xref: /openbmc/linux/arch/x86/kvm/x86.c (revision e42dd3ee)
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 static 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 
814 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
815 {
816 	if (vcpu->guest_xcr0_loaded) {
817 		if (vcpu->arch.xcr0 != host_xcr0)
818 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
819 		vcpu->guest_xcr0_loaded = 0;
820 	}
821 }
822 
823 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
824 {
825 	u64 xcr0 = xcr;
826 	u64 old_xcr0 = vcpu->arch.xcr0;
827 	u64 valid_bits;
828 
829 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
830 	if (index != XCR_XFEATURE_ENABLED_MASK)
831 		return 1;
832 	if (!(xcr0 & XFEATURE_MASK_FP))
833 		return 1;
834 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
835 		return 1;
836 
837 	/*
838 	 * Do not allow the guest to set bits that we do not support
839 	 * saving.  However, xcr0 bit 0 is always set, even if the
840 	 * emulated CPU does not support XSAVE (see fx_init).
841 	 */
842 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
843 	if (xcr0 & ~valid_bits)
844 		return 1;
845 
846 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
847 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
848 		return 1;
849 
850 	if (xcr0 & XFEATURE_MASK_AVX512) {
851 		if (!(xcr0 & XFEATURE_MASK_YMM))
852 			return 1;
853 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
854 			return 1;
855 	}
856 	vcpu->arch.xcr0 = xcr0;
857 
858 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
859 		kvm_update_cpuid(vcpu);
860 	return 0;
861 }
862 
863 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
864 {
865 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
866 	    __kvm_set_xcr(vcpu, index, xcr)) {
867 		kvm_inject_gp(vcpu, 0);
868 		return 1;
869 	}
870 	return 0;
871 }
872 EXPORT_SYMBOL_GPL(kvm_set_xcr);
873 
874 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
875 {
876 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
877 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
878 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
879 
880 	if (cr4 & CR4_RESERVED_BITS)
881 		return 1;
882 
883 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
884 		return 1;
885 
886 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
887 		return 1;
888 
889 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
890 		return 1;
891 
892 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
893 		return 1;
894 
895 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
896 		return 1;
897 
898 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
899 		return 1;
900 
901 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
902 		return 1;
903 
904 	if (is_long_mode(vcpu)) {
905 		if (!(cr4 & X86_CR4_PAE))
906 			return 1;
907 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
908 		   && ((cr4 ^ old_cr4) & pdptr_bits)
909 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
910 				   kvm_read_cr3(vcpu)))
911 		return 1;
912 
913 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
914 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
915 			return 1;
916 
917 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
918 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
919 			return 1;
920 	}
921 
922 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
923 		return 1;
924 
925 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
926 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
927 		kvm_mmu_reset_context(vcpu);
928 
929 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
930 		kvm_update_cpuid(vcpu);
931 
932 	return 0;
933 }
934 EXPORT_SYMBOL_GPL(kvm_set_cr4);
935 
936 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
937 {
938 	bool skip_tlb_flush = false;
939 #ifdef CONFIG_X86_64
940 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
941 
942 	if (pcid_enabled) {
943 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
944 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
945 	}
946 #endif
947 
948 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
949 		if (!skip_tlb_flush) {
950 			kvm_mmu_sync_roots(vcpu);
951 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
952 		}
953 		return 0;
954 	}
955 
956 	if (is_long_mode(vcpu) &&
957 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
958 		return 1;
959 	else if (is_pae(vcpu) && is_paging(vcpu) &&
960 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
961 		return 1;
962 
963 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
964 	vcpu->arch.cr3 = cr3;
965 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
966 
967 	return 0;
968 }
969 EXPORT_SYMBOL_GPL(kvm_set_cr3);
970 
971 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
972 {
973 	if (cr8 & CR8_RESERVED_BITS)
974 		return 1;
975 	if (lapic_in_kernel(vcpu))
976 		kvm_lapic_set_tpr(vcpu, cr8);
977 	else
978 		vcpu->arch.cr8 = cr8;
979 	return 0;
980 }
981 EXPORT_SYMBOL_GPL(kvm_set_cr8);
982 
983 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
984 {
985 	if (lapic_in_kernel(vcpu))
986 		return kvm_lapic_get_cr8(vcpu);
987 	else
988 		return vcpu->arch.cr8;
989 }
990 EXPORT_SYMBOL_GPL(kvm_get_cr8);
991 
992 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
993 {
994 	int i;
995 
996 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
997 		for (i = 0; i < KVM_NR_DB_REGS; i++)
998 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
999 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1000 	}
1001 }
1002 
1003 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1004 {
1005 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1006 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1007 }
1008 
1009 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1010 {
1011 	unsigned long dr7;
1012 
1013 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1014 		dr7 = vcpu->arch.guest_debug_dr7;
1015 	else
1016 		dr7 = vcpu->arch.dr7;
1017 	kvm_x86_ops->set_dr7(vcpu, dr7);
1018 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1019 	if (dr7 & DR7_BP_EN_MASK)
1020 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1021 }
1022 
1023 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1024 {
1025 	u64 fixed = DR6_FIXED_1;
1026 
1027 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1028 		fixed |= DR6_RTM;
1029 	return fixed;
1030 }
1031 
1032 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1033 {
1034 	switch (dr) {
1035 	case 0 ... 3:
1036 		vcpu->arch.db[dr] = val;
1037 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1038 			vcpu->arch.eff_db[dr] = val;
1039 		break;
1040 	case 4:
1041 		/* fall through */
1042 	case 6:
1043 		if (val & 0xffffffff00000000ULL)
1044 			return -1; /* #GP */
1045 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1046 		kvm_update_dr6(vcpu);
1047 		break;
1048 	case 5:
1049 		/* fall through */
1050 	default: /* 7 */
1051 		if (val & 0xffffffff00000000ULL)
1052 			return -1; /* #GP */
1053 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1054 		kvm_update_dr7(vcpu);
1055 		break;
1056 	}
1057 
1058 	return 0;
1059 }
1060 
1061 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1062 {
1063 	if (__kvm_set_dr(vcpu, dr, val)) {
1064 		kvm_inject_gp(vcpu, 0);
1065 		return 1;
1066 	}
1067 	return 0;
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_dr);
1070 
1071 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1072 {
1073 	switch (dr) {
1074 	case 0 ... 3:
1075 		*val = vcpu->arch.db[dr];
1076 		break;
1077 	case 4:
1078 		/* fall through */
1079 	case 6:
1080 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1081 			*val = vcpu->arch.dr6;
1082 		else
1083 			*val = kvm_x86_ops->get_dr6(vcpu);
1084 		break;
1085 	case 5:
1086 		/* fall through */
1087 	default: /* 7 */
1088 		*val = vcpu->arch.dr7;
1089 		break;
1090 	}
1091 	return 0;
1092 }
1093 EXPORT_SYMBOL_GPL(kvm_get_dr);
1094 
1095 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1096 {
1097 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1098 	u64 data;
1099 	int err;
1100 
1101 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1102 	if (err)
1103 		return err;
1104 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1105 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1106 	return err;
1107 }
1108 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1109 
1110 /*
1111  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1112  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1113  *
1114  * This list is modified at module load time to reflect the
1115  * capabilities of the host cpu. This capabilities test skips MSRs that are
1116  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1117  * may depend on host virtualization features rather than host cpu features.
1118  */
1119 
1120 static u32 msrs_to_save[] = {
1121 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1122 	MSR_STAR,
1123 #ifdef CONFIG_X86_64
1124 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1125 #endif
1126 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1127 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1128 	MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES,
1129 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1130 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1131 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1132 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1133 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1134 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1135 };
1136 
1137 static unsigned num_msrs_to_save;
1138 
1139 static u32 emulated_msrs[] = {
1140 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1141 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1142 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1143 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1144 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1145 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1146 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1147 	HV_X64_MSR_RESET,
1148 	HV_X64_MSR_VP_INDEX,
1149 	HV_X64_MSR_VP_RUNTIME,
1150 	HV_X64_MSR_SCONTROL,
1151 	HV_X64_MSR_STIMER0_CONFIG,
1152 	HV_X64_MSR_VP_ASSIST_PAGE,
1153 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1154 	HV_X64_MSR_TSC_EMULATION_STATUS,
1155 
1156 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1157 	MSR_KVM_PV_EOI_EN,
1158 
1159 	MSR_IA32_TSC_ADJUST,
1160 	MSR_IA32_TSCDEADLINE,
1161 	MSR_IA32_MISC_ENABLE,
1162 	MSR_IA32_MCG_STATUS,
1163 	MSR_IA32_MCG_CTL,
1164 	MSR_IA32_MCG_EXT_CTL,
1165 	MSR_IA32_SMBASE,
1166 	MSR_SMI_COUNT,
1167 	MSR_PLATFORM_INFO,
1168 	MSR_MISC_FEATURES_ENABLES,
1169 	MSR_AMD64_VIRT_SPEC_CTRL,
1170 };
1171 
1172 static unsigned num_emulated_msrs;
1173 
1174 /*
1175  * List of msr numbers which are used to expose MSR-based features that
1176  * can be used by a hypervisor to validate requested CPU features.
1177  */
1178 static u32 msr_based_features[] = {
1179 	MSR_IA32_VMX_BASIC,
1180 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1181 	MSR_IA32_VMX_PINBASED_CTLS,
1182 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1183 	MSR_IA32_VMX_PROCBASED_CTLS,
1184 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1185 	MSR_IA32_VMX_EXIT_CTLS,
1186 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1187 	MSR_IA32_VMX_ENTRY_CTLS,
1188 	MSR_IA32_VMX_MISC,
1189 	MSR_IA32_VMX_CR0_FIXED0,
1190 	MSR_IA32_VMX_CR0_FIXED1,
1191 	MSR_IA32_VMX_CR4_FIXED0,
1192 	MSR_IA32_VMX_CR4_FIXED1,
1193 	MSR_IA32_VMX_VMCS_ENUM,
1194 	MSR_IA32_VMX_PROCBASED_CTLS2,
1195 	MSR_IA32_VMX_EPT_VPID_CAP,
1196 	MSR_IA32_VMX_VMFUNC,
1197 
1198 	MSR_F10H_DECFG,
1199 	MSR_IA32_UCODE_REV,
1200 	MSR_IA32_ARCH_CAPABILITIES,
1201 };
1202 
1203 static unsigned int num_msr_based_features;
1204 
1205 u64 kvm_get_arch_capabilities(void)
1206 {
1207 	u64 data;
1208 
1209 	rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1210 
1211 	/*
1212 	 * If we're doing cache flushes (either "always" or "cond")
1213 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1214 	 * If an outer hypervisor is doing the cache flush for us
1215 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1216 	 * capability to the guest too, and if EPT is disabled we're not
1217 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1218 	 * require a nested hypervisor to do a flush of its own.
1219 	 */
1220 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1221 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1222 
1223 	return data;
1224 }
1225 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1226 
1227 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1228 {
1229 	switch (msr->index) {
1230 	case MSR_IA32_ARCH_CAPABILITIES:
1231 		msr->data = kvm_get_arch_capabilities();
1232 		break;
1233 	case MSR_IA32_UCODE_REV:
1234 		rdmsrl_safe(msr->index, &msr->data);
1235 		break;
1236 	default:
1237 		if (kvm_x86_ops->get_msr_feature(msr))
1238 			return 1;
1239 	}
1240 	return 0;
1241 }
1242 
1243 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1244 {
1245 	struct kvm_msr_entry msr;
1246 	int r;
1247 
1248 	msr.index = index;
1249 	r = kvm_get_msr_feature(&msr);
1250 	if (r)
1251 		return r;
1252 
1253 	*data = msr.data;
1254 
1255 	return 0;
1256 }
1257 
1258 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1259 {
1260 	if (efer & efer_reserved_bits)
1261 		return false;
1262 
1263 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1264 			return false;
1265 
1266 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1267 			return false;
1268 
1269 	return true;
1270 }
1271 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1272 
1273 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1274 {
1275 	u64 old_efer = vcpu->arch.efer;
1276 
1277 	if (!kvm_valid_efer(vcpu, efer))
1278 		return 1;
1279 
1280 	if (is_paging(vcpu)
1281 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1282 		return 1;
1283 
1284 	efer &= ~EFER_LMA;
1285 	efer |= vcpu->arch.efer & EFER_LMA;
1286 
1287 	kvm_x86_ops->set_efer(vcpu, efer);
1288 
1289 	/* Update reserved bits */
1290 	if ((efer ^ old_efer) & EFER_NX)
1291 		kvm_mmu_reset_context(vcpu);
1292 
1293 	return 0;
1294 }
1295 
1296 void kvm_enable_efer_bits(u64 mask)
1297 {
1298        efer_reserved_bits &= ~mask;
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1301 
1302 /*
1303  * Writes msr value into into the appropriate "register".
1304  * Returns 0 on success, non-0 otherwise.
1305  * Assumes vcpu_load() was already called.
1306  */
1307 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1308 {
1309 	switch (msr->index) {
1310 	case MSR_FS_BASE:
1311 	case MSR_GS_BASE:
1312 	case MSR_KERNEL_GS_BASE:
1313 	case MSR_CSTAR:
1314 	case MSR_LSTAR:
1315 		if (is_noncanonical_address(msr->data, vcpu))
1316 			return 1;
1317 		break;
1318 	case MSR_IA32_SYSENTER_EIP:
1319 	case MSR_IA32_SYSENTER_ESP:
1320 		/*
1321 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1322 		 * non-canonical address is written on Intel but not on
1323 		 * AMD (which ignores the top 32-bits, because it does
1324 		 * not implement 64-bit SYSENTER).
1325 		 *
1326 		 * 64-bit code should hence be able to write a non-canonical
1327 		 * value on AMD.  Making the address canonical ensures that
1328 		 * vmentry does not fail on Intel after writing a non-canonical
1329 		 * value, and that something deterministic happens if the guest
1330 		 * invokes 64-bit SYSENTER.
1331 		 */
1332 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1333 	}
1334 	return kvm_x86_ops->set_msr(vcpu, msr);
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_set_msr);
1337 
1338 /*
1339  * Adapt set_msr() to msr_io()'s calling convention
1340  */
1341 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1342 {
1343 	struct msr_data msr;
1344 	int r;
1345 
1346 	msr.index = index;
1347 	msr.host_initiated = true;
1348 	r = kvm_get_msr(vcpu, &msr);
1349 	if (r)
1350 		return r;
1351 
1352 	*data = msr.data;
1353 	return 0;
1354 }
1355 
1356 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1357 {
1358 	struct msr_data msr;
1359 
1360 	msr.data = *data;
1361 	msr.index = index;
1362 	msr.host_initiated = true;
1363 	return kvm_set_msr(vcpu, &msr);
1364 }
1365 
1366 #ifdef CONFIG_X86_64
1367 struct pvclock_gtod_data {
1368 	seqcount_t	seq;
1369 
1370 	struct { /* extract of a clocksource struct */
1371 		int vclock_mode;
1372 		u64	cycle_last;
1373 		u64	mask;
1374 		u32	mult;
1375 		u32	shift;
1376 	} clock;
1377 
1378 	u64		boot_ns;
1379 	u64		nsec_base;
1380 	u64		wall_time_sec;
1381 };
1382 
1383 static struct pvclock_gtod_data pvclock_gtod_data;
1384 
1385 static void update_pvclock_gtod(struct timekeeper *tk)
1386 {
1387 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1388 	u64 boot_ns;
1389 
1390 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1391 
1392 	write_seqcount_begin(&vdata->seq);
1393 
1394 	/* copy pvclock gtod data */
1395 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1396 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1397 	vdata->clock.mask		= tk->tkr_mono.mask;
1398 	vdata->clock.mult		= tk->tkr_mono.mult;
1399 	vdata->clock.shift		= tk->tkr_mono.shift;
1400 
1401 	vdata->boot_ns			= boot_ns;
1402 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1403 
1404 	vdata->wall_time_sec            = tk->xtime_sec;
1405 
1406 	write_seqcount_end(&vdata->seq);
1407 }
1408 #endif
1409 
1410 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1411 {
1412 	/*
1413 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1414 	 * vcpu_enter_guest.  This function is only called from
1415 	 * the physical CPU that is running vcpu.
1416 	 */
1417 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1418 }
1419 
1420 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1421 {
1422 	int version;
1423 	int r;
1424 	struct pvclock_wall_clock wc;
1425 	struct timespec64 boot;
1426 
1427 	if (!wall_clock)
1428 		return;
1429 
1430 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1431 	if (r)
1432 		return;
1433 
1434 	if (version & 1)
1435 		++version;  /* first time write, random junk */
1436 
1437 	++version;
1438 
1439 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1440 		return;
1441 
1442 	/*
1443 	 * The guest calculates current wall clock time by adding
1444 	 * system time (updated by kvm_guest_time_update below) to the
1445 	 * wall clock specified here.  guest system time equals host
1446 	 * system time for us, thus we must fill in host boot time here.
1447 	 */
1448 	getboottime64(&boot);
1449 
1450 	if (kvm->arch.kvmclock_offset) {
1451 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1452 		boot = timespec64_sub(boot, ts);
1453 	}
1454 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1455 	wc.nsec = boot.tv_nsec;
1456 	wc.version = version;
1457 
1458 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1459 
1460 	version++;
1461 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1462 }
1463 
1464 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1465 {
1466 	do_shl32_div32(dividend, divisor);
1467 	return dividend;
1468 }
1469 
1470 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1471 			       s8 *pshift, u32 *pmultiplier)
1472 {
1473 	uint64_t scaled64;
1474 	int32_t  shift = 0;
1475 	uint64_t tps64;
1476 	uint32_t tps32;
1477 
1478 	tps64 = base_hz;
1479 	scaled64 = scaled_hz;
1480 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1481 		tps64 >>= 1;
1482 		shift--;
1483 	}
1484 
1485 	tps32 = (uint32_t)tps64;
1486 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1487 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1488 			scaled64 >>= 1;
1489 		else
1490 			tps32 <<= 1;
1491 		shift++;
1492 	}
1493 
1494 	*pshift = shift;
1495 	*pmultiplier = div_frac(scaled64, tps32);
1496 
1497 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1498 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1499 }
1500 
1501 #ifdef CONFIG_X86_64
1502 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1503 #endif
1504 
1505 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1506 static unsigned long max_tsc_khz;
1507 
1508 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1509 {
1510 	u64 v = (u64)khz * (1000000 + ppm);
1511 	do_div(v, 1000000);
1512 	return v;
1513 }
1514 
1515 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1516 {
1517 	u64 ratio;
1518 
1519 	/* Guest TSC same frequency as host TSC? */
1520 	if (!scale) {
1521 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1522 		return 0;
1523 	}
1524 
1525 	/* TSC scaling supported? */
1526 	if (!kvm_has_tsc_control) {
1527 		if (user_tsc_khz > tsc_khz) {
1528 			vcpu->arch.tsc_catchup = 1;
1529 			vcpu->arch.tsc_always_catchup = 1;
1530 			return 0;
1531 		} else {
1532 			WARN(1, "user requested TSC rate below hardware speed\n");
1533 			return -1;
1534 		}
1535 	}
1536 
1537 	/* TSC scaling required  - calculate ratio */
1538 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1539 				user_tsc_khz, tsc_khz);
1540 
1541 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1542 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1543 			  user_tsc_khz);
1544 		return -1;
1545 	}
1546 
1547 	vcpu->arch.tsc_scaling_ratio = ratio;
1548 	return 0;
1549 }
1550 
1551 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1552 {
1553 	u32 thresh_lo, thresh_hi;
1554 	int use_scaling = 0;
1555 
1556 	/* tsc_khz can be zero if TSC calibration fails */
1557 	if (user_tsc_khz == 0) {
1558 		/* set tsc_scaling_ratio to a safe value */
1559 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1560 		return -1;
1561 	}
1562 
1563 	/* Compute a scale to convert nanoseconds in TSC cycles */
1564 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1565 			   &vcpu->arch.virtual_tsc_shift,
1566 			   &vcpu->arch.virtual_tsc_mult);
1567 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1568 
1569 	/*
1570 	 * Compute the variation in TSC rate which is acceptable
1571 	 * within the range of tolerance and decide if the
1572 	 * rate being applied is within that bounds of the hardware
1573 	 * rate.  If so, no scaling or compensation need be done.
1574 	 */
1575 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1576 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1577 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1578 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1579 		use_scaling = 1;
1580 	}
1581 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1582 }
1583 
1584 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1585 {
1586 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1587 				      vcpu->arch.virtual_tsc_mult,
1588 				      vcpu->arch.virtual_tsc_shift);
1589 	tsc += vcpu->arch.this_tsc_write;
1590 	return tsc;
1591 }
1592 
1593 static inline int gtod_is_based_on_tsc(int mode)
1594 {
1595 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1596 }
1597 
1598 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1599 {
1600 #ifdef CONFIG_X86_64
1601 	bool vcpus_matched;
1602 	struct kvm_arch *ka = &vcpu->kvm->arch;
1603 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1604 
1605 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1606 			 atomic_read(&vcpu->kvm->online_vcpus));
1607 
1608 	/*
1609 	 * Once the masterclock is enabled, always perform request in
1610 	 * order to update it.
1611 	 *
1612 	 * In order to enable masterclock, the host clocksource must be TSC
1613 	 * and the vcpus need to have matched TSCs.  When that happens,
1614 	 * perform request to enable masterclock.
1615 	 */
1616 	if (ka->use_master_clock ||
1617 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1618 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1619 
1620 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1621 			    atomic_read(&vcpu->kvm->online_vcpus),
1622 		            ka->use_master_clock, gtod->clock.vclock_mode);
1623 #endif
1624 }
1625 
1626 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1627 {
1628 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1629 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1630 }
1631 
1632 /*
1633  * Multiply tsc by a fixed point number represented by ratio.
1634  *
1635  * The most significant 64-N bits (mult) of ratio represent the
1636  * integral part of the fixed point number; the remaining N bits
1637  * (frac) represent the fractional part, ie. ratio represents a fixed
1638  * point number (mult + frac * 2^(-N)).
1639  *
1640  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1641  */
1642 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1643 {
1644 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1645 }
1646 
1647 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1648 {
1649 	u64 _tsc = tsc;
1650 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1651 
1652 	if (ratio != kvm_default_tsc_scaling_ratio)
1653 		_tsc = __scale_tsc(ratio, tsc);
1654 
1655 	return _tsc;
1656 }
1657 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1658 
1659 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1660 {
1661 	u64 tsc;
1662 
1663 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1664 
1665 	return target_tsc - tsc;
1666 }
1667 
1668 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1669 {
1670 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1671 
1672 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1673 }
1674 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1675 
1676 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1677 {
1678 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1679 }
1680 
1681 static inline bool kvm_check_tsc_unstable(void)
1682 {
1683 #ifdef CONFIG_X86_64
1684 	/*
1685 	 * TSC is marked unstable when we're running on Hyper-V,
1686 	 * 'TSC page' clocksource is good.
1687 	 */
1688 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1689 		return false;
1690 #endif
1691 	return check_tsc_unstable();
1692 }
1693 
1694 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1695 {
1696 	struct kvm *kvm = vcpu->kvm;
1697 	u64 offset, ns, elapsed;
1698 	unsigned long flags;
1699 	bool matched;
1700 	bool already_matched;
1701 	u64 data = msr->data;
1702 	bool synchronizing = false;
1703 
1704 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1705 	offset = kvm_compute_tsc_offset(vcpu, data);
1706 	ns = ktime_get_boot_ns();
1707 	elapsed = ns - kvm->arch.last_tsc_nsec;
1708 
1709 	if (vcpu->arch.virtual_tsc_khz) {
1710 		if (data == 0 && msr->host_initiated) {
1711 			/*
1712 			 * detection of vcpu initialization -- need to sync
1713 			 * with other vCPUs. This particularly helps to keep
1714 			 * kvm_clock stable after CPU hotplug
1715 			 */
1716 			synchronizing = true;
1717 		} else {
1718 			u64 tsc_exp = kvm->arch.last_tsc_write +
1719 						nsec_to_cycles(vcpu, elapsed);
1720 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1721 			/*
1722 			 * Special case: TSC write with a small delta (1 second)
1723 			 * of virtual cycle time against real time is
1724 			 * interpreted as an attempt to synchronize the CPU.
1725 			 */
1726 			synchronizing = data < tsc_exp + tsc_hz &&
1727 					data + tsc_hz > tsc_exp;
1728 		}
1729 	}
1730 
1731 	/*
1732 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1733 	 * TSC, we add elapsed time in this computation.  We could let the
1734 	 * compensation code attempt to catch up if we fall behind, but
1735 	 * it's better to try to match offsets from the beginning.
1736          */
1737 	if (synchronizing &&
1738 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1739 		if (!kvm_check_tsc_unstable()) {
1740 			offset = kvm->arch.cur_tsc_offset;
1741 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1742 		} else {
1743 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1744 			data += delta;
1745 			offset = kvm_compute_tsc_offset(vcpu, data);
1746 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1747 		}
1748 		matched = true;
1749 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1750 	} else {
1751 		/*
1752 		 * We split periods of matched TSC writes into generations.
1753 		 * For each generation, we track the original measured
1754 		 * nanosecond time, offset, and write, so if TSCs are in
1755 		 * sync, we can match exact offset, and if not, we can match
1756 		 * exact software computation in compute_guest_tsc()
1757 		 *
1758 		 * These values are tracked in kvm->arch.cur_xxx variables.
1759 		 */
1760 		kvm->arch.cur_tsc_generation++;
1761 		kvm->arch.cur_tsc_nsec = ns;
1762 		kvm->arch.cur_tsc_write = data;
1763 		kvm->arch.cur_tsc_offset = offset;
1764 		matched = false;
1765 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1766 			 kvm->arch.cur_tsc_generation, data);
1767 	}
1768 
1769 	/*
1770 	 * We also track th most recent recorded KHZ, write and time to
1771 	 * allow the matching interval to be extended at each write.
1772 	 */
1773 	kvm->arch.last_tsc_nsec = ns;
1774 	kvm->arch.last_tsc_write = data;
1775 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1776 
1777 	vcpu->arch.last_guest_tsc = data;
1778 
1779 	/* Keep track of which generation this VCPU has synchronized to */
1780 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1781 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1782 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1783 
1784 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1785 		update_ia32_tsc_adjust_msr(vcpu, offset);
1786 
1787 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1788 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1789 
1790 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1791 	if (!matched) {
1792 		kvm->arch.nr_vcpus_matched_tsc = 0;
1793 	} else if (!already_matched) {
1794 		kvm->arch.nr_vcpus_matched_tsc++;
1795 	}
1796 
1797 	kvm_track_tsc_matching(vcpu);
1798 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1799 }
1800 
1801 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1802 
1803 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1804 					   s64 adjustment)
1805 {
1806 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1807 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1808 }
1809 
1810 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1811 {
1812 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1813 		WARN_ON(adjustment < 0);
1814 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1815 	adjust_tsc_offset_guest(vcpu, adjustment);
1816 }
1817 
1818 #ifdef CONFIG_X86_64
1819 
1820 static u64 read_tsc(void)
1821 {
1822 	u64 ret = (u64)rdtsc_ordered();
1823 	u64 last = pvclock_gtod_data.clock.cycle_last;
1824 
1825 	if (likely(ret >= last))
1826 		return ret;
1827 
1828 	/*
1829 	 * GCC likes to generate cmov here, but this branch is extremely
1830 	 * predictable (it's just a function of time and the likely is
1831 	 * very likely) and there's a data dependence, so force GCC
1832 	 * to generate a branch instead.  I don't barrier() because
1833 	 * we don't actually need a barrier, and if this function
1834 	 * ever gets inlined it will generate worse code.
1835 	 */
1836 	asm volatile ("");
1837 	return last;
1838 }
1839 
1840 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1841 {
1842 	long v;
1843 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1844 	u64 tsc_pg_val;
1845 
1846 	switch (gtod->clock.vclock_mode) {
1847 	case VCLOCK_HVCLOCK:
1848 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1849 						  tsc_timestamp);
1850 		if (tsc_pg_val != U64_MAX) {
1851 			/* TSC page valid */
1852 			*mode = VCLOCK_HVCLOCK;
1853 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1854 				gtod->clock.mask;
1855 		} else {
1856 			/* TSC page invalid */
1857 			*mode = VCLOCK_NONE;
1858 		}
1859 		break;
1860 	case VCLOCK_TSC:
1861 		*mode = VCLOCK_TSC;
1862 		*tsc_timestamp = read_tsc();
1863 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1864 			gtod->clock.mask;
1865 		break;
1866 	default:
1867 		*mode = VCLOCK_NONE;
1868 	}
1869 
1870 	if (*mode == VCLOCK_NONE)
1871 		*tsc_timestamp = v = 0;
1872 
1873 	return v * gtod->clock.mult;
1874 }
1875 
1876 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1877 {
1878 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1879 	unsigned long seq;
1880 	int mode;
1881 	u64 ns;
1882 
1883 	do {
1884 		seq = read_seqcount_begin(&gtod->seq);
1885 		ns = gtod->nsec_base;
1886 		ns += vgettsc(tsc_timestamp, &mode);
1887 		ns >>= gtod->clock.shift;
1888 		ns += gtod->boot_ns;
1889 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1890 	*t = ns;
1891 
1892 	return mode;
1893 }
1894 
1895 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1896 {
1897 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1898 	unsigned long seq;
1899 	int mode;
1900 	u64 ns;
1901 
1902 	do {
1903 		seq = read_seqcount_begin(&gtod->seq);
1904 		ts->tv_sec = gtod->wall_time_sec;
1905 		ns = gtod->nsec_base;
1906 		ns += vgettsc(tsc_timestamp, &mode);
1907 		ns >>= gtod->clock.shift;
1908 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1909 
1910 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1911 	ts->tv_nsec = ns;
1912 
1913 	return mode;
1914 }
1915 
1916 /* returns true if host is using TSC based clocksource */
1917 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1918 {
1919 	/* checked again under seqlock below */
1920 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1921 		return false;
1922 
1923 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1924 						      tsc_timestamp));
1925 }
1926 
1927 /* returns true if host is using TSC based clocksource */
1928 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1929 					   u64 *tsc_timestamp)
1930 {
1931 	/* checked again under seqlock below */
1932 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1933 		return false;
1934 
1935 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1936 }
1937 #endif
1938 
1939 /*
1940  *
1941  * Assuming a stable TSC across physical CPUS, and a stable TSC
1942  * across virtual CPUs, the following condition is possible.
1943  * Each numbered line represents an event visible to both
1944  * CPUs at the next numbered event.
1945  *
1946  * "timespecX" represents host monotonic time. "tscX" represents
1947  * RDTSC value.
1948  *
1949  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1950  *
1951  * 1.  read timespec0,tsc0
1952  * 2.					| timespec1 = timespec0 + N
1953  * 					| tsc1 = tsc0 + M
1954  * 3. transition to guest		| transition to guest
1955  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1956  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1957  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1958  *
1959  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1960  *
1961  * 	- ret0 < ret1
1962  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1963  *		...
1964  *	- 0 < N - M => M < N
1965  *
1966  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1967  * always the case (the difference between two distinct xtime instances
1968  * might be smaller then the difference between corresponding TSC reads,
1969  * when updating guest vcpus pvclock areas).
1970  *
1971  * To avoid that problem, do not allow visibility of distinct
1972  * system_timestamp/tsc_timestamp values simultaneously: use a master
1973  * copy of host monotonic time values. Update that master copy
1974  * in lockstep.
1975  *
1976  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1977  *
1978  */
1979 
1980 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1981 {
1982 #ifdef CONFIG_X86_64
1983 	struct kvm_arch *ka = &kvm->arch;
1984 	int vclock_mode;
1985 	bool host_tsc_clocksource, vcpus_matched;
1986 
1987 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1988 			atomic_read(&kvm->online_vcpus));
1989 
1990 	/*
1991 	 * If the host uses TSC clock, then passthrough TSC as stable
1992 	 * to the guest.
1993 	 */
1994 	host_tsc_clocksource = kvm_get_time_and_clockread(
1995 					&ka->master_kernel_ns,
1996 					&ka->master_cycle_now);
1997 
1998 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1999 				&& !ka->backwards_tsc_observed
2000 				&& !ka->boot_vcpu_runs_old_kvmclock;
2001 
2002 	if (ka->use_master_clock)
2003 		atomic_set(&kvm_guest_has_master_clock, 1);
2004 
2005 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2006 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2007 					vcpus_matched);
2008 #endif
2009 }
2010 
2011 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2012 {
2013 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2014 }
2015 
2016 static void kvm_gen_update_masterclock(struct kvm *kvm)
2017 {
2018 #ifdef CONFIG_X86_64
2019 	int i;
2020 	struct kvm_vcpu *vcpu;
2021 	struct kvm_arch *ka = &kvm->arch;
2022 
2023 	spin_lock(&ka->pvclock_gtod_sync_lock);
2024 	kvm_make_mclock_inprogress_request(kvm);
2025 	/* no guest entries from this point */
2026 	pvclock_update_vm_gtod_copy(kvm);
2027 
2028 	kvm_for_each_vcpu(i, vcpu, kvm)
2029 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2030 
2031 	/* guest entries allowed */
2032 	kvm_for_each_vcpu(i, vcpu, kvm)
2033 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2034 
2035 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2036 #endif
2037 }
2038 
2039 u64 get_kvmclock_ns(struct kvm *kvm)
2040 {
2041 	struct kvm_arch *ka = &kvm->arch;
2042 	struct pvclock_vcpu_time_info hv_clock;
2043 	u64 ret;
2044 
2045 	spin_lock(&ka->pvclock_gtod_sync_lock);
2046 	if (!ka->use_master_clock) {
2047 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2048 		return ktime_get_boot_ns() + ka->kvmclock_offset;
2049 	}
2050 
2051 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2052 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2053 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2054 
2055 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2056 	get_cpu();
2057 
2058 	if (__this_cpu_read(cpu_tsc_khz)) {
2059 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2060 				   &hv_clock.tsc_shift,
2061 				   &hv_clock.tsc_to_system_mul);
2062 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2063 	} else
2064 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2065 
2066 	put_cpu();
2067 
2068 	return ret;
2069 }
2070 
2071 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2072 {
2073 	struct kvm_vcpu_arch *vcpu = &v->arch;
2074 	struct pvclock_vcpu_time_info guest_hv_clock;
2075 
2076 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2077 		&guest_hv_clock, sizeof(guest_hv_clock))))
2078 		return;
2079 
2080 	/* This VCPU is paused, but it's legal for a guest to read another
2081 	 * VCPU's kvmclock, so we really have to follow the specification where
2082 	 * it says that version is odd if data is being modified, and even after
2083 	 * it is consistent.
2084 	 *
2085 	 * Version field updates must be kept separate.  This is because
2086 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2087 	 * writes within a string instruction are weakly ordered.  So there
2088 	 * are three writes overall.
2089 	 *
2090 	 * As a small optimization, only write the version field in the first
2091 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2092 	 * version field is the first in the struct.
2093 	 */
2094 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2095 
2096 	if (guest_hv_clock.version & 1)
2097 		++guest_hv_clock.version;  /* first time write, random junk */
2098 
2099 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2100 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2101 				&vcpu->hv_clock,
2102 				sizeof(vcpu->hv_clock.version));
2103 
2104 	smp_wmb();
2105 
2106 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2107 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2108 
2109 	if (vcpu->pvclock_set_guest_stopped_request) {
2110 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2111 		vcpu->pvclock_set_guest_stopped_request = false;
2112 	}
2113 
2114 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2115 
2116 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2117 				&vcpu->hv_clock,
2118 				sizeof(vcpu->hv_clock));
2119 
2120 	smp_wmb();
2121 
2122 	vcpu->hv_clock.version++;
2123 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2124 				&vcpu->hv_clock,
2125 				sizeof(vcpu->hv_clock.version));
2126 }
2127 
2128 static int kvm_guest_time_update(struct kvm_vcpu *v)
2129 {
2130 	unsigned long flags, tgt_tsc_khz;
2131 	struct kvm_vcpu_arch *vcpu = &v->arch;
2132 	struct kvm_arch *ka = &v->kvm->arch;
2133 	s64 kernel_ns;
2134 	u64 tsc_timestamp, host_tsc;
2135 	u8 pvclock_flags;
2136 	bool use_master_clock;
2137 
2138 	kernel_ns = 0;
2139 	host_tsc = 0;
2140 
2141 	/*
2142 	 * If the host uses TSC clock, then passthrough TSC as stable
2143 	 * to the guest.
2144 	 */
2145 	spin_lock(&ka->pvclock_gtod_sync_lock);
2146 	use_master_clock = ka->use_master_clock;
2147 	if (use_master_clock) {
2148 		host_tsc = ka->master_cycle_now;
2149 		kernel_ns = ka->master_kernel_ns;
2150 	}
2151 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2152 
2153 	/* Keep irq disabled to prevent changes to the clock */
2154 	local_irq_save(flags);
2155 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2156 	if (unlikely(tgt_tsc_khz == 0)) {
2157 		local_irq_restore(flags);
2158 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2159 		return 1;
2160 	}
2161 	if (!use_master_clock) {
2162 		host_tsc = rdtsc();
2163 		kernel_ns = ktime_get_boot_ns();
2164 	}
2165 
2166 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2167 
2168 	/*
2169 	 * We may have to catch up the TSC to match elapsed wall clock
2170 	 * time for two reasons, even if kvmclock is used.
2171 	 *   1) CPU could have been running below the maximum TSC rate
2172 	 *   2) Broken TSC compensation resets the base at each VCPU
2173 	 *      entry to avoid unknown leaps of TSC even when running
2174 	 *      again on the same CPU.  This may cause apparent elapsed
2175 	 *      time to disappear, and the guest to stand still or run
2176 	 *	very slowly.
2177 	 */
2178 	if (vcpu->tsc_catchup) {
2179 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2180 		if (tsc > tsc_timestamp) {
2181 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2182 			tsc_timestamp = tsc;
2183 		}
2184 	}
2185 
2186 	local_irq_restore(flags);
2187 
2188 	/* With all the info we got, fill in the values */
2189 
2190 	if (kvm_has_tsc_control)
2191 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2192 
2193 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2194 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2195 				   &vcpu->hv_clock.tsc_shift,
2196 				   &vcpu->hv_clock.tsc_to_system_mul);
2197 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2198 	}
2199 
2200 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2201 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2202 	vcpu->last_guest_tsc = tsc_timestamp;
2203 
2204 	/* If the host uses TSC clocksource, then it is stable */
2205 	pvclock_flags = 0;
2206 	if (use_master_clock)
2207 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2208 
2209 	vcpu->hv_clock.flags = pvclock_flags;
2210 
2211 	if (vcpu->pv_time_enabled)
2212 		kvm_setup_pvclock_page(v);
2213 	if (v == kvm_get_vcpu(v->kvm, 0))
2214 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2215 	return 0;
2216 }
2217 
2218 /*
2219  * kvmclock updates which are isolated to a given vcpu, such as
2220  * vcpu->cpu migration, should not allow system_timestamp from
2221  * the rest of the vcpus to remain static. Otherwise ntp frequency
2222  * correction applies to one vcpu's system_timestamp but not
2223  * the others.
2224  *
2225  * So in those cases, request a kvmclock update for all vcpus.
2226  * We need to rate-limit these requests though, as they can
2227  * considerably slow guests that have a large number of vcpus.
2228  * The time for a remote vcpu to update its kvmclock is bound
2229  * by the delay we use to rate-limit the updates.
2230  */
2231 
2232 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2233 
2234 static void kvmclock_update_fn(struct work_struct *work)
2235 {
2236 	int i;
2237 	struct delayed_work *dwork = to_delayed_work(work);
2238 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2239 					   kvmclock_update_work);
2240 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2241 	struct kvm_vcpu *vcpu;
2242 
2243 	kvm_for_each_vcpu(i, vcpu, kvm) {
2244 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2245 		kvm_vcpu_kick(vcpu);
2246 	}
2247 }
2248 
2249 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2250 {
2251 	struct kvm *kvm = v->kvm;
2252 
2253 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2254 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2255 					KVMCLOCK_UPDATE_DELAY);
2256 }
2257 
2258 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2259 
2260 static void kvmclock_sync_fn(struct work_struct *work)
2261 {
2262 	struct delayed_work *dwork = to_delayed_work(work);
2263 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2264 					   kvmclock_sync_work);
2265 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2266 
2267 	if (!kvmclock_periodic_sync)
2268 		return;
2269 
2270 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2271 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2272 					KVMCLOCK_SYNC_PERIOD);
2273 }
2274 
2275 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2276 {
2277 	u64 mcg_cap = vcpu->arch.mcg_cap;
2278 	unsigned bank_num = mcg_cap & 0xff;
2279 	u32 msr = msr_info->index;
2280 	u64 data = msr_info->data;
2281 
2282 	switch (msr) {
2283 	case MSR_IA32_MCG_STATUS:
2284 		vcpu->arch.mcg_status = data;
2285 		break;
2286 	case MSR_IA32_MCG_CTL:
2287 		if (!(mcg_cap & MCG_CTL_P) &&
2288 		    (data || !msr_info->host_initiated))
2289 			return 1;
2290 		if (data != 0 && data != ~(u64)0)
2291 			return 1;
2292 		vcpu->arch.mcg_ctl = data;
2293 		break;
2294 	default:
2295 		if (msr >= MSR_IA32_MC0_CTL &&
2296 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2297 			u32 offset = msr - MSR_IA32_MC0_CTL;
2298 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2299 			 * some Linux kernels though clear bit 10 in bank 4 to
2300 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2301 			 * this to avoid an uncatched #GP in the guest
2302 			 */
2303 			if ((offset & 0x3) == 0 &&
2304 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2305 				return -1;
2306 			if (!msr_info->host_initiated &&
2307 				(offset & 0x3) == 1 && data != 0)
2308 				return -1;
2309 			vcpu->arch.mce_banks[offset] = data;
2310 			break;
2311 		}
2312 		return 1;
2313 	}
2314 	return 0;
2315 }
2316 
2317 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2318 {
2319 	struct kvm *kvm = vcpu->kvm;
2320 	int lm = is_long_mode(vcpu);
2321 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2322 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2323 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2324 		: kvm->arch.xen_hvm_config.blob_size_32;
2325 	u32 page_num = data & ~PAGE_MASK;
2326 	u64 page_addr = data & PAGE_MASK;
2327 	u8 *page;
2328 	int r;
2329 
2330 	r = -E2BIG;
2331 	if (page_num >= blob_size)
2332 		goto out;
2333 	r = -ENOMEM;
2334 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2335 	if (IS_ERR(page)) {
2336 		r = PTR_ERR(page);
2337 		goto out;
2338 	}
2339 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2340 		goto out_free;
2341 	r = 0;
2342 out_free:
2343 	kfree(page);
2344 out:
2345 	return r;
2346 }
2347 
2348 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2349 {
2350 	gpa_t gpa = data & ~0x3f;
2351 
2352 	/* Bits 3:5 are reserved, Should be zero */
2353 	if (data & 0x38)
2354 		return 1;
2355 
2356 	vcpu->arch.apf.msr_val = data;
2357 
2358 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2359 		kvm_clear_async_pf_completion_queue(vcpu);
2360 		kvm_async_pf_hash_reset(vcpu);
2361 		return 0;
2362 	}
2363 
2364 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2365 					sizeof(u32)))
2366 		return 1;
2367 
2368 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2369 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2370 	kvm_async_pf_wakeup_all(vcpu);
2371 	return 0;
2372 }
2373 
2374 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2375 {
2376 	vcpu->arch.pv_time_enabled = false;
2377 }
2378 
2379 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2380 {
2381 	++vcpu->stat.tlb_flush;
2382 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2383 }
2384 
2385 static void record_steal_time(struct kvm_vcpu *vcpu)
2386 {
2387 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2388 		return;
2389 
2390 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2391 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2392 		return;
2393 
2394 	/*
2395 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2396 	 * expensive IPIs.
2397 	 */
2398 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2399 		kvm_vcpu_flush_tlb(vcpu, false);
2400 
2401 	if (vcpu->arch.st.steal.version & 1)
2402 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2403 
2404 	vcpu->arch.st.steal.version += 1;
2405 
2406 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2407 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2408 
2409 	smp_wmb();
2410 
2411 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2412 		vcpu->arch.st.last_steal;
2413 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2414 
2415 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2416 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2417 
2418 	smp_wmb();
2419 
2420 	vcpu->arch.st.steal.version += 1;
2421 
2422 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2423 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2424 }
2425 
2426 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2427 {
2428 	bool pr = false;
2429 	u32 msr = msr_info->index;
2430 	u64 data = msr_info->data;
2431 
2432 	switch (msr) {
2433 	case MSR_AMD64_NB_CFG:
2434 	case MSR_IA32_UCODE_WRITE:
2435 	case MSR_VM_HSAVE_PA:
2436 	case MSR_AMD64_PATCH_LOADER:
2437 	case MSR_AMD64_BU_CFG2:
2438 	case MSR_AMD64_DC_CFG:
2439 	case MSR_F15H_EX_CFG:
2440 		break;
2441 
2442 	case MSR_IA32_UCODE_REV:
2443 		if (msr_info->host_initiated)
2444 			vcpu->arch.microcode_version = data;
2445 		break;
2446 	case MSR_EFER:
2447 		return set_efer(vcpu, data);
2448 	case MSR_K7_HWCR:
2449 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2450 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2451 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2452 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2453 		if (data != 0) {
2454 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2455 				    data);
2456 			return 1;
2457 		}
2458 		break;
2459 	case MSR_FAM10H_MMIO_CONF_BASE:
2460 		if (data != 0) {
2461 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2462 				    "0x%llx\n", data);
2463 			return 1;
2464 		}
2465 		break;
2466 	case MSR_IA32_DEBUGCTLMSR:
2467 		if (!data) {
2468 			/* We support the non-activated case already */
2469 			break;
2470 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2471 			/* Values other than LBR and BTF are vendor-specific,
2472 			   thus reserved and should throw a #GP */
2473 			return 1;
2474 		}
2475 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2476 			    __func__, data);
2477 		break;
2478 	case 0x200 ... 0x2ff:
2479 		return kvm_mtrr_set_msr(vcpu, msr, data);
2480 	case MSR_IA32_APICBASE:
2481 		return kvm_set_apic_base(vcpu, msr_info);
2482 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2483 		return kvm_x2apic_msr_write(vcpu, msr, data);
2484 	case MSR_IA32_TSCDEADLINE:
2485 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2486 		break;
2487 	case MSR_IA32_TSC_ADJUST:
2488 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2489 			if (!msr_info->host_initiated) {
2490 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2491 				adjust_tsc_offset_guest(vcpu, adj);
2492 			}
2493 			vcpu->arch.ia32_tsc_adjust_msr = data;
2494 		}
2495 		break;
2496 	case MSR_IA32_MISC_ENABLE:
2497 		vcpu->arch.ia32_misc_enable_msr = data;
2498 		break;
2499 	case MSR_IA32_SMBASE:
2500 		if (!msr_info->host_initiated)
2501 			return 1;
2502 		vcpu->arch.smbase = data;
2503 		break;
2504 	case MSR_IA32_TSC:
2505 		kvm_write_tsc(vcpu, msr_info);
2506 		break;
2507 	case MSR_SMI_COUNT:
2508 		if (!msr_info->host_initiated)
2509 			return 1;
2510 		vcpu->arch.smi_count = data;
2511 		break;
2512 	case MSR_KVM_WALL_CLOCK_NEW:
2513 	case MSR_KVM_WALL_CLOCK:
2514 		vcpu->kvm->arch.wall_clock = data;
2515 		kvm_write_wall_clock(vcpu->kvm, data);
2516 		break;
2517 	case MSR_KVM_SYSTEM_TIME_NEW:
2518 	case MSR_KVM_SYSTEM_TIME: {
2519 		struct kvm_arch *ka = &vcpu->kvm->arch;
2520 
2521 		kvmclock_reset(vcpu);
2522 
2523 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2524 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2525 
2526 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2527 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2528 
2529 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2530 		}
2531 
2532 		vcpu->arch.time = data;
2533 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2534 
2535 		/* we verify if the enable bit is set... */
2536 		if (!(data & 1))
2537 			break;
2538 
2539 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2540 		     &vcpu->arch.pv_time, data & ~1ULL,
2541 		     sizeof(struct pvclock_vcpu_time_info)))
2542 			vcpu->arch.pv_time_enabled = false;
2543 		else
2544 			vcpu->arch.pv_time_enabled = true;
2545 
2546 		break;
2547 	}
2548 	case MSR_KVM_ASYNC_PF_EN:
2549 		if (kvm_pv_enable_async_pf(vcpu, data))
2550 			return 1;
2551 		break;
2552 	case MSR_KVM_STEAL_TIME:
2553 
2554 		if (unlikely(!sched_info_on()))
2555 			return 1;
2556 
2557 		if (data & KVM_STEAL_RESERVED_MASK)
2558 			return 1;
2559 
2560 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2561 						data & KVM_STEAL_VALID_BITS,
2562 						sizeof(struct kvm_steal_time)))
2563 			return 1;
2564 
2565 		vcpu->arch.st.msr_val = data;
2566 
2567 		if (!(data & KVM_MSR_ENABLED))
2568 			break;
2569 
2570 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2571 
2572 		break;
2573 	case MSR_KVM_PV_EOI_EN:
2574 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2575 			return 1;
2576 		break;
2577 
2578 	case MSR_IA32_MCG_CTL:
2579 	case MSR_IA32_MCG_STATUS:
2580 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2581 		return set_msr_mce(vcpu, msr_info);
2582 
2583 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2584 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2585 		pr = true; /* fall through */
2586 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2587 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2588 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2589 			return kvm_pmu_set_msr(vcpu, msr_info);
2590 
2591 		if (pr || data != 0)
2592 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2593 				    "0x%x data 0x%llx\n", msr, data);
2594 		break;
2595 	case MSR_K7_CLK_CTL:
2596 		/*
2597 		 * Ignore all writes to this no longer documented MSR.
2598 		 * Writes are only relevant for old K7 processors,
2599 		 * all pre-dating SVM, but a recommended workaround from
2600 		 * AMD for these chips. It is possible to specify the
2601 		 * affected processor models on the command line, hence
2602 		 * the need to ignore the workaround.
2603 		 */
2604 		break;
2605 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2606 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2607 	case HV_X64_MSR_CRASH_CTL:
2608 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2609 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2610 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2611 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2612 		return kvm_hv_set_msr_common(vcpu, msr, data,
2613 					     msr_info->host_initiated);
2614 	case MSR_IA32_BBL_CR_CTL3:
2615 		/* Drop writes to this legacy MSR -- see rdmsr
2616 		 * counterpart for further detail.
2617 		 */
2618 		if (report_ignored_msrs)
2619 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2620 				msr, data);
2621 		break;
2622 	case MSR_AMD64_OSVW_ID_LENGTH:
2623 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2624 			return 1;
2625 		vcpu->arch.osvw.length = data;
2626 		break;
2627 	case MSR_AMD64_OSVW_STATUS:
2628 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2629 			return 1;
2630 		vcpu->arch.osvw.status = data;
2631 		break;
2632 	case MSR_PLATFORM_INFO:
2633 		if (!msr_info->host_initiated ||
2634 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2635 		     cpuid_fault_enabled(vcpu)))
2636 			return 1;
2637 		vcpu->arch.msr_platform_info = data;
2638 		break;
2639 	case MSR_MISC_FEATURES_ENABLES:
2640 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2641 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2642 		     !supports_cpuid_fault(vcpu)))
2643 			return 1;
2644 		vcpu->arch.msr_misc_features_enables = data;
2645 		break;
2646 	default:
2647 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2648 			return xen_hvm_config(vcpu, data);
2649 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2650 			return kvm_pmu_set_msr(vcpu, msr_info);
2651 		if (!ignore_msrs) {
2652 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2653 				    msr, data);
2654 			return 1;
2655 		} else {
2656 			if (report_ignored_msrs)
2657 				vcpu_unimpl(vcpu,
2658 					"ignored wrmsr: 0x%x data 0x%llx\n",
2659 					msr, data);
2660 			break;
2661 		}
2662 	}
2663 	return 0;
2664 }
2665 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2666 
2667 
2668 /*
2669  * Reads an msr value (of 'msr_index') into 'pdata'.
2670  * Returns 0 on success, non-0 otherwise.
2671  * Assumes vcpu_load() was already called.
2672  */
2673 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2674 {
2675 	return kvm_x86_ops->get_msr(vcpu, msr);
2676 }
2677 EXPORT_SYMBOL_GPL(kvm_get_msr);
2678 
2679 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2680 {
2681 	u64 data;
2682 	u64 mcg_cap = vcpu->arch.mcg_cap;
2683 	unsigned bank_num = mcg_cap & 0xff;
2684 
2685 	switch (msr) {
2686 	case MSR_IA32_P5_MC_ADDR:
2687 	case MSR_IA32_P5_MC_TYPE:
2688 		data = 0;
2689 		break;
2690 	case MSR_IA32_MCG_CAP:
2691 		data = vcpu->arch.mcg_cap;
2692 		break;
2693 	case MSR_IA32_MCG_CTL:
2694 		if (!(mcg_cap & MCG_CTL_P) && !host)
2695 			return 1;
2696 		data = vcpu->arch.mcg_ctl;
2697 		break;
2698 	case MSR_IA32_MCG_STATUS:
2699 		data = vcpu->arch.mcg_status;
2700 		break;
2701 	default:
2702 		if (msr >= MSR_IA32_MC0_CTL &&
2703 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2704 			u32 offset = msr - MSR_IA32_MC0_CTL;
2705 			data = vcpu->arch.mce_banks[offset];
2706 			break;
2707 		}
2708 		return 1;
2709 	}
2710 	*pdata = data;
2711 	return 0;
2712 }
2713 
2714 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2715 {
2716 	switch (msr_info->index) {
2717 	case MSR_IA32_PLATFORM_ID:
2718 	case MSR_IA32_EBL_CR_POWERON:
2719 	case MSR_IA32_DEBUGCTLMSR:
2720 	case MSR_IA32_LASTBRANCHFROMIP:
2721 	case MSR_IA32_LASTBRANCHTOIP:
2722 	case MSR_IA32_LASTINTFROMIP:
2723 	case MSR_IA32_LASTINTTOIP:
2724 	case MSR_K8_SYSCFG:
2725 	case MSR_K8_TSEG_ADDR:
2726 	case MSR_K8_TSEG_MASK:
2727 	case MSR_K7_HWCR:
2728 	case MSR_VM_HSAVE_PA:
2729 	case MSR_K8_INT_PENDING_MSG:
2730 	case MSR_AMD64_NB_CFG:
2731 	case MSR_FAM10H_MMIO_CONF_BASE:
2732 	case MSR_AMD64_BU_CFG2:
2733 	case MSR_IA32_PERF_CTL:
2734 	case MSR_AMD64_DC_CFG:
2735 	case MSR_F15H_EX_CFG:
2736 		msr_info->data = 0;
2737 		break;
2738 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2739 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2740 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2741 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2742 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2743 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2744 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2745 		msr_info->data = 0;
2746 		break;
2747 	case MSR_IA32_UCODE_REV:
2748 		msr_info->data = vcpu->arch.microcode_version;
2749 		break;
2750 	case MSR_IA32_TSC:
2751 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2752 		break;
2753 	case MSR_MTRRcap:
2754 	case 0x200 ... 0x2ff:
2755 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2756 	case 0xcd: /* fsb frequency */
2757 		msr_info->data = 3;
2758 		break;
2759 		/*
2760 		 * MSR_EBC_FREQUENCY_ID
2761 		 * Conservative value valid for even the basic CPU models.
2762 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2763 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2764 		 * and 266MHz for model 3, or 4. Set Core Clock
2765 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2766 		 * 31:24) even though these are only valid for CPU
2767 		 * models > 2, however guests may end up dividing or
2768 		 * multiplying by zero otherwise.
2769 		 */
2770 	case MSR_EBC_FREQUENCY_ID:
2771 		msr_info->data = 1 << 24;
2772 		break;
2773 	case MSR_IA32_APICBASE:
2774 		msr_info->data = kvm_get_apic_base(vcpu);
2775 		break;
2776 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2777 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2778 		break;
2779 	case MSR_IA32_TSCDEADLINE:
2780 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2781 		break;
2782 	case MSR_IA32_TSC_ADJUST:
2783 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2784 		break;
2785 	case MSR_IA32_MISC_ENABLE:
2786 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2787 		break;
2788 	case MSR_IA32_SMBASE:
2789 		if (!msr_info->host_initiated)
2790 			return 1;
2791 		msr_info->data = vcpu->arch.smbase;
2792 		break;
2793 	case MSR_SMI_COUNT:
2794 		msr_info->data = vcpu->arch.smi_count;
2795 		break;
2796 	case MSR_IA32_PERF_STATUS:
2797 		/* TSC increment by tick */
2798 		msr_info->data = 1000ULL;
2799 		/* CPU multiplier */
2800 		msr_info->data |= (((uint64_t)4ULL) << 40);
2801 		break;
2802 	case MSR_EFER:
2803 		msr_info->data = vcpu->arch.efer;
2804 		break;
2805 	case MSR_KVM_WALL_CLOCK:
2806 	case MSR_KVM_WALL_CLOCK_NEW:
2807 		msr_info->data = vcpu->kvm->arch.wall_clock;
2808 		break;
2809 	case MSR_KVM_SYSTEM_TIME:
2810 	case MSR_KVM_SYSTEM_TIME_NEW:
2811 		msr_info->data = vcpu->arch.time;
2812 		break;
2813 	case MSR_KVM_ASYNC_PF_EN:
2814 		msr_info->data = vcpu->arch.apf.msr_val;
2815 		break;
2816 	case MSR_KVM_STEAL_TIME:
2817 		msr_info->data = vcpu->arch.st.msr_val;
2818 		break;
2819 	case MSR_KVM_PV_EOI_EN:
2820 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2821 		break;
2822 	case MSR_IA32_P5_MC_ADDR:
2823 	case MSR_IA32_P5_MC_TYPE:
2824 	case MSR_IA32_MCG_CAP:
2825 	case MSR_IA32_MCG_CTL:
2826 	case MSR_IA32_MCG_STATUS:
2827 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2828 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2829 				   msr_info->host_initiated);
2830 	case MSR_K7_CLK_CTL:
2831 		/*
2832 		 * Provide expected ramp-up count for K7. All other
2833 		 * are set to zero, indicating minimum divisors for
2834 		 * every field.
2835 		 *
2836 		 * This prevents guest kernels on AMD host with CPU
2837 		 * type 6, model 8 and higher from exploding due to
2838 		 * the rdmsr failing.
2839 		 */
2840 		msr_info->data = 0x20000000;
2841 		break;
2842 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2843 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2844 	case HV_X64_MSR_CRASH_CTL:
2845 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2846 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2847 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2848 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2849 		return kvm_hv_get_msr_common(vcpu,
2850 					     msr_info->index, &msr_info->data,
2851 					     msr_info->host_initiated);
2852 		break;
2853 	case MSR_IA32_BBL_CR_CTL3:
2854 		/* This legacy MSR exists but isn't fully documented in current
2855 		 * silicon.  It is however accessed by winxp in very narrow
2856 		 * scenarios where it sets bit #19, itself documented as
2857 		 * a "reserved" bit.  Best effort attempt to source coherent
2858 		 * read data here should the balance of the register be
2859 		 * interpreted by the guest:
2860 		 *
2861 		 * L2 cache control register 3: 64GB range, 256KB size,
2862 		 * enabled, latency 0x1, configured
2863 		 */
2864 		msr_info->data = 0xbe702111;
2865 		break;
2866 	case MSR_AMD64_OSVW_ID_LENGTH:
2867 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2868 			return 1;
2869 		msr_info->data = vcpu->arch.osvw.length;
2870 		break;
2871 	case MSR_AMD64_OSVW_STATUS:
2872 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2873 			return 1;
2874 		msr_info->data = vcpu->arch.osvw.status;
2875 		break;
2876 	case MSR_PLATFORM_INFO:
2877 		if (!msr_info->host_initiated &&
2878 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2879 			return 1;
2880 		msr_info->data = vcpu->arch.msr_platform_info;
2881 		break;
2882 	case MSR_MISC_FEATURES_ENABLES:
2883 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2884 		break;
2885 	default:
2886 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2887 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2888 		if (!ignore_msrs) {
2889 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2890 					       msr_info->index);
2891 			return 1;
2892 		} else {
2893 			if (report_ignored_msrs)
2894 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2895 					msr_info->index);
2896 			msr_info->data = 0;
2897 		}
2898 		break;
2899 	}
2900 	return 0;
2901 }
2902 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2903 
2904 /*
2905  * Read or write a bunch of msrs. All parameters are kernel addresses.
2906  *
2907  * @return number of msrs set successfully.
2908  */
2909 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2910 		    struct kvm_msr_entry *entries,
2911 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2912 				  unsigned index, u64 *data))
2913 {
2914 	int i;
2915 
2916 	for (i = 0; i < msrs->nmsrs; ++i)
2917 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2918 			break;
2919 
2920 	return i;
2921 }
2922 
2923 /*
2924  * Read or write a bunch of msrs. Parameters are user addresses.
2925  *
2926  * @return number of msrs set successfully.
2927  */
2928 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2929 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2930 				unsigned index, u64 *data),
2931 		  int writeback)
2932 {
2933 	struct kvm_msrs msrs;
2934 	struct kvm_msr_entry *entries;
2935 	int r, n;
2936 	unsigned size;
2937 
2938 	r = -EFAULT;
2939 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2940 		goto out;
2941 
2942 	r = -E2BIG;
2943 	if (msrs.nmsrs >= MAX_IO_MSRS)
2944 		goto out;
2945 
2946 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2947 	entries = memdup_user(user_msrs->entries, size);
2948 	if (IS_ERR(entries)) {
2949 		r = PTR_ERR(entries);
2950 		goto out;
2951 	}
2952 
2953 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2954 	if (r < 0)
2955 		goto out_free;
2956 
2957 	r = -EFAULT;
2958 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2959 		goto out_free;
2960 
2961 	r = n;
2962 
2963 out_free:
2964 	kfree(entries);
2965 out:
2966 	return r;
2967 }
2968 
2969 static inline bool kvm_can_mwait_in_guest(void)
2970 {
2971 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2972 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2973 		boot_cpu_has(X86_FEATURE_ARAT);
2974 }
2975 
2976 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2977 {
2978 	int r = 0;
2979 
2980 	switch (ext) {
2981 	case KVM_CAP_IRQCHIP:
2982 	case KVM_CAP_HLT:
2983 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2984 	case KVM_CAP_SET_TSS_ADDR:
2985 	case KVM_CAP_EXT_CPUID:
2986 	case KVM_CAP_EXT_EMUL_CPUID:
2987 	case KVM_CAP_CLOCKSOURCE:
2988 	case KVM_CAP_PIT:
2989 	case KVM_CAP_NOP_IO_DELAY:
2990 	case KVM_CAP_MP_STATE:
2991 	case KVM_CAP_SYNC_MMU:
2992 	case KVM_CAP_USER_NMI:
2993 	case KVM_CAP_REINJECT_CONTROL:
2994 	case KVM_CAP_IRQ_INJECT_STATUS:
2995 	case KVM_CAP_IOEVENTFD:
2996 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2997 	case KVM_CAP_PIT2:
2998 	case KVM_CAP_PIT_STATE2:
2999 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3000 	case KVM_CAP_XEN_HVM:
3001 	case KVM_CAP_VCPU_EVENTS:
3002 	case KVM_CAP_HYPERV:
3003 	case KVM_CAP_HYPERV_VAPIC:
3004 	case KVM_CAP_HYPERV_SPIN:
3005 	case KVM_CAP_HYPERV_SYNIC:
3006 	case KVM_CAP_HYPERV_SYNIC2:
3007 	case KVM_CAP_HYPERV_VP_INDEX:
3008 	case KVM_CAP_HYPERV_EVENTFD:
3009 	case KVM_CAP_HYPERV_TLBFLUSH:
3010 	case KVM_CAP_HYPERV_SEND_IPI:
3011 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3012 	case KVM_CAP_HYPERV_CPUID:
3013 	case KVM_CAP_PCI_SEGMENT:
3014 	case KVM_CAP_DEBUGREGS:
3015 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3016 	case KVM_CAP_XSAVE:
3017 	case KVM_CAP_ASYNC_PF:
3018 	case KVM_CAP_GET_TSC_KHZ:
3019 	case KVM_CAP_KVMCLOCK_CTRL:
3020 	case KVM_CAP_READONLY_MEM:
3021 	case KVM_CAP_HYPERV_TIME:
3022 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3023 	case KVM_CAP_TSC_DEADLINE_TIMER:
3024 	case KVM_CAP_DISABLE_QUIRKS:
3025 	case KVM_CAP_SET_BOOT_CPU_ID:
3026  	case KVM_CAP_SPLIT_IRQCHIP:
3027 	case KVM_CAP_IMMEDIATE_EXIT:
3028 	case KVM_CAP_GET_MSR_FEATURES:
3029 	case KVM_CAP_MSR_PLATFORM_INFO:
3030 	case KVM_CAP_EXCEPTION_PAYLOAD:
3031 		r = 1;
3032 		break;
3033 	case KVM_CAP_SYNC_REGS:
3034 		r = KVM_SYNC_X86_VALID_FIELDS;
3035 		break;
3036 	case KVM_CAP_ADJUST_CLOCK:
3037 		r = KVM_CLOCK_TSC_STABLE;
3038 		break;
3039 	case KVM_CAP_X86_DISABLE_EXITS:
3040 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3041 		if(kvm_can_mwait_in_guest())
3042 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3043 		break;
3044 	case KVM_CAP_X86_SMM:
3045 		/* SMBASE is usually relocated above 1M on modern chipsets,
3046 		 * and SMM handlers might indeed rely on 4G segment limits,
3047 		 * so do not report SMM to be available if real mode is
3048 		 * emulated via vm86 mode.  Still, do not go to great lengths
3049 		 * to avoid userspace's usage of the feature, because it is a
3050 		 * fringe case that is not enabled except via specific settings
3051 		 * of the module parameters.
3052 		 */
3053 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3054 		break;
3055 	case KVM_CAP_VAPIC:
3056 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3057 		break;
3058 	case KVM_CAP_NR_VCPUS:
3059 		r = KVM_SOFT_MAX_VCPUS;
3060 		break;
3061 	case KVM_CAP_MAX_VCPUS:
3062 		r = KVM_MAX_VCPUS;
3063 		break;
3064 	case KVM_CAP_NR_MEMSLOTS:
3065 		r = KVM_USER_MEM_SLOTS;
3066 		break;
3067 	case KVM_CAP_PV_MMU:	/* obsolete */
3068 		r = 0;
3069 		break;
3070 	case KVM_CAP_MCE:
3071 		r = KVM_MAX_MCE_BANKS;
3072 		break;
3073 	case KVM_CAP_XCRS:
3074 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3075 		break;
3076 	case KVM_CAP_TSC_CONTROL:
3077 		r = kvm_has_tsc_control;
3078 		break;
3079 	case KVM_CAP_X2APIC_API:
3080 		r = KVM_X2APIC_API_VALID_FLAGS;
3081 		break;
3082 	case KVM_CAP_NESTED_STATE:
3083 		r = kvm_x86_ops->get_nested_state ?
3084 			kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3085 		break;
3086 	default:
3087 		break;
3088 	}
3089 	return r;
3090 
3091 }
3092 
3093 long kvm_arch_dev_ioctl(struct file *filp,
3094 			unsigned int ioctl, unsigned long arg)
3095 {
3096 	void __user *argp = (void __user *)arg;
3097 	long r;
3098 
3099 	switch (ioctl) {
3100 	case KVM_GET_MSR_INDEX_LIST: {
3101 		struct kvm_msr_list __user *user_msr_list = argp;
3102 		struct kvm_msr_list msr_list;
3103 		unsigned n;
3104 
3105 		r = -EFAULT;
3106 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3107 			goto out;
3108 		n = msr_list.nmsrs;
3109 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3110 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3111 			goto out;
3112 		r = -E2BIG;
3113 		if (n < msr_list.nmsrs)
3114 			goto out;
3115 		r = -EFAULT;
3116 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3117 				 num_msrs_to_save * sizeof(u32)))
3118 			goto out;
3119 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3120 				 &emulated_msrs,
3121 				 num_emulated_msrs * sizeof(u32)))
3122 			goto out;
3123 		r = 0;
3124 		break;
3125 	}
3126 	case KVM_GET_SUPPORTED_CPUID:
3127 	case KVM_GET_EMULATED_CPUID: {
3128 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3129 		struct kvm_cpuid2 cpuid;
3130 
3131 		r = -EFAULT;
3132 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3133 			goto out;
3134 
3135 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3136 					    ioctl);
3137 		if (r)
3138 			goto out;
3139 
3140 		r = -EFAULT;
3141 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3142 			goto out;
3143 		r = 0;
3144 		break;
3145 	}
3146 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3147 		r = -EFAULT;
3148 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3149 				 sizeof(kvm_mce_cap_supported)))
3150 			goto out;
3151 		r = 0;
3152 		break;
3153 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3154 		struct kvm_msr_list __user *user_msr_list = argp;
3155 		struct kvm_msr_list msr_list;
3156 		unsigned int n;
3157 
3158 		r = -EFAULT;
3159 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3160 			goto out;
3161 		n = msr_list.nmsrs;
3162 		msr_list.nmsrs = num_msr_based_features;
3163 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3164 			goto out;
3165 		r = -E2BIG;
3166 		if (n < msr_list.nmsrs)
3167 			goto out;
3168 		r = -EFAULT;
3169 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3170 				 num_msr_based_features * sizeof(u32)))
3171 			goto out;
3172 		r = 0;
3173 		break;
3174 	}
3175 	case KVM_GET_MSRS:
3176 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3177 		break;
3178 	}
3179 	default:
3180 		r = -EINVAL;
3181 	}
3182 out:
3183 	return r;
3184 }
3185 
3186 static void wbinvd_ipi(void *garbage)
3187 {
3188 	wbinvd();
3189 }
3190 
3191 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3192 {
3193 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3194 }
3195 
3196 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3197 {
3198 	/* Address WBINVD may be executed by guest */
3199 	if (need_emulate_wbinvd(vcpu)) {
3200 		if (kvm_x86_ops->has_wbinvd_exit())
3201 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3202 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3203 			smp_call_function_single(vcpu->cpu,
3204 					wbinvd_ipi, NULL, 1);
3205 	}
3206 
3207 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3208 
3209 	/* Apply any externally detected TSC adjustments (due to suspend) */
3210 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3211 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3212 		vcpu->arch.tsc_offset_adjustment = 0;
3213 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3214 	}
3215 
3216 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3217 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3218 				rdtsc() - vcpu->arch.last_host_tsc;
3219 		if (tsc_delta < 0)
3220 			mark_tsc_unstable("KVM discovered backwards TSC");
3221 
3222 		if (kvm_check_tsc_unstable()) {
3223 			u64 offset = kvm_compute_tsc_offset(vcpu,
3224 						vcpu->arch.last_guest_tsc);
3225 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3226 			vcpu->arch.tsc_catchup = 1;
3227 		}
3228 
3229 		if (kvm_lapic_hv_timer_in_use(vcpu))
3230 			kvm_lapic_restart_hv_timer(vcpu);
3231 
3232 		/*
3233 		 * On a host with synchronized TSC, there is no need to update
3234 		 * kvmclock on vcpu->cpu migration
3235 		 */
3236 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3237 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3238 		if (vcpu->cpu != cpu)
3239 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3240 		vcpu->cpu = cpu;
3241 	}
3242 
3243 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3244 }
3245 
3246 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3247 {
3248 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3249 		return;
3250 
3251 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3252 
3253 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3254 			&vcpu->arch.st.steal.preempted,
3255 			offsetof(struct kvm_steal_time, preempted),
3256 			sizeof(vcpu->arch.st.steal.preempted));
3257 }
3258 
3259 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3260 {
3261 	int idx;
3262 
3263 	if (vcpu->preempted)
3264 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3265 
3266 	/*
3267 	 * Disable page faults because we're in atomic context here.
3268 	 * kvm_write_guest_offset_cached() would call might_fault()
3269 	 * that relies on pagefault_disable() to tell if there's a
3270 	 * bug. NOTE: the write to guest memory may not go through if
3271 	 * during postcopy live migration or if there's heavy guest
3272 	 * paging.
3273 	 */
3274 	pagefault_disable();
3275 	/*
3276 	 * kvm_memslots() will be called by
3277 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3278 	 */
3279 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3280 	kvm_steal_time_set_preempted(vcpu);
3281 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3282 	pagefault_enable();
3283 	kvm_x86_ops->vcpu_put(vcpu);
3284 	vcpu->arch.last_host_tsc = rdtsc();
3285 	/*
3286 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3287 	 * on every vmexit, but if not, we might have a stale dr6 from the
3288 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3289 	 */
3290 	set_debugreg(0, 6);
3291 }
3292 
3293 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3294 				    struct kvm_lapic_state *s)
3295 {
3296 	if (vcpu->arch.apicv_active)
3297 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3298 
3299 	return kvm_apic_get_state(vcpu, s);
3300 }
3301 
3302 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3303 				    struct kvm_lapic_state *s)
3304 {
3305 	int r;
3306 
3307 	r = kvm_apic_set_state(vcpu, s);
3308 	if (r)
3309 		return r;
3310 	update_cr8_intercept(vcpu);
3311 
3312 	return 0;
3313 }
3314 
3315 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3316 {
3317 	return (!lapic_in_kernel(vcpu) ||
3318 		kvm_apic_accept_pic_intr(vcpu));
3319 }
3320 
3321 /*
3322  * if userspace requested an interrupt window, check that the
3323  * interrupt window is open.
3324  *
3325  * No need to exit to userspace if we already have an interrupt queued.
3326  */
3327 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3328 {
3329 	return kvm_arch_interrupt_allowed(vcpu) &&
3330 		!kvm_cpu_has_interrupt(vcpu) &&
3331 		!kvm_event_needs_reinjection(vcpu) &&
3332 		kvm_cpu_accept_dm_intr(vcpu);
3333 }
3334 
3335 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3336 				    struct kvm_interrupt *irq)
3337 {
3338 	if (irq->irq >= KVM_NR_INTERRUPTS)
3339 		return -EINVAL;
3340 
3341 	if (!irqchip_in_kernel(vcpu->kvm)) {
3342 		kvm_queue_interrupt(vcpu, irq->irq, false);
3343 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3344 		return 0;
3345 	}
3346 
3347 	/*
3348 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3349 	 * fail for in-kernel 8259.
3350 	 */
3351 	if (pic_in_kernel(vcpu->kvm))
3352 		return -ENXIO;
3353 
3354 	if (vcpu->arch.pending_external_vector != -1)
3355 		return -EEXIST;
3356 
3357 	vcpu->arch.pending_external_vector = irq->irq;
3358 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3359 	return 0;
3360 }
3361 
3362 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3363 {
3364 	kvm_inject_nmi(vcpu);
3365 
3366 	return 0;
3367 }
3368 
3369 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3370 {
3371 	kvm_make_request(KVM_REQ_SMI, vcpu);
3372 
3373 	return 0;
3374 }
3375 
3376 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3377 					   struct kvm_tpr_access_ctl *tac)
3378 {
3379 	if (tac->flags)
3380 		return -EINVAL;
3381 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3382 	return 0;
3383 }
3384 
3385 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3386 					u64 mcg_cap)
3387 {
3388 	int r;
3389 	unsigned bank_num = mcg_cap & 0xff, bank;
3390 
3391 	r = -EINVAL;
3392 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3393 		goto out;
3394 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3395 		goto out;
3396 	r = 0;
3397 	vcpu->arch.mcg_cap = mcg_cap;
3398 	/* Init IA32_MCG_CTL to all 1s */
3399 	if (mcg_cap & MCG_CTL_P)
3400 		vcpu->arch.mcg_ctl = ~(u64)0;
3401 	/* Init IA32_MCi_CTL to all 1s */
3402 	for (bank = 0; bank < bank_num; bank++)
3403 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3404 
3405 	if (kvm_x86_ops->setup_mce)
3406 		kvm_x86_ops->setup_mce(vcpu);
3407 out:
3408 	return r;
3409 }
3410 
3411 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3412 				      struct kvm_x86_mce *mce)
3413 {
3414 	u64 mcg_cap = vcpu->arch.mcg_cap;
3415 	unsigned bank_num = mcg_cap & 0xff;
3416 	u64 *banks = vcpu->arch.mce_banks;
3417 
3418 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3419 		return -EINVAL;
3420 	/*
3421 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3422 	 * reporting is disabled
3423 	 */
3424 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3425 	    vcpu->arch.mcg_ctl != ~(u64)0)
3426 		return 0;
3427 	banks += 4 * mce->bank;
3428 	/*
3429 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3430 	 * reporting is disabled for the bank
3431 	 */
3432 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3433 		return 0;
3434 	if (mce->status & MCI_STATUS_UC) {
3435 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3436 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3437 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3438 			return 0;
3439 		}
3440 		if (banks[1] & MCI_STATUS_VAL)
3441 			mce->status |= MCI_STATUS_OVER;
3442 		banks[2] = mce->addr;
3443 		banks[3] = mce->misc;
3444 		vcpu->arch.mcg_status = mce->mcg_status;
3445 		banks[1] = mce->status;
3446 		kvm_queue_exception(vcpu, MC_VECTOR);
3447 	} else if (!(banks[1] & MCI_STATUS_VAL)
3448 		   || !(banks[1] & MCI_STATUS_UC)) {
3449 		if (banks[1] & MCI_STATUS_VAL)
3450 			mce->status |= MCI_STATUS_OVER;
3451 		banks[2] = mce->addr;
3452 		banks[3] = mce->misc;
3453 		banks[1] = mce->status;
3454 	} else
3455 		banks[1] |= MCI_STATUS_OVER;
3456 	return 0;
3457 }
3458 
3459 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3460 					       struct kvm_vcpu_events *events)
3461 {
3462 	process_nmi(vcpu);
3463 
3464 	/*
3465 	 * The API doesn't provide the instruction length for software
3466 	 * exceptions, so don't report them. As long as the guest RIP
3467 	 * isn't advanced, we should expect to encounter the exception
3468 	 * again.
3469 	 */
3470 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3471 		events->exception.injected = 0;
3472 		events->exception.pending = 0;
3473 	} else {
3474 		events->exception.injected = vcpu->arch.exception.injected;
3475 		events->exception.pending = vcpu->arch.exception.pending;
3476 		/*
3477 		 * For ABI compatibility, deliberately conflate
3478 		 * pending and injected exceptions when
3479 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3480 		 */
3481 		if (!vcpu->kvm->arch.exception_payload_enabled)
3482 			events->exception.injected |=
3483 				vcpu->arch.exception.pending;
3484 	}
3485 	events->exception.nr = vcpu->arch.exception.nr;
3486 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3487 	events->exception.error_code = vcpu->arch.exception.error_code;
3488 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3489 	events->exception_payload = vcpu->arch.exception.payload;
3490 
3491 	events->interrupt.injected =
3492 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3493 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3494 	events->interrupt.soft = 0;
3495 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3496 
3497 	events->nmi.injected = vcpu->arch.nmi_injected;
3498 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3499 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3500 	events->nmi.pad = 0;
3501 
3502 	events->sipi_vector = 0; /* never valid when reporting to user space */
3503 
3504 	events->smi.smm = is_smm(vcpu);
3505 	events->smi.pending = vcpu->arch.smi_pending;
3506 	events->smi.smm_inside_nmi =
3507 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3508 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3509 
3510 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3511 			 | KVM_VCPUEVENT_VALID_SHADOW
3512 			 | KVM_VCPUEVENT_VALID_SMM);
3513 	if (vcpu->kvm->arch.exception_payload_enabled)
3514 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3515 
3516 	memset(&events->reserved, 0, sizeof(events->reserved));
3517 }
3518 
3519 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3520 
3521 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3522 					      struct kvm_vcpu_events *events)
3523 {
3524 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3525 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3526 			      | KVM_VCPUEVENT_VALID_SHADOW
3527 			      | KVM_VCPUEVENT_VALID_SMM
3528 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3529 		return -EINVAL;
3530 
3531 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3532 		if (!vcpu->kvm->arch.exception_payload_enabled)
3533 			return -EINVAL;
3534 		if (events->exception.pending)
3535 			events->exception.injected = 0;
3536 		else
3537 			events->exception_has_payload = 0;
3538 	} else {
3539 		events->exception.pending = 0;
3540 		events->exception_has_payload = 0;
3541 	}
3542 
3543 	if ((events->exception.injected || events->exception.pending) &&
3544 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3545 		return -EINVAL;
3546 
3547 	/* INITs are latched while in SMM */
3548 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3549 	    (events->smi.smm || events->smi.pending) &&
3550 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3551 		return -EINVAL;
3552 
3553 	process_nmi(vcpu);
3554 	vcpu->arch.exception.injected = events->exception.injected;
3555 	vcpu->arch.exception.pending = events->exception.pending;
3556 	vcpu->arch.exception.nr = events->exception.nr;
3557 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3558 	vcpu->arch.exception.error_code = events->exception.error_code;
3559 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3560 	vcpu->arch.exception.payload = events->exception_payload;
3561 
3562 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3563 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3564 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3565 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3566 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3567 						  events->interrupt.shadow);
3568 
3569 	vcpu->arch.nmi_injected = events->nmi.injected;
3570 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3571 		vcpu->arch.nmi_pending = events->nmi.pending;
3572 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3573 
3574 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3575 	    lapic_in_kernel(vcpu))
3576 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3577 
3578 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3579 		u32 hflags = vcpu->arch.hflags;
3580 		if (events->smi.smm)
3581 			hflags |= HF_SMM_MASK;
3582 		else
3583 			hflags &= ~HF_SMM_MASK;
3584 		kvm_set_hflags(vcpu, hflags);
3585 
3586 		vcpu->arch.smi_pending = events->smi.pending;
3587 
3588 		if (events->smi.smm) {
3589 			if (events->smi.smm_inside_nmi)
3590 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3591 			else
3592 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3593 			if (lapic_in_kernel(vcpu)) {
3594 				if (events->smi.latched_init)
3595 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3596 				else
3597 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3598 			}
3599 		}
3600 	}
3601 
3602 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3603 
3604 	return 0;
3605 }
3606 
3607 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3608 					     struct kvm_debugregs *dbgregs)
3609 {
3610 	unsigned long val;
3611 
3612 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3613 	kvm_get_dr(vcpu, 6, &val);
3614 	dbgregs->dr6 = val;
3615 	dbgregs->dr7 = vcpu->arch.dr7;
3616 	dbgregs->flags = 0;
3617 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3618 }
3619 
3620 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3621 					    struct kvm_debugregs *dbgregs)
3622 {
3623 	if (dbgregs->flags)
3624 		return -EINVAL;
3625 
3626 	if (dbgregs->dr6 & ~0xffffffffull)
3627 		return -EINVAL;
3628 	if (dbgregs->dr7 & ~0xffffffffull)
3629 		return -EINVAL;
3630 
3631 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3632 	kvm_update_dr0123(vcpu);
3633 	vcpu->arch.dr6 = dbgregs->dr6;
3634 	kvm_update_dr6(vcpu);
3635 	vcpu->arch.dr7 = dbgregs->dr7;
3636 	kvm_update_dr7(vcpu);
3637 
3638 	return 0;
3639 }
3640 
3641 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3642 
3643 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3644 {
3645 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3646 	u64 xstate_bv = xsave->header.xfeatures;
3647 	u64 valid;
3648 
3649 	/*
3650 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3651 	 * leaves 0 and 1 in the loop below.
3652 	 */
3653 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3654 
3655 	/* Set XSTATE_BV */
3656 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3657 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3658 
3659 	/*
3660 	 * Copy each region from the possibly compacted offset to the
3661 	 * non-compacted offset.
3662 	 */
3663 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3664 	while (valid) {
3665 		u64 feature = valid & -valid;
3666 		int index = fls64(feature) - 1;
3667 		void *src = get_xsave_addr(xsave, feature);
3668 
3669 		if (src) {
3670 			u32 size, offset, ecx, edx;
3671 			cpuid_count(XSTATE_CPUID, index,
3672 				    &size, &offset, &ecx, &edx);
3673 			if (feature == XFEATURE_MASK_PKRU)
3674 				memcpy(dest + offset, &vcpu->arch.pkru,
3675 				       sizeof(vcpu->arch.pkru));
3676 			else
3677 				memcpy(dest + offset, src, size);
3678 
3679 		}
3680 
3681 		valid -= feature;
3682 	}
3683 }
3684 
3685 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3686 {
3687 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3688 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3689 	u64 valid;
3690 
3691 	/*
3692 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3693 	 * leaves 0 and 1 in the loop below.
3694 	 */
3695 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3696 
3697 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3698 	xsave->header.xfeatures = xstate_bv;
3699 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3700 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3701 
3702 	/*
3703 	 * Copy each region from the non-compacted offset to the
3704 	 * possibly compacted offset.
3705 	 */
3706 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3707 	while (valid) {
3708 		u64 feature = valid & -valid;
3709 		int index = fls64(feature) - 1;
3710 		void *dest = get_xsave_addr(xsave, feature);
3711 
3712 		if (dest) {
3713 			u32 size, offset, ecx, edx;
3714 			cpuid_count(XSTATE_CPUID, index,
3715 				    &size, &offset, &ecx, &edx);
3716 			if (feature == XFEATURE_MASK_PKRU)
3717 				memcpy(&vcpu->arch.pkru, src + offset,
3718 				       sizeof(vcpu->arch.pkru));
3719 			else
3720 				memcpy(dest, src + offset, size);
3721 		}
3722 
3723 		valid -= feature;
3724 	}
3725 }
3726 
3727 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3728 					 struct kvm_xsave *guest_xsave)
3729 {
3730 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3731 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3732 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3733 	} else {
3734 		memcpy(guest_xsave->region,
3735 			&vcpu->arch.guest_fpu->state.fxsave,
3736 			sizeof(struct fxregs_state));
3737 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3738 			XFEATURE_MASK_FPSSE;
3739 	}
3740 }
3741 
3742 #define XSAVE_MXCSR_OFFSET 24
3743 
3744 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3745 					struct kvm_xsave *guest_xsave)
3746 {
3747 	u64 xstate_bv =
3748 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3749 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3750 
3751 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3752 		/*
3753 		 * Here we allow setting states that are not present in
3754 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3755 		 * with old userspace.
3756 		 */
3757 		if (xstate_bv & ~kvm_supported_xcr0() ||
3758 			mxcsr & ~mxcsr_feature_mask)
3759 			return -EINVAL;
3760 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3761 	} else {
3762 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3763 			mxcsr & ~mxcsr_feature_mask)
3764 			return -EINVAL;
3765 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3766 			guest_xsave->region, sizeof(struct fxregs_state));
3767 	}
3768 	return 0;
3769 }
3770 
3771 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3772 					struct kvm_xcrs *guest_xcrs)
3773 {
3774 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3775 		guest_xcrs->nr_xcrs = 0;
3776 		return;
3777 	}
3778 
3779 	guest_xcrs->nr_xcrs = 1;
3780 	guest_xcrs->flags = 0;
3781 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3782 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3783 }
3784 
3785 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3786 				       struct kvm_xcrs *guest_xcrs)
3787 {
3788 	int i, r = 0;
3789 
3790 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3791 		return -EINVAL;
3792 
3793 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3794 		return -EINVAL;
3795 
3796 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3797 		/* Only support XCR0 currently */
3798 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3799 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3800 				guest_xcrs->xcrs[i].value);
3801 			break;
3802 		}
3803 	if (r)
3804 		r = -EINVAL;
3805 	return r;
3806 }
3807 
3808 /*
3809  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3810  * stopped by the hypervisor.  This function will be called from the host only.
3811  * EINVAL is returned when the host attempts to set the flag for a guest that
3812  * does not support pv clocks.
3813  */
3814 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3815 {
3816 	if (!vcpu->arch.pv_time_enabled)
3817 		return -EINVAL;
3818 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3819 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3820 	return 0;
3821 }
3822 
3823 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3824 				     struct kvm_enable_cap *cap)
3825 {
3826 	int r;
3827 	uint16_t vmcs_version;
3828 	void __user *user_ptr;
3829 
3830 	if (cap->flags)
3831 		return -EINVAL;
3832 
3833 	switch (cap->cap) {
3834 	case KVM_CAP_HYPERV_SYNIC2:
3835 		if (cap->args[0])
3836 			return -EINVAL;
3837 		/* fall through */
3838 
3839 	case KVM_CAP_HYPERV_SYNIC:
3840 		if (!irqchip_in_kernel(vcpu->kvm))
3841 			return -EINVAL;
3842 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3843 					     KVM_CAP_HYPERV_SYNIC2);
3844 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3845 		if (!kvm_x86_ops->nested_enable_evmcs)
3846 			return -ENOTTY;
3847 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3848 		if (!r) {
3849 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
3850 			if (copy_to_user(user_ptr, &vmcs_version,
3851 					 sizeof(vmcs_version)))
3852 				r = -EFAULT;
3853 		}
3854 		return r;
3855 
3856 	default:
3857 		return -EINVAL;
3858 	}
3859 }
3860 
3861 long kvm_arch_vcpu_ioctl(struct file *filp,
3862 			 unsigned int ioctl, unsigned long arg)
3863 {
3864 	struct kvm_vcpu *vcpu = filp->private_data;
3865 	void __user *argp = (void __user *)arg;
3866 	int r;
3867 	union {
3868 		struct kvm_lapic_state *lapic;
3869 		struct kvm_xsave *xsave;
3870 		struct kvm_xcrs *xcrs;
3871 		void *buffer;
3872 	} u;
3873 
3874 	vcpu_load(vcpu);
3875 
3876 	u.buffer = NULL;
3877 	switch (ioctl) {
3878 	case KVM_GET_LAPIC: {
3879 		r = -EINVAL;
3880 		if (!lapic_in_kernel(vcpu))
3881 			goto out;
3882 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
3883 				GFP_KERNEL_ACCOUNT);
3884 
3885 		r = -ENOMEM;
3886 		if (!u.lapic)
3887 			goto out;
3888 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3889 		if (r)
3890 			goto out;
3891 		r = -EFAULT;
3892 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3893 			goto out;
3894 		r = 0;
3895 		break;
3896 	}
3897 	case KVM_SET_LAPIC: {
3898 		r = -EINVAL;
3899 		if (!lapic_in_kernel(vcpu))
3900 			goto out;
3901 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3902 		if (IS_ERR(u.lapic)) {
3903 			r = PTR_ERR(u.lapic);
3904 			goto out_nofree;
3905 		}
3906 
3907 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3908 		break;
3909 	}
3910 	case KVM_INTERRUPT: {
3911 		struct kvm_interrupt irq;
3912 
3913 		r = -EFAULT;
3914 		if (copy_from_user(&irq, argp, sizeof(irq)))
3915 			goto out;
3916 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3917 		break;
3918 	}
3919 	case KVM_NMI: {
3920 		r = kvm_vcpu_ioctl_nmi(vcpu);
3921 		break;
3922 	}
3923 	case KVM_SMI: {
3924 		r = kvm_vcpu_ioctl_smi(vcpu);
3925 		break;
3926 	}
3927 	case KVM_SET_CPUID: {
3928 		struct kvm_cpuid __user *cpuid_arg = argp;
3929 		struct kvm_cpuid cpuid;
3930 
3931 		r = -EFAULT;
3932 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3933 			goto out;
3934 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3935 		break;
3936 	}
3937 	case KVM_SET_CPUID2: {
3938 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3939 		struct kvm_cpuid2 cpuid;
3940 
3941 		r = -EFAULT;
3942 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3943 			goto out;
3944 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3945 					      cpuid_arg->entries);
3946 		break;
3947 	}
3948 	case KVM_GET_CPUID2: {
3949 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3950 		struct kvm_cpuid2 cpuid;
3951 
3952 		r = -EFAULT;
3953 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3954 			goto out;
3955 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3956 					      cpuid_arg->entries);
3957 		if (r)
3958 			goto out;
3959 		r = -EFAULT;
3960 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3961 			goto out;
3962 		r = 0;
3963 		break;
3964 	}
3965 	case KVM_GET_MSRS: {
3966 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3967 		r = msr_io(vcpu, argp, do_get_msr, 1);
3968 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3969 		break;
3970 	}
3971 	case KVM_SET_MSRS: {
3972 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3973 		r = msr_io(vcpu, argp, do_set_msr, 0);
3974 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3975 		break;
3976 	}
3977 	case KVM_TPR_ACCESS_REPORTING: {
3978 		struct kvm_tpr_access_ctl tac;
3979 
3980 		r = -EFAULT;
3981 		if (copy_from_user(&tac, argp, sizeof(tac)))
3982 			goto out;
3983 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3984 		if (r)
3985 			goto out;
3986 		r = -EFAULT;
3987 		if (copy_to_user(argp, &tac, sizeof(tac)))
3988 			goto out;
3989 		r = 0;
3990 		break;
3991 	};
3992 	case KVM_SET_VAPIC_ADDR: {
3993 		struct kvm_vapic_addr va;
3994 		int idx;
3995 
3996 		r = -EINVAL;
3997 		if (!lapic_in_kernel(vcpu))
3998 			goto out;
3999 		r = -EFAULT;
4000 		if (copy_from_user(&va, argp, sizeof(va)))
4001 			goto out;
4002 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4003 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4004 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4005 		break;
4006 	}
4007 	case KVM_X86_SETUP_MCE: {
4008 		u64 mcg_cap;
4009 
4010 		r = -EFAULT;
4011 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4012 			goto out;
4013 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4014 		break;
4015 	}
4016 	case KVM_X86_SET_MCE: {
4017 		struct kvm_x86_mce mce;
4018 
4019 		r = -EFAULT;
4020 		if (copy_from_user(&mce, argp, sizeof(mce)))
4021 			goto out;
4022 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4023 		break;
4024 	}
4025 	case KVM_GET_VCPU_EVENTS: {
4026 		struct kvm_vcpu_events events;
4027 
4028 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4029 
4030 		r = -EFAULT;
4031 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4032 			break;
4033 		r = 0;
4034 		break;
4035 	}
4036 	case KVM_SET_VCPU_EVENTS: {
4037 		struct kvm_vcpu_events events;
4038 
4039 		r = -EFAULT;
4040 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4041 			break;
4042 
4043 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4044 		break;
4045 	}
4046 	case KVM_GET_DEBUGREGS: {
4047 		struct kvm_debugregs dbgregs;
4048 
4049 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4050 
4051 		r = -EFAULT;
4052 		if (copy_to_user(argp, &dbgregs,
4053 				 sizeof(struct kvm_debugregs)))
4054 			break;
4055 		r = 0;
4056 		break;
4057 	}
4058 	case KVM_SET_DEBUGREGS: {
4059 		struct kvm_debugregs dbgregs;
4060 
4061 		r = -EFAULT;
4062 		if (copy_from_user(&dbgregs, argp,
4063 				   sizeof(struct kvm_debugregs)))
4064 			break;
4065 
4066 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4067 		break;
4068 	}
4069 	case KVM_GET_XSAVE: {
4070 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4071 		r = -ENOMEM;
4072 		if (!u.xsave)
4073 			break;
4074 
4075 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4076 
4077 		r = -EFAULT;
4078 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4079 			break;
4080 		r = 0;
4081 		break;
4082 	}
4083 	case KVM_SET_XSAVE: {
4084 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4085 		if (IS_ERR(u.xsave)) {
4086 			r = PTR_ERR(u.xsave);
4087 			goto out_nofree;
4088 		}
4089 
4090 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4091 		break;
4092 	}
4093 	case KVM_GET_XCRS: {
4094 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4095 		r = -ENOMEM;
4096 		if (!u.xcrs)
4097 			break;
4098 
4099 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4100 
4101 		r = -EFAULT;
4102 		if (copy_to_user(argp, u.xcrs,
4103 				 sizeof(struct kvm_xcrs)))
4104 			break;
4105 		r = 0;
4106 		break;
4107 	}
4108 	case KVM_SET_XCRS: {
4109 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4110 		if (IS_ERR(u.xcrs)) {
4111 			r = PTR_ERR(u.xcrs);
4112 			goto out_nofree;
4113 		}
4114 
4115 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4116 		break;
4117 	}
4118 	case KVM_SET_TSC_KHZ: {
4119 		u32 user_tsc_khz;
4120 
4121 		r = -EINVAL;
4122 		user_tsc_khz = (u32)arg;
4123 
4124 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4125 			goto out;
4126 
4127 		if (user_tsc_khz == 0)
4128 			user_tsc_khz = tsc_khz;
4129 
4130 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4131 			r = 0;
4132 
4133 		goto out;
4134 	}
4135 	case KVM_GET_TSC_KHZ: {
4136 		r = vcpu->arch.virtual_tsc_khz;
4137 		goto out;
4138 	}
4139 	case KVM_KVMCLOCK_CTRL: {
4140 		r = kvm_set_guest_paused(vcpu);
4141 		goto out;
4142 	}
4143 	case KVM_ENABLE_CAP: {
4144 		struct kvm_enable_cap cap;
4145 
4146 		r = -EFAULT;
4147 		if (copy_from_user(&cap, argp, sizeof(cap)))
4148 			goto out;
4149 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4150 		break;
4151 	}
4152 	case KVM_GET_NESTED_STATE: {
4153 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4154 		u32 user_data_size;
4155 
4156 		r = -EINVAL;
4157 		if (!kvm_x86_ops->get_nested_state)
4158 			break;
4159 
4160 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4161 		r = -EFAULT;
4162 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4163 			break;
4164 
4165 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4166 						  user_data_size);
4167 		if (r < 0)
4168 			break;
4169 
4170 		if (r > user_data_size) {
4171 			if (put_user(r, &user_kvm_nested_state->size))
4172 				r = -EFAULT;
4173 			else
4174 				r = -E2BIG;
4175 			break;
4176 		}
4177 
4178 		r = 0;
4179 		break;
4180 	}
4181 	case KVM_SET_NESTED_STATE: {
4182 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4183 		struct kvm_nested_state kvm_state;
4184 
4185 		r = -EINVAL;
4186 		if (!kvm_x86_ops->set_nested_state)
4187 			break;
4188 
4189 		r = -EFAULT;
4190 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4191 			break;
4192 
4193 		r = -EINVAL;
4194 		if (kvm_state.size < sizeof(kvm_state))
4195 			break;
4196 
4197 		if (kvm_state.flags &
4198 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4199 		      | KVM_STATE_NESTED_EVMCS))
4200 			break;
4201 
4202 		/* nested_run_pending implies guest_mode.  */
4203 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4204 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4205 			break;
4206 
4207 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4208 		break;
4209 	}
4210 	case KVM_GET_SUPPORTED_HV_CPUID: {
4211 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4212 		struct kvm_cpuid2 cpuid;
4213 
4214 		r = -EFAULT;
4215 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4216 			goto out;
4217 
4218 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4219 						cpuid_arg->entries);
4220 		if (r)
4221 			goto out;
4222 
4223 		r = -EFAULT;
4224 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4225 			goto out;
4226 		r = 0;
4227 		break;
4228 	}
4229 	default:
4230 		r = -EINVAL;
4231 	}
4232 out:
4233 	kfree(u.buffer);
4234 out_nofree:
4235 	vcpu_put(vcpu);
4236 	return r;
4237 }
4238 
4239 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4240 {
4241 	return VM_FAULT_SIGBUS;
4242 }
4243 
4244 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4245 {
4246 	int ret;
4247 
4248 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4249 		return -EINVAL;
4250 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4251 	return ret;
4252 }
4253 
4254 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4255 					      u64 ident_addr)
4256 {
4257 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4258 }
4259 
4260 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4261 					  u32 kvm_nr_mmu_pages)
4262 {
4263 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4264 		return -EINVAL;
4265 
4266 	mutex_lock(&kvm->slots_lock);
4267 
4268 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4269 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4270 
4271 	mutex_unlock(&kvm->slots_lock);
4272 	return 0;
4273 }
4274 
4275 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4276 {
4277 	return kvm->arch.n_max_mmu_pages;
4278 }
4279 
4280 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4281 {
4282 	struct kvm_pic *pic = kvm->arch.vpic;
4283 	int r;
4284 
4285 	r = 0;
4286 	switch (chip->chip_id) {
4287 	case KVM_IRQCHIP_PIC_MASTER:
4288 		memcpy(&chip->chip.pic, &pic->pics[0],
4289 			sizeof(struct kvm_pic_state));
4290 		break;
4291 	case KVM_IRQCHIP_PIC_SLAVE:
4292 		memcpy(&chip->chip.pic, &pic->pics[1],
4293 			sizeof(struct kvm_pic_state));
4294 		break;
4295 	case KVM_IRQCHIP_IOAPIC:
4296 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4297 		break;
4298 	default:
4299 		r = -EINVAL;
4300 		break;
4301 	}
4302 	return r;
4303 }
4304 
4305 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4306 {
4307 	struct kvm_pic *pic = kvm->arch.vpic;
4308 	int r;
4309 
4310 	r = 0;
4311 	switch (chip->chip_id) {
4312 	case KVM_IRQCHIP_PIC_MASTER:
4313 		spin_lock(&pic->lock);
4314 		memcpy(&pic->pics[0], &chip->chip.pic,
4315 			sizeof(struct kvm_pic_state));
4316 		spin_unlock(&pic->lock);
4317 		break;
4318 	case KVM_IRQCHIP_PIC_SLAVE:
4319 		spin_lock(&pic->lock);
4320 		memcpy(&pic->pics[1], &chip->chip.pic,
4321 			sizeof(struct kvm_pic_state));
4322 		spin_unlock(&pic->lock);
4323 		break;
4324 	case KVM_IRQCHIP_IOAPIC:
4325 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4326 		break;
4327 	default:
4328 		r = -EINVAL;
4329 		break;
4330 	}
4331 	kvm_pic_update_irq(pic);
4332 	return r;
4333 }
4334 
4335 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4336 {
4337 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4338 
4339 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4340 
4341 	mutex_lock(&kps->lock);
4342 	memcpy(ps, &kps->channels, sizeof(*ps));
4343 	mutex_unlock(&kps->lock);
4344 	return 0;
4345 }
4346 
4347 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4348 {
4349 	int i;
4350 	struct kvm_pit *pit = kvm->arch.vpit;
4351 
4352 	mutex_lock(&pit->pit_state.lock);
4353 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4354 	for (i = 0; i < 3; i++)
4355 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4356 	mutex_unlock(&pit->pit_state.lock);
4357 	return 0;
4358 }
4359 
4360 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4361 {
4362 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4363 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4364 		sizeof(ps->channels));
4365 	ps->flags = kvm->arch.vpit->pit_state.flags;
4366 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4367 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4368 	return 0;
4369 }
4370 
4371 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4372 {
4373 	int start = 0;
4374 	int i;
4375 	u32 prev_legacy, cur_legacy;
4376 	struct kvm_pit *pit = kvm->arch.vpit;
4377 
4378 	mutex_lock(&pit->pit_state.lock);
4379 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4380 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4381 	if (!prev_legacy && cur_legacy)
4382 		start = 1;
4383 	memcpy(&pit->pit_state.channels, &ps->channels,
4384 	       sizeof(pit->pit_state.channels));
4385 	pit->pit_state.flags = ps->flags;
4386 	for (i = 0; i < 3; i++)
4387 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4388 				   start && i == 0);
4389 	mutex_unlock(&pit->pit_state.lock);
4390 	return 0;
4391 }
4392 
4393 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4394 				 struct kvm_reinject_control *control)
4395 {
4396 	struct kvm_pit *pit = kvm->arch.vpit;
4397 
4398 	if (!pit)
4399 		return -ENXIO;
4400 
4401 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4402 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4403 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4404 	 */
4405 	mutex_lock(&pit->pit_state.lock);
4406 	kvm_pit_set_reinject(pit, control->pit_reinject);
4407 	mutex_unlock(&pit->pit_state.lock);
4408 
4409 	return 0;
4410 }
4411 
4412 /**
4413  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4414  * @kvm: kvm instance
4415  * @log: slot id and address to which we copy the log
4416  *
4417  * Steps 1-4 below provide general overview of dirty page logging. See
4418  * kvm_get_dirty_log_protect() function description for additional details.
4419  *
4420  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4421  * always flush the TLB (step 4) even if previous step failed  and the dirty
4422  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4423  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4424  * writes will be marked dirty for next log read.
4425  *
4426  *   1. Take a snapshot of the bit and clear it if needed.
4427  *   2. Write protect the corresponding page.
4428  *   3. Copy the snapshot to the userspace.
4429  *   4. Flush TLB's if needed.
4430  */
4431 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4432 {
4433 	bool flush = false;
4434 	int r;
4435 
4436 	mutex_lock(&kvm->slots_lock);
4437 
4438 	/*
4439 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4440 	 */
4441 	if (kvm_x86_ops->flush_log_dirty)
4442 		kvm_x86_ops->flush_log_dirty(kvm);
4443 
4444 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4445 
4446 	/*
4447 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4448 	 * kvm_mmu_slot_remove_write_access().
4449 	 */
4450 	lockdep_assert_held(&kvm->slots_lock);
4451 	if (flush)
4452 		kvm_flush_remote_tlbs(kvm);
4453 
4454 	mutex_unlock(&kvm->slots_lock);
4455 	return r;
4456 }
4457 
4458 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4459 {
4460 	bool flush = false;
4461 	int r;
4462 
4463 	mutex_lock(&kvm->slots_lock);
4464 
4465 	/*
4466 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4467 	 */
4468 	if (kvm_x86_ops->flush_log_dirty)
4469 		kvm_x86_ops->flush_log_dirty(kvm);
4470 
4471 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4472 
4473 	/*
4474 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4475 	 * kvm_mmu_slot_remove_write_access().
4476 	 */
4477 	lockdep_assert_held(&kvm->slots_lock);
4478 	if (flush)
4479 		kvm_flush_remote_tlbs(kvm);
4480 
4481 	mutex_unlock(&kvm->slots_lock);
4482 	return r;
4483 }
4484 
4485 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4486 			bool line_status)
4487 {
4488 	if (!irqchip_in_kernel(kvm))
4489 		return -ENXIO;
4490 
4491 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4492 					irq_event->irq, irq_event->level,
4493 					line_status);
4494 	return 0;
4495 }
4496 
4497 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4498 			    struct kvm_enable_cap *cap)
4499 {
4500 	int r;
4501 
4502 	if (cap->flags)
4503 		return -EINVAL;
4504 
4505 	switch (cap->cap) {
4506 	case KVM_CAP_DISABLE_QUIRKS:
4507 		kvm->arch.disabled_quirks = cap->args[0];
4508 		r = 0;
4509 		break;
4510 	case KVM_CAP_SPLIT_IRQCHIP: {
4511 		mutex_lock(&kvm->lock);
4512 		r = -EINVAL;
4513 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4514 			goto split_irqchip_unlock;
4515 		r = -EEXIST;
4516 		if (irqchip_in_kernel(kvm))
4517 			goto split_irqchip_unlock;
4518 		if (kvm->created_vcpus)
4519 			goto split_irqchip_unlock;
4520 		r = kvm_setup_empty_irq_routing(kvm);
4521 		if (r)
4522 			goto split_irqchip_unlock;
4523 		/* Pairs with irqchip_in_kernel. */
4524 		smp_wmb();
4525 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4526 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4527 		r = 0;
4528 split_irqchip_unlock:
4529 		mutex_unlock(&kvm->lock);
4530 		break;
4531 	}
4532 	case KVM_CAP_X2APIC_API:
4533 		r = -EINVAL;
4534 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4535 			break;
4536 
4537 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4538 			kvm->arch.x2apic_format = true;
4539 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4540 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4541 
4542 		r = 0;
4543 		break;
4544 	case KVM_CAP_X86_DISABLE_EXITS:
4545 		r = -EINVAL;
4546 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4547 			break;
4548 
4549 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4550 			kvm_can_mwait_in_guest())
4551 			kvm->arch.mwait_in_guest = true;
4552 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4553 			kvm->arch.hlt_in_guest = true;
4554 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4555 			kvm->arch.pause_in_guest = true;
4556 		r = 0;
4557 		break;
4558 	case KVM_CAP_MSR_PLATFORM_INFO:
4559 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4560 		r = 0;
4561 		break;
4562 	case KVM_CAP_EXCEPTION_PAYLOAD:
4563 		kvm->arch.exception_payload_enabled = cap->args[0];
4564 		r = 0;
4565 		break;
4566 	default:
4567 		r = -EINVAL;
4568 		break;
4569 	}
4570 	return r;
4571 }
4572 
4573 long kvm_arch_vm_ioctl(struct file *filp,
4574 		       unsigned int ioctl, unsigned long arg)
4575 {
4576 	struct kvm *kvm = filp->private_data;
4577 	void __user *argp = (void __user *)arg;
4578 	int r = -ENOTTY;
4579 	/*
4580 	 * This union makes it completely explicit to gcc-3.x
4581 	 * that these two variables' stack usage should be
4582 	 * combined, not added together.
4583 	 */
4584 	union {
4585 		struct kvm_pit_state ps;
4586 		struct kvm_pit_state2 ps2;
4587 		struct kvm_pit_config pit_config;
4588 	} u;
4589 
4590 	switch (ioctl) {
4591 	case KVM_SET_TSS_ADDR:
4592 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4593 		break;
4594 	case KVM_SET_IDENTITY_MAP_ADDR: {
4595 		u64 ident_addr;
4596 
4597 		mutex_lock(&kvm->lock);
4598 		r = -EINVAL;
4599 		if (kvm->created_vcpus)
4600 			goto set_identity_unlock;
4601 		r = -EFAULT;
4602 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4603 			goto set_identity_unlock;
4604 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4605 set_identity_unlock:
4606 		mutex_unlock(&kvm->lock);
4607 		break;
4608 	}
4609 	case KVM_SET_NR_MMU_PAGES:
4610 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4611 		break;
4612 	case KVM_GET_NR_MMU_PAGES:
4613 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4614 		break;
4615 	case KVM_CREATE_IRQCHIP: {
4616 		mutex_lock(&kvm->lock);
4617 
4618 		r = -EEXIST;
4619 		if (irqchip_in_kernel(kvm))
4620 			goto create_irqchip_unlock;
4621 
4622 		r = -EINVAL;
4623 		if (kvm->created_vcpus)
4624 			goto create_irqchip_unlock;
4625 
4626 		r = kvm_pic_init(kvm);
4627 		if (r)
4628 			goto create_irqchip_unlock;
4629 
4630 		r = kvm_ioapic_init(kvm);
4631 		if (r) {
4632 			kvm_pic_destroy(kvm);
4633 			goto create_irqchip_unlock;
4634 		}
4635 
4636 		r = kvm_setup_default_irq_routing(kvm);
4637 		if (r) {
4638 			kvm_ioapic_destroy(kvm);
4639 			kvm_pic_destroy(kvm);
4640 			goto create_irqchip_unlock;
4641 		}
4642 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4643 		smp_wmb();
4644 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4645 	create_irqchip_unlock:
4646 		mutex_unlock(&kvm->lock);
4647 		break;
4648 	}
4649 	case KVM_CREATE_PIT:
4650 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4651 		goto create_pit;
4652 	case KVM_CREATE_PIT2:
4653 		r = -EFAULT;
4654 		if (copy_from_user(&u.pit_config, argp,
4655 				   sizeof(struct kvm_pit_config)))
4656 			goto out;
4657 	create_pit:
4658 		mutex_lock(&kvm->lock);
4659 		r = -EEXIST;
4660 		if (kvm->arch.vpit)
4661 			goto create_pit_unlock;
4662 		r = -ENOMEM;
4663 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4664 		if (kvm->arch.vpit)
4665 			r = 0;
4666 	create_pit_unlock:
4667 		mutex_unlock(&kvm->lock);
4668 		break;
4669 	case KVM_GET_IRQCHIP: {
4670 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4671 		struct kvm_irqchip *chip;
4672 
4673 		chip = memdup_user(argp, sizeof(*chip));
4674 		if (IS_ERR(chip)) {
4675 			r = PTR_ERR(chip);
4676 			goto out;
4677 		}
4678 
4679 		r = -ENXIO;
4680 		if (!irqchip_kernel(kvm))
4681 			goto get_irqchip_out;
4682 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4683 		if (r)
4684 			goto get_irqchip_out;
4685 		r = -EFAULT;
4686 		if (copy_to_user(argp, chip, sizeof(*chip)))
4687 			goto get_irqchip_out;
4688 		r = 0;
4689 	get_irqchip_out:
4690 		kfree(chip);
4691 		break;
4692 	}
4693 	case KVM_SET_IRQCHIP: {
4694 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4695 		struct kvm_irqchip *chip;
4696 
4697 		chip = memdup_user(argp, sizeof(*chip));
4698 		if (IS_ERR(chip)) {
4699 			r = PTR_ERR(chip);
4700 			goto out;
4701 		}
4702 
4703 		r = -ENXIO;
4704 		if (!irqchip_kernel(kvm))
4705 			goto set_irqchip_out;
4706 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4707 		if (r)
4708 			goto set_irqchip_out;
4709 		r = 0;
4710 	set_irqchip_out:
4711 		kfree(chip);
4712 		break;
4713 	}
4714 	case KVM_GET_PIT: {
4715 		r = -EFAULT;
4716 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4717 			goto out;
4718 		r = -ENXIO;
4719 		if (!kvm->arch.vpit)
4720 			goto out;
4721 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4722 		if (r)
4723 			goto out;
4724 		r = -EFAULT;
4725 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4726 			goto out;
4727 		r = 0;
4728 		break;
4729 	}
4730 	case KVM_SET_PIT: {
4731 		r = -EFAULT;
4732 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4733 			goto out;
4734 		r = -ENXIO;
4735 		if (!kvm->arch.vpit)
4736 			goto out;
4737 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4738 		break;
4739 	}
4740 	case KVM_GET_PIT2: {
4741 		r = -ENXIO;
4742 		if (!kvm->arch.vpit)
4743 			goto out;
4744 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4745 		if (r)
4746 			goto out;
4747 		r = -EFAULT;
4748 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4749 			goto out;
4750 		r = 0;
4751 		break;
4752 	}
4753 	case KVM_SET_PIT2: {
4754 		r = -EFAULT;
4755 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4756 			goto out;
4757 		r = -ENXIO;
4758 		if (!kvm->arch.vpit)
4759 			goto out;
4760 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4761 		break;
4762 	}
4763 	case KVM_REINJECT_CONTROL: {
4764 		struct kvm_reinject_control control;
4765 		r =  -EFAULT;
4766 		if (copy_from_user(&control, argp, sizeof(control)))
4767 			goto out;
4768 		r = kvm_vm_ioctl_reinject(kvm, &control);
4769 		break;
4770 	}
4771 	case KVM_SET_BOOT_CPU_ID:
4772 		r = 0;
4773 		mutex_lock(&kvm->lock);
4774 		if (kvm->created_vcpus)
4775 			r = -EBUSY;
4776 		else
4777 			kvm->arch.bsp_vcpu_id = arg;
4778 		mutex_unlock(&kvm->lock);
4779 		break;
4780 	case KVM_XEN_HVM_CONFIG: {
4781 		struct kvm_xen_hvm_config xhc;
4782 		r = -EFAULT;
4783 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4784 			goto out;
4785 		r = -EINVAL;
4786 		if (xhc.flags)
4787 			goto out;
4788 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4789 		r = 0;
4790 		break;
4791 	}
4792 	case KVM_SET_CLOCK: {
4793 		struct kvm_clock_data user_ns;
4794 		u64 now_ns;
4795 
4796 		r = -EFAULT;
4797 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4798 			goto out;
4799 
4800 		r = -EINVAL;
4801 		if (user_ns.flags)
4802 			goto out;
4803 
4804 		r = 0;
4805 		/*
4806 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4807 		 * kvm_gen_update_masterclock() can be cut down to locked
4808 		 * pvclock_update_vm_gtod_copy().
4809 		 */
4810 		kvm_gen_update_masterclock(kvm);
4811 		now_ns = get_kvmclock_ns(kvm);
4812 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4813 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4814 		break;
4815 	}
4816 	case KVM_GET_CLOCK: {
4817 		struct kvm_clock_data user_ns;
4818 		u64 now_ns;
4819 
4820 		now_ns = get_kvmclock_ns(kvm);
4821 		user_ns.clock = now_ns;
4822 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4823 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4824 
4825 		r = -EFAULT;
4826 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4827 			goto out;
4828 		r = 0;
4829 		break;
4830 	}
4831 	case KVM_MEMORY_ENCRYPT_OP: {
4832 		r = -ENOTTY;
4833 		if (kvm_x86_ops->mem_enc_op)
4834 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4835 		break;
4836 	}
4837 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4838 		struct kvm_enc_region region;
4839 
4840 		r = -EFAULT;
4841 		if (copy_from_user(&region, argp, sizeof(region)))
4842 			goto out;
4843 
4844 		r = -ENOTTY;
4845 		if (kvm_x86_ops->mem_enc_reg_region)
4846 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4847 		break;
4848 	}
4849 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4850 		struct kvm_enc_region region;
4851 
4852 		r = -EFAULT;
4853 		if (copy_from_user(&region, argp, sizeof(region)))
4854 			goto out;
4855 
4856 		r = -ENOTTY;
4857 		if (kvm_x86_ops->mem_enc_unreg_region)
4858 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4859 		break;
4860 	}
4861 	case KVM_HYPERV_EVENTFD: {
4862 		struct kvm_hyperv_eventfd hvevfd;
4863 
4864 		r = -EFAULT;
4865 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4866 			goto out;
4867 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4868 		break;
4869 	}
4870 	default:
4871 		r = -ENOTTY;
4872 	}
4873 out:
4874 	return r;
4875 }
4876 
4877 static void kvm_init_msr_list(void)
4878 {
4879 	u32 dummy[2];
4880 	unsigned i, j;
4881 
4882 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4883 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4884 			continue;
4885 
4886 		/*
4887 		 * Even MSRs that are valid in the host may not be exposed
4888 		 * to the guests in some cases.
4889 		 */
4890 		switch (msrs_to_save[i]) {
4891 		case MSR_IA32_BNDCFGS:
4892 			if (!kvm_mpx_supported())
4893 				continue;
4894 			break;
4895 		case MSR_TSC_AUX:
4896 			if (!kvm_x86_ops->rdtscp_supported())
4897 				continue;
4898 			break;
4899 		case MSR_IA32_RTIT_CTL:
4900 		case MSR_IA32_RTIT_STATUS:
4901 			if (!kvm_x86_ops->pt_supported())
4902 				continue;
4903 			break;
4904 		case MSR_IA32_RTIT_CR3_MATCH:
4905 			if (!kvm_x86_ops->pt_supported() ||
4906 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
4907 				continue;
4908 			break;
4909 		case MSR_IA32_RTIT_OUTPUT_BASE:
4910 		case MSR_IA32_RTIT_OUTPUT_MASK:
4911 			if (!kvm_x86_ops->pt_supported() ||
4912 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
4913 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
4914 				continue;
4915 			break;
4916 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
4917 			if (!kvm_x86_ops->pt_supported() ||
4918 				msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
4919 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
4920 				continue;
4921 			break;
4922 		}
4923 		default:
4924 			break;
4925 		}
4926 
4927 		if (j < i)
4928 			msrs_to_save[j] = msrs_to_save[i];
4929 		j++;
4930 	}
4931 	num_msrs_to_save = j;
4932 
4933 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4934 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4935 			continue;
4936 
4937 		if (j < i)
4938 			emulated_msrs[j] = emulated_msrs[i];
4939 		j++;
4940 	}
4941 	num_emulated_msrs = j;
4942 
4943 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4944 		struct kvm_msr_entry msr;
4945 
4946 		msr.index = msr_based_features[i];
4947 		if (kvm_get_msr_feature(&msr))
4948 			continue;
4949 
4950 		if (j < i)
4951 			msr_based_features[j] = msr_based_features[i];
4952 		j++;
4953 	}
4954 	num_msr_based_features = j;
4955 }
4956 
4957 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4958 			   const void *v)
4959 {
4960 	int handled = 0;
4961 	int n;
4962 
4963 	do {
4964 		n = min(len, 8);
4965 		if (!(lapic_in_kernel(vcpu) &&
4966 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4967 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4968 			break;
4969 		handled += n;
4970 		addr += n;
4971 		len -= n;
4972 		v += n;
4973 	} while (len);
4974 
4975 	return handled;
4976 }
4977 
4978 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4979 {
4980 	int handled = 0;
4981 	int n;
4982 
4983 	do {
4984 		n = min(len, 8);
4985 		if (!(lapic_in_kernel(vcpu) &&
4986 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4987 					 addr, n, v))
4988 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4989 			break;
4990 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4991 		handled += n;
4992 		addr += n;
4993 		len -= n;
4994 		v += n;
4995 	} while (len);
4996 
4997 	return handled;
4998 }
4999 
5000 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5001 			struct kvm_segment *var, int seg)
5002 {
5003 	kvm_x86_ops->set_segment(vcpu, var, seg);
5004 }
5005 
5006 void kvm_get_segment(struct kvm_vcpu *vcpu,
5007 		     struct kvm_segment *var, int seg)
5008 {
5009 	kvm_x86_ops->get_segment(vcpu, var, seg);
5010 }
5011 
5012 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5013 			   struct x86_exception *exception)
5014 {
5015 	gpa_t t_gpa;
5016 
5017 	BUG_ON(!mmu_is_nested(vcpu));
5018 
5019 	/* NPT walks are always user-walks */
5020 	access |= PFERR_USER_MASK;
5021 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5022 
5023 	return t_gpa;
5024 }
5025 
5026 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5027 			      struct x86_exception *exception)
5028 {
5029 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5030 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5031 }
5032 
5033  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5034 				struct x86_exception *exception)
5035 {
5036 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5037 	access |= PFERR_FETCH_MASK;
5038 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5039 }
5040 
5041 gpa_t kvm_mmu_gva_to_gpa_write(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 	access |= PFERR_WRITE_MASK;
5046 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5047 }
5048 
5049 /* uses this to access any guest's mapped memory without checking CPL */
5050 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5051 				struct x86_exception *exception)
5052 {
5053 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5054 }
5055 
5056 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5057 				      struct kvm_vcpu *vcpu, u32 access,
5058 				      struct x86_exception *exception)
5059 {
5060 	void *data = val;
5061 	int r = X86EMUL_CONTINUE;
5062 
5063 	while (bytes) {
5064 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5065 							    exception);
5066 		unsigned offset = addr & (PAGE_SIZE-1);
5067 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5068 		int ret;
5069 
5070 		if (gpa == UNMAPPED_GVA)
5071 			return X86EMUL_PROPAGATE_FAULT;
5072 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5073 					       offset, toread);
5074 		if (ret < 0) {
5075 			r = X86EMUL_IO_NEEDED;
5076 			goto out;
5077 		}
5078 
5079 		bytes -= toread;
5080 		data += toread;
5081 		addr += toread;
5082 	}
5083 out:
5084 	return r;
5085 }
5086 
5087 /* used for instruction fetching */
5088 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5089 				gva_t addr, void *val, unsigned int bytes,
5090 				struct x86_exception *exception)
5091 {
5092 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5093 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5094 	unsigned offset;
5095 	int ret;
5096 
5097 	/* Inline kvm_read_guest_virt_helper for speed.  */
5098 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5099 						    exception);
5100 	if (unlikely(gpa == UNMAPPED_GVA))
5101 		return X86EMUL_PROPAGATE_FAULT;
5102 
5103 	offset = addr & (PAGE_SIZE-1);
5104 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5105 		bytes = (unsigned)PAGE_SIZE - offset;
5106 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5107 				       offset, bytes);
5108 	if (unlikely(ret < 0))
5109 		return X86EMUL_IO_NEEDED;
5110 
5111 	return X86EMUL_CONTINUE;
5112 }
5113 
5114 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5115 			       gva_t addr, void *val, unsigned int bytes,
5116 			       struct x86_exception *exception)
5117 {
5118 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5119 
5120 	/*
5121 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5122 	 * is returned, but our callers are not ready for that and they blindly
5123 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5124 	 * uninitialized kernel stack memory into cr2 and error code.
5125 	 */
5126 	memset(exception, 0, sizeof(*exception));
5127 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5128 					  exception);
5129 }
5130 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5131 
5132 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5133 			     gva_t addr, void *val, unsigned int bytes,
5134 			     struct x86_exception *exception, bool system)
5135 {
5136 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5137 	u32 access = 0;
5138 
5139 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5140 		access |= PFERR_USER_MASK;
5141 
5142 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5143 }
5144 
5145 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5146 		unsigned long addr, void *val, unsigned int bytes)
5147 {
5148 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5149 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5150 
5151 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5152 }
5153 
5154 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5155 				      struct kvm_vcpu *vcpu, u32 access,
5156 				      struct x86_exception *exception)
5157 {
5158 	void *data = val;
5159 	int r = X86EMUL_CONTINUE;
5160 
5161 	while (bytes) {
5162 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5163 							     access,
5164 							     exception);
5165 		unsigned offset = addr & (PAGE_SIZE-1);
5166 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5167 		int ret;
5168 
5169 		if (gpa == UNMAPPED_GVA)
5170 			return X86EMUL_PROPAGATE_FAULT;
5171 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5172 		if (ret < 0) {
5173 			r = X86EMUL_IO_NEEDED;
5174 			goto out;
5175 		}
5176 
5177 		bytes -= towrite;
5178 		data += towrite;
5179 		addr += towrite;
5180 	}
5181 out:
5182 	return r;
5183 }
5184 
5185 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5186 			      unsigned int bytes, struct x86_exception *exception,
5187 			      bool system)
5188 {
5189 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5190 	u32 access = PFERR_WRITE_MASK;
5191 
5192 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5193 		access |= PFERR_USER_MASK;
5194 
5195 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5196 					   access, exception);
5197 }
5198 
5199 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5200 				unsigned int bytes, struct x86_exception *exception)
5201 {
5202 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5203 	vcpu->arch.l1tf_flush_l1d = true;
5204 
5205 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5206 					   PFERR_WRITE_MASK, exception);
5207 }
5208 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5209 
5210 int handle_ud(struct kvm_vcpu *vcpu)
5211 {
5212 	int emul_type = EMULTYPE_TRAP_UD;
5213 	enum emulation_result er;
5214 	char sig[5]; /* ud2; .ascii "kvm" */
5215 	struct x86_exception e;
5216 
5217 	if (force_emulation_prefix &&
5218 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5219 				sig, sizeof(sig), &e) == 0 &&
5220 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5221 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5222 		emul_type = 0;
5223 	}
5224 
5225 	er = kvm_emulate_instruction(vcpu, emul_type);
5226 	if (er == EMULATE_USER_EXIT)
5227 		return 0;
5228 	if (er != EMULATE_DONE)
5229 		kvm_queue_exception(vcpu, UD_VECTOR);
5230 	return 1;
5231 }
5232 EXPORT_SYMBOL_GPL(handle_ud);
5233 
5234 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5235 			    gpa_t gpa, bool write)
5236 {
5237 	/* For APIC access vmexit */
5238 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5239 		return 1;
5240 
5241 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5242 		trace_vcpu_match_mmio(gva, gpa, write, true);
5243 		return 1;
5244 	}
5245 
5246 	return 0;
5247 }
5248 
5249 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5250 				gpa_t *gpa, struct x86_exception *exception,
5251 				bool write)
5252 {
5253 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5254 		| (write ? PFERR_WRITE_MASK : 0);
5255 
5256 	/*
5257 	 * currently PKRU is only applied to ept enabled guest so
5258 	 * there is no pkey in EPT page table for L1 guest or EPT
5259 	 * shadow page table for L2 guest.
5260 	 */
5261 	if (vcpu_match_mmio_gva(vcpu, gva)
5262 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5263 				 vcpu->arch.access, 0, access)) {
5264 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5265 					(gva & (PAGE_SIZE - 1));
5266 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5267 		return 1;
5268 	}
5269 
5270 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5271 
5272 	if (*gpa == UNMAPPED_GVA)
5273 		return -1;
5274 
5275 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5276 }
5277 
5278 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5279 			const void *val, int bytes)
5280 {
5281 	int ret;
5282 
5283 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5284 	if (ret < 0)
5285 		return 0;
5286 	kvm_page_track_write(vcpu, gpa, val, bytes);
5287 	return 1;
5288 }
5289 
5290 struct read_write_emulator_ops {
5291 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5292 				  int bytes);
5293 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5294 				  void *val, int bytes);
5295 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5296 			       int bytes, void *val);
5297 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5298 				    void *val, int bytes);
5299 	bool write;
5300 };
5301 
5302 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5303 {
5304 	if (vcpu->mmio_read_completed) {
5305 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5306 			       vcpu->mmio_fragments[0].gpa, val);
5307 		vcpu->mmio_read_completed = 0;
5308 		return 1;
5309 	}
5310 
5311 	return 0;
5312 }
5313 
5314 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5315 			void *val, int bytes)
5316 {
5317 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5318 }
5319 
5320 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5321 			 void *val, int bytes)
5322 {
5323 	return emulator_write_phys(vcpu, gpa, val, bytes);
5324 }
5325 
5326 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5327 {
5328 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5329 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5330 }
5331 
5332 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5333 			  void *val, int bytes)
5334 {
5335 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5336 	return X86EMUL_IO_NEEDED;
5337 }
5338 
5339 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5340 			   void *val, int bytes)
5341 {
5342 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5343 
5344 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5345 	return X86EMUL_CONTINUE;
5346 }
5347 
5348 static const struct read_write_emulator_ops read_emultor = {
5349 	.read_write_prepare = read_prepare,
5350 	.read_write_emulate = read_emulate,
5351 	.read_write_mmio = vcpu_mmio_read,
5352 	.read_write_exit_mmio = read_exit_mmio,
5353 };
5354 
5355 static const struct read_write_emulator_ops write_emultor = {
5356 	.read_write_emulate = write_emulate,
5357 	.read_write_mmio = write_mmio,
5358 	.read_write_exit_mmio = write_exit_mmio,
5359 	.write = true,
5360 };
5361 
5362 static int emulator_read_write_onepage(unsigned long addr, void *val,
5363 				       unsigned int bytes,
5364 				       struct x86_exception *exception,
5365 				       struct kvm_vcpu *vcpu,
5366 				       const struct read_write_emulator_ops *ops)
5367 {
5368 	gpa_t gpa;
5369 	int handled, ret;
5370 	bool write = ops->write;
5371 	struct kvm_mmio_fragment *frag;
5372 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5373 
5374 	/*
5375 	 * If the exit was due to a NPF we may already have a GPA.
5376 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5377 	 * Note, this cannot be used on string operations since string
5378 	 * operation using rep will only have the initial GPA from the NPF
5379 	 * occurred.
5380 	 */
5381 	if (vcpu->arch.gpa_available &&
5382 	    emulator_can_use_gpa(ctxt) &&
5383 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5384 		gpa = vcpu->arch.gpa_val;
5385 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5386 	} else {
5387 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5388 		if (ret < 0)
5389 			return X86EMUL_PROPAGATE_FAULT;
5390 	}
5391 
5392 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5393 		return X86EMUL_CONTINUE;
5394 
5395 	/*
5396 	 * Is this MMIO handled locally?
5397 	 */
5398 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5399 	if (handled == bytes)
5400 		return X86EMUL_CONTINUE;
5401 
5402 	gpa += handled;
5403 	bytes -= handled;
5404 	val += handled;
5405 
5406 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5407 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5408 	frag->gpa = gpa;
5409 	frag->data = val;
5410 	frag->len = bytes;
5411 	return X86EMUL_CONTINUE;
5412 }
5413 
5414 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5415 			unsigned long addr,
5416 			void *val, unsigned int bytes,
5417 			struct x86_exception *exception,
5418 			const struct read_write_emulator_ops *ops)
5419 {
5420 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5421 	gpa_t gpa;
5422 	int rc;
5423 
5424 	if (ops->read_write_prepare &&
5425 		  ops->read_write_prepare(vcpu, val, bytes))
5426 		return X86EMUL_CONTINUE;
5427 
5428 	vcpu->mmio_nr_fragments = 0;
5429 
5430 	/* Crossing a page boundary? */
5431 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5432 		int now;
5433 
5434 		now = -addr & ~PAGE_MASK;
5435 		rc = emulator_read_write_onepage(addr, val, now, exception,
5436 						 vcpu, ops);
5437 
5438 		if (rc != X86EMUL_CONTINUE)
5439 			return rc;
5440 		addr += now;
5441 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5442 			addr = (u32)addr;
5443 		val += now;
5444 		bytes -= now;
5445 	}
5446 
5447 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5448 					 vcpu, ops);
5449 	if (rc != X86EMUL_CONTINUE)
5450 		return rc;
5451 
5452 	if (!vcpu->mmio_nr_fragments)
5453 		return rc;
5454 
5455 	gpa = vcpu->mmio_fragments[0].gpa;
5456 
5457 	vcpu->mmio_needed = 1;
5458 	vcpu->mmio_cur_fragment = 0;
5459 
5460 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5461 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5462 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5463 	vcpu->run->mmio.phys_addr = gpa;
5464 
5465 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5466 }
5467 
5468 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5469 				  unsigned long addr,
5470 				  void *val,
5471 				  unsigned int bytes,
5472 				  struct x86_exception *exception)
5473 {
5474 	return emulator_read_write(ctxt, addr, val, bytes,
5475 				   exception, &read_emultor);
5476 }
5477 
5478 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5479 			    unsigned long addr,
5480 			    const void *val,
5481 			    unsigned int bytes,
5482 			    struct x86_exception *exception)
5483 {
5484 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5485 				   exception, &write_emultor);
5486 }
5487 
5488 #define CMPXCHG_TYPE(t, ptr, old, new) \
5489 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5490 
5491 #ifdef CONFIG_X86_64
5492 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5493 #else
5494 #  define CMPXCHG64(ptr, old, new) \
5495 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5496 #endif
5497 
5498 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5499 				     unsigned long addr,
5500 				     const void *old,
5501 				     const void *new,
5502 				     unsigned int bytes,
5503 				     struct x86_exception *exception)
5504 {
5505 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5506 	gpa_t gpa;
5507 	struct page *page;
5508 	char *kaddr;
5509 	bool exchanged;
5510 
5511 	/* guests cmpxchg8b have to be emulated atomically */
5512 	if (bytes > 8 || (bytes & (bytes - 1)))
5513 		goto emul_write;
5514 
5515 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5516 
5517 	if (gpa == UNMAPPED_GVA ||
5518 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5519 		goto emul_write;
5520 
5521 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5522 		goto emul_write;
5523 
5524 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5525 	if (is_error_page(page))
5526 		goto emul_write;
5527 
5528 	kaddr = kmap_atomic(page);
5529 	kaddr += offset_in_page(gpa);
5530 	switch (bytes) {
5531 	case 1:
5532 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5533 		break;
5534 	case 2:
5535 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5536 		break;
5537 	case 4:
5538 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5539 		break;
5540 	case 8:
5541 		exchanged = CMPXCHG64(kaddr, old, new);
5542 		break;
5543 	default:
5544 		BUG();
5545 	}
5546 	kunmap_atomic(kaddr);
5547 	kvm_release_page_dirty(page);
5548 
5549 	if (!exchanged)
5550 		return X86EMUL_CMPXCHG_FAILED;
5551 
5552 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5553 	kvm_page_track_write(vcpu, gpa, new, bytes);
5554 
5555 	return X86EMUL_CONTINUE;
5556 
5557 emul_write:
5558 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5559 
5560 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5561 }
5562 
5563 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5564 {
5565 	int r = 0, i;
5566 
5567 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5568 		if (vcpu->arch.pio.in)
5569 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5570 					    vcpu->arch.pio.size, pd);
5571 		else
5572 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5573 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5574 					     pd);
5575 		if (r)
5576 			break;
5577 		pd += vcpu->arch.pio.size;
5578 	}
5579 	return r;
5580 }
5581 
5582 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5583 			       unsigned short port, void *val,
5584 			       unsigned int count, bool in)
5585 {
5586 	vcpu->arch.pio.port = port;
5587 	vcpu->arch.pio.in = in;
5588 	vcpu->arch.pio.count  = count;
5589 	vcpu->arch.pio.size = size;
5590 
5591 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5592 		vcpu->arch.pio.count = 0;
5593 		return 1;
5594 	}
5595 
5596 	vcpu->run->exit_reason = KVM_EXIT_IO;
5597 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5598 	vcpu->run->io.size = size;
5599 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5600 	vcpu->run->io.count = count;
5601 	vcpu->run->io.port = port;
5602 
5603 	return 0;
5604 }
5605 
5606 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5607 				    int size, unsigned short port, void *val,
5608 				    unsigned int count)
5609 {
5610 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5611 	int ret;
5612 
5613 	if (vcpu->arch.pio.count)
5614 		goto data_avail;
5615 
5616 	memset(vcpu->arch.pio_data, 0, size * count);
5617 
5618 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5619 	if (ret) {
5620 data_avail:
5621 		memcpy(val, vcpu->arch.pio_data, size * count);
5622 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5623 		vcpu->arch.pio.count = 0;
5624 		return 1;
5625 	}
5626 
5627 	return 0;
5628 }
5629 
5630 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5631 				     int size, unsigned short port,
5632 				     const void *val, unsigned int count)
5633 {
5634 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5635 
5636 	memcpy(vcpu->arch.pio_data, val, size * count);
5637 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5638 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5639 }
5640 
5641 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5642 {
5643 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5644 }
5645 
5646 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5647 {
5648 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5649 }
5650 
5651 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5652 {
5653 	if (!need_emulate_wbinvd(vcpu))
5654 		return X86EMUL_CONTINUE;
5655 
5656 	if (kvm_x86_ops->has_wbinvd_exit()) {
5657 		int cpu = get_cpu();
5658 
5659 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5660 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5661 				wbinvd_ipi, NULL, 1);
5662 		put_cpu();
5663 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5664 	} else
5665 		wbinvd();
5666 	return X86EMUL_CONTINUE;
5667 }
5668 
5669 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5670 {
5671 	kvm_emulate_wbinvd_noskip(vcpu);
5672 	return kvm_skip_emulated_instruction(vcpu);
5673 }
5674 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5675 
5676 
5677 
5678 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5679 {
5680 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5681 }
5682 
5683 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5684 			   unsigned long *dest)
5685 {
5686 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5687 }
5688 
5689 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5690 			   unsigned long value)
5691 {
5692 
5693 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5694 }
5695 
5696 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5697 {
5698 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5699 }
5700 
5701 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5702 {
5703 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5704 	unsigned long value;
5705 
5706 	switch (cr) {
5707 	case 0:
5708 		value = kvm_read_cr0(vcpu);
5709 		break;
5710 	case 2:
5711 		value = vcpu->arch.cr2;
5712 		break;
5713 	case 3:
5714 		value = kvm_read_cr3(vcpu);
5715 		break;
5716 	case 4:
5717 		value = kvm_read_cr4(vcpu);
5718 		break;
5719 	case 8:
5720 		value = kvm_get_cr8(vcpu);
5721 		break;
5722 	default:
5723 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5724 		return 0;
5725 	}
5726 
5727 	return value;
5728 }
5729 
5730 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5731 {
5732 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5733 	int res = 0;
5734 
5735 	switch (cr) {
5736 	case 0:
5737 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5738 		break;
5739 	case 2:
5740 		vcpu->arch.cr2 = val;
5741 		break;
5742 	case 3:
5743 		res = kvm_set_cr3(vcpu, val);
5744 		break;
5745 	case 4:
5746 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5747 		break;
5748 	case 8:
5749 		res = kvm_set_cr8(vcpu, val);
5750 		break;
5751 	default:
5752 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5753 		res = -1;
5754 	}
5755 
5756 	return res;
5757 }
5758 
5759 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5760 {
5761 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5762 }
5763 
5764 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5765 {
5766 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5767 }
5768 
5769 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5770 {
5771 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5772 }
5773 
5774 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5775 {
5776 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5777 }
5778 
5779 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5780 {
5781 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5782 }
5783 
5784 static unsigned long emulator_get_cached_segment_base(
5785 	struct x86_emulate_ctxt *ctxt, int seg)
5786 {
5787 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5788 }
5789 
5790 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5791 				 struct desc_struct *desc, u32 *base3,
5792 				 int seg)
5793 {
5794 	struct kvm_segment var;
5795 
5796 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5797 	*selector = var.selector;
5798 
5799 	if (var.unusable) {
5800 		memset(desc, 0, sizeof(*desc));
5801 		if (base3)
5802 			*base3 = 0;
5803 		return false;
5804 	}
5805 
5806 	if (var.g)
5807 		var.limit >>= 12;
5808 	set_desc_limit(desc, var.limit);
5809 	set_desc_base(desc, (unsigned long)var.base);
5810 #ifdef CONFIG_X86_64
5811 	if (base3)
5812 		*base3 = var.base >> 32;
5813 #endif
5814 	desc->type = var.type;
5815 	desc->s = var.s;
5816 	desc->dpl = var.dpl;
5817 	desc->p = var.present;
5818 	desc->avl = var.avl;
5819 	desc->l = var.l;
5820 	desc->d = var.db;
5821 	desc->g = var.g;
5822 
5823 	return true;
5824 }
5825 
5826 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5827 				 struct desc_struct *desc, u32 base3,
5828 				 int seg)
5829 {
5830 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5831 	struct kvm_segment var;
5832 
5833 	var.selector = selector;
5834 	var.base = get_desc_base(desc);
5835 #ifdef CONFIG_X86_64
5836 	var.base |= ((u64)base3) << 32;
5837 #endif
5838 	var.limit = get_desc_limit(desc);
5839 	if (desc->g)
5840 		var.limit = (var.limit << 12) | 0xfff;
5841 	var.type = desc->type;
5842 	var.dpl = desc->dpl;
5843 	var.db = desc->d;
5844 	var.s = desc->s;
5845 	var.l = desc->l;
5846 	var.g = desc->g;
5847 	var.avl = desc->avl;
5848 	var.present = desc->p;
5849 	var.unusable = !var.present;
5850 	var.padding = 0;
5851 
5852 	kvm_set_segment(vcpu, &var, seg);
5853 	return;
5854 }
5855 
5856 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5857 			    u32 msr_index, u64 *pdata)
5858 {
5859 	struct msr_data msr;
5860 	int r;
5861 
5862 	msr.index = msr_index;
5863 	msr.host_initiated = false;
5864 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5865 	if (r)
5866 		return r;
5867 
5868 	*pdata = msr.data;
5869 	return 0;
5870 }
5871 
5872 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5873 			    u32 msr_index, u64 data)
5874 {
5875 	struct msr_data msr;
5876 
5877 	msr.data = data;
5878 	msr.index = msr_index;
5879 	msr.host_initiated = false;
5880 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5881 }
5882 
5883 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5884 {
5885 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5886 
5887 	return vcpu->arch.smbase;
5888 }
5889 
5890 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5891 {
5892 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5893 
5894 	vcpu->arch.smbase = smbase;
5895 }
5896 
5897 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5898 			      u32 pmc)
5899 {
5900 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5901 }
5902 
5903 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5904 			     u32 pmc, u64 *pdata)
5905 {
5906 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5907 }
5908 
5909 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5910 {
5911 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5912 }
5913 
5914 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5915 			      struct x86_instruction_info *info,
5916 			      enum x86_intercept_stage stage)
5917 {
5918 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5919 }
5920 
5921 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5922 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5923 {
5924 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5925 }
5926 
5927 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5928 {
5929 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5930 }
5931 
5932 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5933 {
5934 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5935 }
5936 
5937 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5938 {
5939 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5940 }
5941 
5942 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5943 {
5944 	return emul_to_vcpu(ctxt)->arch.hflags;
5945 }
5946 
5947 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5948 {
5949 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5950 }
5951 
5952 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5953 {
5954 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5955 }
5956 
5957 static const struct x86_emulate_ops emulate_ops = {
5958 	.read_gpr            = emulator_read_gpr,
5959 	.write_gpr           = emulator_write_gpr,
5960 	.read_std            = emulator_read_std,
5961 	.write_std           = emulator_write_std,
5962 	.read_phys           = kvm_read_guest_phys_system,
5963 	.fetch               = kvm_fetch_guest_virt,
5964 	.read_emulated       = emulator_read_emulated,
5965 	.write_emulated      = emulator_write_emulated,
5966 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5967 	.invlpg              = emulator_invlpg,
5968 	.pio_in_emulated     = emulator_pio_in_emulated,
5969 	.pio_out_emulated    = emulator_pio_out_emulated,
5970 	.get_segment         = emulator_get_segment,
5971 	.set_segment         = emulator_set_segment,
5972 	.get_cached_segment_base = emulator_get_cached_segment_base,
5973 	.get_gdt             = emulator_get_gdt,
5974 	.get_idt	     = emulator_get_idt,
5975 	.set_gdt             = emulator_set_gdt,
5976 	.set_idt	     = emulator_set_idt,
5977 	.get_cr              = emulator_get_cr,
5978 	.set_cr              = emulator_set_cr,
5979 	.cpl                 = emulator_get_cpl,
5980 	.get_dr              = emulator_get_dr,
5981 	.set_dr              = emulator_set_dr,
5982 	.get_smbase          = emulator_get_smbase,
5983 	.set_smbase          = emulator_set_smbase,
5984 	.set_msr             = emulator_set_msr,
5985 	.get_msr             = emulator_get_msr,
5986 	.check_pmc	     = emulator_check_pmc,
5987 	.read_pmc            = emulator_read_pmc,
5988 	.halt                = emulator_halt,
5989 	.wbinvd              = emulator_wbinvd,
5990 	.fix_hypercall       = emulator_fix_hypercall,
5991 	.intercept           = emulator_intercept,
5992 	.get_cpuid           = emulator_get_cpuid,
5993 	.set_nmi_mask        = emulator_set_nmi_mask,
5994 	.get_hflags          = emulator_get_hflags,
5995 	.set_hflags          = emulator_set_hflags,
5996 	.pre_leave_smm       = emulator_pre_leave_smm,
5997 };
5998 
5999 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6000 {
6001 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6002 	/*
6003 	 * an sti; sti; sequence only disable interrupts for the first
6004 	 * instruction. So, if the last instruction, be it emulated or
6005 	 * not, left the system with the INT_STI flag enabled, it
6006 	 * means that the last instruction is an sti. We should not
6007 	 * leave the flag on in this case. The same goes for mov ss
6008 	 */
6009 	if (int_shadow & mask)
6010 		mask = 0;
6011 	if (unlikely(int_shadow || mask)) {
6012 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6013 		if (!mask)
6014 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6015 	}
6016 }
6017 
6018 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6019 {
6020 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6021 	if (ctxt->exception.vector == PF_VECTOR)
6022 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6023 
6024 	if (ctxt->exception.error_code_valid)
6025 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6026 				      ctxt->exception.error_code);
6027 	else
6028 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6029 	return false;
6030 }
6031 
6032 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6033 {
6034 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6035 	int cs_db, cs_l;
6036 
6037 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6038 
6039 	ctxt->eflags = kvm_get_rflags(vcpu);
6040 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6041 
6042 	ctxt->eip = kvm_rip_read(vcpu);
6043 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6044 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6045 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6046 		     cs_db				? X86EMUL_MODE_PROT32 :
6047 							  X86EMUL_MODE_PROT16;
6048 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6049 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6050 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6051 
6052 	init_decode_cache(ctxt);
6053 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6054 }
6055 
6056 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6057 {
6058 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6059 	int ret;
6060 
6061 	init_emulate_ctxt(vcpu);
6062 
6063 	ctxt->op_bytes = 2;
6064 	ctxt->ad_bytes = 2;
6065 	ctxt->_eip = ctxt->eip + inc_eip;
6066 	ret = emulate_int_real(ctxt, irq);
6067 
6068 	if (ret != X86EMUL_CONTINUE)
6069 		return EMULATE_FAIL;
6070 
6071 	ctxt->eip = ctxt->_eip;
6072 	kvm_rip_write(vcpu, ctxt->eip);
6073 	kvm_set_rflags(vcpu, ctxt->eflags);
6074 
6075 	return EMULATE_DONE;
6076 }
6077 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6078 
6079 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6080 {
6081 	int r = EMULATE_DONE;
6082 
6083 	++vcpu->stat.insn_emulation_fail;
6084 	trace_kvm_emulate_insn_failed(vcpu);
6085 
6086 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6087 		return EMULATE_FAIL;
6088 
6089 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6090 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6091 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6092 		vcpu->run->internal.ndata = 0;
6093 		r = EMULATE_USER_EXIT;
6094 	}
6095 
6096 	kvm_queue_exception(vcpu, UD_VECTOR);
6097 
6098 	return r;
6099 }
6100 
6101 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6102 				  bool write_fault_to_shadow_pgtable,
6103 				  int emulation_type)
6104 {
6105 	gpa_t gpa = cr2;
6106 	kvm_pfn_t pfn;
6107 
6108 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6109 		return false;
6110 
6111 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6112 		return false;
6113 
6114 	if (!vcpu->arch.mmu->direct_map) {
6115 		/*
6116 		 * Write permission should be allowed since only
6117 		 * write access need to be emulated.
6118 		 */
6119 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6120 
6121 		/*
6122 		 * If the mapping is invalid in guest, let cpu retry
6123 		 * it to generate fault.
6124 		 */
6125 		if (gpa == UNMAPPED_GVA)
6126 			return true;
6127 	}
6128 
6129 	/*
6130 	 * Do not retry the unhandleable instruction if it faults on the
6131 	 * readonly host memory, otherwise it will goto a infinite loop:
6132 	 * retry instruction -> write #PF -> emulation fail -> retry
6133 	 * instruction -> ...
6134 	 */
6135 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6136 
6137 	/*
6138 	 * If the instruction failed on the error pfn, it can not be fixed,
6139 	 * report the error to userspace.
6140 	 */
6141 	if (is_error_noslot_pfn(pfn))
6142 		return false;
6143 
6144 	kvm_release_pfn_clean(pfn);
6145 
6146 	/* The instructions are well-emulated on direct mmu. */
6147 	if (vcpu->arch.mmu->direct_map) {
6148 		unsigned int indirect_shadow_pages;
6149 
6150 		spin_lock(&vcpu->kvm->mmu_lock);
6151 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6152 		spin_unlock(&vcpu->kvm->mmu_lock);
6153 
6154 		if (indirect_shadow_pages)
6155 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6156 
6157 		return true;
6158 	}
6159 
6160 	/*
6161 	 * if emulation was due to access to shadowed page table
6162 	 * and it failed try to unshadow page and re-enter the
6163 	 * guest to let CPU execute the instruction.
6164 	 */
6165 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6166 
6167 	/*
6168 	 * If the access faults on its page table, it can not
6169 	 * be fixed by unprotecting shadow page and it should
6170 	 * be reported to userspace.
6171 	 */
6172 	return !write_fault_to_shadow_pgtable;
6173 }
6174 
6175 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6176 			      unsigned long cr2,  int emulation_type)
6177 {
6178 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6179 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6180 
6181 	last_retry_eip = vcpu->arch.last_retry_eip;
6182 	last_retry_addr = vcpu->arch.last_retry_addr;
6183 
6184 	/*
6185 	 * If the emulation is caused by #PF and it is non-page_table
6186 	 * writing instruction, it means the VM-EXIT is caused by shadow
6187 	 * page protected, we can zap the shadow page and retry this
6188 	 * instruction directly.
6189 	 *
6190 	 * Note: if the guest uses a non-page-table modifying instruction
6191 	 * on the PDE that points to the instruction, then we will unmap
6192 	 * the instruction and go to an infinite loop. So, we cache the
6193 	 * last retried eip and the last fault address, if we meet the eip
6194 	 * and the address again, we can break out of the potential infinite
6195 	 * loop.
6196 	 */
6197 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6198 
6199 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6200 		return false;
6201 
6202 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6203 		return false;
6204 
6205 	if (x86_page_table_writing_insn(ctxt))
6206 		return false;
6207 
6208 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6209 		return false;
6210 
6211 	vcpu->arch.last_retry_eip = ctxt->eip;
6212 	vcpu->arch.last_retry_addr = cr2;
6213 
6214 	if (!vcpu->arch.mmu->direct_map)
6215 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6216 
6217 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6218 
6219 	return true;
6220 }
6221 
6222 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6223 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6224 
6225 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6226 {
6227 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6228 		/* This is a good place to trace that we are exiting SMM.  */
6229 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6230 
6231 		/* Process a latched INIT or SMI, if any.  */
6232 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6233 	}
6234 
6235 	kvm_mmu_reset_context(vcpu);
6236 }
6237 
6238 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6239 {
6240 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
6241 
6242 	vcpu->arch.hflags = emul_flags;
6243 
6244 	if (changed & HF_SMM_MASK)
6245 		kvm_smm_changed(vcpu);
6246 }
6247 
6248 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6249 				unsigned long *db)
6250 {
6251 	u32 dr6 = 0;
6252 	int i;
6253 	u32 enable, rwlen;
6254 
6255 	enable = dr7;
6256 	rwlen = dr7 >> 16;
6257 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6258 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6259 			dr6 |= (1 << i);
6260 	return dr6;
6261 }
6262 
6263 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6264 {
6265 	struct kvm_run *kvm_run = vcpu->run;
6266 
6267 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6268 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6269 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6270 		kvm_run->debug.arch.exception = DB_VECTOR;
6271 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6272 		*r = EMULATE_USER_EXIT;
6273 	} else {
6274 		kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6275 	}
6276 }
6277 
6278 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6279 {
6280 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6281 	int r = EMULATE_DONE;
6282 
6283 	kvm_x86_ops->skip_emulated_instruction(vcpu);
6284 
6285 	/*
6286 	 * rflags is the old, "raw" value of the flags.  The new value has
6287 	 * not been saved yet.
6288 	 *
6289 	 * This is correct even for TF set by the guest, because "the
6290 	 * processor will not generate this exception after the instruction
6291 	 * that sets the TF flag".
6292 	 */
6293 	if (unlikely(rflags & X86_EFLAGS_TF))
6294 		kvm_vcpu_do_singlestep(vcpu, &r);
6295 	return r == EMULATE_DONE;
6296 }
6297 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6298 
6299 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6300 {
6301 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6302 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6303 		struct kvm_run *kvm_run = vcpu->run;
6304 		unsigned long eip = kvm_get_linear_rip(vcpu);
6305 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6306 					   vcpu->arch.guest_debug_dr7,
6307 					   vcpu->arch.eff_db);
6308 
6309 		if (dr6 != 0) {
6310 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6311 			kvm_run->debug.arch.pc = eip;
6312 			kvm_run->debug.arch.exception = DB_VECTOR;
6313 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6314 			*r = EMULATE_USER_EXIT;
6315 			return true;
6316 		}
6317 	}
6318 
6319 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6320 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6321 		unsigned long eip = kvm_get_linear_rip(vcpu);
6322 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6323 					   vcpu->arch.dr7,
6324 					   vcpu->arch.db);
6325 
6326 		if (dr6 != 0) {
6327 			vcpu->arch.dr6 &= ~15;
6328 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6329 			kvm_queue_exception(vcpu, DB_VECTOR);
6330 			*r = EMULATE_DONE;
6331 			return true;
6332 		}
6333 	}
6334 
6335 	return false;
6336 }
6337 
6338 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6339 {
6340 	switch (ctxt->opcode_len) {
6341 	case 1:
6342 		switch (ctxt->b) {
6343 		case 0xe4:	/* IN */
6344 		case 0xe5:
6345 		case 0xec:
6346 		case 0xed:
6347 		case 0xe6:	/* OUT */
6348 		case 0xe7:
6349 		case 0xee:
6350 		case 0xef:
6351 		case 0x6c:	/* INS */
6352 		case 0x6d:
6353 		case 0x6e:	/* OUTS */
6354 		case 0x6f:
6355 			return true;
6356 		}
6357 		break;
6358 	case 2:
6359 		switch (ctxt->b) {
6360 		case 0x33:	/* RDPMC */
6361 			return true;
6362 		}
6363 		break;
6364 	}
6365 
6366 	return false;
6367 }
6368 
6369 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6370 			    unsigned long cr2,
6371 			    int emulation_type,
6372 			    void *insn,
6373 			    int insn_len)
6374 {
6375 	int r;
6376 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6377 	bool writeback = true;
6378 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6379 
6380 	vcpu->arch.l1tf_flush_l1d = true;
6381 
6382 	/*
6383 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6384 	 * never reused.
6385 	 */
6386 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6387 	kvm_clear_exception_queue(vcpu);
6388 
6389 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6390 		init_emulate_ctxt(vcpu);
6391 
6392 		/*
6393 		 * We will reenter on the same instruction since
6394 		 * we do not set complete_userspace_io.  This does not
6395 		 * handle watchpoints yet, those would be handled in
6396 		 * the emulate_ops.
6397 		 */
6398 		if (!(emulation_type & EMULTYPE_SKIP) &&
6399 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6400 			return r;
6401 
6402 		ctxt->interruptibility = 0;
6403 		ctxt->have_exception = false;
6404 		ctxt->exception.vector = -1;
6405 		ctxt->perm_ok = false;
6406 
6407 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6408 
6409 		r = x86_decode_insn(ctxt, insn, insn_len);
6410 
6411 		trace_kvm_emulate_insn_start(vcpu);
6412 		++vcpu->stat.insn_emulation;
6413 		if (r != EMULATION_OK)  {
6414 			if (emulation_type & EMULTYPE_TRAP_UD)
6415 				return EMULATE_FAIL;
6416 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6417 						emulation_type))
6418 				return EMULATE_DONE;
6419 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6420 				return EMULATE_DONE;
6421 			if (emulation_type & EMULTYPE_SKIP)
6422 				return EMULATE_FAIL;
6423 			return handle_emulation_failure(vcpu, emulation_type);
6424 		}
6425 	}
6426 
6427 	if ((emulation_type & EMULTYPE_VMWARE) &&
6428 	    !is_vmware_backdoor_opcode(ctxt))
6429 		return EMULATE_FAIL;
6430 
6431 	if (emulation_type & EMULTYPE_SKIP) {
6432 		kvm_rip_write(vcpu, ctxt->_eip);
6433 		if (ctxt->eflags & X86_EFLAGS_RF)
6434 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6435 		return EMULATE_DONE;
6436 	}
6437 
6438 	if (retry_instruction(ctxt, cr2, emulation_type))
6439 		return EMULATE_DONE;
6440 
6441 	/* this is needed for vmware backdoor interface to work since it
6442 	   changes registers values  during IO operation */
6443 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6444 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6445 		emulator_invalidate_register_cache(ctxt);
6446 	}
6447 
6448 restart:
6449 	/* Save the faulting GPA (cr2) in the address field */
6450 	ctxt->exception.address = cr2;
6451 
6452 	r = x86_emulate_insn(ctxt);
6453 
6454 	if (r == EMULATION_INTERCEPTED)
6455 		return EMULATE_DONE;
6456 
6457 	if (r == EMULATION_FAILED) {
6458 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6459 					emulation_type))
6460 			return EMULATE_DONE;
6461 
6462 		return handle_emulation_failure(vcpu, emulation_type);
6463 	}
6464 
6465 	if (ctxt->have_exception) {
6466 		r = EMULATE_DONE;
6467 		if (inject_emulated_exception(vcpu))
6468 			return r;
6469 	} else if (vcpu->arch.pio.count) {
6470 		if (!vcpu->arch.pio.in) {
6471 			/* FIXME: return into emulator if single-stepping.  */
6472 			vcpu->arch.pio.count = 0;
6473 		} else {
6474 			writeback = false;
6475 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6476 		}
6477 		r = EMULATE_USER_EXIT;
6478 	} else if (vcpu->mmio_needed) {
6479 		if (!vcpu->mmio_is_write)
6480 			writeback = false;
6481 		r = EMULATE_USER_EXIT;
6482 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6483 	} else if (r == EMULATION_RESTART)
6484 		goto restart;
6485 	else
6486 		r = EMULATE_DONE;
6487 
6488 	if (writeback) {
6489 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6490 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6491 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6492 		kvm_rip_write(vcpu, ctxt->eip);
6493 		if (r == EMULATE_DONE && ctxt->tf)
6494 			kvm_vcpu_do_singlestep(vcpu, &r);
6495 		if (!ctxt->have_exception ||
6496 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6497 			__kvm_set_rflags(vcpu, ctxt->eflags);
6498 
6499 		/*
6500 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6501 		 * do nothing, and it will be requested again as soon as
6502 		 * the shadow expires.  But we still need to check here,
6503 		 * because POPF has no interrupt shadow.
6504 		 */
6505 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6506 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6507 	} else
6508 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6509 
6510 	return r;
6511 }
6512 
6513 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6514 {
6515 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6516 }
6517 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6518 
6519 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6520 					void *insn, int insn_len)
6521 {
6522 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6523 }
6524 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6525 
6526 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6527 			    unsigned short port)
6528 {
6529 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6530 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6531 					    size, port, &val, 1);
6532 	/* do not return to emulator after return from userspace */
6533 	vcpu->arch.pio.count = 0;
6534 	return ret;
6535 }
6536 
6537 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6538 {
6539 	unsigned long val;
6540 
6541 	/* We should only ever be called with arch.pio.count equal to 1 */
6542 	BUG_ON(vcpu->arch.pio.count != 1);
6543 
6544 	/* For size less than 4 we merge, else we zero extend */
6545 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6546 					: 0;
6547 
6548 	/*
6549 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6550 	 * the copy and tracing
6551 	 */
6552 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6553 				 vcpu->arch.pio.port, &val, 1);
6554 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6555 
6556 	return 1;
6557 }
6558 
6559 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6560 			   unsigned short port)
6561 {
6562 	unsigned long val;
6563 	int ret;
6564 
6565 	/* For size less than 4 we merge, else we zero extend */
6566 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6567 
6568 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6569 				       &val, 1);
6570 	if (ret) {
6571 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6572 		return ret;
6573 	}
6574 
6575 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6576 
6577 	return 0;
6578 }
6579 
6580 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6581 {
6582 	int ret = kvm_skip_emulated_instruction(vcpu);
6583 
6584 	/*
6585 	 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6586 	 * KVM_EXIT_DEBUG here.
6587 	 */
6588 	if (in)
6589 		return kvm_fast_pio_in(vcpu, size, port) && ret;
6590 	else
6591 		return kvm_fast_pio_out(vcpu, size, port) && ret;
6592 }
6593 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6594 
6595 static int kvmclock_cpu_down_prep(unsigned int cpu)
6596 {
6597 	__this_cpu_write(cpu_tsc_khz, 0);
6598 	return 0;
6599 }
6600 
6601 static void tsc_khz_changed(void *data)
6602 {
6603 	struct cpufreq_freqs *freq = data;
6604 	unsigned long khz = 0;
6605 
6606 	if (data)
6607 		khz = freq->new;
6608 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6609 		khz = cpufreq_quick_get(raw_smp_processor_id());
6610 	if (!khz)
6611 		khz = tsc_khz;
6612 	__this_cpu_write(cpu_tsc_khz, khz);
6613 }
6614 
6615 #ifdef CONFIG_X86_64
6616 static void kvm_hyperv_tsc_notifier(void)
6617 {
6618 	struct kvm *kvm;
6619 	struct kvm_vcpu *vcpu;
6620 	int cpu;
6621 
6622 	spin_lock(&kvm_lock);
6623 	list_for_each_entry(kvm, &vm_list, vm_list)
6624 		kvm_make_mclock_inprogress_request(kvm);
6625 
6626 	hyperv_stop_tsc_emulation();
6627 
6628 	/* TSC frequency always matches when on Hyper-V */
6629 	for_each_present_cpu(cpu)
6630 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6631 	kvm_max_guest_tsc_khz = tsc_khz;
6632 
6633 	list_for_each_entry(kvm, &vm_list, vm_list) {
6634 		struct kvm_arch *ka = &kvm->arch;
6635 
6636 		spin_lock(&ka->pvclock_gtod_sync_lock);
6637 
6638 		pvclock_update_vm_gtod_copy(kvm);
6639 
6640 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6641 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6642 
6643 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6644 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6645 
6646 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6647 	}
6648 	spin_unlock(&kvm_lock);
6649 }
6650 #endif
6651 
6652 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6653 				     void *data)
6654 {
6655 	struct cpufreq_freqs *freq = data;
6656 	struct kvm *kvm;
6657 	struct kvm_vcpu *vcpu;
6658 	int i, send_ipi = 0;
6659 
6660 	/*
6661 	 * We allow guests to temporarily run on slowing clocks,
6662 	 * provided we notify them after, or to run on accelerating
6663 	 * clocks, provided we notify them before.  Thus time never
6664 	 * goes backwards.
6665 	 *
6666 	 * However, we have a problem.  We can't atomically update
6667 	 * the frequency of a given CPU from this function; it is
6668 	 * merely a notifier, which can be called from any CPU.
6669 	 * Changing the TSC frequency at arbitrary points in time
6670 	 * requires a recomputation of local variables related to
6671 	 * the TSC for each VCPU.  We must flag these local variables
6672 	 * to be updated and be sure the update takes place with the
6673 	 * new frequency before any guests proceed.
6674 	 *
6675 	 * Unfortunately, the combination of hotplug CPU and frequency
6676 	 * change creates an intractable locking scenario; the order
6677 	 * of when these callouts happen is undefined with respect to
6678 	 * CPU hotplug, and they can race with each other.  As such,
6679 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6680 	 * undefined; you can actually have a CPU frequency change take
6681 	 * place in between the computation of X and the setting of the
6682 	 * variable.  To protect against this problem, all updates of
6683 	 * the per_cpu tsc_khz variable are done in an interrupt
6684 	 * protected IPI, and all callers wishing to update the value
6685 	 * must wait for a synchronous IPI to complete (which is trivial
6686 	 * if the caller is on the CPU already).  This establishes the
6687 	 * necessary total order on variable updates.
6688 	 *
6689 	 * Note that because a guest time update may take place
6690 	 * anytime after the setting of the VCPU's request bit, the
6691 	 * correct TSC value must be set before the request.  However,
6692 	 * to ensure the update actually makes it to any guest which
6693 	 * starts running in hardware virtualization between the set
6694 	 * and the acquisition of the spinlock, we must also ping the
6695 	 * CPU after setting the request bit.
6696 	 *
6697 	 */
6698 
6699 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6700 		return 0;
6701 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6702 		return 0;
6703 
6704 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6705 
6706 	spin_lock(&kvm_lock);
6707 	list_for_each_entry(kvm, &vm_list, vm_list) {
6708 		kvm_for_each_vcpu(i, vcpu, kvm) {
6709 			if (vcpu->cpu != freq->cpu)
6710 				continue;
6711 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6712 			if (vcpu->cpu != smp_processor_id())
6713 				send_ipi = 1;
6714 		}
6715 	}
6716 	spin_unlock(&kvm_lock);
6717 
6718 	if (freq->old < freq->new && send_ipi) {
6719 		/*
6720 		 * We upscale the frequency.  Must make the guest
6721 		 * doesn't see old kvmclock values while running with
6722 		 * the new frequency, otherwise we risk the guest sees
6723 		 * time go backwards.
6724 		 *
6725 		 * In case we update the frequency for another cpu
6726 		 * (which might be in guest context) send an interrupt
6727 		 * to kick the cpu out of guest context.  Next time
6728 		 * guest context is entered kvmclock will be updated,
6729 		 * so the guest will not see stale values.
6730 		 */
6731 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6732 	}
6733 	return 0;
6734 }
6735 
6736 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6737 	.notifier_call  = kvmclock_cpufreq_notifier
6738 };
6739 
6740 static int kvmclock_cpu_online(unsigned int cpu)
6741 {
6742 	tsc_khz_changed(NULL);
6743 	return 0;
6744 }
6745 
6746 static void kvm_timer_init(void)
6747 {
6748 	max_tsc_khz = tsc_khz;
6749 
6750 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6751 #ifdef CONFIG_CPU_FREQ
6752 		struct cpufreq_policy policy;
6753 		int cpu;
6754 
6755 		memset(&policy, 0, sizeof(policy));
6756 		cpu = get_cpu();
6757 		cpufreq_get_policy(&policy, cpu);
6758 		if (policy.cpuinfo.max_freq)
6759 			max_tsc_khz = policy.cpuinfo.max_freq;
6760 		put_cpu();
6761 #endif
6762 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6763 					  CPUFREQ_TRANSITION_NOTIFIER);
6764 	}
6765 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6766 
6767 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6768 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6769 }
6770 
6771 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6772 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6773 
6774 int kvm_is_in_guest(void)
6775 {
6776 	return __this_cpu_read(current_vcpu) != NULL;
6777 }
6778 
6779 static int kvm_is_user_mode(void)
6780 {
6781 	int user_mode = 3;
6782 
6783 	if (__this_cpu_read(current_vcpu))
6784 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6785 
6786 	return user_mode != 0;
6787 }
6788 
6789 static unsigned long kvm_get_guest_ip(void)
6790 {
6791 	unsigned long ip = 0;
6792 
6793 	if (__this_cpu_read(current_vcpu))
6794 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6795 
6796 	return ip;
6797 }
6798 
6799 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6800 	.is_in_guest		= kvm_is_in_guest,
6801 	.is_user_mode		= kvm_is_user_mode,
6802 	.get_guest_ip		= kvm_get_guest_ip,
6803 };
6804 
6805 static void kvm_set_mmio_spte_mask(void)
6806 {
6807 	u64 mask;
6808 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6809 
6810 	/*
6811 	 * Set the reserved bits and the present bit of an paging-structure
6812 	 * entry to generate page fault with PFER.RSV = 1.
6813 	 */
6814 
6815 	/*
6816 	 * Mask the uppermost physical address bit, which would be reserved as
6817 	 * long as the supported physical address width is less than 52.
6818 	 */
6819 	mask = 1ull << 51;
6820 
6821 	/* Set the present bit. */
6822 	mask |= 1ull;
6823 
6824 	/*
6825 	 * If reserved bit is not supported, clear the present bit to disable
6826 	 * mmio page fault.
6827 	 */
6828 	if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6829 		mask &= ~1ull;
6830 
6831 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6832 }
6833 
6834 #ifdef CONFIG_X86_64
6835 static void pvclock_gtod_update_fn(struct work_struct *work)
6836 {
6837 	struct kvm *kvm;
6838 
6839 	struct kvm_vcpu *vcpu;
6840 	int i;
6841 
6842 	spin_lock(&kvm_lock);
6843 	list_for_each_entry(kvm, &vm_list, vm_list)
6844 		kvm_for_each_vcpu(i, vcpu, kvm)
6845 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6846 	atomic_set(&kvm_guest_has_master_clock, 0);
6847 	spin_unlock(&kvm_lock);
6848 }
6849 
6850 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6851 
6852 /*
6853  * Notification about pvclock gtod data update.
6854  */
6855 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6856 			       void *priv)
6857 {
6858 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6859 	struct timekeeper *tk = priv;
6860 
6861 	update_pvclock_gtod(tk);
6862 
6863 	/* disable master clock if host does not trust, or does not
6864 	 * use, TSC based clocksource.
6865 	 */
6866 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6867 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6868 		queue_work(system_long_wq, &pvclock_gtod_work);
6869 
6870 	return 0;
6871 }
6872 
6873 static struct notifier_block pvclock_gtod_notifier = {
6874 	.notifier_call = pvclock_gtod_notify,
6875 };
6876 #endif
6877 
6878 int kvm_arch_init(void *opaque)
6879 {
6880 	int r;
6881 	struct kvm_x86_ops *ops = opaque;
6882 
6883 	if (kvm_x86_ops) {
6884 		printk(KERN_ERR "kvm: already loaded the other module\n");
6885 		r = -EEXIST;
6886 		goto out;
6887 	}
6888 
6889 	if (!ops->cpu_has_kvm_support()) {
6890 		printk(KERN_ERR "kvm: no hardware support\n");
6891 		r = -EOPNOTSUPP;
6892 		goto out;
6893 	}
6894 	if (ops->disabled_by_bios()) {
6895 		printk(KERN_ERR "kvm: disabled by bios\n");
6896 		r = -EOPNOTSUPP;
6897 		goto out;
6898 	}
6899 
6900 	/*
6901 	 * KVM explicitly assumes that the guest has an FPU and
6902 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
6903 	 * vCPU's FPU state as a fxregs_state struct.
6904 	 */
6905 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
6906 		printk(KERN_ERR "kvm: inadequate fpu\n");
6907 		r = -EOPNOTSUPP;
6908 		goto out;
6909 	}
6910 
6911 	r = -ENOMEM;
6912 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
6913 					  __alignof__(struct fpu), SLAB_ACCOUNT,
6914 					  NULL);
6915 	if (!x86_fpu_cache) {
6916 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
6917 		goto out;
6918 	}
6919 
6920 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6921 	if (!shared_msrs) {
6922 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6923 		goto out_free_x86_fpu_cache;
6924 	}
6925 
6926 	r = kvm_mmu_module_init();
6927 	if (r)
6928 		goto out_free_percpu;
6929 
6930 	kvm_set_mmio_spte_mask();
6931 
6932 	kvm_x86_ops = ops;
6933 
6934 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6935 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6936 			PT_PRESENT_MASK, 0, sme_me_mask);
6937 	kvm_timer_init();
6938 
6939 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6940 
6941 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6942 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6943 
6944 	kvm_lapic_init();
6945 #ifdef CONFIG_X86_64
6946 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6947 
6948 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6949 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6950 #endif
6951 
6952 	return 0;
6953 
6954 out_free_percpu:
6955 	free_percpu(shared_msrs);
6956 out_free_x86_fpu_cache:
6957 	kmem_cache_destroy(x86_fpu_cache);
6958 out:
6959 	return r;
6960 }
6961 
6962 void kvm_arch_exit(void)
6963 {
6964 #ifdef CONFIG_X86_64
6965 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6966 		clear_hv_tscchange_cb();
6967 #endif
6968 	kvm_lapic_exit();
6969 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6970 
6971 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6972 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6973 					    CPUFREQ_TRANSITION_NOTIFIER);
6974 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6975 #ifdef CONFIG_X86_64
6976 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6977 #endif
6978 	kvm_x86_ops = NULL;
6979 	kvm_mmu_module_exit();
6980 	free_percpu(shared_msrs);
6981 	kmem_cache_destroy(x86_fpu_cache);
6982 }
6983 
6984 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6985 {
6986 	++vcpu->stat.halt_exits;
6987 	if (lapic_in_kernel(vcpu)) {
6988 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6989 		return 1;
6990 	} else {
6991 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6992 		return 0;
6993 	}
6994 }
6995 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6996 
6997 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6998 {
6999 	int ret = kvm_skip_emulated_instruction(vcpu);
7000 	/*
7001 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7002 	 * KVM_EXIT_DEBUG here.
7003 	 */
7004 	return kvm_vcpu_halt(vcpu) && ret;
7005 }
7006 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7007 
7008 #ifdef CONFIG_X86_64
7009 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7010 			        unsigned long clock_type)
7011 {
7012 	struct kvm_clock_pairing clock_pairing;
7013 	struct timespec64 ts;
7014 	u64 cycle;
7015 	int ret;
7016 
7017 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7018 		return -KVM_EOPNOTSUPP;
7019 
7020 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7021 		return -KVM_EOPNOTSUPP;
7022 
7023 	clock_pairing.sec = ts.tv_sec;
7024 	clock_pairing.nsec = ts.tv_nsec;
7025 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7026 	clock_pairing.flags = 0;
7027 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7028 
7029 	ret = 0;
7030 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7031 			    sizeof(struct kvm_clock_pairing)))
7032 		ret = -KVM_EFAULT;
7033 
7034 	return ret;
7035 }
7036 #endif
7037 
7038 /*
7039  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7040  *
7041  * @apicid - apicid of vcpu to be kicked.
7042  */
7043 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7044 {
7045 	struct kvm_lapic_irq lapic_irq;
7046 
7047 	lapic_irq.shorthand = 0;
7048 	lapic_irq.dest_mode = 0;
7049 	lapic_irq.level = 0;
7050 	lapic_irq.dest_id = apicid;
7051 	lapic_irq.msi_redir_hint = false;
7052 
7053 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7054 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7055 }
7056 
7057 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7058 {
7059 	if (!lapic_in_kernel(vcpu)) {
7060 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7061 		return;
7062 	}
7063 	if (!vcpu->arch.apicv_active)
7064 		return;
7065 
7066 	vcpu->arch.apicv_active = false;
7067 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7068 }
7069 
7070 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7071 {
7072 	unsigned long nr, a0, a1, a2, a3, ret;
7073 	int op_64_bit;
7074 
7075 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7076 		return kvm_hv_hypercall(vcpu);
7077 
7078 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
7079 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
7080 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
7081 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
7082 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
7083 
7084 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7085 
7086 	op_64_bit = is_64_bit_mode(vcpu);
7087 	if (!op_64_bit) {
7088 		nr &= 0xFFFFFFFF;
7089 		a0 &= 0xFFFFFFFF;
7090 		a1 &= 0xFFFFFFFF;
7091 		a2 &= 0xFFFFFFFF;
7092 		a3 &= 0xFFFFFFFF;
7093 	}
7094 
7095 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7096 		ret = -KVM_EPERM;
7097 		goto out;
7098 	}
7099 
7100 	switch (nr) {
7101 	case KVM_HC_VAPIC_POLL_IRQ:
7102 		ret = 0;
7103 		break;
7104 	case KVM_HC_KICK_CPU:
7105 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7106 		ret = 0;
7107 		break;
7108 #ifdef CONFIG_X86_64
7109 	case KVM_HC_CLOCK_PAIRING:
7110 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7111 		break;
7112 #endif
7113 	case KVM_HC_SEND_IPI:
7114 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7115 		break;
7116 	default:
7117 		ret = -KVM_ENOSYS;
7118 		break;
7119 	}
7120 out:
7121 	if (!op_64_bit)
7122 		ret = (u32)ret;
7123 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7124 
7125 	++vcpu->stat.hypercalls;
7126 	return kvm_skip_emulated_instruction(vcpu);
7127 }
7128 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7129 
7130 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7131 {
7132 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7133 	char instruction[3];
7134 	unsigned long rip = kvm_rip_read(vcpu);
7135 
7136 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7137 
7138 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7139 		&ctxt->exception);
7140 }
7141 
7142 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7143 {
7144 	return vcpu->run->request_interrupt_window &&
7145 		likely(!pic_in_kernel(vcpu->kvm));
7146 }
7147 
7148 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7149 {
7150 	struct kvm_run *kvm_run = vcpu->run;
7151 
7152 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7153 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7154 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7155 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7156 	kvm_run->ready_for_interrupt_injection =
7157 		pic_in_kernel(vcpu->kvm) ||
7158 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7159 }
7160 
7161 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7162 {
7163 	int max_irr, tpr;
7164 
7165 	if (!kvm_x86_ops->update_cr8_intercept)
7166 		return;
7167 
7168 	if (!lapic_in_kernel(vcpu))
7169 		return;
7170 
7171 	if (vcpu->arch.apicv_active)
7172 		return;
7173 
7174 	if (!vcpu->arch.apic->vapic_addr)
7175 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7176 	else
7177 		max_irr = -1;
7178 
7179 	if (max_irr != -1)
7180 		max_irr >>= 4;
7181 
7182 	tpr = kvm_lapic_get_cr8(vcpu);
7183 
7184 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7185 }
7186 
7187 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7188 {
7189 	int r;
7190 
7191 	/* try to reinject previous events if any */
7192 
7193 	if (vcpu->arch.exception.injected)
7194 		kvm_x86_ops->queue_exception(vcpu);
7195 	/*
7196 	 * Do not inject an NMI or interrupt if there is a pending
7197 	 * exception.  Exceptions and interrupts are recognized at
7198 	 * instruction boundaries, i.e. the start of an instruction.
7199 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7200 	 * NMIs and interrupts, i.e. traps are recognized before an
7201 	 * NMI/interrupt that's pending on the same instruction.
7202 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7203 	 * priority, but are only generated (pended) during instruction
7204 	 * execution, i.e. a pending fault-like exception means the
7205 	 * fault occurred on the *previous* instruction and must be
7206 	 * serviced prior to recognizing any new events in order to
7207 	 * fully complete the previous instruction.
7208 	 */
7209 	else if (!vcpu->arch.exception.pending) {
7210 		if (vcpu->arch.nmi_injected)
7211 			kvm_x86_ops->set_nmi(vcpu);
7212 		else if (vcpu->arch.interrupt.injected)
7213 			kvm_x86_ops->set_irq(vcpu);
7214 	}
7215 
7216 	/*
7217 	 * Call check_nested_events() even if we reinjected a previous event
7218 	 * in order for caller to determine if it should require immediate-exit
7219 	 * from L2 to L1 due to pending L1 events which require exit
7220 	 * from L2 to L1.
7221 	 */
7222 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7223 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7224 		if (r != 0)
7225 			return r;
7226 	}
7227 
7228 	/* try to inject new event if pending */
7229 	if (vcpu->arch.exception.pending) {
7230 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7231 					vcpu->arch.exception.has_error_code,
7232 					vcpu->arch.exception.error_code);
7233 
7234 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7235 		vcpu->arch.exception.pending = false;
7236 		vcpu->arch.exception.injected = true;
7237 
7238 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7239 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7240 					     X86_EFLAGS_RF);
7241 
7242 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7243 			/*
7244 			 * This code assumes that nSVM doesn't use
7245 			 * check_nested_events(). If it does, the
7246 			 * DR6/DR7 changes should happen before L1
7247 			 * gets a #VMEXIT for an intercepted #DB in
7248 			 * L2.  (Under VMX, on the other hand, the
7249 			 * DR6/DR7 changes should not happen in the
7250 			 * event of a VM-exit to L1 for an intercepted
7251 			 * #DB in L2.)
7252 			 */
7253 			kvm_deliver_exception_payload(vcpu);
7254 			if (vcpu->arch.dr7 & DR7_GD) {
7255 				vcpu->arch.dr7 &= ~DR7_GD;
7256 				kvm_update_dr7(vcpu);
7257 			}
7258 		}
7259 
7260 		kvm_x86_ops->queue_exception(vcpu);
7261 	}
7262 
7263 	/* Don't consider new event if we re-injected an event */
7264 	if (kvm_event_needs_reinjection(vcpu))
7265 		return 0;
7266 
7267 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7268 	    kvm_x86_ops->smi_allowed(vcpu)) {
7269 		vcpu->arch.smi_pending = false;
7270 		++vcpu->arch.smi_count;
7271 		enter_smm(vcpu);
7272 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7273 		--vcpu->arch.nmi_pending;
7274 		vcpu->arch.nmi_injected = true;
7275 		kvm_x86_ops->set_nmi(vcpu);
7276 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7277 		/*
7278 		 * Because interrupts can be injected asynchronously, we are
7279 		 * calling check_nested_events again here to avoid a race condition.
7280 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7281 		 * proposal and current concerns.  Perhaps we should be setting
7282 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7283 		 */
7284 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7285 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7286 			if (r != 0)
7287 				return r;
7288 		}
7289 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7290 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7291 					    false);
7292 			kvm_x86_ops->set_irq(vcpu);
7293 		}
7294 	}
7295 
7296 	return 0;
7297 }
7298 
7299 static void process_nmi(struct kvm_vcpu *vcpu)
7300 {
7301 	unsigned limit = 2;
7302 
7303 	/*
7304 	 * x86 is limited to one NMI running, and one NMI pending after it.
7305 	 * If an NMI is already in progress, limit further NMIs to just one.
7306 	 * Otherwise, allow two (and we'll inject the first one immediately).
7307 	 */
7308 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7309 		limit = 1;
7310 
7311 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7312 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7313 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7314 }
7315 
7316 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7317 {
7318 	u32 flags = 0;
7319 	flags |= seg->g       << 23;
7320 	flags |= seg->db      << 22;
7321 	flags |= seg->l       << 21;
7322 	flags |= seg->avl     << 20;
7323 	flags |= seg->present << 15;
7324 	flags |= seg->dpl     << 13;
7325 	flags |= seg->s       << 12;
7326 	flags |= seg->type    << 8;
7327 	return flags;
7328 }
7329 
7330 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7331 {
7332 	struct kvm_segment seg;
7333 	int offset;
7334 
7335 	kvm_get_segment(vcpu, &seg, n);
7336 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7337 
7338 	if (n < 3)
7339 		offset = 0x7f84 + n * 12;
7340 	else
7341 		offset = 0x7f2c + (n - 3) * 12;
7342 
7343 	put_smstate(u32, buf, offset + 8, seg.base);
7344 	put_smstate(u32, buf, offset + 4, seg.limit);
7345 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7346 }
7347 
7348 #ifdef CONFIG_X86_64
7349 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7350 {
7351 	struct kvm_segment seg;
7352 	int offset;
7353 	u16 flags;
7354 
7355 	kvm_get_segment(vcpu, &seg, n);
7356 	offset = 0x7e00 + n * 16;
7357 
7358 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7359 	put_smstate(u16, buf, offset, seg.selector);
7360 	put_smstate(u16, buf, offset + 2, flags);
7361 	put_smstate(u32, buf, offset + 4, seg.limit);
7362 	put_smstate(u64, buf, offset + 8, seg.base);
7363 }
7364 #endif
7365 
7366 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7367 {
7368 	struct desc_ptr dt;
7369 	struct kvm_segment seg;
7370 	unsigned long val;
7371 	int i;
7372 
7373 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7374 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7375 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7376 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7377 
7378 	for (i = 0; i < 8; i++)
7379 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7380 
7381 	kvm_get_dr(vcpu, 6, &val);
7382 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7383 	kvm_get_dr(vcpu, 7, &val);
7384 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7385 
7386 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7387 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7388 	put_smstate(u32, buf, 0x7f64, seg.base);
7389 	put_smstate(u32, buf, 0x7f60, seg.limit);
7390 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7391 
7392 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7393 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7394 	put_smstate(u32, buf, 0x7f80, seg.base);
7395 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7396 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7397 
7398 	kvm_x86_ops->get_gdt(vcpu, &dt);
7399 	put_smstate(u32, buf, 0x7f74, dt.address);
7400 	put_smstate(u32, buf, 0x7f70, dt.size);
7401 
7402 	kvm_x86_ops->get_idt(vcpu, &dt);
7403 	put_smstate(u32, buf, 0x7f58, dt.address);
7404 	put_smstate(u32, buf, 0x7f54, dt.size);
7405 
7406 	for (i = 0; i < 6; i++)
7407 		enter_smm_save_seg_32(vcpu, buf, i);
7408 
7409 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7410 
7411 	/* revision id */
7412 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7413 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7414 }
7415 
7416 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7417 {
7418 #ifdef CONFIG_X86_64
7419 	struct desc_ptr dt;
7420 	struct kvm_segment seg;
7421 	unsigned long val;
7422 	int i;
7423 
7424 	for (i = 0; i < 16; i++)
7425 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7426 
7427 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7428 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7429 
7430 	kvm_get_dr(vcpu, 6, &val);
7431 	put_smstate(u64, buf, 0x7f68, val);
7432 	kvm_get_dr(vcpu, 7, &val);
7433 	put_smstate(u64, buf, 0x7f60, val);
7434 
7435 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7436 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7437 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7438 
7439 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7440 
7441 	/* revision id */
7442 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7443 
7444 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7445 
7446 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7447 	put_smstate(u16, buf, 0x7e90, seg.selector);
7448 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7449 	put_smstate(u32, buf, 0x7e94, seg.limit);
7450 	put_smstate(u64, buf, 0x7e98, seg.base);
7451 
7452 	kvm_x86_ops->get_idt(vcpu, &dt);
7453 	put_smstate(u32, buf, 0x7e84, dt.size);
7454 	put_smstate(u64, buf, 0x7e88, dt.address);
7455 
7456 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7457 	put_smstate(u16, buf, 0x7e70, seg.selector);
7458 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7459 	put_smstate(u32, buf, 0x7e74, seg.limit);
7460 	put_smstate(u64, buf, 0x7e78, seg.base);
7461 
7462 	kvm_x86_ops->get_gdt(vcpu, &dt);
7463 	put_smstate(u32, buf, 0x7e64, dt.size);
7464 	put_smstate(u64, buf, 0x7e68, dt.address);
7465 
7466 	for (i = 0; i < 6; i++)
7467 		enter_smm_save_seg_64(vcpu, buf, i);
7468 #else
7469 	WARN_ON_ONCE(1);
7470 #endif
7471 }
7472 
7473 static void enter_smm(struct kvm_vcpu *vcpu)
7474 {
7475 	struct kvm_segment cs, ds;
7476 	struct desc_ptr dt;
7477 	char buf[512];
7478 	u32 cr0;
7479 
7480 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7481 	memset(buf, 0, 512);
7482 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7483 		enter_smm_save_state_64(vcpu, buf);
7484 	else
7485 		enter_smm_save_state_32(vcpu, buf);
7486 
7487 	/*
7488 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7489 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7490 	 * the SMM state-save area.
7491 	 */
7492 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7493 
7494 	vcpu->arch.hflags |= HF_SMM_MASK;
7495 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7496 
7497 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7498 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7499 	else
7500 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7501 
7502 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7503 	kvm_rip_write(vcpu, 0x8000);
7504 
7505 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7506 	kvm_x86_ops->set_cr0(vcpu, cr0);
7507 	vcpu->arch.cr0 = cr0;
7508 
7509 	kvm_x86_ops->set_cr4(vcpu, 0);
7510 
7511 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7512 	dt.address = dt.size = 0;
7513 	kvm_x86_ops->set_idt(vcpu, &dt);
7514 
7515 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7516 
7517 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7518 	cs.base = vcpu->arch.smbase;
7519 
7520 	ds.selector = 0;
7521 	ds.base = 0;
7522 
7523 	cs.limit    = ds.limit = 0xffffffff;
7524 	cs.type     = ds.type = 0x3;
7525 	cs.dpl      = ds.dpl = 0;
7526 	cs.db       = ds.db = 0;
7527 	cs.s        = ds.s = 1;
7528 	cs.l        = ds.l = 0;
7529 	cs.g        = ds.g = 1;
7530 	cs.avl      = ds.avl = 0;
7531 	cs.present  = ds.present = 1;
7532 	cs.unusable = ds.unusable = 0;
7533 	cs.padding  = ds.padding = 0;
7534 
7535 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7536 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7537 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7538 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7539 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7540 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7541 
7542 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7543 		kvm_x86_ops->set_efer(vcpu, 0);
7544 
7545 	kvm_update_cpuid(vcpu);
7546 	kvm_mmu_reset_context(vcpu);
7547 }
7548 
7549 static void process_smi(struct kvm_vcpu *vcpu)
7550 {
7551 	vcpu->arch.smi_pending = true;
7552 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7553 }
7554 
7555 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7556 {
7557 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7558 }
7559 
7560 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7561 {
7562 	if (!kvm_apic_present(vcpu))
7563 		return;
7564 
7565 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7566 
7567 	if (irqchip_split(vcpu->kvm))
7568 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7569 	else {
7570 		if (vcpu->arch.apicv_active)
7571 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7572 		if (ioapic_in_kernel(vcpu->kvm))
7573 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7574 	}
7575 
7576 	if (is_guest_mode(vcpu))
7577 		vcpu->arch.load_eoi_exitmap_pending = true;
7578 	else
7579 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7580 }
7581 
7582 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7583 {
7584 	u64 eoi_exit_bitmap[4];
7585 
7586 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7587 		return;
7588 
7589 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7590 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7591 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7592 }
7593 
7594 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7595 		unsigned long start, unsigned long end,
7596 		bool blockable)
7597 {
7598 	unsigned long apic_address;
7599 
7600 	/*
7601 	 * The physical address of apic access page is stored in the VMCS.
7602 	 * Update it when it becomes invalid.
7603 	 */
7604 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7605 	if (start <= apic_address && apic_address < end)
7606 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7607 
7608 	return 0;
7609 }
7610 
7611 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7612 {
7613 	struct page *page = NULL;
7614 
7615 	if (!lapic_in_kernel(vcpu))
7616 		return;
7617 
7618 	if (!kvm_x86_ops->set_apic_access_page_addr)
7619 		return;
7620 
7621 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7622 	if (is_error_page(page))
7623 		return;
7624 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7625 
7626 	/*
7627 	 * Do not pin apic access page in memory, the MMU notifier
7628 	 * will call us again if it is migrated or swapped out.
7629 	 */
7630 	put_page(page);
7631 }
7632 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7633 
7634 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7635 {
7636 	smp_send_reschedule(vcpu->cpu);
7637 }
7638 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7639 
7640 /*
7641  * Returns 1 to let vcpu_run() continue the guest execution loop without
7642  * exiting to the userspace.  Otherwise, the value will be returned to the
7643  * userspace.
7644  */
7645 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7646 {
7647 	int r;
7648 	bool req_int_win =
7649 		dm_request_for_irq_injection(vcpu) &&
7650 		kvm_cpu_accept_dm_intr(vcpu);
7651 
7652 	bool req_immediate_exit = false;
7653 
7654 	if (kvm_request_pending(vcpu)) {
7655 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7656 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7657 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7658 			kvm_mmu_unload(vcpu);
7659 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7660 			__kvm_migrate_timers(vcpu);
7661 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7662 			kvm_gen_update_masterclock(vcpu->kvm);
7663 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7664 			kvm_gen_kvmclock_update(vcpu);
7665 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7666 			r = kvm_guest_time_update(vcpu);
7667 			if (unlikely(r))
7668 				goto out;
7669 		}
7670 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7671 			kvm_mmu_sync_roots(vcpu);
7672 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7673 			kvm_mmu_load_cr3(vcpu);
7674 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7675 			kvm_vcpu_flush_tlb(vcpu, true);
7676 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7677 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7678 			r = 0;
7679 			goto out;
7680 		}
7681 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7682 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7683 			vcpu->mmio_needed = 0;
7684 			r = 0;
7685 			goto out;
7686 		}
7687 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7688 			/* Page is swapped out. Do synthetic halt */
7689 			vcpu->arch.apf.halted = true;
7690 			r = 1;
7691 			goto out;
7692 		}
7693 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7694 			record_steal_time(vcpu);
7695 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7696 			process_smi(vcpu);
7697 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7698 			process_nmi(vcpu);
7699 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7700 			kvm_pmu_handle_event(vcpu);
7701 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7702 			kvm_pmu_deliver_pmi(vcpu);
7703 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7704 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7705 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7706 				     vcpu->arch.ioapic_handled_vectors)) {
7707 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7708 				vcpu->run->eoi.vector =
7709 						vcpu->arch.pending_ioapic_eoi;
7710 				r = 0;
7711 				goto out;
7712 			}
7713 		}
7714 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7715 			vcpu_scan_ioapic(vcpu);
7716 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7717 			vcpu_load_eoi_exitmap(vcpu);
7718 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7719 			kvm_vcpu_reload_apic_access_page(vcpu);
7720 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7721 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7722 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7723 			r = 0;
7724 			goto out;
7725 		}
7726 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7727 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7728 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7729 			r = 0;
7730 			goto out;
7731 		}
7732 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7733 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7734 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7735 			r = 0;
7736 			goto out;
7737 		}
7738 
7739 		/*
7740 		 * KVM_REQ_HV_STIMER has to be processed after
7741 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7742 		 * depend on the guest clock being up-to-date
7743 		 */
7744 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7745 			kvm_hv_process_stimers(vcpu);
7746 	}
7747 
7748 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7749 		++vcpu->stat.req_event;
7750 		kvm_apic_accept_events(vcpu);
7751 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7752 			r = 1;
7753 			goto out;
7754 		}
7755 
7756 		if (inject_pending_event(vcpu, req_int_win) != 0)
7757 			req_immediate_exit = true;
7758 		else {
7759 			/* Enable SMI/NMI/IRQ window open exits if needed.
7760 			 *
7761 			 * SMIs have three cases:
7762 			 * 1) They can be nested, and then there is nothing to
7763 			 *    do here because RSM will cause a vmexit anyway.
7764 			 * 2) There is an ISA-specific reason why SMI cannot be
7765 			 *    injected, and the moment when this changes can be
7766 			 *    intercepted.
7767 			 * 3) Or the SMI can be pending because
7768 			 *    inject_pending_event has completed the injection
7769 			 *    of an IRQ or NMI from the previous vmexit, and
7770 			 *    then we request an immediate exit to inject the
7771 			 *    SMI.
7772 			 */
7773 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7774 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7775 					req_immediate_exit = true;
7776 			if (vcpu->arch.nmi_pending)
7777 				kvm_x86_ops->enable_nmi_window(vcpu);
7778 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7779 				kvm_x86_ops->enable_irq_window(vcpu);
7780 			WARN_ON(vcpu->arch.exception.pending);
7781 		}
7782 
7783 		if (kvm_lapic_enabled(vcpu)) {
7784 			update_cr8_intercept(vcpu);
7785 			kvm_lapic_sync_to_vapic(vcpu);
7786 		}
7787 	}
7788 
7789 	r = kvm_mmu_reload(vcpu);
7790 	if (unlikely(r)) {
7791 		goto cancel_injection;
7792 	}
7793 
7794 	preempt_disable();
7795 
7796 	kvm_x86_ops->prepare_guest_switch(vcpu);
7797 
7798 	/*
7799 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7800 	 * IPI are then delayed after guest entry, which ensures that they
7801 	 * result in virtual interrupt delivery.
7802 	 */
7803 	local_irq_disable();
7804 	vcpu->mode = IN_GUEST_MODE;
7805 
7806 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7807 
7808 	/*
7809 	 * 1) We should set ->mode before checking ->requests.  Please see
7810 	 * the comment in kvm_vcpu_exiting_guest_mode().
7811 	 *
7812 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
7813 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7814 	 * (see vmx_deliver_posted_interrupt).
7815 	 *
7816 	 * 3) This also orders the write to mode from any reads to the page
7817 	 * tables done while the VCPU is running.  Please see the comment
7818 	 * in kvm_flush_remote_tlbs.
7819 	 */
7820 	smp_mb__after_srcu_read_unlock();
7821 
7822 	/*
7823 	 * This handles the case where a posted interrupt was
7824 	 * notified with kvm_vcpu_kick.
7825 	 */
7826 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7827 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7828 
7829 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7830 	    || need_resched() || signal_pending(current)) {
7831 		vcpu->mode = OUTSIDE_GUEST_MODE;
7832 		smp_wmb();
7833 		local_irq_enable();
7834 		preempt_enable();
7835 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7836 		r = 1;
7837 		goto cancel_injection;
7838 	}
7839 
7840 	kvm_load_guest_xcr0(vcpu);
7841 
7842 	if (req_immediate_exit) {
7843 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7844 		kvm_x86_ops->request_immediate_exit(vcpu);
7845 	}
7846 
7847 	trace_kvm_entry(vcpu->vcpu_id);
7848 	if (lapic_timer_advance_ns)
7849 		wait_lapic_expire(vcpu);
7850 	guest_enter_irqoff();
7851 
7852 	if (unlikely(vcpu->arch.switch_db_regs)) {
7853 		set_debugreg(0, 7);
7854 		set_debugreg(vcpu->arch.eff_db[0], 0);
7855 		set_debugreg(vcpu->arch.eff_db[1], 1);
7856 		set_debugreg(vcpu->arch.eff_db[2], 2);
7857 		set_debugreg(vcpu->arch.eff_db[3], 3);
7858 		set_debugreg(vcpu->arch.dr6, 6);
7859 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7860 	}
7861 
7862 	kvm_x86_ops->run(vcpu);
7863 
7864 	/*
7865 	 * Do this here before restoring debug registers on the host.  And
7866 	 * since we do this before handling the vmexit, a DR access vmexit
7867 	 * can (a) read the correct value of the debug registers, (b) set
7868 	 * KVM_DEBUGREG_WONT_EXIT again.
7869 	 */
7870 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7871 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7872 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7873 		kvm_update_dr0123(vcpu);
7874 		kvm_update_dr6(vcpu);
7875 		kvm_update_dr7(vcpu);
7876 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7877 	}
7878 
7879 	/*
7880 	 * If the guest has used debug registers, at least dr7
7881 	 * will be disabled while returning to the host.
7882 	 * If we don't have active breakpoints in the host, we don't
7883 	 * care about the messed up debug address registers. But if
7884 	 * we have some of them active, restore the old state.
7885 	 */
7886 	if (hw_breakpoint_active())
7887 		hw_breakpoint_restore();
7888 
7889 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7890 
7891 	vcpu->mode = OUTSIDE_GUEST_MODE;
7892 	smp_wmb();
7893 
7894 	kvm_put_guest_xcr0(vcpu);
7895 
7896 	kvm_before_interrupt(vcpu);
7897 	kvm_x86_ops->handle_external_intr(vcpu);
7898 	kvm_after_interrupt(vcpu);
7899 
7900 	++vcpu->stat.exits;
7901 
7902 	guest_exit_irqoff();
7903 
7904 	local_irq_enable();
7905 	preempt_enable();
7906 
7907 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7908 
7909 	/*
7910 	 * Profile KVM exit RIPs:
7911 	 */
7912 	if (unlikely(prof_on == KVM_PROFILING)) {
7913 		unsigned long rip = kvm_rip_read(vcpu);
7914 		profile_hit(KVM_PROFILING, (void *)rip);
7915 	}
7916 
7917 	if (unlikely(vcpu->arch.tsc_always_catchup))
7918 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7919 
7920 	if (vcpu->arch.apic_attention)
7921 		kvm_lapic_sync_from_vapic(vcpu);
7922 
7923 	vcpu->arch.gpa_available = false;
7924 	r = kvm_x86_ops->handle_exit(vcpu);
7925 	return r;
7926 
7927 cancel_injection:
7928 	kvm_x86_ops->cancel_injection(vcpu);
7929 	if (unlikely(vcpu->arch.apic_attention))
7930 		kvm_lapic_sync_from_vapic(vcpu);
7931 out:
7932 	return r;
7933 }
7934 
7935 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7936 {
7937 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7938 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7939 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7940 		kvm_vcpu_block(vcpu);
7941 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7942 
7943 		if (kvm_x86_ops->post_block)
7944 			kvm_x86_ops->post_block(vcpu);
7945 
7946 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7947 			return 1;
7948 	}
7949 
7950 	kvm_apic_accept_events(vcpu);
7951 	switch(vcpu->arch.mp_state) {
7952 	case KVM_MP_STATE_HALTED:
7953 		vcpu->arch.pv.pv_unhalted = false;
7954 		vcpu->arch.mp_state =
7955 			KVM_MP_STATE_RUNNABLE;
7956 		/* fall through */
7957 	case KVM_MP_STATE_RUNNABLE:
7958 		vcpu->arch.apf.halted = false;
7959 		break;
7960 	case KVM_MP_STATE_INIT_RECEIVED:
7961 		break;
7962 	default:
7963 		return -EINTR;
7964 		break;
7965 	}
7966 	return 1;
7967 }
7968 
7969 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7970 {
7971 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7972 		kvm_x86_ops->check_nested_events(vcpu, false);
7973 
7974 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7975 		!vcpu->arch.apf.halted);
7976 }
7977 
7978 static int vcpu_run(struct kvm_vcpu *vcpu)
7979 {
7980 	int r;
7981 	struct kvm *kvm = vcpu->kvm;
7982 
7983 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7984 	vcpu->arch.l1tf_flush_l1d = true;
7985 
7986 	for (;;) {
7987 		if (kvm_vcpu_running(vcpu)) {
7988 			r = vcpu_enter_guest(vcpu);
7989 		} else {
7990 			r = vcpu_block(kvm, vcpu);
7991 		}
7992 
7993 		if (r <= 0)
7994 			break;
7995 
7996 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7997 		if (kvm_cpu_has_pending_timer(vcpu))
7998 			kvm_inject_pending_timer_irqs(vcpu);
7999 
8000 		if (dm_request_for_irq_injection(vcpu) &&
8001 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8002 			r = 0;
8003 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8004 			++vcpu->stat.request_irq_exits;
8005 			break;
8006 		}
8007 
8008 		kvm_check_async_pf_completion(vcpu);
8009 
8010 		if (signal_pending(current)) {
8011 			r = -EINTR;
8012 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8013 			++vcpu->stat.signal_exits;
8014 			break;
8015 		}
8016 		if (need_resched()) {
8017 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8018 			cond_resched();
8019 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8020 		}
8021 	}
8022 
8023 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8024 
8025 	return r;
8026 }
8027 
8028 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8029 {
8030 	int r;
8031 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8032 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8033 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8034 	if (r != EMULATE_DONE)
8035 		return 0;
8036 	return 1;
8037 }
8038 
8039 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8040 {
8041 	BUG_ON(!vcpu->arch.pio.count);
8042 
8043 	return complete_emulated_io(vcpu);
8044 }
8045 
8046 /*
8047  * Implements the following, as a state machine:
8048  *
8049  * read:
8050  *   for each fragment
8051  *     for each mmio piece in the fragment
8052  *       write gpa, len
8053  *       exit
8054  *       copy data
8055  *   execute insn
8056  *
8057  * write:
8058  *   for each fragment
8059  *     for each mmio piece in the fragment
8060  *       write gpa, len
8061  *       copy data
8062  *       exit
8063  */
8064 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8065 {
8066 	struct kvm_run *run = vcpu->run;
8067 	struct kvm_mmio_fragment *frag;
8068 	unsigned len;
8069 
8070 	BUG_ON(!vcpu->mmio_needed);
8071 
8072 	/* Complete previous fragment */
8073 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8074 	len = min(8u, frag->len);
8075 	if (!vcpu->mmio_is_write)
8076 		memcpy(frag->data, run->mmio.data, len);
8077 
8078 	if (frag->len <= 8) {
8079 		/* Switch to the next fragment. */
8080 		frag++;
8081 		vcpu->mmio_cur_fragment++;
8082 	} else {
8083 		/* Go forward to the next mmio piece. */
8084 		frag->data += len;
8085 		frag->gpa += len;
8086 		frag->len -= len;
8087 	}
8088 
8089 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8090 		vcpu->mmio_needed = 0;
8091 
8092 		/* FIXME: return into emulator if single-stepping.  */
8093 		if (vcpu->mmio_is_write)
8094 			return 1;
8095 		vcpu->mmio_read_completed = 1;
8096 		return complete_emulated_io(vcpu);
8097 	}
8098 
8099 	run->exit_reason = KVM_EXIT_MMIO;
8100 	run->mmio.phys_addr = frag->gpa;
8101 	if (vcpu->mmio_is_write)
8102 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8103 	run->mmio.len = min(8u, frag->len);
8104 	run->mmio.is_write = vcpu->mmio_is_write;
8105 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8106 	return 0;
8107 }
8108 
8109 /* Swap (qemu) user FPU context for the guest FPU context. */
8110 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8111 {
8112 	preempt_disable();
8113 	copy_fpregs_to_fpstate(&current->thread.fpu);
8114 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8115 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8116 				~XFEATURE_MASK_PKRU);
8117 	preempt_enable();
8118 	trace_kvm_fpu(1);
8119 }
8120 
8121 /* When vcpu_run ends, restore user space FPU context. */
8122 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8123 {
8124 	preempt_disable();
8125 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8126 	copy_kernel_to_fpregs(&current->thread.fpu.state);
8127 	preempt_enable();
8128 	++vcpu->stat.fpu_reload;
8129 	trace_kvm_fpu(0);
8130 }
8131 
8132 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8133 {
8134 	int r;
8135 
8136 	vcpu_load(vcpu);
8137 	kvm_sigset_activate(vcpu);
8138 	kvm_load_guest_fpu(vcpu);
8139 
8140 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8141 		if (kvm_run->immediate_exit) {
8142 			r = -EINTR;
8143 			goto out;
8144 		}
8145 		kvm_vcpu_block(vcpu);
8146 		kvm_apic_accept_events(vcpu);
8147 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8148 		r = -EAGAIN;
8149 		if (signal_pending(current)) {
8150 			r = -EINTR;
8151 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8152 			++vcpu->stat.signal_exits;
8153 		}
8154 		goto out;
8155 	}
8156 
8157 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8158 		r = -EINVAL;
8159 		goto out;
8160 	}
8161 
8162 	if (vcpu->run->kvm_dirty_regs) {
8163 		r = sync_regs(vcpu);
8164 		if (r != 0)
8165 			goto out;
8166 	}
8167 
8168 	/* re-sync apic's tpr */
8169 	if (!lapic_in_kernel(vcpu)) {
8170 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8171 			r = -EINVAL;
8172 			goto out;
8173 		}
8174 	}
8175 
8176 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8177 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8178 		vcpu->arch.complete_userspace_io = NULL;
8179 		r = cui(vcpu);
8180 		if (r <= 0)
8181 			goto out;
8182 	} else
8183 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8184 
8185 	if (kvm_run->immediate_exit)
8186 		r = -EINTR;
8187 	else
8188 		r = vcpu_run(vcpu);
8189 
8190 out:
8191 	kvm_put_guest_fpu(vcpu);
8192 	if (vcpu->run->kvm_valid_regs)
8193 		store_regs(vcpu);
8194 	post_kvm_run_save(vcpu);
8195 	kvm_sigset_deactivate(vcpu);
8196 
8197 	vcpu_put(vcpu);
8198 	return r;
8199 }
8200 
8201 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8202 {
8203 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8204 		/*
8205 		 * We are here if userspace calls get_regs() in the middle of
8206 		 * instruction emulation. Registers state needs to be copied
8207 		 * back from emulation context to vcpu. Userspace shouldn't do
8208 		 * that usually, but some bad designed PV devices (vmware
8209 		 * backdoor interface) need this to work
8210 		 */
8211 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8212 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8213 	}
8214 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8215 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8216 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8217 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8218 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8219 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8220 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8221 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8222 #ifdef CONFIG_X86_64
8223 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8224 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8225 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8226 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8227 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8228 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8229 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8230 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8231 #endif
8232 
8233 	regs->rip = kvm_rip_read(vcpu);
8234 	regs->rflags = kvm_get_rflags(vcpu);
8235 }
8236 
8237 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8238 {
8239 	vcpu_load(vcpu);
8240 	__get_regs(vcpu, regs);
8241 	vcpu_put(vcpu);
8242 	return 0;
8243 }
8244 
8245 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8246 {
8247 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8248 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8249 
8250 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8251 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8252 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8253 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8254 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8255 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8256 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8257 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8258 #ifdef CONFIG_X86_64
8259 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8260 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8261 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8262 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8263 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8264 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8265 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8266 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8267 #endif
8268 
8269 	kvm_rip_write(vcpu, regs->rip);
8270 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8271 
8272 	vcpu->arch.exception.pending = false;
8273 
8274 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8275 }
8276 
8277 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8278 {
8279 	vcpu_load(vcpu);
8280 	__set_regs(vcpu, regs);
8281 	vcpu_put(vcpu);
8282 	return 0;
8283 }
8284 
8285 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8286 {
8287 	struct kvm_segment cs;
8288 
8289 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8290 	*db = cs.db;
8291 	*l = cs.l;
8292 }
8293 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8294 
8295 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8296 {
8297 	struct desc_ptr dt;
8298 
8299 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8300 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8301 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8302 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8303 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8304 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8305 
8306 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8307 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8308 
8309 	kvm_x86_ops->get_idt(vcpu, &dt);
8310 	sregs->idt.limit = dt.size;
8311 	sregs->idt.base = dt.address;
8312 	kvm_x86_ops->get_gdt(vcpu, &dt);
8313 	sregs->gdt.limit = dt.size;
8314 	sregs->gdt.base = dt.address;
8315 
8316 	sregs->cr0 = kvm_read_cr0(vcpu);
8317 	sregs->cr2 = vcpu->arch.cr2;
8318 	sregs->cr3 = kvm_read_cr3(vcpu);
8319 	sregs->cr4 = kvm_read_cr4(vcpu);
8320 	sregs->cr8 = kvm_get_cr8(vcpu);
8321 	sregs->efer = vcpu->arch.efer;
8322 	sregs->apic_base = kvm_get_apic_base(vcpu);
8323 
8324 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8325 
8326 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8327 		set_bit(vcpu->arch.interrupt.nr,
8328 			(unsigned long *)sregs->interrupt_bitmap);
8329 }
8330 
8331 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8332 				  struct kvm_sregs *sregs)
8333 {
8334 	vcpu_load(vcpu);
8335 	__get_sregs(vcpu, sregs);
8336 	vcpu_put(vcpu);
8337 	return 0;
8338 }
8339 
8340 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8341 				    struct kvm_mp_state *mp_state)
8342 {
8343 	vcpu_load(vcpu);
8344 
8345 	kvm_apic_accept_events(vcpu);
8346 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8347 					vcpu->arch.pv.pv_unhalted)
8348 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8349 	else
8350 		mp_state->mp_state = vcpu->arch.mp_state;
8351 
8352 	vcpu_put(vcpu);
8353 	return 0;
8354 }
8355 
8356 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8357 				    struct kvm_mp_state *mp_state)
8358 {
8359 	int ret = -EINVAL;
8360 
8361 	vcpu_load(vcpu);
8362 
8363 	if (!lapic_in_kernel(vcpu) &&
8364 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8365 		goto out;
8366 
8367 	/* INITs are latched while in SMM */
8368 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8369 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8370 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8371 		goto out;
8372 
8373 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8374 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8375 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8376 	} else
8377 		vcpu->arch.mp_state = mp_state->mp_state;
8378 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8379 
8380 	ret = 0;
8381 out:
8382 	vcpu_put(vcpu);
8383 	return ret;
8384 }
8385 
8386 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8387 		    int reason, bool has_error_code, u32 error_code)
8388 {
8389 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8390 	int ret;
8391 
8392 	init_emulate_ctxt(vcpu);
8393 
8394 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8395 				   has_error_code, error_code);
8396 
8397 	if (ret)
8398 		return EMULATE_FAIL;
8399 
8400 	kvm_rip_write(vcpu, ctxt->eip);
8401 	kvm_set_rflags(vcpu, ctxt->eflags);
8402 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8403 	return EMULATE_DONE;
8404 }
8405 EXPORT_SYMBOL_GPL(kvm_task_switch);
8406 
8407 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8408 {
8409 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8410 			(sregs->cr4 & X86_CR4_OSXSAVE))
8411 		return  -EINVAL;
8412 
8413 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8414 		/*
8415 		 * When EFER.LME and CR0.PG are set, the processor is in
8416 		 * 64-bit mode (though maybe in a 32-bit code segment).
8417 		 * CR4.PAE and EFER.LMA must be set.
8418 		 */
8419 		if (!(sregs->cr4 & X86_CR4_PAE)
8420 		    || !(sregs->efer & EFER_LMA))
8421 			return -EINVAL;
8422 	} else {
8423 		/*
8424 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8425 		 * segment cannot be 64-bit.
8426 		 */
8427 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8428 			return -EINVAL;
8429 	}
8430 
8431 	return 0;
8432 }
8433 
8434 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8435 {
8436 	struct msr_data apic_base_msr;
8437 	int mmu_reset_needed = 0;
8438 	int cpuid_update_needed = 0;
8439 	int pending_vec, max_bits, idx;
8440 	struct desc_ptr dt;
8441 	int ret = -EINVAL;
8442 
8443 	if (kvm_valid_sregs(vcpu, sregs))
8444 		goto out;
8445 
8446 	apic_base_msr.data = sregs->apic_base;
8447 	apic_base_msr.host_initiated = true;
8448 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8449 		goto out;
8450 
8451 	dt.size = sregs->idt.limit;
8452 	dt.address = sregs->idt.base;
8453 	kvm_x86_ops->set_idt(vcpu, &dt);
8454 	dt.size = sregs->gdt.limit;
8455 	dt.address = sregs->gdt.base;
8456 	kvm_x86_ops->set_gdt(vcpu, &dt);
8457 
8458 	vcpu->arch.cr2 = sregs->cr2;
8459 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8460 	vcpu->arch.cr3 = sregs->cr3;
8461 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8462 
8463 	kvm_set_cr8(vcpu, sregs->cr8);
8464 
8465 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8466 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8467 
8468 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8469 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8470 	vcpu->arch.cr0 = sregs->cr0;
8471 
8472 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8473 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8474 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8475 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8476 	if (cpuid_update_needed)
8477 		kvm_update_cpuid(vcpu);
8478 
8479 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8480 	if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8481 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8482 		mmu_reset_needed = 1;
8483 	}
8484 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8485 
8486 	if (mmu_reset_needed)
8487 		kvm_mmu_reset_context(vcpu);
8488 
8489 	max_bits = KVM_NR_INTERRUPTS;
8490 	pending_vec = find_first_bit(
8491 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8492 	if (pending_vec < max_bits) {
8493 		kvm_queue_interrupt(vcpu, pending_vec, false);
8494 		pr_debug("Set back pending irq %d\n", pending_vec);
8495 	}
8496 
8497 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8498 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8499 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8500 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8501 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8502 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8503 
8504 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8505 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8506 
8507 	update_cr8_intercept(vcpu);
8508 
8509 	/* Older userspace won't unhalt the vcpu on reset. */
8510 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8511 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8512 	    !is_protmode(vcpu))
8513 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8514 
8515 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8516 
8517 	ret = 0;
8518 out:
8519 	return ret;
8520 }
8521 
8522 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8523 				  struct kvm_sregs *sregs)
8524 {
8525 	int ret;
8526 
8527 	vcpu_load(vcpu);
8528 	ret = __set_sregs(vcpu, sregs);
8529 	vcpu_put(vcpu);
8530 	return ret;
8531 }
8532 
8533 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8534 					struct kvm_guest_debug *dbg)
8535 {
8536 	unsigned long rflags;
8537 	int i, r;
8538 
8539 	vcpu_load(vcpu);
8540 
8541 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8542 		r = -EBUSY;
8543 		if (vcpu->arch.exception.pending)
8544 			goto out;
8545 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8546 			kvm_queue_exception(vcpu, DB_VECTOR);
8547 		else
8548 			kvm_queue_exception(vcpu, BP_VECTOR);
8549 	}
8550 
8551 	/*
8552 	 * Read rflags as long as potentially injected trace flags are still
8553 	 * filtered out.
8554 	 */
8555 	rflags = kvm_get_rflags(vcpu);
8556 
8557 	vcpu->guest_debug = dbg->control;
8558 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8559 		vcpu->guest_debug = 0;
8560 
8561 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8562 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8563 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8564 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8565 	} else {
8566 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8567 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8568 	}
8569 	kvm_update_dr7(vcpu);
8570 
8571 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8572 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8573 			get_segment_base(vcpu, VCPU_SREG_CS);
8574 
8575 	/*
8576 	 * Trigger an rflags update that will inject or remove the trace
8577 	 * flags.
8578 	 */
8579 	kvm_set_rflags(vcpu, rflags);
8580 
8581 	kvm_x86_ops->update_bp_intercept(vcpu);
8582 
8583 	r = 0;
8584 
8585 out:
8586 	vcpu_put(vcpu);
8587 	return r;
8588 }
8589 
8590 /*
8591  * Translate a guest virtual address to a guest physical address.
8592  */
8593 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8594 				    struct kvm_translation *tr)
8595 {
8596 	unsigned long vaddr = tr->linear_address;
8597 	gpa_t gpa;
8598 	int idx;
8599 
8600 	vcpu_load(vcpu);
8601 
8602 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8603 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8604 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8605 	tr->physical_address = gpa;
8606 	tr->valid = gpa != UNMAPPED_GVA;
8607 	tr->writeable = 1;
8608 	tr->usermode = 0;
8609 
8610 	vcpu_put(vcpu);
8611 	return 0;
8612 }
8613 
8614 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8615 {
8616 	struct fxregs_state *fxsave;
8617 
8618 	vcpu_load(vcpu);
8619 
8620 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8621 	memcpy(fpu->fpr, fxsave->st_space, 128);
8622 	fpu->fcw = fxsave->cwd;
8623 	fpu->fsw = fxsave->swd;
8624 	fpu->ftwx = fxsave->twd;
8625 	fpu->last_opcode = fxsave->fop;
8626 	fpu->last_ip = fxsave->rip;
8627 	fpu->last_dp = fxsave->rdp;
8628 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8629 
8630 	vcpu_put(vcpu);
8631 	return 0;
8632 }
8633 
8634 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8635 {
8636 	struct fxregs_state *fxsave;
8637 
8638 	vcpu_load(vcpu);
8639 
8640 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8641 
8642 	memcpy(fxsave->st_space, fpu->fpr, 128);
8643 	fxsave->cwd = fpu->fcw;
8644 	fxsave->swd = fpu->fsw;
8645 	fxsave->twd = fpu->ftwx;
8646 	fxsave->fop = fpu->last_opcode;
8647 	fxsave->rip = fpu->last_ip;
8648 	fxsave->rdp = fpu->last_dp;
8649 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8650 
8651 	vcpu_put(vcpu);
8652 	return 0;
8653 }
8654 
8655 static void store_regs(struct kvm_vcpu *vcpu)
8656 {
8657 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8658 
8659 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8660 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8661 
8662 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8663 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8664 
8665 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8666 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8667 				vcpu, &vcpu->run->s.regs.events);
8668 }
8669 
8670 static int sync_regs(struct kvm_vcpu *vcpu)
8671 {
8672 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8673 		return -EINVAL;
8674 
8675 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8676 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8677 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8678 	}
8679 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8680 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8681 			return -EINVAL;
8682 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8683 	}
8684 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8685 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8686 				vcpu, &vcpu->run->s.regs.events))
8687 			return -EINVAL;
8688 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8689 	}
8690 
8691 	return 0;
8692 }
8693 
8694 static void fx_init(struct kvm_vcpu *vcpu)
8695 {
8696 	fpstate_init(&vcpu->arch.guest_fpu->state);
8697 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8698 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8699 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8700 
8701 	/*
8702 	 * Ensure guest xcr0 is valid for loading
8703 	 */
8704 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8705 
8706 	vcpu->arch.cr0 |= X86_CR0_ET;
8707 }
8708 
8709 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8710 {
8711 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8712 
8713 	kvmclock_reset(vcpu);
8714 
8715 	kvm_x86_ops->vcpu_free(vcpu);
8716 	free_cpumask_var(wbinvd_dirty_mask);
8717 }
8718 
8719 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8720 						unsigned int id)
8721 {
8722 	struct kvm_vcpu *vcpu;
8723 
8724 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8725 		printk_once(KERN_WARNING
8726 		"kvm: SMP vm created on host with unstable TSC; "
8727 		"guest TSC will not be reliable\n");
8728 
8729 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8730 
8731 	return vcpu;
8732 }
8733 
8734 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8735 {
8736 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8737 	kvm_vcpu_mtrr_init(vcpu);
8738 	vcpu_load(vcpu);
8739 	kvm_vcpu_reset(vcpu, false);
8740 	kvm_init_mmu(vcpu, false);
8741 	vcpu_put(vcpu);
8742 	return 0;
8743 }
8744 
8745 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8746 {
8747 	struct msr_data msr;
8748 	struct kvm *kvm = vcpu->kvm;
8749 
8750 	kvm_hv_vcpu_postcreate(vcpu);
8751 
8752 	if (mutex_lock_killable(&vcpu->mutex))
8753 		return;
8754 	vcpu_load(vcpu);
8755 	msr.data = 0x0;
8756 	msr.index = MSR_IA32_TSC;
8757 	msr.host_initiated = true;
8758 	kvm_write_tsc(vcpu, &msr);
8759 	vcpu_put(vcpu);
8760 	mutex_unlock(&vcpu->mutex);
8761 
8762 	if (!kvmclock_periodic_sync)
8763 		return;
8764 
8765 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8766 					KVMCLOCK_SYNC_PERIOD);
8767 }
8768 
8769 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8770 {
8771 	vcpu->arch.apf.msr_val = 0;
8772 
8773 	vcpu_load(vcpu);
8774 	kvm_mmu_unload(vcpu);
8775 	vcpu_put(vcpu);
8776 
8777 	kvm_x86_ops->vcpu_free(vcpu);
8778 }
8779 
8780 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8781 {
8782 	kvm_lapic_reset(vcpu, init_event);
8783 
8784 	vcpu->arch.hflags = 0;
8785 
8786 	vcpu->arch.smi_pending = 0;
8787 	vcpu->arch.smi_count = 0;
8788 	atomic_set(&vcpu->arch.nmi_queued, 0);
8789 	vcpu->arch.nmi_pending = 0;
8790 	vcpu->arch.nmi_injected = false;
8791 	kvm_clear_interrupt_queue(vcpu);
8792 	kvm_clear_exception_queue(vcpu);
8793 	vcpu->arch.exception.pending = false;
8794 
8795 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8796 	kvm_update_dr0123(vcpu);
8797 	vcpu->arch.dr6 = DR6_INIT;
8798 	kvm_update_dr6(vcpu);
8799 	vcpu->arch.dr7 = DR7_FIXED_1;
8800 	kvm_update_dr7(vcpu);
8801 
8802 	vcpu->arch.cr2 = 0;
8803 
8804 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8805 	vcpu->arch.apf.msr_val = 0;
8806 	vcpu->arch.st.msr_val = 0;
8807 
8808 	kvmclock_reset(vcpu);
8809 
8810 	kvm_clear_async_pf_completion_queue(vcpu);
8811 	kvm_async_pf_hash_reset(vcpu);
8812 	vcpu->arch.apf.halted = false;
8813 
8814 	if (kvm_mpx_supported()) {
8815 		void *mpx_state_buffer;
8816 
8817 		/*
8818 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8819 		 * called with loaded FPU and does not let userspace fix the state.
8820 		 */
8821 		if (init_event)
8822 			kvm_put_guest_fpu(vcpu);
8823 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8824 					XFEATURE_MASK_BNDREGS);
8825 		if (mpx_state_buffer)
8826 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8827 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8828 					XFEATURE_MASK_BNDCSR);
8829 		if (mpx_state_buffer)
8830 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8831 		if (init_event)
8832 			kvm_load_guest_fpu(vcpu);
8833 	}
8834 
8835 	if (!init_event) {
8836 		kvm_pmu_reset(vcpu);
8837 		vcpu->arch.smbase = 0x30000;
8838 
8839 		vcpu->arch.msr_misc_features_enables = 0;
8840 
8841 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8842 	}
8843 
8844 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8845 	vcpu->arch.regs_avail = ~0;
8846 	vcpu->arch.regs_dirty = ~0;
8847 
8848 	vcpu->arch.ia32_xss = 0;
8849 
8850 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8851 }
8852 
8853 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8854 {
8855 	struct kvm_segment cs;
8856 
8857 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8858 	cs.selector = vector << 8;
8859 	cs.base = vector << 12;
8860 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8861 	kvm_rip_write(vcpu, 0);
8862 }
8863 
8864 int kvm_arch_hardware_enable(void)
8865 {
8866 	struct kvm *kvm;
8867 	struct kvm_vcpu *vcpu;
8868 	int i;
8869 	int ret;
8870 	u64 local_tsc;
8871 	u64 max_tsc = 0;
8872 	bool stable, backwards_tsc = false;
8873 
8874 	kvm_shared_msr_cpu_online();
8875 	ret = kvm_x86_ops->hardware_enable();
8876 	if (ret != 0)
8877 		return ret;
8878 
8879 	local_tsc = rdtsc();
8880 	stable = !kvm_check_tsc_unstable();
8881 	list_for_each_entry(kvm, &vm_list, vm_list) {
8882 		kvm_for_each_vcpu(i, vcpu, kvm) {
8883 			if (!stable && vcpu->cpu == smp_processor_id())
8884 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8885 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8886 				backwards_tsc = true;
8887 				if (vcpu->arch.last_host_tsc > max_tsc)
8888 					max_tsc = vcpu->arch.last_host_tsc;
8889 			}
8890 		}
8891 	}
8892 
8893 	/*
8894 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8895 	 * platforms that reset TSC during suspend or hibernate actions, but
8896 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8897 	 * detect that condition here, which happens early in CPU bringup,
8898 	 * before any KVM threads can be running.  Unfortunately, we can't
8899 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8900 	 * enough into CPU bringup that we know how much real time has actually
8901 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8902 	 * variables that haven't been updated yet.
8903 	 *
8904 	 * So we simply find the maximum observed TSC above, then record the
8905 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8906 	 * the adjustment will be applied.  Note that we accumulate
8907 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8908 	 * gets a chance to run again.  In the event that no KVM threads get a
8909 	 * chance to run, we will miss the entire elapsed period, as we'll have
8910 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8911 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8912 	 * uniform across all VCPUs (not to mention the scenario is extremely
8913 	 * unlikely). It is possible that a second hibernate recovery happens
8914 	 * much faster than a first, causing the observed TSC here to be
8915 	 * smaller; this would require additional padding adjustment, which is
8916 	 * why we set last_host_tsc to the local tsc observed here.
8917 	 *
8918 	 * N.B. - this code below runs only on platforms with reliable TSC,
8919 	 * as that is the only way backwards_tsc is set above.  Also note
8920 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8921 	 * have the same delta_cyc adjustment applied if backwards_tsc
8922 	 * is detected.  Note further, this adjustment is only done once,
8923 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8924 	 * called multiple times (one for each physical CPU bringup).
8925 	 *
8926 	 * Platforms with unreliable TSCs don't have to deal with this, they
8927 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8928 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8929 	 * guarantee that they stay in perfect synchronization.
8930 	 */
8931 	if (backwards_tsc) {
8932 		u64 delta_cyc = max_tsc - local_tsc;
8933 		list_for_each_entry(kvm, &vm_list, vm_list) {
8934 			kvm->arch.backwards_tsc_observed = true;
8935 			kvm_for_each_vcpu(i, vcpu, kvm) {
8936 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8937 				vcpu->arch.last_host_tsc = local_tsc;
8938 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8939 			}
8940 
8941 			/*
8942 			 * We have to disable TSC offset matching.. if you were
8943 			 * booting a VM while issuing an S4 host suspend....
8944 			 * you may have some problem.  Solving this issue is
8945 			 * left as an exercise to the reader.
8946 			 */
8947 			kvm->arch.last_tsc_nsec = 0;
8948 			kvm->arch.last_tsc_write = 0;
8949 		}
8950 
8951 	}
8952 	return 0;
8953 }
8954 
8955 void kvm_arch_hardware_disable(void)
8956 {
8957 	kvm_x86_ops->hardware_disable();
8958 	drop_user_return_notifiers();
8959 }
8960 
8961 int kvm_arch_hardware_setup(void)
8962 {
8963 	int r;
8964 
8965 	r = kvm_x86_ops->hardware_setup();
8966 	if (r != 0)
8967 		return r;
8968 
8969 	if (kvm_has_tsc_control) {
8970 		/*
8971 		 * Make sure the user can only configure tsc_khz values that
8972 		 * fit into a signed integer.
8973 		 * A min value is not calculated because it will always
8974 		 * be 1 on all machines.
8975 		 */
8976 		u64 max = min(0x7fffffffULL,
8977 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8978 		kvm_max_guest_tsc_khz = max;
8979 
8980 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8981 	}
8982 
8983 	kvm_init_msr_list();
8984 	return 0;
8985 }
8986 
8987 void kvm_arch_hardware_unsetup(void)
8988 {
8989 	kvm_x86_ops->hardware_unsetup();
8990 }
8991 
8992 void kvm_arch_check_processor_compat(void *rtn)
8993 {
8994 	kvm_x86_ops->check_processor_compatibility(rtn);
8995 }
8996 
8997 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8998 {
8999 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9000 }
9001 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9002 
9003 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9004 {
9005 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9006 }
9007 
9008 struct static_key kvm_no_apic_vcpu __read_mostly;
9009 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9010 
9011 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9012 {
9013 	struct page *page;
9014 	int r;
9015 
9016 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9017 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9018 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9019 	else
9020 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9021 
9022 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9023 	if (!page) {
9024 		r = -ENOMEM;
9025 		goto fail;
9026 	}
9027 	vcpu->arch.pio_data = page_address(page);
9028 
9029 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9030 
9031 	r = kvm_mmu_create(vcpu);
9032 	if (r < 0)
9033 		goto fail_free_pio_data;
9034 
9035 	if (irqchip_in_kernel(vcpu->kvm)) {
9036 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9037 		r = kvm_create_lapic(vcpu);
9038 		if (r < 0)
9039 			goto fail_mmu_destroy;
9040 	} else
9041 		static_key_slow_inc(&kvm_no_apic_vcpu);
9042 
9043 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9044 				       GFP_KERNEL_ACCOUNT);
9045 	if (!vcpu->arch.mce_banks) {
9046 		r = -ENOMEM;
9047 		goto fail_free_lapic;
9048 	}
9049 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9050 
9051 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9052 				GFP_KERNEL_ACCOUNT)) {
9053 		r = -ENOMEM;
9054 		goto fail_free_mce_banks;
9055 	}
9056 
9057 	fx_init(vcpu);
9058 
9059 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9060 
9061 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9062 
9063 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9064 
9065 	kvm_async_pf_hash_reset(vcpu);
9066 	kvm_pmu_init(vcpu);
9067 
9068 	vcpu->arch.pending_external_vector = -1;
9069 	vcpu->arch.preempted_in_kernel = false;
9070 
9071 	kvm_hv_vcpu_init(vcpu);
9072 
9073 	return 0;
9074 
9075 fail_free_mce_banks:
9076 	kfree(vcpu->arch.mce_banks);
9077 fail_free_lapic:
9078 	kvm_free_lapic(vcpu);
9079 fail_mmu_destroy:
9080 	kvm_mmu_destroy(vcpu);
9081 fail_free_pio_data:
9082 	free_page((unsigned long)vcpu->arch.pio_data);
9083 fail:
9084 	return r;
9085 }
9086 
9087 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9088 {
9089 	int idx;
9090 
9091 	kvm_hv_vcpu_uninit(vcpu);
9092 	kvm_pmu_destroy(vcpu);
9093 	kfree(vcpu->arch.mce_banks);
9094 	kvm_free_lapic(vcpu);
9095 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9096 	kvm_mmu_destroy(vcpu);
9097 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9098 	free_page((unsigned long)vcpu->arch.pio_data);
9099 	if (!lapic_in_kernel(vcpu))
9100 		static_key_slow_dec(&kvm_no_apic_vcpu);
9101 }
9102 
9103 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9104 {
9105 	vcpu->arch.l1tf_flush_l1d = true;
9106 	kvm_x86_ops->sched_in(vcpu, cpu);
9107 }
9108 
9109 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9110 {
9111 	if (type)
9112 		return -EINVAL;
9113 
9114 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9115 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9116 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9117 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9118 
9119 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9120 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9121 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9122 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9123 		&kvm->arch.irq_sources_bitmap);
9124 
9125 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9126 	mutex_init(&kvm->arch.apic_map_lock);
9127 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9128 
9129 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9130 	pvclock_update_vm_gtod_copy(kvm);
9131 
9132 	kvm->arch.guest_can_read_msr_platform_info = true;
9133 
9134 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9135 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9136 
9137 	kvm_hv_init_vm(kvm);
9138 	kvm_page_track_init(kvm);
9139 	kvm_mmu_init_vm(kvm);
9140 
9141 	if (kvm_x86_ops->vm_init)
9142 		return kvm_x86_ops->vm_init(kvm);
9143 
9144 	return 0;
9145 }
9146 
9147 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9148 {
9149 	vcpu_load(vcpu);
9150 	kvm_mmu_unload(vcpu);
9151 	vcpu_put(vcpu);
9152 }
9153 
9154 static void kvm_free_vcpus(struct kvm *kvm)
9155 {
9156 	unsigned int i;
9157 	struct kvm_vcpu *vcpu;
9158 
9159 	/*
9160 	 * Unpin any mmu pages first.
9161 	 */
9162 	kvm_for_each_vcpu(i, vcpu, kvm) {
9163 		kvm_clear_async_pf_completion_queue(vcpu);
9164 		kvm_unload_vcpu_mmu(vcpu);
9165 	}
9166 	kvm_for_each_vcpu(i, vcpu, kvm)
9167 		kvm_arch_vcpu_free(vcpu);
9168 
9169 	mutex_lock(&kvm->lock);
9170 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9171 		kvm->vcpus[i] = NULL;
9172 
9173 	atomic_set(&kvm->online_vcpus, 0);
9174 	mutex_unlock(&kvm->lock);
9175 }
9176 
9177 void kvm_arch_sync_events(struct kvm *kvm)
9178 {
9179 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9180 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9181 	kvm_free_pit(kvm);
9182 }
9183 
9184 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9185 {
9186 	int i, r;
9187 	unsigned long hva;
9188 	struct kvm_memslots *slots = kvm_memslots(kvm);
9189 	struct kvm_memory_slot *slot, old;
9190 
9191 	/* Called with kvm->slots_lock held.  */
9192 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9193 		return -EINVAL;
9194 
9195 	slot = id_to_memslot(slots, id);
9196 	if (size) {
9197 		if (slot->npages)
9198 			return -EEXIST;
9199 
9200 		/*
9201 		 * MAP_SHARED to prevent internal slot pages from being moved
9202 		 * by fork()/COW.
9203 		 */
9204 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9205 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9206 		if (IS_ERR((void *)hva))
9207 			return PTR_ERR((void *)hva);
9208 	} else {
9209 		if (!slot->npages)
9210 			return 0;
9211 
9212 		hva = 0;
9213 	}
9214 
9215 	old = *slot;
9216 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9217 		struct kvm_userspace_memory_region m;
9218 
9219 		m.slot = id | (i << 16);
9220 		m.flags = 0;
9221 		m.guest_phys_addr = gpa;
9222 		m.userspace_addr = hva;
9223 		m.memory_size = size;
9224 		r = __kvm_set_memory_region(kvm, &m);
9225 		if (r < 0)
9226 			return r;
9227 	}
9228 
9229 	if (!size)
9230 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9231 
9232 	return 0;
9233 }
9234 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9235 
9236 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9237 {
9238 	int r;
9239 
9240 	mutex_lock(&kvm->slots_lock);
9241 	r = __x86_set_memory_region(kvm, id, gpa, size);
9242 	mutex_unlock(&kvm->slots_lock);
9243 
9244 	return r;
9245 }
9246 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9247 
9248 void kvm_arch_destroy_vm(struct kvm *kvm)
9249 {
9250 	if (current->mm == kvm->mm) {
9251 		/*
9252 		 * Free memory regions allocated on behalf of userspace,
9253 		 * unless the the memory map has changed due to process exit
9254 		 * or fd copying.
9255 		 */
9256 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9257 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9258 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9259 	}
9260 	if (kvm_x86_ops->vm_destroy)
9261 		kvm_x86_ops->vm_destroy(kvm);
9262 	kvm_pic_destroy(kvm);
9263 	kvm_ioapic_destroy(kvm);
9264 	kvm_free_vcpus(kvm);
9265 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9266 	kvm_mmu_uninit_vm(kvm);
9267 	kvm_page_track_cleanup(kvm);
9268 	kvm_hv_destroy_vm(kvm);
9269 }
9270 
9271 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9272 			   struct kvm_memory_slot *dont)
9273 {
9274 	int i;
9275 
9276 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9277 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9278 			kvfree(free->arch.rmap[i]);
9279 			free->arch.rmap[i] = NULL;
9280 		}
9281 		if (i == 0)
9282 			continue;
9283 
9284 		if (!dont || free->arch.lpage_info[i - 1] !=
9285 			     dont->arch.lpage_info[i - 1]) {
9286 			kvfree(free->arch.lpage_info[i - 1]);
9287 			free->arch.lpage_info[i - 1] = NULL;
9288 		}
9289 	}
9290 
9291 	kvm_page_track_free_memslot(free, dont);
9292 }
9293 
9294 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9295 			    unsigned long npages)
9296 {
9297 	int i;
9298 
9299 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9300 		struct kvm_lpage_info *linfo;
9301 		unsigned long ugfn;
9302 		int lpages;
9303 		int level = i + 1;
9304 
9305 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9306 				      slot->base_gfn, level) + 1;
9307 
9308 		slot->arch.rmap[i] =
9309 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9310 				 GFP_KERNEL_ACCOUNT);
9311 		if (!slot->arch.rmap[i])
9312 			goto out_free;
9313 		if (i == 0)
9314 			continue;
9315 
9316 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9317 		if (!linfo)
9318 			goto out_free;
9319 
9320 		slot->arch.lpage_info[i - 1] = linfo;
9321 
9322 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9323 			linfo[0].disallow_lpage = 1;
9324 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9325 			linfo[lpages - 1].disallow_lpage = 1;
9326 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9327 		/*
9328 		 * If the gfn and userspace address are not aligned wrt each
9329 		 * other, or if explicitly asked to, disable large page
9330 		 * support for this slot
9331 		 */
9332 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9333 		    !kvm_largepages_enabled()) {
9334 			unsigned long j;
9335 
9336 			for (j = 0; j < lpages; ++j)
9337 				linfo[j].disallow_lpage = 1;
9338 		}
9339 	}
9340 
9341 	if (kvm_page_track_create_memslot(slot, npages))
9342 		goto out_free;
9343 
9344 	return 0;
9345 
9346 out_free:
9347 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9348 		kvfree(slot->arch.rmap[i]);
9349 		slot->arch.rmap[i] = NULL;
9350 		if (i == 0)
9351 			continue;
9352 
9353 		kvfree(slot->arch.lpage_info[i - 1]);
9354 		slot->arch.lpage_info[i - 1] = NULL;
9355 	}
9356 	return -ENOMEM;
9357 }
9358 
9359 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9360 {
9361 	/*
9362 	 * memslots->generation has been incremented.
9363 	 * mmio generation may have reached its maximum value.
9364 	 */
9365 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9366 }
9367 
9368 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9369 				struct kvm_memory_slot *memslot,
9370 				const struct kvm_userspace_memory_region *mem,
9371 				enum kvm_mr_change change)
9372 {
9373 	return 0;
9374 }
9375 
9376 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9377 				     struct kvm_memory_slot *new)
9378 {
9379 	/* Still write protect RO slot */
9380 	if (new->flags & KVM_MEM_READONLY) {
9381 		kvm_mmu_slot_remove_write_access(kvm, new);
9382 		return;
9383 	}
9384 
9385 	/*
9386 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9387 	 *
9388 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9389 	 *
9390 	 *  - KVM_MR_CREATE with dirty logging is disabled
9391 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9392 	 *
9393 	 * The reason is, in case of PML, we need to set D-bit for any slots
9394 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9395 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9396 	 * guarantees leaving PML enabled during guest's lifetime won't have
9397 	 * any additional overhead from PML when guest is running with dirty
9398 	 * logging disabled for memory slots.
9399 	 *
9400 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9401 	 * to dirty logging mode.
9402 	 *
9403 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9404 	 *
9405 	 * In case of write protect:
9406 	 *
9407 	 * Write protect all pages for dirty logging.
9408 	 *
9409 	 * All the sptes including the large sptes which point to this
9410 	 * slot are set to readonly. We can not create any new large
9411 	 * spte on this slot until the end of the logging.
9412 	 *
9413 	 * See the comments in fast_page_fault().
9414 	 */
9415 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9416 		if (kvm_x86_ops->slot_enable_log_dirty)
9417 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9418 		else
9419 			kvm_mmu_slot_remove_write_access(kvm, new);
9420 	} else {
9421 		if (kvm_x86_ops->slot_disable_log_dirty)
9422 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9423 	}
9424 }
9425 
9426 void kvm_arch_commit_memory_region(struct kvm *kvm,
9427 				const struct kvm_userspace_memory_region *mem,
9428 				const struct kvm_memory_slot *old,
9429 				const struct kvm_memory_slot *new,
9430 				enum kvm_mr_change change)
9431 {
9432 	int nr_mmu_pages = 0;
9433 
9434 	if (!kvm->arch.n_requested_mmu_pages)
9435 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9436 
9437 	if (nr_mmu_pages)
9438 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9439 
9440 	/*
9441 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9442 	 * sptes have to be split.  If live migration is successful, the guest
9443 	 * in the source machine will be destroyed and large sptes will be
9444 	 * created in the destination. However, if the guest continues to run
9445 	 * in the source machine (for example if live migration fails), small
9446 	 * sptes will remain around and cause bad performance.
9447 	 *
9448 	 * Scan sptes if dirty logging has been stopped, dropping those
9449 	 * which can be collapsed into a single large-page spte.  Later
9450 	 * page faults will create the large-page sptes.
9451 	 */
9452 	if ((change != KVM_MR_DELETE) &&
9453 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9454 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9455 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9456 
9457 	/*
9458 	 * Set up write protection and/or dirty logging for the new slot.
9459 	 *
9460 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9461 	 * been zapped so no dirty logging staff is needed for old slot. For
9462 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9463 	 * new and it's also covered when dealing with the new slot.
9464 	 *
9465 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9466 	 */
9467 	if (change != KVM_MR_DELETE)
9468 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9469 }
9470 
9471 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9472 {
9473 	kvm_mmu_zap_all(kvm);
9474 }
9475 
9476 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9477 				   struct kvm_memory_slot *slot)
9478 {
9479 	kvm_page_track_flush_slot(kvm, slot);
9480 }
9481 
9482 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9483 {
9484 	return (is_guest_mode(vcpu) &&
9485 			kvm_x86_ops->guest_apic_has_interrupt &&
9486 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9487 }
9488 
9489 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9490 {
9491 	if (!list_empty_careful(&vcpu->async_pf.done))
9492 		return true;
9493 
9494 	if (kvm_apic_has_events(vcpu))
9495 		return true;
9496 
9497 	if (vcpu->arch.pv.pv_unhalted)
9498 		return true;
9499 
9500 	if (vcpu->arch.exception.pending)
9501 		return true;
9502 
9503 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9504 	    (vcpu->arch.nmi_pending &&
9505 	     kvm_x86_ops->nmi_allowed(vcpu)))
9506 		return true;
9507 
9508 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9509 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9510 		return true;
9511 
9512 	if (kvm_arch_interrupt_allowed(vcpu) &&
9513 	    (kvm_cpu_has_interrupt(vcpu) ||
9514 	    kvm_guest_apic_has_interrupt(vcpu)))
9515 		return true;
9516 
9517 	if (kvm_hv_has_stimer_pending(vcpu))
9518 		return true;
9519 
9520 	return false;
9521 }
9522 
9523 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9524 {
9525 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9526 }
9527 
9528 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9529 {
9530 	return vcpu->arch.preempted_in_kernel;
9531 }
9532 
9533 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9534 {
9535 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9536 }
9537 
9538 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9539 {
9540 	return kvm_x86_ops->interrupt_allowed(vcpu);
9541 }
9542 
9543 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9544 {
9545 	if (is_64_bit_mode(vcpu))
9546 		return kvm_rip_read(vcpu);
9547 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9548 		     kvm_rip_read(vcpu));
9549 }
9550 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9551 
9552 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9553 {
9554 	return kvm_get_linear_rip(vcpu) == linear_rip;
9555 }
9556 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9557 
9558 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9559 {
9560 	unsigned long rflags;
9561 
9562 	rflags = kvm_x86_ops->get_rflags(vcpu);
9563 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9564 		rflags &= ~X86_EFLAGS_TF;
9565 	return rflags;
9566 }
9567 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9568 
9569 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9570 {
9571 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9572 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9573 		rflags |= X86_EFLAGS_TF;
9574 	kvm_x86_ops->set_rflags(vcpu, rflags);
9575 }
9576 
9577 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9578 {
9579 	__kvm_set_rflags(vcpu, rflags);
9580 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9581 }
9582 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9583 
9584 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9585 {
9586 	int r;
9587 
9588 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9589 	      work->wakeup_all)
9590 		return;
9591 
9592 	r = kvm_mmu_reload(vcpu);
9593 	if (unlikely(r))
9594 		return;
9595 
9596 	if (!vcpu->arch.mmu->direct_map &&
9597 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9598 		return;
9599 
9600 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9601 }
9602 
9603 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9604 {
9605 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9606 }
9607 
9608 static inline u32 kvm_async_pf_next_probe(u32 key)
9609 {
9610 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9611 }
9612 
9613 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9614 {
9615 	u32 key = kvm_async_pf_hash_fn(gfn);
9616 
9617 	while (vcpu->arch.apf.gfns[key] != ~0)
9618 		key = kvm_async_pf_next_probe(key);
9619 
9620 	vcpu->arch.apf.gfns[key] = gfn;
9621 }
9622 
9623 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9624 {
9625 	int i;
9626 	u32 key = kvm_async_pf_hash_fn(gfn);
9627 
9628 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9629 		     (vcpu->arch.apf.gfns[key] != gfn &&
9630 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9631 		key = kvm_async_pf_next_probe(key);
9632 
9633 	return key;
9634 }
9635 
9636 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9637 {
9638 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9639 }
9640 
9641 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9642 {
9643 	u32 i, j, k;
9644 
9645 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9646 	while (true) {
9647 		vcpu->arch.apf.gfns[i] = ~0;
9648 		do {
9649 			j = kvm_async_pf_next_probe(j);
9650 			if (vcpu->arch.apf.gfns[j] == ~0)
9651 				return;
9652 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9653 			/*
9654 			 * k lies cyclically in ]i,j]
9655 			 * |    i.k.j |
9656 			 * |....j i.k.| or  |.k..j i...|
9657 			 */
9658 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9659 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9660 		i = j;
9661 	}
9662 }
9663 
9664 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9665 {
9666 
9667 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9668 				      sizeof(val));
9669 }
9670 
9671 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9672 {
9673 
9674 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9675 				      sizeof(u32));
9676 }
9677 
9678 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9679 				     struct kvm_async_pf *work)
9680 {
9681 	struct x86_exception fault;
9682 
9683 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9684 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9685 
9686 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9687 	    (vcpu->arch.apf.send_user_only &&
9688 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9689 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9690 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9691 		fault.vector = PF_VECTOR;
9692 		fault.error_code_valid = true;
9693 		fault.error_code = 0;
9694 		fault.nested_page_fault = false;
9695 		fault.address = work->arch.token;
9696 		fault.async_page_fault = true;
9697 		kvm_inject_page_fault(vcpu, &fault);
9698 	}
9699 }
9700 
9701 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9702 				 struct kvm_async_pf *work)
9703 {
9704 	struct x86_exception fault;
9705 	u32 val;
9706 
9707 	if (work->wakeup_all)
9708 		work->arch.token = ~0; /* broadcast wakeup */
9709 	else
9710 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9711 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9712 
9713 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9714 	    !apf_get_user(vcpu, &val)) {
9715 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9716 		    vcpu->arch.exception.pending &&
9717 		    vcpu->arch.exception.nr == PF_VECTOR &&
9718 		    !apf_put_user(vcpu, 0)) {
9719 			vcpu->arch.exception.injected = false;
9720 			vcpu->arch.exception.pending = false;
9721 			vcpu->arch.exception.nr = 0;
9722 			vcpu->arch.exception.has_error_code = false;
9723 			vcpu->arch.exception.error_code = 0;
9724 			vcpu->arch.exception.has_payload = false;
9725 			vcpu->arch.exception.payload = 0;
9726 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9727 			fault.vector = PF_VECTOR;
9728 			fault.error_code_valid = true;
9729 			fault.error_code = 0;
9730 			fault.nested_page_fault = false;
9731 			fault.address = work->arch.token;
9732 			fault.async_page_fault = true;
9733 			kvm_inject_page_fault(vcpu, &fault);
9734 		}
9735 	}
9736 	vcpu->arch.apf.halted = false;
9737 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9738 }
9739 
9740 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9741 {
9742 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9743 		return true;
9744 	else
9745 		return kvm_can_do_async_pf(vcpu);
9746 }
9747 
9748 void kvm_arch_start_assignment(struct kvm *kvm)
9749 {
9750 	atomic_inc(&kvm->arch.assigned_device_count);
9751 }
9752 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9753 
9754 void kvm_arch_end_assignment(struct kvm *kvm)
9755 {
9756 	atomic_dec(&kvm->arch.assigned_device_count);
9757 }
9758 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9759 
9760 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9761 {
9762 	return atomic_read(&kvm->arch.assigned_device_count);
9763 }
9764 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9765 
9766 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9767 {
9768 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9769 }
9770 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9771 
9772 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9773 {
9774 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9775 }
9776 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9777 
9778 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9779 {
9780 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9781 }
9782 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9783 
9784 bool kvm_arch_has_irq_bypass(void)
9785 {
9786 	return kvm_x86_ops->update_pi_irte != NULL;
9787 }
9788 
9789 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9790 				      struct irq_bypass_producer *prod)
9791 {
9792 	struct kvm_kernel_irqfd *irqfd =
9793 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9794 
9795 	irqfd->producer = prod;
9796 
9797 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9798 					   prod->irq, irqfd->gsi, 1);
9799 }
9800 
9801 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9802 				      struct irq_bypass_producer *prod)
9803 {
9804 	int ret;
9805 	struct kvm_kernel_irqfd *irqfd =
9806 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9807 
9808 	WARN_ON(irqfd->producer != prod);
9809 	irqfd->producer = NULL;
9810 
9811 	/*
9812 	 * When producer of consumer is unregistered, we change back to
9813 	 * remapped mode, so we can re-use the current implementation
9814 	 * when the irq is masked/disabled or the consumer side (KVM
9815 	 * int this case doesn't want to receive the interrupts.
9816 	*/
9817 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9818 	if (ret)
9819 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9820 		       " fails: %d\n", irqfd->consumer.token, ret);
9821 }
9822 
9823 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9824 				   uint32_t guest_irq, bool set)
9825 {
9826 	if (!kvm_x86_ops->update_pi_irte)
9827 		return -EINVAL;
9828 
9829 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9830 }
9831 
9832 bool kvm_vector_hashing_enabled(void)
9833 {
9834 	return vector_hashing;
9835 }
9836 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9837 
9838 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9839 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9840 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9841 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9842 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9843 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9844 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9855 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9856 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9857