 {
1542 	return __kvm_get_msr(vcpu, index, data, true);
1543 }
1544 
1545 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1546 {
1547 	return __kvm_set_msr(vcpu, index, *data, true);
1548 }
1549 
1550 #ifdef CONFIG_X86_64
1551 struct pvclock_gtod_data {
1552 	seqcount_t	seq;
1553 
1554 	struct { /* extract of a clocksource struct */
1555 		int vclock_mode;
1556 		u64	cycle_last;
1557 		u64	mask;
1558 		u32	mult;
1559 		u32	shift;
1560 	} clock;
1561 
1562 	u64		boot_ns;
1563 	u64		nsec_base;
1564 	u64		wall_time_sec;
1565 };
1566 
1567 static struct pvclock_gtod_data pvclock_gtod_data;
1568 
1569 static void update_pvclock_gtod(struct timekeeper *tk)
1570 {
1571 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1572 	u64 boot_ns;
1573 
1574 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1575 
1576 	write_seqcount_begin(&vdata->seq);
1577 
1578 	/* copy pvclock gtod data */
1579 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1580 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1581 	vdata->clock.mask		= tk->tkr_mono.mask;
1582 	vdata->clock.mult		= tk->tkr_mono.mult;
1583 	vdata->clock.shift		= tk->tkr_mono.shift;
1584 
1585 	vdata->boot_ns			= boot_ns;
1586 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1587 
1588 	vdata->wall_time_sec            = tk->xtime_sec;
1589 
1590 	write_seqcount_end(&vdata->seq);
1591 }
1592 #endif
1593 
1594 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1595 {
1596 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1597 	kvm_vcpu_kick(vcpu);
1598 }
1599 
1600 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1601 {
1602 	int version;
1603 	int r;
1604 	struct pvclock_wall_clock wc;
1605 	struct timespec64 boot;
1606 
1607 	if (!wall_clock)
1608 		return;
1609 
1610 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1611 	if (r)
1612 		return;
1613 
1614 	if (version & 1)
1615 		++version;  /* first time write, random junk */
1616 
1617 	++version;
1618 
1619 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1620 		return;
1621 
1622 	/*
1623 	 * The guest calculates current wall clock time by adding
1624 	 * system time (updated by kvm_guest_time_update below) to the
1625 	 * wall clock specified here.  guest system time equals host
1626 	 * system time for us, thus we must fill in host boot time here.
1627 	 */
1628 	getboottime64(&boot);
1629 
1630 	if (kvm->arch.kvmclock_offset) {
1631 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1632 		boot = timespec64_sub(boot, ts);
1633 	}
1634 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1635 	wc.nsec = boot.tv_nsec;
1636 	wc.version = version;
1637 
1638 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1639 
1640 	version++;
1641 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1642 }
1643 
1644 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1645 {
1646 	do_shl32_div32(dividend, divisor);
1647 	return dividend;
1648 }
1649 
1650 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1651 			       s8 *pshift, u32 *pmultiplier)
1652 {
1653 	uint64_t scaled64;
1654 	int32_t  shift = 0;
1655 	uint64_t tps64;
1656 	uint32_t tps32;
1657 
1658 	tps64 = base_hz;
1659 	scaled64 = scaled_hz;
1660 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1661 		tps64 >>= 1;
1662 		shift--;
1663 	}
1664 
1665 	tps32 = (uint32_t)tps64;
1666 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1667 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1668 			scaled64 >>= 1;
1669 		else
1670 			tps32 <<= 1;
1671 		shift++;
1672 	}
1673 
1674 	*pshift = shift;
1675 	*pmultiplier = div_frac(scaled64, tps32);
1676 }
1677 
1678 #ifdef CONFIG_X86_64
1679 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1680 #endif
1681 
1682 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1683 static unsigned long max_tsc_khz;
1684 
1685 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1686 {
1687 	u64 v = (u64)khz * (1000000 + ppm);
1688 	do_div(v, 1000000);
1689 	return v;
1690 }
1691 
1692 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1693 {
1694 	u64 ratio;
1695 
1696 	/* Guest TSC same frequency as host TSC? */
1697 	if (!scale) {
1698 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1699 		return 0;
1700 	}
1701 
1702 	/* TSC scaling supported? */
1703 	if (!kvm_has_tsc_control) {
1704 		if (user_tsc_khz > tsc_khz) {
1705 			vcpu->arch.tsc_catchup = 1;
1706 			vcpu->arch.tsc_always_catchup = 1;
1707 			return 0;
1708 		} else {
1709 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1710 			return -1;
1711 		}
1712 	}
1713 
1714 	/* TSC scaling required  - calculate ratio */
1715 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1716 				user_tsc_khz, tsc_khz);
1717 
1718 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1719 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1720 			            user_tsc_khz);
1721 		return -1;
1722 	}
1723 
1724 	vcpu->arch.tsc_scaling_ratio = ratio;
1725 	return 0;
1726 }
1727 
1728 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1729 {
1730 	u32 thresh_lo, thresh_hi;
1731 	int use_scaling = 0;
1732 
1733 	/* tsc_khz can be zero if TSC calibration fails */
1734 	if (user_tsc_khz == 0) {
1735 		/* set tsc_scaling_ratio to a safe value */
1736 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1737 		return -1;
1738 	}
1739 
1740 	/* Compute a scale to convert nanoseconds in TSC cycles */
1741 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1742 			   &vcpu->arch.virtual_tsc_shift,
1743 			   &vcpu->arch.virtual_tsc_mult);
1744 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1745 
1746 	/*
1747 	 * Compute the variation in TSC rate which is acceptable
1748 	 * within the range of tolerance and decide if the
1749 	 * rate being applied is within that bounds of the hardware
1750 	 * rate.  If so, no scaling or compensation need be done.
1751 	 */
1752 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1753 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1754 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1755 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1756 		use_scaling = 1;
1757 	}
1758 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1759 }
1760 
1761 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1762 {
1763 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1764 				      vcpu->arch.virtual_tsc_mult,
1765 				      vcpu->arch.virtual_tsc_shift);
1766 	tsc += vcpu->arch.this_tsc_write;
1767 	return tsc;
1768 }
1769 
1770 static inline int gtod_is_based_on_tsc(int mode)
1771 {
1772 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1773 }
1774 
1775 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1776 {
1777 #ifdef CONFIG_X86_64
1778 	bool vcpus_matched;
1779 	struct kvm_arch *ka = &vcpu->kvm->arch;
1780 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1781 
1782 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1783 			 atomic_read(&vcpu->kvm->online_vcpus));
1784 
1785 	/*
1786 	 * Once the masterclock is enabled, always perform request in
1787 	 * order to update it.
1788 	 *
1789 	 * In order to enable masterclock, the host clocksource must be TSC
1790 	 * and the vcpus need to have matched TSCs.  When that happens,
1791 	 * perform request to enable masterclock.
1792 	 */
1793 	if (ka->use_master_clock ||
1794 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1795 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1796 
1797 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1798 			    atomic_read(&vcpu->kvm->online_vcpus),
1799 		            ka->use_master_clock, gtod->clock.vclock_mode);
1800 #endif
1801 }
1802 
1803 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1804 {
1805 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1806 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1807 }
1808 
1809 /*
1810  * Multiply tsc by a fixed point number represented by ratio.
1811  *
1812  * The most significant 64-N bits (mult) of ratio represent the
1813  * integral part of the fixed point number; the remaining N bits
1814  * (frac) represent the fractional part, ie. ratio represents a fixed
1815  * point number (mult + frac * 2^(-N)).
1816  *
1817  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1818  */
1819 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1820 {
1821 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1822 }
1823 
1824 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1825 {
1826 	u64 _tsc = tsc;
1827 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1828 
1829 	if (ratio != kvm_default_tsc_scaling_ratio)
1830 		_tsc = __scale_tsc(ratio, tsc);
1831 
1832 	return _tsc;
1833 }
1834 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1835 
1836 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1837 {
1838 	u64 tsc;
1839 
1840 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1841 
1842 	return target_tsc - tsc;
1843 }
1844 
1845 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1846 {
1847 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1848 
1849 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1850 }
1851 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1852 
1853 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1854 {
1855 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1856 }
1857 
1858 static inline bool kvm_check_tsc_unstable(void)
1859 {
1860 #ifdef CONFIG_X86_64
1861 	/*
1862 	 * TSC is marked unstable when we're running on Hyper-V,
1863 	 * 'TSC page' clocksource is good.
1864 	 */
1865 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1866 		return false;
1867 #endif
1868 	return check_tsc_unstable();
1869 }
1870 
1871 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1872 {
1873 	struct kvm *kvm = vcpu->kvm;
1874 	u64 offset, ns, elapsed;
1875 	unsigned long flags;
1876 	bool matched;
1877 	bool already_matched;
1878 	u64 data = msr->data;
1879 	bool synchronizing = false;
1880 
1881 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1882 	offset = kvm_compute_tsc_offset(vcpu, data);
1883 	ns = ktime_get_boottime_ns();
1884 	elapsed = ns - kvm->arch.last_tsc_nsec;
1885 
1886 	if (vcpu->arch.virtual_tsc_khz) {
1887 		if (data == 0 && msr->host_initiated) {
1888 			/*
1889 			 * detection of vcpu initialization -- need to sync
1890 			 * with other vCPUs. This particularly helps to keep
1891 			 * kvm_clock stable after CPU hotplug
1892 			 */
1893 			synchronizing = true;
1894 		} else {
1895 			u64 tsc_exp = kvm->arch.last_tsc_write +
1896 						nsec_to_cycles(vcpu, elapsed);
1897 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1898 			/*
1899 			 * Special case: TSC write with a small delta (1 second)
1900 			 * of virtual cycle time against real time is
1901 			 * interpreted as an attempt to synchronize the CPU.
1902 			 */
1903 			synchronizing = data < tsc_exp + tsc_hz &&
1904 					data + tsc_hz > tsc_exp;
1905 		}
1906 	}
1907 
1908 	/*
1909 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1910 	 * TSC, we add elapsed time in this computation.  We could let the
1911 	 * compensation code attempt to catch up if we fall behind, but
1912 	 * it's better to try to match offsets from the beginning.
1913          */
1914 	if (synchronizing &&
1915 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1916 		if (!kvm_check_tsc_unstable()) {
1917 			offset = kvm->arch.cur_tsc_offset;
1918 		} else {
1919 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1920 			data += delta;
1921 			offset = kvm_compute_tsc_offset(vcpu, data);
1922 		}
1923 		matched = true;
1924 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1925 	} else {
1926 		/*
1927 		 * We split periods of matched TSC writes into generations.
1928 		 * For each generation, we track the original measured
1929 		 * nanosecond time, offset, and write, so if TSCs are in
1930 		 * sync, we can match exact offset, and if not, we can match
1931 		 * exact software computation in compute_guest_tsc()
1932 		 *
1933 		 * These values are tracked in kvm->arch.cur_xxx variables.
1934 		 */
1935 		kvm->arch.cur_tsc_generation++;
1936 		kvm->arch.cur_tsc_nsec = ns;
1937 		kvm->arch.cur_tsc_write = data;
1938 		kvm->arch.cur_tsc_offset = offset;
1939 		matched = false;
1940 	}
1941 
1942 	/*
1943 	 * We also track th most recent recorded KHZ, write and time to
1944 	 * allow the matching interval to be extended at each write.
1945 	 */
1946 	kvm->arch.last_tsc_nsec = ns;
1947 	kvm->arch.last_tsc_write = data;
1948 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1949 
1950 	vcpu->arch.last_guest_tsc = data;
1951 
1952 	/* Keep track of which generation this VCPU has synchronized to */
1953 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1954 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1955 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1956 
1957 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1958 		update_ia32_tsc_adjust_msr(vcpu, offset);
1959 
1960 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1961 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1962 
1963 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1964 	if (!matched) {
1965 		kvm->arch.nr_vcpus_matched_tsc = 0;
1966 	} else if (!already_matched) {
1967 		kvm->arch.nr_vcpus_matched_tsc++;
1968 	}
1969 
1970 	kvm_track_tsc_matching(vcpu);
1971 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1972 }
1973 
1974 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1975 
1976 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1977 					   s64 adjustment)
1978 {
1979 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1980 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1981 }
1982 
1983 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1984 {
1985 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1986 		WARN_ON(adjustment < 0);
1987 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1988 	adjust_tsc_offset_guest(vcpu, adjustment);
1989 }
1990 
1991 #ifdef CONFIG_X86_64
1992 
1993 static u64 read_tsc(void)
1994 {
1995 	u64 ret = (u64)rdtsc_ordered();
1996 	u64 last = pvclock_gtod_data.clock.cycle_last;
1997 
1998 	if (likely(ret >= last))
1999 		return ret;
2000 
2001 	/*
2002 	 * GCC likes to generate cmov here, but this branch is extremely
2003 	 * predictable (it's just a function of time and the likely is
2004 	 * very likely) and there's a data dependence, so force GCC
2005 	 * to generate a branch instead.  I don't barrier() because
2006 	 * we don't actually need a barrier, and if this function
2007 	 * ever gets inlined it will generate worse code.
2008 	 */
2009 	asm volatile ("");
2010 	return last;
2011 }
2012 
2013 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
2014 {
2015 	long v;
2016 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2017 	u64 tsc_pg_val;
2018 
2019 	switch (gtod->clock.vclock_mode) {
2020 	case VCLOCK_HVCLOCK:
2021 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2022 						  tsc_timestamp);
2023 		if (tsc_pg_val != U64_MAX) {
2024 			/* TSC page valid */
2025 			*mode = VCLOCK_HVCLOCK;
2026 			v = (tsc_pg_val - gtod->clock.cycle_last) &
2027 				gtod->clock.mask;
2028 		} else {
2029 			/* TSC page invalid */
2030 			*mode = VCLOCK_NONE;
2031 		}
2032 		break;
2033 	case VCLOCK_TSC:
2034 		*mode = VCLOCK_TSC;
2035 		*tsc_timestamp = read_tsc();
2036 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
2037 			gtod->clock.mask;
2038 		break;
2039 	default:
2040 		*mode = VCLOCK_NONE;
2041 	}
2042 
2043 	if (*mode == VCLOCK_NONE)
2044 		*tsc_timestamp = v = 0;
2045 
2046 	return v * gtod->clock.mult;
2047 }
2048 
2049 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2050 {
2051 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2052 	unsigned long seq;
2053 	int mode;
2054 	u64 ns;
2055 
2056 	do {
2057 		seq = read_seqcount_begin(&gtod->seq);
2058 		ns = gtod->nsec_base;
2059 		ns += vgettsc(tsc_timestamp, &mode);
2060 		ns >>= gtod->clock.shift;
2061 		ns += gtod->boot_ns;
2062 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2063 	*t = ns;
2064 
2065 	return mode;
2066 }
2067 
2068 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2069 {
2070 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2071 	unsigned long seq;
2072 	int mode;
2073 	u64 ns;
2074 
2075 	do {
2076 		seq = read_seqcount_begin(&gtod->seq);
2077 		ts->tv_sec = gtod->wall_time_sec;
2078 		ns = gtod->nsec_base;
2079 		ns += vgettsc(tsc_timestamp, &mode);
2080 		ns >>= gtod->clock.shift;
2081 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2082 
2083 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2084 	ts->tv_nsec = ns;
2085 
2086 	return mode;
2087 }
2088 
2089 /* returns true if host is using TSC based clocksource */
2090 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2091 {
2092 	/* checked again under seqlock below */
2093 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2094 		return false;
2095 
2096 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2097 						      tsc_timestamp));
2098 }
2099 
2100 /* returns true if host is using TSC based clocksource */
2101 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2102 					   u64 *tsc_timestamp)
2103 {
2104 	/* checked again under seqlock below */
2105 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2106 		return false;
2107 
2108 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2109 }
2110 #endif
2111 
2112 /*
2113  *
2114  * Assuming a stable TSC across physical CPUS, and a stable TSC
2115  * across virtual CPUs, the following condition is possible.
2116  * Each numbered line represents an event visible to both
2117  * CPUs at the next numbered event.
2118  *
2119  * "timespecX" represents host monotonic time. "tscX" represents
2120  * RDTSC value.
2121  *
2122  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2123  *
2124  * 1.  read timespec0,tsc0
2125  * 2.					| timespec1 = timespec0 + N
2126  * 					| tsc1 = tsc0 + M
2127  * 3. transition to guest		| transition to guest
2128  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2129  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2130  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2131  *
2132  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2133  *
2134  * 	- ret0 < ret1
2135  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2136  *		...
2137  *	- 0 < N - M => M < N
2138  *
2139  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2140  * always the case (the difference between two distinct xtime instances
2141  * might be smaller then the difference between corresponding TSC reads,
2142  * when updating guest vcpus pvclock areas).
2143  *
2144  * To avoid that problem, do not allow visibility of distinct
2145  * system_timestamp/tsc_timestamp values simultaneously: use a master
2146  * copy of host monotonic time values. Update that master copy
2147  * in lockstep.
2148  *
2149  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2150  *
2151  */
2152 
2153 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2154 {
2155 #ifdef CONFIG_X86_64
2156 	struct kvm_arch *ka = &kvm->arch;
2157 	int vclock_mode;
2158 	bool host_tsc_clocksource, vcpus_matched;
2159 
2160 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2161 			atomic_read(&kvm->online_vcpus));
2162 
2163 	/*
2164 	 * If the host uses TSC clock, then passthrough TSC as stable
2165 	 * to the guest.
2166 	 */
2167 	host_tsc_clocksource = kvm_get_time_and_clockread(
2168 					&ka->master_kernel_ns,
2169 					&ka->master_cycle_now);
2170 
2171 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2172 				&& !ka->backwards_tsc_observed
2173 				&& !ka->boot_vcpu_runs_old_kvmclock;
2174 
2175 	if (ka->use_master_clock)
2176 		atomic_set(&kvm_guest_has_master_clock, 1);
2177 
2178 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2179 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2180 					vcpus_matched);
2181 #endif
2182 }
2183 
2184 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2185 {
2186 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2187 }
2188 
2189 static void kvm_gen_update_masterclock(struct kvm *kvm)
2190 {
2191 #ifdef CONFIG_X86_64
2192 	int i;
2193 	struct kvm_vcpu *vcpu;
2194 	struct kvm_arch *ka = &kvm->arch;
2195 
2196 	spin_lock(&ka->pvclock_gtod_sync_lock);
2197 	kvm_make_mclock_inprogress_request(kvm);
2198 	/* no guest entries from this point */
2199 	pvclock_update_vm_gtod_copy(kvm);
2200 
2201 	kvm_for_each_vcpu(i, vcpu, kvm)
2202 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2203 
2204 	/* guest entries allowed */
2205 	kvm_for_each_vcpu(i, vcpu, kvm)
2206 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2207 
2208 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2209 #endif
2210 }
2211 
2212 u64 get_kvmclock_ns(struct kvm *kvm)
2213 {
2214 	struct kvm_arch *ka = &kvm->arch;
2215 	struct pvclock_vcpu_time_info hv_clock;
2216 	u64 ret;
2217 
2218 	spin_lock(&ka->pvclock_gtod_sync_lock);
2219 	if (!ka->use_master_clock) {
2220 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2221 		return ktime_get_boottime_ns() + ka->kvmclock_offset;
2222 	}
2223 
2224 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2225 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2226 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2227 
2228 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2229 	get_cpu();
2230 
2231 	if (__this_cpu_read(cpu_tsc_khz)) {
2232 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2233 				   &hv_clock.tsc_shift,
2234 				   &hv_clock.tsc_to_system_mul);
2235 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2236 	} else
2237 		ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2238 
2239 	put_cpu();
2240 
2241 	return ret;
2242 }
2243 
2244 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2245 {
2246 	struct kvm_vcpu_arch *vcpu = &v->arch;
2247 	struct pvclock_vcpu_time_info guest_hv_clock;
2248 
2249 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2250 		&guest_hv_clock, sizeof(guest_hv_clock))))
2251 		return;
2252 
2253 	/* This VCPU is paused, but it's legal for a guest to read another
2254 	 * VCPU's kvmclock, so we really have to follow the specification where
2255 	 * it says that version is odd if data is being modified, and even after
2256 	 * it is consistent.
2257 	 *
2258 	 * Version field updates must be kept separate.  This is because
2259 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2260 	 * writes within a string instruction are weakly ordered.  So there
2261 	 * are three writes overall.
2262 	 *
2263 	 * As a small optimization, only write the version field in the first
2264 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2265 	 * version field is the first in the struct.
2266 	 */
2267 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2268 
2269 	if (guest_hv_clock.version & 1)
2270 		++guest_hv_clock.version;  /* first time write, random junk */
2271 
2272 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2273 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2274 				&vcpu->hv_clock,
2275 				sizeof(vcpu->hv_clock.version));
2276 
2277 	smp_wmb();
2278 
2279 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2280 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2281 
2282 	if (vcpu->pvclock_set_guest_stopped_request) {
2283 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2284 		vcpu->pvclock_set_guest_stopped_request = false;
2285 	}
2286 
2287 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2288 
2289 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2290 				&vcpu->hv_clock,
2291 				sizeof(vcpu->hv_clock));
2292 
2293 	smp_wmb();
2294 
2295 	vcpu->hv_clock.version++;
2296 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2297 				&vcpu->hv_clock,
2298 				sizeof(vcpu->hv_clock.version));
2299 }
2300 
2301 static int kvm_guest_time_update(struct kvm_vcpu *v)
2302 {
2303 	unsigned long flags, tgt_tsc_khz;
2304 	struct kvm_vcpu_arch *vcpu = &v->arch;
2305 	struct kvm_arch *ka = &v->kvm->arch;
2306 	s64 kernel_ns;
2307 	u64 tsc_timestamp, host_tsc;
2308 	u8 pvclock_flags;
2309 	bool use_master_clock;
2310 
2311 	kernel_ns = 0;
2312 	host_tsc = 0;
2313 
2314 	/*
2315 	 * If the host uses TSC clock, then passthrough TSC as stable
2316 	 * to the guest.
2317 	 */
2318 	spin_lock(&ka->pvclock_gtod_sync_lock);
2319 	use_master_clock = ka->use_master_clock;
2320 	if (use_master_clock) {
2321 		host_tsc = ka->master_cycle_now;
2322 		kernel_ns = ka->master_kernel_ns;
2323 	}
2324 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2325 
2326 	/* Keep irq disabled to prevent changes to the clock */
2327 	local_irq_save(flags);
2328 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2329 	if (unlikely(tgt_tsc_khz == 0)) {
2330 		local_irq_restore(flags);
2331 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2332 		return 1;
2333 	}
2334 	if (!use_master_clock) {
2335 		host_tsc = rdtsc();
2336 		kernel_ns = ktime_get_boottime_ns();
2337 	}
2338 
2339 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2340 
2341 	/*
2342 	 * We may have to catch up the TSC to match elapsed wall clock
2343 	 * time for two reasons, even if kvmclock is used.
2344 	 *   1) CPU could have been running below the maximum TSC rate
2345 	 *   2) Broken TSC compensation resets the base at each VCPU
2346 	 *      entry to avoid unknown leaps of TSC even when running
2347 	 *      again on the same CPU.  This may cause apparent elapsed
2348 	 *      time to disappear, and the guest to stand still or run
2349 	 *	very slowly.
2350 	 */
2351 	if (vcpu->tsc_catchup) {
2352 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2353 		if (tsc > tsc_timestamp) {
2354 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2355 			tsc_timestamp = tsc;
2356 		}
2357 	}
2358 
2359 	local_irq_restore(flags);
2360 
2361 	/* With all the info we got, fill in the values */
2362 
2363 	if (kvm_has_tsc_control)
2364 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2365 
2366 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2367 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2368 				   &vcpu->hv_clock.tsc_shift,
2369 				   &vcpu->hv_clock.tsc_to_system_mul);
2370 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2371 	}
2372 
2373 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2374 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2375 	vcpu->last_guest_tsc = tsc_timestamp;
2376 
2377 	/* If the host uses TSC clocksource, then it is stable */
2378 	pvclock_flags = 0;
2379 	if (use_master_clock)
2380 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2381 
2382 	vcpu->hv_clock.flags = pvclock_flags;
2383 
2384 	if (vcpu->pv_time_enabled)
2385 		kvm_setup_pvclock_page(v);
2386 	if (v == kvm_get_vcpu(v->kvm, 0))
2387 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2388 	return 0;
2389 }
2390 
2391 /*
2392  * kvmclock updates which are isolated to a given vcpu, such as
2393  * vcpu->cpu migration, should not allow system_timestamp from
2394  * the rest of the vcpus to remain static. Otherwise ntp frequency
2395  * correction applies to one vcpu's system_timestamp but not
2396  * the others.
2397  *
2398  * So in those cases, request a kvmclock update for all vcpus.
2399  * We need to rate-limit these requests though, as they can
2400  * considerably slow guests that have a large number of vcpus.
2401  * The time for a remote vcpu to update its kvmclock is bound
2402  * by the delay we use to rate-limit the updates.
2403  */
2404 
2405 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2406 
2407 static void kvmclock_update_fn(struct work_struct *work)
2408 {
2409 	int i;
2410 	struct delayed_work *dwork = to_delayed_work(work);
2411 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2412 					   kvmclock_update_work);
2413 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2414 	struct kvm_vcpu *vcpu;
2415 
2416 	kvm_for_each_vcpu(i, vcpu, kvm) {
2417 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2418 		kvm_vcpu_kick(vcpu);
2419 	}
2420 }
2421 
2422 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2423 {
2424 	struct kvm *kvm = v->kvm;
2425 
2426 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2427 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2428 					KVMCLOCK_UPDATE_DELAY);
2429 }
2430 
2431 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2432 
2433 static void kvmclock_sync_fn(struct work_struct *work)
2434 {
2435 	struct delayed_work *dwork = to_delayed_work(work);
2436 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2437 					   kvmclock_sync_work);
2438 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2439 
2440 	if (!kvmclock_periodic_sync)
2441 		return;
2442 
2443 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2444 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2445 					KVMCLOCK_SYNC_PERIOD);
2446 }
2447 
2448 /*
2449  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2450  */
2451 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2452 {
2453 	/* McStatusWrEn enabled? */
2454 	if (guest_cpuid_is_amd(vcpu))
2455 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2456 
2457 	return false;
2458 }
2459 
2460 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2461 {
2462 	u64 mcg_cap = vcpu->arch.mcg_cap;
2463 	unsigned bank_num = mcg_cap & 0xff;
2464 	u32 msr = msr_info->index;
2465 	u64 data = msr_info->data;
2466 
2467 	switch (msr) {
2468 	case MSR_IA32_MCG_STATUS:
2469 		vcpu->arch.mcg_status = data;
2470 		break;
2471 	case MSR_IA32_MCG_CTL:
2472 		if (!(mcg_cap & MCG_CTL_P) &&
2473 		    (data || !msr_info->host_initiated))
2474 			return 1;
2475 		if (data != 0 && data != ~(u64)0)
2476 			return 1;
2477 		vcpu->arch.mcg_ctl = data;
2478 		break;
2479 	default:
2480 		if (msr >= MSR_IA32_MC0_CTL &&
2481 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2482 			u32 offset = msr - MSR_IA32_MC0_CTL;
2483 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2484 			 * some Linux kernels though clear bit 10 in bank 4 to
2485 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2486 			 * this to avoid an uncatched #GP in the guest
2487 			 */
2488 			if ((offset & 0x3) == 0 &&
2489 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2490 				return -1;
2491 
2492 			/* MCi_STATUS */
2493 			if (!msr_info->host_initiated &&
2494 			    (offset & 0x3) == 1 && data != 0) {
2495 				if (!can_set_mci_status(vcpu))
2496 					return -1;
2497 			}
2498 
2499 			vcpu->arch.mce_banks[offset] = data;
2500 			break;
2501 		}
2502 		return 1;
2503 	}
2504 	return 0;
2505 }
2506 
2507 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2508 {
2509 	struct kvm *kvm = vcpu->kvm;
2510 	int lm = is_long_mode(vcpu);
2511 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2512 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2513 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2514 		: kvm->arch.xen_hvm_config.blob_size_32;
2515 	u32 page_num = data & ~PAGE_MASK;
2516 	u64 page_addr = data & PAGE_MASK;
2517 	u8 *page;
2518 	int r;
2519 
2520 	r = -E2BIG;
2521 	if (page_num >= blob_size)
2522 		goto out;
2523 	r = -ENOMEM;
2524 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2525 	if (IS_ERR(page)) {
2526 		r = PTR_ERR(page);
2527 		goto out;
2528 	}
2529 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2530 		goto out_free;
2531 	r = 0;
2532 out_free:
2533 	kfree(page);
2534 out:
2535 	return r;
2536 }
2537 
2538 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2539 {
2540 	gpa_t gpa = data & ~0x3f;
2541 
2542 	/* Bits 3:5 are reserved, Should be zero */
2543 	if (data & 0x38)
2544 		return 1;
2545 
2546 	vcpu->arch.apf.msr_val = data;
2547 
2548 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2549 		kvm_clear_async_pf_completion_queue(vcpu);
2550 		kvm_async_pf_hash_reset(vcpu);
2551 		return 0;
2552 	}
2553 
2554 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2555 					sizeof(u32)))
2556 		return 1;
2557 
2558 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2559 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2560 	kvm_async_pf_wakeup_all(vcpu);
2561 	return 0;
2562 }
2563 
2564 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2565 {
2566 	vcpu->arch.pv_time_enabled = false;
2567 	vcpu->arch.time = 0;
2568 }
2569 
2570 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2571 {
2572 	++vcpu->stat.tlb_flush;
2573 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2574 }
2575 
2576 static void record_steal_time(struct kvm_vcpu *vcpu)
2577 {
2578 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2579 		return;
2580 
2581 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2582 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2583 		return;
2584 
2585 	/*
2586 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2587 	 * expensive IPIs.
2588 	 */
2589 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2590 		vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2591 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2592 		kvm_vcpu_flush_tlb(vcpu, false);
2593 
2594 	if (vcpu->arch.st.steal.version & 1)
2595 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2596 
2597 	vcpu->arch.st.steal.version += 1;
2598 
2599 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2600 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2601 
2602 	smp_wmb();
2603 
2604 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2605 		vcpu->arch.st.last_steal;
2606 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2607 
2608 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2609 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2610 
2611 	smp_wmb();
2612 
2613 	vcpu->arch.st.steal.version += 1;
2614 
2615 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2616 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2617 }
2618 
2619 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2620 {
2621 	bool pr = false;
2622 	u32 msr = msr_info->index;
2623 	u64 data = msr_info->data;
2624 
2625 	switch (msr) {
2626 	case MSR_AMD64_NB_CFG:
2627 	case MSR_IA32_UCODE_WRITE:
2628 	case MSR_VM_HSAVE_PA:
2629 	case MSR_AMD64_PATCH_LOADER:
2630 	case MSR_AMD64_BU_CFG2:
2631 	case MSR_AMD64_DC_CFG:
2632 	case MSR_F15H_EX_CFG:
2633 		break;
2634 
2635 	case MSR_IA32_UCODE_REV:
2636 		if (msr_info->host_initiated)
2637 			vcpu->arch.microcode_version = data;
2638 		break;
2639 	case MSR_IA32_ARCH_CAPABILITIES:
2640 		if (!msr_info->host_initiated)
2641 			return 1;
2642 		vcpu->arch.arch_capabilities = data;
2643 		break;
2644 	case MSR_EFER:
2645 		return set_efer(vcpu, msr_info);
2646 	case MSR_K7_HWCR:
2647 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2648 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2649 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2650 
2651 		/* Handle McStatusWrEn */
2652 		if (data == BIT_ULL(18)) {
2653 			vcpu->arch.msr_hwcr = data;
2654 		} else if (data != 0) {
2655 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2656 				    data);
2657 			return 1;
2658 		}
2659 		break;
2660 	case MSR_FAM10H_MMIO_CONF_BASE:
2661 		if (data != 0) {
2662 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2663 				    "0x%llx\n", data);
2664 			return 1;
2665 		}
2666 		break;
2667 	case MSR_IA32_DEBUGCTLMSR:
2668 		if (!data) {
2669 			/* We support the non-activated case already */
2670 			break;
2671 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2672 			/* Values other than LBR and BTF are vendor-specific,
2673 			   thus reserved and should throw a #GP */
2674 			return 1;
2675 		}
2676 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2677 			    __func__, data);
2678 		break;
2679 	case 0x200 ... 0x2ff:
2680 		return kvm_mtrr_set_msr(vcpu, msr, data);
2681 	case MSR_IA32_APICBASE:
2682 		return kvm_set_apic_base(vcpu, msr_info);
2683 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2684 		return kvm_x2apic_msr_write(vcpu, msr, data);
2685 	case MSR_IA32_TSCDEADLINE:
2686 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2687 		break;
2688 	case MSR_IA32_TSC_ADJUST:
2689 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2690 			if (!msr_info->host_initiated) {
2691 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2692 				adjust_tsc_offset_guest(vcpu, adj);
2693 			}
2694 			vcpu->arch.ia32_tsc_adjust_msr = data;
2695 		}
2696 		break;
2697 	case MSR_IA32_MISC_ENABLE:
2698 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2699 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2700 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2701 				return 1;
2702 			vcpu->arch.ia32_misc_enable_msr = data;
2703 			kvm_update_cpuid(vcpu);
2704 		} else {
2705 			vcpu->arch.ia32_misc_enable_msr = data;
2706 		}
2707 		break;
2708 	case MSR_IA32_SMBASE:
2709 		if (!msr_info->host_initiated)
2710 			return 1;
2711 		vcpu->arch.smbase = data;
2712 		break;
2713 	case MSR_IA32_POWER_CTL:
2714 		vcpu->arch.msr_ia32_power_ctl = data;
2715 		break;
2716 	case MSR_IA32_TSC:
2717 		kvm_write_tsc(vcpu, msr_info);
2718 		break;
2719 	case MSR_SMI_COUNT:
2720 		if (!msr_info->host_initiated)
2721 			return 1;
2722 		vcpu->arch.smi_count = data;
2723 		break;
2724 	case MSR_KVM_WALL_CLOCK_NEW:
2725 	case MSR_KVM_WALL_CLOCK:
2726 		vcpu->kvm->arch.wall_clock = data;
2727 		kvm_write_wall_clock(vcpu->kvm, data);
2728 		break;
2729 	case MSR_KVM_SYSTEM_TIME_NEW:
2730 	case MSR_KVM_SYSTEM_TIME: {
2731 		struct kvm_arch *ka = &vcpu->kvm->arch;
2732 
2733 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2734 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2735 
2736 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2737 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2738 
2739 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2740 		}
2741 
2742 		vcpu->arch.time = data;
2743 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2744 
2745 		/* we verify if the enable bit is set... */
2746 		vcpu->arch.pv_time_enabled = false;
2747 		if (!(data & 1))
2748 			break;
2749 
2750 		if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2751 		     &vcpu->arch.pv_time, data & ~1ULL,
2752 		     sizeof(struct pvclock_vcpu_time_info)))
2753 			vcpu->arch.pv_time_enabled = true;
2754 
2755 		break;
2756 	}
2757 	case MSR_KVM_ASYNC_PF_EN:
2758 		if (kvm_pv_enable_async_pf(vcpu, data))
2759 			return 1;
2760 		break;
2761 	case MSR_KVM_STEAL_TIME:
2762 
2763 		if (unlikely(!sched_info_on()))
2764 			return 1;
2765 
2766 		if (data & KVM_STEAL_RESERVED_MASK)
2767 			return 1;
2768 
2769 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2770 						data & KVM_STEAL_VALID_BITS,
2771 						sizeof(struct kvm_steal_time)))
2772 			return 1;
2773 
2774 		vcpu->arch.st.msr_val = data;
2775 
2776 		if (!(data & KVM_MSR_ENABLED))
2777 			break;
2778 
2779 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2780 
2781 		break;
2782 	case MSR_KVM_PV_EOI_EN:
2783 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2784 			return 1;
2785 		break;
2786 
2787 	case MSR_KVM_POLL_CONTROL:
2788 		/* only enable bit supported */
2789 		if (data & (-1ULL << 1))
2790 			return 1;
2791 
2792 		vcpu->arch.msr_kvm_poll_control = data;
2793 		break;
2794 
2795 	case MSR_IA32_MCG_CTL:
2796 	case MSR_IA32_MCG_STATUS:
2797 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2798 		return set_msr_mce(vcpu, msr_info);
2799 
2800 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2801 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2802 		pr = true; /* fall through */
2803 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2804 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2805 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2806 			return kvm_pmu_set_msr(vcpu, msr_info);
2807 
2808 		if (pr || data != 0)
2809 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2810 				    "0x%x data 0x%llx\n", msr, data);
2811 		break;
2812 	case MSR_K7_CLK_CTL:
2813 		/*
2814 		 * Ignore all writes to this no longer documented MSR.
2815 		 * Writes are only relevant for old K7 processors,
2816 		 * all pre-dating SVM, but a recommended workaround from
2817 		 * AMD for these chips. It is possible to specify the
2818 		 * affected processor models on the command line, hence
2819 		 * the need to ignore the workaround.
2820 		 */
2821 		break;
2822 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2823 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2824 	case HV_X64_MSR_CRASH_CTL:
2825 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2826 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2827 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2828 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2829 		return kvm_hv_set_msr_common(vcpu, msr, data,
2830 					     msr_info->host_initiated);
2831 	case MSR_IA32_BBL_CR_CTL3:
2832 		/* Drop writes to this legacy MSR -- see rdmsr
2833 		 * counterpart for further detail.
2834 		 */
2835 		if (report_ignored_msrs)
2836 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2837 				msr, data);
2838 		break;
2839 	case MSR_AMD64_OSVW_ID_LENGTH:
2840 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2841 			return 1;
2842 		vcpu->arch.osvw.length = data;
2843 		break;
2844 	case MSR_AMD64_OSVW_STATUS:
2845 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2846 			return 1;
2847 		vcpu->arch.osvw.status = data;
2848 		break;
2849 	case MSR_PLATFORM_INFO:
2850 		if (!msr_info->host_initiated ||
2851 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2852 		     cpuid_fault_enabled(vcpu)))
2853 			return 1;
2854 		vcpu->arch.msr_platform_info = data;
2855 		break;
2856 	case MSR_MISC_FEATURES_ENABLES:
2857 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2858 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2859 		     !supports_cpuid_fault(vcpu)))
2860 			return 1;
2861 		vcpu->arch.msr_misc_features_enables = data;
2862 		break;
2863 	default:
2864 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2865 			return xen_hvm_config(vcpu, data);
2866 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2867 			return kvm_pmu_set_msr(vcpu, msr_info);
2868 		if (!ignore_msrs) {
2869 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2870 				    msr, data);
2871 			return 1;
2872 		} else {
2873 			if (report_ignored_msrs)
2874 				vcpu_unimpl(vcpu,
2875 					"ignored wrmsr: 0x%x data 0x%llx\n",
2876 					msr, data);
2877 			break;
2878 		}
2879 	}
2880 	return 0;
2881 }
2882 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2883 
2884 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2885 {
2886 	u64 data;
2887 	u64 mcg_cap = vcpu->arch.mcg_cap;
2888 	unsigned bank_num = mcg_cap & 0xff;
2889 
2890 	switch (msr) {
2891 	case MSR_IA32_P5_MC_ADDR:
2892 	case MSR_IA32_P5_MC_TYPE:
2893 		data = 0;
2894 		break;
2895 	case MSR_IA32_MCG_CAP:
2896 		data = vcpu->arch.mcg_cap;
2897 		break;
2898 	case MSR_IA32_MCG_CTL:
2899 		if (!(mcg_cap & MCG_CTL_P) && !host)
2900 			return 1;
2901 		data = vcpu->arch.mcg_ctl;
2902 		break;
2903 	case MSR_IA32_MCG_STATUS:
2904 		data = vcpu->arch.mcg_status;
2905 		break;
2906 	default:
2907 		if (msr >= MSR_IA32_MC0_CTL &&
2908 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2909 			u32 offset = msr - MSR_IA32_MC0_CTL;
2910 			data = vcpu->arch.mce_banks[offset];
2911 			break;
2912 		}
2913 		return 1;
2914 	}
2915 	*pdata = data;
2916 	return 0;
2917 }
2918 
2919 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2920 {
2921 	switch (msr_info->index) {
2922 	case MSR_IA32_PLATFORM_ID:
2923 	case MSR_IA32_EBL_CR_POWERON:
2924 	case MSR_IA32_DEBUGCTLMSR:
2925 	case MSR_IA32_LASTBRANCHFROMIP:
2926 	case MSR_IA32_LASTBRANCHTOIP:
2927 	case MSR_IA32_LASTINTFROMIP:
2928 	case MSR_IA32_LASTINTTOIP:
2929 	case MSR_K8_SYSCFG:
2930 	case MSR_K8_TSEG_ADDR:
2931 	case MSR_K8_TSEG_MASK:
2932 	case MSR_VM_HSAVE_PA:
2933 	case MSR_K8_INT_PENDING_MSG:
2934 	case MSR_AMD64_NB_CFG:
2935 	case MSR_FAM10H_MMIO_CONF_BASE:
2936 	case MSR_AMD64_BU_CFG2:
2937 	case MSR_IA32_PERF_CTL:
2938 	case MSR_AMD64_DC_CFG:
2939 	case MSR_F15H_EX_CFG:
2940 		msr_info->data = 0;
2941 		break;
2942 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2943 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2944 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2945 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2946 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2947 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2948 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2949 		msr_info->data = 0;
2950 		break;
2951 	case MSR_IA32_UCODE_REV:
2952 		msr_info->data = vcpu->arch.microcode_version;
2953 		break;
2954 	case MSR_IA32_ARCH_CAPABILITIES:
2955 		if (!msr_info->host_initiated &&
2956 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2957 			return 1;
2958 		msr_info->data = vcpu->arch.arch_capabilities;
2959 		break;
2960 	case MSR_IA32_POWER_CTL:
2961 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2962 		break;
2963 	case MSR_IA32_TSC:
2964 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2965 		break;
2966 	case MSR_MTRRcap:
2967 	case 0x200 ... 0x2ff:
2968 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2969 	case 0xcd: /* fsb frequency */
2970 		msr_info->data = 3;
2971 		break;
2972 		/*
2973 		 * MSR_EBC_FREQUENCY_ID
2974 		 * Conservative value valid for even the basic CPU models.
2975 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2976 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2977 		 * and 266MHz for model 3, or 4. Set Core Clock
2978 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2979 		 * 31:24) even though these are only valid for CPU
2980 		 * models > 2, however guests may end up dividing or
2981 		 * multiplying by zero otherwise.
2982 		 */
2983 	case MSR_EBC_FREQUENCY_ID:
2984 		msr_info->data = 1 << 24;
2985 		break;
2986 	case MSR_IA32_APICBASE:
2987 		msr_info->data = kvm_get_apic_base(vcpu);
2988 		break;
2989 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2990 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2991 		break;
2992 	case MSR_IA32_TSCDEADLINE:
2993 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2994 		break;
2995 	case MSR_IA32_TSC_ADJUST:
2996 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2997 		break;
2998 	case MSR_IA32_MISC_ENABLE:
2999 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3000 		break;
3001 	case MSR_IA32_SMBASE:
3002 		if (!msr_info->host_initiated)
3003 			return 1;
3004 		msr_info->data = vcpu->arch.smbase;
3005 		break;
3006 	case MSR_SMI_COUNT:
3007 		msr_info->data = vcpu->arch.smi_count;
3008 		break;
3009 	case MSR_IA32_PERF_STATUS:
3010 		/* TSC increment by tick */
3011 		msr_info->data = 1000ULL;
3012 		/* CPU multiplier */
3013 		msr_info->data |= (((uint64_t)4ULL) << 40);
3014 		break;
3015 	case MSR_EFER:
3016 		msr_info->data = vcpu->arch.efer;
3017 		break;
3018 	case MSR_KVM_WALL_CLOCK:
3019 	case MSR_KVM_WALL_CLOCK_NEW:
3020 		msr_info->data = vcpu->kvm->arch.wall_clock;
3021 		break;
3022 	case MSR_KVM_SYSTEM_TIME:
3023 	case MSR_KVM_SYSTEM_TIME_NEW:
3024 		msr_info->data = vcpu->arch.time;
3025 		break;
3026 	case MSR_KVM_ASYNC_PF_EN:
3027 		msr_info->data = vcpu->arch.apf.msr_val;
3028 		break;
3029 	case MSR_KVM_STEAL_TIME:
3030 		msr_info->data = vcpu->arch.st.msr_val;
3031 		break;
3032 	case MSR_KVM_PV_EOI_EN:
3033 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3034 		break;
3035 	case MSR_KVM_POLL_CONTROL:
3036 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3037 		break;
3038 	case MSR_IA32_P5_MC_ADDR:
3039 	case MSR_IA32_P5_MC_TYPE:
3040 	case MSR_IA32_MCG_CAP:
3041 	case MSR_IA32_MCG_CTL:
3042 	case MSR_IA32_MCG_STATUS:
3043 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3044 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3045 				   msr_info->host_initiated);
3046 	case MSR_K7_CLK_CTL:
3047 		/*
3048 		 * Provide expected ramp-up count for K7. All other
3049 		 * are set to zero, indicating minimum divisors for
3050 		 * every field.
3051 		 *
3052 		 * This prevents guest kernels on AMD host with CPU
3053 		 * type 6, model 8 and higher from exploding due to
3054 		 * the rdmsr failing.
3055 		 */
3056 		msr_info->data = 0x20000000;
3057 		break;
3058 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3059 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3060 	case HV_X64_MSR_CRASH_CTL:
3061 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3062 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3063 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3064 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3065 		return kvm_hv_get_msr_common(vcpu,
3066 					     msr_info->index, &msr_info->data,
3067 					     msr_info->host_initiated);
3068 		break;
3069 	case MSR_IA32_BBL_CR_CTL3:
3070 		/* This legacy MSR exists but isn't fully documented in current
3071 		 * silicon.  It is however accessed by winxp in very narrow
3072 		 * scenarios where it sets bit #19, itself documented as
3073 		 * a "reserved" bit.  Best effort attempt to source coherent
3074 		 * read data here should the balance of the register be
3075 		 * interpreted by the guest:
3076 		 *
3077 		 * L2 cache control register 3: 64GB range, 256KB size,
3078 		 * enabled, latency 0x1, configured
3079 		 */
3080 		msr_info->data = 0xbe702111;
3081 		break;
3082 	case MSR_AMD64_OSVW_ID_LENGTH:
3083 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3084 			return 1;
3085 		msr_info->data = vcpu->arch.osvw.length;
3086 		break;
3087 	case MSR_AMD64_OSVW_STATUS:
3088 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3089 			return 1;
3090 		msr_info->data = vcpu->arch.osvw.status;
3091 		break;
3092 	case MSR_PLATFORM_INFO:
3093 		if (!msr_info->host_initiated &&
3094 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3095 			return 1;
3096 		msr_info->data = vcpu->arch.msr_platform_info;
3097 		break;
3098 	case MSR_MISC_FEATURES_ENABLES:
3099 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3100 		break;
3101 	case MSR_K7_HWCR:
3102 		msr_info->data = vcpu->arch.msr_hwcr;
3103 		break;
3104 	default:
3105 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3106 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3107 		if (!ignore_msrs) {
3108 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3109 					       msr_info->index);
3110 			return 1;
3111 		} else {
3112 			if (report_ignored_msrs)
3113 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3114 					msr_info->index);
3115 			msr_info->data = 0;
3116 		}
3117 		break;
3118 	}
3119 	return 0;
3120 }
3121 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3122 
3123 /*
3124  * Read or write a bunch of msrs. All parameters are kernel addresses.
3125  *
3126  * @return number of msrs set successfully.
3127  */
3128 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3129 		    struct kvm_msr_entry *entries,
3130 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3131 				  unsigned index, u64 *data))
3132 {
3133 	int i;
3134 
3135 	for (i = 0; i < msrs->nmsrs; ++i)
3136 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3137 			break;
3138 
3139 	return i;
3140 }
3141 
3142 /*
3143  * Read or write a bunch of msrs. Parameters are user addresses.
3144  *
3145  * @return number of msrs set successfully.
3146  */
3147 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3148 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3149 				unsigned index, u64 *data),
3150 		  int writeback)
3151 {
3152 	struct kvm_msrs msrs;
3153 	struct kvm_msr_entry *entries;
3154 	int r, n;
3155 	unsigned size;
3156 
3157 	r = -EFAULT;
3158 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3159 		goto out;
3160 
3161 	r = -E2BIG;
3162 	if (msrs.nmsrs >= MAX_IO_MSRS)
3163 		goto out;
3164 
3165 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3166 	entries = memdup_user(user_msrs->entries, size);
3167 	if (IS_ERR(entries)) {
3168 		r = PTR_ERR(entries);
3169 		goto out;
3170 	}
3171 
3172 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3173 	if (r < 0)
3174 		goto out_free;
3175 
3176 	r = -EFAULT;
3177 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3178 		goto out_free;
3179 
3180 	r = n;
3181 
3182 out_free:
3183 	kfree(entries);
3184 out:
3185 	return r;
3186 }
3187 
3188 static inline bool kvm_can_mwait_in_guest(void)
3189 {
3190 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3191 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3192 		boot_cpu_has(X86_FEATURE_ARAT);
3193 }
3194 
3195 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3196 {
3197 	int r = 0;
3198 
3199 	switch (ext) {
3200 	case KVM_CAP_IRQCHIP:
3201 	case KVM_CAP_HLT:
3202 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3203 	case KVM_CAP_SET_TSS_ADDR:
3204 	case KVM_CAP_EXT_CPUID:
3205 	case KVM_CAP_EXT_EMUL_CPUID:
3206 	case KVM_CAP_CLOCKSOURCE:
3207 	case KVM_CAP_PIT:
3208 	case KVM_CAP_NOP_IO_DELAY:
3209 	case KVM_CAP_MP_STATE:
3210 	case KVM_CAP_SYNC_MMU:
3211 	case KVM_CAP_USER_NMI:
3212 	case KVM_CAP_REINJECT_CONTROL:
3213 	case KVM_CAP_IRQ_INJECT_STATUS:
3214 	case KVM_CAP_IOEVENTFD:
3215 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3216 	case KVM_CAP_PIT2:
3217 	case KVM_CAP_PIT_STATE2:
3218 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3219 	case KVM_CAP_XEN_HVM:
3220 	case KVM_CAP_VCPU_EVENTS:
3221 	case KVM_CAP_HYPERV:
3222 	case KVM_CAP_HYPERV_VAPIC:
3223 	case KVM_CAP_HYPERV_SPIN:
3224 	case KVM_CAP_HYPERV_SYNIC:
3225 	case KVM_CAP_HYPERV_SYNIC2:
3226 	case KVM_CAP_HYPERV_VP_INDEX:
3227 	case KVM_CAP_HYPERV_EVENTFD:
3228 	case KVM_CAP_HYPERV_TLBFLUSH:
3229 	case KVM_CAP_HYPERV_SEND_IPI:
3230 	case KVM_CAP_HYPERV_CPUID:
3231 	case KVM_CAP_PCI_SEGMENT:
3232 	case KVM_CAP_DEBUGREGS:
3233 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3234 	case KVM_CAP_XSAVE:
3235 	case KVM_CAP_ASYNC_PF:
3236 	case KVM_CAP_GET_TSC_KHZ:
3237 	case KVM_CAP_KVMCLOCK_CTRL:
3238 	case KVM_CAP_READONLY_MEM:
3239 	case KVM_CAP_HYPERV_TIME:
3240 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3241 	case KVM_CAP_TSC_DEADLINE_TIMER:
3242 	case KVM_CAP_DISABLE_QUIRKS:
3243 	case KVM_CAP_SET_BOOT_CPU_ID:
3244  	case KVM_CAP_SPLIT_IRQCHIP:
3245 	case KVM_CAP_IMMEDIATE_EXIT:
3246 	case KVM_CAP_PMU_EVENT_FILTER:
3247 	case KVM_CAP_GET_MSR_FEATURES:
3248 	case KVM_CAP_MSR_PLATFORM_INFO:
3249 	case KVM_CAP_EXCEPTION_PAYLOAD:
3250 		r = 1;
3251 		break;
3252 	case KVM_CAP_SYNC_REGS:
3253 		r = KVM_SYNC_X86_VALID_FIELDS;
3254 		break;
3255 	case KVM_CAP_ADJUST_CLOCK:
3256 		r = KVM_CLOCK_TSC_STABLE;
3257 		break;
3258 	case KVM_CAP_X86_DISABLE_EXITS:
3259 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3260 		      KVM_X86_DISABLE_EXITS_CSTATE;
3261 		if(kvm_can_mwait_in_guest())
3262 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3263 		break;
3264 	case KVM_CAP_X86_SMM:
3265 		/* SMBASE is usually relocated above 1M on modern chipsets,
3266 		 * and SMM handlers might indeed rely on 4G segment limits,
3267 		 * so do not report SMM to be available if real mode is
3268 		 * emulated via vm86 mode.  Still, do not go to great lengths
3269 		 * to avoid userspace's usage of the feature, because it is a
3270 		 * fringe case that is not enabled except via specific settings
3271 		 * of the module parameters.
3272 		 */
3273 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3274 		break;
3275 	case KVM_CAP_VAPIC:
3276 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3277 		break;
3278 	case KVM_CAP_NR_VCPUS:
3279 		r = KVM_SOFT_MAX_VCPUS;
3280 		break;
3281 	case KVM_CAP_MAX_VCPUS:
3282 		r = KVM_MAX_VCPUS;
3283 		break;
3284 	case KVM_CAP_MAX_VCPU_ID:
3285 		r = KVM_MAX_VCPU_ID;
3286 		break;
3287 	case KVM_CAP_PV_MMU:	/* obsolete */
3288 		r = 0;
3289 		break;
3290 	case KVM_CAP_MCE:
3291 		r = KVM_MAX_MCE_BANKS;
3292 		break;
3293 	case KVM_CAP_XCRS:
3294 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3295 		break;
3296 	case KVM_CAP_TSC_CONTROL:
3297 		r = kvm_has_tsc_control;
3298 		break;
3299 	case KVM_CAP_X2APIC_API:
3300 		r = KVM_X2APIC_API_VALID_FLAGS;
3301 		break;
3302 	case KVM_CAP_NESTED_STATE:
3303 		r = kvm_x86_ops->get_nested_state ?
3304 			kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3305 		break;
3306 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3307 		r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3308 		break;
3309 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3310 		r = kvm_x86_ops->nested_enable_evmcs != NULL;
3311 		break;
3312 	default:
3313 		break;
3314 	}
3315 	return r;
3316 
3317 }
3318 
3319 long kvm_arch_dev_ioctl(struct file *filp,
3320 			unsigned int ioctl, unsigned long arg)
3321 {
3322 	void __user *argp = (void __user *)arg;
3323 	long r;
3324 
3325 	switch (ioctl) {
3326 	case KVM_GET_MSR_INDEX_LIST: {
3327 		struct kvm_msr_list __user *user_msr_list = argp;
3328 		struct kvm_msr_list msr_list;
3329 		unsigned n;
3330 
3331 		r = -EFAULT;
3332 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3333 			goto out;
3334 		n = msr_list.nmsrs;
3335 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3336 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3337 			goto out;
3338 		r = -E2BIG;
3339 		if (n < msr_list.nmsrs)
3340 			goto out;
3341 		r = -EFAULT;
3342 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3343 				 num_msrs_to_save * sizeof(u32)))
3344 			goto out;
3345 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3346 				 &emulated_msrs,
3347 				 num_emulated_msrs * sizeof(u32)))
3348 			goto out;
3349 		r = 0;
3350 		break;
3351 	}
3352 	case KVM_GET_SUPPORTED_CPUID:
3353 	case KVM_GET_EMULATED_CPUID: {
3354 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3355 		struct kvm_cpuid2 cpuid;
3356 
3357 		r = -EFAULT;
3358 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3359 			goto out;
3360 
3361 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3362 					    ioctl);
3363 		if (r)
3364 			goto out;
3365 
3366 		r = -EFAULT;
3367 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3368 			goto out;
3369 		r = 0;
3370 		break;
3371 	}
3372 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3373 		r = -EFAULT;
3374 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3375 				 sizeof(kvm_mce_cap_supported)))
3376 			goto out;
3377 		r = 0;
3378 		break;
3379 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3380 		struct kvm_msr_list __user *user_msr_list = argp;
3381 		struct kvm_msr_list msr_list;
3382 		unsigned int n;
3383 
3384 		r = -EFAULT;
3385 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3386 			goto out;
3387 		n = msr_list.nmsrs;
3388 		msr_list.nmsrs = num_msr_based_features;
3389 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3390 			goto out;
3391 		r = -E2BIG;
3392 		if (n < msr_list.nmsrs)
3393 			goto out;
3394 		r = -EFAULT;
3395 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3396 				 num_msr_based_features * sizeof(u32)))
3397 			goto out;
3398 		r = 0;
3399 		break;
3400 	}
3401 	case KVM_GET_MSRS:
3402 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3403 		break;
3404 	}
3405 	default:
3406 		r = -EINVAL;
3407 	}
3408 out:
3409 	return r;
3410 }
3411 
3412 static void wbinvd_ipi(void *garbage)
3413 {
3414 	wbinvd();
3415 }
3416 
3417 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3418 {
3419 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3420 }
3421 
3422 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3423 {
3424 	/* Address WBINVD may be executed by guest */
3425 	if (need_emulate_wbinvd(vcpu)) {
3426 		if (kvm_x86_ops->has_wbinvd_exit())
3427 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3428 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3429 			smp_call_function_single(vcpu->cpu,
3430 					wbinvd_ipi, NULL, 1);
3431 	}
3432 
3433 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3434 
3435 	fpregs_assert_state_consistent();
3436 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
3437 		switch_fpu_return();
3438 
3439 	/* Apply any externally detected TSC adjustments (due to suspend) */
3440 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3441 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3442 		vcpu->arch.tsc_offset_adjustment = 0;
3443 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3444 	}
3445 
3446 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3447 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3448 				rdtsc() - vcpu->arch.last_host_tsc;
3449 		if (tsc_delta < 0)
3450 			mark_tsc_unstable("KVM discovered backwards TSC");
3451 
3452 		if (kvm_check_tsc_unstable()) {
3453 			u64 offset = kvm_compute_tsc_offset(vcpu,
3454 						vcpu->arch.last_guest_tsc);
3455 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3456 			vcpu->arch.tsc_catchup = 1;
3457 		}
3458 
3459 		if (kvm_lapic_hv_timer_in_use(vcpu))
3460 			kvm_lapic_restart_hv_timer(vcpu);
3461 
3462 		/*
3463 		 * On a host with synchronized TSC, there is no need to update
3464 		 * kvmclock on vcpu->cpu migration
3465 		 */
3466 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3467 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3468 		if (vcpu->cpu != cpu)
3469 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3470 		vcpu->cpu = cpu;
3471 	}
3472 
3473 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3474 }
3475 
3476 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3477 {
3478 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3479 		return;
3480 
3481 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3482 
3483 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3484 			&vcpu->arch.st.steal.preempted,
3485 			offsetof(struct kvm_steal_time, preempted),
3486 			sizeof(vcpu->arch.st.steal.preempted));
3487 }
3488 
3489 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3490 {
3491 	int idx;
3492 
3493 	if (vcpu->preempted)
3494 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3495 
3496 	/*
3497 	 * Disable page faults because we're in atomic context here.
3498 	 * kvm_write_guest_offset_cached() would call might_fault()
3499 	 * that relies on pagefault_disable() to tell if there's a
3500 	 * bug. NOTE: the write to guest memory may not go through if
3501 	 * during postcopy live migration or if there's heavy guest
3502 	 * paging.
3503 	 */
3504 	pagefault_disable();
3505 	/*
3506 	 * kvm_memslots() will be called by
3507 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3508 	 */
3509 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3510 	kvm_steal_time_set_preempted(vcpu);
3511 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3512 	pagefault_enable();
3513 	kvm_x86_ops->vcpu_put(vcpu);
3514 	vcpu->arch.last_host_tsc = rdtsc();
3515 	/*
3516 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3517 	 * on every vmexit, but if not, we might have a stale dr6 from the
3518 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3519 	 */
3520 	set_debugreg(0, 6);
3521 }
3522 
3523 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3524 				    struct kvm_lapic_state *s)
3525 {
3526 	if (vcpu->arch.apicv_active)
3527 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3528 
3529 	return kvm_apic_get_state(vcpu, s);
3530 }
3531 
3532 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3533 				    struct kvm_lapic_state *s)
3534 {
3535 	int r;
3536 
3537 	r = kvm_apic_set_state(vcpu, s);
3538 	if (r)
3539 		return r;
3540 	update_cr8_intercept(vcpu);
3541 
3542 	return 0;
3543 }
3544 
3545 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3546 {
3547 	return (!lapic_in_kernel(vcpu) ||
3548 		kvm_apic_accept_pic_intr(vcpu));
3549 }
3550 
3551 /*
3552  * if userspace requested an interrupt window, check that the
3553  * interrupt window is open.
3554  *
3555  * No need to exit to userspace if we already have an interrupt queued.
3556  */
3557 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3558 {
3559 	return kvm_arch_interrupt_allowed(vcpu) &&
3560 		!kvm_cpu_has_interrupt(vcpu) &&
3561 		!kvm_event_needs_reinjection(vcpu) &&
3562 		kvm_cpu_accept_dm_intr(vcpu);
3563 }
3564 
3565 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3566 				    struct kvm_interrupt *irq)
3567 {
3568 	if (irq->irq >= KVM_NR_INTERRUPTS)
3569 		return -EINVAL;
3570 
3571 	if (!irqchip_in_kernel(vcpu->kvm)) {
3572 		kvm_queue_interrupt(vcpu, irq->irq, false);
3573 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3574 		return 0;
3575 	}
3576 
3577 	/*
3578 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3579 	 * fail for in-kernel 8259.
3580 	 */
3581 	if (pic_in_kernel(vcpu->kvm))
3582 		return -ENXIO;
3583 
3584 	if (vcpu->arch.pending_external_vector != -1)
3585 		return -EEXIST;
3586 
3587 	vcpu->arch.pending_external_vector = irq->irq;
3588 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3589 	return 0;
3590 }
3591 
3592 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3593 {
3594 	kvm_inject_nmi(vcpu);
3595 
3596 	return 0;
3597 }
3598 
3599 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3600 {
3601 	kvm_make_request(KVM_REQ_SMI, vcpu);
3602 
3603 	return 0;
3604 }
3605 
3606 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3607 					   struct kvm_tpr_access_ctl *tac)
3608 {
3609 	if (tac->flags)
3610 		return -EINVAL;
3611 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3612 	return 0;
3613 }
3614 
3615 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3616 					u64 mcg_cap)
3617 {
3618 	int r;
3619 	unsigned bank_num = mcg_cap & 0xff, bank;
3620 
3621 	r = -EINVAL;
3622 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3623 		goto out;
3624 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3625 		goto out;
3626 	r = 0;
3627 	vcpu->arch.mcg_cap = mcg_cap;
3628 	/* Init IA32_MCG_CTL to all 1s */
3629 	if (mcg_cap & MCG_CTL_P)
3630 		vcpu->arch.mcg_ctl = ~(u64)0;
3631 	/* Init IA32_MCi_CTL to all 1s */
3632 	for (bank = 0; bank < bank_num; bank++)
3633 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3634 
3635 	kvm_x86_ops->setup_mce(vcpu);
3636 out:
3637 	return r;
3638 }
3639 
3640 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3641 				      struct kvm_x86_mce *mce)
3642 {
3643 	u64 mcg_cap = vcpu->arch.mcg_cap;
3644 	unsigned bank_num = mcg_cap & 0xff;
3645 	u64 *banks = vcpu->arch.mce_banks;
3646 
3647 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3648 		return -EINVAL;
3649 	/*
3650 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3651 	 * reporting is disabled
3652 	 */
3653 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3654 	    vcpu->arch.mcg_ctl != ~(u64)0)
3655 		return 0;
3656 	banks += 4 * mce->bank;
3657 	/*
3658 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3659 	 * reporting is disabled for the bank
3660 	 */
3661 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3662 		return 0;
3663 	if (mce->status & MCI_STATUS_UC) {
3664 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3665 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3666 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3667 			return 0;
3668 		}
3669 		if (banks[1] & MCI_STATUS_VAL)
3670 			mce->status |= MCI_STATUS_OVER;
3671 		banks[2] = mce->addr;
3672 		banks[3] = mce->misc;
3673 		vcpu->arch.mcg_status = mce->mcg_status;
3674 		banks[1] = mce->status;
3675 		kvm_queue_exception(vcpu, MC_VECTOR);
3676 	} else if (!(banks[1] & MCI_STATUS_VAL)
3677 		   || !(banks[1] & MCI_STATUS_UC)) {
3678 		if (banks[1] & MCI_STATUS_VAL)
3679 			mce->status |= MCI_STATUS_OVER;
3680 		banks[2] = mce->addr;
3681 		banks[3] = mce->misc;
3682 		banks[1] = mce->status;
3683 	} else
3684 		banks[1] |= MCI_STATUS_OVER;
3685 	return 0;
3686 }
3687 
3688 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3689 					       struct kvm_vcpu_events *events)
3690 {
3691 	process_nmi(vcpu);
3692 
3693 	/*
3694 	 * The API doesn't provide the instruction length for software
3695 	 * exceptions, so don't report them. As long as the guest RIP
3696 	 * isn't advanced, we should expect to encounter the exception
3697 	 * again.
3698 	 */
3699 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3700 		events->exception.injected = 0;
3701 		events->exception.pending = 0;
3702 	} else {
3703 		events->exception.injected = vcpu->arch.exception.injected;
3704 		events->exception.pending = vcpu->arch.exception.pending;
3705 		/*
3706 		 * For ABI compatibility, deliberately conflate
3707 		 * pending and injected exceptions when
3708 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3709 		 */
3710 		if (!vcpu->kvm->arch.exception_payload_enabled)
3711 			events->exception.injected |=
3712 				vcpu->arch.exception.pending;
3713 	}
3714 	events->exception.nr = vcpu->arch.exception.nr;
3715 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3716 	events->exception.error_code = vcpu->arch.exception.error_code;
3717 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3718 	events->exception_payload = vcpu->arch.exception.payload;
3719 
3720 	events->interrupt.injected =
3721 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3722 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3723 	events->interrupt.soft = 0;
3724 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3725 
3726 	events->nmi.injected = vcpu->arch.nmi_injected;
3727 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3728 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3729 	events->nmi.pad = 0;
3730 
3731 	events->sipi_vector = 0; /* never valid when reporting to user space */
3732 
3733 	events->smi.smm = is_smm(vcpu);
3734 	events->smi.pending = vcpu->arch.smi_pending;
3735 	events->smi.smm_inside_nmi =
3736 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3737 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3738 
3739 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3740 			 | KVM_VCPUEVENT_VALID_SHADOW
3741 			 | KVM_VCPUEVENT_VALID_SMM);
3742 	if (vcpu->kvm->arch.exception_payload_enabled)
3743 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3744 
3745 	memset(&events->reserved, 0, sizeof(events->reserved));
3746 }
3747 
3748 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3749 
3750 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3751 					      struct kvm_vcpu_events *events)
3752 {
3753 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3754 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3755 			      | KVM_VCPUEVENT_VALID_SHADOW
3756 			      | KVM_VCPUEVENT_VALID_SMM
3757 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3758 		return -EINVAL;
3759 
3760 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3761 		if (!vcpu->kvm->arch.exception_payload_enabled)
3762 			return -EINVAL;
3763 		if (events->exception.pending)
3764 			events->exception.injected = 0;
3765 		else
3766 			events->exception_has_payload = 0;
3767 	} else {
3768 		events->exception.pending = 0;
3769 		events->exception_has_payload = 0;
3770 	}
3771 
3772 	if ((events->exception.injected || events->exception.pending) &&
3773 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3774 		return -EINVAL;
3775 
3776 	/* INITs are latched while in SMM */
3777 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3778 	    (events->smi.smm || events->smi.pending) &&
3779 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3780 		return -EINVAL;
3781 
3782 	process_nmi(vcpu);
3783 	vcpu->arch.exception.injected = events->exception.injected;
3784 	vcpu->arch.exception.pending = events->exception.pending;
3785 	vcpu->arch.exception.nr = events->exception.nr;
3786 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3787 	vcpu->arch.exception.error_code = events->exception.error_code;
3788 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3789 	vcpu->arch.exception.payload = events->exception_payload;
3790 
3791 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3792 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3793 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3794 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3795 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3796 						  events->interrupt.shadow);
3797 
3798 	vcpu->arch.nmi_injected = events->nmi.injected;
3799 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3800 		vcpu->arch.nmi_pending = events->nmi.pending;
3801 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3802 
3803 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3804 	    lapic_in_kernel(vcpu))
3805 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3806 
3807 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3808 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3809 			if (events->smi.smm)
3810 				vcpu->arch.hflags |= HF_SMM_MASK;
3811 			else
3812 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3813 			kvm_smm_changed(vcpu);
3814 		}
3815 
3816 		vcpu->arch.smi_pending = events->smi.pending;
3817 
3818 		if (events->smi.smm) {
3819 			if (events->smi.smm_inside_nmi)
3820 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3821 			else
3822 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3823 			if (lapic_in_kernel(vcpu)) {
3824 				if (events->smi.latched_init)
3825 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3826 				else
3827 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3828 			}
3829 		}
3830 	}
3831 
3832 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3833 
3834 	return 0;
3835 }
3836 
3837 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3838 					     struct kvm_debugregs *dbgregs)
3839 {
3840 	unsigned long val;
3841 
3842 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3843 	kvm_get_dr(vcpu, 6, &val);
3844 	dbgregs->dr6 = val;
3845 	dbgregs->dr7 = vcpu->arch.dr7;
3846 	dbgregs->flags = 0;
3847 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3848 }
3849 
3850 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3851 					    struct kvm_debugregs *dbgregs)
3852 {
3853 	if (dbgregs->flags)
3854 		return -EINVAL;
3855 
3856 	if (dbgregs->dr6 & ~0xffffffffull)
3857 		return -EINVAL;
3858 	if (dbgregs->dr7 & ~0xffffffffull)
3859 		return -EINVAL;
3860 
3861 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3862 	kvm_update_dr0123(vcpu);
3863 	vcpu->arch.dr6 = dbgregs->dr6;
3864 	kvm_update_dr6(vcpu);
3865 	vcpu->arch.dr7 = dbgregs->dr7;
3866 	kvm_update_dr7(vcpu);
3867 
3868 	return 0;
3869 }
3870 
3871 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3872 
3873 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3874 {
3875 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3876 	u64 xstate_bv = xsave->header.xfeatures;
3877 	u64 valid;
3878 
3879 	/*
3880 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3881 	 * leaves 0 and 1 in the loop below.
3882 	 */
3883 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3884 
3885 	/* Set XSTATE_BV */
3886 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3887 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3888 
3889 	/*
3890 	 * Copy each region from the possibly compacted offset to the
3891 	 * non-compacted offset.
3892 	 */
3893 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3894 	while (valid) {
3895 		u64 xfeature_mask = valid & -valid;
3896 		int xfeature_nr = fls64(xfeature_mask) - 1;
3897 		void *src = get_xsave_addr(xsave, xfeature_nr);
3898 
3899 		if (src) {
3900 			u32 size, offset, ecx, edx;
3901 			cpuid_count(XSTATE_CPUID, xfeature_nr,
3902 				    &size, &offset, &ecx, &edx);
3903 			if (xfeature_nr == XFEATURE_PKRU)
3904 				memcpy(dest + offset, &vcpu->arch.pkru,
3905 				       sizeof(vcpu->arch.pkru));
3906 			else
3907 				memcpy(dest + offset, src, size);
3908 
3909 		}
3910 
3911 		valid -= xfeature_mask;
3912 	}
3913 }
3914 
3915 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3916 {
3917 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3918 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3919 	u64 valid;
3920 
3921 	/*
3922 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3923 	 * leaves 0 and 1 in the loop below.
3924 	 */
3925 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3926 
3927 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3928 	xsave->header.xfeatures = xstate_bv;
3929 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3930 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3931 
3932 	/*
3933 	 * Copy each region from the non-compacted offset to the
3934 	 * possibly compacted offset.
3935 	 */
3936 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3937 	while (valid) {
3938 		u64 xfeature_mask = valid & -valid;
3939 		int xfeature_nr = fls64(xfeature_mask) - 1;
3940 		void *dest = get_xsave_addr(xsave, xfeature_nr);
3941 
3942 		if (dest) {
3943 			u32 size, offset, ecx, edx;
3944 			cpuid_count(XSTATE_CPUID, xfeature_nr,
3945 				    &size, &offset, &ecx, &edx);
3946 			if (xfeature_nr == XFEATURE_PKRU)
3947 				memcpy(&vcpu->arch.pkru, src + offset,
3948 				       sizeof(vcpu->arch.pkru));
3949 			else
3950 				memcpy(dest, src + offset, size);
3951 		}
3952 
3953 		valid -= xfeature_mask;
3954 	}
3955 }
3956 
3957 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3958 					 struct kvm_xsave *guest_xsave)
3959 {
3960 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3961 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3962 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3963 	} else {
3964 		memcpy(guest_xsave->region,
3965 			&vcpu->arch.guest_fpu->state.fxsave,
3966 			sizeof(struct fxregs_state));
3967 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3968 			XFEATURE_MASK_FPSSE;
3969 	}
3970 }
3971 
3972 #define XSAVE_MXCSR_OFFSET 24
3973 
3974 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3975 					struct kvm_xsave *guest_xsave)
3976 {
3977 	u64 xstate_bv =
3978 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3979 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3980 
3981 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3982 		/*
3983 		 * Here we allow setting states that are not present in
3984 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3985 		 * with old userspace.
3986 		 */
3987 		if (xstate_bv & ~kvm_supported_xcr0() ||
3988 			mxcsr & ~mxcsr_feature_mask)
3989 			return -EINVAL;
3990 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3991 	} else {
3992 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3993 			mxcsr & ~mxcsr_feature_mask)
3994 			return -EINVAL;
3995 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3996 			guest_xsave->region, sizeof(struct fxregs_state));
3997 	}
3998 	return 0;
3999 }
4000 
4001 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4002 					struct kvm_xcrs *guest_xcrs)
4003 {
4004 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4005 		guest_xcrs->nr_xcrs = 0;
4006 		return;
4007 	}
4008 
4009 	guest_xcrs->nr_xcrs = 1;
4010 	guest_xcrs->flags = 0;
4011 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4012 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4013 }
4014 
4015 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4016 				       struct kvm_xcrs *guest_xcrs)
4017 {
4018 	int i, r = 0;
4019 
4020 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4021 		return -EINVAL;
4022 
4023 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4024 		return -EINVAL;
4025 
4026 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4027 		/* Only support XCR0 currently */
4028 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4029 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4030 				guest_xcrs->xcrs[i].value);
4031 			break;
4032 		}
4033 	if (r)
4034 		r = -EINVAL;
4035 	return r;
4036 }
4037 
4038 /*
4039  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4040  * stopped by the hypervisor.  This function will be called from the host only.
4041  * EINVAL is returned when the host attempts to set the flag for a guest that
4042  * does not support pv clocks.
4043  */
4044 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4045 {
4046 	if (!vcpu->arch.pv_time_enabled)
4047 		return -EINVAL;
4048 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4049 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4050 	return 0;
4051 }
4052 
4053 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4054 				     struct kvm_enable_cap *cap)
4055 {
4056 	int r;
4057 	uint16_t vmcs_version;
4058 	void __user *user_ptr;
4059 
4060 	if (cap->flags)
4061 		return -EINVAL;
4062 
4063 	switch (cap->cap) {
4064 	case KVM_CAP_HYPERV_SYNIC2:
4065 		if (cap->args[0])
4066 			return -EINVAL;
4067 		/* fall through */
4068 
4069 	case KVM_CAP_HYPERV_SYNIC:
4070 		if (!irqchip_in_kernel(vcpu->kvm))
4071 			return -EINVAL;
4072 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4073 					     KVM_CAP_HYPERV_SYNIC2);
4074 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4075 		if (!kvm_x86_ops->nested_enable_evmcs)
4076 			return -ENOTTY;
4077 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4078 		if (!r) {
4079 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4080 			if (copy_to_user(user_ptr, &vmcs_version,
4081 					 sizeof(vmcs_version)))
4082 				r = -EFAULT;
4083 		}
4084 		return r;
4085 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4086 		if (!kvm_x86_ops->enable_direct_tlbflush)
4087 			return -ENOTTY;
4088 
4089 		return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4090 
4091 	default:
4092 		return -EINVAL;
4093 	}
4094 }
4095 
4096 long kvm_arch_vcpu_ioctl(struct file *filp,
4097 			 unsigned int ioctl, unsigned long arg)
4098 {
4099 	struct kvm_vcpu *vcpu = filp->private_data;
4100 	void __user *argp = (void __user *)arg;
4101 	int r;
4102 	union {
4103 		struct kvm_lapic_state *lapic;
4104 		struct kvm_xsave *xsave;
4105 		struct kvm_xcrs *xcrs;
4106 		void *buffer;
4107 	} u;
4108 
4109 	vcpu_load(vcpu);
4110 
4111 	u.buffer = NULL;
4112 	switch (ioctl) {
4113 	case KVM_GET_LAPIC: {
4114 		r = -EINVAL;
4115 		if (!lapic_in_kernel(vcpu))
4116 			goto out;
4117 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4118 				GFP_KERNEL_ACCOUNT);
4119 
4120 		r = -ENOMEM;
4121 		if (!u.lapic)
4122 			goto out;
4123 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4124 		if (r)
4125 			goto out;
4126 		r = -EFAULT;
4127 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4128 			goto out;
4129 		r = 0;
4130 		break;
4131 	}
4132 	case KVM_SET_LAPIC: {
4133 		r = -EINVAL;
4134 		if (!lapic_in_kernel(vcpu))
4135 			goto out;
4136 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4137 		if (IS_ERR(u.lapic)) {
4138 			r = PTR_ERR(u.lapic);
4139 			goto out_nofree;
4140 		}
4141 
4142 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4143 		break;
4144 	}
4145 	case KVM_INTERRUPT: {
4146 		struct kvm_interrupt irq;
4147 
4148 		r = -EFAULT;
4149 		if (copy_from_user(&irq, argp, sizeof(irq)))
4150 			goto out;
4151 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4152 		break;
4153 	}
4154 	case KVM_NMI: {
4155 		r = kvm_vcpu_ioctl_nmi(vcpu);
4156 		break;
4157 	}
4158 	case KVM_SMI: {
4159 		r = kvm_vcpu_ioctl_smi(vcpu);
4160 		break;
4161 	}
4162 	case KVM_SET_CPUID: {
4163 		struct kvm_cpuid __user *cpuid_arg = argp;
4164 		struct kvm_cpuid cpuid;
4165 
4166 		r = -EFAULT;
4167 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4168 			goto out;
4169 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4170 		break;
4171 	}
4172 	case KVM_SET_CPUID2: {
4173 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4174 		struct kvm_cpuid2 cpuid;
4175 
4176 		r = -EFAULT;
4177 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4178 			goto out;
4179 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4180 					      cpuid_arg->entries);
4181 		break;
4182 	}
4183 	case KVM_GET_CPUID2: {
4184 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4185 		struct kvm_cpuid2 cpuid;
4186 
4187 		r = -EFAULT;
4188 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4189 			goto out;
4190 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4191 					      cpuid_arg->entries);
4192 		if (r)
4193 			goto out;
4194 		r = -EFAULT;
4195 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4196 			goto out;
4197 		r = 0;
4198 		break;
4199 	}
4200 	case KVM_GET_MSRS: {
4201 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4202 		r = msr_io(vcpu, argp, do_get_msr, 1);
4203 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4204 		break;
4205 	}
4206 	case KVM_SET_MSRS: {
4207 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4208 		r = msr_io(vcpu, argp, do_set_msr, 0);
4209 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4210 		break;
4211 	}
4212 	case KVM_TPR_ACCESS_REPORTING: {
4213 		struct kvm_tpr_access_ctl tac;
4214 
4215 		r = -EFAULT;
4216 		if (copy_from_user(&tac, argp, sizeof(tac)))
4217 			goto out;
4218 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4219 		if (r)
4220 			goto out;
4221 		r = -EFAULT;
4222 		if (copy_to_user(argp, &tac, sizeof(tac)))
4223 			goto out;
4224 		r = 0;
4225 		break;
4226 	};
4227 	case KVM_SET_VAPIC_ADDR: {
4228 		struct kvm_vapic_addr va;
4229 		int idx;
4230 
4231 		r = -EINVAL;
4232 		if (!lapic_in_kernel(vcpu))
4233 			goto out;
4234 		r = -EFAULT;
4235 		if (copy_from_user(&va, argp, sizeof(va)))
4236 			goto out;
4237 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4238 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4239 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4240 		break;
4241 	}
4242 	case KVM_X86_SETUP_MCE: {
4243 		u64 mcg_cap;
4244 
4245 		r = -EFAULT;
4246 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4247 			goto out;
4248 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4249 		break;
4250 	}
4251 	case KVM_X86_SET_MCE: {
4252 		struct kvm_x86_mce mce;
4253 
4254 		r = -EFAULT;
4255 		if (copy_from_user(&mce, argp, sizeof(mce)))
4256 			goto out;
4257 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4258 		break;
4259 	}
4260 	case KVM_GET_VCPU_EVENTS: {
4261 		struct kvm_vcpu_events events;
4262 
4263 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4264 
4265 		r = -EFAULT;
4266 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4267 			break;
4268 		r = 0;
4269 		break;
4270 	}
4271 	case KVM_SET_VCPU_EVENTS: {
4272 		struct kvm_vcpu_events events;
4273 
4274 		r = -EFAULT;
4275 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4276 			break;
4277 
4278 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4279 		break;
4280 	}
4281 	case KVM_GET_DEBUGREGS: {
4282 		struct kvm_debugregs dbgregs;
4283 
4284 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4285 
4286 		r = -EFAULT;
4287 		if (copy_to_user(argp, &dbgregs,
4288 				 sizeof(struct kvm_debugregs)))
4289 			break;
4290 		r = 0;
4291 		break;
4292 	}
4293 	case KVM_SET_DEBUGREGS: {
4294 		struct kvm_debugregs dbgregs;
4295 
4296 		r = -EFAULT;
4297 		if (copy_from_user(&dbgregs, argp,
4298 				   sizeof(struct kvm_debugregs)))
4299 			break;
4300 
4301 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4302 		break;
4303 	}
4304 	case KVM_GET_XSAVE: {
4305 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4306 		r = -ENOMEM;
4307 		if (!u.xsave)
4308 			break;
4309 
4310 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4311 
4312 		r = -EFAULT;
4313 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4314 			break;
4315 		r = 0;
4316 		break;
4317 	}
4318 	case KVM_SET_XSAVE: {
4319 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4320 		if (IS_ERR(u.xsave)) {
4321 			r = PTR_ERR(u.xsave);
4322 			goto out_nofree;
4323 		}
4324 
4325 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4326 		break;
4327 	}
4328 	case KVM_GET_XCRS: {
4329 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4330 		r = -ENOMEM;
4331 		if (!u.xcrs)
4332 			break;
4333 
4334 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4335 
4336 		r = -EFAULT;
4337 		if (copy_to_user(argp, u.xcrs,
4338 				 sizeof(struct kvm_xcrs)))
4339 			break;
4340 		r = 0;
4341 		break;
4342 	}
4343 	case KVM_SET_XCRS: {
4344 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4345 		if (IS_ERR(u.xcrs)) {
4346 			r = PTR_ERR(u.xcrs);
4347 			goto out_nofree;
4348 		}
4349 
4350 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4351 		break;
4352 	}
4353 	case KVM_SET_TSC_KHZ: {
4354 		u32 user_tsc_khz;
4355 
4356 		r = -EINVAL;
4357 		user_tsc_khz = (u32)arg;
4358 
4359 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4360 			goto out;
4361 
4362 		if (user_tsc_khz == 0)
4363 			user_tsc_khz = tsc_khz;
4364 
4365 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4366 			r = 0;
4367 
4368 		goto out;
4369 	}
4370 	case KVM_GET_TSC_KHZ: {
4371 		r = vcpu->arch.virtual_tsc_khz;
4372 		goto out;
4373 	}
4374 	case KVM_KVMCLOCK_CTRL: {
4375 		r = kvm_set_guest_paused(vcpu);
4376 		goto out;
4377 	}
4378 	case KVM_ENABLE_CAP: {
4379 		struct kvm_enable_cap cap;
4380 
4381 		r = -EFAULT;
4382 		if (copy_from_user(&cap, argp, sizeof(cap)))
4383 			goto out;
4384 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4385 		break;
4386 	}
4387 	case KVM_GET_NESTED_STATE: {
4388 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4389 		u32 user_data_size;
4390 
4391 		r = -EINVAL;
4392 		if (!kvm_x86_ops->get_nested_state)
4393 			break;
4394 
4395 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4396 		r = -EFAULT;
4397 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4398 			break;
4399 
4400 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4401 						  user_data_size);
4402 		if (r < 0)
4403 			break;
4404 
4405 		if (r > user_data_size) {
4406 			if (put_user(r, &user_kvm_nested_state->size))
4407 				r = -EFAULT;
4408 			else
4409 				r = -E2BIG;
4410 			break;
4411 		}
4412 
4413 		r = 0;
4414 		break;
4415 	}
4416 	case KVM_SET_NESTED_STATE: {
4417 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4418 		struct kvm_nested_state kvm_state;
4419 
4420 		r = -EINVAL;
4421 		if (!kvm_x86_ops->set_nested_state)
4422 			break;
4423 
4424 		r = -EFAULT;
4425 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4426 			break;
4427 
4428 		r = -EINVAL;
4429 		if (kvm_state.size < sizeof(kvm_state))
4430 			break;
4431 
4432 		if (kvm_state.flags &
4433 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4434 		      | KVM_STATE_NESTED_EVMCS))
4435 			break;
4436 
4437 		/* nested_run_pending implies guest_mode.  */
4438 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4439 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4440 			break;
4441 
4442 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4443 		break;
4444 	}
4445 	case KVM_GET_SUPPORTED_HV_CPUID: {
4446 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4447 		struct kvm_cpuid2 cpuid;
4448 
4449 		r = -EFAULT;
4450 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4451 			goto out;
4452 
4453 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4454 						cpuid_arg->entries);
4455 		if (r)
4456 			goto out;
4457 
4458 		r = -EFAULT;
4459 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4460 			goto out;
4461 		r = 0;
4462 		break;
4463 	}
4464 	default:
4465 		r = -EINVAL;
4466 	}
4467 out:
4468 	kfree(u.buffer);
4469 out_nofree:
4470 	vcpu_put(vcpu);
4471 	return r;
4472 }
4473 
4474 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4475 {
4476 	return VM_FAULT_SIGBUS;
4477 }
4478 
4479 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4480 {
4481 	int ret;
4482 
4483 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4484 		return -EINVAL;
4485 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4486 	return ret;
4487 }
4488 
4489 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4490 					      u64 ident_addr)
4491 {
4492 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4493 }
4494 
4495 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4496 					 unsigned long kvm_nr_mmu_pages)
4497 {
4498 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4499 		return -EINVAL;
4500 
4501 	mutex_lock(&kvm->slots_lock);
4502 
4503 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4504 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4505 
4506 	mutex_unlock(&kvm->slots_lock);
4507 	return 0;
4508 }
4509 
4510 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4511 {
4512 	return kvm->arch.n_max_mmu_pages;
4513 }
4514 
4515 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4516 {
4517 	struct kvm_pic *pic = kvm->arch.vpic;
4518 	int r;
4519 
4520 	r = 0;
4521 	switch (chip->chip_id) {
4522 	case KVM_IRQCHIP_PIC_MASTER:
4523 		memcpy(&chip->chip.pic, &pic->pics[0],
4524 			sizeof(struct kvm_pic_state));
4525 		break;
4526 	case KVM_IRQCHIP_PIC_SLAVE:
4527 		memcpy(&chip->chip.pic, &pic->pics[1],
4528 			sizeof(struct kvm_pic_state));
4529 		break;
4530 	case KVM_IRQCHIP_IOAPIC:
4531 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4532 		break;
4533 	default:
4534 		r = -EINVAL;
4535 		break;
4536 	}
4537 	return r;
4538 }
4539 
4540 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4541 {
4542 	struct kvm_pic *pic = kvm->arch.vpic;
4543 	int r;
4544 
4545 	r = 0;
4546 	switch (chip->chip_id) {
4547 	case KVM_IRQCHIP_PIC_MASTER:
4548 		spin_lock(&pic->lock);
4549 		memcpy(&pic->pics[0], &chip->chip.pic,
4550 			sizeof(struct kvm_pic_state));
4551 		spin_unlock(&pic->lock);
4552 		break;
4553 	case KVM_IRQCHIP_PIC_SLAVE:
4554 		spin_lock(&pic->lock);
4555 		memcpy(&pic->pics[1], &chip->chip.pic,
4556 			sizeof(struct kvm_pic_state));
4557 		spin_unlock(&pic->lock);
4558 		break;
4559 	case KVM_IRQCHIP_IOAPIC:
4560 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4561 		break;
4562 	default:
4563 		r = -EINVAL;
4564 		break;
4565 	}
4566 	kvm_pic_update_irq(pic);
4567 	return r;
4568 }
4569 
4570 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4571 {
4572 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4573 
4574 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4575 
4576 	mutex_lock(&kps->lock);
4577 	memcpy(ps, &kps->channels, sizeof(*ps));
4578 	mutex_unlock(&kps->lock);
4579 	return 0;
4580 }
4581 
4582 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4583 {
4584 	int i;
4585 	struct kvm_pit *pit = kvm->arch.vpit;
4586 
4587 	mutex_lock(&pit->pit_state.lock);
4588 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4589 	for (i = 0; i < 3; i++)
4590 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4591 	mutex_unlock(&pit->pit_state.lock);
4592 	return 0;
4593 }
4594 
4595 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4596 {
4597 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4598 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4599 		sizeof(ps->channels));
4600 	ps->flags = kvm->arch.vpit->pit_state.flags;
4601 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4602 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4603 	return 0;
4604 }
4605 
4606 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4607 {
4608 	int start = 0;
4609 	int i;
4610 	u32 prev_legacy, cur_legacy;
4611 	struct kvm_pit *pit = kvm->arch.vpit;
4612 
4613 	mutex_lock(&pit->pit_state.lock);
4614 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4615 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4616 	if (!prev_legacy && cur_legacy)
4617 		start = 1;
4618 	memcpy(&pit->pit_state.channels, &ps->channels,
4619 	       sizeof(pit->pit_state.channels));
4620 	pit->pit_state.flags = ps->flags;
4621 	for (i = 0; i < 3; i++)
4622 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4623 				   start && i == 0);
4624 	mutex_unlock(&pit->pit_state.lock);
4625 	return 0;
4626 }
4627 
4628 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4629 				 struct kvm_reinject_control *control)
4630 {
4631 	struct kvm_pit *pit = kvm->arch.vpit;
4632 
4633 	if (!pit)
4634 		return -ENXIO;
4635 
4636 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4637 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4638 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4639 	 */
4640 	mutex_lock(&pit->pit_state.lock);
4641 	kvm_pit_set_reinject(pit, control->pit_reinject);
4642 	mutex_unlock(&pit->pit_state.lock);
4643 
4644 	return 0;
4645 }
4646 
4647 /**
4648  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4649  * @kvm: kvm instance
4650  * @log: slot id and address to which we copy the log
4651  *
4652  * Steps 1-4 below provide general overview of dirty page logging. See
4653  * kvm_get_dirty_log_protect() function description for additional details.
4654  *
4655  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4656  * always flush the TLB (step 4) even if previous step failed  and the dirty
4657  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4658  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4659  * writes will be marked dirty for next log read.
4660  *
4661  *   1. Take a snapshot of the bit and clear it if needed.
4662  *   2. Write protect the corresponding page.
4663  *   3. Copy the snapshot to the userspace.
4664  *   4. Flush TLB's if needed.
4665  */
4666 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4667 {
4668 	bool flush = false;
4669 	int r;
4670 
4671 	mutex_lock(&kvm->slots_lock);
4672 
4673 	/*
4674 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4675 	 */
4676 	if (kvm_x86_ops->flush_log_dirty)
4677 		kvm_x86_ops->flush_log_dirty(kvm);
4678 
4679 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4680 
4681 	/*
4682 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4683 	 * kvm_mmu_slot_remove_write_access().
4684 	 */
4685 	lockdep_assert_held(&kvm->slots_lock);
4686 	if (flush)
4687 		kvm_flush_remote_tlbs(kvm);
4688 
4689 	mutex_unlock(&kvm->slots_lock);
4690 	return r;
4691 }
4692 
4693 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4694 {
4695 	bool flush = false;
4696 	int r;
4697 
4698 	mutex_lock(&kvm->slots_lock);
4699 
4700 	/*
4701 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4702 	 */
4703 	if (kvm_x86_ops->flush_log_dirty)
4704 		kvm_x86_ops->flush_log_dirty(kvm);
4705 
4706 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4707 
4708 	/*
4709 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4710 	 * kvm_mmu_slot_remove_write_access().
4711 	 */
4712 	lockdep_assert_held(&kvm->slots_lock);
4713 	if (flush)
4714 		kvm_flush_remote_tlbs(kvm);
4715 
4716 	mutex_unlock(&kvm->slots_lock);
4717 	return r;
4718 }
4719 
4720 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4721 			bool line_status)
4722 {
4723 	if (!irqchip_in_kernel(kvm))
4724 		return -ENXIO;
4725 
4726 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4727 					irq_event->irq, irq_event->level,
4728 					line_status);
4729 	return 0;
4730 }
4731 
4732 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4733 			    struct kvm_enable_cap *cap)
4734 {
4735 	int r;
4736 
4737 	if (cap->flags)
4738 		return -EINVAL;
4739 
4740 	switch (cap->cap) {
4741 	case KVM_CAP_DISABLE_QUIRKS:
4742 		kvm->arch.disabled_quirks = cap->args[0];
4743 		r = 0;
4744 		break;
4745 	case KVM_CAP_SPLIT_IRQCHIP: {
4746 		mutex_lock(&kvm->lock);
4747 		r = -EINVAL;
4748 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4749 			goto split_irqchip_unlock;
4750 		r = -EEXIST;
4751 		if (irqchip_in_kernel(kvm))
4752 			goto split_irqchip_unlock;
4753 		if (kvm->created_vcpus)
4754 			goto split_irqchip_unlock;
4755 		r = kvm_setup_empty_irq_routing(kvm);
4756 		if (r)
4757 			goto split_irqchip_unlock;
4758 		/* Pairs with irqchip_in_kernel. */
4759 		smp_wmb();
4760 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4761 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4762 		r = 0;
4763 split_irqchip_unlock:
4764 		mutex_unlock(&kvm->lock);
4765 		break;
4766 	}
4767 	case KVM_CAP_X2APIC_API:
4768 		r = -EINVAL;
4769 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4770 			break;
4771 
4772 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4773 			kvm->arch.x2apic_format = true;
4774 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4775 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4776 
4777 		r = 0;
4778 		break;
4779 	case KVM_CAP_X86_DISABLE_EXITS:
4780 		r = -EINVAL;
4781 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4782 			break;
4783 
4784 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4785 			kvm_can_mwait_in_guest())
4786 			kvm->arch.mwait_in_guest = true;
4787 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4788 			kvm->arch.hlt_in_guest = true;
4789 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4790 			kvm->arch.pause_in_guest = true;
4791 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4792 			kvm->arch.cstate_in_guest = true;
4793 		r = 0;
4794 		break;
4795 	case KVM_CAP_MSR_PLATFORM_INFO:
4796 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4797 		r = 0;
4798 		break;
4799 	case KVM_CAP_EXCEPTION_PAYLOAD:
4800 		kvm->arch.exception_payload_enabled = cap->args[0];
4801 		r = 0;
4802 		break;
4803 	default:
4804 		r = -EINVAL;
4805 		break;
4806 	}
4807 	return r;
4808 }
4809 
4810 long kvm_arch_vm_ioctl(struct file *filp,
4811 		       unsigned int ioctl, unsigned long arg)
4812 {
4813 	struct kvm *kvm = filp->private_data;
4814 	void __user *argp = (void __user *)arg;
4815 	int r = -ENOTTY;
4816 	/*
4817 	 * This union makes it completely explicit to gcc-3.x
4818 	 * that these two variables' stack usage should be
4819 	 * combined, not added together.
4820 	 */
4821 	union {
4822 		struct kvm_pit_state ps;
4823 		struct kvm_pit_state2 ps2;
4824 		struct kvm_pit_config pit_config;
4825 	} u;
4826 
4827 	switch (ioctl) {
4828 	case KVM_SET_TSS_ADDR:
4829 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4830 		break;
4831 	case KVM_SET_IDENTITY_MAP_ADDR: {
4832 		u64 ident_addr;
4833 
4834 		mutex_lock(&kvm->lock);
4835 		r = -EINVAL;
4836 		if (kvm->created_vcpus)
4837 			goto set_identity_unlock;
4838 		r = -EFAULT;
4839 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4840 			goto set_identity_unlock;
4841 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4842 set_identity_unlock:
4843 		mutex_unlock(&kvm->lock);
4844 		break;
4845 	}
4846 	case KVM_SET_NR_MMU_PAGES:
4847 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4848 		break;
4849 	case KVM_GET_NR_MMU_PAGES:
4850 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4851 		break;
4852 	case KVM_CREATE_IRQCHIP: {
4853 		mutex_lock(&kvm->lock);
4854 
4855 		r = -EEXIST;
4856 		if (irqchip_in_kernel(kvm))
4857 			goto create_irqchip_unlock;
4858 
4859 		r = -EINVAL;
4860 		if (kvm->created_vcpus)
4861 			goto create_irqchip_unlock;
4862 
4863 		r = kvm_pic_init(kvm);
4864 		if (r)
4865 			goto create_irqchip_unlock;
4866 
4867 		r = kvm_ioapic_init(kvm);
4868 		if (r) {
4869 			kvm_pic_destroy(kvm);
4870 			goto create_irqchip_unlock;
4871 		}
4872 
4873 		r = kvm_setup_default_irq_routing(kvm);
4874 		if (r) {
4875 			kvm_ioapic_destroy(kvm);
4876 			kvm_pic_destroy(kvm);
4877 			goto create_irqchip_unlock;
4878 		}
4879 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4880 		smp_wmb();
4881 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4882 	create_irqchip_unlock:
4883 		mutex_unlock(&kvm->lock);
4884 		break;
4885 	}
4886 	case KVM_CREATE_PIT:
4887 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4888 		goto create_pit;
4889 	case KVM_CREATE_PIT2:
4890 		r = -EFAULT;
4891 		if (copy_from_user(&u.pit_config, argp,
4892 				   sizeof(struct kvm_pit_config)))
4893 			goto out;
4894 	create_pit:
4895 		mutex_lock(&kvm->lock);
4896 		r = -EEXIST;
4897 		if (kvm->arch.vpit)
4898 			goto create_pit_unlock;
4899 		r = -ENOMEM;
4900 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4901 		if (kvm->arch.vpit)
4902 			r = 0;
4903 	create_pit_unlock:
4904 		mutex_unlock(&kvm->lock);
4905 		break;
4906 	case KVM_GET_IRQCHIP: {
4907 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4908 		struct kvm_irqchip *chip;
4909 
4910 		chip = memdup_user(argp, sizeof(*chip));
4911 		if (IS_ERR(chip)) {
4912 			r = PTR_ERR(chip);
4913 			goto out;
4914 		}
4915 
4916 		r = -ENXIO;
4917 		if (!irqchip_kernel(kvm))
4918 			goto get_irqchip_out;
4919 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4920 		if (r)
4921 			goto get_irqchip_out;
4922 		r = -EFAULT;
4923 		if (copy_to_user(argp, chip, sizeof(*chip)))
4924 			goto get_irqchip_out;
4925 		r = 0;
4926 	get_irqchip_out:
4927 		kfree(chip);
4928 		break;
4929 	}
4930 	case KVM_SET_IRQCHIP: {
4931 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4932 		struct kvm_irqchip *chip;
4933 
4934 		chip = memdup_user(argp, sizeof(*chip));
4935 		if (IS_ERR(chip)) {
4936 			r = PTR_ERR(chip);
4937 			goto out;
4938 		}
4939 
4940 		r = -ENXIO;
4941 		if (!irqchip_kernel(kvm))
4942 			goto set_irqchip_out;
4943 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4944 		if (r)
4945 			goto set_irqchip_out;
4946 		r = 0;
4947 	set_irqchip_out:
4948 		kfree(chip);
4949 		break;
4950 	}
4951 	case KVM_GET_PIT: {
4952 		r = -EFAULT;
4953 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4954 			goto out;
4955 		r = -ENXIO;
4956 		if (!kvm->arch.vpit)
4957 			goto out;
4958 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4959 		if (r)
4960 			goto out;
4961 		r = -EFAULT;
4962 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4963 			goto out;
4964 		r = 0;
4965 		break;
4966 	}
4967 	case KVM_SET_PIT: {
4968 		r = -EFAULT;
4969 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4970 			goto out;
4971 		r = -ENXIO;
4972 		if (!kvm->arch.vpit)
4973 			goto out;
4974 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4975 		break;
4976 	}
4977 	case KVM_GET_PIT2: {
4978 		r = -ENXIO;
4979 		if (!kvm->arch.vpit)
4980 			goto out;
4981 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4982 		if (r)
4983 			goto out;
4984 		r = -EFAULT;
4985 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4986 			goto out;
4987 		r = 0;
4988 		break;
4989 	}
4990 	case KVM_SET_PIT2: {
4991 		r = -EFAULT;
4992 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4993 			goto out;
4994 		r = -ENXIO;
4995 		if (!kvm->arch.vpit)
4996 			goto out;
4997 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4998 		break;
4999 	}
5000 	case KVM_REINJECT_CONTROL: {
5001 		struct kvm_reinject_control control;
5002 		r =  -EFAULT;
5003 		if (copy_from_user(&control, argp, sizeof(control)))
5004 			goto out;
5005 		r = kvm_vm_ioctl_reinject(kvm, &control);
5006 		break;
5007 	}
5008 	case KVM_SET_BOOT_CPU_ID:
5009 		r = 0;
5010 		mutex_lock(&kvm->lock);
5011 		if (kvm->created_vcpus)
5012 			r = -EBUSY;
5013 		else
5014 			kvm->arch.bsp_vcpu_id = arg;
5015 		mutex_unlock(&kvm->lock);
5016 		break;
5017 	case KVM_XEN_HVM_CONFIG: {
5018 		struct kvm_xen_hvm_config xhc;
5019 		r = -EFAULT;
5020 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5021 			goto out;
5022 		r = -EINVAL;
5023 		if (xhc.flags)
5024 			goto out;
5025 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5026 		r = 0;
5027 		break;
5028 	}
5029 	case KVM_SET_CLOCK: {
5030 		struct kvm_clock_data user_ns;
5031 		u64 now_ns;
5032 
5033 		r = -EFAULT;
5034 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5035 			goto out;
5036 
5037 		r = -EINVAL;
5038 		if (user_ns.flags)
5039 			goto out;
5040 
5041 		r = 0;
5042 		/*
5043 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5044 		 * kvm_gen_update_masterclock() can be cut down to locked
5045 		 * pvclock_update_vm_gtod_copy().
5046 		 */
5047 		kvm_gen_update_masterclock(kvm);
5048 		now_ns = get_kvmclock_ns(kvm);
5049 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5050 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5051 		break;
5052 	}
5053 	case KVM_GET_CLOCK: {
5054 		struct kvm_clock_data user_ns;
5055 		u64 now_ns;
5056 
5057 		now_ns = get_kvmclock_ns(kvm);
5058 		user_ns.clock = now_ns;
5059 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5060 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5061 
5062 		r = -EFAULT;
5063 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5064 			goto out;
5065 		r = 0;
5066 		break;
5067 	}
5068 	case KVM_MEMORY_ENCRYPT_OP: {
5069 		r = -ENOTTY;
5070 		if (kvm_x86_ops->mem_enc_op)
5071 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
5072 		break;
5073 	}
5074 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5075 		struct kvm_enc_region region;
5076 
5077 		r = -EFAULT;
5078 		if (copy_from_user(&region, argp, sizeof(region)))
5079 			goto out;
5080 
5081 		r = -ENOTTY;
5082 		if (kvm_x86_ops->mem_enc_reg_region)
5083 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
5084 		break;
5085 	}
5086 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5087 		struct kvm_enc_region region;
5088 
5089 		r = -EFAULT;
5090 		if (copy_from_user(&region, argp, sizeof(region)))
5091 			goto out;
5092 
5093 		r = -ENOTTY;
5094 		if (kvm_x86_ops->mem_enc_unreg_region)
5095 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
5096 		break;
5097 	}
5098 	case KVM_HYPERV_EVENTFD: {
5099 		struct kvm_hyperv_eventfd hvevfd;
5100 
5101 		r = -EFAULT;
5102 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5103 			goto out;
5104 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5105 		break;
5106 	}
5107 	case KVM_SET_PMU_EVENT_FILTER:
5108 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5109 		break;
5110 	default:
5111 		r = -ENOTTY;
5112 	}
5113 out:
5114 	return r;
5115 }
5116 
5117 static void kvm_init_msr_list(void)
5118 {
5119 	struct x86_pmu_capability x86_pmu;
5120 	u32 dummy[2];
5121 	unsigned i, j;
5122 
5123 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5124 			 "Please update the fixed PMCs in msrs_to_save[]");
5125 
5126 	perf_get_x86_pmu_capability(&x86_pmu);
5127 
5128 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5129 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5130 			continue;
5131 
5132 		/*
5133 		 * Even MSRs that are valid in the host may not be exposed
5134 		 * to the guests in some cases.
5135 		 */
5136 		switch (msrs_to_save[i]) {
5137 		case MSR_IA32_BNDCFGS:
5138 			if (!kvm_mpx_supported())
5139 				continue;
5140 			break;
5141 		case MSR_TSC_AUX:
5142 			if (!kvm_x86_ops->rdtscp_supported())
5143 				continue;
5144 			break;
5145 		case MSR_IA32_RTIT_CTL:
5146 		case MSR_IA32_RTIT_STATUS:
5147 			if (!kvm_x86_ops->pt_supported())
5148 				continue;
5149 			break;
5150 		case MSR_IA32_RTIT_CR3_MATCH:
5151 			if (!kvm_x86_ops->pt_supported() ||
5152 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5153 				continue;
5154 			break;
5155 		case MSR_IA32_RTIT_OUTPUT_BASE:
5156 		case MSR_IA32_RTIT_OUTPUT_MASK:
5157 			if (!kvm_x86_ops->pt_supported() ||
5158 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5159 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5160 				continue;
5161 			break;
5162 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5163 			if (!kvm_x86_ops->pt_supported() ||
5164 				msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5165 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5166 				continue;
5167 			break;
5168 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5169 			if (msrs_to_save[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5170 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5171 				continue;
5172 			break;
5173 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5174 			if (msrs_to_save[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5175 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5176 				continue;
5177 		}
5178 		default:
5179 			break;
5180 		}
5181 
5182 		if (j < i)
5183 			msrs_to_save[j] = msrs_to_save[i];
5184 		j++;
5185 	}
5186 	num_msrs_to_save = j;
5187 
5188 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5189 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5190 			continue;
5191 
5192 		if (j < i)
5193 			emulated_msrs[j] = emulated_msrs[i];
5194 		j++;
5195 	}
5196 	num_emulated_msrs = j;
5197 
5198 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5199 		struct kvm_msr_entry msr;
5200 
5201 		msr.index = msr_based_features[i];
5202 		if (kvm_get_msr_feature(&msr))
5203 			continue;
5204 
5205 		if (j < i)
5206 			msr_based_features[j] = msr_based_features[i];
5207 		j++;
5208 	}
5209 	num_msr_based_features = j;
5210 }
5211 
5212 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5213 			   const void *v)
5214 {
5215 	int handled = 0;
5216 	int n;
5217 
5218 	do {
5219 		n = min(len, 8);
5220 		if (!(lapic_in_kernel(vcpu) &&
5221 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5222 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5223 			break;
5224 		handled += n;
5225 		addr += n;
5226 		len -= n;
5227 		v += n;
5228 	} while (len);
5229 
5230 	return handled;
5231 }
5232 
5233 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5234 {
5235 	int handled = 0;
5236 	int n;
5237 
5238 	do {
5239 		n = min(len, 8);
5240 		if (!(lapic_in_kernel(vcpu) &&
5241 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5242 					 addr, n, v))
5243 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5244 			break;
5245 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5246 		handled += n;
5247 		addr += n;
5248 		len -= n;
5249 		v += n;
5250 	} while (len);
5251 
5252 	return handled;
5253 }
5254 
5255 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5256 			struct kvm_segment *var, int seg)
5257 {
5258 	kvm_x86_ops->set_segment(vcpu, var, seg);
5259 }
5260 
5261 void kvm_get_segment(struct kvm_vcpu *vcpu,
5262 		     struct kvm_segment *var, int seg)
5263 {
5264 	kvm_x86_ops->get_segment(vcpu, var, seg);
5265 }
5266 
5267 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5268 			   struct x86_exception *exception)
5269 {
5270 	gpa_t t_gpa;
5271 
5272 	BUG_ON(!mmu_is_nested(vcpu));
5273 
5274 	/* NPT walks are always user-walks */
5275 	access |= PFERR_USER_MASK;
5276 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5277 
5278 	return t_gpa;
5279 }
5280 
5281 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5282 			      struct x86_exception *exception)
5283 {
5284 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5285 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5286 }
5287 
5288  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5289 				struct x86_exception *exception)
5290 {
5291 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5292 	access |= PFERR_FETCH_MASK;
5293 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5294 }
5295 
5296 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5297 			       struct x86_exception *exception)
5298 {
5299 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5300 	access |= PFERR_WRITE_MASK;
5301 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5302 }
5303 
5304 /* uses this to access any guest's mapped memory without checking CPL */
5305 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5306 				struct x86_exception *exception)
5307 {
5308 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5309 }
5310 
5311 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5312 				      struct kvm_vcpu *vcpu, u32 access,
5313 				      struct x86_exception *exception)
5314 {
5315 	void *data = val;
5316 	int r = X86EMUL_CONTINUE;
5317 
5318 	while (bytes) {
5319 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5320 							    exception);
5321 		unsigned offset = addr & (PAGE_SIZE-1);
5322 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5323 		int ret;
5324 
5325 		if (gpa == UNMAPPED_GVA)
5326 			return X86EMUL_PROPAGATE_FAULT;
5327 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5328 					       offset, toread);
5329 		if (ret < 0) {
5330 			r = X86EMUL_IO_NEEDED;
5331 			goto out;
5332 		}
5333 
5334 		bytes -= toread;
5335 		data += toread;
5336 		addr += toread;
5337 	}
5338 out:
5339 	return r;
5340 }
5341 
5342 /* used for instruction fetching */
5343 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5344 				gva_t addr, void *val, unsigned int bytes,
5345 				struct x86_exception *exception)
5346 {
5347 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5348 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5349 	unsigned offset;
5350 	int ret;
5351 
5352 	/* Inline kvm_read_guest_virt_helper for speed.  */
5353 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5354 						    exception);
5355 	if (unlikely(gpa == UNMAPPED_GVA))
5356 		return X86EMUL_PROPAGATE_FAULT;
5357 
5358 	offset = addr & (PAGE_SIZE-1);
5359 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5360 		bytes = (unsigned)PAGE_SIZE - offset;
5361 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5362 				       offset, bytes);
5363 	if (unlikely(ret < 0))
5364 		return X86EMUL_IO_NEEDED;
5365 
5366 	return X86EMUL_CONTINUE;
5367 }
5368 
5369 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5370 			       gva_t addr, void *val, unsigned int bytes,
5371 			       struct x86_exception *exception)
5372 {
5373 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5374 
5375 	/*
5376 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5377 	 * is returned, but our callers are not ready for that and they blindly
5378 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5379 	 * uninitialized kernel stack memory into cr2 and error code.
5380 	 */
5381 	memset(exception, 0, sizeof(*exception));
5382 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5383 					  exception);
5384 }
5385 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5386 
5387 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5388 			     gva_t addr, void *val, unsigned int bytes,
5389 			     struct x86_exception *exception, bool system)
5390 {
5391 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5392 	u32 access = 0;
5393 
5394 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5395 		access |= PFERR_USER_MASK;
5396 
5397 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5398 }
5399 
5400 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5401 		unsigned long addr, void *val, unsigned int bytes)
5402 {
5403 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5404 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5405 
5406 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5407 }
5408 
5409 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5410 				      struct kvm_vcpu *vcpu, u32 access,
5411 				      struct x86_exception *exception)
5412 {
5413 	void *data = val;
5414 	int r = X86EMUL_CONTINUE;
5415 
5416 	while (bytes) {
5417 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5418 							     access,
5419 							     exception);
5420 		unsigned offset = addr & (PAGE_SIZE-1);
5421 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5422 		int ret;
5423 
5424 		if (gpa == UNMAPPED_GVA)
5425 			return X86EMUL_PROPAGATE_FAULT;
5426 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5427 		if (ret < 0) {
5428 			r = X86EMUL_IO_NEEDED;
5429 			goto out;
5430 		}
5431 
5432 		bytes -= towrite;
5433 		data += towrite;
5434 		addr += towrite;
5435 	}
5436 out:
5437 	return r;
5438 }
5439 
5440 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5441 			      unsigned int bytes, struct x86_exception *exception,
5442 			      bool system)
5443 {
5444 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5445 	u32 access = PFERR_WRITE_MASK;
5446 
5447 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5448 		access |= PFERR_USER_MASK;
5449 
5450 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5451 					   access, exception);
5452 }
5453 
5454 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5455 				unsigned int bytes, struct x86_exception *exception)
5456 {
5457 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5458 	vcpu->arch.l1tf_flush_l1d = true;
5459 
5460 	/*
5461 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5462 	 * is returned, but our callers are not ready for that and they blindly
5463 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5464 	 * uninitialized kernel stack memory into cr2 and error code.
5465 	 */
5466 	memset(exception, 0, sizeof(*exception));
5467 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5468 					   PFERR_WRITE_MASK, exception);
5469 }
5470 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5471 
5472 int handle_ud(struct kvm_vcpu *vcpu)
5473 {
5474 	int emul_type = EMULTYPE_TRAP_UD;
5475 	char sig[5]; /* ud2; .ascii "kvm" */
5476 	struct x86_exception e;
5477 
5478 	if (force_emulation_prefix &&
5479 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5480 				sig, sizeof(sig), &e) == 0 &&
5481 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5482 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5483 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5484 	}
5485 
5486 	return kvm_emulate_instruction(vcpu, emul_type);
5487 }
5488 EXPORT_SYMBOL_GPL(handle_ud);
5489 
5490 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5491 			    gpa_t gpa, bool write)
5492 {
5493 	/* For APIC access vmexit */
5494 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5495 		return 1;
5496 
5497 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5498 		trace_vcpu_match_mmio(gva, gpa, write, true);
5499 		return 1;
5500 	}
5501 
5502 	return 0;
5503 }
5504 
5505 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5506 				gpa_t *gpa, struct x86_exception *exception,
5507 				bool write)
5508 {
5509 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5510 		| (write ? PFERR_WRITE_MASK : 0);
5511 
5512 	/*
5513 	 * currently PKRU is only applied to ept enabled guest so
5514 	 * there is no pkey in EPT page table for L1 guest or EPT
5515 	 * shadow page table for L2 guest.
5516 	 */
5517 	if (vcpu_match_mmio_gva(vcpu, gva)
5518 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5519 				 vcpu->arch.mmio_access, 0, access)) {
5520 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5521 					(gva & (PAGE_SIZE - 1));
5522 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5523 		return 1;
5524 	}
5525 
5526 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5527 
5528 	if (*gpa == UNMAPPED_GVA)
5529 		return -1;
5530 
5531 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5532 }
5533 
5534 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5535 			const void *val, int bytes)
5536 {
5537 	int ret;
5538 
5539 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5540 	if (ret < 0)
5541 		return 0;
5542 	kvm_page_track_write(vcpu, gpa, val, bytes);
5543 	return 1;
5544 }
5545 
5546 struct read_write_emulator_ops {
5547 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5548 				  int bytes);
5549 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5550 				  void *val, int bytes);
5551 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5552 			       int bytes, void *val);
5553 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5554 				    void *val, int bytes);
5555 	bool write;
5556 };
5557 
5558 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5559 {
5560 	if (vcpu->mmio_read_completed) {
5561 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5562 			       vcpu->mmio_fragments[0].gpa, val);
5563 		vcpu->mmio_read_completed = 0;
5564 		return 1;
5565 	}
5566 
5567 	return 0;
5568 }
5569 
5570 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5571 			void *val, int bytes)
5572 {
5573 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5574 }
5575 
5576 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5577 			 void *val, int bytes)
5578 {
5579 	return emulator_write_phys(vcpu, gpa, val, bytes);
5580 }
5581 
5582 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5583 {
5584 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5585 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5586 }
5587 
5588 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5589 			  void *val, int bytes)
5590 {
5591 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5592 	return X86EMUL_IO_NEEDED;
5593 }
5594 
5595 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5596 			   void *val, int bytes)
5597 {
5598 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5599 
5600 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5601 	return X86EMUL_CONTINUE;
5602 }
5603 
5604 static const struct read_write_emulator_ops read_emultor = {
5605 	.read_write_prepare = read_prepare,
5606 	.read_write_emulate = read_emulate,
5607 	.read_write_mmio = vcpu_mmio_read,
5608 	.read_write_exit_mmio = read_exit_mmio,
5609 };
5610 
5611 static const struct read_write_emulator_ops write_emultor = {
5612 	.read_write_emulate = write_emulate,
5613 	.read_write_mmio = write_mmio,
5614 	.read_write_exit_mmio = write_exit_mmio,
5615 	.write = true,
5616 };
5617 
5618 static int emulator_read_write_onepage(unsigned long addr, void *val,
5619 				       unsigned int bytes,
5620 				       struct x86_exception *exception,
5621 				       struct kvm_vcpu *vcpu,
5622 				       const struct read_write_emulator_ops *ops)
5623 {
5624 	gpa_t gpa;
5625 	int handled, ret;
5626 	bool write = ops->write;
5627 	struct kvm_mmio_fragment *frag;
5628 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5629 
5630 	/*
5631 	 * If the exit was due to a NPF we may already have a GPA.
5632 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5633 	 * Note, this cannot be used on string operations since string
5634 	 * operation using rep will only have the initial GPA from the NPF
5635 	 * occurred.
5636 	 */
5637 	if (vcpu->arch.gpa_available &&
5638 	    emulator_can_use_gpa(ctxt) &&
5639 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5640 		gpa = vcpu->arch.gpa_val;
5641 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5642 	} else {
5643 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5644 		if (ret < 0)
5645 			return X86EMUL_PROPAGATE_FAULT;
5646 	}
5647 
5648 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5649 		return X86EMUL_CONTINUE;
5650 
5651 	/*
5652 	 * Is this MMIO handled locally?
5653 	 */
5654 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5655 	if (handled == bytes)
5656 		return X86EMUL_CONTINUE;
5657 
5658 	gpa += handled;
5659 	bytes -= handled;
5660 	val += handled;
5661 
5662 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5663 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5664 	frag->gpa = gpa;
5665 	frag->data = val;
5666 	frag->len = bytes;
5667 	return X86EMUL_CONTINUE;
5668 }
5669 
5670 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5671 			unsigned long addr,
5672 			void *val, unsigned int bytes,
5673 			struct x86_exception *exception,
5674 			const struct read_write_emulator_ops *ops)
5675 {
5676 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5677 	gpa_t gpa;
5678 	int rc;
5679 
5680 	if (ops->read_write_prepare &&
5681 		  ops->read_write_prepare(vcpu, val, bytes))
5682 		return X86EMUL_CONTINUE;
5683 
5684 	vcpu->mmio_nr_fragments = 0;
5685 
5686 	/* Crossing a page boundary? */
5687 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5688 		int now;
5689 
5690 		now = -addr & ~PAGE_MASK;
5691 		rc = emulator_read_write_onepage(addr, val, now, exception,
5692 						 vcpu, ops);
5693 
5694 		if (rc != X86EMUL_CONTINUE)
5695 			return rc;
5696 		addr += now;
5697 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5698 			addr = (u32)addr;
5699 		val += now;
5700 		bytes -= now;
5701 	}
5702 
5703 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5704 					 vcpu, ops);
5705 	if (rc != X86EMUL_CONTINUE)
5706 		return rc;
5707 
5708 	if (!vcpu->mmio_nr_fragments)
5709 		return rc;
5710 
5711 	gpa = vcpu->mmio_fragments[0].gpa;
5712 
5713 	vcpu->mmio_needed = 1;
5714 	vcpu->mmio_cur_fragment = 0;
5715 
5716 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5717 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5718 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5719 	vcpu->run->mmio.phys_addr = gpa;
5720 
5721 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5722 }
5723 
5724 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5725 				  unsigned long addr,
5726 				  void *val,
5727 				  unsigned int bytes,
5728 				  struct x86_exception *exception)
5729 {
5730 	return emulator_read_write(ctxt, addr, val, bytes,
5731 				   exception, &read_emultor);
5732 }
5733 
5734 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5735 			    unsigned long addr,
5736 			    const void *val,
5737 			    unsigned int bytes,
5738 			    struct x86_exception *exception)
5739 {
5740 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5741 				   exception, &write_emultor);
5742 }
5743 
5744 #define CMPXCHG_TYPE(t, ptr, old, new) \
5745 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5746 
5747 #ifdef CONFIG_X86_64
5748 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5749 #else
5750 #  define CMPXCHG64(ptr, old, new) \
5751 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5752 #endif
5753 
5754 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5755 				     unsigned long addr,
5756 				     const void *old,
5757 				     const void *new,
5758 				     unsigned int bytes,
5759 				     struct x86_exception *exception)
5760 {
5761 	struct kvm_host_map map;
5762 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5763 	gpa_t gpa;
5764 	char *kaddr;
5765 	bool exchanged;
5766 
5767 	/* guests cmpxchg8b have to be emulated atomically */
5768 	if (bytes > 8 || (bytes & (bytes - 1)))
5769 		goto emul_write;
5770 
5771 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5772 
5773 	if (gpa == UNMAPPED_GVA ||
5774 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5775 		goto emul_write;
5776 
5777 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5778 		goto emul_write;
5779 
5780 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5781 		goto emul_write;
5782 
5783 	kaddr = map.hva + offset_in_page(gpa);
5784 
5785 	switch (bytes) {
5786 	case 1:
5787 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5788 		break;
5789 	case 2:
5790 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5791 		break;
5792 	case 4:
5793 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5794 		break;
5795 	case 8:
5796 		exchanged = CMPXCHG64(kaddr, old, new);
5797 		break;
5798 	default:
5799 		BUG();
5800 	}
5801 
5802 	kvm_vcpu_unmap(vcpu, &map, true);
5803 
5804 	if (!exchanged)
5805 		return X86EMUL_CMPXCHG_FAILED;
5806 
5807 	kvm_page_track_write(vcpu, gpa, new, bytes);
5808 
5809 	return X86EMUL_CONTINUE;
5810 
5811 emul_write:
5812 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5813 
5814 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5815 }
5816 
5817 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5818 {
5819 	int r = 0, i;
5820 
5821 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5822 		if (vcpu->arch.pio.in)
5823 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5824 					    vcpu->arch.pio.size, pd);
5825 		else
5826 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5827 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5828 					     pd);
5829 		if (r)
5830 			break;
5831 		pd += vcpu->arch.pio.size;
5832 	}
5833 	return r;
5834 }
5835 
5836 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5837 			       unsigned short port, void *val,
5838 			       unsigned int count, bool in)
5839 {
5840 	vcpu->arch.pio.port = port;
5841 	vcpu->arch.pio.in = in;
5842 	vcpu->arch.pio.count  = count;
5843 	vcpu->arch.pio.size = size;
5844 
5845 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5846 		vcpu->arch.pio.count = 0;
5847 		return 1;
5848 	}
5849 
5850 	vcpu->run->exit_reason = KVM_EXIT_IO;
5851 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5852 	vcpu->run->io.size = size;
5853 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5854 	vcpu->run->io.count = count;
5855 	vcpu->run->io.port = port;
5856 
5857 	return 0;
5858 }
5859 
5860 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5861 				    int size, unsigned short port, void *val,
5862 				    unsigned int count)
5863 {
5864 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5865 	int ret;
5866 
5867 	if (vcpu->arch.pio.count)
5868 		goto data_avail;
5869 
5870 	memset(vcpu->arch.pio_data, 0, size * count);
5871 
5872 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5873 	if (ret) {
5874 data_avail:
5875 		memcpy(val, vcpu->arch.pio_data, size * count);
5876 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5877 		vcpu->arch.pio.count = 0;
5878 		return 1;
5879 	}
5880 
5881 	return 0;
5882 }
5883 
5884 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5885 				     int size, unsigned short port,
5886 				     const void *val, unsigned int count)
5887 {
5888 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5889 
5890 	memcpy(vcpu->arch.pio_data, val, size * count);
5891 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5892 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5893 }
5894 
5895 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5896 {
5897 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5898 }
5899 
5900 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5901 {
5902 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5903 }
5904 
5905 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5906 {
5907 	if (!need_emulate_wbinvd(vcpu))
5908 		return X86EMUL_CONTINUE;
5909 
5910 	if (kvm_x86_ops->has_wbinvd_exit()) {
5911 		int cpu = get_cpu();
5912 
5913 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5914 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5915 				wbinvd_ipi, NULL, 1);
5916 		put_cpu();
5917 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5918 	} else
5919 		wbinvd();
5920 	return X86EMUL_CONTINUE;
5921 }
5922 
5923 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5924 {
5925 	kvm_emulate_wbinvd_noskip(vcpu);
5926 	return kvm_skip_emulated_instruction(vcpu);
5927 }
5928 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5929 
5930 
5931 
5932 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5933 {
5934 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5935 }
5936 
5937 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5938 			   unsigned long *dest)
5939 {
5940 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5941 }
5942 
5943 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5944 			   unsigned long value)
5945 {
5946 
5947 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5948 }
5949 
5950 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5951 {
5952 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5953 }
5954 
5955 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5956 {
5957 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5958 	unsigned long value;
5959 
5960 	switch (cr) {
5961 	case 0:
5962 		value = kvm_read_cr0(vcpu);
5963 		break;
5964 	case 2:
5965 		value = vcpu->arch.cr2;
5966 		break;
5967 	case 3:
5968 		value = kvm_read_cr3(vcpu);
5969 		break;
5970 	case 4:
5971 		value = kvm_read_cr4(vcpu);
5972 		break;
5973 	case 8:
5974 		value = kvm_get_cr8(vcpu);
5975 		break;
5976 	default:
5977 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5978 		return 0;
5979 	}
5980 
5981 	return value;
5982 }
5983 
5984 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5985 {
5986 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5987 	int res = 0;
5988 
5989 	switch (cr) {
5990 	case 0:
5991 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5992 		break;
5993 	case 2:
5994 		vcpu->arch.cr2 = val;
5995 		break;
5996 	case 3:
5997 		res = kvm_set_cr3(vcpu, val);
5998 		break;
5999 	case 4:
6000 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6001 		break;
6002 	case 8:
6003 		res = kvm_set_cr8(vcpu, val);
6004 		break;
6005 	default:
6006 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6007 		res = -1;
6008 	}
6009 
6010 	return res;
6011 }
6012 
6013 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6014 {
6015 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6016 }
6017 
6018 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6019 {
6020 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6021 }
6022 
6023 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6024 {
6025 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6026 }
6027 
6028 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6029 {
6030 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6031 }
6032 
6033 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6034 {
6035 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6036 }
6037 
6038 static unsigned long emulator_get_cached_segment_base(
6039 	struct x86_emulate_ctxt *ctxt, int seg)
6040 {
6041 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6042 }
6043 
6044 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6045 				 struct desc_struct *desc, u32 *base3,
6046 				 int seg)
6047 {
6048 	struct kvm_segment var;
6049 
6050 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6051 	*selector = var.selector;
6052 
6053 	if (var.unusable) {
6054 		memset(desc, 0, sizeof(*desc));
6055 		if (base3)
6056 			*base3 = 0;
6057 		return false;
6058 	}
6059 
6060 	if (var.g)
6061 		var.limit >>= 12;
6062 	set_desc_limit(desc, var.limit);
6063 	set_desc_base(desc, (unsigned long)var.base);
6064 #ifdef CONFIG_X86_64
6065 	if (base3)
6066 		*base3 = var.base >> 32;
6067 #endif
6068 	desc->type = var.type;
6069 	desc->s = var.s;
6070 	desc->dpl = var.dpl;
6071 	desc->p = var.present;
6072 	desc->avl = var.avl;
6073 	desc->l = var.l;
6074 	desc->d = var.db;
6075 	desc->g = var.g;
6076 
6077 	return true;
6078 }
6079 
6080 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6081 				 struct desc_struct *desc, u32 base3,
6082 				 int seg)
6083 {
6084 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6085 	struct kvm_segment var;
6086 
6087 	var.selector = selector;
6088 	var.base = get_desc_base(desc);
6089 #ifdef CONFIG_X86_64
6090 	var.base |= ((u64)base3) << 32;
6091 #endif
6092 	var.limit = get_desc_limit(desc);
6093 	if (desc->g)
6094 		var.limit = (var.limit << 12) | 0xfff;
6095 	var.type = desc->type;
6096 	var.dpl = desc->dpl;
6097 	var.db = desc->d;
6098 	var.s = desc->s;
6099 	var.l = desc->l;
6100 	var.g = desc->g;
6101 	var.avl = desc->avl;
6102 	var.present = desc->p;
6103 	var.unusable = !var.present;
6104 	var.padding = 0;
6105 
6106 	kvm_set_segment(vcpu, &var, seg);
6107 	return;
6108 }
6109 
6110 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6111 			    u32 msr_index, u64 *pdata)
6112 {
6113 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6114 }
6115 
6116 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6117 			    u32 msr_index, u64 data)
6118 {
6119 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6120 }
6121 
6122 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6123 {
6124 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6125 
6126 	return vcpu->arch.smbase;
6127 }
6128 
6129 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6130 {
6131 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6132 
6133 	vcpu->arch.smbase = smbase;
6134 }
6135 
6136 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6137 			      u32 pmc)
6138 {
6139 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6140 }
6141 
6142 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6143 			     u32 pmc, u64 *pdata)
6144 {
6145 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6146 }
6147 
6148 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6149 {
6150 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6151 }
6152 
6153 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6154 			      struct x86_instruction_info *info,
6155 			      enum x86_intercept_stage stage)
6156 {
6157 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6158 }
6159 
6160 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6161 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6162 {
6163 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6164 }
6165 
6166 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6167 {
6168 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6169 }
6170 
6171 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6172 {
6173 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6174 }
6175 
6176 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6177 {
6178 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6179 }
6180 
6181 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6182 {
6183 	return emul_to_vcpu(ctxt)->arch.hflags;
6184 }
6185 
6186 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6187 {
6188 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6189 }
6190 
6191 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6192 				  const char *smstate)
6193 {
6194 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6195 }
6196 
6197 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6198 {
6199 	kvm_smm_changed(emul_to_vcpu(ctxt));
6200 }
6201 
6202 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6203 {
6204 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6205 }
6206 
6207 static const struct x86_emulate_ops emulate_ops = {
6208 	.read_gpr            = emulator_read_gpr,
6209 	.write_gpr           = emulator_write_gpr,
6210 	.read_std            = emulator_read_std,
6211 	.write_std           = emulator_write_std,
6212 	.read_phys           = kvm_read_guest_phys_system,
6213 	.fetch               = kvm_fetch_guest_virt,
6214 	.read_emulated       = emulator_read_emulated,
6215 	.write_emulated      = emulator_write_emulated,
6216 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6217 	.invlpg              = emulator_invlpg,
6218 	.pio_in_emulated     = emulator_pio_in_emulated,
6219 	.pio_out_emulated    = emulator_pio_out_emulated,
6220 	.get_segment         = emulator_get_segment,
6221 	.set_segment         = emulator_set_segment,
6222 	.get_cached_segment_base = emulator_get_cached_segment_base,
6223 	.get_gdt             = emulator_get_gdt,
6224 	.get_idt	     = emulator_get_idt,
6225 	.set_gdt             = emulator_set_gdt,
6226 	.set_idt	     = emulator_set_idt,
6227 	.get_cr              = emulator_get_cr,
6228 	.set_cr              = emulator_set_cr,
6229 	.cpl                 = emulator_get_cpl,
6230 	.get_dr              = emulator_get_dr,
6231 	.set_dr              = emulator_set_dr,
6232 	.get_smbase          = emulator_get_smbase,
6233 	.set_smbase          = emulator_set_smbase,
6234 	.set_msr             = emulator_set_msr,
6235 	.get_msr             = emulator_get_msr,
6236 	.check_pmc	     = emulator_check_pmc,
6237 	.read_pmc            = emulator_read_pmc,
6238 	.halt                = emulator_halt,
6239 	.wbinvd              = emulator_wbinvd,
6240 	.fix_hypercall       = emulator_fix_hypercall,
6241 	.intercept           = emulator_intercept,
6242 	.get_cpuid           = emulator_get_cpuid,
6243 	.set_nmi_mask        = emulator_set_nmi_mask,
6244 	.get_hflags          = emulator_get_hflags,
6245 	.set_hflags          = emulator_set_hflags,
6246 	.pre_leave_smm       = emulator_pre_leave_smm,
6247 	.post_leave_smm      = emulator_post_leave_smm,
6248 	.set_xcr             = emulator_set_xcr,
6249 };
6250 
6251 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6252 {
6253 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6254 	/*
6255 	 * an sti; sti; sequence only disable interrupts for the first
6256 	 * instruction. So, if the last instruction, be it emulated or
6257 	 * not, left the system with the INT_STI flag enabled, it
6258 	 * means that the last instruction is an sti. We should not
6259 	 * leave the flag on in this case. The same goes for mov ss
6260 	 */
6261 	if (int_shadow & mask)
6262 		mask = 0;
6263 	if (unlikely(int_shadow || mask)) {
6264 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6265 		if (!mask)
6266 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6267 	}
6268 }
6269 
6270 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6271 {
6272 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6273 	if (ctxt->exception.vector == PF_VECTOR)
6274 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6275 
6276 	if (ctxt->exception.error_code_valid)
6277 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6278 				      ctxt->exception.error_code);
6279 	else
6280 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6281 	return false;
6282 }
6283 
6284 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6285 {
6286 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6287 	int cs_db, cs_l;
6288 
6289 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6290 
6291 	ctxt->eflags = kvm_get_rflags(vcpu);
6292 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6293 
6294 	ctxt->eip = kvm_rip_read(vcpu);
6295 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6296 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6297 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6298 		     cs_db				? X86EMUL_MODE_PROT32 :
6299 							  X86EMUL_MODE_PROT16;
6300 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6301 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6302 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6303 
6304 	init_decode_cache(ctxt);
6305 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6306 }
6307 
6308 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6309 {
6310 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6311 	int ret;
6312 
6313 	init_emulate_ctxt(vcpu);
6314 
6315 	ctxt->op_bytes = 2;
6316 	ctxt->ad_bytes = 2;
6317 	ctxt->_eip = ctxt->eip + inc_eip;
6318 	ret = emulate_int_real(ctxt, irq);
6319 
6320 	if (ret != X86EMUL_CONTINUE) {
6321 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6322 	} else {
6323 		ctxt->eip = ctxt->_eip;
6324 		kvm_rip_write(vcpu, ctxt->eip);
6325 		kvm_set_rflags(vcpu, ctxt->eflags);
6326 	}
6327 }
6328 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6329 
6330 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6331 {
6332 	++vcpu->stat.insn_emulation_fail;
6333 	trace_kvm_emulate_insn_failed(vcpu);
6334 
6335 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6336 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6337 		return 1;
6338 	}
6339 
6340 	if (emulation_type & EMULTYPE_SKIP) {
6341 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6342 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6343 		vcpu->run->internal.ndata = 0;
6344 		return 0;
6345 	}
6346 
6347 	kvm_queue_exception(vcpu, UD_VECTOR);
6348 
6349 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6350 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6351 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6352 		vcpu->run->internal.ndata = 0;
6353 		return 0;
6354 	}
6355 
6356 	return 1;
6357 }
6358 
6359 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6360 				  bool write_fault_to_shadow_pgtable,
6361 				  int emulation_type)
6362 {
6363 	gpa_t gpa = cr2;
6364 	kvm_pfn_t pfn;
6365 
6366 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6367 		return false;
6368 
6369 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6370 		return false;
6371 
6372 	if (!vcpu->arch.mmu->direct_map) {
6373 		/*
6374 		 * Write permission should be allowed since only
6375 		 * write access need to be emulated.
6376 		 */
6377 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6378 
6379 		/*
6380 		 * If the mapping is invalid in guest, let cpu retry
6381 		 * it to generate fault.
6382 		 */
6383 		if (gpa == UNMAPPED_GVA)
6384 			return true;
6385 	}
6386 
6387 	/*
6388 	 * Do not retry the unhandleable instruction if it faults on the
6389 	 * readonly host memory, otherwise it will goto a infinite loop:
6390 	 * retry instruction -> write #PF -> emulation fail -> retry
6391 	 * instruction -> ...
6392 	 */
6393 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6394 
6395 	/*
6396 	 * If the instruction failed on the error pfn, it can not be fixed,
6397 	 * report the error to userspace.
6398 	 */
6399 	if (is_error_noslot_pfn(pfn))
6400 		return false;
6401 
6402 	kvm_release_pfn_clean(pfn);
6403 
6404 	/* The instructions are well-emulated on direct mmu. */
6405 	if (vcpu->arch.mmu->direct_map) {
6406 		unsigned int indirect_shadow_pages;
6407 
6408 		spin_lock(&vcpu->kvm->mmu_lock);
6409 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6410 		spin_unlock(&vcpu->kvm->mmu_lock);
6411 
6412 		if (indirect_shadow_pages)
6413 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6414 
6415 		return true;
6416 	}
6417 
6418 	/*
6419 	 * if emulation was due to access to shadowed page table
6420 	 * and it failed try to unshadow page and re-enter the
6421 	 * guest to let CPU execute the instruction.
6422 	 */
6423 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6424 
6425 	/*
6426 	 * If the access faults on its page table, it can not
6427 	 * be fixed by unprotecting shadow page and it should
6428 	 * be reported to userspace.
6429 	 */
6430 	return !write_fault_to_shadow_pgtable;
6431 }
6432 
6433 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6434 			      unsigned long cr2,  int emulation_type)
6435 {
6436 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6437 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6438 
6439 	last_retry_eip = vcpu->arch.last_retry_eip;
6440 	last_retry_addr = vcpu->arch.last_retry_addr;
6441 
6442 	/*
6443 	 * If the emulation is caused by #PF and it is non-page_table
6444 	 * writing instruction, it means the VM-EXIT is caused by shadow
6445 	 * page protected, we can zap the shadow page and retry this
6446 	 * instruction directly.
6447 	 *
6448 	 * Note: if the guest uses a non-page-table modifying instruction
6449 	 * on the PDE that points to the instruction, then we will unmap
6450 	 * the instruction and go to an infinite loop. So, we cache the
6451 	 * last retried eip and the last fault address, if we meet the eip
6452 	 * and the address again, we can break out of the potential infinite
6453 	 * loop.
6454 	 */
6455 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6456 
6457 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6458 		return false;
6459 
6460 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6461 		return false;
6462 
6463 	if (x86_page_table_writing_insn(ctxt))
6464 		return false;
6465 
6466 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6467 		return false;
6468 
6469 	vcpu->arch.last_retry_eip = ctxt->eip;
6470 	vcpu->arch.last_retry_addr = cr2;
6471 
6472 	if (!vcpu->arch.mmu->direct_map)
6473 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6474 
6475 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6476 
6477 	return true;
6478 }
6479 
6480 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6481 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6482 
6483 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6484 {
6485 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6486 		/* This is a good place to trace that we are exiting SMM.  */
6487 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6488 
6489 		/* Process a latched INIT or SMI, if any.  */
6490 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6491 	}
6492 
6493 	kvm_mmu_reset_context(vcpu);
6494 }
6495 
6496 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6497 				unsigned long *db)
6498 {
6499 	u32 dr6 = 0;
6500 	int i;
6501 	u32 enable, rwlen;
6502 
6503 	enable = dr7;
6504 	rwlen = dr7 >> 16;
6505 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6506 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6507 			dr6 |= (1 << i);
6508 	return dr6;
6509 }
6510 
6511 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6512 {
6513 	struct kvm_run *kvm_run = vcpu->run;
6514 
6515 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6516 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6517 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6518 		kvm_run->debug.arch.exception = DB_VECTOR;
6519 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6520 		return 0;
6521 	}
6522 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6523 	return 1;
6524 }
6525 
6526 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6527 {
6528 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6529 	int r;
6530 
6531 	r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6532 	if (unlikely(!r))
6533 		return 0;
6534 
6535 	/*
6536 	 * rflags is the old, "raw" value of the flags.  The new value has
6537 	 * not been saved yet.
6538 	 *
6539 	 * This is correct even for TF set by the guest, because "the
6540 	 * processor will not generate this exception after the instruction
6541 	 * that sets the TF flag".
6542 	 */
6543 	if (unlikely(rflags & X86_EFLAGS_TF))
6544 		r = kvm_vcpu_do_singlestep(vcpu);
6545 	return r;
6546 }
6547 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6548 
6549 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6550 {
6551 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6552 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6553 		struct kvm_run *kvm_run = vcpu->run;
6554 		unsigned long eip = kvm_get_linear_rip(vcpu);
6555 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6556 					   vcpu->arch.guest_debug_dr7,
6557 					   vcpu->arch.eff_db);
6558 
6559 		if (dr6 != 0) {
6560 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6561 			kvm_run->debug.arch.pc = eip;
6562 			kvm_run->debug.arch.exception = DB_VECTOR;
6563 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6564 			*r = 0;
6565 			return true;
6566 		}
6567 	}
6568 
6569 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6570 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6571 		unsigned long eip = kvm_get_linear_rip(vcpu);
6572 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6573 					   vcpu->arch.dr7,
6574 					   vcpu->arch.db);
6575 
6576 		if (dr6 != 0) {
6577 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6578 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6579 			kvm_queue_exception(vcpu, DB_VECTOR);
6580 			*r = 1;
6581 			return true;
6582 		}
6583 	}
6584 
6585 	return false;
6586 }
6587 
6588 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6589 {
6590 	switch (ctxt->opcode_len) {
6591 	case 1:
6592 		switch (ctxt->b) {
6593 		case 0xe4:	/* IN */
6594 		case 0xe5:
6595 		case 0xec:
6596 		case 0xed:
6597 		case 0xe6:	/* OUT */
6598 		case 0xe7:
6599 		case 0xee:
6600 		case 0xef:
6601 		case 0x6c:	/* INS */
6602 		case 0x6d:
6603 		case 0x6e:	/* OUTS */
6604 		case 0x6f:
6605 			return true;
6606 		}
6607 		break;
6608 	case 2:
6609 		switch (ctxt->b) {
6610 		case 0x33:	/* RDPMC */
6611 			return true;
6612 		}
6613 		break;
6614 	}
6615 
6616 	return false;
6617 }
6618 
6619 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6620 			    unsigned long cr2,
6621 			    int emulation_type,
6622 			    void *insn,
6623 			    int insn_len)
6624 {
6625 	int r;
6626 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6627 	bool writeback = true;
6628 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6629 
6630 	vcpu->arch.l1tf_flush_l1d = true;
6631 
6632 	/*
6633 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6634 	 * never reused.
6635 	 */
6636 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6637 	kvm_clear_exception_queue(vcpu);
6638 
6639 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6640 		init_emulate_ctxt(vcpu);
6641 
6642 		/*
6643 		 * We will reenter on the same instruction since
6644 		 * we do not set complete_userspace_io.  This does not
6645 		 * handle watchpoints yet, those would be handled in
6646 		 * the emulate_ops.
6647 		 */
6648 		if (!(emulation_type & EMULTYPE_SKIP) &&
6649 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6650 			return r;
6651 
6652 		ctxt->interruptibility = 0;
6653 		ctxt->have_exception = false;
6654 		ctxt->exception.vector = -1;
6655 		ctxt->perm_ok = false;
6656 
6657 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6658 
6659 		r = x86_decode_insn(ctxt, insn, insn_len);
6660 
6661 		trace_kvm_emulate_insn_start(vcpu);
6662 		++vcpu->stat.insn_emulation;
6663 		if (r != EMULATION_OK)  {
6664 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6665 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6666 				kvm_queue_exception(vcpu, UD_VECTOR);
6667 				return 1;
6668 			}
6669 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6670 						emulation_type))
6671 				return 1;
6672 			if (ctxt->have_exception) {
6673 				/*
6674 				 * #UD should result in just EMULATION_FAILED, and trap-like
6675 				 * exception should not be encountered during decode.
6676 				 */
6677 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6678 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6679 				inject_emulated_exception(vcpu);
6680 				return 1;
6681 			}
6682 			return handle_emulation_failure(vcpu, emulation_type);
6683 		}
6684 	}
6685 
6686 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6687 	    !is_vmware_backdoor_opcode(ctxt)) {
6688 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6689 		return 1;
6690 	}
6691 
6692 	/*
6693 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6694 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6695 	 * updating interruptibility state and injecting single-step #DBs.
6696 	 */
6697 	if (emulation_type & EMULTYPE_SKIP) {
6698 		kvm_rip_write(vcpu, ctxt->_eip);
6699 		if (ctxt->eflags & X86_EFLAGS_RF)
6700 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6701 		return 1;
6702 	}
6703 
6704 	if (retry_instruction(ctxt, cr2, emulation_type))
6705 		return 1;
6706 
6707 	/* this is needed for vmware backdoor interface to work since it
6708 	   changes registers values  during IO operation */
6709 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6710 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6711 		emulator_invalidate_register_cache(ctxt);
6712 	}
6713 
6714 restart:
6715 	/* Save the faulting GPA (cr2) in the address field */
6716 	ctxt->exception.address = cr2;
6717 
6718 	r = x86_emulate_insn(ctxt);
6719 
6720 	if (r == EMULATION_INTERCEPTED)
6721 		return 1;
6722 
6723 	if (r == EMULATION_FAILED) {
6724 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6725 					emulation_type))
6726 			return 1;
6727 
6728 		return handle_emulation_failure(vcpu, emulation_type);
6729 	}
6730 
6731 	if (ctxt->have_exception) {
6732 		r = 1;
6733 		if (inject_emulated_exception(vcpu))
6734 			return r;
6735 	} else if (vcpu->arch.pio.count) {
6736 		if (!vcpu->arch.pio.in) {
6737 			/* FIXME: return into emulator if single-stepping.  */
6738 			vcpu->arch.pio.count = 0;
6739 		} else {
6740 			writeback = false;
6741 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6742 		}
6743 		r = 0;
6744 	} else if (vcpu->mmio_needed) {
6745 		++vcpu->stat.mmio_exits;
6746 
6747 		if (!vcpu->mmio_is_write)
6748 			writeback = false;
6749 		r = 0;
6750 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6751 	} else if (r == EMULATION_RESTART)
6752 		goto restart;
6753 	else
6754 		r = 1;
6755 
6756 	if (writeback) {
6757 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6758 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6759 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6760 		if (!ctxt->have_exception ||
6761 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6762 			kvm_rip_write(vcpu, ctxt->eip);
6763 			if (r && ctxt->tf)
6764 				r = kvm_vcpu_do_singlestep(vcpu);
6765 			__kvm_set_rflags(vcpu, ctxt->eflags);
6766 		}
6767 
6768 		/*
6769 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6770 		 * do nothing, and it will be requested again as soon as
6771 		 * the shadow expires.  But we still need to check here,
6772 		 * because POPF has no interrupt shadow.
6773 		 */
6774 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6775 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6776 	} else
6777 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6778 
6779 	return r;
6780 }
6781 
6782 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6783 {
6784 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6785 }
6786 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6787 
6788 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6789 					void *insn, int insn_len)
6790 {
6791 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6792 }
6793 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6794 
6795 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6796 {
6797 	vcpu->arch.pio.count = 0;
6798 	return 1;
6799 }
6800 
6801 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6802 {
6803 	vcpu->arch.pio.count = 0;
6804 
6805 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6806 		return 1;
6807 
6808 	return kvm_skip_emulated_instruction(vcpu);
6809 }
6810 
6811 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6812 			    unsigned short port)
6813 {
6814 	unsigned long val = kvm_rax_read(vcpu);
6815 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6816 					    size, port, &val, 1);
6817 	if (ret)
6818 		return ret;
6819 
6820 	/*
6821 	 * Workaround userspace that relies on old KVM behavior of %rip being
6822 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6823 	 */
6824 	if (port == 0x7e &&
6825 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6826 		vcpu->arch.complete_userspace_io =
6827 			complete_fast_pio_out_port_0x7e;
6828 		kvm_skip_emulated_instruction(vcpu);
6829 	} else {
6830 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6831 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6832 	}
6833 	return 0;
6834 }
6835 
6836 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6837 {
6838 	unsigned long val;
6839 
6840 	/* We should only ever be called with arch.pio.count equal to 1 */
6841 	BUG_ON(vcpu->arch.pio.count != 1);
6842 
6843 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6844 		vcpu->arch.pio.count = 0;
6845 		return 1;
6846 	}
6847 
6848 	/* For size less than 4 we merge, else we zero extend */
6849 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6850 
6851 	/*
6852 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6853 	 * the copy and tracing
6854 	 */
6855 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6856 				 vcpu->arch.pio.port, &val, 1);
6857 	kvm_rax_write(vcpu, val);
6858 
6859 	return kvm_skip_emulated_instruction(vcpu);
6860 }
6861 
6862 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6863 			   unsigned short port)
6864 {
6865 	unsigned long val;
6866 	int ret;
6867 
6868 	/* For size less than 4 we merge, else we zero extend */
6869 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6870 
6871 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6872 				       &val, 1);
6873 	if (ret) {
6874 		kvm_rax_write(vcpu, val);
6875 		return ret;
6876 	}
6877 
6878 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6879 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6880 
6881 	return 0;
6882 }
6883 
6884 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6885 {
6886 	int ret;
6887 
6888 	if (in)
6889 		ret = kvm_fast_pio_in(vcpu, size, port);
6890 	else
6891 		ret = kvm_fast_pio_out(vcpu, size, port);
6892 	return ret && kvm_skip_emulated_instruction(vcpu);
6893 }
6894 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6895 
6896 static int kvmclock_cpu_down_prep(unsigned int cpu)
6897 {
6898 	__this_cpu_write(cpu_tsc_khz, 0);
6899 	return 0;
6900 }
6901 
6902 static void tsc_khz_changed(void *data)
6903 {
6904 	struct cpufreq_freqs *freq = data;
6905 	unsigned long khz = 0;
6906 
6907 	if (data)
6908 		khz = freq->new;
6909 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6910 		khz = cpufreq_quick_get(raw_smp_processor_id());
6911 	if (!khz)
6912 		khz = tsc_khz;
6913 	__this_cpu_write(cpu_tsc_khz, khz);
6914 }
6915 
6916 #ifdef CONFIG_X86_64
6917 static void kvm_hyperv_tsc_notifier(void)
6918 {
6919 	struct kvm *kvm;
6920 	struct kvm_vcpu *vcpu;
6921 	int cpu;
6922 
6923 	mutex_lock(&kvm_lock);
6924 	list_for_each_entry(kvm, &vm_list, vm_list)
6925 		kvm_make_mclock_inprogress_request(kvm);
6926 
6927 	hyperv_stop_tsc_emulation();
6928 
6929 	/* TSC frequency always matches when on Hyper-V */
6930 	for_each_present_cpu(cpu)
6931 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6932 	kvm_max_guest_tsc_khz = tsc_khz;
6933 
6934 	list_for_each_entry(kvm, &vm_list, vm_list) {
6935 		struct kvm_arch *ka = &kvm->arch;
6936 
6937 		spin_lock(&ka->pvclock_gtod_sync_lock);
6938 
6939 		pvclock_update_vm_gtod_copy(kvm);
6940 
6941 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6942 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6943 
6944 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6945 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6946 
6947 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6948 	}
6949 	mutex_unlock(&kvm_lock);
6950 }
6951 #endif
6952 
6953 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6954 {
6955 	struct kvm *kvm;
6956 	struct kvm_vcpu *vcpu;
6957 	int i, send_ipi = 0;
6958 
6959 	/*
6960 	 * We allow guests to temporarily run on slowing clocks,
6961 	 * provided we notify them after, or to run on accelerating
6962 	 * clocks, provided we notify them before.  Thus time never
6963 	 * goes backwards.
6964 	 *
6965 	 * However, we have a problem.  We can't atomically update
6966 	 * the frequency of a given CPU from this function; it is
6967 	 * merely a notifier, which can be called from any CPU.
6968 	 * Changing the TSC frequency at arbitrary points in time
6969 	 * requires a recomputation of local variables related to
6970 	 * the TSC for each VCPU.  We must flag these local variables
6971 	 * to be updated and be sure the update takes place with the
6972 	 * new frequency before any guests proceed.
6973 	 *
6974 	 * Unfortunately, the combination of hotplug CPU and frequency
6975 	 * change creates an intractable locking scenario; the order
6976 	 * of when these callouts happen is undefined with respect to
6977 	 * CPU hotplug, and they can race with each other.  As such,
6978 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6979 	 * undefined; you can actually have a CPU frequency change take
6980 	 * place in between the computation of X and the setting of the
6981 	 * variable.  To protect against this problem, all updates of
6982 	 * the per_cpu tsc_khz variable are done in an interrupt
6983 	 * protected IPI, and all callers wishing to update the value
6984 	 * must wait for a synchronous IPI to complete (which is trivial
6985 	 * if the caller is on the CPU already).  This establishes the
6986 	 * necessary total order on variable updates.
6987 	 *
6988 	 * Note that because a guest time update may take place
6989 	 * anytime after the setting of the VCPU's request bit, the
6990 	 * correct TSC value must be set before the request.  However,
6991 	 * to ensure the update actually makes it to any guest which
6992 	 * starts running in hardware virtualization between the set
6993 	 * and the acquisition of the spinlock, we must also ping the
6994 	 * CPU after setting the request bit.
6995 	 *
6996 	 */
6997 
6998 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6999 
7000 	mutex_lock(&kvm_lock);
7001 	list_for_each_entry(kvm, &vm_list, vm_list) {
7002 		kvm_for_each_vcpu(i, vcpu, kvm) {
7003 			if (vcpu->cpu != cpu)
7004 				continue;
7005 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7006 			if (vcpu->cpu != raw_smp_processor_id())
7007 				send_ipi = 1;
7008 		}
7009 	}
7010 	mutex_unlock(&kvm_lock);
7011 
7012 	if (freq->old < freq->new && send_ipi) {
7013 		/*
7014 		 * We upscale the frequency.  Must make the guest
7015 		 * doesn't see old kvmclock values while running with
7016 		 * the new frequency, otherwise we risk the guest sees
7017 		 * time go backwards.
7018 		 *
7019 		 * In case we update the frequency for another cpu
7020 		 * (which might be in guest context) send an interrupt
7021 		 * to kick the cpu out of guest context.  Next time
7022 		 * guest context is entered kvmclock will be updated,
7023 		 * so the guest will not see stale values.
7024 		 */
7025 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7026 	}
7027 }
7028 
7029 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7030 				     void *data)
7031 {
7032 	struct cpufreq_freqs *freq = data;
7033 	int cpu;
7034 
7035 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7036 		return 0;
7037 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7038 		return 0;
7039 
7040 	for_each_cpu(cpu, freq->policy->cpus)
7041 		__kvmclock_cpufreq_notifier(freq, cpu);
7042 
7043 	return 0;
7044 }
7045 
7046 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7047 	.notifier_call  = kvmclock_cpufreq_notifier
7048 };
7049 
7050 static int kvmclock_cpu_online(unsigned int cpu)
7051 {
7052 	tsc_khz_changed(NULL);
7053 	return 0;
7054 }
7055 
7056 static void kvm_timer_init(void)
7057 {
7058 	max_tsc_khz = tsc_khz;
7059 
7060 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7061 #ifdef CONFIG_CPU_FREQ
7062 		struct cpufreq_policy policy;
7063 		int cpu;
7064 
7065 		memset(&policy, 0, sizeof(policy));
7066 		cpu = get_cpu();
7067 		cpufreq_get_policy(&policy, cpu);
7068 		if (policy.cpuinfo.max_freq)
7069 			max_tsc_khz = policy.cpuinfo.max_freq;
7070 		put_cpu();
7071 #endif
7072 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7073 					  CPUFREQ_TRANSITION_NOTIFIER);
7074 	}
7075 
7076 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7077 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7078 }
7079 
7080 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7081 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7082 
7083 int kvm_is_in_guest(void)
7084 {
7085 	return __this_cpu_read(current_vcpu) != NULL;
7086 }
7087 
7088 static int kvm_is_user_mode(void)
7089 {
7090 	int user_mode = 3;
7091 
7092 	if (__this_cpu_read(current_vcpu))
7093 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7094 
7095 	return user_mode != 0;
7096 }
7097 
7098 static unsigned long kvm_get_guest_ip(void)
7099 {
7100 	unsigned long ip = 0;
7101 
7102 	if (__this_cpu_read(current_vcpu))
7103 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7104 
7105 	return ip;
7106 }
7107 
7108 static void kvm_handle_intel_pt_intr(void)
7109 {
7110 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7111 
7112 	kvm_make_request(KVM_REQ_PMI, vcpu);
7113 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7114 			(unsigned long *)&vcpu->arch.pmu.global_status);
7115 }
7116 
7117 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7118 	.is_in_guest		= kvm_is_in_guest,
7119 	.is_user_mode		= kvm_is_user_mode,
7120 	.get_guest_ip		= kvm_get_guest_ip,
7121 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7122 };
7123 
7124 #ifdef CONFIG_X86_64
7125 static void pvclock_gtod_update_fn(struct work_struct *work)
7126 {
7127 	struct kvm *kvm;
7128 
7129 	struct kvm_vcpu *vcpu;
7130 	int i;
7131 
7132 	mutex_lock(&kvm_lock);
7133 	list_for_each_entry(kvm, &vm_list, vm_list)
7134 		kvm_for_each_vcpu(i, vcpu, kvm)
7135 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7136 	atomic_set(&kvm_guest_has_master_clock, 0);
7137 	mutex_unlock(&kvm_lock);
7138 }
7139 
7140 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7141 
7142 /*
7143  * Notification about pvclock gtod data update.
7144  */
7145 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7146 			       void *priv)
7147 {
7148 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7149 	struct timekeeper *tk = priv;
7150 
7151 	update_pvclock_gtod(tk);
7152 
7153 	/* disable master clock if host does not trust, or does not
7154 	 * use, TSC based clocksource.
7155 	 */
7156 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7157 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7158 		queue_work(system_long_wq, &pvclock_gtod_work);
7159 
7160 	return 0;
7161 }
7162 
7163 static struct notifier_block pvclock_gtod_notifier = {
7164 	.notifier_call = pvclock_gtod_notify,
7165 };
7166 #endif
7167 
7168 int kvm_arch_init(void *opaque)
7169 {
7170 	int r;
7171 	struct kvm_x86_ops *ops = opaque;
7172 
7173 	if (kvm_x86_ops) {
7174 		printk(KERN_ERR "kvm: already loaded the other module\n");
7175 		r = -EEXIST;
7176 		goto out;
7177 	}
7178 
7179 	if (!ops->cpu_has_kvm_support()) {
7180 		printk(KERN_ERR "kvm: no hardware support\n");
7181 		r = -EOPNOTSUPP;
7182 		goto out;
7183 	}
7184 	if (ops->disabled_by_bios()) {
7185 		printk(KERN_ERR "kvm: disabled by bios\n");
7186 		r = -EOPNOTSUPP;
7187 		goto out;
7188 	}
7189 
7190 	/*
7191 	 * KVM explicitly assumes that the guest has an FPU and
7192 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7193 	 * vCPU's FPU state as a fxregs_state struct.
7194 	 */
7195 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7196 		printk(KERN_ERR "kvm: inadequate fpu\n");
7197 		r = -EOPNOTSUPP;
7198 		goto out;
7199 	}
7200 
7201 	r = -ENOMEM;
7202 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7203 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7204 					  NULL);
7205 	if (!x86_fpu_cache) {
7206 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7207 		goto out;
7208 	}
7209 
7210 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7211 	if (!shared_msrs) {
7212 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7213 		goto out_free_x86_fpu_cache;
7214 	}
7215 
7216 	r = kvm_mmu_module_init();
7217 	if (r)
7218 		goto out_free_percpu;
7219 
7220 	kvm_x86_ops = ops;
7221 
7222 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7223 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7224 			PT_PRESENT_MASK, 0, sme_me_mask);
7225 	kvm_timer_init();
7226 
7227 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7228 
7229 	if (boot_cpu_has(X86_FEATURE_XSAVE))
7230 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7231 
7232 	kvm_lapic_init();
7233 	if (pi_inject_timer == -1)
7234 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7235 #ifdef CONFIG_X86_64
7236 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7237 
7238 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7239 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7240 #endif
7241 
7242 	return 0;
7243 
7244 out_free_percpu:
7245 	free_percpu(shared_msrs);
7246 out_free_x86_fpu_cache:
7247 	kmem_cache_destroy(x86_fpu_cache);
7248 out:
7249 	return r;
7250 }
7251 
7252 void kvm_arch_exit(void)
7253 {
7254 #ifdef CONFIG_X86_64
7255 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7256 		clear_hv_tscchange_cb();
7257 #endif
7258 	kvm_lapic_exit();
7259 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7260 
7261 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7262 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7263 					    CPUFREQ_TRANSITION_NOTIFIER);
7264 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7265 #ifdef CONFIG_X86_64
7266 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7267 #endif
7268 	kvm_x86_ops = NULL;
7269 	kvm_mmu_module_exit();
7270 	free_percpu(shared_msrs);
7271 	kmem_cache_destroy(x86_fpu_cache);
7272 }
7273 
7274 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7275 {
7276 	++vcpu->stat.halt_exits;
7277 	if (lapic_in_kernel(vcpu)) {
7278 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7279 		return 1;
7280 	} else {
7281 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7282 		return 0;
7283 	}
7284 }
7285 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7286 
7287 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7288 {
7289 	int ret = kvm_skip_emulated_instruction(vcpu);
7290 	/*
7291 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7292 	 * KVM_EXIT_DEBUG here.
7293 	 */
7294 	return kvm_vcpu_halt(vcpu) && ret;
7295 }
7296 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7297 
7298 #ifdef CONFIG_X86_64
7299 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7300 			        unsigned long clock_type)
7301 {
7302 	struct kvm_clock_pairing clock_pairing;
7303 	struct timespec64 ts;
7304 	u64 cycle;
7305 	int ret;
7306 
7307 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7308 		return -KVM_EOPNOTSUPP;
7309 
7310 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7311 		return -KVM_EOPNOTSUPP;
7312 
7313 	clock_pairing.sec = ts.tv_sec;
7314 	clock_pairing.nsec = ts.tv_nsec;
7315 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7316 	clock_pairing.flags = 0;
7317 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7318 
7319 	ret = 0;
7320 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7321 			    sizeof(struct kvm_clock_pairing)))
7322 		ret = -KVM_EFAULT;
7323 
7324 	return ret;
7325 }
7326 #endif
7327 
7328 /*
7329  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7330  *
7331  * @apicid - apicid of vcpu to be kicked.
7332  */
7333 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7334 {
7335 	struct kvm_lapic_irq lapic_irq;
7336 
7337 	lapic_irq.shorthand = 0;
7338 	lapic_irq.dest_mode = 0;
7339 	lapic_irq.level = 0;
7340 	lapic_irq.dest_id = apicid;
7341 	lapic_irq.msi_redir_hint = false;
7342 
7343 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7344 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7345 }
7346 
7347 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7348 {
7349 	if (!lapic_in_kernel(vcpu)) {
7350 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7351 		return;
7352 	}
7353 	if (!vcpu->arch.apicv_active)
7354 		return;
7355 
7356 	vcpu->arch.apicv_active = false;
7357 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7358 }
7359 
7360 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7361 {
7362 	struct kvm_vcpu *target = NULL;
7363 	struct kvm_apic_map *map;
7364 
7365 	rcu_read_lock();
7366 	map = rcu_dereference(kvm->arch.apic_map);
7367 
7368 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7369 		target = map->phys_map[dest_id]->vcpu;
7370 
7371 	rcu_read_unlock();
7372 
7373 	if (target && READ_ONCE(target->ready))
7374 		kvm_vcpu_yield_to(target);
7375 }
7376 
7377 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7378 {
7379 	unsigned long nr, a0, a1, a2, a3, ret;
7380 	int op_64_bit;
7381 
7382 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7383 		return kvm_hv_hypercall(vcpu);
7384 
7385 	nr = kvm_rax_read(vcpu);
7386 	a0 = kvm_rbx_read(vcpu);
7387 	a1 = kvm_rcx_read(vcpu);
7388 	a2 = kvm_rdx_read(vcpu);
7389 	a3 = kvm_rsi_read(vcpu);
7390 
7391 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7392 
7393 	op_64_bit = is_64_bit_mode(vcpu);
7394 	if (!op_64_bit) {
7395 		nr &= 0xFFFFFFFF;
7396 		a0 &= 0xFFFFFFFF;
7397 		a1 &= 0xFFFFFFFF;
7398 		a2 &= 0xFFFFFFFF;
7399 		a3 &= 0xFFFFFFFF;
7400 	}
7401 
7402 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7403 		ret = -KVM_EPERM;
7404 		goto out;
7405 	}
7406 
7407 	switch (nr) {
7408 	case KVM_HC_VAPIC_POLL_IRQ:
7409 		ret = 0;
7410 		break;
7411 	case KVM_HC_KICK_CPU:
7412 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7413 		kvm_sched_yield(vcpu->kvm, a1);
7414 		ret = 0;
7415 		break;
7416 #ifdef CONFIG_X86_64
7417 	case KVM_HC_CLOCK_PAIRING:
7418 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7419 		break;
7420 #endif
7421 	case KVM_HC_SEND_IPI:
7422 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7423 		break;
7424 	case KVM_HC_SCHED_YIELD:
7425 		kvm_sched_yield(vcpu->kvm, a0);
7426 		ret = 0;
7427 		break;
7428 	default:
7429 		ret = -KVM_ENOSYS;
7430 		break;
7431 	}
7432 out:
7433 	if (!op_64_bit)
7434 		ret = (u32)ret;
7435 	kvm_rax_write(vcpu, ret);
7436 
7437 	++vcpu->stat.hypercalls;
7438 	return kvm_skip_emulated_instruction(vcpu);
7439 }
7440 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7441 
7442 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7443 {
7444 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7445 	char instruction[3];
7446 	unsigned long rip = kvm_rip_read(vcpu);
7447 
7448 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7449 
7450 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7451 		&ctxt->exception);
7452 }
7453 
7454 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7455 {
7456 	return vcpu->run->request_interrupt_window &&
7457 		likely(!pic_in_kernel(vcpu->kvm));
7458 }
7459 
7460 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7461 {
7462 	struct kvm_run *kvm_run = vcpu->run;
7463 
7464 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7465 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7466 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7467 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7468 	kvm_run->ready_for_interrupt_injection =
7469 		pic_in_kernel(vcpu->kvm) ||
7470 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7471 }
7472 
7473 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7474 {
7475 	int max_irr, tpr;
7476 
7477 	if (!kvm_x86_ops->update_cr8_intercept)
7478 		return;
7479 
7480 	if (!lapic_in_kernel(vcpu))
7481 		return;
7482 
7483 	if (vcpu->arch.apicv_active)
7484 		return;
7485 
7486 	if (!vcpu->arch.apic->vapic_addr)
7487 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7488 	else
7489 		max_irr = -1;
7490 
7491 	if (max_irr != -1)
7492 		max_irr >>= 4;
7493 
7494 	tpr = kvm_lapic_get_cr8(vcpu);
7495 
7496 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7497 }
7498 
7499 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7500 {
7501 	int r;
7502 
7503 	/* try to reinject previous events if any */
7504 
7505 	if (vcpu->arch.exception.injected)
7506 		kvm_x86_ops->queue_exception(vcpu);
7507 	/*
7508 	 * Do not inject an NMI or interrupt if there is a pending
7509 	 * exception.  Exceptions and interrupts are recognized at
7510 	 * instruction boundaries, i.e. the start of an instruction.
7511 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7512 	 * NMIs and interrupts, i.e. traps are recognized before an
7513 	 * NMI/interrupt that's pending on the same instruction.
7514 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7515 	 * priority, but are only generated (pended) during instruction
7516 	 * execution, i.e. a pending fault-like exception means the
7517 	 * fault occurred on the *previous* instruction and must be
7518 	 * serviced prior to recognizing any new events in order to
7519 	 * fully complete the previous instruction.
7520 	 */
7521 	else if (!vcpu->arch.exception.pending) {
7522 		if (vcpu->arch.nmi_injected)
7523 			kvm_x86_ops->set_nmi(vcpu);
7524 		else if (vcpu->arch.interrupt.injected)
7525 			kvm_x86_ops->set_irq(vcpu);
7526 	}
7527 
7528 	/*
7529 	 * Call check_nested_events() even if we reinjected a previous event
7530 	 * in order for caller to determine if it should require immediate-exit
7531 	 * from L2 to L1 due to pending L1 events which require exit
7532 	 * from L2 to L1.
7533 	 */
7534 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7535 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7536 		if (r != 0)
7537 			return r;
7538 	}
7539 
7540 	/* try to inject new event if pending */
7541 	if (vcpu->arch.exception.pending) {
7542 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7543 					vcpu->arch.exception.has_error_code,
7544 					vcpu->arch.exception.error_code);
7545 
7546 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7547 		vcpu->arch.exception.pending = false;
7548 		vcpu->arch.exception.injected = true;
7549 
7550 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7551 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7552 					     X86_EFLAGS_RF);
7553 
7554 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7555 			/*
7556 			 * This code assumes that nSVM doesn't use
7557 			 * check_nested_events(). If it does, the
7558 			 * DR6/DR7 changes should happen before L1
7559 			 * gets a #VMEXIT for an intercepted #DB in
7560 			 * L2.  (Under VMX, on the other hand, the
7561 			 * DR6/DR7 changes should not happen in the
7562 			 * event of a VM-exit to L1 for an intercepted
7563 			 * #DB in L2.)
7564 			 */
7565 			kvm_deliver_exception_payload(vcpu);
7566 			if (vcpu->arch.dr7 & DR7_GD) {
7567 				vcpu->arch.dr7 &= ~DR7_GD;
7568 				kvm_update_dr7(vcpu);
7569 			}
7570 		}
7571 
7572 		kvm_x86_ops->queue_exception(vcpu);
7573 	}
7574 
7575 	/* Don't consider new event if we re-injected an event */
7576 	if (kvm_event_needs_reinjection(vcpu))
7577 		return 0;
7578 
7579 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7580 	    kvm_x86_ops->smi_allowed(vcpu)) {
7581 		vcpu->arch.smi_pending = false;
7582 		++vcpu->arch.smi_count;
7583 		enter_smm(vcpu);
7584 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7585 		--vcpu->arch.nmi_pending;
7586 		vcpu->arch.nmi_injected = true;
7587 		kvm_x86_ops->set_nmi(vcpu);
7588 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7589 		/*
7590 		 * Because interrupts can be injected asynchronously, we are
7591 		 * calling check_nested_events again here to avoid a race condition.
7592 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7593 		 * proposal and current concerns.  Perhaps we should be setting
7594 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7595 		 */
7596 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7597 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7598 			if (r != 0)
7599 				return r;
7600 		}
7601 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7602 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7603 					    false);
7604 			kvm_x86_ops->set_irq(vcpu);
7605 		}
7606 	}
7607 
7608 	return 0;
7609 }
7610 
7611 static void process_nmi(struct kvm_vcpu *vcpu)
7612 {
7613 	unsigned limit = 2;
7614 
7615 	/*
7616 	 * x86 is limited to one NMI running, and one NMI pending after it.
7617 	 * If an NMI is already in progress, limit further NMIs to just one.
7618 	 * Otherwise, allow two (and we'll inject the first one immediately).
7619 	 */
7620 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7621 		limit = 1;
7622 
7623 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7624 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7625 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7626 }
7627 
7628 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7629 {
7630 	u32 flags = 0;
7631 	flags |= seg->g       << 23;
7632 	flags |= seg->db      << 22;
7633 	flags |= seg->l       << 21;
7634 	flags |= seg->avl     << 20;
7635 	flags |= seg->present << 15;
7636 	flags |= seg->dpl     << 13;
7637 	flags |= seg->s       << 12;
7638 	flags |= seg->type    << 8;
7639 	return flags;
7640 }
7641 
7642 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7643 {
7644 	struct kvm_segment seg;
7645 	int offset;
7646 
7647 	kvm_get_segment(vcpu, &seg, n);
7648 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7649 
7650 	if (n < 3)
7651 		offset = 0x7f84 + n * 12;
7652 	else
7653 		offset = 0x7f2c + (n - 3) * 12;
7654 
7655 	put_smstate(u32, buf, offset + 8, seg.base);
7656 	put_smstate(u32, buf, offset + 4, seg.limit);
7657 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7658 }
7659 
7660 #ifdef CONFIG_X86_64
7661 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7662 {
7663 	struct kvm_segment seg;
7664 	int offset;
7665 	u16 flags;
7666 
7667 	kvm_get_segment(vcpu, &seg, n);
7668 	offset = 0x7e00 + n * 16;
7669 
7670 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7671 	put_smstate(u16, buf, offset, seg.selector);
7672 	put_smstate(u16, buf, offset + 2, flags);
7673 	put_smstate(u32, buf, offset + 4, seg.limit);
7674 	put_smstate(u64, buf, offset + 8, seg.base);
7675 }
7676 #endif
7677 
7678 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7679 {
7680 	struct desc_ptr dt;
7681 	struct kvm_segment seg;
7682 	unsigned long val;
7683 	int i;
7684 
7685 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7686 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7687 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7688 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7689 
7690 	for (i = 0; i < 8; i++)
7691 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7692 
7693 	kvm_get_dr(vcpu, 6, &val);
7694 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7695 	kvm_get_dr(vcpu, 7, &val);
7696 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7697 
7698 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7699 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7700 	put_smstate(u32, buf, 0x7f64, seg.base);
7701 	put_smstate(u32, buf, 0x7f60, seg.limit);
7702 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7703 
7704 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7705 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7706 	put_smstate(u32, buf, 0x7f80, seg.base);
7707 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7708 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7709 
7710 	kvm_x86_ops->get_gdt(vcpu, &dt);
7711 	put_smstate(u32, buf, 0x7f74, dt.address);
7712 	put_smstate(u32, buf, 0x7f70, dt.size);
7713 
7714 	kvm_x86_ops->get_idt(vcpu, &dt);
7715 	put_smstate(u32, buf, 0x7f58, dt.address);
7716 	put_smstate(u32, buf, 0x7f54, dt.size);
7717 
7718 	for (i = 0; i < 6; i++)
7719 		enter_smm_save_seg_32(vcpu, buf, i);
7720 
7721 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7722 
7723 	/* revision id */
7724 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7725 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7726 }
7727 
7728 #ifdef CONFIG_X86_64
7729 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7730 {
7731 	struct desc_ptr dt;
7732 	struct kvm_segment seg;
7733 	unsigned long val;
7734 	int i;
7735 
7736 	for (i = 0; i < 16; i++)
7737 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7738 
7739 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7740 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7741 
7742 	kvm_get_dr(vcpu, 6, &val);
7743 	put_smstate(u64, buf, 0x7f68, val);
7744 	kvm_get_dr(vcpu, 7, &val);
7745 	put_smstate(u64, buf, 0x7f60, val);
7746 
7747 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7748 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7749 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7750 
7751 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7752 
7753 	/* revision id */
7754 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7755 
7756 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7757 
7758 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7759 	put_smstate(u16, buf, 0x7e90, seg.selector);
7760 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7761 	put_smstate(u32, buf, 0x7e94, seg.limit);
7762 	put_smstate(u64, buf, 0x7e98, seg.base);
7763 
7764 	kvm_x86_ops->get_idt(vcpu, &dt);
7765 	put_smstate(u32, buf, 0x7e84, dt.size);
7766 	put_smstate(u64, buf, 0x7e88, dt.address);
7767 
7768 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7769 	put_smstate(u16, buf, 0x7e70, seg.selector);
7770 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7771 	put_smstate(u32, buf, 0x7e74, seg.limit);
7772 	put_smstate(u64, buf, 0x7e78, seg.base);
7773 
7774 	kvm_x86_ops->get_gdt(vcpu, &dt);
7775 	put_smstate(u32, buf, 0x7e64, dt.size);
7776 	put_smstate(u64, buf, 0x7e68, dt.address);
7777 
7778 	for (i = 0; i < 6; i++)
7779 		enter_smm_save_seg_64(vcpu, buf, i);
7780 }
7781 #endif
7782 
7783 static void enter_smm(struct kvm_vcpu *vcpu)
7784 {
7785 	struct kvm_segment cs, ds;
7786 	struct desc_ptr dt;
7787 	char buf[512];
7788 	u32 cr0;
7789 
7790 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7791 	memset(buf, 0, 512);
7792 #ifdef CONFIG_X86_64
7793 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7794 		enter_smm_save_state_64(vcpu, buf);
7795 	else
7796 #endif
7797 		enter_smm_save_state_32(vcpu, buf);
7798 
7799 	/*
7800 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7801 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7802 	 * the SMM state-save area.
7803 	 */
7804 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7805 
7806 	vcpu->arch.hflags |= HF_SMM_MASK;
7807 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7808 
7809 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7810 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7811 	else
7812 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7813 
7814 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7815 	kvm_rip_write(vcpu, 0x8000);
7816 
7817 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7818 	kvm_x86_ops->set_cr0(vcpu, cr0);
7819 	vcpu->arch.cr0 = cr0;
7820 
7821 	kvm_x86_ops->set_cr4(vcpu, 0);
7822 
7823 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7824 	dt.address = dt.size = 0;
7825 	kvm_x86_ops->set_idt(vcpu, &dt);
7826 
7827 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7828 
7829 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7830 	cs.base = vcpu->arch.smbase;
7831 
7832 	ds.selector = 0;
7833 	ds.base = 0;
7834 
7835 	cs.limit    = ds.limit = 0xffffffff;
7836 	cs.type     = ds.type = 0x3;
7837 	cs.dpl      = ds.dpl = 0;
7838 	cs.db       = ds.db = 0;
7839 	cs.s        = ds.s = 1;
7840 	cs.l        = ds.l = 0;
7841 	cs.g        = ds.g = 1;
7842 	cs.avl      = ds.avl = 0;
7843 	cs.present  = ds.present = 1;
7844 	cs.unusable = ds.unusable = 0;
7845 	cs.padding  = ds.padding = 0;
7846 
7847 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7848 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7849 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7850 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7851 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7852 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7853 
7854 #ifdef CONFIG_X86_64
7855 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7856 		kvm_x86_ops->set_efer(vcpu, 0);
7857 #endif
7858 
7859 	kvm_update_cpuid(vcpu);
7860 	kvm_mmu_reset_context(vcpu);
7861 }
7862 
7863 static void process_smi(struct kvm_vcpu *vcpu)
7864 {
7865 	vcpu->arch.smi_pending = true;
7866 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7867 }
7868 
7869 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7870 {
7871 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7872 }
7873 
7874 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7875 {
7876 	if (!kvm_apic_present(vcpu))
7877 		return;
7878 
7879 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7880 
7881 	if (irqchip_split(vcpu->kvm))
7882 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7883 	else {
7884 		if (vcpu->arch.apicv_active)
7885 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7886 		if (ioapic_in_kernel(vcpu->kvm))
7887 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7888 	}
7889 
7890 	if (is_guest_mode(vcpu))
7891 		vcpu->arch.load_eoi_exitmap_pending = true;
7892 	else
7893 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7894 }
7895 
7896 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7897 {
7898 	u64 eoi_exit_bitmap[4];
7899 
7900 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7901 		return;
7902 
7903 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7904 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7905 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7906 }
7907 
7908 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7909 		unsigned long start, unsigned long end,
7910 		bool blockable)
7911 {
7912 	unsigned long apic_address;
7913 
7914 	/*
7915 	 * The physical address of apic access page is stored in the VMCS.
7916 	 * Update it when it becomes invalid.
7917 	 */
7918 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7919 	if (start <= apic_address && apic_address < end)
7920 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7921 
7922 	return 0;
7923 }
7924 
7925 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7926 {
7927 	struct page *page = NULL;
7928 
7929 	if (!lapic_in_kernel(vcpu))
7930 		return;
7931 
7932 	if (!kvm_x86_ops->set_apic_access_page_addr)
7933 		return;
7934 
7935 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7936 	if (is_error_page(page))
7937 		return;
7938 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7939 
7940 	/*
7941 	 * Do not pin apic access page in memory, the MMU notifier
7942 	 * will call us again if it is migrated or swapped out.
7943 	 */
7944 	put_page(page);
7945 }
7946 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7947 
7948 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7949 {
7950 	smp_send_reschedule(vcpu->cpu);
7951 }
7952 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7953 
7954 /*
7955  * Returns 1 to let vcpu_run() continue the guest execution loop without
7956  * exiting to the userspace.  Otherwise, the value will be returned to the
7957  * userspace.
7958  */
7959 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7960 {
7961 	int r;
7962 	bool req_int_win =
7963 		dm_request_for_irq_injection(vcpu) &&
7964 		kvm_cpu_accept_dm_intr(vcpu);
7965 
7966 	bool req_immediate_exit = false;
7967 
7968 	if (kvm_request_pending(vcpu)) {
7969 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
7970 			if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
7971 				r = 0;
7972 				goto out;
7973 			}
7974 		}
7975 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7976 			kvm_mmu_unload(vcpu);
7977 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7978 			__kvm_migrate_timers(vcpu);
7979 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7980 			kvm_gen_update_masterclock(vcpu->kvm);
7981 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7982 			kvm_gen_kvmclock_update(vcpu);
7983 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7984 			r = kvm_guest_time_update(vcpu);
7985 			if (unlikely(r))
7986 				goto out;
7987 		}
7988 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7989 			kvm_mmu_sync_roots(vcpu);
7990 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7991 			kvm_mmu_load_cr3(vcpu);
7992 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7993 			kvm_vcpu_flush_tlb(vcpu, true);
7994 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7995 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7996 			r = 0;
7997 			goto out;
7998 		}
7999 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8000 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8001 			vcpu->mmio_needed = 0;
8002 			r = 0;
8003 			goto out;
8004 		}
8005 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8006 			/* Page is swapped out. Do synthetic halt */
8007 			vcpu->arch.apf.halted = true;
8008 			r = 1;
8009 			goto out;
8010 		}
8011 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8012 			record_steal_time(vcpu);
8013 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8014 			process_smi(vcpu);
8015 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8016 			process_nmi(vcpu);
8017 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8018 			kvm_pmu_handle_event(vcpu);
8019 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8020 			kvm_pmu_deliver_pmi(vcpu);
8021 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8022 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8023 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8024 				     vcpu->arch.ioapic_handled_vectors)) {
8025 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8026 				vcpu->run->eoi.vector =
8027 						vcpu->arch.pending_ioapic_eoi;
8028 				r = 0;
8029 				goto out;
8030 			}
8031 		}
8032 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8033 			vcpu_scan_ioapic(vcpu);
8034 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8035 			vcpu_load_eoi_exitmap(vcpu);
8036 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8037 			kvm_vcpu_reload_apic_access_page(vcpu);
8038 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8039 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8040 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8041 			r = 0;
8042 			goto out;
8043 		}
8044 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8045 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8046 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8047 			r = 0;
8048 			goto out;
8049 		}
8050 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8051 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8052 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8053 			r = 0;
8054 			goto out;
8055 		}
8056 
8057 		/*
8058 		 * KVM_REQ_HV_STIMER has to be processed after
8059 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8060 		 * depend on the guest clock being up-to-date
8061 		 */
8062 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8063 			kvm_hv_process_stimers(vcpu);
8064 	}
8065 
8066 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8067 		++vcpu->stat.req_event;
8068 		kvm_apic_accept_events(vcpu);
8069 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8070 			r = 1;
8071 			goto out;
8072 		}
8073 
8074 		if (inject_pending_event(vcpu, req_int_win) != 0)
8075 			req_immediate_exit = true;
8076 		else {
8077 			/* Enable SMI/NMI/IRQ window open exits if needed.
8078 			 *
8079 			 * SMIs have three cases:
8080 			 * 1) They can be nested, and then there is nothing to
8081 			 *    do here because RSM will cause a vmexit anyway.
8082 			 * 2) There is an ISA-specific reason why SMI cannot be
8083 			 *    injected, and the moment when this changes can be
8084 			 *    intercepted.
8085 			 * 3) Or the SMI can be pending because
8086 			 *    inject_pending_event has completed the injection
8087 			 *    of an IRQ or NMI from the previous vmexit, and
8088 			 *    then we request an immediate exit to inject the
8089 			 *    SMI.
8090 			 */
8091 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
8092 				if (!kvm_x86_ops->enable_smi_window(vcpu))
8093 					req_immediate_exit = true;
8094 			if (vcpu->arch.nmi_pending)
8095 				kvm_x86_ops->enable_nmi_window(vcpu);
8096 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8097 				kvm_x86_ops->enable_irq_window(vcpu);
8098 			WARN_ON(vcpu->arch.exception.pending);
8099 		}
8100 
8101 		if (kvm_lapic_enabled(vcpu)) {
8102 			update_cr8_intercept(vcpu);
8103 			kvm_lapic_sync_to_vapic(vcpu);
8104 		}
8105 	}
8106 
8107 	r = kvm_mmu_reload(vcpu);
8108 	if (unlikely(r)) {
8109 		goto cancel_injection;
8110 	}
8111 
8112 	preempt_disable();
8113 
8114 	kvm_x86_ops->prepare_guest_switch(vcpu);
8115 
8116 	/*
8117 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8118 	 * IPI are then delayed after guest entry, which ensures that they
8119 	 * result in virtual interrupt delivery.
8120 	 */
8121 	local_irq_disable();
8122 	vcpu->mode = IN_GUEST_MODE;
8123 
8124 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8125 
8126 	/*
8127 	 * 1) We should set ->mode before checking ->requests.  Please see
8128 	 * the comment in kvm_vcpu_exiting_guest_mode().
8129 	 *
8130 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8131 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8132 	 * (see vmx_deliver_posted_interrupt).
8133 	 *
8134 	 * 3) This also orders the write to mode from any reads to the page
8135 	 * tables done while the VCPU is running.  Please see the comment
8136 	 * in kvm_flush_remote_tlbs.
8137 	 */
8138 	smp_mb__after_srcu_read_unlock();
8139 
8140 	/*
8141 	 * This handles the case where a posted interrupt was
8142 	 * notified with kvm_vcpu_kick.
8143 	 */
8144 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8145 		kvm_x86_ops->sync_pir_to_irr(vcpu);
8146 
8147 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8148 	    || need_resched() || signal_pending(current)) {
8149 		vcpu->mode = OUTSIDE_GUEST_MODE;
8150 		smp_wmb();
8151 		local_irq_enable();
8152 		preempt_enable();
8153 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8154 		r = 1;
8155 		goto cancel_injection;
8156 	}
8157 
8158 	if (req_immediate_exit) {
8159 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8160 		kvm_x86_ops->request_immediate_exit(vcpu);
8161 	}
8162 
8163 	trace_kvm_entry(vcpu->vcpu_id);
8164 	guest_enter_irqoff();
8165 
8166 	/* The preempt notifier should have taken care of the FPU already.  */
8167 	WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8168 
8169 	if (unlikely(vcpu->arch.switch_db_regs)) {
8170 		set_debugreg(0, 7);
8171 		set_debugreg(vcpu->arch.eff_db[0], 0);
8172 		set_debugreg(vcpu->arch.eff_db[1], 1);
8173 		set_debugreg(vcpu->arch.eff_db[2], 2);
8174 		set_debugreg(vcpu->arch.eff_db[3], 3);
8175 		set_debugreg(vcpu->arch.dr6, 6);
8176 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8177 	}
8178 
8179 	kvm_x86_ops->run(vcpu);
8180 
8181 	/*
8182 	 * Do this here before restoring debug registers on the host.  And
8183 	 * since we do this before handling the vmexit, a DR access vmexit
8184 	 * can (a) read the correct value of the debug registers, (b) set
8185 	 * KVM_DEBUGREG_WONT_EXIT again.
8186 	 */
8187 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8188 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8189 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8190 		kvm_update_dr0123(vcpu);
8191 		kvm_update_dr6(vcpu);
8192 		kvm_update_dr7(vcpu);
8193 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8194 	}
8195 
8196 	/*
8197 	 * If the guest has used debug registers, at least dr7
8198 	 * will be disabled while returning to the host.
8199 	 * If we don't have active breakpoints in the host, we don't
8200 	 * care about the messed up debug address registers. But if
8201 	 * we have some of them active, restore the old state.
8202 	 */
8203 	if (hw_breakpoint_active())
8204 		hw_breakpoint_restore();
8205 
8206 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8207 
8208 	vcpu->mode = OUTSIDE_GUEST_MODE;
8209 	smp_wmb();
8210 
8211 	kvm_x86_ops->handle_exit_irqoff(vcpu);
8212 
8213 	/*
8214 	 * Consume any pending interrupts, including the possible source of
8215 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8216 	 * An instruction is required after local_irq_enable() to fully unblock
8217 	 * interrupts on processors that implement an interrupt shadow, the
8218 	 * stat.exits increment will do nicely.
8219 	 */
8220 	kvm_before_interrupt(vcpu);
8221 	local_irq_enable();
8222 	++vcpu->stat.exits;
8223 	local_irq_disable();
8224 	kvm_after_interrupt(vcpu);
8225 
8226 	guest_exit_irqoff();
8227 	if (lapic_in_kernel(vcpu)) {
8228 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8229 		if (delta != S64_MIN) {
8230 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8231 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8232 		}
8233 	}
8234 
8235 	local_irq_enable();
8236 	preempt_enable();
8237 
8238 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8239 
8240 	/*
8241 	 * Profile KVM exit RIPs:
8242 	 */
8243 	if (unlikely(prof_on == KVM_PROFILING)) {
8244 		unsigned long rip = kvm_rip_read(vcpu);
8245 		profile_hit(KVM_PROFILING, (void *)rip);
8246 	}
8247 
8248 	if (unlikely(vcpu->arch.tsc_always_catchup))
8249 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8250 
8251 	if (vcpu->arch.apic_attention)
8252 		kvm_lapic_sync_from_vapic(vcpu);
8253 
8254 	vcpu->arch.gpa_available = false;
8255 	r = kvm_x86_ops->handle_exit(vcpu);
8256 	return r;
8257 
8258 cancel_injection:
8259 	kvm_x86_ops->cancel_injection(vcpu);
8260 	if (unlikely(vcpu->arch.apic_attention))
8261 		kvm_lapic_sync_from_vapic(vcpu);
8262 out:
8263 	return r;
8264 }
8265 
8266 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8267 {
8268 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8269 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8270 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8271 		kvm_vcpu_block(vcpu);
8272 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8273 
8274 		if (kvm_x86_ops->post_block)
8275 			kvm_x86_ops->post_block(vcpu);
8276 
8277 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8278 			return 1;
8279 	}
8280 
8281 	kvm_apic_accept_events(vcpu);
8282 	switch(vcpu->arch.mp_state) {
8283 	case KVM_MP_STATE_HALTED:
8284 		vcpu->arch.pv.pv_unhalted = false;
8285 		vcpu->arch.mp_state =
8286 			KVM_MP_STATE_RUNNABLE;
8287 		/* fall through */
8288 	case KVM_MP_STATE_RUNNABLE:
8289 		vcpu->arch.apf.halted = false;
8290 		break;
8291 	case KVM_MP_STATE_INIT_RECEIVED:
8292 		break;
8293 	default:
8294 		return -EINTR;
8295 		break;
8296 	}
8297 	return 1;
8298 }
8299 
8300 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8301 {
8302 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8303 		kvm_x86_ops->check_nested_events(vcpu, false);
8304 
8305 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8306 		!vcpu->arch.apf.halted);
8307 }
8308 
8309 static int vcpu_run(struct kvm_vcpu *vcpu)
8310 {
8311 	int r;
8312 	struct kvm *kvm = vcpu->kvm;
8313 
8314 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8315 	vcpu->arch.l1tf_flush_l1d = true;
8316 
8317 	for (;;) {
8318 		if (kvm_vcpu_running(vcpu)) {
8319 			r = vcpu_enter_guest(vcpu);
8320 		} else {
8321 			r = vcpu_block(kvm, vcpu);
8322 		}
8323 
8324 		if (r <= 0)
8325 			break;
8326 
8327 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8328 		if (kvm_cpu_has_pending_timer(vcpu))
8329 			kvm_inject_pending_timer_irqs(vcpu);
8330 
8331 		if (dm_request_for_irq_injection(vcpu) &&
8332 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8333 			r = 0;
8334 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8335 			++vcpu->stat.request_irq_exits;
8336 			break;
8337 		}
8338 
8339 		kvm_check_async_pf_completion(vcpu);
8340 
8341 		if (signal_pending(current)) {
8342 			r = -EINTR;
8343 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8344 			++vcpu->stat.signal_exits;
8345 			break;
8346 		}
8347 		if (need_resched()) {
8348 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8349 			cond_resched();
8350 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8351 		}
8352 	}
8353 
8354 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8355 
8356 	return r;
8357 }
8358 
8359 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8360 {
8361 	int r;
8362 
8363 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8364 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8365 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8366 	return r;
8367 }
8368 
8369 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8370 {
8371 	BUG_ON(!vcpu->arch.pio.count);
8372 
8373 	return complete_emulated_io(vcpu);
8374 }
8375 
8376 /*
8377  * Implements the following, as a state machine:
8378  *
8379  * read:
8380  *   for each fragment
8381  *     for each mmio piece in the fragment
8382  *       write gpa, len
8383  *       exit
8384  *       copy data
8385  *   execute insn
8386  *
8387  * write:
8388  *   for each fragment
8389  *     for each mmio piece in the fragment
8390  *       write gpa, len
8391  *       copy data
8392  *       exit
8393  */
8394 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8395 {
8396 	struct kvm_run *run = vcpu->run;
8397 	struct kvm_mmio_fragment *frag;
8398 	unsigned len;
8399 
8400 	BUG_ON(!vcpu->mmio_needed);
8401 
8402 	/* Complete previous fragment */
8403 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8404 	len = min(8u, frag->len);
8405 	if (!vcpu->mmio_is_write)
8406 		memcpy(frag->data, run->mmio.data, len);
8407 
8408 	if (frag->len <= 8) {
8409 		/* Switch to the next fragment. */
8410 		frag++;
8411 		vcpu->mmio_cur_fragment++;
8412 	} else {
8413 		/* Go forward to the next mmio piece. */
8414 		frag->data += len;
8415 		frag->gpa += len;
8416 		frag->len -= len;
8417 	}
8418 
8419 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8420 		vcpu->mmio_needed = 0;
8421 
8422 		/* FIXME: return into emulator if single-stepping.  */
8423 		if (vcpu->mmio_is_write)
8424 			return 1;
8425 		vcpu->mmio_read_completed = 1;
8426 		return complete_emulated_io(vcpu);
8427 	}
8428 
8429 	run->exit_reason = KVM_EXIT_MMIO;
8430 	run->mmio.phys_addr = frag->gpa;
8431 	if (vcpu->mmio_is_write)
8432 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8433 	run->mmio.len = min(8u, frag->len);
8434 	run->mmio.is_write = vcpu->mmio_is_write;
8435 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8436 	return 0;
8437 }
8438 
8439 /* Swap (qemu) user FPU context for the guest FPU context. */
8440 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8441 {
8442 	fpregs_lock();
8443 
8444 	copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8445 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8446 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8447 				~XFEATURE_MASK_PKRU);
8448 
8449 	fpregs_mark_activate();
8450 	fpregs_unlock();
8451 
8452 	trace_kvm_fpu(1);
8453 }
8454 
8455 /* When vcpu_run ends, restore user space FPU context. */
8456 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8457 {
8458 	fpregs_lock();
8459 
8460 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8461 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8462 
8463 	fpregs_mark_activate();
8464 	fpregs_unlock();
8465 
8466 	++vcpu->stat.fpu_reload;
8467 	trace_kvm_fpu(0);
8468 }
8469 
8470 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8471 {
8472 	int r;
8473 
8474 	vcpu_load(vcpu);
8475 	kvm_sigset_activate(vcpu);
8476 	kvm_load_guest_fpu(vcpu);
8477 
8478 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8479 		if (kvm_run->immediate_exit) {
8480 			r = -EINTR;
8481 			goto out;
8482 		}
8483 		kvm_vcpu_block(vcpu);
8484 		kvm_apic_accept_events(vcpu);
8485 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8486 		r = -EAGAIN;
8487 		if (signal_pending(current)) {
8488 			r = -EINTR;
8489 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8490 			++vcpu->stat.signal_exits;
8491 		}
8492 		goto out;
8493 	}
8494 
8495 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8496 		r = -EINVAL;
8497 		goto out;
8498 	}
8499 
8500 	if (vcpu->run->kvm_dirty_regs) {
8501 		r = sync_regs(vcpu);
8502 		if (r != 0)
8503 			goto out;
8504 	}
8505 
8506 	/* re-sync apic's tpr */
8507 	if (!lapic_in_kernel(vcpu)) {
8508 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8509 			r = -EINVAL;
8510 			goto out;
8511 		}
8512 	}
8513 
8514 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8515 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8516 		vcpu->arch.complete_userspace_io = NULL;
8517 		r = cui(vcpu);
8518 		if (r <= 0)
8519 			goto out;
8520 	} else
8521 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8522 
8523 	if (kvm_run->immediate_exit)
8524 		r = -EINTR;
8525 	else
8526 		r = vcpu_run(vcpu);
8527 
8528 out:
8529 	kvm_put_guest_fpu(vcpu);
8530 	if (vcpu->run->kvm_valid_regs)
8531 		store_regs(vcpu);
8532 	post_kvm_run_save(vcpu);
8533 	kvm_sigset_deactivate(vcpu);
8534 
8535 	vcpu_put(vcpu);
8536 	return r;
8537 }
8538 
8539 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8540 {
8541 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8542 		/*
8543 		 * We are here if userspace calls get_regs() in the middle of
8544 		 * instruction emulation. Registers state needs to be copied
8545 		 * back from emulation context to vcpu. Userspace shouldn't do
8546 		 * that usually, but some bad designed PV devices (vmware
8547 		 * backdoor interface) need this to work
8548 		 */
8549 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8550 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8551 	}
8552 	regs->rax = kvm_rax_read(vcpu);
8553 	regs->rbx = kvm_rbx_read(vcpu);
8554 	regs->rcx = kvm_rcx_read(vcpu);
8555 	regs->rdx = kvm_rdx_read(vcpu);
8556 	regs->rsi = kvm_rsi_read(vcpu);
8557 	regs->rdi = kvm_rdi_read(vcpu);
8558 	regs->rsp = kvm_rsp_read(vcpu);
8559 	regs->rbp = kvm_rbp_read(vcpu);
8560 #ifdef CONFIG_X86_64
8561 	regs->r8 = kvm_r8_read(vcpu);
8562 	regs->r9 = kvm_r9_read(vcpu);
8563 	regs->r10 = kvm_r10_read(vcpu);
8564 	regs->r11 = kvm_r11_read(vcpu);
8565 	regs->r12 = kvm_r12_read(vcpu);
8566 	regs->r13 = kvm_r13_read(vcpu);
8567 	regs->r14 = kvm_r14_read(vcpu);
8568 	regs->r15 = kvm_r15_read(vcpu);
8569 #endif
8570 
8571 	regs->rip = kvm_rip_read(vcpu);
8572 	regs->rflags = kvm_get_rflags(vcpu);
8573 }
8574 
8575 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8576 {
8577 	vcpu_load(vcpu);
8578 	__get_regs(vcpu, regs);
8579 	vcpu_put(vcpu);
8580 	return 0;
8581 }
8582 
8583 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8584 {
8585 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8586 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8587 
8588 	kvm_rax_write(vcpu, regs->rax);
8589 	kvm_rbx_write(vcpu, regs->rbx);
8590 	kvm_rcx_write(vcpu, regs->rcx);
8591 	kvm_rdx_write(vcpu, regs->rdx);
8592 	kvm_rsi_write(vcpu, regs->rsi);
8593 	kvm_rdi_write(vcpu, regs->rdi);
8594 	kvm_rsp_write(vcpu, regs->rsp);
8595 	kvm_rbp_write(vcpu, regs->rbp);
8596 #ifdef CONFIG_X86_64
8597 	kvm_r8_write(vcpu, regs->r8);
8598 	kvm_r9_write(vcpu, regs->r9);
8599 	kvm_r10_write(vcpu, regs->r10);
8600 	kvm_r11_write(vcpu, regs->r11);
8601 	kvm_r12_write(vcpu, regs->r12);
8602 	kvm_r13_write(vcpu, regs->r13);
8603 	kvm_r14_write(vcpu, regs->r14);
8604 	kvm_r15_write(vcpu, regs->r15);
8605 #endif
8606 
8607 	kvm_rip_write(vcpu, regs->rip);
8608 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8609 
8610 	vcpu->arch.exception.pending = false;
8611 
8612 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8613 }
8614 
8615 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8616 {
8617 	vcpu_load(vcpu);
8618 	__set_regs(vcpu, regs);
8619 	vcpu_put(vcpu);
8620 	return 0;
8621 }
8622 
8623 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8624 {
8625 	struct kvm_segment cs;
8626 
8627 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8628 	*db = cs.db;
8629 	*l = cs.l;
8630 }
8631 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8632 
8633 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8634 {
8635 	struct desc_ptr dt;
8636 
8637 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8638 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8639 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8640 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8641 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8642 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8643 
8644 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8645 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8646 
8647 	kvm_x86_ops->get_idt(vcpu, &dt);
8648 	sregs->idt.limit = dt.size;
8649 	sregs->idt.base = dt.address;
8650 	kvm_x86_ops->get_gdt(vcpu, &dt);
8651 	sregs->gdt.limit = dt.size;
8652 	sregs->gdt.base = dt.address;
8653 
8654 	sregs->cr0 = kvm_read_cr0(vcpu);
8655 	sregs->cr2 = vcpu->arch.cr2;
8656 	sregs->cr3 = kvm_read_cr3(vcpu);
8657 	sregs->cr4 = kvm_read_cr4(vcpu);
8658 	sregs->cr8 = kvm_get_cr8(vcpu);
8659 	sregs->efer = vcpu->arch.efer;
8660 	sregs->apic_base = kvm_get_apic_base(vcpu);
8661 
8662 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8663 
8664 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8665 		set_bit(vcpu->arch.interrupt.nr,
8666 			(unsigned long *)sregs->interrupt_bitmap);
8667 }
8668 
8669 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8670 				  struct kvm_sregs *sregs)
8671 {
8672 	vcpu_load(vcpu);
8673 	__get_sregs(vcpu, sregs);
8674 	vcpu_put(vcpu);
8675 	return 0;
8676 }
8677 
8678 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8679 				    struct kvm_mp_state *mp_state)
8680 {
8681 	vcpu_load(vcpu);
8682 
8683 	kvm_apic_accept_events(vcpu);
8684 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8685 					vcpu->arch.pv.pv_unhalted)
8686 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8687 	else
8688 		mp_state->mp_state = vcpu->arch.mp_state;
8689 
8690 	vcpu_put(vcpu);
8691 	return 0;
8692 }
8693 
8694 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8695 				    struct kvm_mp_state *mp_state)
8696 {
8697 	int ret = -EINVAL;
8698 
8699 	vcpu_load(vcpu);
8700 
8701 	if (!lapic_in_kernel(vcpu) &&
8702 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8703 		goto out;
8704 
8705 	/* INITs are latched while in SMM */
8706 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8707 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8708 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8709 		goto out;
8710 
8711 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8712 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8713 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8714 	} else
8715 		vcpu->arch.mp_state = mp_state->mp_state;
8716 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8717 
8718 	ret = 0;
8719 out:
8720 	vcpu_put(vcpu);
8721 	return ret;
8722 }
8723 
8724 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8725 		    int reason, bool has_error_code, u32 error_code)
8726 {
8727 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8728 	int ret;
8729 
8730 	init_emulate_ctxt(vcpu);
8731 
8732 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8733 				   has_error_code, error_code);
8734 	if (ret) {
8735 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8736 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8737 		vcpu->run->internal.ndata = 0;
8738 		return 0;
8739 	}
8740 
8741 	kvm_rip_write(vcpu, ctxt->eip);
8742 	kvm_set_rflags(vcpu, ctxt->eflags);
8743 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8744 	return 1;
8745 }
8746 EXPORT_SYMBOL_GPL(kvm_task_switch);
8747 
8748 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8749 {
8750 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8751 		/*
8752 		 * When EFER.LME and CR0.PG are set, the processor is in
8753 		 * 64-bit mode (though maybe in a 32-bit code segment).
8754 		 * CR4.PAE and EFER.LMA must be set.
8755 		 */
8756 		if (!(sregs->cr4 & X86_CR4_PAE)
8757 		    || !(sregs->efer & EFER_LMA))
8758 			return -EINVAL;
8759 	} else {
8760 		/*
8761 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8762 		 * segment cannot be 64-bit.
8763 		 */
8764 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8765 			return -EINVAL;
8766 	}
8767 
8768 	return kvm_valid_cr4(vcpu, sregs->cr4);
8769 }
8770 
8771 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8772 {
8773 	struct msr_data apic_base_msr;
8774 	int mmu_reset_needed = 0;
8775 	int cpuid_update_needed = 0;
8776 	int pending_vec, max_bits, idx;
8777 	struct desc_ptr dt;
8778 	int ret = -EINVAL;
8779 
8780 	if (kvm_valid_sregs(vcpu, sregs))
8781 		goto out;
8782 
8783 	apic_base_msr.data = sregs->apic_base;
8784 	apic_base_msr.host_initiated = true;
8785 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8786 		goto out;
8787 
8788 	dt.size = sregs->idt.limit;
8789 	dt.address = sregs->idt.base;
8790 	kvm_x86_ops->set_idt(vcpu, &dt);
8791 	dt.size = sregs->gdt.limit;
8792 	dt.address = sregs->gdt.base;
8793 	kvm_x86_ops->set_gdt(vcpu, &dt);
8794 
8795 	vcpu->arch.cr2 = sregs->cr2;
8796 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8797 	vcpu->arch.cr3 = sregs->cr3;
8798 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8799 
8800 	kvm_set_cr8(vcpu, sregs->cr8);
8801 
8802 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8803 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8804 
8805 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8806 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8807 	vcpu->arch.cr0 = sregs->cr0;
8808 
8809 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8810 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8811 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8812 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8813 	if (cpuid_update_needed)
8814 		kvm_update_cpuid(vcpu);
8815 
8816 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8817 	if (is_pae_paging(vcpu)) {
8818 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8819 		mmu_reset_needed = 1;
8820 	}
8821 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8822 
8823 	if (mmu_reset_needed)
8824 		kvm_mmu_reset_context(vcpu);
8825 
8826 	max_bits = KVM_NR_INTERRUPTS;
8827 	pending_vec = find_first_bit(
8828 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8829 	if (pending_vec < max_bits) {
8830 		kvm_queue_interrupt(vcpu, pending_vec, false);
8831 		pr_debug("Set back pending irq %d\n", pending_vec);
8832 	}
8833 
8834 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8835 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8836 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8837 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8838 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8839 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8840 
8841 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8842 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8843 
8844 	update_cr8_intercept(vcpu);
8845 
8846 	/* Older userspace won't unhalt the vcpu on reset. */
8847 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8848 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8849 	    !is_protmode(vcpu))
8850 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8851 
8852 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8853 
8854 	ret = 0;
8855 out:
8856 	return ret;
8857 }
8858 
8859 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8860 				  struct kvm_sregs *sregs)
8861 {
8862 	int ret;
8863 
8864 	vcpu_load(vcpu);
8865 	ret = __set_sregs(vcpu, sregs);
8866 	vcpu_put(vcpu);
8867 	return ret;
8868 }
8869 
8870 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8871 					struct kvm_guest_debug *dbg)
8872 {
8873 	unsigned long rflags;
8874 	int i, r;
8875 
8876 	vcpu_load(vcpu);
8877 
8878 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8879 		r = -EBUSY;
8880 		if (vcpu->arch.exception.pending)
8881 			goto out;
8882 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8883 			kvm_queue_exception(vcpu, DB_VECTOR);
8884 		else
8885 			kvm_queue_exception(vcpu, BP_VECTOR);
8886 	}
8887 
8888 	/*
8889 	 * Read rflags as long as potentially injected trace flags are still
8890 	 * filtered out.
8891 	 */
8892 	rflags = kvm_get_rflags(vcpu);
8893 
8894 	vcpu->guest_debug = dbg->control;
8895 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8896 		vcpu->guest_debug = 0;
8897 
8898 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8899 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8900 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8901 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8902 	} else {
8903 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8904 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8905 	}
8906 	kvm_update_dr7(vcpu);
8907 
8908 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8909 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8910 			get_segment_base(vcpu, VCPU_SREG_CS);
8911 
8912 	/*
8913 	 * Trigger an rflags update that will inject or remove the trace
8914 	 * flags.
8915 	 */
8916 	kvm_set_rflags(vcpu, rflags);
8917 
8918 	kvm_x86_ops->update_bp_intercept(vcpu);
8919 
8920 	r = 0;
8921 
8922 out:
8923 	vcpu_put(vcpu);
8924 	return r;
8925 }
8926 
8927 /*
8928  * Translate a guest virtual address to a guest physical address.
8929  */
8930 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8931 				    struct kvm_translation *tr)
8932 {
8933 	unsigned long vaddr = tr->linear_address;
8934 	gpa_t gpa;
8935 	int idx;
8936 
8937 	vcpu_load(vcpu);
8938 
8939 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8940 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8941 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8942 	tr->physical_address = gpa;
8943 	tr->valid = gpa != UNMAPPED_GVA;
8944 	tr->writeable = 1;
8945 	tr->usermode = 0;
8946 
8947 	vcpu_put(vcpu);
8948 	return 0;
8949 }
8950 
8951 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8952 {
8953 	struct fxregs_state *fxsave;
8954 
8955 	vcpu_load(vcpu);
8956 
8957 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8958 	memcpy(fpu->fpr, fxsave->st_space, 128);
8959 	fpu->fcw = fxsave->cwd;
8960 	fpu->fsw = fxsave->swd;
8961 	fpu->ftwx = fxsave->twd;
8962 	fpu->last_opcode = fxsave->fop;
8963 	fpu->last_ip = fxsave->rip;
8964 	fpu->last_dp = fxsave->rdp;
8965 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8966 
8967 	vcpu_put(vcpu);
8968 	return 0;
8969 }
8970 
8971 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8972 {
8973 	struct fxregs_state *fxsave;
8974 
8975 	vcpu_load(vcpu);
8976 
8977 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8978 
8979 	memcpy(fxsave->st_space, fpu->fpr, 128);
8980 	fxsave->cwd = fpu->fcw;
8981 	fxsave->swd = fpu->fsw;
8982 	fxsave->twd = fpu->ftwx;
8983 	fxsave->fop = fpu->last_opcode;
8984 	fxsave->rip = fpu->last_ip;
8985 	fxsave->rdp = fpu->last_dp;
8986 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8987 
8988 	vcpu_put(vcpu);
8989 	return 0;
8990 }
8991 
8992 static void store_regs(struct kvm_vcpu *vcpu)
8993 {
8994 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8995 
8996 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8997 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8998 
8999 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9000 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9001 
9002 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9003 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9004 				vcpu, &vcpu->run->s.regs.events);
9005 }
9006 
9007 static int sync_regs(struct kvm_vcpu *vcpu)
9008 {
9009 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9010 		return -EINVAL;
9011 
9012 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9013 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9014 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9015 	}
9016 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9017 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9018 			return -EINVAL;
9019 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9020 	}
9021 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9022 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9023 				vcpu, &vcpu->run->s.regs.events))
9024 			return -EINVAL;
9025 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9026 	}
9027 
9028 	return 0;
9029 }
9030 
9031 static void fx_init(struct kvm_vcpu *vcpu)
9032 {
9033 	fpstate_init(&vcpu->arch.guest_fpu->state);
9034 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9035 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9036 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9037 
9038 	/*
9039 	 * Ensure guest xcr0 is valid for loading
9040 	 */
9041 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9042 
9043 	vcpu->arch.cr0 |= X86_CR0_ET;
9044 }
9045 
9046 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9047 {
9048 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9049 
9050 	kvmclock_reset(vcpu);
9051 
9052 	kvm_x86_ops->vcpu_free(vcpu);
9053 	free_cpumask_var(wbinvd_dirty_mask);
9054 }
9055 
9056 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9057 						unsigned int id)
9058 {
9059 	struct kvm_vcpu *vcpu;
9060 
9061 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9062 		printk_once(KERN_WARNING
9063 		"kvm: SMP vm created on host with unstable TSC; "
9064 		"guest TSC will not be reliable\n");
9065 
9066 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9067 
9068 	return vcpu;
9069 }
9070 
9071 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9072 {
9073 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9074 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9075 	kvm_vcpu_mtrr_init(vcpu);
9076 	vcpu_load(vcpu);
9077 	kvm_vcpu_reset(vcpu, false);
9078 	kvm_init_mmu(vcpu, false);
9079 	vcpu_put(vcpu);
9080 	return 0;
9081 }
9082 
9083 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9084 {
9085 	struct msr_data msr;
9086 	struct kvm *kvm = vcpu->kvm;
9087 
9088 	kvm_hv_vcpu_postcreate(vcpu);
9089 
9090 	if (mutex_lock_killable(&vcpu->mutex))
9091 		return;
9092 	vcpu_load(vcpu);
9093 	msr.data = 0x0;
9094 	msr.index = MSR_IA32_TSC;
9095 	msr.host_initiated = true;
9096 	kvm_write_tsc(vcpu, &msr);
9097 	vcpu_put(vcpu);
9098 
9099 	/* poll control enabled by default */
9100 	vcpu->arch.msr_kvm_poll_control = 1;
9101 
9102 	mutex_unlock(&vcpu->mutex);
9103 
9104 	if (!kvmclock_periodic_sync)
9105 		return;
9106 
9107 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9108 					KVMCLOCK_SYNC_PERIOD);
9109 }
9110 
9111 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9112 {
9113 	vcpu->arch.apf.msr_val = 0;
9114 
9115 	vcpu_load(vcpu);
9116 	kvm_mmu_unload(vcpu);
9117 	vcpu_put(vcpu);
9118 
9119 	kvm_x86_ops->vcpu_free(vcpu);
9120 }
9121 
9122 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9123 {
9124 	kvm_lapic_reset(vcpu, init_event);
9125 
9126 	vcpu->arch.hflags = 0;
9127 
9128 	vcpu->arch.smi_pending = 0;
9129 	vcpu->arch.smi_count = 0;
9130 	atomic_set(&vcpu->arch.nmi_queued, 0);
9131 	vcpu->arch.nmi_pending = 0;
9132 	vcpu->arch.nmi_injected = false;
9133 	kvm_clear_interrupt_queue(vcpu);
9134 	kvm_clear_exception_queue(vcpu);
9135 	vcpu->arch.exception.pending = false;
9136 
9137 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9138 	kvm_update_dr0123(vcpu);
9139 	vcpu->arch.dr6 = DR6_INIT;
9140 	kvm_update_dr6(vcpu);
9141 	vcpu->arch.dr7 = DR7_FIXED_1;
9142 	kvm_update_dr7(vcpu);
9143 
9144 	vcpu->arch.cr2 = 0;
9145 
9146 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9147 	vcpu->arch.apf.msr_val = 0;
9148 	vcpu->arch.st.msr_val = 0;
9149 
9150 	kvmclock_reset(vcpu);
9151 
9152 	kvm_clear_async_pf_completion_queue(vcpu);
9153 	kvm_async_pf_hash_reset(vcpu);
9154 	vcpu->arch.apf.halted = false;
9155 
9156 	if (kvm_mpx_supported()) {
9157 		void *mpx_state_buffer;
9158 
9159 		/*
9160 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9161 		 * called with loaded FPU and does not let userspace fix the state.
9162 		 */
9163 		if (init_event)
9164 			kvm_put_guest_fpu(vcpu);
9165 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9166 					XFEATURE_BNDREGS);
9167 		if (mpx_state_buffer)
9168 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9169 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9170 					XFEATURE_BNDCSR);
9171 		if (mpx_state_buffer)
9172 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9173 		if (init_event)
9174 			kvm_load_guest_fpu(vcpu);
9175 	}
9176 
9177 	if (!init_event) {
9178 		kvm_pmu_reset(vcpu);
9179 		vcpu->arch.smbase = 0x30000;
9180 
9181 		vcpu->arch.msr_misc_features_enables = 0;
9182 
9183 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9184 	}
9185 
9186 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9187 	vcpu->arch.regs_avail = ~0;
9188 	vcpu->arch.regs_dirty = ~0;
9189 
9190 	vcpu->arch.ia32_xss = 0;
9191 
9192 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
9193 }
9194 
9195 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9196 {
9197 	struct kvm_segment cs;
9198 
9199 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9200 	cs.selector = vector << 8;
9201 	cs.base = vector << 12;
9202 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9203 	kvm_rip_write(vcpu, 0);
9204 }
9205 
9206 int kvm_arch_hardware_enable(void)
9207 {
9208 	struct kvm *kvm;
9209 	struct kvm_vcpu *vcpu;
9210 	int i;
9211 	int ret;
9212 	u64 local_tsc;
9213 	u64 max_tsc = 0;
9214 	bool stable, backwards_tsc = false;
9215 
9216 	kvm_shared_msr_cpu_online();
9217 	ret = kvm_x86_ops->hardware_enable();
9218 	if (ret != 0)
9219 		return ret;
9220 
9221 	local_tsc = rdtsc();
9222 	stable = !kvm_check_tsc_unstable();
9223 	list_for_each_entry(kvm, &vm_list, vm_list) {
9224 		kvm_for_each_vcpu(i, vcpu, kvm) {
9225 			if (!stable && vcpu->cpu == smp_processor_id())
9226 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9227 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9228 				backwards_tsc = true;
9229 				if (vcpu->arch.last_host_tsc > max_tsc)
9230 					max_tsc = vcpu->arch.last_host_tsc;
9231 			}
9232 		}
9233 	}
9234 
9235 	/*
9236 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9237 	 * platforms that reset TSC during suspend or hibernate actions, but
9238 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9239 	 * detect that condition here, which happens early in CPU bringup,
9240 	 * before any KVM threads can be running.  Unfortunately, we can't
9241 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9242 	 * enough into CPU bringup that we know how much real time has actually
9243 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9244 	 * variables that haven't been updated yet.
9245 	 *
9246 	 * So we simply find the maximum observed TSC above, then record the
9247 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9248 	 * the adjustment will be applied.  Note that we accumulate
9249 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9250 	 * gets a chance to run again.  In the event that no KVM threads get a
9251 	 * chance to run, we will miss the entire elapsed period, as we'll have
9252 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9253 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9254 	 * uniform across all VCPUs (not to mention the scenario is extremely
9255 	 * unlikely). It is possible that a second hibernate recovery happens
9256 	 * much faster than a first, causing the observed TSC here to be
9257 	 * smaller; this would require additional padding adjustment, which is
9258 	 * why we set last_host_tsc to the local tsc observed here.
9259 	 *
9260 	 * N.B. - this code below runs only on platforms with reliable TSC,
9261 	 * as that is the only way backwards_tsc is set above.  Also note
9262 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9263 	 * have the same delta_cyc adjustment applied if backwards_tsc
9264 	 * is detected.  Note further, this adjustment is only done once,
9265 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9266 	 * called multiple times (one for each physical CPU bringup).
9267 	 *
9268 	 * Platforms with unreliable TSCs don't have to deal with this, they
9269 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9270 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9271 	 * guarantee that they stay in perfect synchronization.
9272 	 */
9273 	if (backwards_tsc) {
9274 		u64 delta_cyc = max_tsc - local_tsc;
9275 		list_for_each_entry(kvm, &vm_list, vm_list) {
9276 			kvm->arch.backwards_tsc_observed = true;
9277 			kvm_for_each_vcpu(i, vcpu, kvm) {
9278 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9279 				vcpu->arch.last_host_tsc = local_tsc;
9280 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9281 			}
9282 
9283 			/*
9284 			 * We have to disable TSC offset matching.. if you were
9285 			 * booting a VM while issuing an S4 host suspend....
9286 			 * you may have some problem.  Solving this issue is
9287 			 * left as an exercise to the reader.
9288 			 */
9289 			kvm->arch.last_tsc_nsec = 0;
9290 			kvm->arch.last_tsc_write = 0;
9291 		}
9292 
9293 	}
9294 	return 0;
9295 }
9296 
9297 void kvm_arch_hardware_disable(void)
9298 {
9299 	kvm_x86_ops->hardware_disable();
9300 	drop_user_return_notifiers();
9301 }
9302 
9303 int kvm_arch_hardware_setup(void)
9304 {
9305 	int r;
9306 
9307 	r = kvm_x86_ops->hardware_setup();
9308 	if (r != 0)
9309 		return r;
9310 
9311 	if (kvm_has_tsc_control) {
9312 		/*
9313 		 * Make sure the user can only configure tsc_khz values that
9314 		 * fit into a signed integer.
9315 		 * A min value is not calculated because it will always
9316 		 * be 1 on all machines.
9317 		 */
9318 		u64 max = min(0x7fffffffULL,
9319 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9320 		kvm_max_guest_tsc_khz = max;
9321 
9322 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9323 	}
9324 
9325 	kvm_init_msr_list();
9326 	return 0;
9327 }
9328 
9329 void kvm_arch_hardware_unsetup(void)
9330 {
9331 	kvm_x86_ops->hardware_unsetup();
9332 }
9333 
9334 int kvm_arch_check_processor_compat(void)
9335 {
9336 	return kvm_x86_ops->check_processor_compatibility();
9337 }
9338 
9339 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9340 {
9341 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9342 }
9343 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9344 
9345 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9346 {
9347 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9348 }
9349 
9350 struct static_key kvm_no_apic_vcpu __read_mostly;
9351 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9352 
9353 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9354 {
9355 	struct page *page;
9356 	int r;
9357 
9358 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9359 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9360 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9361 	else
9362 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9363 
9364 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9365 	if (!page) {
9366 		r = -ENOMEM;
9367 		goto fail;
9368 	}
9369 	vcpu->arch.pio_data = page_address(page);
9370 
9371 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9372 
9373 	r = kvm_mmu_create(vcpu);
9374 	if (r < 0)
9375 		goto fail_free_pio_data;
9376 
9377 	if (irqchip_in_kernel(vcpu->kvm)) {
9378 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9379 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9380 		if (r < 0)
9381 			goto fail_mmu_destroy;
9382 	} else
9383 		static_key_slow_inc(&kvm_no_apic_vcpu);
9384 
9385 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9386 				       GFP_KERNEL_ACCOUNT);
9387 	if (!vcpu->arch.mce_banks) {
9388 		r = -ENOMEM;
9389 		goto fail_free_lapic;
9390 	}
9391 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9392 
9393 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9394 				GFP_KERNEL_ACCOUNT)) {
9395 		r = -ENOMEM;
9396 		goto fail_free_mce_banks;
9397 	}
9398 
9399 	fx_init(vcpu);
9400 
9401 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9402 
9403 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9404 
9405 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9406 
9407 	kvm_async_pf_hash_reset(vcpu);
9408 	kvm_pmu_init(vcpu);
9409 
9410 	vcpu->arch.pending_external_vector = -1;
9411 	vcpu->arch.preempted_in_kernel = false;
9412 
9413 	kvm_hv_vcpu_init(vcpu);
9414 
9415 	return 0;
9416 
9417 fail_free_mce_banks:
9418 	kfree(vcpu->arch.mce_banks);
9419 fail_free_lapic:
9420 	kvm_free_lapic(vcpu);
9421 fail_mmu_destroy:
9422 	kvm_mmu_destroy(vcpu);
9423 fail_free_pio_data:
9424 	free_page((unsigned long)vcpu->arch.pio_data);
9425 fail:
9426 	return r;
9427 }
9428 
9429 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9430 {
9431 	int idx;
9432 
9433 	kvm_hv_vcpu_uninit(vcpu);
9434 	kvm_pmu_destroy(vcpu);
9435 	kfree(vcpu->arch.mce_banks);
9436 	kvm_free_lapic(vcpu);
9437 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9438 	kvm_mmu_destroy(vcpu);
9439 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9440 	free_page((unsigned long)vcpu->arch.pio_data);
9441 	if (!lapic_in_kernel(vcpu))
9442 		static_key_slow_dec(&kvm_no_apic_vcpu);
9443 }
9444 
9445 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9446 {
9447 	vcpu->arch.l1tf_flush_l1d = true;
9448 	kvm_x86_ops->sched_in(vcpu, cpu);
9449 }
9450 
9451 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9452 {
9453 	if (type)
9454 		return -EINVAL;
9455 
9456 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9457 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9458 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9459 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9460 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9461 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9462 
9463 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9464 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9465 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9466 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9467 		&kvm->arch.irq_sources_bitmap);
9468 
9469 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9470 	mutex_init(&kvm->arch.apic_map_lock);
9471 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9472 
9473 	kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9474 	pvclock_update_vm_gtod_copy(kvm);
9475 
9476 	kvm->arch.guest_can_read_msr_platform_info = true;
9477 
9478 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9479 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9480 
9481 	kvm_hv_init_vm(kvm);
9482 	kvm_page_track_init(kvm);
9483 	kvm_mmu_init_vm(kvm);
9484 
9485 	return kvm_x86_ops->vm_init(kvm);
9486 }
9487 
9488 int kvm_arch_post_init_vm(struct kvm *kvm)
9489 {
9490 	return kvm_mmu_post_init_vm(kvm);
9491 }
9492 
9493 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9494 {
9495 	vcpu_load(vcpu);
9496 	kvm_mmu_unload(vcpu);
9497 	vcpu_put(vcpu);
9498 }
9499 
9500 static void kvm_free_vcpus(struct kvm *kvm)
9501 {
9502 	unsigned int i;
9503 	struct kvm_vcpu *vcpu;
9504 
9505 	/*
9506 	 * Unpin any mmu pages first.
9507 	 */
9508 	kvm_for_each_vcpu(i, vcpu, kvm) {
9509 		kvm_clear_async_pf_completion_queue(vcpu);
9510 		kvm_unload_vcpu_mmu(vcpu);
9511 	}
9512 	kvm_for_each_vcpu(i, vcpu, kvm)
9513 		kvm_arch_vcpu_free(vcpu);
9514 
9515 	mutex_lock(&kvm->lock);
9516 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9517 		kvm->vcpus[i] = NULL;
9518 
9519 	atomic_set(&kvm->online_vcpus, 0);
9520 	mutex_unlock(&kvm->lock);
9521 }
9522 
9523 void kvm_arch_sync_events(struct kvm *kvm)
9524 {
9525 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9526 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9527 	kvm_free_pit(kvm);
9528 }
9529 
9530 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9531 {
9532 	int i, r;
9533 	unsigned long hva;
9534 	struct kvm_memslots *slots = kvm_memslots(kvm);
9535 	struct kvm_memory_slot *slot, old;
9536 
9537 	/* Called with kvm->slots_lock held.  */
9538 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9539 		return -EINVAL;
9540 
9541 	slot = id_to_memslot(slots, id);
9542 	if (size) {
9543 		if (slot->npages)
9544 			return -EEXIST;
9545 
9546 		/*
9547 		 * MAP_SHARED to prevent internal slot pages from being moved
9548 		 * by fork()/COW.
9549 		 */
9550 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9551 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9552 		if (IS_ERR((void *)hva))
9553 			return PTR_ERR((void *)hva);
9554 	} else {
9555 		if (!slot->npages)
9556 			return 0;
9557 
9558 		hva = 0;
9559 	}
9560 
9561 	old = *slot;
9562 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9563 		struct kvm_userspace_memory_region m;
9564 
9565 		m.slot = id | (i << 16);
9566 		m.flags = 0;
9567 		m.guest_phys_addr = gpa;
9568 		m.userspace_addr = hva;
9569 		m.memory_size = size;
9570 		r = __kvm_set_memory_region(kvm, &m);
9571 		if (r < 0)
9572 			return r;
9573 	}
9574 
9575 	if (!size)
9576 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9577 
9578 	return 0;
9579 }
9580 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9581 
9582 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9583 {
9584 	int r;
9585 
9586 	mutex_lock(&kvm->slots_lock);
9587 	r = __x86_set_memory_region(kvm, id, gpa, size);
9588 	mutex_unlock(&kvm->slots_lock);
9589 
9590 	return r;
9591 }
9592 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9593 
9594 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9595 {
9596 	kvm_mmu_pre_destroy_vm(kvm);
9597 }
9598 
9599 void kvm_arch_destroy_vm(struct kvm *kvm)
9600 {
9601 	if (current->mm == kvm->mm) {
9602 		/*
9603 		 * Free memory regions allocated on behalf of userspace,
9604 		 * unless the the memory map has changed due to process exit
9605 		 * or fd copying.
9606 		 */
9607 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9608 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9609 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9610 	}
9611 	if (kvm_x86_ops->vm_destroy)
9612 		kvm_x86_ops->vm_destroy(kvm);
9613 	kvm_pic_destroy(kvm);
9614 	kvm_ioapic_destroy(kvm);
9615 	kvm_free_vcpus(kvm);
9616 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9617 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9618 	kvm_mmu_uninit_vm(kvm);
9619 	kvm_page_track_cleanup(kvm);
9620 	kvm_hv_destroy_vm(kvm);
9621 }
9622 
9623 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9624 			   struct kvm_memory_slot *dont)
9625 {
9626 	int i;
9627 
9628 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9629 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9630 			kvfree(free->arch.rmap[i]);
9631 			free->arch.rmap[i] = NULL;
9632 		}
9633 		if (i == 0)
9634 			continue;
9635 
9636 		if (!dont || free->arch.lpage_info[i - 1] !=
9637 			     dont->arch.lpage_info[i - 1]) {
9638 			kvfree(free->arch.lpage_info[i - 1]);
9639 			free->arch.lpage_info[i - 1] = NULL;
9640 		}
9641 	}
9642 
9643 	kvm_page_track_free_memslot(free, dont);
9644 }
9645 
9646 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9647 			    unsigned long npages)
9648 {
9649 	int i;
9650 
9651 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9652 		struct kvm_lpage_info *linfo;
9653 		unsigned long ugfn;
9654 		int lpages;
9655 		int level = i + 1;
9656 
9657 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9658 				      slot->base_gfn, level) + 1;
9659 
9660 		slot->arch.rmap[i] =
9661 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9662 				 GFP_KERNEL_ACCOUNT);
9663 		if (!slot->arch.rmap[i])
9664 			goto out_free;
9665 		if (i == 0)
9666 			continue;
9667 
9668 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9669 		if (!linfo)
9670 			goto out_free;
9671 
9672 		slot->arch.lpage_info[i - 1] = linfo;
9673 
9674 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9675 			linfo[0].disallow_lpage = 1;
9676 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9677 			linfo[lpages - 1].disallow_lpage = 1;
9678 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9679 		/*
9680 		 * If the gfn and userspace address are not aligned wrt each
9681 		 * other, or if explicitly asked to, disable large page
9682 		 * support for this slot
9683 		 */
9684 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9685 		    !kvm_largepages_enabled()) {
9686 			unsigned long j;
9687 
9688 			for (j = 0; j < lpages; ++j)
9689 				linfo[j].disallow_lpage = 1;
9690 		}
9691 	}
9692 
9693 	if (kvm_page_track_create_memslot(slot, npages))
9694 		goto out_free;
9695 
9696 	return 0;
9697 
9698 out_free:
9699 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9700 		kvfree(slot->arch.rmap[i]);
9701 		slot->arch.rmap[i] = NULL;
9702 		if (i == 0)
9703 			continue;
9704 
9705 		kvfree(slot->arch.lpage_info[i - 1]);
9706 		slot->arch.lpage_info[i - 1] = NULL;
9707 	}
9708 	return -ENOMEM;
9709 }
9710 
9711 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9712 {
9713 	/*
9714 	 * memslots->generation has been incremented.
9715 	 * mmio generation may have reached its maximum value.
9716 	 */
9717 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9718 }
9719 
9720 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9721 				struct kvm_memory_slot *memslot,
9722 				const struct kvm_userspace_memory_region *mem,
9723 				enum kvm_mr_change change)
9724 {
9725 	return 0;
9726 }
9727 
9728 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9729 				     struct kvm_memory_slot *new)
9730 {
9731 	/* Still write protect RO slot */
9732 	if (new->flags & KVM_MEM_READONLY) {
9733 		kvm_mmu_slot_remove_write_access(kvm, new);
9734 		return;
9735 	}
9736 
9737 	/*
9738 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9739 	 *
9740 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9741 	 *
9742 	 *  - KVM_MR_CREATE with dirty logging is disabled
9743 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9744 	 *
9745 	 * The reason is, in case of PML, we need to set D-bit for any slots
9746 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9747 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9748 	 * guarantees leaving PML enabled during guest's lifetime won't have
9749 	 * any additional overhead from PML when guest is running with dirty
9750 	 * logging disabled for memory slots.
9751 	 *
9752 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9753 	 * to dirty logging mode.
9754 	 *
9755 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9756 	 *
9757 	 * In case of write protect:
9758 	 *
9759 	 * Write protect all pages for dirty logging.
9760 	 *
9761 	 * All the sptes including the large sptes which point to this
9762 	 * slot are set to readonly. We can not create any new large
9763 	 * spte on this slot until the end of the logging.
9764 	 *
9765 	 * See the comments in fast_page_fault().
9766 	 */
9767 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9768 		if (kvm_x86_ops->slot_enable_log_dirty)
9769 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9770 		else
9771 			kvm_mmu_slot_remove_write_access(kvm, new);
9772 	} else {
9773 		if (kvm_x86_ops->slot_disable_log_dirty)
9774 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9775 	}
9776 }
9777 
9778 void kvm_arch_commit_memory_region(struct kvm *kvm,
9779 				const struct kvm_userspace_memory_region *mem,
9780 				const struct kvm_memory_slot *old,
9781 				const struct kvm_memory_slot *new,
9782 				enum kvm_mr_change change)
9783 {
9784 	if (!kvm->arch.n_requested_mmu_pages)
9785 		kvm_mmu_change_mmu_pages(kvm,
9786 				kvm_mmu_calculate_default_mmu_pages(kvm));
9787 
9788 	/*
9789 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9790 	 * sptes have to be split.  If live migration is successful, the guest
9791 	 * in the source machine will be destroyed and large sptes will be
9792 	 * created in the destination. However, if the guest continues to run
9793 	 * in the source machine (for example if live migration fails), small
9794 	 * sptes will remain around and cause bad performance.
9795 	 *
9796 	 * Scan sptes if dirty logging has been stopped, dropping those
9797 	 * which can be collapsed into a single large-page spte.  Later
9798 	 * page faults will create the large-page sptes.
9799 	 *
9800 	 * There is no need to do this in any of the following cases:
9801 	 * CREATE:	No dirty mappings will already exist.
9802 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
9803 	 *		kvm_arch_flush_shadow_memslot()
9804 	 */
9805 	if (change == KVM_MR_FLAGS_ONLY &&
9806 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9807 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9808 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9809 
9810 	/*
9811 	 * Set up write protection and/or dirty logging for the new slot.
9812 	 *
9813 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9814 	 * been zapped so no dirty logging staff is needed for old slot. For
9815 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9816 	 * new and it's also covered when dealing with the new slot.
9817 	 *
9818 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9819 	 */
9820 	if (change != KVM_MR_DELETE)
9821 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9822 }
9823 
9824 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9825 {
9826 	kvm_mmu_zap_all(kvm);
9827 }
9828 
9829 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9830 				   struct kvm_memory_slot *slot)
9831 {
9832 	kvm_page_track_flush_slot(kvm, slot);
9833 }
9834 
9835 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9836 {
9837 	return (is_guest_mode(vcpu) &&
9838 			kvm_x86_ops->guest_apic_has_interrupt &&
9839 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9840 }
9841 
9842 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9843 {
9844 	if (!list_empty_careful(&vcpu->async_pf.done))
9845 		return true;
9846 
9847 	if (kvm_apic_has_events(vcpu))
9848 		return true;
9849 
9850 	if (vcpu->arch.pv.pv_unhalted)
9851 		return true;
9852 
9853 	if (vcpu->arch.exception.pending)
9854 		return true;
9855 
9856 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9857 	    (vcpu->arch.nmi_pending &&
9858 	     kvm_x86_ops->nmi_allowed(vcpu)))
9859 		return true;
9860 
9861 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9862 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9863 		return true;
9864 
9865 	if (kvm_arch_interrupt_allowed(vcpu) &&
9866 	    (kvm_cpu_has_interrupt(vcpu) ||
9867 	    kvm_guest_apic_has_interrupt(vcpu)))
9868 		return true;
9869 
9870 	if (kvm_hv_has_stimer_pending(vcpu))
9871 		return true;
9872 
9873 	return false;
9874 }
9875 
9876 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9877 {
9878 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9879 }
9880 
9881 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9882 {
9883 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9884 		return true;
9885 
9886 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9887 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
9888 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
9889 		return true;
9890 
9891 	if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9892 		return true;
9893 
9894 	return false;
9895 }
9896 
9897 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9898 {
9899 	return vcpu->arch.preempted_in_kernel;
9900 }
9901 
9902 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9903 {
9904 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9905 }
9906 
9907 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9908 {
9909 	return kvm_x86_ops->interrupt_allowed(vcpu);
9910 }
9911 
9912 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9913 {
9914 	if (is_64_bit_mode(vcpu))
9915 		return kvm_rip_read(vcpu);
9916 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9917 		     kvm_rip_read(vcpu));
9918 }
9919 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9920 
9921 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9922 {
9923 	return kvm_get_linear_rip(vcpu) == linear_rip;
9924 }
9925 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9926 
9927 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9928 {
9929 	unsigned long rflags;
9930 
9931 	rflags = kvm_x86_ops->get_rflags(vcpu);
9932 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9933 		rflags &= ~X86_EFLAGS_TF;
9934 	return rflags;
9935 }
9936 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9937 
9938 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9939 {
9940 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9941 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9942 		rflags |= X86_EFLAGS_TF;
9943 	kvm_x86_ops->set_rflags(vcpu, rflags);
9944 }
9945 
9946 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9947 {
9948 	__kvm_set_rflags(vcpu, rflags);
9949 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9950 }
9951 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9952 
9953 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9954 {
9955 	int r;
9956 
9957 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9958 	      work->wakeup_all)
9959 		return;
9960 
9961 	r = kvm_mmu_reload(vcpu);
9962 	if (unlikely(r))
9963 		return;
9964 
9965 	if (!vcpu->arch.mmu->direct_map &&
9966 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9967 		return;
9968 
9969 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9970 }
9971 
9972 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9973 {
9974 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9975 }
9976 
9977 static inline u32 kvm_async_pf_next_probe(u32 key)
9978 {
9979 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9980 }
9981 
9982 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9983 {
9984 	u32 key = kvm_async_pf_hash_fn(gfn);
9985 
9986 	while (vcpu->arch.apf.gfns[key] != ~0)
9987 		key = kvm_async_pf_next_probe(key);
9988 
9989 	vcpu->arch.apf.gfns[key] = gfn;
9990 }
9991 
9992 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9993 {
9994 	int i;
9995 	u32 key = kvm_async_pf_hash_fn(gfn);
9996 
9997 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9998 		     (vcpu->arch.apf.gfns[key] != gfn &&
9999 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10000 		key = kvm_async_pf_next_probe(key);
10001 
10002 	return key;
10003 }
10004 
10005 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10006 {
10007 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10008 }
10009 
10010 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10011 {
10012 	u32 i, j, k;
10013 
10014 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10015 	while (true) {
10016 		vcpu->arch.apf.gfns[i] = ~0;
10017 		do {
10018 			j = kvm_async_pf_next_probe(j);
10019 			if (vcpu->arch.apf.gfns[j] == ~0)
10020 				return;
10021 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10022 			/*
10023 			 * k lies cyclically in ]i,j]
10024 			 * |    i.k.j |
10025 			 * |....j i.k.| or  |.k..j i...|
10026 			 */
10027 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10028 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10029 		i = j;
10030 	}
10031 }
10032 
10033 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10034 {
10035 
10036 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10037 				      sizeof(val));
10038 }
10039 
10040 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10041 {
10042 
10043 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10044 				      sizeof(u32));
10045 }
10046 
10047 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10048 {
10049 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10050 		return false;
10051 
10052 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10053 	    (vcpu->arch.apf.send_user_only &&
10054 	     kvm_x86_ops->get_cpl(vcpu) == 0))
10055 		return false;
10056 
10057 	return true;
10058 }
10059 
10060 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10061 {
10062 	if (unlikely(!lapic_in_kernel(vcpu) ||
10063 		     kvm_event_needs_reinjection(vcpu) ||
10064 		     vcpu->arch.exception.pending))
10065 		return false;
10066 
10067 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10068 		return false;
10069 
10070 	/*
10071 	 * If interrupts are off we cannot even use an artificial
10072 	 * halt state.
10073 	 */
10074 	return kvm_x86_ops->interrupt_allowed(vcpu);
10075 }
10076 
10077 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10078 				     struct kvm_async_pf *work)
10079 {
10080 	struct x86_exception fault;
10081 
10082 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10083 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10084 
10085 	if (kvm_can_deliver_async_pf(vcpu) &&
10086 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10087 		fault.vector = PF_VECTOR;
10088 		fault.error_code_valid = true;
10089 		fault.error_code = 0;
10090 		fault.nested_page_fault = false;
10091 		fault.address = work->arch.token;
10092 		fault.async_page_fault = true;
10093 		kvm_inject_page_fault(vcpu, &fault);
10094 	} else {
10095 		/*
10096 		 * It is not possible to deliver a paravirtualized asynchronous
10097 		 * page fault, but putting the guest in an artificial halt state
10098 		 * can be beneficial nevertheless: if an interrupt arrives, we
10099 		 * can deliver it timely and perhaps the guest will schedule
10100 		 * another process.  When the instruction that triggered a page
10101 		 * fault is retried, hopefully the page will be ready in the host.
10102 		 */
10103 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10104 	}
10105 }
10106 
10107 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10108 				 struct kvm_async_pf *work)
10109 {
10110 	struct x86_exception fault;
10111 	u32 val;
10112 
10113 	if (work->wakeup_all)
10114 		work->arch.token = ~0; /* broadcast wakeup */
10115 	else
10116 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10117 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
10118 
10119 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10120 	    !apf_get_user(vcpu, &val)) {
10121 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10122 		    vcpu->arch.exception.pending &&
10123 		    vcpu->arch.exception.nr == PF_VECTOR &&
10124 		    !apf_put_user(vcpu, 0)) {
10125 			vcpu->arch.exception.injected = false;
10126 			vcpu->arch.exception.pending = false;
10127 			vcpu->arch.exception.nr = 0;
10128 			vcpu->arch.exception.has_error_code = false;
10129 			vcpu->arch.exception.error_code = 0;
10130 			vcpu->arch.exception.has_payload = false;
10131 			vcpu->arch.exception.payload = 0;
10132 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10133 			fault.vector = PF_VECTOR;
10134 			fault.error_code_valid = true;
10135 			fault.error_code = 0;
10136 			fault.nested_page_fault = false;
10137 			fault.address = work->arch.token;
10138 			fault.async_page_fault = true;
10139 			kvm_inject_page_fault(vcpu, &fault);
10140 		}
10141 	}
10142 	vcpu->arch.apf.halted = false;
10143 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10144 }
10145 
10146 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10147 {
10148 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10149 		return true;
10150 	else
10151 		return kvm_can_do_async_pf(vcpu);
10152 }
10153 
10154 void kvm_arch_start_assignment(struct kvm *kvm)
10155 {
10156 	atomic_inc(&kvm->arch.assigned_device_count);
10157 }
10158 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10159 
10160 void kvm_arch_end_assignment(struct kvm *kvm)
10161 {
10162 	atomic_dec(&kvm->arch.assigned_device_count);
10163 }
10164 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10165 
10166 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10167 {
10168 	return atomic_read(&kvm->arch.assigned_device_count);
10169 }
10170 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10171 
10172 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10173 {
10174 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10175 }
10176 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10177 
10178 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10179 {
10180 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10181 }
10182 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10183 
10184 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10185 {
10186 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10187 }
10188 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10189 
10190 bool kvm_arch_has_irq_bypass(void)
10191 {
10192 	return true;
10193 }
10194 
10195 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10196 				      struct irq_bypass_producer *prod)
10197 {
10198 	struct kvm_kernel_irqfd *irqfd =
10199 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10200 
10201 	irqfd->producer = prod;
10202 
10203 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10204 					   prod->irq, irqfd->gsi, 1);
10205 }
10206 
10207 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10208 				      struct irq_bypass_producer *prod)
10209 {
10210 	int ret;
10211 	struct kvm_kernel_irqfd *irqfd =
10212 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10213 
10214 	WARN_ON(irqfd->producer != prod);
10215 	irqfd->producer = NULL;
10216 
10217 	/*
10218 	 * When producer of consumer is unregistered, we change back to
10219 	 * remapped mode, so we can re-use the current implementation
10220 	 * when the irq is masked/disabled or the consumer side (KVM
10221 	 * int this case doesn't want to receive the interrupts.
10222 	*/
10223 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10224 	if (ret)
10225 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10226 		       " fails: %d\n", irqfd->consumer.token, ret);
10227 }
10228 
10229 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10230 				   uint32_t guest_irq, bool set)
10231 {
10232 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10233 }
10234 
10235 bool kvm_vector_hashing_enabled(void)
10236 {
10237 	return vector_hashing;
10238 }
10239 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10240 
10241 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10242 {
10243 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10244 }
10245 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10246 
10247 
10248 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10249 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10250 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10251 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10252 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10253 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10254 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10255 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10256 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10257 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10258 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10259 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10260 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10261 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10262 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10263 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10264 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10265 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10268