xref: /openbmc/linux/arch/x86/kvm/x86.c (revision d003d772)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
58 
59 #include <trace/events/kvm.h>
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
73 
74 #define CREATE_TRACE_POINTS
75 #include "trace.h"
76 
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 
82 #define emul_to_vcpu(ctxt) \
83 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
84 
85 /* EFER defaults:
86  * - enable syscall per default because its emulated by KVM
87  * - enable LME and LMA per default on 64 bit KVM
88  */
89 #ifdef CONFIG_X86_64
90 static
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 #else
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
94 #endif
95 
96 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
97 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
108 
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32  __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64  __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
138 
139 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
140 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
141 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
143 
144 static bool __read_mostly vector_hashing = true;
145 module_param(vector_hashing, bool, S_IRUGO);
146 
147 bool __read_mostly enable_vmware_backdoor = false;
148 module_param(enable_vmware_backdoor, bool, S_IRUGO);
149 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
150 
151 static bool __read_mostly force_emulation_prefix = false;
152 module_param(force_emulation_prefix, bool, S_IRUGO);
153 
154 #define KVM_NR_SHARED_MSRS 16
155 
156 struct kvm_shared_msrs_global {
157 	int nr;
158 	u32 msrs[KVM_NR_SHARED_MSRS];
159 };
160 
161 struct kvm_shared_msrs {
162 	struct user_return_notifier urn;
163 	bool registered;
164 	struct kvm_shared_msr_values {
165 		u64 host;
166 		u64 curr;
167 	} values[KVM_NR_SHARED_MSRS];
168 };
169 
170 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
171 static struct kvm_shared_msrs __percpu *shared_msrs;
172 
173 struct kvm_stats_debugfs_item debugfs_entries[] = {
174 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
175 	{ "pf_guest", VCPU_STAT(pf_guest) },
176 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
177 	{ "invlpg", VCPU_STAT(invlpg) },
178 	{ "exits", VCPU_STAT(exits) },
179 	{ "io_exits", VCPU_STAT(io_exits) },
180 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
181 	{ "signal_exits", VCPU_STAT(signal_exits) },
182 	{ "irq_window", VCPU_STAT(irq_window_exits) },
183 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
184 	{ "halt_exits", VCPU_STAT(halt_exits) },
185 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
186 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
187 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
188 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
189 	{ "hypercalls", VCPU_STAT(hypercalls) },
190 	{ "request_irq", VCPU_STAT(request_irq_exits) },
191 	{ "irq_exits", VCPU_STAT(irq_exits) },
192 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
193 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
194 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
195 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
196 	{ "irq_injections", VCPU_STAT(irq_injections) },
197 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
198 	{ "req_event", VCPU_STAT(req_event) },
199 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
200 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
201 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
202 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
203 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
204 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
205 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
206 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
207 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
208 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
209 	{ "largepages", VM_STAT(lpages) },
210 	{ "max_mmu_page_hash_collisions",
211 		VM_STAT(max_mmu_page_hash_collisions) },
212 	{ NULL }
213 };
214 
215 u64 __read_mostly host_xcr0;
216 
217 struct kmem_cache *x86_fpu_cache;
218 EXPORT_SYMBOL_GPL(x86_fpu_cache);
219 
220 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
221 
222 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
223 {
224 	int i;
225 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
226 		vcpu->arch.apf.gfns[i] = ~0;
227 }
228 
229 static void kvm_on_user_return(struct user_return_notifier *urn)
230 {
231 	unsigned slot;
232 	struct kvm_shared_msrs *locals
233 		= container_of(urn, struct kvm_shared_msrs, urn);
234 	struct kvm_shared_msr_values *values;
235 	unsigned long flags;
236 
237 	/*
238 	 * Disabling irqs at this point since the following code could be
239 	 * interrupted and executed through kvm_arch_hardware_disable()
240 	 */
241 	local_irq_save(flags);
242 	if (locals->registered) {
243 		locals->registered = false;
244 		user_return_notifier_unregister(urn);
245 	}
246 	local_irq_restore(flags);
247 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
248 		values = &locals->values[slot];
249 		if (values->host != values->curr) {
250 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
251 			values->curr = values->host;
252 		}
253 	}
254 }
255 
256 static void shared_msr_update(unsigned slot, u32 msr)
257 {
258 	u64 value;
259 	unsigned int cpu = smp_processor_id();
260 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
261 
262 	/* only read, and nobody should modify it at this time,
263 	 * so don't need lock */
264 	if (slot >= shared_msrs_global.nr) {
265 		printk(KERN_ERR "kvm: invalid MSR slot!");
266 		return;
267 	}
268 	rdmsrl_safe(msr, &value);
269 	smsr->values[slot].host = value;
270 	smsr->values[slot].curr = value;
271 }
272 
273 void kvm_define_shared_msr(unsigned slot, u32 msr)
274 {
275 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
276 	shared_msrs_global.msrs[slot] = msr;
277 	if (slot >= shared_msrs_global.nr)
278 		shared_msrs_global.nr = slot + 1;
279 }
280 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
281 
282 static void kvm_shared_msr_cpu_online(void)
283 {
284 	unsigned i;
285 
286 	for (i = 0; i < shared_msrs_global.nr; ++i)
287 		shared_msr_update(i, shared_msrs_global.msrs[i]);
288 }
289 
290 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
291 {
292 	unsigned int cpu = smp_processor_id();
293 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
294 	int err;
295 
296 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
297 		return 0;
298 	smsr->values[slot].curr = value;
299 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
300 	if (err)
301 		return 1;
302 
303 	if (!smsr->registered) {
304 		smsr->urn.on_user_return = kvm_on_user_return;
305 		user_return_notifier_register(&smsr->urn);
306 		smsr->registered = true;
307 	}
308 	return 0;
309 }
310 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
311 
312 static void drop_user_return_notifiers(void)
313 {
314 	unsigned int cpu = smp_processor_id();
315 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
316 
317 	if (smsr->registered)
318 		kvm_on_user_return(&smsr->urn);
319 }
320 
321 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
322 {
323 	return vcpu->arch.apic_base;
324 }
325 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
326 
327 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
328 {
329 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
330 }
331 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
332 
333 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
334 {
335 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
336 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
337 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
338 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
339 
340 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
341 		return 1;
342 	if (!msr_info->host_initiated) {
343 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
344 			return 1;
345 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
346 			return 1;
347 	}
348 
349 	kvm_lapic_set_base(vcpu, msr_info->data);
350 	return 0;
351 }
352 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
353 
354 asmlinkage __visible void kvm_spurious_fault(void)
355 {
356 	/* Fault while not rebooting.  We want the trace. */
357 	BUG();
358 }
359 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
360 
361 #define EXCPT_BENIGN		0
362 #define EXCPT_CONTRIBUTORY	1
363 #define EXCPT_PF		2
364 
365 static int exception_class(int vector)
366 {
367 	switch (vector) {
368 	case PF_VECTOR:
369 		return EXCPT_PF;
370 	case DE_VECTOR:
371 	case TS_VECTOR:
372 	case NP_VECTOR:
373 	case SS_VECTOR:
374 	case GP_VECTOR:
375 		return EXCPT_CONTRIBUTORY;
376 	default:
377 		break;
378 	}
379 	return EXCPT_BENIGN;
380 }
381 
382 #define EXCPT_FAULT		0
383 #define EXCPT_TRAP		1
384 #define EXCPT_ABORT		2
385 #define EXCPT_INTERRUPT		3
386 
387 static int exception_type(int vector)
388 {
389 	unsigned int mask;
390 
391 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
392 		return EXCPT_INTERRUPT;
393 
394 	mask = 1 << vector;
395 
396 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
397 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
398 		return EXCPT_TRAP;
399 
400 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
401 		return EXCPT_ABORT;
402 
403 	/* Reserved exceptions will result in fault */
404 	return EXCPT_FAULT;
405 }
406 
407 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
408 {
409 	unsigned nr = vcpu->arch.exception.nr;
410 	bool has_payload = vcpu->arch.exception.has_payload;
411 	unsigned long payload = vcpu->arch.exception.payload;
412 
413 	if (!has_payload)
414 		return;
415 
416 	switch (nr) {
417 	case DB_VECTOR:
418 		/*
419 		 * "Certain debug exceptions may clear bit 0-3.  The
420 		 * remaining contents of the DR6 register are never
421 		 * cleared by the processor".
422 		 */
423 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
424 		/*
425 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
426 		 */
427 		vcpu->arch.dr6 |= DR6_RTM;
428 		vcpu->arch.dr6 |= payload;
429 		/*
430 		 * Bit 16 should be set in the payload whenever the #DB
431 		 * exception should clear DR6.RTM. This makes the payload
432 		 * compatible with the pending debug exceptions under VMX.
433 		 * Though not currently documented in the SDM, this also
434 		 * makes the payload compatible with the exit qualification
435 		 * for #DB exceptions under VMX.
436 		 */
437 		vcpu->arch.dr6 ^= payload & DR6_RTM;
438 		break;
439 	case PF_VECTOR:
440 		vcpu->arch.cr2 = payload;
441 		break;
442 	}
443 
444 	vcpu->arch.exception.has_payload = false;
445 	vcpu->arch.exception.payload = 0;
446 }
447 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
448 
449 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
450 		unsigned nr, bool has_error, u32 error_code,
451 	        bool has_payload, unsigned long payload, bool reinject)
452 {
453 	u32 prev_nr;
454 	int class1, class2;
455 
456 	kvm_make_request(KVM_REQ_EVENT, vcpu);
457 
458 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
459 	queue:
460 		if (has_error && !is_protmode(vcpu))
461 			has_error = false;
462 		if (reinject) {
463 			/*
464 			 * On vmentry, vcpu->arch.exception.pending is only
465 			 * true if an event injection was blocked by
466 			 * nested_run_pending.  In that case, however,
467 			 * vcpu_enter_guest requests an immediate exit,
468 			 * and the guest shouldn't proceed far enough to
469 			 * need reinjection.
470 			 */
471 			WARN_ON_ONCE(vcpu->arch.exception.pending);
472 			vcpu->arch.exception.injected = true;
473 			if (WARN_ON_ONCE(has_payload)) {
474 				/*
475 				 * A reinjected event has already
476 				 * delivered its payload.
477 				 */
478 				has_payload = false;
479 				payload = 0;
480 			}
481 		} else {
482 			vcpu->arch.exception.pending = true;
483 			vcpu->arch.exception.injected = false;
484 		}
485 		vcpu->arch.exception.has_error_code = has_error;
486 		vcpu->arch.exception.nr = nr;
487 		vcpu->arch.exception.error_code = error_code;
488 		vcpu->arch.exception.has_payload = has_payload;
489 		vcpu->arch.exception.payload = payload;
490 		/*
491 		 * In guest mode, payload delivery should be deferred,
492 		 * so that the L1 hypervisor can intercept #PF before
493 		 * CR2 is modified (or intercept #DB before DR6 is
494 		 * modified under nVMX).  However, for ABI
495 		 * compatibility with KVM_GET_VCPU_EVENTS and
496 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
497 		 * delivery unless userspace has enabled this
498 		 * functionality via the per-VM capability,
499 		 * KVM_CAP_EXCEPTION_PAYLOAD.
500 		 */
501 		if (!vcpu->kvm->arch.exception_payload_enabled ||
502 		    !is_guest_mode(vcpu))
503 			kvm_deliver_exception_payload(vcpu);
504 		return;
505 	}
506 
507 	/* to check exception */
508 	prev_nr = vcpu->arch.exception.nr;
509 	if (prev_nr == DF_VECTOR) {
510 		/* triple fault -> shutdown */
511 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
512 		return;
513 	}
514 	class1 = exception_class(prev_nr);
515 	class2 = exception_class(nr);
516 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
517 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
518 		/*
519 		 * Generate double fault per SDM Table 5-5.  Set
520 		 * exception.pending = true so that the double fault
521 		 * can trigger a nested vmexit.
522 		 */
523 		vcpu->arch.exception.pending = true;
524 		vcpu->arch.exception.injected = false;
525 		vcpu->arch.exception.has_error_code = true;
526 		vcpu->arch.exception.nr = DF_VECTOR;
527 		vcpu->arch.exception.error_code = 0;
528 		vcpu->arch.exception.has_payload = false;
529 		vcpu->arch.exception.payload = 0;
530 	} else
531 		/* replace previous exception with a new one in a hope
532 		   that instruction re-execution will regenerate lost
533 		   exception */
534 		goto queue;
535 }
536 
537 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
538 {
539 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
540 }
541 EXPORT_SYMBOL_GPL(kvm_queue_exception);
542 
543 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
544 {
545 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
546 }
547 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
548 
549 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
550 				  unsigned long payload)
551 {
552 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
553 }
554 
555 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
556 				    u32 error_code, unsigned long payload)
557 {
558 	kvm_multiple_exception(vcpu, nr, true, error_code,
559 			       true, payload, false);
560 }
561 
562 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
563 {
564 	if (err)
565 		kvm_inject_gp(vcpu, 0);
566 	else
567 		return kvm_skip_emulated_instruction(vcpu);
568 
569 	return 1;
570 }
571 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
572 
573 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
574 {
575 	++vcpu->stat.pf_guest;
576 	vcpu->arch.exception.nested_apf =
577 		is_guest_mode(vcpu) && fault->async_page_fault;
578 	if (vcpu->arch.exception.nested_apf) {
579 		vcpu->arch.apf.nested_apf_token = fault->address;
580 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
581 	} else {
582 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
583 					fault->address);
584 	}
585 }
586 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
587 
588 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
589 {
590 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
591 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
592 	else
593 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
594 
595 	return fault->nested_page_fault;
596 }
597 
598 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
599 {
600 	atomic_inc(&vcpu->arch.nmi_queued);
601 	kvm_make_request(KVM_REQ_NMI, vcpu);
602 }
603 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
604 
605 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
606 {
607 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
608 }
609 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
610 
611 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
612 {
613 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
614 }
615 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
616 
617 /*
618  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
619  * a #GP and return false.
620  */
621 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
622 {
623 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
624 		return true;
625 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
626 	return false;
627 }
628 EXPORT_SYMBOL_GPL(kvm_require_cpl);
629 
630 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
631 {
632 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
633 		return true;
634 
635 	kvm_queue_exception(vcpu, UD_VECTOR);
636 	return false;
637 }
638 EXPORT_SYMBOL_GPL(kvm_require_dr);
639 
640 /*
641  * This function will be used to read from the physical memory of the currently
642  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
643  * can read from guest physical or from the guest's guest physical memory.
644  */
645 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
646 			    gfn_t ngfn, void *data, int offset, int len,
647 			    u32 access)
648 {
649 	struct x86_exception exception;
650 	gfn_t real_gfn;
651 	gpa_t ngpa;
652 
653 	ngpa     = gfn_to_gpa(ngfn);
654 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
655 	if (real_gfn == UNMAPPED_GVA)
656 		return -EFAULT;
657 
658 	real_gfn = gpa_to_gfn(real_gfn);
659 
660 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
661 }
662 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
663 
664 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
665 			       void *data, int offset, int len, u32 access)
666 {
667 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
668 				       data, offset, len, access);
669 }
670 
671 /*
672  * Load the pae pdptrs.  Return true is they are all valid.
673  */
674 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
675 {
676 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
677 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
678 	int i;
679 	int ret;
680 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
681 
682 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
683 				      offset * sizeof(u64), sizeof(pdpte),
684 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
685 	if (ret < 0) {
686 		ret = 0;
687 		goto out;
688 	}
689 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
690 		if ((pdpte[i] & PT_PRESENT_MASK) &&
691 		    (pdpte[i] &
692 		     vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
693 			ret = 0;
694 			goto out;
695 		}
696 	}
697 	ret = 1;
698 
699 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
700 	__set_bit(VCPU_EXREG_PDPTR,
701 		  (unsigned long *)&vcpu->arch.regs_avail);
702 	__set_bit(VCPU_EXREG_PDPTR,
703 		  (unsigned long *)&vcpu->arch.regs_dirty);
704 out:
705 
706 	return ret;
707 }
708 EXPORT_SYMBOL_GPL(load_pdptrs);
709 
710 bool pdptrs_changed(struct kvm_vcpu *vcpu)
711 {
712 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
713 	bool changed = true;
714 	int offset;
715 	gfn_t gfn;
716 	int r;
717 
718 	if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
719 		return false;
720 
721 	if (!test_bit(VCPU_EXREG_PDPTR,
722 		      (unsigned long *)&vcpu->arch.regs_avail))
723 		return true;
724 
725 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
726 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
727 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
728 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
729 	if (r < 0)
730 		goto out;
731 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
732 out:
733 
734 	return changed;
735 }
736 EXPORT_SYMBOL_GPL(pdptrs_changed);
737 
738 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
739 {
740 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
741 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
742 
743 	cr0 |= X86_CR0_ET;
744 
745 #ifdef CONFIG_X86_64
746 	if (cr0 & 0xffffffff00000000UL)
747 		return 1;
748 #endif
749 
750 	cr0 &= ~CR0_RESERVED_BITS;
751 
752 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
753 		return 1;
754 
755 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
756 		return 1;
757 
758 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
759 #ifdef CONFIG_X86_64
760 		if ((vcpu->arch.efer & EFER_LME)) {
761 			int cs_db, cs_l;
762 
763 			if (!is_pae(vcpu))
764 				return 1;
765 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
766 			if (cs_l)
767 				return 1;
768 		} else
769 #endif
770 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
771 						 kvm_read_cr3(vcpu)))
772 			return 1;
773 	}
774 
775 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
776 		return 1;
777 
778 	kvm_x86_ops->set_cr0(vcpu, cr0);
779 
780 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
781 		kvm_clear_async_pf_completion_queue(vcpu);
782 		kvm_async_pf_hash_reset(vcpu);
783 	}
784 
785 	if ((cr0 ^ old_cr0) & update_bits)
786 		kvm_mmu_reset_context(vcpu);
787 
788 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
789 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
790 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
791 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
792 
793 	return 0;
794 }
795 EXPORT_SYMBOL_GPL(kvm_set_cr0);
796 
797 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
798 {
799 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
800 }
801 EXPORT_SYMBOL_GPL(kvm_lmsw);
802 
803 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
804 {
805 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
806 			!vcpu->guest_xcr0_loaded) {
807 		/* kvm_set_xcr() also depends on this */
808 		if (vcpu->arch.xcr0 != host_xcr0)
809 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
810 		vcpu->guest_xcr0_loaded = 1;
811 	}
812 }
813 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
814 
815 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
816 {
817 	if (vcpu->guest_xcr0_loaded) {
818 		if (vcpu->arch.xcr0 != host_xcr0)
819 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
820 		vcpu->guest_xcr0_loaded = 0;
821 	}
822 }
823 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
824 
825 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
826 {
827 	u64 xcr0 = xcr;
828 	u64 old_xcr0 = vcpu->arch.xcr0;
829 	u64 valid_bits;
830 
831 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
832 	if (index != XCR_XFEATURE_ENABLED_MASK)
833 		return 1;
834 	if (!(xcr0 & XFEATURE_MASK_FP))
835 		return 1;
836 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
837 		return 1;
838 
839 	/*
840 	 * Do not allow the guest to set bits that we do not support
841 	 * saving.  However, xcr0 bit 0 is always set, even if the
842 	 * emulated CPU does not support XSAVE (see fx_init).
843 	 */
844 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
845 	if (xcr0 & ~valid_bits)
846 		return 1;
847 
848 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
849 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
850 		return 1;
851 
852 	if (xcr0 & XFEATURE_MASK_AVX512) {
853 		if (!(xcr0 & XFEATURE_MASK_YMM))
854 			return 1;
855 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
856 			return 1;
857 	}
858 	vcpu->arch.xcr0 = xcr0;
859 
860 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
861 		kvm_update_cpuid(vcpu);
862 	return 0;
863 }
864 
865 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
866 {
867 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
868 	    __kvm_set_xcr(vcpu, index, xcr)) {
869 		kvm_inject_gp(vcpu, 0);
870 		return 1;
871 	}
872 	return 0;
873 }
874 EXPORT_SYMBOL_GPL(kvm_set_xcr);
875 
876 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
877 {
878 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
879 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
880 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
881 
882 	if (cr4 & CR4_RESERVED_BITS)
883 		return 1;
884 
885 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
886 		return 1;
887 
888 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
889 		return 1;
890 
891 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
892 		return 1;
893 
894 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
895 		return 1;
896 
897 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
898 		return 1;
899 
900 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
901 		return 1;
902 
903 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
904 		return 1;
905 
906 	if (is_long_mode(vcpu)) {
907 		if (!(cr4 & X86_CR4_PAE))
908 			return 1;
909 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
910 		   && ((cr4 ^ old_cr4) & pdptr_bits)
911 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
912 				   kvm_read_cr3(vcpu)))
913 		return 1;
914 
915 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
916 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
917 			return 1;
918 
919 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
920 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
921 			return 1;
922 	}
923 
924 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
925 		return 1;
926 
927 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
928 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
929 		kvm_mmu_reset_context(vcpu);
930 
931 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
932 		kvm_update_cpuid(vcpu);
933 
934 	return 0;
935 }
936 EXPORT_SYMBOL_GPL(kvm_set_cr4);
937 
938 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
939 {
940 	bool skip_tlb_flush = false;
941 #ifdef CONFIG_X86_64
942 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
943 
944 	if (pcid_enabled) {
945 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
946 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
947 	}
948 #endif
949 
950 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
951 		if (!skip_tlb_flush) {
952 			kvm_mmu_sync_roots(vcpu);
953 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
954 		}
955 		return 0;
956 	}
957 
958 	if (is_long_mode(vcpu) &&
959 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
960 		return 1;
961 	else if (is_pae(vcpu) && is_paging(vcpu) &&
962 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
963 		return 1;
964 
965 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
966 	vcpu->arch.cr3 = cr3;
967 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
968 
969 	return 0;
970 }
971 EXPORT_SYMBOL_GPL(kvm_set_cr3);
972 
973 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
974 {
975 	if (cr8 & CR8_RESERVED_BITS)
976 		return 1;
977 	if (lapic_in_kernel(vcpu))
978 		kvm_lapic_set_tpr(vcpu, cr8);
979 	else
980 		vcpu->arch.cr8 = cr8;
981 	return 0;
982 }
983 EXPORT_SYMBOL_GPL(kvm_set_cr8);
984 
985 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
986 {
987 	if (lapic_in_kernel(vcpu))
988 		return kvm_lapic_get_cr8(vcpu);
989 	else
990 		return vcpu->arch.cr8;
991 }
992 EXPORT_SYMBOL_GPL(kvm_get_cr8);
993 
994 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
995 {
996 	int i;
997 
998 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
999 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1000 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1001 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1002 	}
1003 }
1004 
1005 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1006 {
1007 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1008 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1009 }
1010 
1011 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1012 {
1013 	unsigned long dr7;
1014 
1015 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1016 		dr7 = vcpu->arch.guest_debug_dr7;
1017 	else
1018 		dr7 = vcpu->arch.dr7;
1019 	kvm_x86_ops->set_dr7(vcpu, dr7);
1020 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1021 	if (dr7 & DR7_BP_EN_MASK)
1022 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1023 }
1024 
1025 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1026 {
1027 	u64 fixed = DR6_FIXED_1;
1028 
1029 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1030 		fixed |= DR6_RTM;
1031 	return fixed;
1032 }
1033 
1034 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1035 {
1036 	switch (dr) {
1037 	case 0 ... 3:
1038 		vcpu->arch.db[dr] = val;
1039 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1040 			vcpu->arch.eff_db[dr] = val;
1041 		break;
1042 	case 4:
1043 		/* fall through */
1044 	case 6:
1045 		if (val & 0xffffffff00000000ULL)
1046 			return -1; /* #GP */
1047 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1048 		kvm_update_dr6(vcpu);
1049 		break;
1050 	case 5:
1051 		/* fall through */
1052 	default: /* 7 */
1053 		if (val & 0xffffffff00000000ULL)
1054 			return -1; /* #GP */
1055 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1056 		kvm_update_dr7(vcpu);
1057 		break;
1058 	}
1059 
1060 	return 0;
1061 }
1062 
1063 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1064 {
1065 	if (__kvm_set_dr(vcpu, dr, val)) {
1066 		kvm_inject_gp(vcpu, 0);
1067 		return 1;
1068 	}
1069 	return 0;
1070 }
1071 EXPORT_SYMBOL_GPL(kvm_set_dr);
1072 
1073 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1074 {
1075 	switch (dr) {
1076 	case 0 ... 3:
1077 		*val = vcpu->arch.db[dr];
1078 		break;
1079 	case 4:
1080 		/* fall through */
1081 	case 6:
1082 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1083 			*val = vcpu->arch.dr6;
1084 		else
1085 			*val = kvm_x86_ops->get_dr6(vcpu);
1086 		break;
1087 	case 5:
1088 		/* fall through */
1089 	default: /* 7 */
1090 		*val = vcpu->arch.dr7;
1091 		break;
1092 	}
1093 	return 0;
1094 }
1095 EXPORT_SYMBOL_GPL(kvm_get_dr);
1096 
1097 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1098 {
1099 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1100 	u64 data;
1101 	int err;
1102 
1103 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1104 	if (err)
1105 		return err;
1106 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1107 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1108 	return err;
1109 }
1110 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1111 
1112 /*
1113  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1114  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1115  *
1116  * This list is modified at module load time to reflect the
1117  * capabilities of the host cpu. This capabilities test skips MSRs that are
1118  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1119  * may depend on host virtualization features rather than host cpu features.
1120  */
1121 
1122 static u32 msrs_to_save[] = {
1123 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1124 	MSR_STAR,
1125 #ifdef CONFIG_X86_64
1126 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1127 #endif
1128 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1129 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1130 	MSR_IA32_SPEC_CTRL,
1131 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1132 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1133 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1134 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1135 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1136 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1137 };
1138 
1139 static unsigned num_msrs_to_save;
1140 
1141 static u32 emulated_msrs[] = {
1142 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1143 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1144 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1145 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1146 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1147 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1148 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1149 	HV_X64_MSR_RESET,
1150 	HV_X64_MSR_VP_INDEX,
1151 	HV_X64_MSR_VP_RUNTIME,
1152 	HV_X64_MSR_SCONTROL,
1153 	HV_X64_MSR_STIMER0_CONFIG,
1154 	HV_X64_MSR_VP_ASSIST_PAGE,
1155 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1156 	HV_X64_MSR_TSC_EMULATION_STATUS,
1157 
1158 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1159 	MSR_KVM_PV_EOI_EN,
1160 
1161 	MSR_IA32_TSC_ADJUST,
1162 	MSR_IA32_TSCDEADLINE,
1163 	MSR_IA32_ARCH_CAPABILITIES,
1164 	MSR_IA32_MISC_ENABLE,
1165 	MSR_IA32_MCG_STATUS,
1166 	MSR_IA32_MCG_CTL,
1167 	MSR_IA32_MCG_EXT_CTL,
1168 	MSR_IA32_SMBASE,
1169 	MSR_SMI_COUNT,
1170 	MSR_PLATFORM_INFO,
1171 	MSR_MISC_FEATURES_ENABLES,
1172 	MSR_AMD64_VIRT_SPEC_CTRL,
1173 };
1174 
1175 static unsigned num_emulated_msrs;
1176 
1177 /*
1178  * List of msr numbers which are used to expose MSR-based features that
1179  * can be used by a hypervisor to validate requested CPU features.
1180  */
1181 static u32 msr_based_features[] = {
1182 	MSR_IA32_VMX_BASIC,
1183 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1184 	MSR_IA32_VMX_PINBASED_CTLS,
1185 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1186 	MSR_IA32_VMX_PROCBASED_CTLS,
1187 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1188 	MSR_IA32_VMX_EXIT_CTLS,
1189 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1190 	MSR_IA32_VMX_ENTRY_CTLS,
1191 	MSR_IA32_VMX_MISC,
1192 	MSR_IA32_VMX_CR0_FIXED0,
1193 	MSR_IA32_VMX_CR0_FIXED1,
1194 	MSR_IA32_VMX_CR4_FIXED0,
1195 	MSR_IA32_VMX_CR4_FIXED1,
1196 	MSR_IA32_VMX_VMCS_ENUM,
1197 	MSR_IA32_VMX_PROCBASED_CTLS2,
1198 	MSR_IA32_VMX_EPT_VPID_CAP,
1199 	MSR_IA32_VMX_VMFUNC,
1200 
1201 	MSR_F10H_DECFG,
1202 	MSR_IA32_UCODE_REV,
1203 	MSR_IA32_ARCH_CAPABILITIES,
1204 };
1205 
1206 static unsigned int num_msr_based_features;
1207 
1208 u64 kvm_get_arch_capabilities(void)
1209 {
1210 	u64 data;
1211 
1212 	rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1213 
1214 	/*
1215 	 * If we're doing cache flushes (either "always" or "cond")
1216 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1217 	 * If an outer hypervisor is doing the cache flush for us
1218 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1219 	 * capability to the guest too, and if EPT is disabled we're not
1220 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1221 	 * require a nested hypervisor to do a flush of its own.
1222 	 */
1223 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1224 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1225 
1226 	return data;
1227 }
1228 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1229 
1230 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1231 {
1232 	switch (msr->index) {
1233 	case MSR_IA32_ARCH_CAPABILITIES:
1234 		msr->data = kvm_get_arch_capabilities();
1235 		break;
1236 	case MSR_IA32_UCODE_REV:
1237 		rdmsrl_safe(msr->index, &msr->data);
1238 		break;
1239 	default:
1240 		if (kvm_x86_ops->get_msr_feature(msr))
1241 			return 1;
1242 	}
1243 	return 0;
1244 }
1245 
1246 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1247 {
1248 	struct kvm_msr_entry msr;
1249 	int r;
1250 
1251 	msr.index = index;
1252 	r = kvm_get_msr_feature(&msr);
1253 	if (r)
1254 		return r;
1255 
1256 	*data = msr.data;
1257 
1258 	return 0;
1259 }
1260 
1261 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1262 {
1263 	if (efer & efer_reserved_bits)
1264 		return false;
1265 
1266 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1267 			return false;
1268 
1269 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1270 			return false;
1271 
1272 	return true;
1273 }
1274 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1275 
1276 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1277 {
1278 	u64 old_efer = vcpu->arch.efer;
1279 
1280 	if (!kvm_valid_efer(vcpu, efer))
1281 		return 1;
1282 
1283 	if (is_paging(vcpu)
1284 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1285 		return 1;
1286 
1287 	efer &= ~EFER_LMA;
1288 	efer |= vcpu->arch.efer & EFER_LMA;
1289 
1290 	kvm_x86_ops->set_efer(vcpu, efer);
1291 
1292 	/* Update reserved bits */
1293 	if ((efer ^ old_efer) & EFER_NX)
1294 		kvm_mmu_reset_context(vcpu);
1295 
1296 	return 0;
1297 }
1298 
1299 void kvm_enable_efer_bits(u64 mask)
1300 {
1301        efer_reserved_bits &= ~mask;
1302 }
1303 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1304 
1305 /*
1306  * Writes msr value into into the appropriate "register".
1307  * Returns 0 on success, non-0 otherwise.
1308  * Assumes vcpu_load() was already called.
1309  */
1310 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1311 {
1312 	switch (msr->index) {
1313 	case MSR_FS_BASE:
1314 	case MSR_GS_BASE:
1315 	case MSR_KERNEL_GS_BASE:
1316 	case MSR_CSTAR:
1317 	case MSR_LSTAR:
1318 		if (is_noncanonical_address(msr->data, vcpu))
1319 			return 1;
1320 		break;
1321 	case MSR_IA32_SYSENTER_EIP:
1322 	case MSR_IA32_SYSENTER_ESP:
1323 		/*
1324 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1325 		 * non-canonical address is written on Intel but not on
1326 		 * AMD (which ignores the top 32-bits, because it does
1327 		 * not implement 64-bit SYSENTER).
1328 		 *
1329 		 * 64-bit code should hence be able to write a non-canonical
1330 		 * value on AMD.  Making the address canonical ensures that
1331 		 * vmentry does not fail on Intel after writing a non-canonical
1332 		 * value, and that something deterministic happens if the guest
1333 		 * invokes 64-bit SYSENTER.
1334 		 */
1335 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1336 	}
1337 	return kvm_x86_ops->set_msr(vcpu, msr);
1338 }
1339 EXPORT_SYMBOL_GPL(kvm_set_msr);
1340 
1341 /*
1342  * Adapt set_msr() to msr_io()'s calling convention
1343  */
1344 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1345 {
1346 	struct msr_data msr;
1347 	int r;
1348 
1349 	msr.index = index;
1350 	msr.host_initiated = true;
1351 	r = kvm_get_msr(vcpu, &msr);
1352 	if (r)
1353 		return r;
1354 
1355 	*data = msr.data;
1356 	return 0;
1357 }
1358 
1359 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1360 {
1361 	struct msr_data msr;
1362 
1363 	msr.data = *data;
1364 	msr.index = index;
1365 	msr.host_initiated = true;
1366 	return kvm_set_msr(vcpu, &msr);
1367 }
1368 
1369 #ifdef CONFIG_X86_64
1370 struct pvclock_gtod_data {
1371 	seqcount_t	seq;
1372 
1373 	struct { /* extract of a clocksource struct */
1374 		int vclock_mode;
1375 		u64	cycle_last;
1376 		u64	mask;
1377 		u32	mult;
1378 		u32	shift;
1379 	} clock;
1380 
1381 	u64		boot_ns;
1382 	u64		nsec_base;
1383 	u64		wall_time_sec;
1384 };
1385 
1386 static struct pvclock_gtod_data pvclock_gtod_data;
1387 
1388 static void update_pvclock_gtod(struct timekeeper *tk)
1389 {
1390 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1391 	u64 boot_ns;
1392 
1393 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1394 
1395 	write_seqcount_begin(&vdata->seq);
1396 
1397 	/* copy pvclock gtod data */
1398 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1399 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1400 	vdata->clock.mask		= tk->tkr_mono.mask;
1401 	vdata->clock.mult		= tk->tkr_mono.mult;
1402 	vdata->clock.shift		= tk->tkr_mono.shift;
1403 
1404 	vdata->boot_ns			= boot_ns;
1405 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1406 
1407 	vdata->wall_time_sec            = tk->xtime_sec;
1408 
1409 	write_seqcount_end(&vdata->seq);
1410 }
1411 #endif
1412 
1413 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1414 {
1415 	/*
1416 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1417 	 * vcpu_enter_guest.  This function is only called from
1418 	 * the physical CPU that is running vcpu.
1419 	 */
1420 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1421 }
1422 
1423 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1424 {
1425 	int version;
1426 	int r;
1427 	struct pvclock_wall_clock wc;
1428 	struct timespec64 boot;
1429 
1430 	if (!wall_clock)
1431 		return;
1432 
1433 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1434 	if (r)
1435 		return;
1436 
1437 	if (version & 1)
1438 		++version;  /* first time write, random junk */
1439 
1440 	++version;
1441 
1442 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1443 		return;
1444 
1445 	/*
1446 	 * The guest calculates current wall clock time by adding
1447 	 * system time (updated by kvm_guest_time_update below) to the
1448 	 * wall clock specified here.  guest system time equals host
1449 	 * system time for us, thus we must fill in host boot time here.
1450 	 */
1451 	getboottime64(&boot);
1452 
1453 	if (kvm->arch.kvmclock_offset) {
1454 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1455 		boot = timespec64_sub(boot, ts);
1456 	}
1457 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1458 	wc.nsec = boot.tv_nsec;
1459 	wc.version = version;
1460 
1461 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1462 
1463 	version++;
1464 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1465 }
1466 
1467 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1468 {
1469 	do_shl32_div32(dividend, divisor);
1470 	return dividend;
1471 }
1472 
1473 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1474 			       s8 *pshift, u32 *pmultiplier)
1475 {
1476 	uint64_t scaled64;
1477 	int32_t  shift = 0;
1478 	uint64_t tps64;
1479 	uint32_t tps32;
1480 
1481 	tps64 = base_hz;
1482 	scaled64 = scaled_hz;
1483 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1484 		tps64 >>= 1;
1485 		shift--;
1486 	}
1487 
1488 	tps32 = (uint32_t)tps64;
1489 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1490 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1491 			scaled64 >>= 1;
1492 		else
1493 			tps32 <<= 1;
1494 		shift++;
1495 	}
1496 
1497 	*pshift = shift;
1498 	*pmultiplier = div_frac(scaled64, tps32);
1499 
1500 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1501 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1502 }
1503 
1504 #ifdef CONFIG_X86_64
1505 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1506 #endif
1507 
1508 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1509 static unsigned long max_tsc_khz;
1510 
1511 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1512 {
1513 	u64 v = (u64)khz * (1000000 + ppm);
1514 	do_div(v, 1000000);
1515 	return v;
1516 }
1517 
1518 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1519 {
1520 	u64 ratio;
1521 
1522 	/* Guest TSC same frequency as host TSC? */
1523 	if (!scale) {
1524 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1525 		return 0;
1526 	}
1527 
1528 	/* TSC scaling supported? */
1529 	if (!kvm_has_tsc_control) {
1530 		if (user_tsc_khz > tsc_khz) {
1531 			vcpu->arch.tsc_catchup = 1;
1532 			vcpu->arch.tsc_always_catchup = 1;
1533 			return 0;
1534 		} else {
1535 			WARN(1, "user requested TSC rate below hardware speed\n");
1536 			return -1;
1537 		}
1538 	}
1539 
1540 	/* TSC scaling required  - calculate ratio */
1541 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1542 				user_tsc_khz, tsc_khz);
1543 
1544 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1545 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1546 			  user_tsc_khz);
1547 		return -1;
1548 	}
1549 
1550 	vcpu->arch.tsc_scaling_ratio = ratio;
1551 	return 0;
1552 }
1553 
1554 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1555 {
1556 	u32 thresh_lo, thresh_hi;
1557 	int use_scaling = 0;
1558 
1559 	/* tsc_khz can be zero if TSC calibration fails */
1560 	if (user_tsc_khz == 0) {
1561 		/* set tsc_scaling_ratio to a safe value */
1562 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1563 		return -1;
1564 	}
1565 
1566 	/* Compute a scale to convert nanoseconds in TSC cycles */
1567 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1568 			   &vcpu->arch.virtual_tsc_shift,
1569 			   &vcpu->arch.virtual_tsc_mult);
1570 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1571 
1572 	/*
1573 	 * Compute the variation in TSC rate which is acceptable
1574 	 * within the range of tolerance and decide if the
1575 	 * rate being applied is within that bounds of the hardware
1576 	 * rate.  If so, no scaling or compensation need be done.
1577 	 */
1578 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1579 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1580 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1581 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1582 		use_scaling = 1;
1583 	}
1584 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1585 }
1586 
1587 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1588 {
1589 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1590 				      vcpu->arch.virtual_tsc_mult,
1591 				      vcpu->arch.virtual_tsc_shift);
1592 	tsc += vcpu->arch.this_tsc_write;
1593 	return tsc;
1594 }
1595 
1596 static inline int gtod_is_based_on_tsc(int mode)
1597 {
1598 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1599 }
1600 
1601 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1602 {
1603 #ifdef CONFIG_X86_64
1604 	bool vcpus_matched;
1605 	struct kvm_arch *ka = &vcpu->kvm->arch;
1606 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1607 
1608 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1609 			 atomic_read(&vcpu->kvm->online_vcpus));
1610 
1611 	/*
1612 	 * Once the masterclock is enabled, always perform request in
1613 	 * order to update it.
1614 	 *
1615 	 * In order to enable masterclock, the host clocksource must be TSC
1616 	 * and the vcpus need to have matched TSCs.  When that happens,
1617 	 * perform request to enable masterclock.
1618 	 */
1619 	if (ka->use_master_clock ||
1620 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1621 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1622 
1623 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1624 			    atomic_read(&vcpu->kvm->online_vcpus),
1625 		            ka->use_master_clock, gtod->clock.vclock_mode);
1626 #endif
1627 }
1628 
1629 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1630 {
1631 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1632 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1633 }
1634 
1635 /*
1636  * Multiply tsc by a fixed point number represented by ratio.
1637  *
1638  * The most significant 64-N bits (mult) of ratio represent the
1639  * integral part of the fixed point number; the remaining N bits
1640  * (frac) represent the fractional part, ie. ratio represents a fixed
1641  * point number (mult + frac * 2^(-N)).
1642  *
1643  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1644  */
1645 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1646 {
1647 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1648 }
1649 
1650 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1651 {
1652 	u64 _tsc = tsc;
1653 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1654 
1655 	if (ratio != kvm_default_tsc_scaling_ratio)
1656 		_tsc = __scale_tsc(ratio, tsc);
1657 
1658 	return _tsc;
1659 }
1660 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1661 
1662 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1663 {
1664 	u64 tsc;
1665 
1666 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1667 
1668 	return target_tsc - tsc;
1669 }
1670 
1671 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1672 {
1673 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1674 
1675 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1676 }
1677 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1678 
1679 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1680 {
1681 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1682 }
1683 
1684 static inline bool kvm_check_tsc_unstable(void)
1685 {
1686 #ifdef CONFIG_X86_64
1687 	/*
1688 	 * TSC is marked unstable when we're running on Hyper-V,
1689 	 * 'TSC page' clocksource is good.
1690 	 */
1691 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1692 		return false;
1693 #endif
1694 	return check_tsc_unstable();
1695 }
1696 
1697 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1698 {
1699 	struct kvm *kvm = vcpu->kvm;
1700 	u64 offset, ns, elapsed;
1701 	unsigned long flags;
1702 	bool matched;
1703 	bool already_matched;
1704 	u64 data = msr->data;
1705 	bool synchronizing = false;
1706 
1707 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1708 	offset = kvm_compute_tsc_offset(vcpu, data);
1709 	ns = ktime_get_boot_ns();
1710 	elapsed = ns - kvm->arch.last_tsc_nsec;
1711 
1712 	if (vcpu->arch.virtual_tsc_khz) {
1713 		if (data == 0 && msr->host_initiated) {
1714 			/*
1715 			 * detection of vcpu initialization -- need to sync
1716 			 * with other vCPUs. This particularly helps to keep
1717 			 * kvm_clock stable after CPU hotplug
1718 			 */
1719 			synchronizing = true;
1720 		} else {
1721 			u64 tsc_exp = kvm->arch.last_tsc_write +
1722 						nsec_to_cycles(vcpu, elapsed);
1723 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1724 			/*
1725 			 * Special case: TSC write with a small delta (1 second)
1726 			 * of virtual cycle time against real time is
1727 			 * interpreted as an attempt to synchronize the CPU.
1728 			 */
1729 			synchronizing = data < tsc_exp + tsc_hz &&
1730 					data + tsc_hz > tsc_exp;
1731 		}
1732 	}
1733 
1734 	/*
1735 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1736 	 * TSC, we add elapsed time in this computation.  We could let the
1737 	 * compensation code attempt to catch up if we fall behind, but
1738 	 * it's better to try to match offsets from the beginning.
1739          */
1740 	if (synchronizing &&
1741 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1742 		if (!kvm_check_tsc_unstable()) {
1743 			offset = kvm->arch.cur_tsc_offset;
1744 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1745 		} else {
1746 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1747 			data += delta;
1748 			offset = kvm_compute_tsc_offset(vcpu, data);
1749 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1750 		}
1751 		matched = true;
1752 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1753 	} else {
1754 		/*
1755 		 * We split periods of matched TSC writes into generations.
1756 		 * For each generation, we track the original measured
1757 		 * nanosecond time, offset, and write, so if TSCs are in
1758 		 * sync, we can match exact offset, and if not, we can match
1759 		 * exact software computation in compute_guest_tsc()
1760 		 *
1761 		 * These values are tracked in kvm->arch.cur_xxx variables.
1762 		 */
1763 		kvm->arch.cur_tsc_generation++;
1764 		kvm->arch.cur_tsc_nsec = ns;
1765 		kvm->arch.cur_tsc_write = data;
1766 		kvm->arch.cur_tsc_offset = offset;
1767 		matched = false;
1768 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1769 			 kvm->arch.cur_tsc_generation, data);
1770 	}
1771 
1772 	/*
1773 	 * We also track th most recent recorded KHZ, write and time to
1774 	 * allow the matching interval to be extended at each write.
1775 	 */
1776 	kvm->arch.last_tsc_nsec = ns;
1777 	kvm->arch.last_tsc_write = data;
1778 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1779 
1780 	vcpu->arch.last_guest_tsc = data;
1781 
1782 	/* Keep track of which generation this VCPU has synchronized to */
1783 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1784 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1785 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1786 
1787 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1788 		update_ia32_tsc_adjust_msr(vcpu, offset);
1789 
1790 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1791 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1792 
1793 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1794 	if (!matched) {
1795 		kvm->arch.nr_vcpus_matched_tsc = 0;
1796 	} else if (!already_matched) {
1797 		kvm->arch.nr_vcpus_matched_tsc++;
1798 	}
1799 
1800 	kvm_track_tsc_matching(vcpu);
1801 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1802 }
1803 
1804 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1805 
1806 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1807 					   s64 adjustment)
1808 {
1809 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1810 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1811 }
1812 
1813 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1814 {
1815 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1816 		WARN_ON(adjustment < 0);
1817 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1818 	adjust_tsc_offset_guest(vcpu, adjustment);
1819 }
1820 
1821 #ifdef CONFIG_X86_64
1822 
1823 static u64 read_tsc(void)
1824 {
1825 	u64 ret = (u64)rdtsc_ordered();
1826 	u64 last = pvclock_gtod_data.clock.cycle_last;
1827 
1828 	if (likely(ret >= last))
1829 		return ret;
1830 
1831 	/*
1832 	 * GCC likes to generate cmov here, but this branch is extremely
1833 	 * predictable (it's just a function of time and the likely is
1834 	 * very likely) and there's a data dependence, so force GCC
1835 	 * to generate a branch instead.  I don't barrier() because
1836 	 * we don't actually need a barrier, and if this function
1837 	 * ever gets inlined it will generate worse code.
1838 	 */
1839 	asm volatile ("");
1840 	return last;
1841 }
1842 
1843 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1844 {
1845 	long v;
1846 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1847 	u64 tsc_pg_val;
1848 
1849 	switch (gtod->clock.vclock_mode) {
1850 	case VCLOCK_HVCLOCK:
1851 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1852 						  tsc_timestamp);
1853 		if (tsc_pg_val != U64_MAX) {
1854 			/* TSC page valid */
1855 			*mode = VCLOCK_HVCLOCK;
1856 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1857 				gtod->clock.mask;
1858 		} else {
1859 			/* TSC page invalid */
1860 			*mode = VCLOCK_NONE;
1861 		}
1862 		break;
1863 	case VCLOCK_TSC:
1864 		*mode = VCLOCK_TSC;
1865 		*tsc_timestamp = read_tsc();
1866 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1867 			gtod->clock.mask;
1868 		break;
1869 	default:
1870 		*mode = VCLOCK_NONE;
1871 	}
1872 
1873 	if (*mode == VCLOCK_NONE)
1874 		*tsc_timestamp = v = 0;
1875 
1876 	return v * gtod->clock.mult;
1877 }
1878 
1879 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1880 {
1881 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1882 	unsigned long seq;
1883 	int mode;
1884 	u64 ns;
1885 
1886 	do {
1887 		seq = read_seqcount_begin(&gtod->seq);
1888 		ns = gtod->nsec_base;
1889 		ns += vgettsc(tsc_timestamp, &mode);
1890 		ns >>= gtod->clock.shift;
1891 		ns += gtod->boot_ns;
1892 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1893 	*t = ns;
1894 
1895 	return mode;
1896 }
1897 
1898 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1899 {
1900 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1901 	unsigned long seq;
1902 	int mode;
1903 	u64 ns;
1904 
1905 	do {
1906 		seq = read_seqcount_begin(&gtod->seq);
1907 		ts->tv_sec = gtod->wall_time_sec;
1908 		ns = gtod->nsec_base;
1909 		ns += vgettsc(tsc_timestamp, &mode);
1910 		ns >>= gtod->clock.shift;
1911 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1912 
1913 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1914 	ts->tv_nsec = ns;
1915 
1916 	return mode;
1917 }
1918 
1919 /* returns true if host is using TSC based clocksource */
1920 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1921 {
1922 	/* checked again under seqlock below */
1923 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1924 		return false;
1925 
1926 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1927 						      tsc_timestamp));
1928 }
1929 
1930 /* returns true if host is using TSC based clocksource */
1931 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1932 					   u64 *tsc_timestamp)
1933 {
1934 	/* checked again under seqlock below */
1935 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1936 		return false;
1937 
1938 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1939 }
1940 #endif
1941 
1942 /*
1943  *
1944  * Assuming a stable TSC across physical CPUS, and a stable TSC
1945  * across virtual CPUs, the following condition is possible.
1946  * Each numbered line represents an event visible to both
1947  * CPUs at the next numbered event.
1948  *
1949  * "timespecX" represents host monotonic time. "tscX" represents
1950  * RDTSC value.
1951  *
1952  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1953  *
1954  * 1.  read timespec0,tsc0
1955  * 2.					| timespec1 = timespec0 + N
1956  * 					| tsc1 = tsc0 + M
1957  * 3. transition to guest		| transition to guest
1958  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1959  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1960  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1961  *
1962  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1963  *
1964  * 	- ret0 < ret1
1965  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1966  *		...
1967  *	- 0 < N - M => M < N
1968  *
1969  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1970  * always the case (the difference between two distinct xtime instances
1971  * might be smaller then the difference between corresponding TSC reads,
1972  * when updating guest vcpus pvclock areas).
1973  *
1974  * To avoid that problem, do not allow visibility of distinct
1975  * system_timestamp/tsc_timestamp values simultaneously: use a master
1976  * copy of host monotonic time values. Update that master copy
1977  * in lockstep.
1978  *
1979  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1980  *
1981  */
1982 
1983 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1984 {
1985 #ifdef CONFIG_X86_64
1986 	struct kvm_arch *ka = &kvm->arch;
1987 	int vclock_mode;
1988 	bool host_tsc_clocksource, vcpus_matched;
1989 
1990 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1991 			atomic_read(&kvm->online_vcpus));
1992 
1993 	/*
1994 	 * If the host uses TSC clock, then passthrough TSC as stable
1995 	 * to the guest.
1996 	 */
1997 	host_tsc_clocksource = kvm_get_time_and_clockread(
1998 					&ka->master_kernel_ns,
1999 					&ka->master_cycle_now);
2000 
2001 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2002 				&& !ka->backwards_tsc_observed
2003 				&& !ka->boot_vcpu_runs_old_kvmclock;
2004 
2005 	if (ka->use_master_clock)
2006 		atomic_set(&kvm_guest_has_master_clock, 1);
2007 
2008 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2009 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2010 					vcpus_matched);
2011 #endif
2012 }
2013 
2014 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2015 {
2016 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2017 }
2018 
2019 static void kvm_gen_update_masterclock(struct kvm *kvm)
2020 {
2021 #ifdef CONFIG_X86_64
2022 	int i;
2023 	struct kvm_vcpu *vcpu;
2024 	struct kvm_arch *ka = &kvm->arch;
2025 
2026 	spin_lock(&ka->pvclock_gtod_sync_lock);
2027 	kvm_make_mclock_inprogress_request(kvm);
2028 	/* no guest entries from this point */
2029 	pvclock_update_vm_gtod_copy(kvm);
2030 
2031 	kvm_for_each_vcpu(i, vcpu, kvm)
2032 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2033 
2034 	/* guest entries allowed */
2035 	kvm_for_each_vcpu(i, vcpu, kvm)
2036 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2037 
2038 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2039 #endif
2040 }
2041 
2042 u64 get_kvmclock_ns(struct kvm *kvm)
2043 {
2044 	struct kvm_arch *ka = &kvm->arch;
2045 	struct pvclock_vcpu_time_info hv_clock;
2046 	u64 ret;
2047 
2048 	spin_lock(&ka->pvclock_gtod_sync_lock);
2049 	if (!ka->use_master_clock) {
2050 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2051 		return ktime_get_boot_ns() + ka->kvmclock_offset;
2052 	}
2053 
2054 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2055 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2056 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2057 
2058 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2059 	get_cpu();
2060 
2061 	if (__this_cpu_read(cpu_tsc_khz)) {
2062 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2063 				   &hv_clock.tsc_shift,
2064 				   &hv_clock.tsc_to_system_mul);
2065 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2066 	} else
2067 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2068 
2069 	put_cpu();
2070 
2071 	return ret;
2072 }
2073 
2074 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2075 {
2076 	struct kvm_vcpu_arch *vcpu = &v->arch;
2077 	struct pvclock_vcpu_time_info guest_hv_clock;
2078 
2079 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2080 		&guest_hv_clock, sizeof(guest_hv_clock))))
2081 		return;
2082 
2083 	/* This VCPU is paused, but it's legal for a guest to read another
2084 	 * VCPU's kvmclock, so we really have to follow the specification where
2085 	 * it says that version is odd if data is being modified, and even after
2086 	 * it is consistent.
2087 	 *
2088 	 * Version field updates must be kept separate.  This is because
2089 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2090 	 * writes within a string instruction are weakly ordered.  So there
2091 	 * are three writes overall.
2092 	 *
2093 	 * As a small optimization, only write the version field in the first
2094 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2095 	 * version field is the first in the struct.
2096 	 */
2097 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2098 
2099 	if (guest_hv_clock.version & 1)
2100 		++guest_hv_clock.version;  /* first time write, random junk */
2101 
2102 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2103 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2104 				&vcpu->hv_clock,
2105 				sizeof(vcpu->hv_clock.version));
2106 
2107 	smp_wmb();
2108 
2109 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2110 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2111 
2112 	if (vcpu->pvclock_set_guest_stopped_request) {
2113 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2114 		vcpu->pvclock_set_guest_stopped_request = false;
2115 	}
2116 
2117 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2118 
2119 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2120 				&vcpu->hv_clock,
2121 				sizeof(vcpu->hv_clock));
2122 
2123 	smp_wmb();
2124 
2125 	vcpu->hv_clock.version++;
2126 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2127 				&vcpu->hv_clock,
2128 				sizeof(vcpu->hv_clock.version));
2129 }
2130 
2131 static int kvm_guest_time_update(struct kvm_vcpu *v)
2132 {
2133 	unsigned long flags, tgt_tsc_khz;
2134 	struct kvm_vcpu_arch *vcpu = &v->arch;
2135 	struct kvm_arch *ka = &v->kvm->arch;
2136 	s64 kernel_ns;
2137 	u64 tsc_timestamp, host_tsc;
2138 	u8 pvclock_flags;
2139 	bool use_master_clock;
2140 
2141 	kernel_ns = 0;
2142 	host_tsc = 0;
2143 
2144 	/*
2145 	 * If the host uses TSC clock, then passthrough TSC as stable
2146 	 * to the guest.
2147 	 */
2148 	spin_lock(&ka->pvclock_gtod_sync_lock);
2149 	use_master_clock = ka->use_master_clock;
2150 	if (use_master_clock) {
2151 		host_tsc = ka->master_cycle_now;
2152 		kernel_ns = ka->master_kernel_ns;
2153 	}
2154 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2155 
2156 	/* Keep irq disabled to prevent changes to the clock */
2157 	local_irq_save(flags);
2158 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2159 	if (unlikely(tgt_tsc_khz == 0)) {
2160 		local_irq_restore(flags);
2161 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2162 		return 1;
2163 	}
2164 	if (!use_master_clock) {
2165 		host_tsc = rdtsc();
2166 		kernel_ns = ktime_get_boot_ns();
2167 	}
2168 
2169 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2170 
2171 	/*
2172 	 * We may have to catch up the TSC to match elapsed wall clock
2173 	 * time for two reasons, even if kvmclock is used.
2174 	 *   1) CPU could have been running below the maximum TSC rate
2175 	 *   2) Broken TSC compensation resets the base at each VCPU
2176 	 *      entry to avoid unknown leaps of TSC even when running
2177 	 *      again on the same CPU.  This may cause apparent elapsed
2178 	 *      time to disappear, and the guest to stand still or run
2179 	 *	very slowly.
2180 	 */
2181 	if (vcpu->tsc_catchup) {
2182 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2183 		if (tsc > tsc_timestamp) {
2184 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2185 			tsc_timestamp = tsc;
2186 		}
2187 	}
2188 
2189 	local_irq_restore(flags);
2190 
2191 	/* With all the info we got, fill in the values */
2192 
2193 	if (kvm_has_tsc_control)
2194 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2195 
2196 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2197 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2198 				   &vcpu->hv_clock.tsc_shift,
2199 				   &vcpu->hv_clock.tsc_to_system_mul);
2200 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2201 	}
2202 
2203 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2204 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2205 	vcpu->last_guest_tsc = tsc_timestamp;
2206 
2207 	/* If the host uses TSC clocksource, then it is stable */
2208 	pvclock_flags = 0;
2209 	if (use_master_clock)
2210 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2211 
2212 	vcpu->hv_clock.flags = pvclock_flags;
2213 
2214 	if (vcpu->pv_time_enabled)
2215 		kvm_setup_pvclock_page(v);
2216 	if (v == kvm_get_vcpu(v->kvm, 0))
2217 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2218 	return 0;
2219 }
2220 
2221 /*
2222  * kvmclock updates which are isolated to a given vcpu, such as
2223  * vcpu->cpu migration, should not allow system_timestamp from
2224  * the rest of the vcpus to remain static. Otherwise ntp frequency
2225  * correction applies to one vcpu's system_timestamp but not
2226  * the others.
2227  *
2228  * So in those cases, request a kvmclock update for all vcpus.
2229  * We need to rate-limit these requests though, as they can
2230  * considerably slow guests that have a large number of vcpus.
2231  * The time for a remote vcpu to update its kvmclock is bound
2232  * by the delay we use to rate-limit the updates.
2233  */
2234 
2235 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2236 
2237 static void kvmclock_update_fn(struct work_struct *work)
2238 {
2239 	int i;
2240 	struct delayed_work *dwork = to_delayed_work(work);
2241 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2242 					   kvmclock_update_work);
2243 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2244 	struct kvm_vcpu *vcpu;
2245 
2246 	kvm_for_each_vcpu(i, vcpu, kvm) {
2247 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2248 		kvm_vcpu_kick(vcpu);
2249 	}
2250 }
2251 
2252 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2253 {
2254 	struct kvm *kvm = v->kvm;
2255 
2256 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2257 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2258 					KVMCLOCK_UPDATE_DELAY);
2259 }
2260 
2261 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2262 
2263 static void kvmclock_sync_fn(struct work_struct *work)
2264 {
2265 	struct delayed_work *dwork = to_delayed_work(work);
2266 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2267 					   kvmclock_sync_work);
2268 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2269 
2270 	if (!kvmclock_periodic_sync)
2271 		return;
2272 
2273 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2274 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2275 					KVMCLOCK_SYNC_PERIOD);
2276 }
2277 
2278 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2279 {
2280 	u64 mcg_cap = vcpu->arch.mcg_cap;
2281 	unsigned bank_num = mcg_cap & 0xff;
2282 	u32 msr = msr_info->index;
2283 	u64 data = msr_info->data;
2284 
2285 	switch (msr) {
2286 	case MSR_IA32_MCG_STATUS:
2287 		vcpu->arch.mcg_status = data;
2288 		break;
2289 	case MSR_IA32_MCG_CTL:
2290 		if (!(mcg_cap & MCG_CTL_P) &&
2291 		    (data || !msr_info->host_initiated))
2292 			return 1;
2293 		if (data != 0 && data != ~(u64)0)
2294 			return 1;
2295 		vcpu->arch.mcg_ctl = data;
2296 		break;
2297 	default:
2298 		if (msr >= MSR_IA32_MC0_CTL &&
2299 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2300 			u32 offset = msr - MSR_IA32_MC0_CTL;
2301 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2302 			 * some Linux kernels though clear bit 10 in bank 4 to
2303 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2304 			 * this to avoid an uncatched #GP in the guest
2305 			 */
2306 			if ((offset & 0x3) == 0 &&
2307 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2308 				return -1;
2309 			if (!msr_info->host_initiated &&
2310 				(offset & 0x3) == 1 && data != 0)
2311 				return -1;
2312 			vcpu->arch.mce_banks[offset] = data;
2313 			break;
2314 		}
2315 		return 1;
2316 	}
2317 	return 0;
2318 }
2319 
2320 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2321 {
2322 	struct kvm *kvm = vcpu->kvm;
2323 	int lm = is_long_mode(vcpu);
2324 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2325 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2326 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2327 		: kvm->arch.xen_hvm_config.blob_size_32;
2328 	u32 page_num = data & ~PAGE_MASK;
2329 	u64 page_addr = data & PAGE_MASK;
2330 	u8 *page;
2331 	int r;
2332 
2333 	r = -E2BIG;
2334 	if (page_num >= blob_size)
2335 		goto out;
2336 	r = -ENOMEM;
2337 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2338 	if (IS_ERR(page)) {
2339 		r = PTR_ERR(page);
2340 		goto out;
2341 	}
2342 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2343 		goto out_free;
2344 	r = 0;
2345 out_free:
2346 	kfree(page);
2347 out:
2348 	return r;
2349 }
2350 
2351 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2352 {
2353 	gpa_t gpa = data & ~0x3f;
2354 
2355 	/* Bits 3:5 are reserved, Should be zero */
2356 	if (data & 0x38)
2357 		return 1;
2358 
2359 	vcpu->arch.apf.msr_val = data;
2360 
2361 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2362 		kvm_clear_async_pf_completion_queue(vcpu);
2363 		kvm_async_pf_hash_reset(vcpu);
2364 		return 0;
2365 	}
2366 
2367 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2368 					sizeof(u32)))
2369 		return 1;
2370 
2371 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2372 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2373 	kvm_async_pf_wakeup_all(vcpu);
2374 	return 0;
2375 }
2376 
2377 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2378 {
2379 	vcpu->arch.pv_time_enabled = false;
2380 }
2381 
2382 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2383 {
2384 	++vcpu->stat.tlb_flush;
2385 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2386 }
2387 
2388 static void record_steal_time(struct kvm_vcpu *vcpu)
2389 {
2390 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2391 		return;
2392 
2393 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2394 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2395 		return;
2396 
2397 	/*
2398 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2399 	 * expensive IPIs.
2400 	 */
2401 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2402 		kvm_vcpu_flush_tlb(vcpu, false);
2403 
2404 	if (vcpu->arch.st.steal.version & 1)
2405 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2406 
2407 	vcpu->arch.st.steal.version += 1;
2408 
2409 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2410 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2411 
2412 	smp_wmb();
2413 
2414 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2415 		vcpu->arch.st.last_steal;
2416 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2417 
2418 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2419 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2420 
2421 	smp_wmb();
2422 
2423 	vcpu->arch.st.steal.version += 1;
2424 
2425 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2426 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2427 }
2428 
2429 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2430 {
2431 	bool pr = false;
2432 	u32 msr = msr_info->index;
2433 	u64 data = msr_info->data;
2434 
2435 	switch (msr) {
2436 	case MSR_AMD64_NB_CFG:
2437 	case MSR_IA32_UCODE_WRITE:
2438 	case MSR_VM_HSAVE_PA:
2439 	case MSR_AMD64_PATCH_LOADER:
2440 	case MSR_AMD64_BU_CFG2:
2441 	case MSR_AMD64_DC_CFG:
2442 	case MSR_F15H_EX_CFG:
2443 		break;
2444 
2445 	case MSR_IA32_UCODE_REV:
2446 		if (msr_info->host_initiated)
2447 			vcpu->arch.microcode_version = data;
2448 		break;
2449 	case MSR_IA32_ARCH_CAPABILITIES:
2450 		if (!msr_info->host_initiated)
2451 			return 1;
2452 		vcpu->arch.arch_capabilities = data;
2453 		break;
2454 	case MSR_EFER:
2455 		return set_efer(vcpu, data);
2456 	case MSR_K7_HWCR:
2457 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2458 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2459 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2460 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2461 		if (data != 0) {
2462 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2463 				    data);
2464 			return 1;
2465 		}
2466 		break;
2467 	case MSR_FAM10H_MMIO_CONF_BASE:
2468 		if (data != 0) {
2469 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2470 				    "0x%llx\n", data);
2471 			return 1;
2472 		}
2473 		break;
2474 	case MSR_IA32_DEBUGCTLMSR:
2475 		if (!data) {
2476 			/* We support the non-activated case already */
2477 			break;
2478 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2479 			/* Values other than LBR and BTF are vendor-specific,
2480 			   thus reserved and should throw a #GP */
2481 			return 1;
2482 		}
2483 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2484 			    __func__, data);
2485 		break;
2486 	case 0x200 ... 0x2ff:
2487 		return kvm_mtrr_set_msr(vcpu, msr, data);
2488 	case MSR_IA32_APICBASE:
2489 		return kvm_set_apic_base(vcpu, msr_info);
2490 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2491 		return kvm_x2apic_msr_write(vcpu, msr, data);
2492 	case MSR_IA32_TSCDEADLINE:
2493 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2494 		break;
2495 	case MSR_IA32_TSC_ADJUST:
2496 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2497 			if (!msr_info->host_initiated) {
2498 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2499 				adjust_tsc_offset_guest(vcpu, adj);
2500 			}
2501 			vcpu->arch.ia32_tsc_adjust_msr = data;
2502 		}
2503 		break;
2504 	case MSR_IA32_MISC_ENABLE:
2505 		vcpu->arch.ia32_misc_enable_msr = data;
2506 		break;
2507 	case MSR_IA32_SMBASE:
2508 		if (!msr_info->host_initiated)
2509 			return 1;
2510 		vcpu->arch.smbase = data;
2511 		break;
2512 	case MSR_IA32_TSC:
2513 		kvm_write_tsc(vcpu, msr_info);
2514 		break;
2515 	case MSR_SMI_COUNT:
2516 		if (!msr_info->host_initiated)
2517 			return 1;
2518 		vcpu->arch.smi_count = data;
2519 		break;
2520 	case MSR_KVM_WALL_CLOCK_NEW:
2521 	case MSR_KVM_WALL_CLOCK:
2522 		vcpu->kvm->arch.wall_clock = data;
2523 		kvm_write_wall_clock(vcpu->kvm, data);
2524 		break;
2525 	case MSR_KVM_SYSTEM_TIME_NEW:
2526 	case MSR_KVM_SYSTEM_TIME: {
2527 		struct kvm_arch *ka = &vcpu->kvm->arch;
2528 
2529 		kvmclock_reset(vcpu);
2530 
2531 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2532 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2533 
2534 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2535 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2536 
2537 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2538 		}
2539 
2540 		vcpu->arch.time = data;
2541 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2542 
2543 		/* we verify if the enable bit is set... */
2544 		if (!(data & 1))
2545 			break;
2546 
2547 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2548 		     &vcpu->arch.pv_time, data & ~1ULL,
2549 		     sizeof(struct pvclock_vcpu_time_info)))
2550 			vcpu->arch.pv_time_enabled = false;
2551 		else
2552 			vcpu->arch.pv_time_enabled = true;
2553 
2554 		break;
2555 	}
2556 	case MSR_KVM_ASYNC_PF_EN:
2557 		if (kvm_pv_enable_async_pf(vcpu, data))
2558 			return 1;
2559 		break;
2560 	case MSR_KVM_STEAL_TIME:
2561 
2562 		if (unlikely(!sched_info_on()))
2563 			return 1;
2564 
2565 		if (data & KVM_STEAL_RESERVED_MASK)
2566 			return 1;
2567 
2568 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2569 						data & KVM_STEAL_VALID_BITS,
2570 						sizeof(struct kvm_steal_time)))
2571 			return 1;
2572 
2573 		vcpu->arch.st.msr_val = data;
2574 
2575 		if (!(data & KVM_MSR_ENABLED))
2576 			break;
2577 
2578 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2579 
2580 		break;
2581 	case MSR_KVM_PV_EOI_EN:
2582 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2583 			return 1;
2584 		break;
2585 
2586 	case MSR_IA32_MCG_CTL:
2587 	case MSR_IA32_MCG_STATUS:
2588 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2589 		return set_msr_mce(vcpu, msr_info);
2590 
2591 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2592 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2593 		pr = true; /* fall through */
2594 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2595 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2596 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2597 			return kvm_pmu_set_msr(vcpu, msr_info);
2598 
2599 		if (pr || data != 0)
2600 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2601 				    "0x%x data 0x%llx\n", msr, data);
2602 		break;
2603 	case MSR_K7_CLK_CTL:
2604 		/*
2605 		 * Ignore all writes to this no longer documented MSR.
2606 		 * Writes are only relevant for old K7 processors,
2607 		 * all pre-dating SVM, but a recommended workaround from
2608 		 * AMD for these chips. It is possible to specify the
2609 		 * affected processor models on the command line, hence
2610 		 * the need to ignore the workaround.
2611 		 */
2612 		break;
2613 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2614 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2615 	case HV_X64_MSR_CRASH_CTL:
2616 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2617 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2618 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2619 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2620 		return kvm_hv_set_msr_common(vcpu, msr, data,
2621 					     msr_info->host_initiated);
2622 	case MSR_IA32_BBL_CR_CTL3:
2623 		/* Drop writes to this legacy MSR -- see rdmsr
2624 		 * counterpart for further detail.
2625 		 */
2626 		if (report_ignored_msrs)
2627 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2628 				msr, data);
2629 		break;
2630 	case MSR_AMD64_OSVW_ID_LENGTH:
2631 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2632 			return 1;
2633 		vcpu->arch.osvw.length = data;
2634 		break;
2635 	case MSR_AMD64_OSVW_STATUS:
2636 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2637 			return 1;
2638 		vcpu->arch.osvw.status = data;
2639 		break;
2640 	case MSR_PLATFORM_INFO:
2641 		if (!msr_info->host_initiated ||
2642 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2643 		     cpuid_fault_enabled(vcpu)))
2644 			return 1;
2645 		vcpu->arch.msr_platform_info = data;
2646 		break;
2647 	case MSR_MISC_FEATURES_ENABLES:
2648 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2649 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2650 		     !supports_cpuid_fault(vcpu)))
2651 			return 1;
2652 		vcpu->arch.msr_misc_features_enables = data;
2653 		break;
2654 	default:
2655 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2656 			return xen_hvm_config(vcpu, data);
2657 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2658 			return kvm_pmu_set_msr(vcpu, msr_info);
2659 		if (!ignore_msrs) {
2660 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2661 				    msr, data);
2662 			return 1;
2663 		} else {
2664 			if (report_ignored_msrs)
2665 				vcpu_unimpl(vcpu,
2666 					"ignored wrmsr: 0x%x data 0x%llx\n",
2667 					msr, data);
2668 			break;
2669 		}
2670 	}
2671 	return 0;
2672 }
2673 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2674 
2675 
2676 /*
2677  * Reads an msr value (of 'msr_index') into 'pdata'.
2678  * Returns 0 on success, non-0 otherwise.
2679  * Assumes vcpu_load() was already called.
2680  */
2681 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2682 {
2683 	return kvm_x86_ops->get_msr(vcpu, msr);
2684 }
2685 EXPORT_SYMBOL_GPL(kvm_get_msr);
2686 
2687 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2688 {
2689 	u64 data;
2690 	u64 mcg_cap = vcpu->arch.mcg_cap;
2691 	unsigned bank_num = mcg_cap & 0xff;
2692 
2693 	switch (msr) {
2694 	case MSR_IA32_P5_MC_ADDR:
2695 	case MSR_IA32_P5_MC_TYPE:
2696 		data = 0;
2697 		break;
2698 	case MSR_IA32_MCG_CAP:
2699 		data = vcpu->arch.mcg_cap;
2700 		break;
2701 	case MSR_IA32_MCG_CTL:
2702 		if (!(mcg_cap & MCG_CTL_P) && !host)
2703 			return 1;
2704 		data = vcpu->arch.mcg_ctl;
2705 		break;
2706 	case MSR_IA32_MCG_STATUS:
2707 		data = vcpu->arch.mcg_status;
2708 		break;
2709 	default:
2710 		if (msr >= MSR_IA32_MC0_CTL &&
2711 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2712 			u32 offset = msr - MSR_IA32_MC0_CTL;
2713 			data = vcpu->arch.mce_banks[offset];
2714 			break;
2715 		}
2716 		return 1;
2717 	}
2718 	*pdata = data;
2719 	return 0;
2720 }
2721 
2722 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2723 {
2724 	switch (msr_info->index) {
2725 	case MSR_IA32_PLATFORM_ID:
2726 	case MSR_IA32_EBL_CR_POWERON:
2727 	case MSR_IA32_DEBUGCTLMSR:
2728 	case MSR_IA32_LASTBRANCHFROMIP:
2729 	case MSR_IA32_LASTBRANCHTOIP:
2730 	case MSR_IA32_LASTINTFROMIP:
2731 	case MSR_IA32_LASTINTTOIP:
2732 	case MSR_K8_SYSCFG:
2733 	case MSR_K8_TSEG_ADDR:
2734 	case MSR_K8_TSEG_MASK:
2735 	case MSR_K7_HWCR:
2736 	case MSR_VM_HSAVE_PA:
2737 	case MSR_K8_INT_PENDING_MSG:
2738 	case MSR_AMD64_NB_CFG:
2739 	case MSR_FAM10H_MMIO_CONF_BASE:
2740 	case MSR_AMD64_BU_CFG2:
2741 	case MSR_IA32_PERF_CTL:
2742 	case MSR_AMD64_DC_CFG:
2743 	case MSR_F15H_EX_CFG:
2744 		msr_info->data = 0;
2745 		break;
2746 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2747 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2748 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2749 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2750 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2751 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2752 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2753 		msr_info->data = 0;
2754 		break;
2755 	case MSR_IA32_UCODE_REV:
2756 		msr_info->data = vcpu->arch.microcode_version;
2757 		break;
2758 	case MSR_IA32_ARCH_CAPABILITIES:
2759 		if (!msr_info->host_initiated &&
2760 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2761 			return 1;
2762 		msr_info->data = vcpu->arch.arch_capabilities;
2763 		break;
2764 	case MSR_IA32_TSC:
2765 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2766 		break;
2767 	case MSR_MTRRcap:
2768 	case 0x200 ... 0x2ff:
2769 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2770 	case 0xcd: /* fsb frequency */
2771 		msr_info->data = 3;
2772 		break;
2773 		/*
2774 		 * MSR_EBC_FREQUENCY_ID
2775 		 * Conservative value valid for even the basic CPU models.
2776 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2777 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2778 		 * and 266MHz for model 3, or 4. Set Core Clock
2779 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2780 		 * 31:24) even though these are only valid for CPU
2781 		 * models > 2, however guests may end up dividing or
2782 		 * multiplying by zero otherwise.
2783 		 */
2784 	case MSR_EBC_FREQUENCY_ID:
2785 		msr_info->data = 1 << 24;
2786 		break;
2787 	case MSR_IA32_APICBASE:
2788 		msr_info->data = kvm_get_apic_base(vcpu);
2789 		break;
2790 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2791 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2792 		break;
2793 	case MSR_IA32_TSCDEADLINE:
2794 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2795 		break;
2796 	case MSR_IA32_TSC_ADJUST:
2797 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2798 		break;
2799 	case MSR_IA32_MISC_ENABLE:
2800 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2801 		break;
2802 	case MSR_IA32_SMBASE:
2803 		if (!msr_info->host_initiated)
2804 			return 1;
2805 		msr_info->data = vcpu->arch.smbase;
2806 		break;
2807 	case MSR_SMI_COUNT:
2808 		msr_info->data = vcpu->arch.smi_count;
2809 		break;
2810 	case MSR_IA32_PERF_STATUS:
2811 		/* TSC increment by tick */
2812 		msr_info->data = 1000ULL;
2813 		/* CPU multiplier */
2814 		msr_info->data |= (((uint64_t)4ULL) << 40);
2815 		break;
2816 	case MSR_EFER:
2817 		msr_info->data = vcpu->arch.efer;
2818 		break;
2819 	case MSR_KVM_WALL_CLOCK:
2820 	case MSR_KVM_WALL_CLOCK_NEW:
2821 		msr_info->data = vcpu->kvm->arch.wall_clock;
2822 		break;
2823 	case MSR_KVM_SYSTEM_TIME:
2824 	case MSR_KVM_SYSTEM_TIME_NEW:
2825 		msr_info->data = vcpu->arch.time;
2826 		break;
2827 	case MSR_KVM_ASYNC_PF_EN:
2828 		msr_info->data = vcpu->arch.apf.msr_val;
2829 		break;
2830 	case MSR_KVM_STEAL_TIME:
2831 		msr_info->data = vcpu->arch.st.msr_val;
2832 		break;
2833 	case MSR_KVM_PV_EOI_EN:
2834 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2835 		break;
2836 	case MSR_IA32_P5_MC_ADDR:
2837 	case MSR_IA32_P5_MC_TYPE:
2838 	case MSR_IA32_MCG_CAP:
2839 	case MSR_IA32_MCG_CTL:
2840 	case MSR_IA32_MCG_STATUS:
2841 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2842 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2843 				   msr_info->host_initiated);
2844 	case MSR_K7_CLK_CTL:
2845 		/*
2846 		 * Provide expected ramp-up count for K7. All other
2847 		 * are set to zero, indicating minimum divisors for
2848 		 * every field.
2849 		 *
2850 		 * This prevents guest kernels on AMD host with CPU
2851 		 * type 6, model 8 and higher from exploding due to
2852 		 * the rdmsr failing.
2853 		 */
2854 		msr_info->data = 0x20000000;
2855 		break;
2856 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2857 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2858 	case HV_X64_MSR_CRASH_CTL:
2859 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2860 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2861 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2862 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2863 		return kvm_hv_get_msr_common(vcpu,
2864 					     msr_info->index, &msr_info->data,
2865 					     msr_info->host_initiated);
2866 		break;
2867 	case MSR_IA32_BBL_CR_CTL3:
2868 		/* This legacy MSR exists but isn't fully documented in current
2869 		 * silicon.  It is however accessed by winxp in very narrow
2870 		 * scenarios where it sets bit #19, itself documented as
2871 		 * a "reserved" bit.  Best effort attempt to source coherent
2872 		 * read data here should the balance of the register be
2873 		 * interpreted by the guest:
2874 		 *
2875 		 * L2 cache control register 3: 64GB range, 256KB size,
2876 		 * enabled, latency 0x1, configured
2877 		 */
2878 		msr_info->data = 0xbe702111;
2879 		break;
2880 	case MSR_AMD64_OSVW_ID_LENGTH:
2881 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2882 			return 1;
2883 		msr_info->data = vcpu->arch.osvw.length;
2884 		break;
2885 	case MSR_AMD64_OSVW_STATUS:
2886 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2887 			return 1;
2888 		msr_info->data = vcpu->arch.osvw.status;
2889 		break;
2890 	case MSR_PLATFORM_INFO:
2891 		if (!msr_info->host_initiated &&
2892 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2893 			return 1;
2894 		msr_info->data = vcpu->arch.msr_platform_info;
2895 		break;
2896 	case MSR_MISC_FEATURES_ENABLES:
2897 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2898 		break;
2899 	default:
2900 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2901 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2902 		if (!ignore_msrs) {
2903 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2904 					       msr_info->index);
2905 			return 1;
2906 		} else {
2907 			if (report_ignored_msrs)
2908 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2909 					msr_info->index);
2910 			msr_info->data = 0;
2911 		}
2912 		break;
2913 	}
2914 	return 0;
2915 }
2916 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2917 
2918 /*
2919  * Read or write a bunch of msrs. All parameters are kernel addresses.
2920  *
2921  * @return number of msrs set successfully.
2922  */
2923 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2924 		    struct kvm_msr_entry *entries,
2925 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2926 				  unsigned index, u64 *data))
2927 {
2928 	int i;
2929 
2930 	for (i = 0; i < msrs->nmsrs; ++i)
2931 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2932 			break;
2933 
2934 	return i;
2935 }
2936 
2937 /*
2938  * Read or write a bunch of msrs. Parameters are user addresses.
2939  *
2940  * @return number of msrs set successfully.
2941  */
2942 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2943 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2944 				unsigned index, u64 *data),
2945 		  int writeback)
2946 {
2947 	struct kvm_msrs msrs;
2948 	struct kvm_msr_entry *entries;
2949 	int r, n;
2950 	unsigned size;
2951 
2952 	r = -EFAULT;
2953 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2954 		goto out;
2955 
2956 	r = -E2BIG;
2957 	if (msrs.nmsrs >= MAX_IO_MSRS)
2958 		goto out;
2959 
2960 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2961 	entries = memdup_user(user_msrs->entries, size);
2962 	if (IS_ERR(entries)) {
2963 		r = PTR_ERR(entries);
2964 		goto out;
2965 	}
2966 
2967 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2968 	if (r < 0)
2969 		goto out_free;
2970 
2971 	r = -EFAULT;
2972 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2973 		goto out_free;
2974 
2975 	r = n;
2976 
2977 out_free:
2978 	kfree(entries);
2979 out:
2980 	return r;
2981 }
2982 
2983 static inline bool kvm_can_mwait_in_guest(void)
2984 {
2985 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2986 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2987 		boot_cpu_has(X86_FEATURE_ARAT);
2988 }
2989 
2990 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2991 {
2992 	int r = 0;
2993 
2994 	switch (ext) {
2995 	case KVM_CAP_IRQCHIP:
2996 	case KVM_CAP_HLT:
2997 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2998 	case KVM_CAP_SET_TSS_ADDR:
2999 	case KVM_CAP_EXT_CPUID:
3000 	case KVM_CAP_EXT_EMUL_CPUID:
3001 	case KVM_CAP_CLOCKSOURCE:
3002 	case KVM_CAP_PIT:
3003 	case KVM_CAP_NOP_IO_DELAY:
3004 	case KVM_CAP_MP_STATE:
3005 	case KVM_CAP_SYNC_MMU:
3006 	case KVM_CAP_USER_NMI:
3007 	case KVM_CAP_REINJECT_CONTROL:
3008 	case KVM_CAP_IRQ_INJECT_STATUS:
3009 	case KVM_CAP_IOEVENTFD:
3010 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3011 	case KVM_CAP_PIT2:
3012 	case KVM_CAP_PIT_STATE2:
3013 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3014 	case KVM_CAP_XEN_HVM:
3015 	case KVM_CAP_VCPU_EVENTS:
3016 	case KVM_CAP_HYPERV:
3017 	case KVM_CAP_HYPERV_VAPIC:
3018 	case KVM_CAP_HYPERV_SPIN:
3019 	case KVM_CAP_HYPERV_SYNIC:
3020 	case KVM_CAP_HYPERV_SYNIC2:
3021 	case KVM_CAP_HYPERV_VP_INDEX:
3022 	case KVM_CAP_HYPERV_EVENTFD:
3023 	case KVM_CAP_HYPERV_TLBFLUSH:
3024 	case KVM_CAP_HYPERV_SEND_IPI:
3025 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3026 	case KVM_CAP_HYPERV_CPUID:
3027 	case KVM_CAP_PCI_SEGMENT:
3028 	case KVM_CAP_DEBUGREGS:
3029 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3030 	case KVM_CAP_XSAVE:
3031 	case KVM_CAP_ASYNC_PF:
3032 	case KVM_CAP_GET_TSC_KHZ:
3033 	case KVM_CAP_KVMCLOCK_CTRL:
3034 	case KVM_CAP_READONLY_MEM:
3035 	case KVM_CAP_HYPERV_TIME:
3036 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3037 	case KVM_CAP_TSC_DEADLINE_TIMER:
3038 	case KVM_CAP_DISABLE_QUIRKS:
3039 	case KVM_CAP_SET_BOOT_CPU_ID:
3040  	case KVM_CAP_SPLIT_IRQCHIP:
3041 	case KVM_CAP_IMMEDIATE_EXIT:
3042 	case KVM_CAP_GET_MSR_FEATURES:
3043 	case KVM_CAP_MSR_PLATFORM_INFO:
3044 	case KVM_CAP_EXCEPTION_PAYLOAD:
3045 		r = 1;
3046 		break;
3047 	case KVM_CAP_SYNC_REGS:
3048 		r = KVM_SYNC_X86_VALID_FIELDS;
3049 		break;
3050 	case KVM_CAP_ADJUST_CLOCK:
3051 		r = KVM_CLOCK_TSC_STABLE;
3052 		break;
3053 	case KVM_CAP_X86_DISABLE_EXITS:
3054 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3055 		if(kvm_can_mwait_in_guest())
3056 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3057 		break;
3058 	case KVM_CAP_X86_SMM:
3059 		/* SMBASE is usually relocated above 1M on modern chipsets,
3060 		 * and SMM handlers might indeed rely on 4G segment limits,
3061 		 * so do not report SMM to be available if real mode is
3062 		 * emulated via vm86 mode.  Still, do not go to great lengths
3063 		 * to avoid userspace's usage of the feature, because it is a
3064 		 * fringe case that is not enabled except via specific settings
3065 		 * of the module parameters.
3066 		 */
3067 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3068 		break;
3069 	case KVM_CAP_VAPIC:
3070 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3071 		break;
3072 	case KVM_CAP_NR_VCPUS:
3073 		r = KVM_SOFT_MAX_VCPUS;
3074 		break;
3075 	case KVM_CAP_MAX_VCPUS:
3076 		r = KVM_MAX_VCPUS;
3077 		break;
3078 	case KVM_CAP_NR_MEMSLOTS:
3079 		r = KVM_USER_MEM_SLOTS;
3080 		break;
3081 	case KVM_CAP_PV_MMU:	/* obsolete */
3082 		r = 0;
3083 		break;
3084 	case KVM_CAP_MCE:
3085 		r = KVM_MAX_MCE_BANKS;
3086 		break;
3087 	case KVM_CAP_XCRS:
3088 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3089 		break;
3090 	case KVM_CAP_TSC_CONTROL:
3091 		r = kvm_has_tsc_control;
3092 		break;
3093 	case KVM_CAP_X2APIC_API:
3094 		r = KVM_X2APIC_API_VALID_FLAGS;
3095 		break;
3096 	case KVM_CAP_NESTED_STATE:
3097 		r = kvm_x86_ops->get_nested_state ?
3098 			kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3099 		break;
3100 	default:
3101 		break;
3102 	}
3103 	return r;
3104 
3105 }
3106 
3107 long kvm_arch_dev_ioctl(struct file *filp,
3108 			unsigned int ioctl, unsigned long arg)
3109 {
3110 	void __user *argp = (void __user *)arg;
3111 	long r;
3112 
3113 	switch (ioctl) {
3114 	case KVM_GET_MSR_INDEX_LIST: {
3115 		struct kvm_msr_list __user *user_msr_list = argp;
3116 		struct kvm_msr_list msr_list;
3117 		unsigned n;
3118 
3119 		r = -EFAULT;
3120 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3121 			goto out;
3122 		n = msr_list.nmsrs;
3123 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3124 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3125 			goto out;
3126 		r = -E2BIG;
3127 		if (n < msr_list.nmsrs)
3128 			goto out;
3129 		r = -EFAULT;
3130 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3131 				 num_msrs_to_save * sizeof(u32)))
3132 			goto out;
3133 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3134 				 &emulated_msrs,
3135 				 num_emulated_msrs * sizeof(u32)))
3136 			goto out;
3137 		r = 0;
3138 		break;
3139 	}
3140 	case KVM_GET_SUPPORTED_CPUID:
3141 	case KVM_GET_EMULATED_CPUID: {
3142 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3143 		struct kvm_cpuid2 cpuid;
3144 
3145 		r = -EFAULT;
3146 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3147 			goto out;
3148 
3149 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3150 					    ioctl);
3151 		if (r)
3152 			goto out;
3153 
3154 		r = -EFAULT;
3155 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3156 			goto out;
3157 		r = 0;
3158 		break;
3159 	}
3160 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3161 		r = -EFAULT;
3162 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3163 				 sizeof(kvm_mce_cap_supported)))
3164 			goto out;
3165 		r = 0;
3166 		break;
3167 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3168 		struct kvm_msr_list __user *user_msr_list = argp;
3169 		struct kvm_msr_list msr_list;
3170 		unsigned int n;
3171 
3172 		r = -EFAULT;
3173 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3174 			goto out;
3175 		n = msr_list.nmsrs;
3176 		msr_list.nmsrs = num_msr_based_features;
3177 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3178 			goto out;
3179 		r = -E2BIG;
3180 		if (n < msr_list.nmsrs)
3181 			goto out;
3182 		r = -EFAULT;
3183 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3184 				 num_msr_based_features * sizeof(u32)))
3185 			goto out;
3186 		r = 0;
3187 		break;
3188 	}
3189 	case KVM_GET_MSRS:
3190 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3191 		break;
3192 	}
3193 	default:
3194 		r = -EINVAL;
3195 	}
3196 out:
3197 	return r;
3198 }
3199 
3200 static void wbinvd_ipi(void *garbage)
3201 {
3202 	wbinvd();
3203 }
3204 
3205 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3206 {
3207 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3208 }
3209 
3210 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3211 {
3212 	/* Address WBINVD may be executed by guest */
3213 	if (need_emulate_wbinvd(vcpu)) {
3214 		if (kvm_x86_ops->has_wbinvd_exit())
3215 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3216 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3217 			smp_call_function_single(vcpu->cpu,
3218 					wbinvd_ipi, NULL, 1);
3219 	}
3220 
3221 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3222 
3223 	/* Apply any externally detected TSC adjustments (due to suspend) */
3224 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3225 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3226 		vcpu->arch.tsc_offset_adjustment = 0;
3227 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3228 	}
3229 
3230 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3231 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3232 				rdtsc() - vcpu->arch.last_host_tsc;
3233 		if (tsc_delta < 0)
3234 			mark_tsc_unstable("KVM discovered backwards TSC");
3235 
3236 		if (kvm_check_tsc_unstable()) {
3237 			u64 offset = kvm_compute_tsc_offset(vcpu,
3238 						vcpu->arch.last_guest_tsc);
3239 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3240 			vcpu->arch.tsc_catchup = 1;
3241 		}
3242 
3243 		if (kvm_lapic_hv_timer_in_use(vcpu))
3244 			kvm_lapic_restart_hv_timer(vcpu);
3245 
3246 		/*
3247 		 * On a host with synchronized TSC, there is no need to update
3248 		 * kvmclock on vcpu->cpu migration
3249 		 */
3250 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3251 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3252 		if (vcpu->cpu != cpu)
3253 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3254 		vcpu->cpu = cpu;
3255 	}
3256 
3257 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3258 }
3259 
3260 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3261 {
3262 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3263 		return;
3264 
3265 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3266 
3267 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3268 			&vcpu->arch.st.steal.preempted,
3269 			offsetof(struct kvm_steal_time, preempted),
3270 			sizeof(vcpu->arch.st.steal.preempted));
3271 }
3272 
3273 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3274 {
3275 	int idx;
3276 
3277 	if (vcpu->preempted)
3278 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3279 
3280 	/*
3281 	 * Disable page faults because we're in atomic context here.
3282 	 * kvm_write_guest_offset_cached() would call might_fault()
3283 	 * that relies on pagefault_disable() to tell if there's a
3284 	 * bug. NOTE: the write to guest memory may not go through if
3285 	 * during postcopy live migration or if there's heavy guest
3286 	 * paging.
3287 	 */
3288 	pagefault_disable();
3289 	/*
3290 	 * kvm_memslots() will be called by
3291 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3292 	 */
3293 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3294 	kvm_steal_time_set_preempted(vcpu);
3295 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3296 	pagefault_enable();
3297 	kvm_x86_ops->vcpu_put(vcpu);
3298 	vcpu->arch.last_host_tsc = rdtsc();
3299 	/*
3300 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3301 	 * on every vmexit, but if not, we might have a stale dr6 from the
3302 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3303 	 */
3304 	set_debugreg(0, 6);
3305 }
3306 
3307 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3308 				    struct kvm_lapic_state *s)
3309 {
3310 	if (vcpu->arch.apicv_active)
3311 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3312 
3313 	return kvm_apic_get_state(vcpu, s);
3314 }
3315 
3316 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3317 				    struct kvm_lapic_state *s)
3318 {
3319 	int r;
3320 
3321 	r = kvm_apic_set_state(vcpu, s);
3322 	if (r)
3323 		return r;
3324 	update_cr8_intercept(vcpu);
3325 
3326 	return 0;
3327 }
3328 
3329 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3330 {
3331 	return (!lapic_in_kernel(vcpu) ||
3332 		kvm_apic_accept_pic_intr(vcpu));
3333 }
3334 
3335 /*
3336  * if userspace requested an interrupt window, check that the
3337  * interrupt window is open.
3338  *
3339  * No need to exit to userspace if we already have an interrupt queued.
3340  */
3341 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3342 {
3343 	return kvm_arch_interrupt_allowed(vcpu) &&
3344 		!kvm_cpu_has_interrupt(vcpu) &&
3345 		!kvm_event_needs_reinjection(vcpu) &&
3346 		kvm_cpu_accept_dm_intr(vcpu);
3347 }
3348 
3349 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3350 				    struct kvm_interrupt *irq)
3351 {
3352 	if (irq->irq >= KVM_NR_INTERRUPTS)
3353 		return -EINVAL;
3354 
3355 	if (!irqchip_in_kernel(vcpu->kvm)) {
3356 		kvm_queue_interrupt(vcpu, irq->irq, false);
3357 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3358 		return 0;
3359 	}
3360 
3361 	/*
3362 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3363 	 * fail for in-kernel 8259.
3364 	 */
3365 	if (pic_in_kernel(vcpu->kvm))
3366 		return -ENXIO;
3367 
3368 	if (vcpu->arch.pending_external_vector != -1)
3369 		return -EEXIST;
3370 
3371 	vcpu->arch.pending_external_vector = irq->irq;
3372 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3373 	return 0;
3374 }
3375 
3376 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3377 {
3378 	kvm_inject_nmi(vcpu);
3379 
3380 	return 0;
3381 }
3382 
3383 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3384 {
3385 	kvm_make_request(KVM_REQ_SMI, vcpu);
3386 
3387 	return 0;
3388 }
3389 
3390 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3391 					   struct kvm_tpr_access_ctl *tac)
3392 {
3393 	if (tac->flags)
3394 		return -EINVAL;
3395 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3396 	return 0;
3397 }
3398 
3399 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3400 					u64 mcg_cap)
3401 {
3402 	int r;
3403 	unsigned bank_num = mcg_cap & 0xff, bank;
3404 
3405 	r = -EINVAL;
3406 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3407 		goto out;
3408 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3409 		goto out;
3410 	r = 0;
3411 	vcpu->arch.mcg_cap = mcg_cap;
3412 	/* Init IA32_MCG_CTL to all 1s */
3413 	if (mcg_cap & MCG_CTL_P)
3414 		vcpu->arch.mcg_ctl = ~(u64)0;
3415 	/* Init IA32_MCi_CTL to all 1s */
3416 	for (bank = 0; bank < bank_num; bank++)
3417 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3418 
3419 	if (kvm_x86_ops->setup_mce)
3420 		kvm_x86_ops->setup_mce(vcpu);
3421 out:
3422 	return r;
3423 }
3424 
3425 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3426 				      struct kvm_x86_mce *mce)
3427 {
3428 	u64 mcg_cap = vcpu->arch.mcg_cap;
3429 	unsigned bank_num = mcg_cap & 0xff;
3430 	u64 *banks = vcpu->arch.mce_banks;
3431 
3432 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3433 		return -EINVAL;
3434 	/*
3435 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3436 	 * reporting is disabled
3437 	 */
3438 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3439 	    vcpu->arch.mcg_ctl != ~(u64)0)
3440 		return 0;
3441 	banks += 4 * mce->bank;
3442 	/*
3443 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3444 	 * reporting is disabled for the bank
3445 	 */
3446 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3447 		return 0;
3448 	if (mce->status & MCI_STATUS_UC) {
3449 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3450 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3451 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3452 			return 0;
3453 		}
3454 		if (banks[1] & MCI_STATUS_VAL)
3455 			mce->status |= MCI_STATUS_OVER;
3456 		banks[2] = mce->addr;
3457 		banks[3] = mce->misc;
3458 		vcpu->arch.mcg_status = mce->mcg_status;
3459 		banks[1] = mce->status;
3460 		kvm_queue_exception(vcpu, MC_VECTOR);
3461 	} else if (!(banks[1] & MCI_STATUS_VAL)
3462 		   || !(banks[1] & MCI_STATUS_UC)) {
3463 		if (banks[1] & MCI_STATUS_VAL)
3464 			mce->status |= MCI_STATUS_OVER;
3465 		banks[2] = mce->addr;
3466 		banks[3] = mce->misc;
3467 		banks[1] = mce->status;
3468 	} else
3469 		banks[1] |= MCI_STATUS_OVER;
3470 	return 0;
3471 }
3472 
3473 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3474 					       struct kvm_vcpu_events *events)
3475 {
3476 	process_nmi(vcpu);
3477 
3478 	/*
3479 	 * The API doesn't provide the instruction length for software
3480 	 * exceptions, so don't report them. As long as the guest RIP
3481 	 * isn't advanced, we should expect to encounter the exception
3482 	 * again.
3483 	 */
3484 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3485 		events->exception.injected = 0;
3486 		events->exception.pending = 0;
3487 	} else {
3488 		events->exception.injected = vcpu->arch.exception.injected;
3489 		events->exception.pending = vcpu->arch.exception.pending;
3490 		/*
3491 		 * For ABI compatibility, deliberately conflate
3492 		 * pending and injected exceptions when
3493 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3494 		 */
3495 		if (!vcpu->kvm->arch.exception_payload_enabled)
3496 			events->exception.injected |=
3497 				vcpu->arch.exception.pending;
3498 	}
3499 	events->exception.nr = vcpu->arch.exception.nr;
3500 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3501 	events->exception.error_code = vcpu->arch.exception.error_code;
3502 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3503 	events->exception_payload = vcpu->arch.exception.payload;
3504 
3505 	events->interrupt.injected =
3506 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3507 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3508 	events->interrupt.soft = 0;
3509 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3510 
3511 	events->nmi.injected = vcpu->arch.nmi_injected;
3512 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3513 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3514 	events->nmi.pad = 0;
3515 
3516 	events->sipi_vector = 0; /* never valid when reporting to user space */
3517 
3518 	events->smi.smm = is_smm(vcpu);
3519 	events->smi.pending = vcpu->arch.smi_pending;
3520 	events->smi.smm_inside_nmi =
3521 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3522 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3523 
3524 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3525 			 | KVM_VCPUEVENT_VALID_SHADOW
3526 			 | KVM_VCPUEVENT_VALID_SMM);
3527 	if (vcpu->kvm->arch.exception_payload_enabled)
3528 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3529 
3530 	memset(&events->reserved, 0, sizeof(events->reserved));
3531 }
3532 
3533 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3534 
3535 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3536 					      struct kvm_vcpu_events *events)
3537 {
3538 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3539 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3540 			      | KVM_VCPUEVENT_VALID_SHADOW
3541 			      | KVM_VCPUEVENT_VALID_SMM
3542 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3543 		return -EINVAL;
3544 
3545 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3546 		if (!vcpu->kvm->arch.exception_payload_enabled)
3547 			return -EINVAL;
3548 		if (events->exception.pending)
3549 			events->exception.injected = 0;
3550 		else
3551 			events->exception_has_payload = 0;
3552 	} else {
3553 		events->exception.pending = 0;
3554 		events->exception_has_payload = 0;
3555 	}
3556 
3557 	if ((events->exception.injected || events->exception.pending) &&
3558 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3559 		return -EINVAL;
3560 
3561 	/* INITs are latched while in SMM */
3562 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3563 	    (events->smi.smm || events->smi.pending) &&
3564 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3565 		return -EINVAL;
3566 
3567 	process_nmi(vcpu);
3568 	vcpu->arch.exception.injected = events->exception.injected;
3569 	vcpu->arch.exception.pending = events->exception.pending;
3570 	vcpu->arch.exception.nr = events->exception.nr;
3571 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3572 	vcpu->arch.exception.error_code = events->exception.error_code;
3573 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3574 	vcpu->arch.exception.payload = events->exception_payload;
3575 
3576 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3577 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3578 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3579 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3580 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3581 						  events->interrupt.shadow);
3582 
3583 	vcpu->arch.nmi_injected = events->nmi.injected;
3584 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3585 		vcpu->arch.nmi_pending = events->nmi.pending;
3586 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3587 
3588 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3589 	    lapic_in_kernel(vcpu))
3590 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3591 
3592 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3593 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3594 			if (events->smi.smm)
3595 				vcpu->arch.hflags |= HF_SMM_MASK;
3596 			else
3597 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3598 			kvm_smm_changed(vcpu);
3599 		}
3600 
3601 		vcpu->arch.smi_pending = events->smi.pending;
3602 
3603 		if (events->smi.smm) {
3604 			if (events->smi.smm_inside_nmi)
3605 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3606 			else
3607 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3608 			if (lapic_in_kernel(vcpu)) {
3609 				if (events->smi.latched_init)
3610 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3611 				else
3612 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3613 			}
3614 		}
3615 	}
3616 
3617 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3618 
3619 	return 0;
3620 }
3621 
3622 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3623 					     struct kvm_debugregs *dbgregs)
3624 {
3625 	unsigned long val;
3626 
3627 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3628 	kvm_get_dr(vcpu, 6, &val);
3629 	dbgregs->dr6 = val;
3630 	dbgregs->dr7 = vcpu->arch.dr7;
3631 	dbgregs->flags = 0;
3632 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3633 }
3634 
3635 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3636 					    struct kvm_debugregs *dbgregs)
3637 {
3638 	if (dbgregs->flags)
3639 		return -EINVAL;
3640 
3641 	if (dbgregs->dr6 & ~0xffffffffull)
3642 		return -EINVAL;
3643 	if (dbgregs->dr7 & ~0xffffffffull)
3644 		return -EINVAL;
3645 
3646 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3647 	kvm_update_dr0123(vcpu);
3648 	vcpu->arch.dr6 = dbgregs->dr6;
3649 	kvm_update_dr6(vcpu);
3650 	vcpu->arch.dr7 = dbgregs->dr7;
3651 	kvm_update_dr7(vcpu);
3652 
3653 	return 0;
3654 }
3655 
3656 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3657 
3658 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3659 {
3660 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3661 	u64 xstate_bv = xsave->header.xfeatures;
3662 	u64 valid;
3663 
3664 	/*
3665 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3666 	 * leaves 0 and 1 in the loop below.
3667 	 */
3668 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3669 
3670 	/* Set XSTATE_BV */
3671 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3672 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3673 
3674 	/*
3675 	 * Copy each region from the possibly compacted offset to the
3676 	 * non-compacted offset.
3677 	 */
3678 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3679 	while (valid) {
3680 		u64 feature = valid & -valid;
3681 		int index = fls64(feature) - 1;
3682 		void *src = get_xsave_addr(xsave, feature);
3683 
3684 		if (src) {
3685 			u32 size, offset, ecx, edx;
3686 			cpuid_count(XSTATE_CPUID, index,
3687 				    &size, &offset, &ecx, &edx);
3688 			if (feature == XFEATURE_MASK_PKRU)
3689 				memcpy(dest + offset, &vcpu->arch.pkru,
3690 				       sizeof(vcpu->arch.pkru));
3691 			else
3692 				memcpy(dest + offset, src, size);
3693 
3694 		}
3695 
3696 		valid -= feature;
3697 	}
3698 }
3699 
3700 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3701 {
3702 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3703 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3704 	u64 valid;
3705 
3706 	/*
3707 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3708 	 * leaves 0 and 1 in the loop below.
3709 	 */
3710 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3711 
3712 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3713 	xsave->header.xfeatures = xstate_bv;
3714 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3715 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3716 
3717 	/*
3718 	 * Copy each region from the non-compacted offset to the
3719 	 * possibly compacted offset.
3720 	 */
3721 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3722 	while (valid) {
3723 		u64 feature = valid & -valid;
3724 		int index = fls64(feature) - 1;
3725 		void *dest = get_xsave_addr(xsave, feature);
3726 
3727 		if (dest) {
3728 			u32 size, offset, ecx, edx;
3729 			cpuid_count(XSTATE_CPUID, index,
3730 				    &size, &offset, &ecx, &edx);
3731 			if (feature == XFEATURE_MASK_PKRU)
3732 				memcpy(&vcpu->arch.pkru, src + offset,
3733 				       sizeof(vcpu->arch.pkru));
3734 			else
3735 				memcpy(dest, src + offset, size);
3736 		}
3737 
3738 		valid -= feature;
3739 	}
3740 }
3741 
3742 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3743 					 struct kvm_xsave *guest_xsave)
3744 {
3745 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3746 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3747 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3748 	} else {
3749 		memcpy(guest_xsave->region,
3750 			&vcpu->arch.guest_fpu->state.fxsave,
3751 			sizeof(struct fxregs_state));
3752 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3753 			XFEATURE_MASK_FPSSE;
3754 	}
3755 }
3756 
3757 #define XSAVE_MXCSR_OFFSET 24
3758 
3759 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3760 					struct kvm_xsave *guest_xsave)
3761 {
3762 	u64 xstate_bv =
3763 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3764 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3765 
3766 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3767 		/*
3768 		 * Here we allow setting states that are not present in
3769 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3770 		 * with old userspace.
3771 		 */
3772 		if (xstate_bv & ~kvm_supported_xcr0() ||
3773 			mxcsr & ~mxcsr_feature_mask)
3774 			return -EINVAL;
3775 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3776 	} else {
3777 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3778 			mxcsr & ~mxcsr_feature_mask)
3779 			return -EINVAL;
3780 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3781 			guest_xsave->region, sizeof(struct fxregs_state));
3782 	}
3783 	return 0;
3784 }
3785 
3786 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3787 					struct kvm_xcrs *guest_xcrs)
3788 {
3789 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3790 		guest_xcrs->nr_xcrs = 0;
3791 		return;
3792 	}
3793 
3794 	guest_xcrs->nr_xcrs = 1;
3795 	guest_xcrs->flags = 0;
3796 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3797 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3798 }
3799 
3800 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3801 				       struct kvm_xcrs *guest_xcrs)
3802 {
3803 	int i, r = 0;
3804 
3805 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3806 		return -EINVAL;
3807 
3808 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3809 		return -EINVAL;
3810 
3811 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3812 		/* Only support XCR0 currently */
3813 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3814 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3815 				guest_xcrs->xcrs[i].value);
3816 			break;
3817 		}
3818 	if (r)
3819 		r = -EINVAL;
3820 	return r;
3821 }
3822 
3823 /*
3824  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3825  * stopped by the hypervisor.  This function will be called from the host only.
3826  * EINVAL is returned when the host attempts to set the flag for a guest that
3827  * does not support pv clocks.
3828  */
3829 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3830 {
3831 	if (!vcpu->arch.pv_time_enabled)
3832 		return -EINVAL;
3833 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3834 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3835 	return 0;
3836 }
3837 
3838 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3839 				     struct kvm_enable_cap *cap)
3840 {
3841 	int r;
3842 	uint16_t vmcs_version;
3843 	void __user *user_ptr;
3844 
3845 	if (cap->flags)
3846 		return -EINVAL;
3847 
3848 	switch (cap->cap) {
3849 	case KVM_CAP_HYPERV_SYNIC2:
3850 		if (cap->args[0])
3851 			return -EINVAL;
3852 		/* fall through */
3853 
3854 	case KVM_CAP_HYPERV_SYNIC:
3855 		if (!irqchip_in_kernel(vcpu->kvm))
3856 			return -EINVAL;
3857 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3858 					     KVM_CAP_HYPERV_SYNIC2);
3859 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3860 		if (!kvm_x86_ops->nested_enable_evmcs)
3861 			return -ENOTTY;
3862 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3863 		if (!r) {
3864 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
3865 			if (copy_to_user(user_ptr, &vmcs_version,
3866 					 sizeof(vmcs_version)))
3867 				r = -EFAULT;
3868 		}
3869 		return r;
3870 
3871 	default:
3872 		return -EINVAL;
3873 	}
3874 }
3875 
3876 long kvm_arch_vcpu_ioctl(struct file *filp,
3877 			 unsigned int ioctl, unsigned long arg)
3878 {
3879 	struct kvm_vcpu *vcpu = filp->private_data;
3880 	void __user *argp = (void __user *)arg;
3881 	int r;
3882 	union {
3883 		struct kvm_lapic_state *lapic;
3884 		struct kvm_xsave *xsave;
3885 		struct kvm_xcrs *xcrs;
3886 		void *buffer;
3887 	} u;
3888 
3889 	vcpu_load(vcpu);
3890 
3891 	u.buffer = NULL;
3892 	switch (ioctl) {
3893 	case KVM_GET_LAPIC: {
3894 		r = -EINVAL;
3895 		if (!lapic_in_kernel(vcpu))
3896 			goto out;
3897 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
3898 				GFP_KERNEL_ACCOUNT);
3899 
3900 		r = -ENOMEM;
3901 		if (!u.lapic)
3902 			goto out;
3903 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3904 		if (r)
3905 			goto out;
3906 		r = -EFAULT;
3907 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3908 			goto out;
3909 		r = 0;
3910 		break;
3911 	}
3912 	case KVM_SET_LAPIC: {
3913 		r = -EINVAL;
3914 		if (!lapic_in_kernel(vcpu))
3915 			goto out;
3916 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3917 		if (IS_ERR(u.lapic)) {
3918 			r = PTR_ERR(u.lapic);
3919 			goto out_nofree;
3920 		}
3921 
3922 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3923 		break;
3924 	}
3925 	case KVM_INTERRUPT: {
3926 		struct kvm_interrupt irq;
3927 
3928 		r = -EFAULT;
3929 		if (copy_from_user(&irq, argp, sizeof(irq)))
3930 			goto out;
3931 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3932 		break;
3933 	}
3934 	case KVM_NMI: {
3935 		r = kvm_vcpu_ioctl_nmi(vcpu);
3936 		break;
3937 	}
3938 	case KVM_SMI: {
3939 		r = kvm_vcpu_ioctl_smi(vcpu);
3940 		break;
3941 	}
3942 	case KVM_SET_CPUID: {
3943 		struct kvm_cpuid __user *cpuid_arg = argp;
3944 		struct kvm_cpuid cpuid;
3945 
3946 		r = -EFAULT;
3947 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3948 			goto out;
3949 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3950 		break;
3951 	}
3952 	case KVM_SET_CPUID2: {
3953 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3954 		struct kvm_cpuid2 cpuid;
3955 
3956 		r = -EFAULT;
3957 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3958 			goto out;
3959 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3960 					      cpuid_arg->entries);
3961 		break;
3962 	}
3963 	case KVM_GET_CPUID2: {
3964 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3965 		struct kvm_cpuid2 cpuid;
3966 
3967 		r = -EFAULT;
3968 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3969 			goto out;
3970 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3971 					      cpuid_arg->entries);
3972 		if (r)
3973 			goto out;
3974 		r = -EFAULT;
3975 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3976 			goto out;
3977 		r = 0;
3978 		break;
3979 	}
3980 	case KVM_GET_MSRS: {
3981 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3982 		r = msr_io(vcpu, argp, do_get_msr, 1);
3983 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3984 		break;
3985 	}
3986 	case KVM_SET_MSRS: {
3987 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3988 		r = msr_io(vcpu, argp, do_set_msr, 0);
3989 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3990 		break;
3991 	}
3992 	case KVM_TPR_ACCESS_REPORTING: {
3993 		struct kvm_tpr_access_ctl tac;
3994 
3995 		r = -EFAULT;
3996 		if (copy_from_user(&tac, argp, sizeof(tac)))
3997 			goto out;
3998 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3999 		if (r)
4000 			goto out;
4001 		r = -EFAULT;
4002 		if (copy_to_user(argp, &tac, sizeof(tac)))
4003 			goto out;
4004 		r = 0;
4005 		break;
4006 	};
4007 	case KVM_SET_VAPIC_ADDR: {
4008 		struct kvm_vapic_addr va;
4009 		int idx;
4010 
4011 		r = -EINVAL;
4012 		if (!lapic_in_kernel(vcpu))
4013 			goto out;
4014 		r = -EFAULT;
4015 		if (copy_from_user(&va, argp, sizeof(va)))
4016 			goto out;
4017 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4018 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4019 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4020 		break;
4021 	}
4022 	case KVM_X86_SETUP_MCE: {
4023 		u64 mcg_cap;
4024 
4025 		r = -EFAULT;
4026 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4027 			goto out;
4028 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4029 		break;
4030 	}
4031 	case KVM_X86_SET_MCE: {
4032 		struct kvm_x86_mce mce;
4033 
4034 		r = -EFAULT;
4035 		if (copy_from_user(&mce, argp, sizeof(mce)))
4036 			goto out;
4037 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4038 		break;
4039 	}
4040 	case KVM_GET_VCPU_EVENTS: {
4041 		struct kvm_vcpu_events events;
4042 
4043 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4044 
4045 		r = -EFAULT;
4046 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4047 			break;
4048 		r = 0;
4049 		break;
4050 	}
4051 	case KVM_SET_VCPU_EVENTS: {
4052 		struct kvm_vcpu_events events;
4053 
4054 		r = -EFAULT;
4055 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4056 			break;
4057 
4058 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4059 		break;
4060 	}
4061 	case KVM_GET_DEBUGREGS: {
4062 		struct kvm_debugregs dbgregs;
4063 
4064 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4065 
4066 		r = -EFAULT;
4067 		if (copy_to_user(argp, &dbgregs,
4068 				 sizeof(struct kvm_debugregs)))
4069 			break;
4070 		r = 0;
4071 		break;
4072 	}
4073 	case KVM_SET_DEBUGREGS: {
4074 		struct kvm_debugregs dbgregs;
4075 
4076 		r = -EFAULT;
4077 		if (copy_from_user(&dbgregs, argp,
4078 				   sizeof(struct kvm_debugregs)))
4079 			break;
4080 
4081 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4082 		break;
4083 	}
4084 	case KVM_GET_XSAVE: {
4085 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4086 		r = -ENOMEM;
4087 		if (!u.xsave)
4088 			break;
4089 
4090 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4091 
4092 		r = -EFAULT;
4093 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4094 			break;
4095 		r = 0;
4096 		break;
4097 	}
4098 	case KVM_SET_XSAVE: {
4099 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4100 		if (IS_ERR(u.xsave)) {
4101 			r = PTR_ERR(u.xsave);
4102 			goto out_nofree;
4103 		}
4104 
4105 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4106 		break;
4107 	}
4108 	case KVM_GET_XCRS: {
4109 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4110 		r = -ENOMEM;
4111 		if (!u.xcrs)
4112 			break;
4113 
4114 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4115 
4116 		r = -EFAULT;
4117 		if (copy_to_user(argp, u.xcrs,
4118 				 sizeof(struct kvm_xcrs)))
4119 			break;
4120 		r = 0;
4121 		break;
4122 	}
4123 	case KVM_SET_XCRS: {
4124 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4125 		if (IS_ERR(u.xcrs)) {
4126 			r = PTR_ERR(u.xcrs);
4127 			goto out_nofree;
4128 		}
4129 
4130 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4131 		break;
4132 	}
4133 	case KVM_SET_TSC_KHZ: {
4134 		u32 user_tsc_khz;
4135 
4136 		r = -EINVAL;
4137 		user_tsc_khz = (u32)arg;
4138 
4139 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4140 			goto out;
4141 
4142 		if (user_tsc_khz == 0)
4143 			user_tsc_khz = tsc_khz;
4144 
4145 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4146 			r = 0;
4147 
4148 		goto out;
4149 	}
4150 	case KVM_GET_TSC_KHZ: {
4151 		r = vcpu->arch.virtual_tsc_khz;
4152 		goto out;
4153 	}
4154 	case KVM_KVMCLOCK_CTRL: {
4155 		r = kvm_set_guest_paused(vcpu);
4156 		goto out;
4157 	}
4158 	case KVM_ENABLE_CAP: {
4159 		struct kvm_enable_cap cap;
4160 
4161 		r = -EFAULT;
4162 		if (copy_from_user(&cap, argp, sizeof(cap)))
4163 			goto out;
4164 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4165 		break;
4166 	}
4167 	case KVM_GET_NESTED_STATE: {
4168 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4169 		u32 user_data_size;
4170 
4171 		r = -EINVAL;
4172 		if (!kvm_x86_ops->get_nested_state)
4173 			break;
4174 
4175 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4176 		r = -EFAULT;
4177 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4178 			break;
4179 
4180 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4181 						  user_data_size);
4182 		if (r < 0)
4183 			break;
4184 
4185 		if (r > user_data_size) {
4186 			if (put_user(r, &user_kvm_nested_state->size))
4187 				r = -EFAULT;
4188 			else
4189 				r = -E2BIG;
4190 			break;
4191 		}
4192 
4193 		r = 0;
4194 		break;
4195 	}
4196 	case KVM_SET_NESTED_STATE: {
4197 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4198 		struct kvm_nested_state kvm_state;
4199 
4200 		r = -EINVAL;
4201 		if (!kvm_x86_ops->set_nested_state)
4202 			break;
4203 
4204 		r = -EFAULT;
4205 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4206 			break;
4207 
4208 		r = -EINVAL;
4209 		if (kvm_state.size < sizeof(kvm_state))
4210 			break;
4211 
4212 		if (kvm_state.flags &
4213 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4214 		      | KVM_STATE_NESTED_EVMCS))
4215 			break;
4216 
4217 		/* nested_run_pending implies guest_mode.  */
4218 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4219 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4220 			break;
4221 
4222 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4223 		break;
4224 	}
4225 	case KVM_GET_SUPPORTED_HV_CPUID: {
4226 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4227 		struct kvm_cpuid2 cpuid;
4228 
4229 		r = -EFAULT;
4230 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4231 			goto out;
4232 
4233 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4234 						cpuid_arg->entries);
4235 		if (r)
4236 			goto out;
4237 
4238 		r = -EFAULT;
4239 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4240 			goto out;
4241 		r = 0;
4242 		break;
4243 	}
4244 	default:
4245 		r = -EINVAL;
4246 	}
4247 out:
4248 	kfree(u.buffer);
4249 out_nofree:
4250 	vcpu_put(vcpu);
4251 	return r;
4252 }
4253 
4254 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4255 {
4256 	return VM_FAULT_SIGBUS;
4257 }
4258 
4259 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4260 {
4261 	int ret;
4262 
4263 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4264 		return -EINVAL;
4265 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4266 	return ret;
4267 }
4268 
4269 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4270 					      u64 ident_addr)
4271 {
4272 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4273 }
4274 
4275 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4276 					 unsigned long kvm_nr_mmu_pages)
4277 {
4278 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4279 		return -EINVAL;
4280 
4281 	mutex_lock(&kvm->slots_lock);
4282 
4283 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4284 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4285 
4286 	mutex_unlock(&kvm->slots_lock);
4287 	return 0;
4288 }
4289 
4290 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4291 {
4292 	return kvm->arch.n_max_mmu_pages;
4293 }
4294 
4295 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4296 {
4297 	struct kvm_pic *pic = kvm->arch.vpic;
4298 	int r;
4299 
4300 	r = 0;
4301 	switch (chip->chip_id) {
4302 	case KVM_IRQCHIP_PIC_MASTER:
4303 		memcpy(&chip->chip.pic, &pic->pics[0],
4304 			sizeof(struct kvm_pic_state));
4305 		break;
4306 	case KVM_IRQCHIP_PIC_SLAVE:
4307 		memcpy(&chip->chip.pic, &pic->pics[1],
4308 			sizeof(struct kvm_pic_state));
4309 		break;
4310 	case KVM_IRQCHIP_IOAPIC:
4311 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4312 		break;
4313 	default:
4314 		r = -EINVAL;
4315 		break;
4316 	}
4317 	return r;
4318 }
4319 
4320 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4321 {
4322 	struct kvm_pic *pic = kvm->arch.vpic;
4323 	int r;
4324 
4325 	r = 0;
4326 	switch (chip->chip_id) {
4327 	case KVM_IRQCHIP_PIC_MASTER:
4328 		spin_lock(&pic->lock);
4329 		memcpy(&pic->pics[0], &chip->chip.pic,
4330 			sizeof(struct kvm_pic_state));
4331 		spin_unlock(&pic->lock);
4332 		break;
4333 	case KVM_IRQCHIP_PIC_SLAVE:
4334 		spin_lock(&pic->lock);
4335 		memcpy(&pic->pics[1], &chip->chip.pic,
4336 			sizeof(struct kvm_pic_state));
4337 		spin_unlock(&pic->lock);
4338 		break;
4339 	case KVM_IRQCHIP_IOAPIC:
4340 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4341 		break;
4342 	default:
4343 		r = -EINVAL;
4344 		break;
4345 	}
4346 	kvm_pic_update_irq(pic);
4347 	return r;
4348 }
4349 
4350 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4351 {
4352 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4353 
4354 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4355 
4356 	mutex_lock(&kps->lock);
4357 	memcpy(ps, &kps->channels, sizeof(*ps));
4358 	mutex_unlock(&kps->lock);
4359 	return 0;
4360 }
4361 
4362 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4363 {
4364 	int i;
4365 	struct kvm_pit *pit = kvm->arch.vpit;
4366 
4367 	mutex_lock(&pit->pit_state.lock);
4368 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4369 	for (i = 0; i < 3; i++)
4370 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4371 	mutex_unlock(&pit->pit_state.lock);
4372 	return 0;
4373 }
4374 
4375 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4376 {
4377 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4378 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4379 		sizeof(ps->channels));
4380 	ps->flags = kvm->arch.vpit->pit_state.flags;
4381 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4382 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4383 	return 0;
4384 }
4385 
4386 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4387 {
4388 	int start = 0;
4389 	int i;
4390 	u32 prev_legacy, cur_legacy;
4391 	struct kvm_pit *pit = kvm->arch.vpit;
4392 
4393 	mutex_lock(&pit->pit_state.lock);
4394 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4395 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4396 	if (!prev_legacy && cur_legacy)
4397 		start = 1;
4398 	memcpy(&pit->pit_state.channels, &ps->channels,
4399 	       sizeof(pit->pit_state.channels));
4400 	pit->pit_state.flags = ps->flags;
4401 	for (i = 0; i < 3; i++)
4402 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4403 				   start && i == 0);
4404 	mutex_unlock(&pit->pit_state.lock);
4405 	return 0;
4406 }
4407 
4408 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4409 				 struct kvm_reinject_control *control)
4410 {
4411 	struct kvm_pit *pit = kvm->arch.vpit;
4412 
4413 	if (!pit)
4414 		return -ENXIO;
4415 
4416 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4417 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4418 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4419 	 */
4420 	mutex_lock(&pit->pit_state.lock);
4421 	kvm_pit_set_reinject(pit, control->pit_reinject);
4422 	mutex_unlock(&pit->pit_state.lock);
4423 
4424 	return 0;
4425 }
4426 
4427 /**
4428  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4429  * @kvm: kvm instance
4430  * @log: slot id and address to which we copy the log
4431  *
4432  * Steps 1-4 below provide general overview of dirty page logging. See
4433  * kvm_get_dirty_log_protect() function description for additional details.
4434  *
4435  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4436  * always flush the TLB (step 4) even if previous step failed  and the dirty
4437  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4438  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4439  * writes will be marked dirty for next log read.
4440  *
4441  *   1. Take a snapshot of the bit and clear it if needed.
4442  *   2. Write protect the corresponding page.
4443  *   3. Copy the snapshot to the userspace.
4444  *   4. Flush TLB's if needed.
4445  */
4446 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4447 {
4448 	bool flush = false;
4449 	int r;
4450 
4451 	mutex_lock(&kvm->slots_lock);
4452 
4453 	/*
4454 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4455 	 */
4456 	if (kvm_x86_ops->flush_log_dirty)
4457 		kvm_x86_ops->flush_log_dirty(kvm);
4458 
4459 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4460 
4461 	/*
4462 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4463 	 * kvm_mmu_slot_remove_write_access().
4464 	 */
4465 	lockdep_assert_held(&kvm->slots_lock);
4466 	if (flush)
4467 		kvm_flush_remote_tlbs(kvm);
4468 
4469 	mutex_unlock(&kvm->slots_lock);
4470 	return r;
4471 }
4472 
4473 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4474 {
4475 	bool flush = false;
4476 	int r;
4477 
4478 	mutex_lock(&kvm->slots_lock);
4479 
4480 	/*
4481 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4482 	 */
4483 	if (kvm_x86_ops->flush_log_dirty)
4484 		kvm_x86_ops->flush_log_dirty(kvm);
4485 
4486 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4487 
4488 	/*
4489 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4490 	 * kvm_mmu_slot_remove_write_access().
4491 	 */
4492 	lockdep_assert_held(&kvm->slots_lock);
4493 	if (flush)
4494 		kvm_flush_remote_tlbs(kvm);
4495 
4496 	mutex_unlock(&kvm->slots_lock);
4497 	return r;
4498 }
4499 
4500 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4501 			bool line_status)
4502 {
4503 	if (!irqchip_in_kernel(kvm))
4504 		return -ENXIO;
4505 
4506 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4507 					irq_event->irq, irq_event->level,
4508 					line_status);
4509 	return 0;
4510 }
4511 
4512 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4513 			    struct kvm_enable_cap *cap)
4514 {
4515 	int r;
4516 
4517 	if (cap->flags)
4518 		return -EINVAL;
4519 
4520 	switch (cap->cap) {
4521 	case KVM_CAP_DISABLE_QUIRKS:
4522 		kvm->arch.disabled_quirks = cap->args[0];
4523 		r = 0;
4524 		break;
4525 	case KVM_CAP_SPLIT_IRQCHIP: {
4526 		mutex_lock(&kvm->lock);
4527 		r = -EINVAL;
4528 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4529 			goto split_irqchip_unlock;
4530 		r = -EEXIST;
4531 		if (irqchip_in_kernel(kvm))
4532 			goto split_irqchip_unlock;
4533 		if (kvm->created_vcpus)
4534 			goto split_irqchip_unlock;
4535 		r = kvm_setup_empty_irq_routing(kvm);
4536 		if (r)
4537 			goto split_irqchip_unlock;
4538 		/* Pairs with irqchip_in_kernel. */
4539 		smp_wmb();
4540 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4541 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4542 		r = 0;
4543 split_irqchip_unlock:
4544 		mutex_unlock(&kvm->lock);
4545 		break;
4546 	}
4547 	case KVM_CAP_X2APIC_API:
4548 		r = -EINVAL;
4549 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4550 			break;
4551 
4552 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4553 			kvm->arch.x2apic_format = true;
4554 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4555 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4556 
4557 		r = 0;
4558 		break;
4559 	case KVM_CAP_X86_DISABLE_EXITS:
4560 		r = -EINVAL;
4561 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4562 			break;
4563 
4564 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4565 			kvm_can_mwait_in_guest())
4566 			kvm->arch.mwait_in_guest = true;
4567 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4568 			kvm->arch.hlt_in_guest = true;
4569 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4570 			kvm->arch.pause_in_guest = true;
4571 		r = 0;
4572 		break;
4573 	case KVM_CAP_MSR_PLATFORM_INFO:
4574 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4575 		r = 0;
4576 		break;
4577 	case KVM_CAP_EXCEPTION_PAYLOAD:
4578 		kvm->arch.exception_payload_enabled = cap->args[0];
4579 		r = 0;
4580 		break;
4581 	default:
4582 		r = -EINVAL;
4583 		break;
4584 	}
4585 	return r;
4586 }
4587 
4588 long kvm_arch_vm_ioctl(struct file *filp,
4589 		       unsigned int ioctl, unsigned long arg)
4590 {
4591 	struct kvm *kvm = filp->private_data;
4592 	void __user *argp = (void __user *)arg;
4593 	int r = -ENOTTY;
4594 	/*
4595 	 * This union makes it completely explicit to gcc-3.x
4596 	 * that these two variables' stack usage should be
4597 	 * combined, not added together.
4598 	 */
4599 	union {
4600 		struct kvm_pit_state ps;
4601 		struct kvm_pit_state2 ps2;
4602 		struct kvm_pit_config pit_config;
4603 	} u;
4604 
4605 	switch (ioctl) {
4606 	case KVM_SET_TSS_ADDR:
4607 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4608 		break;
4609 	case KVM_SET_IDENTITY_MAP_ADDR: {
4610 		u64 ident_addr;
4611 
4612 		mutex_lock(&kvm->lock);
4613 		r = -EINVAL;
4614 		if (kvm->created_vcpus)
4615 			goto set_identity_unlock;
4616 		r = -EFAULT;
4617 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4618 			goto set_identity_unlock;
4619 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4620 set_identity_unlock:
4621 		mutex_unlock(&kvm->lock);
4622 		break;
4623 	}
4624 	case KVM_SET_NR_MMU_PAGES:
4625 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4626 		break;
4627 	case KVM_GET_NR_MMU_PAGES:
4628 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4629 		break;
4630 	case KVM_CREATE_IRQCHIP: {
4631 		mutex_lock(&kvm->lock);
4632 
4633 		r = -EEXIST;
4634 		if (irqchip_in_kernel(kvm))
4635 			goto create_irqchip_unlock;
4636 
4637 		r = -EINVAL;
4638 		if (kvm->created_vcpus)
4639 			goto create_irqchip_unlock;
4640 
4641 		r = kvm_pic_init(kvm);
4642 		if (r)
4643 			goto create_irqchip_unlock;
4644 
4645 		r = kvm_ioapic_init(kvm);
4646 		if (r) {
4647 			kvm_pic_destroy(kvm);
4648 			goto create_irqchip_unlock;
4649 		}
4650 
4651 		r = kvm_setup_default_irq_routing(kvm);
4652 		if (r) {
4653 			kvm_ioapic_destroy(kvm);
4654 			kvm_pic_destroy(kvm);
4655 			goto create_irqchip_unlock;
4656 		}
4657 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4658 		smp_wmb();
4659 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4660 	create_irqchip_unlock:
4661 		mutex_unlock(&kvm->lock);
4662 		break;
4663 	}
4664 	case KVM_CREATE_PIT:
4665 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4666 		goto create_pit;
4667 	case KVM_CREATE_PIT2:
4668 		r = -EFAULT;
4669 		if (copy_from_user(&u.pit_config, argp,
4670 				   sizeof(struct kvm_pit_config)))
4671 			goto out;
4672 	create_pit:
4673 		mutex_lock(&kvm->lock);
4674 		r = -EEXIST;
4675 		if (kvm->arch.vpit)
4676 			goto create_pit_unlock;
4677 		r = -ENOMEM;
4678 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4679 		if (kvm->arch.vpit)
4680 			r = 0;
4681 	create_pit_unlock:
4682 		mutex_unlock(&kvm->lock);
4683 		break;
4684 	case KVM_GET_IRQCHIP: {
4685 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4686 		struct kvm_irqchip *chip;
4687 
4688 		chip = memdup_user(argp, sizeof(*chip));
4689 		if (IS_ERR(chip)) {
4690 			r = PTR_ERR(chip);
4691 			goto out;
4692 		}
4693 
4694 		r = -ENXIO;
4695 		if (!irqchip_kernel(kvm))
4696 			goto get_irqchip_out;
4697 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4698 		if (r)
4699 			goto get_irqchip_out;
4700 		r = -EFAULT;
4701 		if (copy_to_user(argp, chip, sizeof(*chip)))
4702 			goto get_irqchip_out;
4703 		r = 0;
4704 	get_irqchip_out:
4705 		kfree(chip);
4706 		break;
4707 	}
4708 	case KVM_SET_IRQCHIP: {
4709 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4710 		struct kvm_irqchip *chip;
4711 
4712 		chip = memdup_user(argp, sizeof(*chip));
4713 		if (IS_ERR(chip)) {
4714 			r = PTR_ERR(chip);
4715 			goto out;
4716 		}
4717 
4718 		r = -ENXIO;
4719 		if (!irqchip_kernel(kvm))
4720 			goto set_irqchip_out;
4721 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4722 		if (r)
4723 			goto set_irqchip_out;
4724 		r = 0;
4725 	set_irqchip_out:
4726 		kfree(chip);
4727 		break;
4728 	}
4729 	case KVM_GET_PIT: {
4730 		r = -EFAULT;
4731 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4732 			goto out;
4733 		r = -ENXIO;
4734 		if (!kvm->arch.vpit)
4735 			goto out;
4736 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4737 		if (r)
4738 			goto out;
4739 		r = -EFAULT;
4740 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4741 			goto out;
4742 		r = 0;
4743 		break;
4744 	}
4745 	case KVM_SET_PIT: {
4746 		r = -EFAULT;
4747 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4748 			goto out;
4749 		r = -ENXIO;
4750 		if (!kvm->arch.vpit)
4751 			goto out;
4752 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4753 		break;
4754 	}
4755 	case KVM_GET_PIT2: {
4756 		r = -ENXIO;
4757 		if (!kvm->arch.vpit)
4758 			goto out;
4759 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4760 		if (r)
4761 			goto out;
4762 		r = -EFAULT;
4763 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4764 			goto out;
4765 		r = 0;
4766 		break;
4767 	}
4768 	case KVM_SET_PIT2: {
4769 		r = -EFAULT;
4770 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4771 			goto out;
4772 		r = -ENXIO;
4773 		if (!kvm->arch.vpit)
4774 			goto out;
4775 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4776 		break;
4777 	}
4778 	case KVM_REINJECT_CONTROL: {
4779 		struct kvm_reinject_control control;
4780 		r =  -EFAULT;
4781 		if (copy_from_user(&control, argp, sizeof(control)))
4782 			goto out;
4783 		r = kvm_vm_ioctl_reinject(kvm, &control);
4784 		break;
4785 	}
4786 	case KVM_SET_BOOT_CPU_ID:
4787 		r = 0;
4788 		mutex_lock(&kvm->lock);
4789 		if (kvm->created_vcpus)
4790 			r = -EBUSY;
4791 		else
4792 			kvm->arch.bsp_vcpu_id = arg;
4793 		mutex_unlock(&kvm->lock);
4794 		break;
4795 	case KVM_XEN_HVM_CONFIG: {
4796 		struct kvm_xen_hvm_config xhc;
4797 		r = -EFAULT;
4798 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4799 			goto out;
4800 		r = -EINVAL;
4801 		if (xhc.flags)
4802 			goto out;
4803 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4804 		r = 0;
4805 		break;
4806 	}
4807 	case KVM_SET_CLOCK: {
4808 		struct kvm_clock_data user_ns;
4809 		u64 now_ns;
4810 
4811 		r = -EFAULT;
4812 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4813 			goto out;
4814 
4815 		r = -EINVAL;
4816 		if (user_ns.flags)
4817 			goto out;
4818 
4819 		r = 0;
4820 		/*
4821 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4822 		 * kvm_gen_update_masterclock() can be cut down to locked
4823 		 * pvclock_update_vm_gtod_copy().
4824 		 */
4825 		kvm_gen_update_masterclock(kvm);
4826 		now_ns = get_kvmclock_ns(kvm);
4827 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4828 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4829 		break;
4830 	}
4831 	case KVM_GET_CLOCK: {
4832 		struct kvm_clock_data user_ns;
4833 		u64 now_ns;
4834 
4835 		now_ns = get_kvmclock_ns(kvm);
4836 		user_ns.clock = now_ns;
4837 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4838 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4839 
4840 		r = -EFAULT;
4841 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4842 			goto out;
4843 		r = 0;
4844 		break;
4845 	}
4846 	case KVM_MEMORY_ENCRYPT_OP: {
4847 		r = -ENOTTY;
4848 		if (kvm_x86_ops->mem_enc_op)
4849 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4850 		break;
4851 	}
4852 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4853 		struct kvm_enc_region region;
4854 
4855 		r = -EFAULT;
4856 		if (copy_from_user(&region, argp, sizeof(region)))
4857 			goto out;
4858 
4859 		r = -ENOTTY;
4860 		if (kvm_x86_ops->mem_enc_reg_region)
4861 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4862 		break;
4863 	}
4864 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4865 		struct kvm_enc_region region;
4866 
4867 		r = -EFAULT;
4868 		if (copy_from_user(&region, argp, sizeof(region)))
4869 			goto out;
4870 
4871 		r = -ENOTTY;
4872 		if (kvm_x86_ops->mem_enc_unreg_region)
4873 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4874 		break;
4875 	}
4876 	case KVM_HYPERV_EVENTFD: {
4877 		struct kvm_hyperv_eventfd hvevfd;
4878 
4879 		r = -EFAULT;
4880 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4881 			goto out;
4882 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4883 		break;
4884 	}
4885 	default:
4886 		r = -ENOTTY;
4887 	}
4888 out:
4889 	return r;
4890 }
4891 
4892 static void kvm_init_msr_list(void)
4893 {
4894 	u32 dummy[2];
4895 	unsigned i, j;
4896 
4897 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4898 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4899 			continue;
4900 
4901 		/*
4902 		 * Even MSRs that are valid in the host may not be exposed
4903 		 * to the guests in some cases.
4904 		 */
4905 		switch (msrs_to_save[i]) {
4906 		case MSR_IA32_BNDCFGS:
4907 			if (!kvm_mpx_supported())
4908 				continue;
4909 			break;
4910 		case MSR_TSC_AUX:
4911 			if (!kvm_x86_ops->rdtscp_supported())
4912 				continue;
4913 			break;
4914 		case MSR_IA32_RTIT_CTL:
4915 		case MSR_IA32_RTIT_STATUS:
4916 			if (!kvm_x86_ops->pt_supported())
4917 				continue;
4918 			break;
4919 		case MSR_IA32_RTIT_CR3_MATCH:
4920 			if (!kvm_x86_ops->pt_supported() ||
4921 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
4922 				continue;
4923 			break;
4924 		case MSR_IA32_RTIT_OUTPUT_BASE:
4925 		case MSR_IA32_RTIT_OUTPUT_MASK:
4926 			if (!kvm_x86_ops->pt_supported() ||
4927 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
4928 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
4929 				continue;
4930 			break;
4931 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
4932 			if (!kvm_x86_ops->pt_supported() ||
4933 				msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
4934 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
4935 				continue;
4936 			break;
4937 		}
4938 		default:
4939 			break;
4940 		}
4941 
4942 		if (j < i)
4943 			msrs_to_save[j] = msrs_to_save[i];
4944 		j++;
4945 	}
4946 	num_msrs_to_save = j;
4947 
4948 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4949 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4950 			continue;
4951 
4952 		if (j < i)
4953 			emulated_msrs[j] = emulated_msrs[i];
4954 		j++;
4955 	}
4956 	num_emulated_msrs = j;
4957 
4958 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4959 		struct kvm_msr_entry msr;
4960 
4961 		msr.index = msr_based_features[i];
4962 		if (kvm_get_msr_feature(&msr))
4963 			continue;
4964 
4965 		if (j < i)
4966 			msr_based_features[j] = msr_based_features[i];
4967 		j++;
4968 	}
4969 	num_msr_based_features = j;
4970 }
4971 
4972 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4973 			   const void *v)
4974 {
4975 	int handled = 0;
4976 	int n;
4977 
4978 	do {
4979 		n = min(len, 8);
4980 		if (!(lapic_in_kernel(vcpu) &&
4981 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4982 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4983 			break;
4984 		handled += n;
4985 		addr += n;
4986 		len -= n;
4987 		v += n;
4988 	} while (len);
4989 
4990 	return handled;
4991 }
4992 
4993 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4994 {
4995 	int handled = 0;
4996 	int n;
4997 
4998 	do {
4999 		n = min(len, 8);
5000 		if (!(lapic_in_kernel(vcpu) &&
5001 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5002 					 addr, n, v))
5003 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5004 			break;
5005 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5006 		handled += n;
5007 		addr += n;
5008 		len -= n;
5009 		v += n;
5010 	} while (len);
5011 
5012 	return handled;
5013 }
5014 
5015 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5016 			struct kvm_segment *var, int seg)
5017 {
5018 	kvm_x86_ops->set_segment(vcpu, var, seg);
5019 }
5020 
5021 void kvm_get_segment(struct kvm_vcpu *vcpu,
5022 		     struct kvm_segment *var, int seg)
5023 {
5024 	kvm_x86_ops->get_segment(vcpu, var, seg);
5025 }
5026 
5027 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5028 			   struct x86_exception *exception)
5029 {
5030 	gpa_t t_gpa;
5031 
5032 	BUG_ON(!mmu_is_nested(vcpu));
5033 
5034 	/* NPT walks are always user-walks */
5035 	access |= PFERR_USER_MASK;
5036 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5037 
5038 	return t_gpa;
5039 }
5040 
5041 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5042 			      struct x86_exception *exception)
5043 {
5044 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5045 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5046 }
5047 
5048  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5049 				struct x86_exception *exception)
5050 {
5051 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5052 	access |= PFERR_FETCH_MASK;
5053 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5054 }
5055 
5056 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5057 			       struct x86_exception *exception)
5058 {
5059 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5060 	access |= PFERR_WRITE_MASK;
5061 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5062 }
5063 
5064 /* uses this to access any guest's mapped memory without checking CPL */
5065 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5066 				struct x86_exception *exception)
5067 {
5068 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5069 }
5070 
5071 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5072 				      struct kvm_vcpu *vcpu, u32 access,
5073 				      struct x86_exception *exception)
5074 {
5075 	void *data = val;
5076 	int r = X86EMUL_CONTINUE;
5077 
5078 	while (bytes) {
5079 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5080 							    exception);
5081 		unsigned offset = addr & (PAGE_SIZE-1);
5082 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5083 		int ret;
5084 
5085 		if (gpa == UNMAPPED_GVA)
5086 			return X86EMUL_PROPAGATE_FAULT;
5087 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5088 					       offset, toread);
5089 		if (ret < 0) {
5090 			r = X86EMUL_IO_NEEDED;
5091 			goto out;
5092 		}
5093 
5094 		bytes -= toread;
5095 		data += toread;
5096 		addr += toread;
5097 	}
5098 out:
5099 	return r;
5100 }
5101 
5102 /* used for instruction fetching */
5103 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5104 				gva_t addr, void *val, unsigned int bytes,
5105 				struct x86_exception *exception)
5106 {
5107 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5108 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5109 	unsigned offset;
5110 	int ret;
5111 
5112 	/* Inline kvm_read_guest_virt_helper for speed.  */
5113 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5114 						    exception);
5115 	if (unlikely(gpa == UNMAPPED_GVA))
5116 		return X86EMUL_PROPAGATE_FAULT;
5117 
5118 	offset = addr & (PAGE_SIZE-1);
5119 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5120 		bytes = (unsigned)PAGE_SIZE - offset;
5121 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5122 				       offset, bytes);
5123 	if (unlikely(ret < 0))
5124 		return X86EMUL_IO_NEEDED;
5125 
5126 	return X86EMUL_CONTINUE;
5127 }
5128 
5129 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5130 			       gva_t addr, void *val, unsigned int bytes,
5131 			       struct x86_exception *exception)
5132 {
5133 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5134 
5135 	/*
5136 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5137 	 * is returned, but our callers are not ready for that and they blindly
5138 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5139 	 * uninitialized kernel stack memory into cr2 and error code.
5140 	 */
5141 	memset(exception, 0, sizeof(*exception));
5142 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5143 					  exception);
5144 }
5145 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5146 
5147 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5148 			     gva_t addr, void *val, unsigned int bytes,
5149 			     struct x86_exception *exception, bool system)
5150 {
5151 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5152 	u32 access = 0;
5153 
5154 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5155 		access |= PFERR_USER_MASK;
5156 
5157 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5158 }
5159 
5160 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5161 		unsigned long addr, void *val, unsigned int bytes)
5162 {
5163 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5164 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5165 
5166 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5167 }
5168 
5169 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5170 				      struct kvm_vcpu *vcpu, u32 access,
5171 				      struct x86_exception *exception)
5172 {
5173 	void *data = val;
5174 	int r = X86EMUL_CONTINUE;
5175 
5176 	while (bytes) {
5177 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5178 							     access,
5179 							     exception);
5180 		unsigned offset = addr & (PAGE_SIZE-1);
5181 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5182 		int ret;
5183 
5184 		if (gpa == UNMAPPED_GVA)
5185 			return X86EMUL_PROPAGATE_FAULT;
5186 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5187 		if (ret < 0) {
5188 			r = X86EMUL_IO_NEEDED;
5189 			goto out;
5190 		}
5191 
5192 		bytes -= towrite;
5193 		data += towrite;
5194 		addr += towrite;
5195 	}
5196 out:
5197 	return r;
5198 }
5199 
5200 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5201 			      unsigned int bytes, struct x86_exception *exception,
5202 			      bool system)
5203 {
5204 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5205 	u32 access = PFERR_WRITE_MASK;
5206 
5207 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5208 		access |= PFERR_USER_MASK;
5209 
5210 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5211 					   access, exception);
5212 }
5213 
5214 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5215 				unsigned int bytes, struct x86_exception *exception)
5216 {
5217 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5218 	vcpu->arch.l1tf_flush_l1d = true;
5219 
5220 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5221 					   PFERR_WRITE_MASK, exception);
5222 }
5223 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5224 
5225 int handle_ud(struct kvm_vcpu *vcpu)
5226 {
5227 	int emul_type = EMULTYPE_TRAP_UD;
5228 	enum emulation_result er;
5229 	char sig[5]; /* ud2; .ascii "kvm" */
5230 	struct x86_exception e;
5231 
5232 	if (force_emulation_prefix &&
5233 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5234 				sig, sizeof(sig), &e) == 0 &&
5235 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5236 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5237 		emul_type = 0;
5238 	}
5239 
5240 	er = kvm_emulate_instruction(vcpu, emul_type);
5241 	if (er == EMULATE_USER_EXIT)
5242 		return 0;
5243 	if (er != EMULATE_DONE)
5244 		kvm_queue_exception(vcpu, UD_VECTOR);
5245 	return 1;
5246 }
5247 EXPORT_SYMBOL_GPL(handle_ud);
5248 
5249 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5250 			    gpa_t gpa, bool write)
5251 {
5252 	/* For APIC access vmexit */
5253 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5254 		return 1;
5255 
5256 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5257 		trace_vcpu_match_mmio(gva, gpa, write, true);
5258 		return 1;
5259 	}
5260 
5261 	return 0;
5262 }
5263 
5264 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5265 				gpa_t *gpa, struct x86_exception *exception,
5266 				bool write)
5267 {
5268 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5269 		| (write ? PFERR_WRITE_MASK : 0);
5270 
5271 	/*
5272 	 * currently PKRU is only applied to ept enabled guest so
5273 	 * there is no pkey in EPT page table for L1 guest or EPT
5274 	 * shadow page table for L2 guest.
5275 	 */
5276 	if (vcpu_match_mmio_gva(vcpu, gva)
5277 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5278 				 vcpu->arch.access, 0, access)) {
5279 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5280 					(gva & (PAGE_SIZE - 1));
5281 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5282 		return 1;
5283 	}
5284 
5285 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5286 
5287 	if (*gpa == UNMAPPED_GVA)
5288 		return -1;
5289 
5290 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5291 }
5292 
5293 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5294 			const void *val, int bytes)
5295 {
5296 	int ret;
5297 
5298 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5299 	if (ret < 0)
5300 		return 0;
5301 	kvm_page_track_write(vcpu, gpa, val, bytes);
5302 	return 1;
5303 }
5304 
5305 struct read_write_emulator_ops {
5306 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5307 				  int bytes);
5308 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5309 				  void *val, int bytes);
5310 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5311 			       int bytes, void *val);
5312 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5313 				    void *val, int bytes);
5314 	bool write;
5315 };
5316 
5317 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5318 {
5319 	if (vcpu->mmio_read_completed) {
5320 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5321 			       vcpu->mmio_fragments[0].gpa, val);
5322 		vcpu->mmio_read_completed = 0;
5323 		return 1;
5324 	}
5325 
5326 	return 0;
5327 }
5328 
5329 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5330 			void *val, int bytes)
5331 {
5332 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5333 }
5334 
5335 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5336 			 void *val, int bytes)
5337 {
5338 	return emulator_write_phys(vcpu, gpa, val, bytes);
5339 }
5340 
5341 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5342 {
5343 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5344 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5345 }
5346 
5347 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5348 			  void *val, int bytes)
5349 {
5350 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5351 	return X86EMUL_IO_NEEDED;
5352 }
5353 
5354 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5355 			   void *val, int bytes)
5356 {
5357 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5358 
5359 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5360 	return X86EMUL_CONTINUE;
5361 }
5362 
5363 static const struct read_write_emulator_ops read_emultor = {
5364 	.read_write_prepare = read_prepare,
5365 	.read_write_emulate = read_emulate,
5366 	.read_write_mmio = vcpu_mmio_read,
5367 	.read_write_exit_mmio = read_exit_mmio,
5368 };
5369 
5370 static const struct read_write_emulator_ops write_emultor = {
5371 	.read_write_emulate = write_emulate,
5372 	.read_write_mmio = write_mmio,
5373 	.read_write_exit_mmio = write_exit_mmio,
5374 	.write = true,
5375 };
5376 
5377 static int emulator_read_write_onepage(unsigned long addr, void *val,
5378 				       unsigned int bytes,
5379 				       struct x86_exception *exception,
5380 				       struct kvm_vcpu *vcpu,
5381 				       const struct read_write_emulator_ops *ops)
5382 {
5383 	gpa_t gpa;
5384 	int handled, ret;
5385 	bool write = ops->write;
5386 	struct kvm_mmio_fragment *frag;
5387 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5388 
5389 	/*
5390 	 * If the exit was due to a NPF we may already have a GPA.
5391 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5392 	 * Note, this cannot be used on string operations since string
5393 	 * operation using rep will only have the initial GPA from the NPF
5394 	 * occurred.
5395 	 */
5396 	if (vcpu->arch.gpa_available &&
5397 	    emulator_can_use_gpa(ctxt) &&
5398 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5399 		gpa = vcpu->arch.gpa_val;
5400 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5401 	} else {
5402 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5403 		if (ret < 0)
5404 			return X86EMUL_PROPAGATE_FAULT;
5405 	}
5406 
5407 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5408 		return X86EMUL_CONTINUE;
5409 
5410 	/*
5411 	 * Is this MMIO handled locally?
5412 	 */
5413 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5414 	if (handled == bytes)
5415 		return X86EMUL_CONTINUE;
5416 
5417 	gpa += handled;
5418 	bytes -= handled;
5419 	val += handled;
5420 
5421 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5422 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5423 	frag->gpa = gpa;
5424 	frag->data = val;
5425 	frag->len = bytes;
5426 	return X86EMUL_CONTINUE;
5427 }
5428 
5429 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5430 			unsigned long addr,
5431 			void *val, unsigned int bytes,
5432 			struct x86_exception *exception,
5433 			const struct read_write_emulator_ops *ops)
5434 {
5435 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5436 	gpa_t gpa;
5437 	int rc;
5438 
5439 	if (ops->read_write_prepare &&
5440 		  ops->read_write_prepare(vcpu, val, bytes))
5441 		return X86EMUL_CONTINUE;
5442 
5443 	vcpu->mmio_nr_fragments = 0;
5444 
5445 	/* Crossing a page boundary? */
5446 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5447 		int now;
5448 
5449 		now = -addr & ~PAGE_MASK;
5450 		rc = emulator_read_write_onepage(addr, val, now, exception,
5451 						 vcpu, ops);
5452 
5453 		if (rc != X86EMUL_CONTINUE)
5454 			return rc;
5455 		addr += now;
5456 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5457 			addr = (u32)addr;
5458 		val += now;
5459 		bytes -= now;
5460 	}
5461 
5462 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5463 					 vcpu, ops);
5464 	if (rc != X86EMUL_CONTINUE)
5465 		return rc;
5466 
5467 	if (!vcpu->mmio_nr_fragments)
5468 		return rc;
5469 
5470 	gpa = vcpu->mmio_fragments[0].gpa;
5471 
5472 	vcpu->mmio_needed = 1;
5473 	vcpu->mmio_cur_fragment = 0;
5474 
5475 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5476 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5477 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5478 	vcpu->run->mmio.phys_addr = gpa;
5479 
5480 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5481 }
5482 
5483 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5484 				  unsigned long addr,
5485 				  void *val,
5486 				  unsigned int bytes,
5487 				  struct x86_exception *exception)
5488 {
5489 	return emulator_read_write(ctxt, addr, val, bytes,
5490 				   exception, &read_emultor);
5491 }
5492 
5493 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5494 			    unsigned long addr,
5495 			    const void *val,
5496 			    unsigned int bytes,
5497 			    struct x86_exception *exception)
5498 {
5499 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5500 				   exception, &write_emultor);
5501 }
5502 
5503 #define CMPXCHG_TYPE(t, ptr, old, new) \
5504 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5505 
5506 #ifdef CONFIG_X86_64
5507 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5508 #else
5509 #  define CMPXCHG64(ptr, old, new) \
5510 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5511 #endif
5512 
5513 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5514 				     unsigned long addr,
5515 				     const void *old,
5516 				     const void *new,
5517 				     unsigned int bytes,
5518 				     struct x86_exception *exception)
5519 {
5520 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5521 	gpa_t gpa;
5522 	struct page *page;
5523 	char *kaddr;
5524 	bool exchanged;
5525 
5526 	/* guests cmpxchg8b have to be emulated atomically */
5527 	if (bytes > 8 || (bytes & (bytes - 1)))
5528 		goto emul_write;
5529 
5530 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5531 
5532 	if (gpa == UNMAPPED_GVA ||
5533 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5534 		goto emul_write;
5535 
5536 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5537 		goto emul_write;
5538 
5539 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5540 	if (is_error_page(page))
5541 		goto emul_write;
5542 
5543 	kaddr = kmap_atomic(page);
5544 	kaddr += offset_in_page(gpa);
5545 	switch (bytes) {
5546 	case 1:
5547 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5548 		break;
5549 	case 2:
5550 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5551 		break;
5552 	case 4:
5553 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5554 		break;
5555 	case 8:
5556 		exchanged = CMPXCHG64(kaddr, old, new);
5557 		break;
5558 	default:
5559 		BUG();
5560 	}
5561 	kunmap_atomic(kaddr);
5562 	kvm_release_page_dirty(page);
5563 
5564 	if (!exchanged)
5565 		return X86EMUL_CMPXCHG_FAILED;
5566 
5567 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5568 	kvm_page_track_write(vcpu, gpa, new, bytes);
5569 
5570 	return X86EMUL_CONTINUE;
5571 
5572 emul_write:
5573 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5574 
5575 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5576 }
5577 
5578 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5579 {
5580 	int r = 0, i;
5581 
5582 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5583 		if (vcpu->arch.pio.in)
5584 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5585 					    vcpu->arch.pio.size, pd);
5586 		else
5587 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5588 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5589 					     pd);
5590 		if (r)
5591 			break;
5592 		pd += vcpu->arch.pio.size;
5593 	}
5594 	return r;
5595 }
5596 
5597 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5598 			       unsigned short port, void *val,
5599 			       unsigned int count, bool in)
5600 {
5601 	vcpu->arch.pio.port = port;
5602 	vcpu->arch.pio.in = in;
5603 	vcpu->arch.pio.count  = count;
5604 	vcpu->arch.pio.size = size;
5605 
5606 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5607 		vcpu->arch.pio.count = 0;
5608 		return 1;
5609 	}
5610 
5611 	vcpu->run->exit_reason = KVM_EXIT_IO;
5612 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5613 	vcpu->run->io.size = size;
5614 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5615 	vcpu->run->io.count = count;
5616 	vcpu->run->io.port = port;
5617 
5618 	return 0;
5619 }
5620 
5621 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5622 				    int size, unsigned short port, void *val,
5623 				    unsigned int count)
5624 {
5625 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5626 	int ret;
5627 
5628 	if (vcpu->arch.pio.count)
5629 		goto data_avail;
5630 
5631 	memset(vcpu->arch.pio_data, 0, size * count);
5632 
5633 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5634 	if (ret) {
5635 data_avail:
5636 		memcpy(val, vcpu->arch.pio_data, size * count);
5637 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5638 		vcpu->arch.pio.count = 0;
5639 		return 1;
5640 	}
5641 
5642 	return 0;
5643 }
5644 
5645 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5646 				     int size, unsigned short port,
5647 				     const void *val, unsigned int count)
5648 {
5649 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5650 
5651 	memcpy(vcpu->arch.pio_data, val, size * count);
5652 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5653 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5654 }
5655 
5656 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5657 {
5658 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5659 }
5660 
5661 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5662 {
5663 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5664 }
5665 
5666 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5667 {
5668 	if (!need_emulate_wbinvd(vcpu))
5669 		return X86EMUL_CONTINUE;
5670 
5671 	if (kvm_x86_ops->has_wbinvd_exit()) {
5672 		int cpu = get_cpu();
5673 
5674 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5675 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5676 				wbinvd_ipi, NULL, 1);
5677 		put_cpu();
5678 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5679 	} else
5680 		wbinvd();
5681 	return X86EMUL_CONTINUE;
5682 }
5683 
5684 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5685 {
5686 	kvm_emulate_wbinvd_noskip(vcpu);
5687 	return kvm_skip_emulated_instruction(vcpu);
5688 }
5689 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5690 
5691 
5692 
5693 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5694 {
5695 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5696 }
5697 
5698 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5699 			   unsigned long *dest)
5700 {
5701 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5702 }
5703 
5704 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5705 			   unsigned long value)
5706 {
5707 
5708 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5709 }
5710 
5711 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5712 {
5713 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5714 }
5715 
5716 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5717 {
5718 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5719 	unsigned long value;
5720 
5721 	switch (cr) {
5722 	case 0:
5723 		value = kvm_read_cr0(vcpu);
5724 		break;
5725 	case 2:
5726 		value = vcpu->arch.cr2;
5727 		break;
5728 	case 3:
5729 		value = kvm_read_cr3(vcpu);
5730 		break;
5731 	case 4:
5732 		value = kvm_read_cr4(vcpu);
5733 		break;
5734 	case 8:
5735 		value = kvm_get_cr8(vcpu);
5736 		break;
5737 	default:
5738 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5739 		return 0;
5740 	}
5741 
5742 	return value;
5743 }
5744 
5745 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5746 {
5747 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5748 	int res = 0;
5749 
5750 	switch (cr) {
5751 	case 0:
5752 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5753 		break;
5754 	case 2:
5755 		vcpu->arch.cr2 = val;
5756 		break;
5757 	case 3:
5758 		res = kvm_set_cr3(vcpu, val);
5759 		break;
5760 	case 4:
5761 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5762 		break;
5763 	case 8:
5764 		res = kvm_set_cr8(vcpu, val);
5765 		break;
5766 	default:
5767 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5768 		res = -1;
5769 	}
5770 
5771 	return res;
5772 }
5773 
5774 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5775 {
5776 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5777 }
5778 
5779 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5780 {
5781 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5782 }
5783 
5784 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5785 {
5786 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5787 }
5788 
5789 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5790 {
5791 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5792 }
5793 
5794 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5795 {
5796 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5797 }
5798 
5799 static unsigned long emulator_get_cached_segment_base(
5800 	struct x86_emulate_ctxt *ctxt, int seg)
5801 {
5802 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5803 }
5804 
5805 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5806 				 struct desc_struct *desc, u32 *base3,
5807 				 int seg)
5808 {
5809 	struct kvm_segment var;
5810 
5811 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5812 	*selector = var.selector;
5813 
5814 	if (var.unusable) {
5815 		memset(desc, 0, sizeof(*desc));
5816 		if (base3)
5817 			*base3 = 0;
5818 		return false;
5819 	}
5820 
5821 	if (var.g)
5822 		var.limit >>= 12;
5823 	set_desc_limit(desc, var.limit);
5824 	set_desc_base(desc, (unsigned long)var.base);
5825 #ifdef CONFIG_X86_64
5826 	if (base3)
5827 		*base3 = var.base >> 32;
5828 #endif
5829 	desc->type = var.type;
5830 	desc->s = var.s;
5831 	desc->dpl = var.dpl;
5832 	desc->p = var.present;
5833 	desc->avl = var.avl;
5834 	desc->l = var.l;
5835 	desc->d = var.db;
5836 	desc->g = var.g;
5837 
5838 	return true;
5839 }
5840 
5841 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5842 				 struct desc_struct *desc, u32 base3,
5843 				 int seg)
5844 {
5845 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5846 	struct kvm_segment var;
5847 
5848 	var.selector = selector;
5849 	var.base = get_desc_base(desc);
5850 #ifdef CONFIG_X86_64
5851 	var.base |= ((u64)base3) << 32;
5852 #endif
5853 	var.limit = get_desc_limit(desc);
5854 	if (desc->g)
5855 		var.limit = (var.limit << 12) | 0xfff;
5856 	var.type = desc->type;
5857 	var.dpl = desc->dpl;
5858 	var.db = desc->d;
5859 	var.s = desc->s;
5860 	var.l = desc->l;
5861 	var.g = desc->g;
5862 	var.avl = desc->avl;
5863 	var.present = desc->p;
5864 	var.unusable = !var.present;
5865 	var.padding = 0;
5866 
5867 	kvm_set_segment(vcpu, &var, seg);
5868 	return;
5869 }
5870 
5871 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5872 			    u32 msr_index, u64 *pdata)
5873 {
5874 	struct msr_data msr;
5875 	int r;
5876 
5877 	msr.index = msr_index;
5878 	msr.host_initiated = false;
5879 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5880 	if (r)
5881 		return r;
5882 
5883 	*pdata = msr.data;
5884 	return 0;
5885 }
5886 
5887 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5888 			    u32 msr_index, u64 data)
5889 {
5890 	struct msr_data msr;
5891 
5892 	msr.data = data;
5893 	msr.index = msr_index;
5894 	msr.host_initiated = false;
5895 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5896 }
5897 
5898 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5899 {
5900 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5901 
5902 	return vcpu->arch.smbase;
5903 }
5904 
5905 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5906 {
5907 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5908 
5909 	vcpu->arch.smbase = smbase;
5910 }
5911 
5912 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5913 			      u32 pmc)
5914 {
5915 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5916 }
5917 
5918 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5919 			     u32 pmc, u64 *pdata)
5920 {
5921 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5922 }
5923 
5924 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5925 {
5926 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5927 }
5928 
5929 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5930 			      struct x86_instruction_info *info,
5931 			      enum x86_intercept_stage stage)
5932 {
5933 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5934 }
5935 
5936 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5937 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5938 {
5939 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5940 }
5941 
5942 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5943 {
5944 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5945 }
5946 
5947 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5948 {
5949 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5950 }
5951 
5952 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5953 {
5954 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5955 }
5956 
5957 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5958 {
5959 	return emul_to_vcpu(ctxt)->arch.hflags;
5960 }
5961 
5962 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5963 {
5964 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
5965 }
5966 
5967 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
5968 				  const char *smstate)
5969 {
5970 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
5971 }
5972 
5973 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
5974 {
5975 	kvm_smm_changed(emul_to_vcpu(ctxt));
5976 }
5977 
5978 static const struct x86_emulate_ops emulate_ops = {
5979 	.read_gpr            = emulator_read_gpr,
5980 	.write_gpr           = emulator_write_gpr,
5981 	.read_std            = emulator_read_std,
5982 	.write_std           = emulator_write_std,
5983 	.read_phys           = kvm_read_guest_phys_system,
5984 	.fetch               = kvm_fetch_guest_virt,
5985 	.read_emulated       = emulator_read_emulated,
5986 	.write_emulated      = emulator_write_emulated,
5987 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5988 	.invlpg              = emulator_invlpg,
5989 	.pio_in_emulated     = emulator_pio_in_emulated,
5990 	.pio_out_emulated    = emulator_pio_out_emulated,
5991 	.get_segment         = emulator_get_segment,
5992 	.set_segment         = emulator_set_segment,
5993 	.get_cached_segment_base = emulator_get_cached_segment_base,
5994 	.get_gdt             = emulator_get_gdt,
5995 	.get_idt	     = emulator_get_idt,
5996 	.set_gdt             = emulator_set_gdt,
5997 	.set_idt	     = emulator_set_idt,
5998 	.get_cr              = emulator_get_cr,
5999 	.set_cr              = emulator_set_cr,
6000 	.cpl                 = emulator_get_cpl,
6001 	.get_dr              = emulator_get_dr,
6002 	.set_dr              = emulator_set_dr,
6003 	.get_smbase          = emulator_get_smbase,
6004 	.set_smbase          = emulator_set_smbase,
6005 	.set_msr             = emulator_set_msr,
6006 	.get_msr             = emulator_get_msr,
6007 	.check_pmc	     = emulator_check_pmc,
6008 	.read_pmc            = emulator_read_pmc,
6009 	.halt                = emulator_halt,
6010 	.wbinvd              = emulator_wbinvd,
6011 	.fix_hypercall       = emulator_fix_hypercall,
6012 	.intercept           = emulator_intercept,
6013 	.get_cpuid           = emulator_get_cpuid,
6014 	.set_nmi_mask        = emulator_set_nmi_mask,
6015 	.get_hflags          = emulator_get_hflags,
6016 	.set_hflags          = emulator_set_hflags,
6017 	.pre_leave_smm       = emulator_pre_leave_smm,
6018 	.post_leave_smm      = emulator_post_leave_smm,
6019 };
6020 
6021 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6022 {
6023 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6024 	/*
6025 	 * an sti; sti; sequence only disable interrupts for the first
6026 	 * instruction. So, if the last instruction, be it emulated or
6027 	 * not, left the system with the INT_STI flag enabled, it
6028 	 * means that the last instruction is an sti. We should not
6029 	 * leave the flag on in this case. The same goes for mov ss
6030 	 */
6031 	if (int_shadow & mask)
6032 		mask = 0;
6033 	if (unlikely(int_shadow || mask)) {
6034 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6035 		if (!mask)
6036 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6037 	}
6038 }
6039 
6040 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6041 {
6042 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6043 	if (ctxt->exception.vector == PF_VECTOR)
6044 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6045 
6046 	if (ctxt->exception.error_code_valid)
6047 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6048 				      ctxt->exception.error_code);
6049 	else
6050 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6051 	return false;
6052 }
6053 
6054 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6055 {
6056 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6057 	int cs_db, cs_l;
6058 
6059 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6060 
6061 	ctxt->eflags = kvm_get_rflags(vcpu);
6062 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6063 
6064 	ctxt->eip = kvm_rip_read(vcpu);
6065 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6066 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6067 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6068 		     cs_db				? X86EMUL_MODE_PROT32 :
6069 							  X86EMUL_MODE_PROT16;
6070 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6071 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6072 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6073 
6074 	init_decode_cache(ctxt);
6075 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6076 }
6077 
6078 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6079 {
6080 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6081 	int ret;
6082 
6083 	init_emulate_ctxt(vcpu);
6084 
6085 	ctxt->op_bytes = 2;
6086 	ctxt->ad_bytes = 2;
6087 	ctxt->_eip = ctxt->eip + inc_eip;
6088 	ret = emulate_int_real(ctxt, irq);
6089 
6090 	if (ret != X86EMUL_CONTINUE)
6091 		return EMULATE_FAIL;
6092 
6093 	ctxt->eip = ctxt->_eip;
6094 	kvm_rip_write(vcpu, ctxt->eip);
6095 	kvm_set_rflags(vcpu, ctxt->eflags);
6096 
6097 	return EMULATE_DONE;
6098 }
6099 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6100 
6101 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6102 {
6103 	int r = EMULATE_DONE;
6104 
6105 	++vcpu->stat.insn_emulation_fail;
6106 	trace_kvm_emulate_insn_failed(vcpu);
6107 
6108 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6109 		return EMULATE_FAIL;
6110 
6111 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6112 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6113 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6114 		vcpu->run->internal.ndata = 0;
6115 		r = EMULATE_USER_EXIT;
6116 	}
6117 
6118 	kvm_queue_exception(vcpu, UD_VECTOR);
6119 
6120 	return r;
6121 }
6122 
6123 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6124 				  bool write_fault_to_shadow_pgtable,
6125 				  int emulation_type)
6126 {
6127 	gpa_t gpa = cr2;
6128 	kvm_pfn_t pfn;
6129 
6130 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6131 		return false;
6132 
6133 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6134 		return false;
6135 
6136 	if (!vcpu->arch.mmu->direct_map) {
6137 		/*
6138 		 * Write permission should be allowed since only
6139 		 * write access need to be emulated.
6140 		 */
6141 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6142 
6143 		/*
6144 		 * If the mapping is invalid in guest, let cpu retry
6145 		 * it to generate fault.
6146 		 */
6147 		if (gpa == UNMAPPED_GVA)
6148 			return true;
6149 	}
6150 
6151 	/*
6152 	 * Do not retry the unhandleable instruction if it faults on the
6153 	 * readonly host memory, otherwise it will goto a infinite loop:
6154 	 * retry instruction -> write #PF -> emulation fail -> retry
6155 	 * instruction -> ...
6156 	 */
6157 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6158 
6159 	/*
6160 	 * If the instruction failed on the error pfn, it can not be fixed,
6161 	 * report the error to userspace.
6162 	 */
6163 	if (is_error_noslot_pfn(pfn))
6164 		return false;
6165 
6166 	kvm_release_pfn_clean(pfn);
6167 
6168 	/* The instructions are well-emulated on direct mmu. */
6169 	if (vcpu->arch.mmu->direct_map) {
6170 		unsigned int indirect_shadow_pages;
6171 
6172 		spin_lock(&vcpu->kvm->mmu_lock);
6173 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6174 		spin_unlock(&vcpu->kvm->mmu_lock);
6175 
6176 		if (indirect_shadow_pages)
6177 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6178 
6179 		return true;
6180 	}
6181 
6182 	/*
6183 	 * if emulation was due to access to shadowed page table
6184 	 * and it failed try to unshadow page and re-enter the
6185 	 * guest to let CPU execute the instruction.
6186 	 */
6187 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6188 
6189 	/*
6190 	 * If the access faults on its page table, it can not
6191 	 * be fixed by unprotecting shadow page and it should
6192 	 * be reported to userspace.
6193 	 */
6194 	return !write_fault_to_shadow_pgtable;
6195 }
6196 
6197 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6198 			      unsigned long cr2,  int emulation_type)
6199 {
6200 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6201 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6202 
6203 	last_retry_eip = vcpu->arch.last_retry_eip;
6204 	last_retry_addr = vcpu->arch.last_retry_addr;
6205 
6206 	/*
6207 	 * If the emulation is caused by #PF and it is non-page_table
6208 	 * writing instruction, it means the VM-EXIT is caused by shadow
6209 	 * page protected, we can zap the shadow page and retry this
6210 	 * instruction directly.
6211 	 *
6212 	 * Note: if the guest uses a non-page-table modifying instruction
6213 	 * on the PDE that points to the instruction, then we will unmap
6214 	 * the instruction and go to an infinite loop. So, we cache the
6215 	 * last retried eip and the last fault address, if we meet the eip
6216 	 * and the address again, we can break out of the potential infinite
6217 	 * loop.
6218 	 */
6219 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6220 
6221 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6222 		return false;
6223 
6224 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6225 		return false;
6226 
6227 	if (x86_page_table_writing_insn(ctxt))
6228 		return false;
6229 
6230 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6231 		return false;
6232 
6233 	vcpu->arch.last_retry_eip = ctxt->eip;
6234 	vcpu->arch.last_retry_addr = cr2;
6235 
6236 	if (!vcpu->arch.mmu->direct_map)
6237 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6238 
6239 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6240 
6241 	return true;
6242 }
6243 
6244 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6245 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6246 
6247 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6248 {
6249 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6250 		/* This is a good place to trace that we are exiting SMM.  */
6251 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6252 
6253 		/* Process a latched INIT or SMI, if any.  */
6254 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6255 	}
6256 
6257 	kvm_mmu_reset_context(vcpu);
6258 }
6259 
6260 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6261 				unsigned long *db)
6262 {
6263 	u32 dr6 = 0;
6264 	int i;
6265 	u32 enable, rwlen;
6266 
6267 	enable = dr7;
6268 	rwlen = dr7 >> 16;
6269 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6270 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6271 			dr6 |= (1 << i);
6272 	return dr6;
6273 }
6274 
6275 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6276 {
6277 	struct kvm_run *kvm_run = vcpu->run;
6278 
6279 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6280 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6281 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6282 		kvm_run->debug.arch.exception = DB_VECTOR;
6283 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6284 		*r = EMULATE_USER_EXIT;
6285 	} else {
6286 		kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6287 	}
6288 }
6289 
6290 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6291 {
6292 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6293 	int r = EMULATE_DONE;
6294 
6295 	kvm_x86_ops->skip_emulated_instruction(vcpu);
6296 
6297 	/*
6298 	 * rflags is the old, "raw" value of the flags.  The new value has
6299 	 * not been saved yet.
6300 	 *
6301 	 * This is correct even for TF set by the guest, because "the
6302 	 * processor will not generate this exception after the instruction
6303 	 * that sets the TF flag".
6304 	 */
6305 	if (unlikely(rflags & X86_EFLAGS_TF))
6306 		kvm_vcpu_do_singlestep(vcpu, &r);
6307 	return r == EMULATE_DONE;
6308 }
6309 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6310 
6311 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6312 {
6313 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6314 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6315 		struct kvm_run *kvm_run = vcpu->run;
6316 		unsigned long eip = kvm_get_linear_rip(vcpu);
6317 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6318 					   vcpu->arch.guest_debug_dr7,
6319 					   vcpu->arch.eff_db);
6320 
6321 		if (dr6 != 0) {
6322 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6323 			kvm_run->debug.arch.pc = eip;
6324 			kvm_run->debug.arch.exception = DB_VECTOR;
6325 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6326 			*r = EMULATE_USER_EXIT;
6327 			return true;
6328 		}
6329 	}
6330 
6331 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6332 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6333 		unsigned long eip = kvm_get_linear_rip(vcpu);
6334 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6335 					   vcpu->arch.dr7,
6336 					   vcpu->arch.db);
6337 
6338 		if (dr6 != 0) {
6339 			vcpu->arch.dr6 &= ~15;
6340 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6341 			kvm_queue_exception(vcpu, DB_VECTOR);
6342 			*r = EMULATE_DONE;
6343 			return true;
6344 		}
6345 	}
6346 
6347 	return false;
6348 }
6349 
6350 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6351 {
6352 	switch (ctxt->opcode_len) {
6353 	case 1:
6354 		switch (ctxt->b) {
6355 		case 0xe4:	/* IN */
6356 		case 0xe5:
6357 		case 0xec:
6358 		case 0xed:
6359 		case 0xe6:	/* OUT */
6360 		case 0xe7:
6361 		case 0xee:
6362 		case 0xef:
6363 		case 0x6c:	/* INS */
6364 		case 0x6d:
6365 		case 0x6e:	/* OUTS */
6366 		case 0x6f:
6367 			return true;
6368 		}
6369 		break;
6370 	case 2:
6371 		switch (ctxt->b) {
6372 		case 0x33:	/* RDPMC */
6373 			return true;
6374 		}
6375 		break;
6376 	}
6377 
6378 	return false;
6379 }
6380 
6381 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6382 			    unsigned long cr2,
6383 			    int emulation_type,
6384 			    void *insn,
6385 			    int insn_len)
6386 {
6387 	int r;
6388 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6389 	bool writeback = true;
6390 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6391 
6392 	vcpu->arch.l1tf_flush_l1d = true;
6393 
6394 	/*
6395 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6396 	 * never reused.
6397 	 */
6398 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6399 	kvm_clear_exception_queue(vcpu);
6400 
6401 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6402 		init_emulate_ctxt(vcpu);
6403 
6404 		/*
6405 		 * We will reenter on the same instruction since
6406 		 * we do not set complete_userspace_io.  This does not
6407 		 * handle watchpoints yet, those would be handled in
6408 		 * the emulate_ops.
6409 		 */
6410 		if (!(emulation_type & EMULTYPE_SKIP) &&
6411 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6412 			return r;
6413 
6414 		ctxt->interruptibility = 0;
6415 		ctxt->have_exception = false;
6416 		ctxt->exception.vector = -1;
6417 		ctxt->perm_ok = false;
6418 
6419 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6420 
6421 		r = x86_decode_insn(ctxt, insn, insn_len);
6422 
6423 		trace_kvm_emulate_insn_start(vcpu);
6424 		++vcpu->stat.insn_emulation;
6425 		if (r != EMULATION_OK)  {
6426 			if (emulation_type & EMULTYPE_TRAP_UD)
6427 				return EMULATE_FAIL;
6428 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6429 						emulation_type))
6430 				return EMULATE_DONE;
6431 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6432 				return EMULATE_DONE;
6433 			if (emulation_type & EMULTYPE_SKIP)
6434 				return EMULATE_FAIL;
6435 			return handle_emulation_failure(vcpu, emulation_type);
6436 		}
6437 	}
6438 
6439 	if ((emulation_type & EMULTYPE_VMWARE) &&
6440 	    !is_vmware_backdoor_opcode(ctxt))
6441 		return EMULATE_FAIL;
6442 
6443 	if (emulation_type & EMULTYPE_SKIP) {
6444 		kvm_rip_write(vcpu, ctxt->_eip);
6445 		if (ctxt->eflags & X86_EFLAGS_RF)
6446 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6447 		return EMULATE_DONE;
6448 	}
6449 
6450 	if (retry_instruction(ctxt, cr2, emulation_type))
6451 		return EMULATE_DONE;
6452 
6453 	/* this is needed for vmware backdoor interface to work since it
6454 	   changes registers values  during IO operation */
6455 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6456 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6457 		emulator_invalidate_register_cache(ctxt);
6458 	}
6459 
6460 restart:
6461 	/* Save the faulting GPA (cr2) in the address field */
6462 	ctxt->exception.address = cr2;
6463 
6464 	r = x86_emulate_insn(ctxt);
6465 
6466 	if (r == EMULATION_INTERCEPTED)
6467 		return EMULATE_DONE;
6468 
6469 	if (r == EMULATION_FAILED) {
6470 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6471 					emulation_type))
6472 			return EMULATE_DONE;
6473 
6474 		return handle_emulation_failure(vcpu, emulation_type);
6475 	}
6476 
6477 	if (ctxt->have_exception) {
6478 		r = EMULATE_DONE;
6479 		if (inject_emulated_exception(vcpu))
6480 			return r;
6481 	} else if (vcpu->arch.pio.count) {
6482 		if (!vcpu->arch.pio.in) {
6483 			/* FIXME: return into emulator if single-stepping.  */
6484 			vcpu->arch.pio.count = 0;
6485 		} else {
6486 			writeback = false;
6487 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6488 		}
6489 		r = EMULATE_USER_EXIT;
6490 	} else if (vcpu->mmio_needed) {
6491 		if (!vcpu->mmio_is_write)
6492 			writeback = false;
6493 		r = EMULATE_USER_EXIT;
6494 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6495 	} else if (r == EMULATION_RESTART)
6496 		goto restart;
6497 	else
6498 		r = EMULATE_DONE;
6499 
6500 	if (writeback) {
6501 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6502 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6503 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6504 		kvm_rip_write(vcpu, ctxt->eip);
6505 		if (r == EMULATE_DONE && ctxt->tf)
6506 			kvm_vcpu_do_singlestep(vcpu, &r);
6507 		if (!ctxt->have_exception ||
6508 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6509 			__kvm_set_rflags(vcpu, ctxt->eflags);
6510 
6511 		/*
6512 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6513 		 * do nothing, and it will be requested again as soon as
6514 		 * the shadow expires.  But we still need to check here,
6515 		 * because POPF has no interrupt shadow.
6516 		 */
6517 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6518 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6519 	} else
6520 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6521 
6522 	return r;
6523 }
6524 
6525 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6526 {
6527 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6528 }
6529 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6530 
6531 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6532 					void *insn, int insn_len)
6533 {
6534 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6535 }
6536 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6537 
6538 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6539 {
6540 	vcpu->arch.pio.count = 0;
6541 
6542 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6543 		return 1;
6544 
6545 	return kvm_skip_emulated_instruction(vcpu);
6546 }
6547 
6548 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6549 			    unsigned short port)
6550 {
6551 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6552 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6553 					    size, port, &val, 1);
6554 
6555 	if (!ret) {
6556 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6557 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6558 	}
6559 	return ret;
6560 }
6561 
6562 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6563 {
6564 	unsigned long val;
6565 
6566 	/* We should only ever be called with arch.pio.count equal to 1 */
6567 	BUG_ON(vcpu->arch.pio.count != 1);
6568 
6569 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6570 		vcpu->arch.pio.count = 0;
6571 		return 1;
6572 	}
6573 
6574 	/* For size less than 4 we merge, else we zero extend */
6575 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6576 					: 0;
6577 
6578 	/*
6579 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6580 	 * the copy and tracing
6581 	 */
6582 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6583 				 vcpu->arch.pio.port, &val, 1);
6584 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6585 
6586 	return kvm_skip_emulated_instruction(vcpu);
6587 }
6588 
6589 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6590 			   unsigned short port)
6591 {
6592 	unsigned long val;
6593 	int ret;
6594 
6595 	/* For size less than 4 we merge, else we zero extend */
6596 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6597 
6598 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6599 				       &val, 1);
6600 	if (ret) {
6601 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6602 		return ret;
6603 	}
6604 
6605 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6606 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6607 
6608 	return 0;
6609 }
6610 
6611 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6612 {
6613 	int ret;
6614 
6615 	if (in)
6616 		ret = kvm_fast_pio_in(vcpu, size, port);
6617 	else
6618 		ret = kvm_fast_pio_out(vcpu, size, port);
6619 	return ret && kvm_skip_emulated_instruction(vcpu);
6620 }
6621 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6622 
6623 static int kvmclock_cpu_down_prep(unsigned int cpu)
6624 {
6625 	__this_cpu_write(cpu_tsc_khz, 0);
6626 	return 0;
6627 }
6628 
6629 static void tsc_khz_changed(void *data)
6630 {
6631 	struct cpufreq_freqs *freq = data;
6632 	unsigned long khz = 0;
6633 
6634 	if (data)
6635 		khz = freq->new;
6636 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6637 		khz = cpufreq_quick_get(raw_smp_processor_id());
6638 	if (!khz)
6639 		khz = tsc_khz;
6640 	__this_cpu_write(cpu_tsc_khz, khz);
6641 }
6642 
6643 #ifdef CONFIG_X86_64
6644 static void kvm_hyperv_tsc_notifier(void)
6645 {
6646 	struct kvm *kvm;
6647 	struct kvm_vcpu *vcpu;
6648 	int cpu;
6649 
6650 	spin_lock(&kvm_lock);
6651 	list_for_each_entry(kvm, &vm_list, vm_list)
6652 		kvm_make_mclock_inprogress_request(kvm);
6653 
6654 	hyperv_stop_tsc_emulation();
6655 
6656 	/* TSC frequency always matches when on Hyper-V */
6657 	for_each_present_cpu(cpu)
6658 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6659 	kvm_max_guest_tsc_khz = tsc_khz;
6660 
6661 	list_for_each_entry(kvm, &vm_list, vm_list) {
6662 		struct kvm_arch *ka = &kvm->arch;
6663 
6664 		spin_lock(&ka->pvclock_gtod_sync_lock);
6665 
6666 		pvclock_update_vm_gtod_copy(kvm);
6667 
6668 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6669 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6670 
6671 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6672 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6673 
6674 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6675 	}
6676 	spin_unlock(&kvm_lock);
6677 }
6678 #endif
6679 
6680 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6681 				     void *data)
6682 {
6683 	struct cpufreq_freqs *freq = data;
6684 	struct kvm *kvm;
6685 	struct kvm_vcpu *vcpu;
6686 	int i, send_ipi = 0;
6687 
6688 	/*
6689 	 * We allow guests to temporarily run on slowing clocks,
6690 	 * provided we notify them after, or to run on accelerating
6691 	 * clocks, provided we notify them before.  Thus time never
6692 	 * goes backwards.
6693 	 *
6694 	 * However, we have a problem.  We can't atomically update
6695 	 * the frequency of a given CPU from this function; it is
6696 	 * merely a notifier, which can be called from any CPU.
6697 	 * Changing the TSC frequency at arbitrary points in time
6698 	 * requires a recomputation of local variables related to
6699 	 * the TSC for each VCPU.  We must flag these local variables
6700 	 * to be updated and be sure the update takes place with the
6701 	 * new frequency before any guests proceed.
6702 	 *
6703 	 * Unfortunately, the combination of hotplug CPU and frequency
6704 	 * change creates an intractable locking scenario; the order
6705 	 * of when these callouts happen is undefined with respect to
6706 	 * CPU hotplug, and they can race with each other.  As such,
6707 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6708 	 * undefined; you can actually have a CPU frequency change take
6709 	 * place in between the computation of X and the setting of the
6710 	 * variable.  To protect against this problem, all updates of
6711 	 * the per_cpu tsc_khz variable are done in an interrupt
6712 	 * protected IPI, and all callers wishing to update the value
6713 	 * must wait for a synchronous IPI to complete (which is trivial
6714 	 * if the caller is on the CPU already).  This establishes the
6715 	 * necessary total order on variable updates.
6716 	 *
6717 	 * Note that because a guest time update may take place
6718 	 * anytime after the setting of the VCPU's request bit, the
6719 	 * correct TSC value must be set before the request.  However,
6720 	 * to ensure the update actually makes it to any guest which
6721 	 * starts running in hardware virtualization between the set
6722 	 * and the acquisition of the spinlock, we must also ping the
6723 	 * CPU after setting the request bit.
6724 	 *
6725 	 */
6726 
6727 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6728 		return 0;
6729 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6730 		return 0;
6731 
6732 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6733 
6734 	spin_lock(&kvm_lock);
6735 	list_for_each_entry(kvm, &vm_list, vm_list) {
6736 		kvm_for_each_vcpu(i, vcpu, kvm) {
6737 			if (vcpu->cpu != freq->cpu)
6738 				continue;
6739 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6740 			if (vcpu->cpu != smp_processor_id())
6741 				send_ipi = 1;
6742 		}
6743 	}
6744 	spin_unlock(&kvm_lock);
6745 
6746 	if (freq->old < freq->new && send_ipi) {
6747 		/*
6748 		 * We upscale the frequency.  Must make the guest
6749 		 * doesn't see old kvmclock values while running with
6750 		 * the new frequency, otherwise we risk the guest sees
6751 		 * time go backwards.
6752 		 *
6753 		 * In case we update the frequency for another cpu
6754 		 * (which might be in guest context) send an interrupt
6755 		 * to kick the cpu out of guest context.  Next time
6756 		 * guest context is entered kvmclock will be updated,
6757 		 * so the guest will not see stale values.
6758 		 */
6759 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6760 	}
6761 	return 0;
6762 }
6763 
6764 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6765 	.notifier_call  = kvmclock_cpufreq_notifier
6766 };
6767 
6768 static int kvmclock_cpu_online(unsigned int cpu)
6769 {
6770 	tsc_khz_changed(NULL);
6771 	return 0;
6772 }
6773 
6774 static void kvm_timer_init(void)
6775 {
6776 	max_tsc_khz = tsc_khz;
6777 
6778 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6779 #ifdef CONFIG_CPU_FREQ
6780 		struct cpufreq_policy policy;
6781 		int cpu;
6782 
6783 		memset(&policy, 0, sizeof(policy));
6784 		cpu = get_cpu();
6785 		cpufreq_get_policy(&policy, cpu);
6786 		if (policy.cpuinfo.max_freq)
6787 			max_tsc_khz = policy.cpuinfo.max_freq;
6788 		put_cpu();
6789 #endif
6790 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6791 					  CPUFREQ_TRANSITION_NOTIFIER);
6792 	}
6793 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6794 
6795 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6796 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6797 }
6798 
6799 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6800 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6801 
6802 int kvm_is_in_guest(void)
6803 {
6804 	return __this_cpu_read(current_vcpu) != NULL;
6805 }
6806 
6807 static int kvm_is_user_mode(void)
6808 {
6809 	int user_mode = 3;
6810 
6811 	if (__this_cpu_read(current_vcpu))
6812 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6813 
6814 	return user_mode != 0;
6815 }
6816 
6817 static unsigned long kvm_get_guest_ip(void)
6818 {
6819 	unsigned long ip = 0;
6820 
6821 	if (__this_cpu_read(current_vcpu))
6822 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6823 
6824 	return ip;
6825 }
6826 
6827 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6828 	.is_in_guest		= kvm_is_in_guest,
6829 	.is_user_mode		= kvm_is_user_mode,
6830 	.get_guest_ip		= kvm_get_guest_ip,
6831 };
6832 
6833 static void kvm_set_mmio_spte_mask(void)
6834 {
6835 	u64 mask;
6836 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6837 
6838 	/*
6839 	 * Set the reserved bits and the present bit of an paging-structure
6840 	 * entry to generate page fault with PFER.RSV = 1.
6841 	 */
6842 
6843 	/*
6844 	 * Mask the uppermost physical address bit, which would be reserved as
6845 	 * long as the supported physical address width is less than 52.
6846 	 */
6847 	mask = 1ull << 51;
6848 
6849 	/* Set the present bit. */
6850 	mask |= 1ull;
6851 
6852 	/*
6853 	 * If reserved bit is not supported, clear the present bit to disable
6854 	 * mmio page fault.
6855 	 */
6856 	if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6857 		mask &= ~1ull;
6858 
6859 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6860 }
6861 
6862 #ifdef CONFIG_X86_64
6863 static void pvclock_gtod_update_fn(struct work_struct *work)
6864 {
6865 	struct kvm *kvm;
6866 
6867 	struct kvm_vcpu *vcpu;
6868 	int i;
6869 
6870 	spin_lock(&kvm_lock);
6871 	list_for_each_entry(kvm, &vm_list, vm_list)
6872 		kvm_for_each_vcpu(i, vcpu, kvm)
6873 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6874 	atomic_set(&kvm_guest_has_master_clock, 0);
6875 	spin_unlock(&kvm_lock);
6876 }
6877 
6878 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6879 
6880 /*
6881  * Notification about pvclock gtod data update.
6882  */
6883 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6884 			       void *priv)
6885 {
6886 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6887 	struct timekeeper *tk = priv;
6888 
6889 	update_pvclock_gtod(tk);
6890 
6891 	/* disable master clock if host does not trust, or does not
6892 	 * use, TSC based clocksource.
6893 	 */
6894 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6895 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6896 		queue_work(system_long_wq, &pvclock_gtod_work);
6897 
6898 	return 0;
6899 }
6900 
6901 static struct notifier_block pvclock_gtod_notifier = {
6902 	.notifier_call = pvclock_gtod_notify,
6903 };
6904 #endif
6905 
6906 int kvm_arch_init(void *opaque)
6907 {
6908 	int r;
6909 	struct kvm_x86_ops *ops = opaque;
6910 
6911 	if (kvm_x86_ops) {
6912 		printk(KERN_ERR "kvm: already loaded the other module\n");
6913 		r = -EEXIST;
6914 		goto out;
6915 	}
6916 
6917 	if (!ops->cpu_has_kvm_support()) {
6918 		printk(KERN_ERR "kvm: no hardware support\n");
6919 		r = -EOPNOTSUPP;
6920 		goto out;
6921 	}
6922 	if (ops->disabled_by_bios()) {
6923 		printk(KERN_ERR "kvm: disabled by bios\n");
6924 		r = -EOPNOTSUPP;
6925 		goto out;
6926 	}
6927 
6928 	/*
6929 	 * KVM explicitly assumes that the guest has an FPU and
6930 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
6931 	 * vCPU's FPU state as a fxregs_state struct.
6932 	 */
6933 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
6934 		printk(KERN_ERR "kvm: inadequate fpu\n");
6935 		r = -EOPNOTSUPP;
6936 		goto out;
6937 	}
6938 
6939 	r = -ENOMEM;
6940 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
6941 					  __alignof__(struct fpu), SLAB_ACCOUNT,
6942 					  NULL);
6943 	if (!x86_fpu_cache) {
6944 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
6945 		goto out;
6946 	}
6947 
6948 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6949 	if (!shared_msrs) {
6950 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6951 		goto out_free_x86_fpu_cache;
6952 	}
6953 
6954 	r = kvm_mmu_module_init();
6955 	if (r)
6956 		goto out_free_percpu;
6957 
6958 	kvm_set_mmio_spte_mask();
6959 
6960 	kvm_x86_ops = ops;
6961 
6962 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6963 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6964 			PT_PRESENT_MASK, 0, sme_me_mask);
6965 	kvm_timer_init();
6966 
6967 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6968 
6969 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6970 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6971 
6972 	kvm_lapic_init();
6973 #ifdef CONFIG_X86_64
6974 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6975 
6976 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6977 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6978 #endif
6979 
6980 	return 0;
6981 
6982 out_free_percpu:
6983 	free_percpu(shared_msrs);
6984 out_free_x86_fpu_cache:
6985 	kmem_cache_destroy(x86_fpu_cache);
6986 out:
6987 	return r;
6988 }
6989 
6990 void kvm_arch_exit(void)
6991 {
6992 #ifdef CONFIG_X86_64
6993 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6994 		clear_hv_tscchange_cb();
6995 #endif
6996 	kvm_lapic_exit();
6997 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6998 
6999 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7000 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7001 					    CPUFREQ_TRANSITION_NOTIFIER);
7002 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7003 #ifdef CONFIG_X86_64
7004 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7005 #endif
7006 	kvm_x86_ops = NULL;
7007 	kvm_mmu_module_exit();
7008 	free_percpu(shared_msrs);
7009 	kmem_cache_destroy(x86_fpu_cache);
7010 }
7011 
7012 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7013 {
7014 	++vcpu->stat.halt_exits;
7015 	if (lapic_in_kernel(vcpu)) {
7016 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7017 		return 1;
7018 	} else {
7019 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7020 		return 0;
7021 	}
7022 }
7023 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7024 
7025 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7026 {
7027 	int ret = kvm_skip_emulated_instruction(vcpu);
7028 	/*
7029 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7030 	 * KVM_EXIT_DEBUG here.
7031 	 */
7032 	return kvm_vcpu_halt(vcpu) && ret;
7033 }
7034 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7035 
7036 #ifdef CONFIG_X86_64
7037 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7038 			        unsigned long clock_type)
7039 {
7040 	struct kvm_clock_pairing clock_pairing;
7041 	struct timespec64 ts;
7042 	u64 cycle;
7043 	int ret;
7044 
7045 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7046 		return -KVM_EOPNOTSUPP;
7047 
7048 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7049 		return -KVM_EOPNOTSUPP;
7050 
7051 	clock_pairing.sec = ts.tv_sec;
7052 	clock_pairing.nsec = ts.tv_nsec;
7053 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7054 	clock_pairing.flags = 0;
7055 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7056 
7057 	ret = 0;
7058 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7059 			    sizeof(struct kvm_clock_pairing)))
7060 		ret = -KVM_EFAULT;
7061 
7062 	return ret;
7063 }
7064 #endif
7065 
7066 /*
7067  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7068  *
7069  * @apicid - apicid of vcpu to be kicked.
7070  */
7071 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7072 {
7073 	struct kvm_lapic_irq lapic_irq;
7074 
7075 	lapic_irq.shorthand = 0;
7076 	lapic_irq.dest_mode = 0;
7077 	lapic_irq.level = 0;
7078 	lapic_irq.dest_id = apicid;
7079 	lapic_irq.msi_redir_hint = false;
7080 
7081 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7082 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7083 }
7084 
7085 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7086 {
7087 	if (!lapic_in_kernel(vcpu)) {
7088 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7089 		return;
7090 	}
7091 	if (!vcpu->arch.apicv_active)
7092 		return;
7093 
7094 	vcpu->arch.apicv_active = false;
7095 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7096 }
7097 
7098 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7099 {
7100 	unsigned long nr, a0, a1, a2, a3, ret;
7101 	int op_64_bit;
7102 
7103 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7104 		return kvm_hv_hypercall(vcpu);
7105 
7106 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
7107 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
7108 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
7109 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
7110 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
7111 
7112 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7113 
7114 	op_64_bit = is_64_bit_mode(vcpu);
7115 	if (!op_64_bit) {
7116 		nr &= 0xFFFFFFFF;
7117 		a0 &= 0xFFFFFFFF;
7118 		a1 &= 0xFFFFFFFF;
7119 		a2 &= 0xFFFFFFFF;
7120 		a3 &= 0xFFFFFFFF;
7121 	}
7122 
7123 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7124 		ret = -KVM_EPERM;
7125 		goto out;
7126 	}
7127 
7128 	switch (nr) {
7129 	case KVM_HC_VAPIC_POLL_IRQ:
7130 		ret = 0;
7131 		break;
7132 	case KVM_HC_KICK_CPU:
7133 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7134 		ret = 0;
7135 		break;
7136 #ifdef CONFIG_X86_64
7137 	case KVM_HC_CLOCK_PAIRING:
7138 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7139 		break;
7140 #endif
7141 	case KVM_HC_SEND_IPI:
7142 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7143 		break;
7144 	default:
7145 		ret = -KVM_ENOSYS;
7146 		break;
7147 	}
7148 out:
7149 	if (!op_64_bit)
7150 		ret = (u32)ret;
7151 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7152 
7153 	++vcpu->stat.hypercalls;
7154 	return kvm_skip_emulated_instruction(vcpu);
7155 }
7156 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7157 
7158 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7159 {
7160 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7161 	char instruction[3];
7162 	unsigned long rip = kvm_rip_read(vcpu);
7163 
7164 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7165 
7166 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7167 		&ctxt->exception);
7168 }
7169 
7170 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7171 {
7172 	return vcpu->run->request_interrupt_window &&
7173 		likely(!pic_in_kernel(vcpu->kvm));
7174 }
7175 
7176 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7177 {
7178 	struct kvm_run *kvm_run = vcpu->run;
7179 
7180 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7181 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7182 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7183 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7184 	kvm_run->ready_for_interrupt_injection =
7185 		pic_in_kernel(vcpu->kvm) ||
7186 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7187 }
7188 
7189 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7190 {
7191 	int max_irr, tpr;
7192 
7193 	if (!kvm_x86_ops->update_cr8_intercept)
7194 		return;
7195 
7196 	if (!lapic_in_kernel(vcpu))
7197 		return;
7198 
7199 	if (vcpu->arch.apicv_active)
7200 		return;
7201 
7202 	if (!vcpu->arch.apic->vapic_addr)
7203 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7204 	else
7205 		max_irr = -1;
7206 
7207 	if (max_irr != -1)
7208 		max_irr >>= 4;
7209 
7210 	tpr = kvm_lapic_get_cr8(vcpu);
7211 
7212 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7213 }
7214 
7215 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7216 {
7217 	int r;
7218 
7219 	/* try to reinject previous events if any */
7220 
7221 	if (vcpu->arch.exception.injected)
7222 		kvm_x86_ops->queue_exception(vcpu);
7223 	/*
7224 	 * Do not inject an NMI or interrupt if there is a pending
7225 	 * exception.  Exceptions and interrupts are recognized at
7226 	 * instruction boundaries, i.e. the start of an instruction.
7227 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7228 	 * NMIs and interrupts, i.e. traps are recognized before an
7229 	 * NMI/interrupt that's pending on the same instruction.
7230 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7231 	 * priority, but are only generated (pended) during instruction
7232 	 * execution, i.e. a pending fault-like exception means the
7233 	 * fault occurred on the *previous* instruction and must be
7234 	 * serviced prior to recognizing any new events in order to
7235 	 * fully complete the previous instruction.
7236 	 */
7237 	else if (!vcpu->arch.exception.pending) {
7238 		if (vcpu->arch.nmi_injected)
7239 			kvm_x86_ops->set_nmi(vcpu);
7240 		else if (vcpu->arch.interrupt.injected)
7241 			kvm_x86_ops->set_irq(vcpu);
7242 	}
7243 
7244 	/*
7245 	 * Call check_nested_events() even if we reinjected a previous event
7246 	 * in order for caller to determine if it should require immediate-exit
7247 	 * from L2 to L1 due to pending L1 events which require exit
7248 	 * from L2 to L1.
7249 	 */
7250 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7251 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7252 		if (r != 0)
7253 			return r;
7254 	}
7255 
7256 	/* try to inject new event if pending */
7257 	if (vcpu->arch.exception.pending) {
7258 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7259 					vcpu->arch.exception.has_error_code,
7260 					vcpu->arch.exception.error_code);
7261 
7262 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7263 		vcpu->arch.exception.pending = false;
7264 		vcpu->arch.exception.injected = true;
7265 
7266 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7267 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7268 					     X86_EFLAGS_RF);
7269 
7270 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7271 			/*
7272 			 * This code assumes that nSVM doesn't use
7273 			 * check_nested_events(). If it does, the
7274 			 * DR6/DR7 changes should happen before L1
7275 			 * gets a #VMEXIT for an intercepted #DB in
7276 			 * L2.  (Under VMX, on the other hand, the
7277 			 * DR6/DR7 changes should not happen in the
7278 			 * event of a VM-exit to L1 for an intercepted
7279 			 * #DB in L2.)
7280 			 */
7281 			kvm_deliver_exception_payload(vcpu);
7282 			if (vcpu->arch.dr7 & DR7_GD) {
7283 				vcpu->arch.dr7 &= ~DR7_GD;
7284 				kvm_update_dr7(vcpu);
7285 			}
7286 		}
7287 
7288 		kvm_x86_ops->queue_exception(vcpu);
7289 	}
7290 
7291 	/* Don't consider new event if we re-injected an event */
7292 	if (kvm_event_needs_reinjection(vcpu))
7293 		return 0;
7294 
7295 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7296 	    kvm_x86_ops->smi_allowed(vcpu)) {
7297 		vcpu->arch.smi_pending = false;
7298 		++vcpu->arch.smi_count;
7299 		enter_smm(vcpu);
7300 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7301 		--vcpu->arch.nmi_pending;
7302 		vcpu->arch.nmi_injected = true;
7303 		kvm_x86_ops->set_nmi(vcpu);
7304 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7305 		/*
7306 		 * Because interrupts can be injected asynchronously, we are
7307 		 * calling check_nested_events again here to avoid a race condition.
7308 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7309 		 * proposal and current concerns.  Perhaps we should be setting
7310 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7311 		 */
7312 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7313 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7314 			if (r != 0)
7315 				return r;
7316 		}
7317 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7318 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7319 					    false);
7320 			kvm_x86_ops->set_irq(vcpu);
7321 		}
7322 	}
7323 
7324 	return 0;
7325 }
7326 
7327 static void process_nmi(struct kvm_vcpu *vcpu)
7328 {
7329 	unsigned limit = 2;
7330 
7331 	/*
7332 	 * x86 is limited to one NMI running, and one NMI pending after it.
7333 	 * If an NMI is already in progress, limit further NMIs to just one.
7334 	 * Otherwise, allow two (and we'll inject the first one immediately).
7335 	 */
7336 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7337 		limit = 1;
7338 
7339 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7340 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7341 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7342 }
7343 
7344 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7345 {
7346 	u32 flags = 0;
7347 	flags |= seg->g       << 23;
7348 	flags |= seg->db      << 22;
7349 	flags |= seg->l       << 21;
7350 	flags |= seg->avl     << 20;
7351 	flags |= seg->present << 15;
7352 	flags |= seg->dpl     << 13;
7353 	flags |= seg->s       << 12;
7354 	flags |= seg->type    << 8;
7355 	return flags;
7356 }
7357 
7358 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7359 {
7360 	struct kvm_segment seg;
7361 	int offset;
7362 
7363 	kvm_get_segment(vcpu, &seg, n);
7364 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7365 
7366 	if (n < 3)
7367 		offset = 0x7f84 + n * 12;
7368 	else
7369 		offset = 0x7f2c + (n - 3) * 12;
7370 
7371 	put_smstate(u32, buf, offset + 8, seg.base);
7372 	put_smstate(u32, buf, offset + 4, seg.limit);
7373 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7374 }
7375 
7376 #ifdef CONFIG_X86_64
7377 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7378 {
7379 	struct kvm_segment seg;
7380 	int offset;
7381 	u16 flags;
7382 
7383 	kvm_get_segment(vcpu, &seg, n);
7384 	offset = 0x7e00 + n * 16;
7385 
7386 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7387 	put_smstate(u16, buf, offset, seg.selector);
7388 	put_smstate(u16, buf, offset + 2, flags);
7389 	put_smstate(u32, buf, offset + 4, seg.limit);
7390 	put_smstate(u64, buf, offset + 8, seg.base);
7391 }
7392 #endif
7393 
7394 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7395 {
7396 	struct desc_ptr dt;
7397 	struct kvm_segment seg;
7398 	unsigned long val;
7399 	int i;
7400 
7401 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7402 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7403 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7404 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7405 
7406 	for (i = 0; i < 8; i++)
7407 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7408 
7409 	kvm_get_dr(vcpu, 6, &val);
7410 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7411 	kvm_get_dr(vcpu, 7, &val);
7412 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7413 
7414 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7415 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7416 	put_smstate(u32, buf, 0x7f64, seg.base);
7417 	put_smstate(u32, buf, 0x7f60, seg.limit);
7418 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7419 
7420 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7421 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7422 	put_smstate(u32, buf, 0x7f80, seg.base);
7423 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7424 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7425 
7426 	kvm_x86_ops->get_gdt(vcpu, &dt);
7427 	put_smstate(u32, buf, 0x7f74, dt.address);
7428 	put_smstate(u32, buf, 0x7f70, dt.size);
7429 
7430 	kvm_x86_ops->get_idt(vcpu, &dt);
7431 	put_smstate(u32, buf, 0x7f58, dt.address);
7432 	put_smstate(u32, buf, 0x7f54, dt.size);
7433 
7434 	for (i = 0; i < 6; i++)
7435 		enter_smm_save_seg_32(vcpu, buf, i);
7436 
7437 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7438 
7439 	/* revision id */
7440 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7441 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7442 }
7443 
7444 #ifdef CONFIG_X86_64
7445 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7446 {
7447 	struct desc_ptr dt;
7448 	struct kvm_segment seg;
7449 	unsigned long val;
7450 	int i;
7451 
7452 	for (i = 0; i < 16; i++)
7453 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7454 
7455 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7456 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7457 
7458 	kvm_get_dr(vcpu, 6, &val);
7459 	put_smstate(u64, buf, 0x7f68, val);
7460 	kvm_get_dr(vcpu, 7, &val);
7461 	put_smstate(u64, buf, 0x7f60, val);
7462 
7463 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7464 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7465 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7466 
7467 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7468 
7469 	/* revision id */
7470 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7471 
7472 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7473 
7474 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7475 	put_smstate(u16, buf, 0x7e90, seg.selector);
7476 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7477 	put_smstate(u32, buf, 0x7e94, seg.limit);
7478 	put_smstate(u64, buf, 0x7e98, seg.base);
7479 
7480 	kvm_x86_ops->get_idt(vcpu, &dt);
7481 	put_smstate(u32, buf, 0x7e84, dt.size);
7482 	put_smstate(u64, buf, 0x7e88, dt.address);
7483 
7484 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7485 	put_smstate(u16, buf, 0x7e70, seg.selector);
7486 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7487 	put_smstate(u32, buf, 0x7e74, seg.limit);
7488 	put_smstate(u64, buf, 0x7e78, seg.base);
7489 
7490 	kvm_x86_ops->get_gdt(vcpu, &dt);
7491 	put_smstate(u32, buf, 0x7e64, dt.size);
7492 	put_smstate(u64, buf, 0x7e68, dt.address);
7493 
7494 	for (i = 0; i < 6; i++)
7495 		enter_smm_save_seg_64(vcpu, buf, i);
7496 }
7497 #endif
7498 
7499 static void enter_smm(struct kvm_vcpu *vcpu)
7500 {
7501 	struct kvm_segment cs, ds;
7502 	struct desc_ptr dt;
7503 	char buf[512];
7504 	u32 cr0;
7505 
7506 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7507 	memset(buf, 0, 512);
7508 #ifdef CONFIG_X86_64
7509 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7510 		enter_smm_save_state_64(vcpu, buf);
7511 	else
7512 #endif
7513 		enter_smm_save_state_32(vcpu, buf);
7514 
7515 	/*
7516 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7517 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7518 	 * the SMM state-save area.
7519 	 */
7520 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7521 
7522 	vcpu->arch.hflags |= HF_SMM_MASK;
7523 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7524 
7525 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7526 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7527 	else
7528 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7529 
7530 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7531 	kvm_rip_write(vcpu, 0x8000);
7532 
7533 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7534 	kvm_x86_ops->set_cr0(vcpu, cr0);
7535 	vcpu->arch.cr0 = cr0;
7536 
7537 	kvm_x86_ops->set_cr4(vcpu, 0);
7538 
7539 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7540 	dt.address = dt.size = 0;
7541 	kvm_x86_ops->set_idt(vcpu, &dt);
7542 
7543 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7544 
7545 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7546 	cs.base = vcpu->arch.smbase;
7547 
7548 	ds.selector = 0;
7549 	ds.base = 0;
7550 
7551 	cs.limit    = ds.limit = 0xffffffff;
7552 	cs.type     = ds.type = 0x3;
7553 	cs.dpl      = ds.dpl = 0;
7554 	cs.db       = ds.db = 0;
7555 	cs.s        = ds.s = 1;
7556 	cs.l        = ds.l = 0;
7557 	cs.g        = ds.g = 1;
7558 	cs.avl      = ds.avl = 0;
7559 	cs.present  = ds.present = 1;
7560 	cs.unusable = ds.unusable = 0;
7561 	cs.padding  = ds.padding = 0;
7562 
7563 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7564 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7565 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7566 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7567 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7568 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7569 
7570 #ifdef CONFIG_X86_64
7571 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7572 		kvm_x86_ops->set_efer(vcpu, 0);
7573 #endif
7574 
7575 	kvm_update_cpuid(vcpu);
7576 	kvm_mmu_reset_context(vcpu);
7577 }
7578 
7579 static void process_smi(struct kvm_vcpu *vcpu)
7580 {
7581 	vcpu->arch.smi_pending = true;
7582 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7583 }
7584 
7585 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7586 {
7587 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7588 }
7589 
7590 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7591 {
7592 	if (!kvm_apic_present(vcpu))
7593 		return;
7594 
7595 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7596 
7597 	if (irqchip_split(vcpu->kvm))
7598 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7599 	else {
7600 		if (vcpu->arch.apicv_active)
7601 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7602 		if (ioapic_in_kernel(vcpu->kvm))
7603 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7604 	}
7605 
7606 	if (is_guest_mode(vcpu))
7607 		vcpu->arch.load_eoi_exitmap_pending = true;
7608 	else
7609 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7610 }
7611 
7612 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7613 {
7614 	u64 eoi_exit_bitmap[4];
7615 
7616 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7617 		return;
7618 
7619 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7620 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7621 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7622 }
7623 
7624 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7625 		unsigned long start, unsigned long end,
7626 		bool blockable)
7627 {
7628 	unsigned long apic_address;
7629 
7630 	/*
7631 	 * The physical address of apic access page is stored in the VMCS.
7632 	 * Update it when it becomes invalid.
7633 	 */
7634 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7635 	if (start <= apic_address && apic_address < end)
7636 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7637 
7638 	return 0;
7639 }
7640 
7641 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7642 {
7643 	struct page *page = NULL;
7644 
7645 	if (!lapic_in_kernel(vcpu))
7646 		return;
7647 
7648 	if (!kvm_x86_ops->set_apic_access_page_addr)
7649 		return;
7650 
7651 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7652 	if (is_error_page(page))
7653 		return;
7654 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7655 
7656 	/*
7657 	 * Do not pin apic access page in memory, the MMU notifier
7658 	 * will call us again if it is migrated or swapped out.
7659 	 */
7660 	put_page(page);
7661 }
7662 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7663 
7664 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7665 {
7666 	smp_send_reschedule(vcpu->cpu);
7667 }
7668 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7669 
7670 /*
7671  * Returns 1 to let vcpu_run() continue the guest execution loop without
7672  * exiting to the userspace.  Otherwise, the value will be returned to the
7673  * userspace.
7674  */
7675 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7676 {
7677 	int r;
7678 	bool req_int_win =
7679 		dm_request_for_irq_injection(vcpu) &&
7680 		kvm_cpu_accept_dm_intr(vcpu);
7681 
7682 	bool req_immediate_exit = false;
7683 
7684 	if (kvm_request_pending(vcpu)) {
7685 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7686 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7687 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7688 			kvm_mmu_unload(vcpu);
7689 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7690 			__kvm_migrate_timers(vcpu);
7691 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7692 			kvm_gen_update_masterclock(vcpu->kvm);
7693 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7694 			kvm_gen_kvmclock_update(vcpu);
7695 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7696 			r = kvm_guest_time_update(vcpu);
7697 			if (unlikely(r))
7698 				goto out;
7699 		}
7700 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7701 			kvm_mmu_sync_roots(vcpu);
7702 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7703 			kvm_mmu_load_cr3(vcpu);
7704 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7705 			kvm_vcpu_flush_tlb(vcpu, true);
7706 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7707 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7708 			r = 0;
7709 			goto out;
7710 		}
7711 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7712 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7713 			vcpu->mmio_needed = 0;
7714 			r = 0;
7715 			goto out;
7716 		}
7717 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7718 			/* Page is swapped out. Do synthetic halt */
7719 			vcpu->arch.apf.halted = true;
7720 			r = 1;
7721 			goto out;
7722 		}
7723 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7724 			record_steal_time(vcpu);
7725 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7726 			process_smi(vcpu);
7727 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7728 			process_nmi(vcpu);
7729 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7730 			kvm_pmu_handle_event(vcpu);
7731 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7732 			kvm_pmu_deliver_pmi(vcpu);
7733 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7734 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7735 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7736 				     vcpu->arch.ioapic_handled_vectors)) {
7737 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7738 				vcpu->run->eoi.vector =
7739 						vcpu->arch.pending_ioapic_eoi;
7740 				r = 0;
7741 				goto out;
7742 			}
7743 		}
7744 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7745 			vcpu_scan_ioapic(vcpu);
7746 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7747 			vcpu_load_eoi_exitmap(vcpu);
7748 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7749 			kvm_vcpu_reload_apic_access_page(vcpu);
7750 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7751 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7752 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7753 			r = 0;
7754 			goto out;
7755 		}
7756 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7757 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7758 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7759 			r = 0;
7760 			goto out;
7761 		}
7762 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7763 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7764 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7765 			r = 0;
7766 			goto out;
7767 		}
7768 
7769 		/*
7770 		 * KVM_REQ_HV_STIMER has to be processed after
7771 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7772 		 * depend on the guest clock being up-to-date
7773 		 */
7774 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7775 			kvm_hv_process_stimers(vcpu);
7776 	}
7777 
7778 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7779 		++vcpu->stat.req_event;
7780 		kvm_apic_accept_events(vcpu);
7781 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7782 			r = 1;
7783 			goto out;
7784 		}
7785 
7786 		if (inject_pending_event(vcpu, req_int_win) != 0)
7787 			req_immediate_exit = true;
7788 		else {
7789 			/* Enable SMI/NMI/IRQ window open exits if needed.
7790 			 *
7791 			 * SMIs have three cases:
7792 			 * 1) They can be nested, and then there is nothing to
7793 			 *    do here because RSM will cause a vmexit anyway.
7794 			 * 2) There is an ISA-specific reason why SMI cannot be
7795 			 *    injected, and the moment when this changes can be
7796 			 *    intercepted.
7797 			 * 3) Or the SMI can be pending because
7798 			 *    inject_pending_event has completed the injection
7799 			 *    of an IRQ or NMI from the previous vmexit, and
7800 			 *    then we request an immediate exit to inject the
7801 			 *    SMI.
7802 			 */
7803 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7804 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7805 					req_immediate_exit = true;
7806 			if (vcpu->arch.nmi_pending)
7807 				kvm_x86_ops->enable_nmi_window(vcpu);
7808 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7809 				kvm_x86_ops->enable_irq_window(vcpu);
7810 			WARN_ON(vcpu->arch.exception.pending);
7811 		}
7812 
7813 		if (kvm_lapic_enabled(vcpu)) {
7814 			update_cr8_intercept(vcpu);
7815 			kvm_lapic_sync_to_vapic(vcpu);
7816 		}
7817 	}
7818 
7819 	r = kvm_mmu_reload(vcpu);
7820 	if (unlikely(r)) {
7821 		goto cancel_injection;
7822 	}
7823 
7824 	preempt_disable();
7825 
7826 	kvm_x86_ops->prepare_guest_switch(vcpu);
7827 
7828 	/*
7829 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7830 	 * IPI are then delayed after guest entry, which ensures that they
7831 	 * result in virtual interrupt delivery.
7832 	 */
7833 	local_irq_disable();
7834 	vcpu->mode = IN_GUEST_MODE;
7835 
7836 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7837 
7838 	/*
7839 	 * 1) We should set ->mode before checking ->requests.  Please see
7840 	 * the comment in kvm_vcpu_exiting_guest_mode().
7841 	 *
7842 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
7843 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7844 	 * (see vmx_deliver_posted_interrupt).
7845 	 *
7846 	 * 3) This also orders the write to mode from any reads to the page
7847 	 * tables done while the VCPU is running.  Please see the comment
7848 	 * in kvm_flush_remote_tlbs.
7849 	 */
7850 	smp_mb__after_srcu_read_unlock();
7851 
7852 	/*
7853 	 * This handles the case where a posted interrupt was
7854 	 * notified with kvm_vcpu_kick.
7855 	 */
7856 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7857 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7858 
7859 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7860 	    || need_resched() || signal_pending(current)) {
7861 		vcpu->mode = OUTSIDE_GUEST_MODE;
7862 		smp_wmb();
7863 		local_irq_enable();
7864 		preempt_enable();
7865 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7866 		r = 1;
7867 		goto cancel_injection;
7868 	}
7869 
7870 	if (req_immediate_exit) {
7871 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7872 		kvm_x86_ops->request_immediate_exit(vcpu);
7873 	}
7874 
7875 	trace_kvm_entry(vcpu->vcpu_id);
7876 	if (lapic_timer_advance_ns)
7877 		wait_lapic_expire(vcpu);
7878 	guest_enter_irqoff();
7879 
7880 	if (unlikely(vcpu->arch.switch_db_regs)) {
7881 		set_debugreg(0, 7);
7882 		set_debugreg(vcpu->arch.eff_db[0], 0);
7883 		set_debugreg(vcpu->arch.eff_db[1], 1);
7884 		set_debugreg(vcpu->arch.eff_db[2], 2);
7885 		set_debugreg(vcpu->arch.eff_db[3], 3);
7886 		set_debugreg(vcpu->arch.dr6, 6);
7887 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7888 	}
7889 
7890 	kvm_x86_ops->run(vcpu);
7891 
7892 	/*
7893 	 * Do this here before restoring debug registers on the host.  And
7894 	 * since we do this before handling the vmexit, a DR access vmexit
7895 	 * can (a) read the correct value of the debug registers, (b) set
7896 	 * KVM_DEBUGREG_WONT_EXIT again.
7897 	 */
7898 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7899 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7900 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7901 		kvm_update_dr0123(vcpu);
7902 		kvm_update_dr6(vcpu);
7903 		kvm_update_dr7(vcpu);
7904 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7905 	}
7906 
7907 	/*
7908 	 * If the guest has used debug registers, at least dr7
7909 	 * will be disabled while returning to the host.
7910 	 * If we don't have active breakpoints in the host, we don't
7911 	 * care about the messed up debug address registers. But if
7912 	 * we have some of them active, restore the old state.
7913 	 */
7914 	if (hw_breakpoint_active())
7915 		hw_breakpoint_restore();
7916 
7917 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7918 
7919 	vcpu->mode = OUTSIDE_GUEST_MODE;
7920 	smp_wmb();
7921 
7922 	kvm_before_interrupt(vcpu);
7923 	kvm_x86_ops->handle_external_intr(vcpu);
7924 	kvm_after_interrupt(vcpu);
7925 
7926 	++vcpu->stat.exits;
7927 
7928 	guest_exit_irqoff();
7929 
7930 	local_irq_enable();
7931 	preempt_enable();
7932 
7933 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7934 
7935 	/*
7936 	 * Profile KVM exit RIPs:
7937 	 */
7938 	if (unlikely(prof_on == KVM_PROFILING)) {
7939 		unsigned long rip = kvm_rip_read(vcpu);
7940 		profile_hit(KVM_PROFILING, (void *)rip);
7941 	}
7942 
7943 	if (unlikely(vcpu->arch.tsc_always_catchup))
7944 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7945 
7946 	if (vcpu->arch.apic_attention)
7947 		kvm_lapic_sync_from_vapic(vcpu);
7948 
7949 	vcpu->arch.gpa_available = false;
7950 	r = kvm_x86_ops->handle_exit(vcpu);
7951 	return r;
7952 
7953 cancel_injection:
7954 	kvm_x86_ops->cancel_injection(vcpu);
7955 	if (unlikely(vcpu->arch.apic_attention))
7956 		kvm_lapic_sync_from_vapic(vcpu);
7957 out:
7958 	return r;
7959 }
7960 
7961 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7962 {
7963 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7964 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7965 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7966 		kvm_vcpu_block(vcpu);
7967 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7968 
7969 		if (kvm_x86_ops->post_block)
7970 			kvm_x86_ops->post_block(vcpu);
7971 
7972 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7973 			return 1;
7974 	}
7975 
7976 	kvm_apic_accept_events(vcpu);
7977 	switch(vcpu->arch.mp_state) {
7978 	case KVM_MP_STATE_HALTED:
7979 		vcpu->arch.pv.pv_unhalted = false;
7980 		vcpu->arch.mp_state =
7981 			KVM_MP_STATE_RUNNABLE;
7982 		/* fall through */
7983 	case KVM_MP_STATE_RUNNABLE:
7984 		vcpu->arch.apf.halted = false;
7985 		break;
7986 	case KVM_MP_STATE_INIT_RECEIVED:
7987 		break;
7988 	default:
7989 		return -EINTR;
7990 		break;
7991 	}
7992 	return 1;
7993 }
7994 
7995 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7996 {
7997 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7998 		kvm_x86_ops->check_nested_events(vcpu, false);
7999 
8000 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8001 		!vcpu->arch.apf.halted);
8002 }
8003 
8004 static int vcpu_run(struct kvm_vcpu *vcpu)
8005 {
8006 	int r;
8007 	struct kvm *kvm = vcpu->kvm;
8008 
8009 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8010 	vcpu->arch.l1tf_flush_l1d = true;
8011 
8012 	for (;;) {
8013 		if (kvm_vcpu_running(vcpu)) {
8014 			r = vcpu_enter_guest(vcpu);
8015 		} else {
8016 			r = vcpu_block(kvm, vcpu);
8017 		}
8018 
8019 		if (r <= 0)
8020 			break;
8021 
8022 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8023 		if (kvm_cpu_has_pending_timer(vcpu))
8024 			kvm_inject_pending_timer_irqs(vcpu);
8025 
8026 		if (dm_request_for_irq_injection(vcpu) &&
8027 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8028 			r = 0;
8029 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8030 			++vcpu->stat.request_irq_exits;
8031 			break;
8032 		}
8033 
8034 		kvm_check_async_pf_completion(vcpu);
8035 
8036 		if (signal_pending(current)) {
8037 			r = -EINTR;
8038 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8039 			++vcpu->stat.signal_exits;
8040 			break;
8041 		}
8042 		if (need_resched()) {
8043 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8044 			cond_resched();
8045 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8046 		}
8047 	}
8048 
8049 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8050 
8051 	return r;
8052 }
8053 
8054 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8055 {
8056 	int r;
8057 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8058 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8059 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8060 	if (r != EMULATE_DONE)
8061 		return 0;
8062 	return 1;
8063 }
8064 
8065 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8066 {
8067 	BUG_ON(!vcpu->arch.pio.count);
8068 
8069 	return complete_emulated_io(vcpu);
8070 }
8071 
8072 /*
8073  * Implements the following, as a state machine:
8074  *
8075  * read:
8076  *   for each fragment
8077  *     for each mmio piece in the fragment
8078  *       write gpa, len
8079  *       exit
8080  *       copy data
8081  *   execute insn
8082  *
8083  * write:
8084  *   for each fragment
8085  *     for each mmio piece in the fragment
8086  *       write gpa, len
8087  *       copy data
8088  *       exit
8089  */
8090 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8091 {
8092 	struct kvm_run *run = vcpu->run;
8093 	struct kvm_mmio_fragment *frag;
8094 	unsigned len;
8095 
8096 	BUG_ON(!vcpu->mmio_needed);
8097 
8098 	/* Complete previous fragment */
8099 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8100 	len = min(8u, frag->len);
8101 	if (!vcpu->mmio_is_write)
8102 		memcpy(frag->data, run->mmio.data, len);
8103 
8104 	if (frag->len <= 8) {
8105 		/* Switch to the next fragment. */
8106 		frag++;
8107 		vcpu->mmio_cur_fragment++;
8108 	} else {
8109 		/* Go forward to the next mmio piece. */
8110 		frag->data += len;
8111 		frag->gpa += len;
8112 		frag->len -= len;
8113 	}
8114 
8115 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8116 		vcpu->mmio_needed = 0;
8117 
8118 		/* FIXME: return into emulator if single-stepping.  */
8119 		if (vcpu->mmio_is_write)
8120 			return 1;
8121 		vcpu->mmio_read_completed = 1;
8122 		return complete_emulated_io(vcpu);
8123 	}
8124 
8125 	run->exit_reason = KVM_EXIT_MMIO;
8126 	run->mmio.phys_addr = frag->gpa;
8127 	if (vcpu->mmio_is_write)
8128 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8129 	run->mmio.len = min(8u, frag->len);
8130 	run->mmio.is_write = vcpu->mmio_is_write;
8131 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8132 	return 0;
8133 }
8134 
8135 /* Swap (qemu) user FPU context for the guest FPU context. */
8136 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8137 {
8138 	preempt_disable();
8139 	copy_fpregs_to_fpstate(&current->thread.fpu);
8140 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8141 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8142 				~XFEATURE_MASK_PKRU);
8143 	preempt_enable();
8144 	trace_kvm_fpu(1);
8145 }
8146 
8147 /* When vcpu_run ends, restore user space FPU context. */
8148 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8149 {
8150 	preempt_disable();
8151 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8152 	copy_kernel_to_fpregs(&current->thread.fpu.state);
8153 	preempt_enable();
8154 	++vcpu->stat.fpu_reload;
8155 	trace_kvm_fpu(0);
8156 }
8157 
8158 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8159 {
8160 	int r;
8161 
8162 	vcpu_load(vcpu);
8163 	kvm_sigset_activate(vcpu);
8164 	kvm_load_guest_fpu(vcpu);
8165 
8166 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8167 		if (kvm_run->immediate_exit) {
8168 			r = -EINTR;
8169 			goto out;
8170 		}
8171 		kvm_vcpu_block(vcpu);
8172 		kvm_apic_accept_events(vcpu);
8173 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8174 		r = -EAGAIN;
8175 		if (signal_pending(current)) {
8176 			r = -EINTR;
8177 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8178 			++vcpu->stat.signal_exits;
8179 		}
8180 		goto out;
8181 	}
8182 
8183 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8184 		r = -EINVAL;
8185 		goto out;
8186 	}
8187 
8188 	if (vcpu->run->kvm_dirty_regs) {
8189 		r = sync_regs(vcpu);
8190 		if (r != 0)
8191 			goto out;
8192 	}
8193 
8194 	/* re-sync apic's tpr */
8195 	if (!lapic_in_kernel(vcpu)) {
8196 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8197 			r = -EINVAL;
8198 			goto out;
8199 		}
8200 	}
8201 
8202 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8203 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8204 		vcpu->arch.complete_userspace_io = NULL;
8205 		r = cui(vcpu);
8206 		if (r <= 0)
8207 			goto out;
8208 	} else
8209 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8210 
8211 	if (kvm_run->immediate_exit)
8212 		r = -EINTR;
8213 	else
8214 		r = vcpu_run(vcpu);
8215 
8216 out:
8217 	kvm_put_guest_fpu(vcpu);
8218 	if (vcpu->run->kvm_valid_regs)
8219 		store_regs(vcpu);
8220 	post_kvm_run_save(vcpu);
8221 	kvm_sigset_deactivate(vcpu);
8222 
8223 	vcpu_put(vcpu);
8224 	return r;
8225 }
8226 
8227 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8228 {
8229 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8230 		/*
8231 		 * We are here if userspace calls get_regs() in the middle of
8232 		 * instruction emulation. Registers state needs to be copied
8233 		 * back from emulation context to vcpu. Userspace shouldn't do
8234 		 * that usually, but some bad designed PV devices (vmware
8235 		 * backdoor interface) need this to work
8236 		 */
8237 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8238 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8239 	}
8240 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8241 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8242 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8243 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8244 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8245 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8246 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8247 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8248 #ifdef CONFIG_X86_64
8249 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8250 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8251 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8252 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8253 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8254 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8255 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8256 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8257 #endif
8258 
8259 	regs->rip = kvm_rip_read(vcpu);
8260 	regs->rflags = kvm_get_rflags(vcpu);
8261 }
8262 
8263 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8264 {
8265 	vcpu_load(vcpu);
8266 	__get_regs(vcpu, regs);
8267 	vcpu_put(vcpu);
8268 	return 0;
8269 }
8270 
8271 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8272 {
8273 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8274 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8275 
8276 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8277 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8278 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8279 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8280 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8281 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8282 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8283 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8284 #ifdef CONFIG_X86_64
8285 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8286 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8287 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8288 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8289 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8290 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8291 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8292 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8293 #endif
8294 
8295 	kvm_rip_write(vcpu, regs->rip);
8296 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8297 
8298 	vcpu->arch.exception.pending = false;
8299 
8300 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8301 }
8302 
8303 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8304 {
8305 	vcpu_load(vcpu);
8306 	__set_regs(vcpu, regs);
8307 	vcpu_put(vcpu);
8308 	return 0;
8309 }
8310 
8311 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8312 {
8313 	struct kvm_segment cs;
8314 
8315 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8316 	*db = cs.db;
8317 	*l = cs.l;
8318 }
8319 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8320 
8321 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8322 {
8323 	struct desc_ptr dt;
8324 
8325 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8326 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8327 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8328 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8329 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8330 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8331 
8332 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8333 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8334 
8335 	kvm_x86_ops->get_idt(vcpu, &dt);
8336 	sregs->idt.limit = dt.size;
8337 	sregs->idt.base = dt.address;
8338 	kvm_x86_ops->get_gdt(vcpu, &dt);
8339 	sregs->gdt.limit = dt.size;
8340 	sregs->gdt.base = dt.address;
8341 
8342 	sregs->cr0 = kvm_read_cr0(vcpu);
8343 	sregs->cr2 = vcpu->arch.cr2;
8344 	sregs->cr3 = kvm_read_cr3(vcpu);
8345 	sregs->cr4 = kvm_read_cr4(vcpu);
8346 	sregs->cr8 = kvm_get_cr8(vcpu);
8347 	sregs->efer = vcpu->arch.efer;
8348 	sregs->apic_base = kvm_get_apic_base(vcpu);
8349 
8350 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8351 
8352 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8353 		set_bit(vcpu->arch.interrupt.nr,
8354 			(unsigned long *)sregs->interrupt_bitmap);
8355 }
8356 
8357 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8358 				  struct kvm_sregs *sregs)
8359 {
8360 	vcpu_load(vcpu);
8361 	__get_sregs(vcpu, sregs);
8362 	vcpu_put(vcpu);
8363 	return 0;
8364 }
8365 
8366 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8367 				    struct kvm_mp_state *mp_state)
8368 {
8369 	vcpu_load(vcpu);
8370 
8371 	kvm_apic_accept_events(vcpu);
8372 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8373 					vcpu->arch.pv.pv_unhalted)
8374 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8375 	else
8376 		mp_state->mp_state = vcpu->arch.mp_state;
8377 
8378 	vcpu_put(vcpu);
8379 	return 0;
8380 }
8381 
8382 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8383 				    struct kvm_mp_state *mp_state)
8384 {
8385 	int ret = -EINVAL;
8386 
8387 	vcpu_load(vcpu);
8388 
8389 	if (!lapic_in_kernel(vcpu) &&
8390 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8391 		goto out;
8392 
8393 	/* INITs are latched while in SMM */
8394 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8395 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8396 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8397 		goto out;
8398 
8399 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8400 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8401 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8402 	} else
8403 		vcpu->arch.mp_state = mp_state->mp_state;
8404 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8405 
8406 	ret = 0;
8407 out:
8408 	vcpu_put(vcpu);
8409 	return ret;
8410 }
8411 
8412 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8413 		    int reason, bool has_error_code, u32 error_code)
8414 {
8415 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8416 	int ret;
8417 
8418 	init_emulate_ctxt(vcpu);
8419 
8420 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8421 				   has_error_code, error_code);
8422 
8423 	if (ret)
8424 		return EMULATE_FAIL;
8425 
8426 	kvm_rip_write(vcpu, ctxt->eip);
8427 	kvm_set_rflags(vcpu, ctxt->eflags);
8428 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8429 	return EMULATE_DONE;
8430 }
8431 EXPORT_SYMBOL_GPL(kvm_task_switch);
8432 
8433 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8434 {
8435 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8436 			(sregs->cr4 & X86_CR4_OSXSAVE))
8437 		return  -EINVAL;
8438 
8439 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8440 		/*
8441 		 * When EFER.LME and CR0.PG are set, the processor is in
8442 		 * 64-bit mode (though maybe in a 32-bit code segment).
8443 		 * CR4.PAE and EFER.LMA must be set.
8444 		 */
8445 		if (!(sregs->cr4 & X86_CR4_PAE)
8446 		    || !(sregs->efer & EFER_LMA))
8447 			return -EINVAL;
8448 	} else {
8449 		/*
8450 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8451 		 * segment cannot be 64-bit.
8452 		 */
8453 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8454 			return -EINVAL;
8455 	}
8456 
8457 	return 0;
8458 }
8459 
8460 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8461 {
8462 	struct msr_data apic_base_msr;
8463 	int mmu_reset_needed = 0;
8464 	int cpuid_update_needed = 0;
8465 	int pending_vec, max_bits, idx;
8466 	struct desc_ptr dt;
8467 	int ret = -EINVAL;
8468 
8469 	if (kvm_valid_sregs(vcpu, sregs))
8470 		goto out;
8471 
8472 	apic_base_msr.data = sregs->apic_base;
8473 	apic_base_msr.host_initiated = true;
8474 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8475 		goto out;
8476 
8477 	dt.size = sregs->idt.limit;
8478 	dt.address = sregs->idt.base;
8479 	kvm_x86_ops->set_idt(vcpu, &dt);
8480 	dt.size = sregs->gdt.limit;
8481 	dt.address = sregs->gdt.base;
8482 	kvm_x86_ops->set_gdt(vcpu, &dt);
8483 
8484 	vcpu->arch.cr2 = sregs->cr2;
8485 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8486 	vcpu->arch.cr3 = sregs->cr3;
8487 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8488 
8489 	kvm_set_cr8(vcpu, sregs->cr8);
8490 
8491 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8492 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8493 
8494 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8495 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8496 	vcpu->arch.cr0 = sregs->cr0;
8497 
8498 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8499 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8500 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8501 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8502 	if (cpuid_update_needed)
8503 		kvm_update_cpuid(vcpu);
8504 
8505 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8506 	if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8507 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8508 		mmu_reset_needed = 1;
8509 	}
8510 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8511 
8512 	if (mmu_reset_needed)
8513 		kvm_mmu_reset_context(vcpu);
8514 
8515 	max_bits = KVM_NR_INTERRUPTS;
8516 	pending_vec = find_first_bit(
8517 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8518 	if (pending_vec < max_bits) {
8519 		kvm_queue_interrupt(vcpu, pending_vec, false);
8520 		pr_debug("Set back pending irq %d\n", pending_vec);
8521 	}
8522 
8523 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8524 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8525 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8526 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8527 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8528 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8529 
8530 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8531 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8532 
8533 	update_cr8_intercept(vcpu);
8534 
8535 	/* Older userspace won't unhalt the vcpu on reset. */
8536 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8537 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8538 	    !is_protmode(vcpu))
8539 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8540 
8541 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8542 
8543 	ret = 0;
8544 out:
8545 	return ret;
8546 }
8547 
8548 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8549 				  struct kvm_sregs *sregs)
8550 {
8551 	int ret;
8552 
8553 	vcpu_load(vcpu);
8554 	ret = __set_sregs(vcpu, sregs);
8555 	vcpu_put(vcpu);
8556 	return ret;
8557 }
8558 
8559 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8560 					struct kvm_guest_debug *dbg)
8561 {
8562 	unsigned long rflags;
8563 	int i, r;
8564 
8565 	vcpu_load(vcpu);
8566 
8567 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8568 		r = -EBUSY;
8569 		if (vcpu->arch.exception.pending)
8570 			goto out;
8571 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8572 			kvm_queue_exception(vcpu, DB_VECTOR);
8573 		else
8574 			kvm_queue_exception(vcpu, BP_VECTOR);
8575 	}
8576 
8577 	/*
8578 	 * Read rflags as long as potentially injected trace flags are still
8579 	 * filtered out.
8580 	 */
8581 	rflags = kvm_get_rflags(vcpu);
8582 
8583 	vcpu->guest_debug = dbg->control;
8584 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8585 		vcpu->guest_debug = 0;
8586 
8587 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8588 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8589 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8590 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8591 	} else {
8592 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8593 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8594 	}
8595 	kvm_update_dr7(vcpu);
8596 
8597 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8598 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8599 			get_segment_base(vcpu, VCPU_SREG_CS);
8600 
8601 	/*
8602 	 * Trigger an rflags update that will inject or remove the trace
8603 	 * flags.
8604 	 */
8605 	kvm_set_rflags(vcpu, rflags);
8606 
8607 	kvm_x86_ops->update_bp_intercept(vcpu);
8608 
8609 	r = 0;
8610 
8611 out:
8612 	vcpu_put(vcpu);
8613 	return r;
8614 }
8615 
8616 /*
8617  * Translate a guest virtual address to a guest physical address.
8618  */
8619 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8620 				    struct kvm_translation *tr)
8621 {
8622 	unsigned long vaddr = tr->linear_address;
8623 	gpa_t gpa;
8624 	int idx;
8625 
8626 	vcpu_load(vcpu);
8627 
8628 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8629 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8630 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8631 	tr->physical_address = gpa;
8632 	tr->valid = gpa != UNMAPPED_GVA;
8633 	tr->writeable = 1;
8634 	tr->usermode = 0;
8635 
8636 	vcpu_put(vcpu);
8637 	return 0;
8638 }
8639 
8640 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8641 {
8642 	struct fxregs_state *fxsave;
8643 
8644 	vcpu_load(vcpu);
8645 
8646 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8647 	memcpy(fpu->fpr, fxsave->st_space, 128);
8648 	fpu->fcw = fxsave->cwd;
8649 	fpu->fsw = fxsave->swd;
8650 	fpu->ftwx = fxsave->twd;
8651 	fpu->last_opcode = fxsave->fop;
8652 	fpu->last_ip = fxsave->rip;
8653 	fpu->last_dp = fxsave->rdp;
8654 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8655 
8656 	vcpu_put(vcpu);
8657 	return 0;
8658 }
8659 
8660 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8661 {
8662 	struct fxregs_state *fxsave;
8663 
8664 	vcpu_load(vcpu);
8665 
8666 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8667 
8668 	memcpy(fxsave->st_space, fpu->fpr, 128);
8669 	fxsave->cwd = fpu->fcw;
8670 	fxsave->swd = fpu->fsw;
8671 	fxsave->twd = fpu->ftwx;
8672 	fxsave->fop = fpu->last_opcode;
8673 	fxsave->rip = fpu->last_ip;
8674 	fxsave->rdp = fpu->last_dp;
8675 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8676 
8677 	vcpu_put(vcpu);
8678 	return 0;
8679 }
8680 
8681 static void store_regs(struct kvm_vcpu *vcpu)
8682 {
8683 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8684 
8685 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8686 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8687 
8688 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8689 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8690 
8691 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8692 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8693 				vcpu, &vcpu->run->s.regs.events);
8694 }
8695 
8696 static int sync_regs(struct kvm_vcpu *vcpu)
8697 {
8698 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8699 		return -EINVAL;
8700 
8701 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8702 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8703 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8704 	}
8705 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8706 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8707 			return -EINVAL;
8708 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8709 	}
8710 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8711 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8712 				vcpu, &vcpu->run->s.regs.events))
8713 			return -EINVAL;
8714 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8715 	}
8716 
8717 	return 0;
8718 }
8719 
8720 static void fx_init(struct kvm_vcpu *vcpu)
8721 {
8722 	fpstate_init(&vcpu->arch.guest_fpu->state);
8723 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8724 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8725 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8726 
8727 	/*
8728 	 * Ensure guest xcr0 is valid for loading
8729 	 */
8730 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8731 
8732 	vcpu->arch.cr0 |= X86_CR0_ET;
8733 }
8734 
8735 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8736 {
8737 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8738 
8739 	kvmclock_reset(vcpu);
8740 
8741 	kvm_x86_ops->vcpu_free(vcpu);
8742 	free_cpumask_var(wbinvd_dirty_mask);
8743 }
8744 
8745 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8746 						unsigned int id)
8747 {
8748 	struct kvm_vcpu *vcpu;
8749 
8750 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8751 		printk_once(KERN_WARNING
8752 		"kvm: SMP vm created on host with unstable TSC; "
8753 		"guest TSC will not be reliable\n");
8754 
8755 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8756 
8757 	return vcpu;
8758 }
8759 
8760 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8761 {
8762 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
8763 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8764 	kvm_vcpu_mtrr_init(vcpu);
8765 	vcpu_load(vcpu);
8766 	kvm_vcpu_reset(vcpu, false);
8767 	kvm_init_mmu(vcpu, false);
8768 	vcpu_put(vcpu);
8769 	return 0;
8770 }
8771 
8772 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8773 {
8774 	struct msr_data msr;
8775 	struct kvm *kvm = vcpu->kvm;
8776 
8777 	kvm_hv_vcpu_postcreate(vcpu);
8778 
8779 	if (mutex_lock_killable(&vcpu->mutex))
8780 		return;
8781 	vcpu_load(vcpu);
8782 	msr.data = 0x0;
8783 	msr.index = MSR_IA32_TSC;
8784 	msr.host_initiated = true;
8785 	kvm_write_tsc(vcpu, &msr);
8786 	vcpu_put(vcpu);
8787 	mutex_unlock(&vcpu->mutex);
8788 
8789 	if (!kvmclock_periodic_sync)
8790 		return;
8791 
8792 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8793 					KVMCLOCK_SYNC_PERIOD);
8794 }
8795 
8796 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8797 {
8798 	vcpu->arch.apf.msr_val = 0;
8799 
8800 	vcpu_load(vcpu);
8801 	kvm_mmu_unload(vcpu);
8802 	vcpu_put(vcpu);
8803 
8804 	kvm_x86_ops->vcpu_free(vcpu);
8805 }
8806 
8807 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8808 {
8809 	kvm_lapic_reset(vcpu, init_event);
8810 
8811 	vcpu->arch.hflags = 0;
8812 
8813 	vcpu->arch.smi_pending = 0;
8814 	vcpu->arch.smi_count = 0;
8815 	atomic_set(&vcpu->arch.nmi_queued, 0);
8816 	vcpu->arch.nmi_pending = 0;
8817 	vcpu->arch.nmi_injected = false;
8818 	kvm_clear_interrupt_queue(vcpu);
8819 	kvm_clear_exception_queue(vcpu);
8820 	vcpu->arch.exception.pending = false;
8821 
8822 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8823 	kvm_update_dr0123(vcpu);
8824 	vcpu->arch.dr6 = DR6_INIT;
8825 	kvm_update_dr6(vcpu);
8826 	vcpu->arch.dr7 = DR7_FIXED_1;
8827 	kvm_update_dr7(vcpu);
8828 
8829 	vcpu->arch.cr2 = 0;
8830 
8831 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8832 	vcpu->arch.apf.msr_val = 0;
8833 	vcpu->arch.st.msr_val = 0;
8834 
8835 	kvmclock_reset(vcpu);
8836 
8837 	kvm_clear_async_pf_completion_queue(vcpu);
8838 	kvm_async_pf_hash_reset(vcpu);
8839 	vcpu->arch.apf.halted = false;
8840 
8841 	if (kvm_mpx_supported()) {
8842 		void *mpx_state_buffer;
8843 
8844 		/*
8845 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8846 		 * called with loaded FPU and does not let userspace fix the state.
8847 		 */
8848 		if (init_event)
8849 			kvm_put_guest_fpu(vcpu);
8850 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8851 					XFEATURE_MASK_BNDREGS);
8852 		if (mpx_state_buffer)
8853 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8854 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8855 					XFEATURE_MASK_BNDCSR);
8856 		if (mpx_state_buffer)
8857 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8858 		if (init_event)
8859 			kvm_load_guest_fpu(vcpu);
8860 	}
8861 
8862 	if (!init_event) {
8863 		kvm_pmu_reset(vcpu);
8864 		vcpu->arch.smbase = 0x30000;
8865 
8866 		vcpu->arch.msr_misc_features_enables = 0;
8867 
8868 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8869 	}
8870 
8871 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8872 	vcpu->arch.regs_avail = ~0;
8873 	vcpu->arch.regs_dirty = ~0;
8874 
8875 	vcpu->arch.ia32_xss = 0;
8876 
8877 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8878 }
8879 
8880 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8881 {
8882 	struct kvm_segment cs;
8883 
8884 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8885 	cs.selector = vector << 8;
8886 	cs.base = vector << 12;
8887 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8888 	kvm_rip_write(vcpu, 0);
8889 }
8890 
8891 int kvm_arch_hardware_enable(void)
8892 {
8893 	struct kvm *kvm;
8894 	struct kvm_vcpu *vcpu;
8895 	int i;
8896 	int ret;
8897 	u64 local_tsc;
8898 	u64 max_tsc = 0;
8899 	bool stable, backwards_tsc = false;
8900 
8901 	kvm_shared_msr_cpu_online();
8902 	ret = kvm_x86_ops->hardware_enable();
8903 	if (ret != 0)
8904 		return ret;
8905 
8906 	local_tsc = rdtsc();
8907 	stable = !kvm_check_tsc_unstable();
8908 	list_for_each_entry(kvm, &vm_list, vm_list) {
8909 		kvm_for_each_vcpu(i, vcpu, kvm) {
8910 			if (!stable && vcpu->cpu == smp_processor_id())
8911 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8912 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8913 				backwards_tsc = true;
8914 				if (vcpu->arch.last_host_tsc > max_tsc)
8915 					max_tsc = vcpu->arch.last_host_tsc;
8916 			}
8917 		}
8918 	}
8919 
8920 	/*
8921 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8922 	 * platforms that reset TSC during suspend or hibernate actions, but
8923 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8924 	 * detect that condition here, which happens early in CPU bringup,
8925 	 * before any KVM threads can be running.  Unfortunately, we can't
8926 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8927 	 * enough into CPU bringup that we know how much real time has actually
8928 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8929 	 * variables that haven't been updated yet.
8930 	 *
8931 	 * So we simply find the maximum observed TSC above, then record the
8932 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8933 	 * the adjustment will be applied.  Note that we accumulate
8934 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8935 	 * gets a chance to run again.  In the event that no KVM threads get a
8936 	 * chance to run, we will miss the entire elapsed period, as we'll have
8937 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8938 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8939 	 * uniform across all VCPUs (not to mention the scenario is extremely
8940 	 * unlikely). It is possible that a second hibernate recovery happens
8941 	 * much faster than a first, causing the observed TSC here to be
8942 	 * smaller; this would require additional padding adjustment, which is
8943 	 * why we set last_host_tsc to the local tsc observed here.
8944 	 *
8945 	 * N.B. - this code below runs only on platforms with reliable TSC,
8946 	 * as that is the only way backwards_tsc is set above.  Also note
8947 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8948 	 * have the same delta_cyc adjustment applied if backwards_tsc
8949 	 * is detected.  Note further, this adjustment is only done once,
8950 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8951 	 * called multiple times (one for each physical CPU bringup).
8952 	 *
8953 	 * Platforms with unreliable TSCs don't have to deal with this, they
8954 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8955 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8956 	 * guarantee that they stay in perfect synchronization.
8957 	 */
8958 	if (backwards_tsc) {
8959 		u64 delta_cyc = max_tsc - local_tsc;
8960 		list_for_each_entry(kvm, &vm_list, vm_list) {
8961 			kvm->arch.backwards_tsc_observed = true;
8962 			kvm_for_each_vcpu(i, vcpu, kvm) {
8963 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8964 				vcpu->arch.last_host_tsc = local_tsc;
8965 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8966 			}
8967 
8968 			/*
8969 			 * We have to disable TSC offset matching.. if you were
8970 			 * booting a VM while issuing an S4 host suspend....
8971 			 * you may have some problem.  Solving this issue is
8972 			 * left as an exercise to the reader.
8973 			 */
8974 			kvm->arch.last_tsc_nsec = 0;
8975 			kvm->arch.last_tsc_write = 0;
8976 		}
8977 
8978 	}
8979 	return 0;
8980 }
8981 
8982 void kvm_arch_hardware_disable(void)
8983 {
8984 	kvm_x86_ops->hardware_disable();
8985 	drop_user_return_notifiers();
8986 }
8987 
8988 int kvm_arch_hardware_setup(void)
8989 {
8990 	int r;
8991 
8992 	r = kvm_x86_ops->hardware_setup();
8993 	if (r != 0)
8994 		return r;
8995 
8996 	if (kvm_has_tsc_control) {
8997 		/*
8998 		 * Make sure the user can only configure tsc_khz values that
8999 		 * fit into a signed integer.
9000 		 * A min value is not calculated because it will always
9001 		 * be 1 on all machines.
9002 		 */
9003 		u64 max = min(0x7fffffffULL,
9004 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9005 		kvm_max_guest_tsc_khz = max;
9006 
9007 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9008 	}
9009 
9010 	kvm_init_msr_list();
9011 	return 0;
9012 }
9013 
9014 void kvm_arch_hardware_unsetup(void)
9015 {
9016 	kvm_x86_ops->hardware_unsetup();
9017 }
9018 
9019 void kvm_arch_check_processor_compat(void *rtn)
9020 {
9021 	kvm_x86_ops->check_processor_compatibility(rtn);
9022 }
9023 
9024 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9025 {
9026 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9027 }
9028 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9029 
9030 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9031 {
9032 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9033 }
9034 
9035 struct static_key kvm_no_apic_vcpu __read_mostly;
9036 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9037 
9038 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9039 {
9040 	struct page *page;
9041 	int r;
9042 
9043 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9044 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9045 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9046 	else
9047 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9048 
9049 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9050 	if (!page) {
9051 		r = -ENOMEM;
9052 		goto fail;
9053 	}
9054 	vcpu->arch.pio_data = page_address(page);
9055 
9056 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9057 
9058 	r = kvm_mmu_create(vcpu);
9059 	if (r < 0)
9060 		goto fail_free_pio_data;
9061 
9062 	if (irqchip_in_kernel(vcpu->kvm)) {
9063 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9064 		r = kvm_create_lapic(vcpu);
9065 		if (r < 0)
9066 			goto fail_mmu_destroy;
9067 	} else
9068 		static_key_slow_inc(&kvm_no_apic_vcpu);
9069 
9070 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9071 				       GFP_KERNEL_ACCOUNT);
9072 	if (!vcpu->arch.mce_banks) {
9073 		r = -ENOMEM;
9074 		goto fail_free_lapic;
9075 	}
9076 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9077 
9078 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9079 				GFP_KERNEL_ACCOUNT)) {
9080 		r = -ENOMEM;
9081 		goto fail_free_mce_banks;
9082 	}
9083 
9084 	fx_init(vcpu);
9085 
9086 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9087 
9088 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9089 
9090 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9091 
9092 	kvm_async_pf_hash_reset(vcpu);
9093 	kvm_pmu_init(vcpu);
9094 
9095 	vcpu->arch.pending_external_vector = -1;
9096 	vcpu->arch.preempted_in_kernel = false;
9097 
9098 	kvm_hv_vcpu_init(vcpu);
9099 
9100 	return 0;
9101 
9102 fail_free_mce_banks:
9103 	kfree(vcpu->arch.mce_banks);
9104 fail_free_lapic:
9105 	kvm_free_lapic(vcpu);
9106 fail_mmu_destroy:
9107 	kvm_mmu_destroy(vcpu);
9108 fail_free_pio_data:
9109 	free_page((unsigned long)vcpu->arch.pio_data);
9110 fail:
9111 	return r;
9112 }
9113 
9114 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9115 {
9116 	int idx;
9117 
9118 	kvm_hv_vcpu_uninit(vcpu);
9119 	kvm_pmu_destroy(vcpu);
9120 	kfree(vcpu->arch.mce_banks);
9121 	kvm_free_lapic(vcpu);
9122 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9123 	kvm_mmu_destroy(vcpu);
9124 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9125 	free_page((unsigned long)vcpu->arch.pio_data);
9126 	if (!lapic_in_kernel(vcpu))
9127 		static_key_slow_dec(&kvm_no_apic_vcpu);
9128 }
9129 
9130 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9131 {
9132 	vcpu->arch.l1tf_flush_l1d = true;
9133 	kvm_x86_ops->sched_in(vcpu, cpu);
9134 }
9135 
9136 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9137 {
9138 	if (type)
9139 		return -EINVAL;
9140 
9141 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9142 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9143 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9144 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9145 
9146 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9147 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9148 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9149 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9150 		&kvm->arch.irq_sources_bitmap);
9151 
9152 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9153 	mutex_init(&kvm->arch.apic_map_lock);
9154 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9155 
9156 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9157 	pvclock_update_vm_gtod_copy(kvm);
9158 
9159 	kvm->arch.guest_can_read_msr_platform_info = true;
9160 
9161 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9162 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9163 
9164 	kvm_hv_init_vm(kvm);
9165 	kvm_page_track_init(kvm);
9166 	kvm_mmu_init_vm(kvm);
9167 
9168 	if (kvm_x86_ops->vm_init)
9169 		return kvm_x86_ops->vm_init(kvm);
9170 
9171 	return 0;
9172 }
9173 
9174 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9175 {
9176 	vcpu_load(vcpu);
9177 	kvm_mmu_unload(vcpu);
9178 	vcpu_put(vcpu);
9179 }
9180 
9181 static void kvm_free_vcpus(struct kvm *kvm)
9182 {
9183 	unsigned int i;
9184 	struct kvm_vcpu *vcpu;
9185 
9186 	/*
9187 	 * Unpin any mmu pages first.
9188 	 */
9189 	kvm_for_each_vcpu(i, vcpu, kvm) {
9190 		kvm_clear_async_pf_completion_queue(vcpu);
9191 		kvm_unload_vcpu_mmu(vcpu);
9192 	}
9193 	kvm_for_each_vcpu(i, vcpu, kvm)
9194 		kvm_arch_vcpu_free(vcpu);
9195 
9196 	mutex_lock(&kvm->lock);
9197 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9198 		kvm->vcpus[i] = NULL;
9199 
9200 	atomic_set(&kvm->online_vcpus, 0);
9201 	mutex_unlock(&kvm->lock);
9202 }
9203 
9204 void kvm_arch_sync_events(struct kvm *kvm)
9205 {
9206 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9207 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9208 	kvm_free_pit(kvm);
9209 }
9210 
9211 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9212 {
9213 	int i, r;
9214 	unsigned long hva;
9215 	struct kvm_memslots *slots = kvm_memslots(kvm);
9216 	struct kvm_memory_slot *slot, old;
9217 
9218 	/* Called with kvm->slots_lock held.  */
9219 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9220 		return -EINVAL;
9221 
9222 	slot = id_to_memslot(slots, id);
9223 	if (size) {
9224 		if (slot->npages)
9225 			return -EEXIST;
9226 
9227 		/*
9228 		 * MAP_SHARED to prevent internal slot pages from being moved
9229 		 * by fork()/COW.
9230 		 */
9231 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9232 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9233 		if (IS_ERR((void *)hva))
9234 			return PTR_ERR((void *)hva);
9235 	} else {
9236 		if (!slot->npages)
9237 			return 0;
9238 
9239 		hva = 0;
9240 	}
9241 
9242 	old = *slot;
9243 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9244 		struct kvm_userspace_memory_region m;
9245 
9246 		m.slot = id | (i << 16);
9247 		m.flags = 0;
9248 		m.guest_phys_addr = gpa;
9249 		m.userspace_addr = hva;
9250 		m.memory_size = size;
9251 		r = __kvm_set_memory_region(kvm, &m);
9252 		if (r < 0)
9253 			return r;
9254 	}
9255 
9256 	if (!size)
9257 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9258 
9259 	return 0;
9260 }
9261 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9262 
9263 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9264 {
9265 	int r;
9266 
9267 	mutex_lock(&kvm->slots_lock);
9268 	r = __x86_set_memory_region(kvm, id, gpa, size);
9269 	mutex_unlock(&kvm->slots_lock);
9270 
9271 	return r;
9272 }
9273 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9274 
9275 void kvm_arch_destroy_vm(struct kvm *kvm)
9276 {
9277 	if (current->mm == kvm->mm) {
9278 		/*
9279 		 * Free memory regions allocated on behalf of userspace,
9280 		 * unless the the memory map has changed due to process exit
9281 		 * or fd copying.
9282 		 */
9283 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9284 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9285 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9286 	}
9287 	if (kvm_x86_ops->vm_destroy)
9288 		kvm_x86_ops->vm_destroy(kvm);
9289 	kvm_pic_destroy(kvm);
9290 	kvm_ioapic_destroy(kvm);
9291 	kvm_free_vcpus(kvm);
9292 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9293 	kvm_mmu_uninit_vm(kvm);
9294 	kvm_page_track_cleanup(kvm);
9295 	kvm_hv_destroy_vm(kvm);
9296 }
9297 
9298 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9299 			   struct kvm_memory_slot *dont)
9300 {
9301 	int i;
9302 
9303 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9304 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9305 			kvfree(free->arch.rmap[i]);
9306 			free->arch.rmap[i] = NULL;
9307 		}
9308 		if (i == 0)
9309 			continue;
9310 
9311 		if (!dont || free->arch.lpage_info[i - 1] !=
9312 			     dont->arch.lpage_info[i - 1]) {
9313 			kvfree(free->arch.lpage_info[i - 1]);
9314 			free->arch.lpage_info[i - 1] = NULL;
9315 		}
9316 	}
9317 
9318 	kvm_page_track_free_memslot(free, dont);
9319 }
9320 
9321 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9322 			    unsigned long npages)
9323 {
9324 	int i;
9325 
9326 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9327 		struct kvm_lpage_info *linfo;
9328 		unsigned long ugfn;
9329 		int lpages;
9330 		int level = i + 1;
9331 
9332 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9333 				      slot->base_gfn, level) + 1;
9334 
9335 		slot->arch.rmap[i] =
9336 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9337 				 GFP_KERNEL_ACCOUNT);
9338 		if (!slot->arch.rmap[i])
9339 			goto out_free;
9340 		if (i == 0)
9341 			continue;
9342 
9343 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9344 		if (!linfo)
9345 			goto out_free;
9346 
9347 		slot->arch.lpage_info[i - 1] = linfo;
9348 
9349 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9350 			linfo[0].disallow_lpage = 1;
9351 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9352 			linfo[lpages - 1].disallow_lpage = 1;
9353 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9354 		/*
9355 		 * If the gfn and userspace address are not aligned wrt each
9356 		 * other, or if explicitly asked to, disable large page
9357 		 * support for this slot
9358 		 */
9359 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9360 		    !kvm_largepages_enabled()) {
9361 			unsigned long j;
9362 
9363 			for (j = 0; j < lpages; ++j)
9364 				linfo[j].disallow_lpage = 1;
9365 		}
9366 	}
9367 
9368 	if (kvm_page_track_create_memslot(slot, npages))
9369 		goto out_free;
9370 
9371 	return 0;
9372 
9373 out_free:
9374 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9375 		kvfree(slot->arch.rmap[i]);
9376 		slot->arch.rmap[i] = NULL;
9377 		if (i == 0)
9378 			continue;
9379 
9380 		kvfree(slot->arch.lpage_info[i - 1]);
9381 		slot->arch.lpage_info[i - 1] = NULL;
9382 	}
9383 	return -ENOMEM;
9384 }
9385 
9386 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9387 {
9388 	/*
9389 	 * memslots->generation has been incremented.
9390 	 * mmio generation may have reached its maximum value.
9391 	 */
9392 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9393 }
9394 
9395 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9396 				struct kvm_memory_slot *memslot,
9397 				const struct kvm_userspace_memory_region *mem,
9398 				enum kvm_mr_change change)
9399 {
9400 	return 0;
9401 }
9402 
9403 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9404 				     struct kvm_memory_slot *new)
9405 {
9406 	/* Still write protect RO slot */
9407 	if (new->flags & KVM_MEM_READONLY) {
9408 		kvm_mmu_slot_remove_write_access(kvm, new);
9409 		return;
9410 	}
9411 
9412 	/*
9413 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9414 	 *
9415 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9416 	 *
9417 	 *  - KVM_MR_CREATE with dirty logging is disabled
9418 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9419 	 *
9420 	 * The reason is, in case of PML, we need to set D-bit for any slots
9421 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9422 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9423 	 * guarantees leaving PML enabled during guest's lifetime won't have
9424 	 * any additional overhead from PML when guest is running with dirty
9425 	 * logging disabled for memory slots.
9426 	 *
9427 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9428 	 * to dirty logging mode.
9429 	 *
9430 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9431 	 *
9432 	 * In case of write protect:
9433 	 *
9434 	 * Write protect all pages for dirty logging.
9435 	 *
9436 	 * All the sptes including the large sptes which point to this
9437 	 * slot are set to readonly. We can not create any new large
9438 	 * spte on this slot until the end of the logging.
9439 	 *
9440 	 * See the comments in fast_page_fault().
9441 	 */
9442 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9443 		if (kvm_x86_ops->slot_enable_log_dirty)
9444 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9445 		else
9446 			kvm_mmu_slot_remove_write_access(kvm, new);
9447 	} else {
9448 		if (kvm_x86_ops->slot_disable_log_dirty)
9449 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9450 	}
9451 }
9452 
9453 void kvm_arch_commit_memory_region(struct kvm *kvm,
9454 				const struct kvm_userspace_memory_region *mem,
9455 				const struct kvm_memory_slot *old,
9456 				const struct kvm_memory_slot *new,
9457 				enum kvm_mr_change change)
9458 {
9459 	if (!kvm->arch.n_requested_mmu_pages)
9460 		kvm_mmu_change_mmu_pages(kvm,
9461 				kvm_mmu_calculate_default_mmu_pages(kvm));
9462 
9463 	/*
9464 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9465 	 * sptes have to be split.  If live migration is successful, the guest
9466 	 * in the source machine will be destroyed and large sptes will be
9467 	 * created in the destination. However, if the guest continues to run
9468 	 * in the source machine (for example if live migration fails), small
9469 	 * sptes will remain around and cause bad performance.
9470 	 *
9471 	 * Scan sptes if dirty logging has been stopped, dropping those
9472 	 * which can be collapsed into a single large-page spte.  Later
9473 	 * page faults will create the large-page sptes.
9474 	 */
9475 	if ((change != KVM_MR_DELETE) &&
9476 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9477 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9478 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9479 
9480 	/*
9481 	 * Set up write protection and/or dirty logging for the new slot.
9482 	 *
9483 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9484 	 * been zapped so no dirty logging staff is needed for old slot. For
9485 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9486 	 * new and it's also covered when dealing with the new slot.
9487 	 *
9488 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9489 	 */
9490 	if (change != KVM_MR_DELETE)
9491 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9492 }
9493 
9494 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9495 {
9496 	kvm_mmu_zap_all(kvm);
9497 }
9498 
9499 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9500 				   struct kvm_memory_slot *slot)
9501 {
9502 	kvm_page_track_flush_slot(kvm, slot);
9503 }
9504 
9505 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9506 {
9507 	return (is_guest_mode(vcpu) &&
9508 			kvm_x86_ops->guest_apic_has_interrupt &&
9509 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9510 }
9511 
9512 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9513 {
9514 	if (!list_empty_careful(&vcpu->async_pf.done))
9515 		return true;
9516 
9517 	if (kvm_apic_has_events(vcpu))
9518 		return true;
9519 
9520 	if (vcpu->arch.pv.pv_unhalted)
9521 		return true;
9522 
9523 	if (vcpu->arch.exception.pending)
9524 		return true;
9525 
9526 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9527 	    (vcpu->arch.nmi_pending &&
9528 	     kvm_x86_ops->nmi_allowed(vcpu)))
9529 		return true;
9530 
9531 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9532 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9533 		return true;
9534 
9535 	if (kvm_arch_interrupt_allowed(vcpu) &&
9536 	    (kvm_cpu_has_interrupt(vcpu) ||
9537 	    kvm_guest_apic_has_interrupt(vcpu)))
9538 		return true;
9539 
9540 	if (kvm_hv_has_stimer_pending(vcpu))
9541 		return true;
9542 
9543 	return false;
9544 }
9545 
9546 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9547 {
9548 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9549 }
9550 
9551 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9552 {
9553 	return vcpu->arch.preempted_in_kernel;
9554 }
9555 
9556 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9557 {
9558 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9559 }
9560 
9561 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9562 {
9563 	return kvm_x86_ops->interrupt_allowed(vcpu);
9564 }
9565 
9566 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9567 {
9568 	if (is_64_bit_mode(vcpu))
9569 		return kvm_rip_read(vcpu);
9570 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9571 		     kvm_rip_read(vcpu));
9572 }
9573 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9574 
9575 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9576 {
9577 	return kvm_get_linear_rip(vcpu) == linear_rip;
9578 }
9579 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9580 
9581 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9582 {
9583 	unsigned long rflags;
9584 
9585 	rflags = kvm_x86_ops->get_rflags(vcpu);
9586 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9587 		rflags &= ~X86_EFLAGS_TF;
9588 	return rflags;
9589 }
9590 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9591 
9592 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9593 {
9594 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9595 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9596 		rflags |= X86_EFLAGS_TF;
9597 	kvm_x86_ops->set_rflags(vcpu, rflags);
9598 }
9599 
9600 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9601 {
9602 	__kvm_set_rflags(vcpu, rflags);
9603 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9604 }
9605 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9606 
9607 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9608 {
9609 	int r;
9610 
9611 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9612 	      work->wakeup_all)
9613 		return;
9614 
9615 	r = kvm_mmu_reload(vcpu);
9616 	if (unlikely(r))
9617 		return;
9618 
9619 	if (!vcpu->arch.mmu->direct_map &&
9620 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9621 		return;
9622 
9623 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9624 }
9625 
9626 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9627 {
9628 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9629 }
9630 
9631 static inline u32 kvm_async_pf_next_probe(u32 key)
9632 {
9633 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9634 }
9635 
9636 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9637 {
9638 	u32 key = kvm_async_pf_hash_fn(gfn);
9639 
9640 	while (vcpu->arch.apf.gfns[key] != ~0)
9641 		key = kvm_async_pf_next_probe(key);
9642 
9643 	vcpu->arch.apf.gfns[key] = gfn;
9644 }
9645 
9646 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9647 {
9648 	int i;
9649 	u32 key = kvm_async_pf_hash_fn(gfn);
9650 
9651 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9652 		     (vcpu->arch.apf.gfns[key] != gfn &&
9653 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9654 		key = kvm_async_pf_next_probe(key);
9655 
9656 	return key;
9657 }
9658 
9659 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9660 {
9661 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9662 }
9663 
9664 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9665 {
9666 	u32 i, j, k;
9667 
9668 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9669 	while (true) {
9670 		vcpu->arch.apf.gfns[i] = ~0;
9671 		do {
9672 			j = kvm_async_pf_next_probe(j);
9673 			if (vcpu->arch.apf.gfns[j] == ~0)
9674 				return;
9675 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9676 			/*
9677 			 * k lies cyclically in ]i,j]
9678 			 * |    i.k.j |
9679 			 * |....j i.k.| or  |.k..j i...|
9680 			 */
9681 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9682 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9683 		i = j;
9684 	}
9685 }
9686 
9687 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9688 {
9689 
9690 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9691 				      sizeof(val));
9692 }
9693 
9694 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9695 {
9696 
9697 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9698 				      sizeof(u32));
9699 }
9700 
9701 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9702 				     struct kvm_async_pf *work)
9703 {
9704 	struct x86_exception fault;
9705 
9706 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9707 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9708 
9709 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9710 	    (vcpu->arch.apf.send_user_only &&
9711 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9712 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9713 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9714 		fault.vector = PF_VECTOR;
9715 		fault.error_code_valid = true;
9716 		fault.error_code = 0;
9717 		fault.nested_page_fault = false;
9718 		fault.address = work->arch.token;
9719 		fault.async_page_fault = true;
9720 		kvm_inject_page_fault(vcpu, &fault);
9721 	}
9722 }
9723 
9724 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9725 				 struct kvm_async_pf *work)
9726 {
9727 	struct x86_exception fault;
9728 	u32 val;
9729 
9730 	if (work->wakeup_all)
9731 		work->arch.token = ~0; /* broadcast wakeup */
9732 	else
9733 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9734 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9735 
9736 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9737 	    !apf_get_user(vcpu, &val)) {
9738 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9739 		    vcpu->arch.exception.pending &&
9740 		    vcpu->arch.exception.nr == PF_VECTOR &&
9741 		    !apf_put_user(vcpu, 0)) {
9742 			vcpu->arch.exception.injected = false;
9743 			vcpu->arch.exception.pending = false;
9744 			vcpu->arch.exception.nr = 0;
9745 			vcpu->arch.exception.has_error_code = false;
9746 			vcpu->arch.exception.error_code = 0;
9747 			vcpu->arch.exception.has_payload = false;
9748 			vcpu->arch.exception.payload = 0;
9749 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9750 			fault.vector = PF_VECTOR;
9751 			fault.error_code_valid = true;
9752 			fault.error_code = 0;
9753 			fault.nested_page_fault = false;
9754 			fault.address = work->arch.token;
9755 			fault.async_page_fault = true;
9756 			kvm_inject_page_fault(vcpu, &fault);
9757 		}
9758 	}
9759 	vcpu->arch.apf.halted = false;
9760 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9761 }
9762 
9763 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9764 {
9765 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9766 		return true;
9767 	else
9768 		return kvm_can_do_async_pf(vcpu);
9769 }
9770 
9771 void kvm_arch_start_assignment(struct kvm *kvm)
9772 {
9773 	atomic_inc(&kvm->arch.assigned_device_count);
9774 }
9775 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9776 
9777 void kvm_arch_end_assignment(struct kvm *kvm)
9778 {
9779 	atomic_dec(&kvm->arch.assigned_device_count);
9780 }
9781 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9782 
9783 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9784 {
9785 	return atomic_read(&kvm->arch.assigned_device_count);
9786 }
9787 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9788 
9789 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9790 {
9791 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9792 }
9793 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9794 
9795 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9796 {
9797 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9798 }
9799 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9800 
9801 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9802 {
9803 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9804 }
9805 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9806 
9807 bool kvm_arch_has_irq_bypass(void)
9808 {
9809 	return kvm_x86_ops->update_pi_irte != NULL;
9810 }
9811 
9812 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9813 				      struct irq_bypass_producer *prod)
9814 {
9815 	struct kvm_kernel_irqfd *irqfd =
9816 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9817 
9818 	irqfd->producer = prod;
9819 
9820 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9821 					   prod->irq, irqfd->gsi, 1);
9822 }
9823 
9824 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9825 				      struct irq_bypass_producer *prod)
9826 {
9827 	int ret;
9828 	struct kvm_kernel_irqfd *irqfd =
9829 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9830 
9831 	WARN_ON(irqfd->producer != prod);
9832 	irqfd->producer = NULL;
9833 
9834 	/*
9835 	 * When producer of consumer is unregistered, we change back to
9836 	 * remapped mode, so we can re-use the current implementation
9837 	 * when the irq is masked/disabled or the consumer side (KVM
9838 	 * int this case doesn't want to receive the interrupts.
9839 	*/
9840 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9841 	if (ret)
9842 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9843 		       " fails: %d\n", irqfd->consumer.token, ret);
9844 }
9845 
9846 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9847 				   uint32_t guest_irq, bool set)
9848 {
9849 	if (!kvm_x86_ops->update_pi_irte)
9850 		return -EINVAL;
9851 
9852 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9853 }
9854 
9855 bool kvm_vector_hashing_enabled(void)
9856 {
9857 	return vector_hashing;
9858 }
9859 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9860 
9861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9875 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9876 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9877 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9878 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9879 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9880