xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 367e5927)
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,
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_ARCH_CAPABILITIES,
1162 	MSR_IA32_MISC_ENABLE,
1163 	MSR_IA32_MCG_STATUS,
1164 	MSR_IA32_MCG_CTL,
1165 	MSR_IA32_MCG_EXT_CTL,
1166 	MSR_IA32_SMBASE,
1167 	MSR_SMI_COUNT,
1168 	MSR_PLATFORM_INFO,
1169 	MSR_MISC_FEATURES_ENABLES,
1170 	MSR_AMD64_VIRT_SPEC_CTRL,
1171 };
1172 
1173 static unsigned num_emulated_msrs;
1174 
1175 /*
1176  * List of msr numbers which are used to expose MSR-based features that
1177  * can be used by a hypervisor to validate requested CPU features.
1178  */
1179 static u32 msr_based_features[] = {
1180 	MSR_IA32_VMX_BASIC,
1181 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1182 	MSR_IA32_VMX_PINBASED_CTLS,
1183 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1184 	MSR_IA32_VMX_PROCBASED_CTLS,
1185 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1186 	MSR_IA32_VMX_EXIT_CTLS,
1187 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1188 	MSR_IA32_VMX_ENTRY_CTLS,
1189 	MSR_IA32_VMX_MISC,
1190 	MSR_IA32_VMX_CR0_FIXED0,
1191 	MSR_IA32_VMX_CR0_FIXED1,
1192 	MSR_IA32_VMX_CR4_FIXED0,
1193 	MSR_IA32_VMX_CR4_FIXED1,
1194 	MSR_IA32_VMX_VMCS_ENUM,
1195 	MSR_IA32_VMX_PROCBASED_CTLS2,
1196 	MSR_IA32_VMX_EPT_VPID_CAP,
1197 	MSR_IA32_VMX_VMFUNC,
1198 
1199 	MSR_F10H_DECFG,
1200 	MSR_IA32_UCODE_REV,
1201 	MSR_IA32_ARCH_CAPABILITIES,
1202 };
1203 
1204 static unsigned int num_msr_based_features;
1205 
1206 u64 kvm_get_arch_capabilities(void)
1207 {
1208 	u64 data;
1209 
1210 	rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1211 
1212 	/*
1213 	 * If we're doing cache flushes (either "always" or "cond")
1214 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1215 	 * If an outer hypervisor is doing the cache flush for us
1216 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1217 	 * capability to the guest too, and if EPT is disabled we're not
1218 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1219 	 * require a nested hypervisor to do a flush of its own.
1220 	 */
1221 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1222 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1223 
1224 	return data;
1225 }
1226 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1227 
1228 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1229 {
1230 	switch (msr->index) {
1231 	case MSR_IA32_ARCH_CAPABILITIES:
1232 		msr->data = kvm_get_arch_capabilities();
1233 		break;
1234 	case MSR_IA32_UCODE_REV:
1235 		rdmsrl_safe(msr->index, &msr->data);
1236 		break;
1237 	default:
1238 		if (kvm_x86_ops->get_msr_feature(msr))
1239 			return 1;
1240 	}
1241 	return 0;
1242 }
1243 
1244 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1245 {
1246 	struct kvm_msr_entry msr;
1247 	int r;
1248 
1249 	msr.index = index;
1250 	r = kvm_get_msr_feature(&msr);
1251 	if (r)
1252 		return r;
1253 
1254 	*data = msr.data;
1255 
1256 	return 0;
1257 }
1258 
1259 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1260 {
1261 	if (efer & efer_reserved_bits)
1262 		return false;
1263 
1264 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1265 			return false;
1266 
1267 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1268 			return false;
1269 
1270 	return true;
1271 }
1272 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1273 
1274 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1275 {
1276 	u64 old_efer = vcpu->arch.efer;
1277 
1278 	if (!kvm_valid_efer(vcpu, efer))
1279 		return 1;
1280 
1281 	if (is_paging(vcpu)
1282 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1283 		return 1;
1284 
1285 	efer &= ~EFER_LMA;
1286 	efer |= vcpu->arch.efer & EFER_LMA;
1287 
1288 	kvm_x86_ops->set_efer(vcpu, efer);
1289 
1290 	/* Update reserved bits */
1291 	if ((efer ^ old_efer) & EFER_NX)
1292 		kvm_mmu_reset_context(vcpu);
1293 
1294 	return 0;
1295 }
1296 
1297 void kvm_enable_efer_bits(u64 mask)
1298 {
1299        efer_reserved_bits &= ~mask;
1300 }
1301 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1302 
1303 /*
1304  * Writes msr value into into the appropriate "register".
1305  * Returns 0 on success, non-0 otherwise.
1306  * Assumes vcpu_load() was already called.
1307  */
1308 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1309 {
1310 	switch (msr->index) {
1311 	case MSR_FS_BASE:
1312 	case MSR_GS_BASE:
1313 	case MSR_KERNEL_GS_BASE:
1314 	case MSR_CSTAR:
1315 	case MSR_LSTAR:
1316 		if (is_noncanonical_address(msr->data, vcpu))
1317 			return 1;
1318 		break;
1319 	case MSR_IA32_SYSENTER_EIP:
1320 	case MSR_IA32_SYSENTER_ESP:
1321 		/*
1322 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1323 		 * non-canonical address is written on Intel but not on
1324 		 * AMD (which ignores the top 32-bits, because it does
1325 		 * not implement 64-bit SYSENTER).
1326 		 *
1327 		 * 64-bit code should hence be able to write a non-canonical
1328 		 * value on AMD.  Making the address canonical ensures that
1329 		 * vmentry does not fail on Intel after writing a non-canonical
1330 		 * value, and that something deterministic happens if the guest
1331 		 * invokes 64-bit SYSENTER.
1332 		 */
1333 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1334 	}
1335 	return kvm_x86_ops->set_msr(vcpu, msr);
1336 }
1337 EXPORT_SYMBOL_GPL(kvm_set_msr);
1338 
1339 /*
1340  * Adapt set_msr() to msr_io()'s calling convention
1341  */
1342 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1343 {
1344 	struct msr_data msr;
1345 	int r;
1346 
1347 	msr.index = index;
1348 	msr.host_initiated = true;
1349 	r = kvm_get_msr(vcpu, &msr);
1350 	if (r)
1351 		return r;
1352 
1353 	*data = msr.data;
1354 	return 0;
1355 }
1356 
1357 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1358 {
1359 	struct msr_data msr;
1360 
1361 	msr.data = *data;
1362 	msr.index = index;
1363 	msr.host_initiated = true;
1364 	return kvm_set_msr(vcpu, &msr);
1365 }
1366 
1367 #ifdef CONFIG_X86_64
1368 struct pvclock_gtod_data {
1369 	seqcount_t	seq;
1370 
1371 	struct { /* extract of a clocksource struct */
1372 		int vclock_mode;
1373 		u64	cycle_last;
1374 		u64	mask;
1375 		u32	mult;
1376 		u32	shift;
1377 	} clock;
1378 
1379 	u64		boot_ns;
1380 	u64		nsec_base;
1381 	u64		wall_time_sec;
1382 };
1383 
1384 static struct pvclock_gtod_data pvclock_gtod_data;
1385 
1386 static void update_pvclock_gtod(struct timekeeper *tk)
1387 {
1388 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1389 	u64 boot_ns;
1390 
1391 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1392 
1393 	write_seqcount_begin(&vdata->seq);
1394 
1395 	/* copy pvclock gtod data */
1396 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1397 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1398 	vdata->clock.mask		= tk->tkr_mono.mask;
1399 	vdata->clock.mult		= tk->tkr_mono.mult;
1400 	vdata->clock.shift		= tk->tkr_mono.shift;
1401 
1402 	vdata->boot_ns			= boot_ns;
1403 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1404 
1405 	vdata->wall_time_sec            = tk->xtime_sec;
1406 
1407 	write_seqcount_end(&vdata->seq);
1408 }
1409 #endif
1410 
1411 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1412 {
1413 	/*
1414 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1415 	 * vcpu_enter_guest.  This function is only called from
1416 	 * the physical CPU that is running vcpu.
1417 	 */
1418 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1419 }
1420 
1421 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1422 {
1423 	int version;
1424 	int r;
1425 	struct pvclock_wall_clock wc;
1426 	struct timespec64 boot;
1427 
1428 	if (!wall_clock)
1429 		return;
1430 
1431 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1432 	if (r)
1433 		return;
1434 
1435 	if (version & 1)
1436 		++version;  /* first time write, random junk */
1437 
1438 	++version;
1439 
1440 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1441 		return;
1442 
1443 	/*
1444 	 * The guest calculates current wall clock time by adding
1445 	 * system time (updated by kvm_guest_time_update below) to the
1446 	 * wall clock specified here.  guest system time equals host
1447 	 * system time for us, thus we must fill in host boot time here.
1448 	 */
1449 	getboottime64(&boot);
1450 
1451 	if (kvm->arch.kvmclock_offset) {
1452 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1453 		boot = timespec64_sub(boot, ts);
1454 	}
1455 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1456 	wc.nsec = boot.tv_nsec;
1457 	wc.version = version;
1458 
1459 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1460 
1461 	version++;
1462 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1463 }
1464 
1465 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1466 {
1467 	do_shl32_div32(dividend, divisor);
1468 	return dividend;
1469 }
1470 
1471 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1472 			       s8 *pshift, u32 *pmultiplier)
1473 {
1474 	uint64_t scaled64;
1475 	int32_t  shift = 0;
1476 	uint64_t tps64;
1477 	uint32_t tps32;
1478 
1479 	tps64 = base_hz;
1480 	scaled64 = scaled_hz;
1481 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1482 		tps64 >>= 1;
1483 		shift--;
1484 	}
1485 
1486 	tps32 = (uint32_t)tps64;
1487 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1488 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1489 			scaled64 >>= 1;
1490 		else
1491 			tps32 <<= 1;
1492 		shift++;
1493 	}
1494 
1495 	*pshift = shift;
1496 	*pmultiplier = div_frac(scaled64, tps32);
1497 
1498 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1499 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1500 }
1501 
1502 #ifdef CONFIG_X86_64
1503 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1504 #endif
1505 
1506 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1507 static unsigned long max_tsc_khz;
1508 
1509 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1510 {
1511 	u64 v = (u64)khz * (1000000 + ppm);
1512 	do_div(v, 1000000);
1513 	return v;
1514 }
1515 
1516 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1517 {
1518 	u64 ratio;
1519 
1520 	/* Guest TSC same frequency as host TSC? */
1521 	if (!scale) {
1522 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1523 		return 0;
1524 	}
1525 
1526 	/* TSC scaling supported? */
1527 	if (!kvm_has_tsc_control) {
1528 		if (user_tsc_khz > tsc_khz) {
1529 			vcpu->arch.tsc_catchup = 1;
1530 			vcpu->arch.tsc_always_catchup = 1;
1531 			return 0;
1532 		} else {
1533 			WARN(1, "user requested TSC rate below hardware speed\n");
1534 			return -1;
1535 		}
1536 	}
1537 
1538 	/* TSC scaling required  - calculate ratio */
1539 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1540 				user_tsc_khz, tsc_khz);
1541 
1542 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1543 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1544 			  user_tsc_khz);
1545 		return -1;
1546 	}
1547 
1548 	vcpu->arch.tsc_scaling_ratio = ratio;
1549 	return 0;
1550 }
1551 
1552 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1553 {
1554 	u32 thresh_lo, thresh_hi;
1555 	int use_scaling = 0;
1556 
1557 	/* tsc_khz can be zero if TSC calibration fails */
1558 	if (user_tsc_khz == 0) {
1559 		/* set tsc_scaling_ratio to a safe value */
1560 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1561 		return -1;
1562 	}
1563 
1564 	/* Compute a scale to convert nanoseconds in TSC cycles */
1565 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1566 			   &vcpu->arch.virtual_tsc_shift,
1567 			   &vcpu->arch.virtual_tsc_mult);
1568 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1569 
1570 	/*
1571 	 * Compute the variation in TSC rate which is acceptable
1572 	 * within the range of tolerance and decide if the
1573 	 * rate being applied is within that bounds of the hardware
1574 	 * rate.  If so, no scaling or compensation need be done.
1575 	 */
1576 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1577 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1578 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1579 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1580 		use_scaling = 1;
1581 	}
1582 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1583 }
1584 
1585 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1586 {
1587 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1588 				      vcpu->arch.virtual_tsc_mult,
1589 				      vcpu->arch.virtual_tsc_shift);
1590 	tsc += vcpu->arch.this_tsc_write;
1591 	return tsc;
1592 }
1593 
1594 static inline int gtod_is_based_on_tsc(int mode)
1595 {
1596 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1597 }
1598 
1599 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1600 {
1601 #ifdef CONFIG_X86_64
1602 	bool vcpus_matched;
1603 	struct kvm_arch *ka = &vcpu->kvm->arch;
1604 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1605 
1606 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1607 			 atomic_read(&vcpu->kvm->online_vcpus));
1608 
1609 	/*
1610 	 * Once the masterclock is enabled, always perform request in
1611 	 * order to update it.
1612 	 *
1613 	 * In order to enable masterclock, the host clocksource must be TSC
1614 	 * and the vcpus need to have matched TSCs.  When that happens,
1615 	 * perform request to enable masterclock.
1616 	 */
1617 	if (ka->use_master_clock ||
1618 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1619 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1620 
1621 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1622 			    atomic_read(&vcpu->kvm->online_vcpus),
1623 		            ka->use_master_clock, gtod->clock.vclock_mode);
1624 #endif
1625 }
1626 
1627 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1628 {
1629 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1630 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1631 }
1632 
1633 /*
1634  * Multiply tsc by a fixed point number represented by ratio.
1635  *
1636  * The most significant 64-N bits (mult) of ratio represent the
1637  * integral part of the fixed point number; the remaining N bits
1638  * (frac) represent the fractional part, ie. ratio represents a fixed
1639  * point number (mult + frac * 2^(-N)).
1640  *
1641  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1642  */
1643 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1644 {
1645 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1646 }
1647 
1648 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1649 {
1650 	u64 _tsc = tsc;
1651 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1652 
1653 	if (ratio != kvm_default_tsc_scaling_ratio)
1654 		_tsc = __scale_tsc(ratio, tsc);
1655 
1656 	return _tsc;
1657 }
1658 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1659 
1660 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1661 {
1662 	u64 tsc;
1663 
1664 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1665 
1666 	return target_tsc - tsc;
1667 }
1668 
1669 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1670 {
1671 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1672 
1673 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1674 }
1675 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1676 
1677 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1678 {
1679 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1680 }
1681 
1682 static inline bool kvm_check_tsc_unstable(void)
1683 {
1684 #ifdef CONFIG_X86_64
1685 	/*
1686 	 * TSC is marked unstable when we're running on Hyper-V,
1687 	 * 'TSC page' clocksource is good.
1688 	 */
1689 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1690 		return false;
1691 #endif
1692 	return check_tsc_unstable();
1693 }
1694 
1695 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1696 {
1697 	struct kvm *kvm = vcpu->kvm;
1698 	u64 offset, ns, elapsed;
1699 	unsigned long flags;
1700 	bool matched;
1701 	bool already_matched;
1702 	u64 data = msr->data;
1703 	bool synchronizing = false;
1704 
1705 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1706 	offset = kvm_compute_tsc_offset(vcpu, data);
1707 	ns = ktime_get_boot_ns();
1708 	elapsed = ns - kvm->arch.last_tsc_nsec;
1709 
1710 	if (vcpu->arch.virtual_tsc_khz) {
1711 		if (data == 0 && msr->host_initiated) {
1712 			/*
1713 			 * detection of vcpu initialization -- need to sync
1714 			 * with other vCPUs. This particularly helps to keep
1715 			 * kvm_clock stable after CPU hotplug
1716 			 */
1717 			synchronizing = true;
1718 		} else {
1719 			u64 tsc_exp = kvm->arch.last_tsc_write +
1720 						nsec_to_cycles(vcpu, elapsed);
1721 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1722 			/*
1723 			 * Special case: TSC write with a small delta (1 second)
1724 			 * of virtual cycle time against real time is
1725 			 * interpreted as an attempt to synchronize the CPU.
1726 			 */
1727 			synchronizing = data < tsc_exp + tsc_hz &&
1728 					data + tsc_hz > tsc_exp;
1729 		}
1730 	}
1731 
1732 	/*
1733 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1734 	 * TSC, we add elapsed time in this computation.  We could let the
1735 	 * compensation code attempt to catch up if we fall behind, but
1736 	 * it's better to try to match offsets from the beginning.
1737          */
1738 	if (synchronizing &&
1739 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1740 		if (!kvm_check_tsc_unstable()) {
1741 			offset = kvm->arch.cur_tsc_offset;
1742 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1743 		} else {
1744 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1745 			data += delta;
1746 			offset = kvm_compute_tsc_offset(vcpu, data);
1747 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1748 		}
1749 		matched = true;
1750 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1751 	} else {
1752 		/*
1753 		 * We split periods of matched TSC writes into generations.
1754 		 * For each generation, we track the original measured
1755 		 * nanosecond time, offset, and write, so if TSCs are in
1756 		 * sync, we can match exact offset, and if not, we can match
1757 		 * exact software computation in compute_guest_tsc()
1758 		 *
1759 		 * These values are tracked in kvm->arch.cur_xxx variables.
1760 		 */
1761 		kvm->arch.cur_tsc_generation++;
1762 		kvm->arch.cur_tsc_nsec = ns;
1763 		kvm->arch.cur_tsc_write = data;
1764 		kvm->arch.cur_tsc_offset = offset;
1765 		matched = false;
1766 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1767 			 kvm->arch.cur_tsc_generation, data);
1768 	}
1769 
1770 	/*
1771 	 * We also track th most recent recorded KHZ, write and time to
1772 	 * allow the matching interval to be extended at each write.
1773 	 */
1774 	kvm->arch.last_tsc_nsec = ns;
1775 	kvm->arch.last_tsc_write = data;
1776 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1777 
1778 	vcpu->arch.last_guest_tsc = data;
1779 
1780 	/* Keep track of which generation this VCPU has synchronized to */
1781 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1782 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1783 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1784 
1785 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1786 		update_ia32_tsc_adjust_msr(vcpu, offset);
1787 
1788 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1789 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1790 
1791 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1792 	if (!matched) {
1793 		kvm->arch.nr_vcpus_matched_tsc = 0;
1794 	} else if (!already_matched) {
1795 		kvm->arch.nr_vcpus_matched_tsc++;
1796 	}
1797 
1798 	kvm_track_tsc_matching(vcpu);
1799 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1800 }
1801 
1802 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1803 
1804 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1805 					   s64 adjustment)
1806 {
1807 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1808 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1809 }
1810 
1811 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1812 {
1813 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1814 		WARN_ON(adjustment < 0);
1815 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1816 	adjust_tsc_offset_guest(vcpu, adjustment);
1817 }
1818 
1819 #ifdef CONFIG_X86_64
1820 
1821 static u64 read_tsc(void)
1822 {
1823 	u64 ret = (u64)rdtsc_ordered();
1824 	u64 last = pvclock_gtod_data.clock.cycle_last;
1825 
1826 	if (likely(ret >= last))
1827 		return ret;
1828 
1829 	/*
1830 	 * GCC likes to generate cmov here, but this branch is extremely
1831 	 * predictable (it's just a function of time and the likely is
1832 	 * very likely) and there's a data dependence, so force GCC
1833 	 * to generate a branch instead.  I don't barrier() because
1834 	 * we don't actually need a barrier, and if this function
1835 	 * ever gets inlined it will generate worse code.
1836 	 */
1837 	asm volatile ("");
1838 	return last;
1839 }
1840 
1841 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1842 {
1843 	long v;
1844 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1845 	u64 tsc_pg_val;
1846 
1847 	switch (gtod->clock.vclock_mode) {
1848 	case VCLOCK_HVCLOCK:
1849 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1850 						  tsc_timestamp);
1851 		if (tsc_pg_val != U64_MAX) {
1852 			/* TSC page valid */
1853 			*mode = VCLOCK_HVCLOCK;
1854 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1855 				gtod->clock.mask;
1856 		} else {
1857 			/* TSC page invalid */
1858 			*mode = VCLOCK_NONE;
1859 		}
1860 		break;
1861 	case VCLOCK_TSC:
1862 		*mode = VCLOCK_TSC;
1863 		*tsc_timestamp = read_tsc();
1864 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1865 			gtod->clock.mask;
1866 		break;
1867 	default:
1868 		*mode = VCLOCK_NONE;
1869 	}
1870 
1871 	if (*mode == VCLOCK_NONE)
1872 		*tsc_timestamp = v = 0;
1873 
1874 	return v * gtod->clock.mult;
1875 }
1876 
1877 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1878 {
1879 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1880 	unsigned long seq;
1881 	int mode;
1882 	u64 ns;
1883 
1884 	do {
1885 		seq = read_seqcount_begin(&gtod->seq);
1886 		ns = gtod->nsec_base;
1887 		ns += vgettsc(tsc_timestamp, &mode);
1888 		ns >>= gtod->clock.shift;
1889 		ns += gtod->boot_ns;
1890 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1891 	*t = ns;
1892 
1893 	return mode;
1894 }
1895 
1896 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1897 {
1898 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1899 	unsigned long seq;
1900 	int mode;
1901 	u64 ns;
1902 
1903 	do {
1904 		seq = read_seqcount_begin(&gtod->seq);
1905 		ts->tv_sec = gtod->wall_time_sec;
1906 		ns = gtod->nsec_base;
1907 		ns += vgettsc(tsc_timestamp, &mode);
1908 		ns >>= gtod->clock.shift;
1909 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1910 
1911 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1912 	ts->tv_nsec = ns;
1913 
1914 	return mode;
1915 }
1916 
1917 /* returns true if host is using TSC based clocksource */
1918 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1919 {
1920 	/* checked again under seqlock below */
1921 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1922 		return false;
1923 
1924 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1925 						      tsc_timestamp));
1926 }
1927 
1928 /* returns true if host is using TSC based clocksource */
1929 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1930 					   u64 *tsc_timestamp)
1931 {
1932 	/* checked again under seqlock below */
1933 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1934 		return false;
1935 
1936 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1937 }
1938 #endif
1939 
1940 /*
1941  *
1942  * Assuming a stable TSC across physical CPUS, and a stable TSC
1943  * across virtual CPUs, the following condition is possible.
1944  * Each numbered line represents an event visible to both
1945  * CPUs at the next numbered event.
1946  *
1947  * "timespecX" represents host monotonic time. "tscX" represents
1948  * RDTSC value.
1949  *
1950  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1951  *
1952  * 1.  read timespec0,tsc0
1953  * 2.					| timespec1 = timespec0 + N
1954  * 					| tsc1 = tsc0 + M
1955  * 3. transition to guest		| transition to guest
1956  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1957  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1958  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1959  *
1960  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1961  *
1962  * 	- ret0 < ret1
1963  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1964  *		...
1965  *	- 0 < N - M => M < N
1966  *
1967  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1968  * always the case (the difference between two distinct xtime instances
1969  * might be smaller then the difference between corresponding TSC reads,
1970  * when updating guest vcpus pvclock areas).
1971  *
1972  * To avoid that problem, do not allow visibility of distinct
1973  * system_timestamp/tsc_timestamp values simultaneously: use a master
1974  * copy of host monotonic time values. Update that master copy
1975  * in lockstep.
1976  *
1977  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1978  *
1979  */
1980 
1981 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1982 {
1983 #ifdef CONFIG_X86_64
1984 	struct kvm_arch *ka = &kvm->arch;
1985 	int vclock_mode;
1986 	bool host_tsc_clocksource, vcpus_matched;
1987 
1988 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1989 			atomic_read(&kvm->online_vcpus));
1990 
1991 	/*
1992 	 * If the host uses TSC clock, then passthrough TSC as stable
1993 	 * to the guest.
1994 	 */
1995 	host_tsc_clocksource = kvm_get_time_and_clockread(
1996 					&ka->master_kernel_ns,
1997 					&ka->master_cycle_now);
1998 
1999 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2000 				&& !ka->backwards_tsc_observed
2001 				&& !ka->boot_vcpu_runs_old_kvmclock;
2002 
2003 	if (ka->use_master_clock)
2004 		atomic_set(&kvm_guest_has_master_clock, 1);
2005 
2006 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2007 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2008 					vcpus_matched);
2009 #endif
2010 }
2011 
2012 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2013 {
2014 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2015 }
2016 
2017 static void kvm_gen_update_masterclock(struct kvm *kvm)
2018 {
2019 #ifdef CONFIG_X86_64
2020 	int i;
2021 	struct kvm_vcpu *vcpu;
2022 	struct kvm_arch *ka = &kvm->arch;
2023 
2024 	spin_lock(&ka->pvclock_gtod_sync_lock);
2025 	kvm_make_mclock_inprogress_request(kvm);
2026 	/* no guest entries from this point */
2027 	pvclock_update_vm_gtod_copy(kvm);
2028 
2029 	kvm_for_each_vcpu(i, vcpu, kvm)
2030 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2031 
2032 	/* guest entries allowed */
2033 	kvm_for_each_vcpu(i, vcpu, kvm)
2034 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2035 
2036 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2037 #endif
2038 }
2039 
2040 u64 get_kvmclock_ns(struct kvm *kvm)
2041 {
2042 	struct kvm_arch *ka = &kvm->arch;
2043 	struct pvclock_vcpu_time_info hv_clock;
2044 	u64 ret;
2045 
2046 	spin_lock(&ka->pvclock_gtod_sync_lock);
2047 	if (!ka->use_master_clock) {
2048 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2049 		return ktime_get_boot_ns() + ka->kvmclock_offset;
2050 	}
2051 
2052 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2053 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2054 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2055 
2056 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2057 	get_cpu();
2058 
2059 	if (__this_cpu_read(cpu_tsc_khz)) {
2060 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2061 				   &hv_clock.tsc_shift,
2062 				   &hv_clock.tsc_to_system_mul);
2063 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2064 	} else
2065 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2066 
2067 	put_cpu();
2068 
2069 	return ret;
2070 }
2071 
2072 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2073 {
2074 	struct kvm_vcpu_arch *vcpu = &v->arch;
2075 	struct pvclock_vcpu_time_info guest_hv_clock;
2076 
2077 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2078 		&guest_hv_clock, sizeof(guest_hv_clock))))
2079 		return;
2080 
2081 	/* This VCPU is paused, but it's legal for a guest to read another
2082 	 * VCPU's kvmclock, so we really have to follow the specification where
2083 	 * it says that version is odd if data is being modified, and even after
2084 	 * it is consistent.
2085 	 *
2086 	 * Version field updates must be kept separate.  This is because
2087 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2088 	 * writes within a string instruction are weakly ordered.  So there
2089 	 * are three writes overall.
2090 	 *
2091 	 * As a small optimization, only write the version field in the first
2092 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2093 	 * version field is the first in the struct.
2094 	 */
2095 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2096 
2097 	if (guest_hv_clock.version & 1)
2098 		++guest_hv_clock.version;  /* first time write, random junk */
2099 
2100 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2101 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2102 				&vcpu->hv_clock,
2103 				sizeof(vcpu->hv_clock.version));
2104 
2105 	smp_wmb();
2106 
2107 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2108 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2109 
2110 	if (vcpu->pvclock_set_guest_stopped_request) {
2111 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2112 		vcpu->pvclock_set_guest_stopped_request = false;
2113 	}
2114 
2115 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2116 
2117 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2118 				&vcpu->hv_clock,
2119 				sizeof(vcpu->hv_clock));
2120 
2121 	smp_wmb();
2122 
2123 	vcpu->hv_clock.version++;
2124 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2125 				&vcpu->hv_clock,
2126 				sizeof(vcpu->hv_clock.version));
2127 }
2128 
2129 static int kvm_guest_time_update(struct kvm_vcpu *v)
2130 {
2131 	unsigned long flags, tgt_tsc_khz;
2132 	struct kvm_vcpu_arch *vcpu = &v->arch;
2133 	struct kvm_arch *ka = &v->kvm->arch;
2134 	s64 kernel_ns;
2135 	u64 tsc_timestamp, host_tsc;
2136 	u8 pvclock_flags;
2137 	bool use_master_clock;
2138 
2139 	kernel_ns = 0;
2140 	host_tsc = 0;
2141 
2142 	/*
2143 	 * If the host uses TSC clock, then passthrough TSC as stable
2144 	 * to the guest.
2145 	 */
2146 	spin_lock(&ka->pvclock_gtod_sync_lock);
2147 	use_master_clock = ka->use_master_clock;
2148 	if (use_master_clock) {
2149 		host_tsc = ka->master_cycle_now;
2150 		kernel_ns = ka->master_kernel_ns;
2151 	}
2152 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2153 
2154 	/* Keep irq disabled to prevent changes to the clock */
2155 	local_irq_save(flags);
2156 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2157 	if (unlikely(tgt_tsc_khz == 0)) {
2158 		local_irq_restore(flags);
2159 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2160 		return 1;
2161 	}
2162 	if (!use_master_clock) {
2163 		host_tsc = rdtsc();
2164 		kernel_ns = ktime_get_boot_ns();
2165 	}
2166 
2167 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2168 
2169 	/*
2170 	 * We may have to catch up the TSC to match elapsed wall clock
2171 	 * time for two reasons, even if kvmclock is used.
2172 	 *   1) CPU could have been running below the maximum TSC rate
2173 	 *   2) Broken TSC compensation resets the base at each VCPU
2174 	 *      entry to avoid unknown leaps of TSC even when running
2175 	 *      again on the same CPU.  This may cause apparent elapsed
2176 	 *      time to disappear, and the guest to stand still or run
2177 	 *	very slowly.
2178 	 */
2179 	if (vcpu->tsc_catchup) {
2180 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2181 		if (tsc > tsc_timestamp) {
2182 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2183 			tsc_timestamp = tsc;
2184 		}
2185 	}
2186 
2187 	local_irq_restore(flags);
2188 
2189 	/* With all the info we got, fill in the values */
2190 
2191 	if (kvm_has_tsc_control)
2192 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2193 
2194 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2195 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2196 				   &vcpu->hv_clock.tsc_shift,
2197 				   &vcpu->hv_clock.tsc_to_system_mul);
2198 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2199 	}
2200 
2201 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2202 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2203 	vcpu->last_guest_tsc = tsc_timestamp;
2204 
2205 	/* If the host uses TSC clocksource, then it is stable */
2206 	pvclock_flags = 0;
2207 	if (use_master_clock)
2208 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2209 
2210 	vcpu->hv_clock.flags = pvclock_flags;
2211 
2212 	if (vcpu->pv_time_enabled)
2213 		kvm_setup_pvclock_page(v);
2214 	if (v == kvm_get_vcpu(v->kvm, 0))
2215 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2216 	return 0;
2217 }
2218 
2219 /*
2220  * kvmclock updates which are isolated to a given vcpu, such as
2221  * vcpu->cpu migration, should not allow system_timestamp from
2222  * the rest of the vcpus to remain static. Otherwise ntp frequency
2223  * correction applies to one vcpu's system_timestamp but not
2224  * the others.
2225  *
2226  * So in those cases, request a kvmclock update for all vcpus.
2227  * We need to rate-limit these requests though, as they can
2228  * considerably slow guests that have a large number of vcpus.
2229  * The time for a remote vcpu to update its kvmclock is bound
2230  * by the delay we use to rate-limit the updates.
2231  */
2232 
2233 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2234 
2235 static void kvmclock_update_fn(struct work_struct *work)
2236 {
2237 	int i;
2238 	struct delayed_work *dwork = to_delayed_work(work);
2239 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2240 					   kvmclock_update_work);
2241 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2242 	struct kvm_vcpu *vcpu;
2243 
2244 	kvm_for_each_vcpu(i, vcpu, kvm) {
2245 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2246 		kvm_vcpu_kick(vcpu);
2247 	}
2248 }
2249 
2250 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2251 {
2252 	struct kvm *kvm = v->kvm;
2253 
2254 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2255 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2256 					KVMCLOCK_UPDATE_DELAY);
2257 }
2258 
2259 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2260 
2261 static void kvmclock_sync_fn(struct work_struct *work)
2262 {
2263 	struct delayed_work *dwork = to_delayed_work(work);
2264 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2265 					   kvmclock_sync_work);
2266 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2267 
2268 	if (!kvmclock_periodic_sync)
2269 		return;
2270 
2271 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2272 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2273 					KVMCLOCK_SYNC_PERIOD);
2274 }
2275 
2276 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2277 {
2278 	u64 mcg_cap = vcpu->arch.mcg_cap;
2279 	unsigned bank_num = mcg_cap & 0xff;
2280 	u32 msr = msr_info->index;
2281 	u64 data = msr_info->data;
2282 
2283 	switch (msr) {
2284 	case MSR_IA32_MCG_STATUS:
2285 		vcpu->arch.mcg_status = data;
2286 		break;
2287 	case MSR_IA32_MCG_CTL:
2288 		if (!(mcg_cap & MCG_CTL_P) &&
2289 		    (data || !msr_info->host_initiated))
2290 			return 1;
2291 		if (data != 0 && data != ~(u64)0)
2292 			return 1;
2293 		vcpu->arch.mcg_ctl = data;
2294 		break;
2295 	default:
2296 		if (msr >= MSR_IA32_MC0_CTL &&
2297 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2298 			u32 offset = msr - MSR_IA32_MC0_CTL;
2299 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2300 			 * some Linux kernels though clear bit 10 in bank 4 to
2301 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2302 			 * this to avoid an uncatched #GP in the guest
2303 			 */
2304 			if ((offset & 0x3) == 0 &&
2305 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2306 				return -1;
2307 			if (!msr_info->host_initiated &&
2308 				(offset & 0x3) == 1 && data != 0)
2309 				return -1;
2310 			vcpu->arch.mce_banks[offset] = data;
2311 			break;
2312 		}
2313 		return 1;
2314 	}
2315 	return 0;
2316 }
2317 
2318 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2319 {
2320 	struct kvm *kvm = vcpu->kvm;
2321 	int lm = is_long_mode(vcpu);
2322 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2323 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2324 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2325 		: kvm->arch.xen_hvm_config.blob_size_32;
2326 	u32 page_num = data & ~PAGE_MASK;
2327 	u64 page_addr = data & PAGE_MASK;
2328 	u8 *page;
2329 	int r;
2330 
2331 	r = -E2BIG;
2332 	if (page_num >= blob_size)
2333 		goto out;
2334 	r = -ENOMEM;
2335 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2336 	if (IS_ERR(page)) {
2337 		r = PTR_ERR(page);
2338 		goto out;
2339 	}
2340 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2341 		goto out_free;
2342 	r = 0;
2343 out_free:
2344 	kfree(page);
2345 out:
2346 	return r;
2347 }
2348 
2349 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2350 {
2351 	gpa_t gpa = data & ~0x3f;
2352 
2353 	/* Bits 3:5 are reserved, Should be zero */
2354 	if (data & 0x38)
2355 		return 1;
2356 
2357 	vcpu->arch.apf.msr_val = data;
2358 
2359 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2360 		kvm_clear_async_pf_completion_queue(vcpu);
2361 		kvm_async_pf_hash_reset(vcpu);
2362 		return 0;
2363 	}
2364 
2365 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2366 					sizeof(u32)))
2367 		return 1;
2368 
2369 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2370 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2371 	kvm_async_pf_wakeup_all(vcpu);
2372 	return 0;
2373 }
2374 
2375 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2376 {
2377 	vcpu->arch.pv_time_enabled = false;
2378 }
2379 
2380 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2381 {
2382 	++vcpu->stat.tlb_flush;
2383 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2384 }
2385 
2386 static void record_steal_time(struct kvm_vcpu *vcpu)
2387 {
2388 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2389 		return;
2390 
2391 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2392 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2393 		return;
2394 
2395 	/*
2396 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2397 	 * expensive IPIs.
2398 	 */
2399 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2400 		kvm_vcpu_flush_tlb(vcpu, false);
2401 
2402 	if (vcpu->arch.st.steal.version & 1)
2403 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2404 
2405 	vcpu->arch.st.steal.version += 1;
2406 
2407 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2408 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2409 
2410 	smp_wmb();
2411 
2412 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2413 		vcpu->arch.st.last_steal;
2414 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2415 
2416 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2417 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2418 
2419 	smp_wmb();
2420 
2421 	vcpu->arch.st.steal.version += 1;
2422 
2423 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2424 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2425 }
2426 
2427 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2428 {
2429 	bool pr = false;
2430 	u32 msr = msr_info->index;
2431 	u64 data = msr_info->data;
2432 
2433 	switch (msr) {
2434 	case MSR_AMD64_NB_CFG:
2435 	case MSR_IA32_UCODE_WRITE:
2436 	case MSR_VM_HSAVE_PA:
2437 	case MSR_AMD64_PATCH_LOADER:
2438 	case MSR_AMD64_BU_CFG2:
2439 	case MSR_AMD64_DC_CFG:
2440 	case MSR_F15H_EX_CFG:
2441 		break;
2442 
2443 	case MSR_IA32_UCODE_REV:
2444 		if (msr_info->host_initiated)
2445 			vcpu->arch.microcode_version = data;
2446 		break;
2447 	case MSR_IA32_ARCH_CAPABILITIES:
2448 		if (!msr_info->host_initiated)
2449 			return 1;
2450 		vcpu->arch.arch_capabilities = data;
2451 		break;
2452 	case MSR_EFER:
2453 		return set_efer(vcpu, data);
2454 	case MSR_K7_HWCR:
2455 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2456 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2457 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2458 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2459 		if (data != 0) {
2460 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2461 				    data);
2462 			return 1;
2463 		}
2464 		break;
2465 	case MSR_FAM10H_MMIO_CONF_BASE:
2466 		if (data != 0) {
2467 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2468 				    "0x%llx\n", data);
2469 			return 1;
2470 		}
2471 		break;
2472 	case MSR_IA32_DEBUGCTLMSR:
2473 		if (!data) {
2474 			/* We support the non-activated case already */
2475 			break;
2476 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2477 			/* Values other than LBR and BTF are vendor-specific,
2478 			   thus reserved and should throw a #GP */
2479 			return 1;
2480 		}
2481 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2482 			    __func__, data);
2483 		break;
2484 	case 0x200 ... 0x2ff:
2485 		return kvm_mtrr_set_msr(vcpu, msr, data);
2486 	case MSR_IA32_APICBASE:
2487 		return kvm_set_apic_base(vcpu, msr_info);
2488 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2489 		return kvm_x2apic_msr_write(vcpu, msr, data);
2490 	case MSR_IA32_TSCDEADLINE:
2491 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2492 		break;
2493 	case MSR_IA32_TSC_ADJUST:
2494 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2495 			if (!msr_info->host_initiated) {
2496 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2497 				adjust_tsc_offset_guest(vcpu, adj);
2498 			}
2499 			vcpu->arch.ia32_tsc_adjust_msr = data;
2500 		}
2501 		break;
2502 	case MSR_IA32_MISC_ENABLE:
2503 		vcpu->arch.ia32_misc_enable_msr = data;
2504 		break;
2505 	case MSR_IA32_SMBASE:
2506 		if (!msr_info->host_initiated)
2507 			return 1;
2508 		vcpu->arch.smbase = data;
2509 		break;
2510 	case MSR_IA32_TSC:
2511 		kvm_write_tsc(vcpu, msr_info);
2512 		break;
2513 	case MSR_SMI_COUNT:
2514 		if (!msr_info->host_initiated)
2515 			return 1;
2516 		vcpu->arch.smi_count = data;
2517 		break;
2518 	case MSR_KVM_WALL_CLOCK_NEW:
2519 	case MSR_KVM_WALL_CLOCK:
2520 		vcpu->kvm->arch.wall_clock = data;
2521 		kvm_write_wall_clock(vcpu->kvm, data);
2522 		break;
2523 	case MSR_KVM_SYSTEM_TIME_NEW:
2524 	case MSR_KVM_SYSTEM_TIME: {
2525 		struct kvm_arch *ka = &vcpu->kvm->arch;
2526 
2527 		kvmclock_reset(vcpu);
2528 
2529 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2530 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2531 
2532 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2533 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2534 
2535 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2536 		}
2537 
2538 		vcpu->arch.time = data;
2539 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2540 
2541 		/* we verify if the enable bit is set... */
2542 		if (!(data & 1))
2543 			break;
2544 
2545 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2546 		     &vcpu->arch.pv_time, data & ~1ULL,
2547 		     sizeof(struct pvclock_vcpu_time_info)))
2548 			vcpu->arch.pv_time_enabled = false;
2549 		else
2550 			vcpu->arch.pv_time_enabled = true;
2551 
2552 		break;
2553 	}
2554 	case MSR_KVM_ASYNC_PF_EN:
2555 		if (kvm_pv_enable_async_pf(vcpu, data))
2556 			return 1;
2557 		break;
2558 	case MSR_KVM_STEAL_TIME:
2559 
2560 		if (unlikely(!sched_info_on()))
2561 			return 1;
2562 
2563 		if (data & KVM_STEAL_RESERVED_MASK)
2564 			return 1;
2565 
2566 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2567 						data & KVM_STEAL_VALID_BITS,
2568 						sizeof(struct kvm_steal_time)))
2569 			return 1;
2570 
2571 		vcpu->arch.st.msr_val = data;
2572 
2573 		if (!(data & KVM_MSR_ENABLED))
2574 			break;
2575 
2576 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2577 
2578 		break;
2579 	case MSR_KVM_PV_EOI_EN:
2580 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2581 			return 1;
2582 		break;
2583 
2584 	case MSR_IA32_MCG_CTL:
2585 	case MSR_IA32_MCG_STATUS:
2586 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2587 		return set_msr_mce(vcpu, msr_info);
2588 
2589 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2590 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2591 		pr = true; /* fall through */
2592 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2593 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2594 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2595 			return kvm_pmu_set_msr(vcpu, msr_info);
2596 
2597 		if (pr || data != 0)
2598 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2599 				    "0x%x data 0x%llx\n", msr, data);
2600 		break;
2601 	case MSR_K7_CLK_CTL:
2602 		/*
2603 		 * Ignore all writes to this no longer documented MSR.
2604 		 * Writes are only relevant for old K7 processors,
2605 		 * all pre-dating SVM, but a recommended workaround from
2606 		 * AMD for these chips. It is possible to specify the
2607 		 * affected processor models on the command line, hence
2608 		 * the need to ignore the workaround.
2609 		 */
2610 		break;
2611 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2612 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2613 	case HV_X64_MSR_CRASH_CTL:
2614 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2615 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2616 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2617 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2618 		return kvm_hv_set_msr_common(vcpu, msr, data,
2619 					     msr_info->host_initiated);
2620 	case MSR_IA32_BBL_CR_CTL3:
2621 		/* Drop writes to this legacy MSR -- see rdmsr
2622 		 * counterpart for further detail.
2623 		 */
2624 		if (report_ignored_msrs)
2625 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2626 				msr, data);
2627 		break;
2628 	case MSR_AMD64_OSVW_ID_LENGTH:
2629 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2630 			return 1;
2631 		vcpu->arch.osvw.length = data;
2632 		break;
2633 	case MSR_AMD64_OSVW_STATUS:
2634 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2635 			return 1;
2636 		vcpu->arch.osvw.status = data;
2637 		break;
2638 	case MSR_PLATFORM_INFO:
2639 		if (!msr_info->host_initiated ||
2640 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2641 		     cpuid_fault_enabled(vcpu)))
2642 			return 1;
2643 		vcpu->arch.msr_platform_info = data;
2644 		break;
2645 	case MSR_MISC_FEATURES_ENABLES:
2646 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2647 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2648 		     !supports_cpuid_fault(vcpu)))
2649 			return 1;
2650 		vcpu->arch.msr_misc_features_enables = data;
2651 		break;
2652 	default:
2653 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2654 			return xen_hvm_config(vcpu, data);
2655 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2656 			return kvm_pmu_set_msr(vcpu, msr_info);
2657 		if (!ignore_msrs) {
2658 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2659 				    msr, data);
2660 			return 1;
2661 		} else {
2662 			if (report_ignored_msrs)
2663 				vcpu_unimpl(vcpu,
2664 					"ignored wrmsr: 0x%x data 0x%llx\n",
2665 					msr, data);
2666 			break;
2667 		}
2668 	}
2669 	return 0;
2670 }
2671 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2672 
2673 
2674 /*
2675  * Reads an msr value (of 'msr_index') into 'pdata'.
2676  * Returns 0 on success, non-0 otherwise.
2677  * Assumes vcpu_load() was already called.
2678  */
2679 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2680 {
2681 	return kvm_x86_ops->get_msr(vcpu, msr);
2682 }
2683 EXPORT_SYMBOL_GPL(kvm_get_msr);
2684 
2685 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2686 {
2687 	u64 data;
2688 	u64 mcg_cap = vcpu->arch.mcg_cap;
2689 	unsigned bank_num = mcg_cap & 0xff;
2690 
2691 	switch (msr) {
2692 	case MSR_IA32_P5_MC_ADDR:
2693 	case MSR_IA32_P5_MC_TYPE:
2694 		data = 0;
2695 		break;
2696 	case MSR_IA32_MCG_CAP:
2697 		data = vcpu->arch.mcg_cap;
2698 		break;
2699 	case MSR_IA32_MCG_CTL:
2700 		if (!(mcg_cap & MCG_CTL_P) && !host)
2701 			return 1;
2702 		data = vcpu->arch.mcg_ctl;
2703 		break;
2704 	case MSR_IA32_MCG_STATUS:
2705 		data = vcpu->arch.mcg_status;
2706 		break;
2707 	default:
2708 		if (msr >= MSR_IA32_MC0_CTL &&
2709 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2710 			u32 offset = msr - MSR_IA32_MC0_CTL;
2711 			data = vcpu->arch.mce_banks[offset];
2712 			break;
2713 		}
2714 		return 1;
2715 	}
2716 	*pdata = data;
2717 	return 0;
2718 }
2719 
2720 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2721 {
2722 	switch (msr_info->index) {
2723 	case MSR_IA32_PLATFORM_ID:
2724 	case MSR_IA32_EBL_CR_POWERON:
2725 	case MSR_IA32_DEBUGCTLMSR:
2726 	case MSR_IA32_LASTBRANCHFROMIP:
2727 	case MSR_IA32_LASTBRANCHTOIP:
2728 	case MSR_IA32_LASTINTFROMIP:
2729 	case MSR_IA32_LASTINTTOIP:
2730 	case MSR_K8_SYSCFG:
2731 	case MSR_K8_TSEG_ADDR:
2732 	case MSR_K8_TSEG_MASK:
2733 	case MSR_K7_HWCR:
2734 	case MSR_VM_HSAVE_PA:
2735 	case MSR_K8_INT_PENDING_MSG:
2736 	case MSR_AMD64_NB_CFG:
2737 	case MSR_FAM10H_MMIO_CONF_BASE:
2738 	case MSR_AMD64_BU_CFG2:
2739 	case MSR_IA32_PERF_CTL:
2740 	case MSR_AMD64_DC_CFG:
2741 	case MSR_F15H_EX_CFG:
2742 		msr_info->data = 0;
2743 		break;
2744 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2745 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2746 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2747 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2748 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2749 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2750 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2751 		msr_info->data = 0;
2752 		break;
2753 	case MSR_IA32_UCODE_REV:
2754 		msr_info->data = vcpu->arch.microcode_version;
2755 		break;
2756 	case MSR_IA32_ARCH_CAPABILITIES:
2757 		if (!msr_info->host_initiated &&
2758 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2759 			return 1;
2760 		msr_info->data = vcpu->arch.arch_capabilities;
2761 		break;
2762 	case MSR_IA32_TSC:
2763 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2764 		break;
2765 	case MSR_MTRRcap:
2766 	case 0x200 ... 0x2ff:
2767 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2768 	case 0xcd: /* fsb frequency */
2769 		msr_info->data = 3;
2770 		break;
2771 		/*
2772 		 * MSR_EBC_FREQUENCY_ID
2773 		 * Conservative value valid for even the basic CPU models.
2774 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2775 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2776 		 * and 266MHz for model 3, or 4. Set Core Clock
2777 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2778 		 * 31:24) even though these are only valid for CPU
2779 		 * models > 2, however guests may end up dividing or
2780 		 * multiplying by zero otherwise.
2781 		 */
2782 	case MSR_EBC_FREQUENCY_ID:
2783 		msr_info->data = 1 << 24;
2784 		break;
2785 	case MSR_IA32_APICBASE:
2786 		msr_info->data = kvm_get_apic_base(vcpu);
2787 		break;
2788 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2789 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2790 		break;
2791 	case MSR_IA32_TSCDEADLINE:
2792 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2793 		break;
2794 	case MSR_IA32_TSC_ADJUST:
2795 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2796 		break;
2797 	case MSR_IA32_MISC_ENABLE:
2798 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2799 		break;
2800 	case MSR_IA32_SMBASE:
2801 		if (!msr_info->host_initiated)
2802 			return 1;
2803 		msr_info->data = vcpu->arch.smbase;
2804 		break;
2805 	case MSR_SMI_COUNT:
2806 		msr_info->data = vcpu->arch.smi_count;
2807 		break;
2808 	case MSR_IA32_PERF_STATUS:
2809 		/* TSC increment by tick */
2810 		msr_info->data = 1000ULL;
2811 		/* CPU multiplier */
2812 		msr_info->data |= (((uint64_t)4ULL) << 40);
2813 		break;
2814 	case MSR_EFER:
2815 		msr_info->data = vcpu->arch.efer;
2816 		break;
2817 	case MSR_KVM_WALL_CLOCK:
2818 	case MSR_KVM_WALL_CLOCK_NEW:
2819 		msr_info->data = vcpu->kvm->arch.wall_clock;
2820 		break;
2821 	case MSR_KVM_SYSTEM_TIME:
2822 	case MSR_KVM_SYSTEM_TIME_NEW:
2823 		msr_info->data = vcpu->arch.time;
2824 		break;
2825 	case MSR_KVM_ASYNC_PF_EN:
2826 		msr_info->data = vcpu->arch.apf.msr_val;
2827 		break;
2828 	case MSR_KVM_STEAL_TIME:
2829 		msr_info->data = vcpu->arch.st.msr_val;
2830 		break;
2831 	case MSR_KVM_PV_EOI_EN:
2832 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2833 		break;
2834 	case MSR_IA32_P5_MC_ADDR:
2835 	case MSR_IA32_P5_MC_TYPE:
2836 	case MSR_IA32_MCG_CAP:
2837 	case MSR_IA32_MCG_CTL:
2838 	case MSR_IA32_MCG_STATUS:
2839 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2840 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2841 				   msr_info->host_initiated);
2842 	case MSR_K7_CLK_CTL:
2843 		/*
2844 		 * Provide expected ramp-up count for K7. All other
2845 		 * are set to zero, indicating minimum divisors for
2846 		 * every field.
2847 		 *
2848 		 * This prevents guest kernels on AMD host with CPU
2849 		 * type 6, model 8 and higher from exploding due to
2850 		 * the rdmsr failing.
2851 		 */
2852 		msr_info->data = 0x20000000;
2853 		break;
2854 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2855 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2856 	case HV_X64_MSR_CRASH_CTL:
2857 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2858 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2859 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2860 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2861 		return kvm_hv_get_msr_common(vcpu,
2862 					     msr_info->index, &msr_info->data,
2863 					     msr_info->host_initiated);
2864 		break;
2865 	case MSR_IA32_BBL_CR_CTL3:
2866 		/* This legacy MSR exists but isn't fully documented in current
2867 		 * silicon.  It is however accessed by winxp in very narrow
2868 		 * scenarios where it sets bit #19, itself documented as
2869 		 * a "reserved" bit.  Best effort attempt to source coherent
2870 		 * read data here should the balance of the register be
2871 		 * interpreted by the guest:
2872 		 *
2873 		 * L2 cache control register 3: 64GB range, 256KB size,
2874 		 * enabled, latency 0x1, configured
2875 		 */
2876 		msr_info->data = 0xbe702111;
2877 		break;
2878 	case MSR_AMD64_OSVW_ID_LENGTH:
2879 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2880 			return 1;
2881 		msr_info->data = vcpu->arch.osvw.length;
2882 		break;
2883 	case MSR_AMD64_OSVW_STATUS:
2884 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2885 			return 1;
2886 		msr_info->data = vcpu->arch.osvw.status;
2887 		break;
2888 	case MSR_PLATFORM_INFO:
2889 		if (!msr_info->host_initiated &&
2890 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2891 			return 1;
2892 		msr_info->data = vcpu->arch.msr_platform_info;
2893 		break;
2894 	case MSR_MISC_FEATURES_ENABLES:
2895 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2896 		break;
2897 	default:
2898 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2899 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2900 		if (!ignore_msrs) {
2901 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2902 					       msr_info->index);
2903 			return 1;
2904 		} else {
2905 			if (report_ignored_msrs)
2906 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2907 					msr_info->index);
2908 			msr_info->data = 0;
2909 		}
2910 		break;
2911 	}
2912 	return 0;
2913 }
2914 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2915 
2916 /*
2917  * Read or write a bunch of msrs. All parameters are kernel addresses.
2918  *
2919  * @return number of msrs set successfully.
2920  */
2921 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2922 		    struct kvm_msr_entry *entries,
2923 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2924 				  unsigned index, u64 *data))
2925 {
2926 	int i;
2927 
2928 	for (i = 0; i < msrs->nmsrs; ++i)
2929 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2930 			break;
2931 
2932 	return i;
2933 }
2934 
2935 /*
2936  * Read or write a bunch of msrs. Parameters are user addresses.
2937  *
2938  * @return number of msrs set successfully.
2939  */
2940 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2941 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2942 				unsigned index, u64 *data),
2943 		  int writeback)
2944 {
2945 	struct kvm_msrs msrs;
2946 	struct kvm_msr_entry *entries;
2947 	int r, n;
2948 	unsigned size;
2949 
2950 	r = -EFAULT;
2951 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2952 		goto out;
2953 
2954 	r = -E2BIG;
2955 	if (msrs.nmsrs >= MAX_IO_MSRS)
2956 		goto out;
2957 
2958 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2959 	entries = memdup_user(user_msrs->entries, size);
2960 	if (IS_ERR(entries)) {
2961 		r = PTR_ERR(entries);
2962 		goto out;
2963 	}
2964 
2965 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2966 	if (r < 0)
2967 		goto out_free;
2968 
2969 	r = -EFAULT;
2970 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2971 		goto out_free;
2972 
2973 	r = n;
2974 
2975 out_free:
2976 	kfree(entries);
2977 out:
2978 	return r;
2979 }
2980 
2981 static inline bool kvm_can_mwait_in_guest(void)
2982 {
2983 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2984 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2985 		boot_cpu_has(X86_FEATURE_ARAT);
2986 }
2987 
2988 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2989 {
2990 	int r = 0;
2991 
2992 	switch (ext) {
2993 	case KVM_CAP_IRQCHIP:
2994 	case KVM_CAP_HLT:
2995 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2996 	case KVM_CAP_SET_TSS_ADDR:
2997 	case KVM_CAP_EXT_CPUID:
2998 	case KVM_CAP_EXT_EMUL_CPUID:
2999 	case KVM_CAP_CLOCKSOURCE:
3000 	case KVM_CAP_PIT:
3001 	case KVM_CAP_NOP_IO_DELAY:
3002 	case KVM_CAP_MP_STATE:
3003 	case KVM_CAP_SYNC_MMU:
3004 	case KVM_CAP_USER_NMI:
3005 	case KVM_CAP_REINJECT_CONTROL:
3006 	case KVM_CAP_IRQ_INJECT_STATUS:
3007 	case KVM_CAP_IOEVENTFD:
3008 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3009 	case KVM_CAP_PIT2:
3010 	case KVM_CAP_PIT_STATE2:
3011 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3012 	case KVM_CAP_XEN_HVM:
3013 	case KVM_CAP_VCPU_EVENTS:
3014 	case KVM_CAP_HYPERV:
3015 	case KVM_CAP_HYPERV_VAPIC:
3016 	case KVM_CAP_HYPERV_SPIN:
3017 	case KVM_CAP_HYPERV_SYNIC:
3018 	case KVM_CAP_HYPERV_SYNIC2:
3019 	case KVM_CAP_HYPERV_VP_INDEX:
3020 	case KVM_CAP_HYPERV_EVENTFD:
3021 	case KVM_CAP_HYPERV_TLBFLUSH:
3022 	case KVM_CAP_HYPERV_SEND_IPI:
3023 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3024 	case KVM_CAP_HYPERV_CPUID:
3025 	case KVM_CAP_PCI_SEGMENT:
3026 	case KVM_CAP_DEBUGREGS:
3027 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3028 	case KVM_CAP_XSAVE:
3029 	case KVM_CAP_ASYNC_PF:
3030 	case KVM_CAP_GET_TSC_KHZ:
3031 	case KVM_CAP_KVMCLOCK_CTRL:
3032 	case KVM_CAP_READONLY_MEM:
3033 	case KVM_CAP_HYPERV_TIME:
3034 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3035 	case KVM_CAP_TSC_DEADLINE_TIMER:
3036 	case KVM_CAP_DISABLE_QUIRKS:
3037 	case KVM_CAP_SET_BOOT_CPU_ID:
3038  	case KVM_CAP_SPLIT_IRQCHIP:
3039 	case KVM_CAP_IMMEDIATE_EXIT:
3040 	case KVM_CAP_GET_MSR_FEATURES:
3041 	case KVM_CAP_MSR_PLATFORM_INFO:
3042 	case KVM_CAP_EXCEPTION_PAYLOAD:
3043 		r = 1;
3044 		break;
3045 	case KVM_CAP_SYNC_REGS:
3046 		r = KVM_SYNC_X86_VALID_FIELDS;
3047 		break;
3048 	case KVM_CAP_ADJUST_CLOCK:
3049 		r = KVM_CLOCK_TSC_STABLE;
3050 		break;
3051 	case KVM_CAP_X86_DISABLE_EXITS:
3052 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3053 		if(kvm_can_mwait_in_guest())
3054 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3055 		break;
3056 	case KVM_CAP_X86_SMM:
3057 		/* SMBASE is usually relocated above 1M on modern chipsets,
3058 		 * and SMM handlers might indeed rely on 4G segment limits,
3059 		 * so do not report SMM to be available if real mode is
3060 		 * emulated via vm86 mode.  Still, do not go to great lengths
3061 		 * to avoid userspace's usage of the feature, because it is a
3062 		 * fringe case that is not enabled except via specific settings
3063 		 * of the module parameters.
3064 		 */
3065 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3066 		break;
3067 	case KVM_CAP_VAPIC:
3068 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3069 		break;
3070 	case KVM_CAP_NR_VCPUS:
3071 		r = KVM_SOFT_MAX_VCPUS;
3072 		break;
3073 	case KVM_CAP_MAX_VCPUS:
3074 		r = KVM_MAX_VCPUS;
3075 		break;
3076 	case KVM_CAP_NR_MEMSLOTS:
3077 		r = KVM_USER_MEM_SLOTS;
3078 		break;
3079 	case KVM_CAP_PV_MMU:	/* obsolete */
3080 		r = 0;
3081 		break;
3082 	case KVM_CAP_MCE:
3083 		r = KVM_MAX_MCE_BANKS;
3084 		break;
3085 	case KVM_CAP_XCRS:
3086 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3087 		break;
3088 	case KVM_CAP_TSC_CONTROL:
3089 		r = kvm_has_tsc_control;
3090 		break;
3091 	case KVM_CAP_X2APIC_API:
3092 		r = KVM_X2APIC_API_VALID_FLAGS;
3093 		break;
3094 	case KVM_CAP_NESTED_STATE:
3095 		r = kvm_x86_ops->get_nested_state ?
3096 			kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3097 		break;
3098 	default:
3099 		break;
3100 	}
3101 	return r;
3102 
3103 }
3104 
3105 long kvm_arch_dev_ioctl(struct file *filp,
3106 			unsigned int ioctl, unsigned long arg)
3107 {
3108 	void __user *argp = (void __user *)arg;
3109 	long r;
3110 
3111 	switch (ioctl) {
3112 	case KVM_GET_MSR_INDEX_LIST: {
3113 		struct kvm_msr_list __user *user_msr_list = argp;
3114 		struct kvm_msr_list msr_list;
3115 		unsigned n;
3116 
3117 		r = -EFAULT;
3118 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3119 			goto out;
3120 		n = msr_list.nmsrs;
3121 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3122 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3123 			goto out;
3124 		r = -E2BIG;
3125 		if (n < msr_list.nmsrs)
3126 			goto out;
3127 		r = -EFAULT;
3128 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3129 				 num_msrs_to_save * sizeof(u32)))
3130 			goto out;
3131 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3132 				 &emulated_msrs,
3133 				 num_emulated_msrs * sizeof(u32)))
3134 			goto out;
3135 		r = 0;
3136 		break;
3137 	}
3138 	case KVM_GET_SUPPORTED_CPUID:
3139 	case KVM_GET_EMULATED_CPUID: {
3140 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3141 		struct kvm_cpuid2 cpuid;
3142 
3143 		r = -EFAULT;
3144 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3145 			goto out;
3146 
3147 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3148 					    ioctl);
3149 		if (r)
3150 			goto out;
3151 
3152 		r = -EFAULT;
3153 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3154 			goto out;
3155 		r = 0;
3156 		break;
3157 	}
3158 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3159 		r = -EFAULT;
3160 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3161 				 sizeof(kvm_mce_cap_supported)))
3162 			goto out;
3163 		r = 0;
3164 		break;
3165 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3166 		struct kvm_msr_list __user *user_msr_list = argp;
3167 		struct kvm_msr_list msr_list;
3168 		unsigned int n;
3169 
3170 		r = -EFAULT;
3171 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3172 			goto out;
3173 		n = msr_list.nmsrs;
3174 		msr_list.nmsrs = num_msr_based_features;
3175 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3176 			goto out;
3177 		r = -E2BIG;
3178 		if (n < msr_list.nmsrs)
3179 			goto out;
3180 		r = -EFAULT;
3181 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3182 				 num_msr_based_features * sizeof(u32)))
3183 			goto out;
3184 		r = 0;
3185 		break;
3186 	}
3187 	case KVM_GET_MSRS:
3188 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3189 		break;
3190 	}
3191 	default:
3192 		r = -EINVAL;
3193 	}
3194 out:
3195 	return r;
3196 }
3197 
3198 static void wbinvd_ipi(void *garbage)
3199 {
3200 	wbinvd();
3201 }
3202 
3203 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3204 {
3205 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3206 }
3207 
3208 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3209 {
3210 	/* Address WBINVD may be executed by guest */
3211 	if (need_emulate_wbinvd(vcpu)) {
3212 		if (kvm_x86_ops->has_wbinvd_exit())
3213 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3214 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3215 			smp_call_function_single(vcpu->cpu,
3216 					wbinvd_ipi, NULL, 1);
3217 	}
3218 
3219 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3220 
3221 	/* Apply any externally detected TSC adjustments (due to suspend) */
3222 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3223 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3224 		vcpu->arch.tsc_offset_adjustment = 0;
3225 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3226 	}
3227 
3228 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3229 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3230 				rdtsc() - vcpu->arch.last_host_tsc;
3231 		if (tsc_delta < 0)
3232 			mark_tsc_unstable("KVM discovered backwards TSC");
3233 
3234 		if (kvm_check_tsc_unstable()) {
3235 			u64 offset = kvm_compute_tsc_offset(vcpu,
3236 						vcpu->arch.last_guest_tsc);
3237 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3238 			vcpu->arch.tsc_catchup = 1;
3239 		}
3240 
3241 		if (kvm_lapic_hv_timer_in_use(vcpu))
3242 			kvm_lapic_restart_hv_timer(vcpu);
3243 
3244 		/*
3245 		 * On a host with synchronized TSC, there is no need to update
3246 		 * kvmclock on vcpu->cpu migration
3247 		 */
3248 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3249 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3250 		if (vcpu->cpu != cpu)
3251 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3252 		vcpu->cpu = cpu;
3253 	}
3254 
3255 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3256 }
3257 
3258 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3259 {
3260 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3261 		return;
3262 
3263 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3264 
3265 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3266 			&vcpu->arch.st.steal.preempted,
3267 			offsetof(struct kvm_steal_time, preempted),
3268 			sizeof(vcpu->arch.st.steal.preempted));
3269 }
3270 
3271 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3272 {
3273 	int idx;
3274 
3275 	if (vcpu->preempted)
3276 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3277 
3278 	/*
3279 	 * Disable page faults because we're in atomic context here.
3280 	 * kvm_write_guest_offset_cached() would call might_fault()
3281 	 * that relies on pagefault_disable() to tell if there's a
3282 	 * bug. NOTE: the write to guest memory may not go through if
3283 	 * during postcopy live migration or if there's heavy guest
3284 	 * paging.
3285 	 */
3286 	pagefault_disable();
3287 	/*
3288 	 * kvm_memslots() will be called by
3289 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3290 	 */
3291 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3292 	kvm_steal_time_set_preempted(vcpu);
3293 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3294 	pagefault_enable();
3295 	kvm_x86_ops->vcpu_put(vcpu);
3296 	vcpu->arch.last_host_tsc = rdtsc();
3297 	/*
3298 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3299 	 * on every vmexit, but if not, we might have a stale dr6 from the
3300 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3301 	 */
3302 	set_debugreg(0, 6);
3303 }
3304 
3305 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3306 				    struct kvm_lapic_state *s)
3307 {
3308 	if (vcpu->arch.apicv_active)
3309 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3310 
3311 	return kvm_apic_get_state(vcpu, s);
3312 }
3313 
3314 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3315 				    struct kvm_lapic_state *s)
3316 {
3317 	int r;
3318 
3319 	r = kvm_apic_set_state(vcpu, s);
3320 	if (r)
3321 		return r;
3322 	update_cr8_intercept(vcpu);
3323 
3324 	return 0;
3325 }
3326 
3327 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3328 {
3329 	return (!lapic_in_kernel(vcpu) ||
3330 		kvm_apic_accept_pic_intr(vcpu));
3331 }
3332 
3333 /*
3334  * if userspace requested an interrupt window, check that the
3335  * interrupt window is open.
3336  *
3337  * No need to exit to userspace if we already have an interrupt queued.
3338  */
3339 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3340 {
3341 	return kvm_arch_interrupt_allowed(vcpu) &&
3342 		!kvm_cpu_has_interrupt(vcpu) &&
3343 		!kvm_event_needs_reinjection(vcpu) &&
3344 		kvm_cpu_accept_dm_intr(vcpu);
3345 }
3346 
3347 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3348 				    struct kvm_interrupt *irq)
3349 {
3350 	if (irq->irq >= KVM_NR_INTERRUPTS)
3351 		return -EINVAL;
3352 
3353 	if (!irqchip_in_kernel(vcpu->kvm)) {
3354 		kvm_queue_interrupt(vcpu, irq->irq, false);
3355 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3356 		return 0;
3357 	}
3358 
3359 	/*
3360 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3361 	 * fail for in-kernel 8259.
3362 	 */
3363 	if (pic_in_kernel(vcpu->kvm))
3364 		return -ENXIO;
3365 
3366 	if (vcpu->arch.pending_external_vector != -1)
3367 		return -EEXIST;
3368 
3369 	vcpu->arch.pending_external_vector = irq->irq;
3370 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3371 	return 0;
3372 }
3373 
3374 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3375 {
3376 	kvm_inject_nmi(vcpu);
3377 
3378 	return 0;
3379 }
3380 
3381 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3382 {
3383 	kvm_make_request(KVM_REQ_SMI, vcpu);
3384 
3385 	return 0;
3386 }
3387 
3388 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3389 					   struct kvm_tpr_access_ctl *tac)
3390 {
3391 	if (tac->flags)
3392 		return -EINVAL;
3393 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3394 	return 0;
3395 }
3396 
3397 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3398 					u64 mcg_cap)
3399 {
3400 	int r;
3401 	unsigned bank_num = mcg_cap & 0xff, bank;
3402 
3403 	r = -EINVAL;
3404 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3405 		goto out;
3406 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3407 		goto out;
3408 	r = 0;
3409 	vcpu->arch.mcg_cap = mcg_cap;
3410 	/* Init IA32_MCG_CTL to all 1s */
3411 	if (mcg_cap & MCG_CTL_P)
3412 		vcpu->arch.mcg_ctl = ~(u64)0;
3413 	/* Init IA32_MCi_CTL to all 1s */
3414 	for (bank = 0; bank < bank_num; bank++)
3415 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3416 
3417 	if (kvm_x86_ops->setup_mce)
3418 		kvm_x86_ops->setup_mce(vcpu);
3419 out:
3420 	return r;
3421 }
3422 
3423 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3424 				      struct kvm_x86_mce *mce)
3425 {
3426 	u64 mcg_cap = vcpu->arch.mcg_cap;
3427 	unsigned bank_num = mcg_cap & 0xff;
3428 	u64 *banks = vcpu->arch.mce_banks;
3429 
3430 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3431 		return -EINVAL;
3432 	/*
3433 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3434 	 * reporting is disabled
3435 	 */
3436 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3437 	    vcpu->arch.mcg_ctl != ~(u64)0)
3438 		return 0;
3439 	banks += 4 * mce->bank;
3440 	/*
3441 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3442 	 * reporting is disabled for the bank
3443 	 */
3444 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3445 		return 0;
3446 	if (mce->status & MCI_STATUS_UC) {
3447 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3448 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3449 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3450 			return 0;
3451 		}
3452 		if (banks[1] & MCI_STATUS_VAL)
3453 			mce->status |= MCI_STATUS_OVER;
3454 		banks[2] = mce->addr;
3455 		banks[3] = mce->misc;
3456 		vcpu->arch.mcg_status = mce->mcg_status;
3457 		banks[1] = mce->status;
3458 		kvm_queue_exception(vcpu, MC_VECTOR);
3459 	} else if (!(banks[1] & MCI_STATUS_VAL)
3460 		   || !(banks[1] & MCI_STATUS_UC)) {
3461 		if (banks[1] & MCI_STATUS_VAL)
3462 			mce->status |= MCI_STATUS_OVER;
3463 		banks[2] = mce->addr;
3464 		banks[3] = mce->misc;
3465 		banks[1] = mce->status;
3466 	} else
3467 		banks[1] |= MCI_STATUS_OVER;
3468 	return 0;
3469 }
3470 
3471 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3472 					       struct kvm_vcpu_events *events)
3473 {
3474 	process_nmi(vcpu);
3475 
3476 	/*
3477 	 * The API doesn't provide the instruction length for software
3478 	 * exceptions, so don't report them. As long as the guest RIP
3479 	 * isn't advanced, we should expect to encounter the exception
3480 	 * again.
3481 	 */
3482 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3483 		events->exception.injected = 0;
3484 		events->exception.pending = 0;
3485 	} else {
3486 		events->exception.injected = vcpu->arch.exception.injected;
3487 		events->exception.pending = vcpu->arch.exception.pending;
3488 		/*
3489 		 * For ABI compatibility, deliberately conflate
3490 		 * pending and injected exceptions when
3491 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3492 		 */
3493 		if (!vcpu->kvm->arch.exception_payload_enabled)
3494 			events->exception.injected |=
3495 				vcpu->arch.exception.pending;
3496 	}
3497 	events->exception.nr = vcpu->arch.exception.nr;
3498 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3499 	events->exception.error_code = vcpu->arch.exception.error_code;
3500 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3501 	events->exception_payload = vcpu->arch.exception.payload;
3502 
3503 	events->interrupt.injected =
3504 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3505 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3506 	events->interrupt.soft = 0;
3507 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3508 
3509 	events->nmi.injected = vcpu->arch.nmi_injected;
3510 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3511 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3512 	events->nmi.pad = 0;
3513 
3514 	events->sipi_vector = 0; /* never valid when reporting to user space */
3515 
3516 	events->smi.smm = is_smm(vcpu);
3517 	events->smi.pending = vcpu->arch.smi_pending;
3518 	events->smi.smm_inside_nmi =
3519 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3520 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3521 
3522 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3523 			 | KVM_VCPUEVENT_VALID_SHADOW
3524 			 | KVM_VCPUEVENT_VALID_SMM);
3525 	if (vcpu->kvm->arch.exception_payload_enabled)
3526 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3527 
3528 	memset(&events->reserved, 0, sizeof(events->reserved));
3529 }
3530 
3531 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3532 
3533 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3534 					      struct kvm_vcpu_events *events)
3535 {
3536 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3537 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3538 			      | KVM_VCPUEVENT_VALID_SHADOW
3539 			      | KVM_VCPUEVENT_VALID_SMM
3540 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3541 		return -EINVAL;
3542 
3543 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3544 		if (!vcpu->kvm->arch.exception_payload_enabled)
3545 			return -EINVAL;
3546 		if (events->exception.pending)
3547 			events->exception.injected = 0;
3548 		else
3549 			events->exception_has_payload = 0;
3550 	} else {
3551 		events->exception.pending = 0;
3552 		events->exception_has_payload = 0;
3553 	}
3554 
3555 	if ((events->exception.injected || events->exception.pending) &&
3556 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3557 		return -EINVAL;
3558 
3559 	/* INITs are latched while in SMM */
3560 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3561 	    (events->smi.smm || events->smi.pending) &&
3562 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3563 		return -EINVAL;
3564 
3565 	process_nmi(vcpu);
3566 	vcpu->arch.exception.injected = events->exception.injected;
3567 	vcpu->arch.exception.pending = events->exception.pending;
3568 	vcpu->arch.exception.nr = events->exception.nr;
3569 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3570 	vcpu->arch.exception.error_code = events->exception.error_code;
3571 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3572 	vcpu->arch.exception.payload = events->exception_payload;
3573 
3574 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3575 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3576 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3577 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3578 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3579 						  events->interrupt.shadow);
3580 
3581 	vcpu->arch.nmi_injected = events->nmi.injected;
3582 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3583 		vcpu->arch.nmi_pending = events->nmi.pending;
3584 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3585 
3586 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3587 	    lapic_in_kernel(vcpu))
3588 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3589 
3590 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3591 		u32 hflags = vcpu->arch.hflags;
3592 		if (events->smi.smm)
3593 			hflags |= HF_SMM_MASK;
3594 		else
3595 			hflags &= ~HF_SMM_MASK;
3596 		kvm_set_hflags(vcpu, hflags);
3597 
3598 		vcpu->arch.smi_pending = events->smi.pending;
3599 
3600 		if (events->smi.smm) {
3601 			if (events->smi.smm_inside_nmi)
3602 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3603 			else
3604 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3605 			if (lapic_in_kernel(vcpu)) {
3606 				if (events->smi.latched_init)
3607 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3608 				else
3609 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3610 			}
3611 		}
3612 	}
3613 
3614 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3615 
3616 	return 0;
3617 }
3618 
3619 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3620 					     struct kvm_debugregs *dbgregs)
3621 {
3622 	unsigned long val;
3623 
3624 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3625 	kvm_get_dr(vcpu, 6, &val);
3626 	dbgregs->dr6 = val;
3627 	dbgregs->dr7 = vcpu->arch.dr7;
3628 	dbgregs->flags = 0;
3629 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3630 }
3631 
3632 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3633 					    struct kvm_debugregs *dbgregs)
3634 {
3635 	if (dbgregs->flags)
3636 		return -EINVAL;
3637 
3638 	if (dbgregs->dr6 & ~0xffffffffull)
3639 		return -EINVAL;
3640 	if (dbgregs->dr7 & ~0xffffffffull)
3641 		return -EINVAL;
3642 
3643 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3644 	kvm_update_dr0123(vcpu);
3645 	vcpu->arch.dr6 = dbgregs->dr6;
3646 	kvm_update_dr6(vcpu);
3647 	vcpu->arch.dr7 = dbgregs->dr7;
3648 	kvm_update_dr7(vcpu);
3649 
3650 	return 0;
3651 }
3652 
3653 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3654 
3655 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3656 {
3657 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3658 	u64 xstate_bv = xsave->header.xfeatures;
3659 	u64 valid;
3660 
3661 	/*
3662 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3663 	 * leaves 0 and 1 in the loop below.
3664 	 */
3665 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3666 
3667 	/* Set XSTATE_BV */
3668 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3669 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3670 
3671 	/*
3672 	 * Copy each region from the possibly compacted offset to the
3673 	 * non-compacted offset.
3674 	 */
3675 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3676 	while (valid) {
3677 		u64 feature = valid & -valid;
3678 		int index = fls64(feature) - 1;
3679 		void *src = get_xsave_addr(xsave, feature);
3680 
3681 		if (src) {
3682 			u32 size, offset, ecx, edx;
3683 			cpuid_count(XSTATE_CPUID, index,
3684 				    &size, &offset, &ecx, &edx);
3685 			if (feature == XFEATURE_MASK_PKRU)
3686 				memcpy(dest + offset, &vcpu->arch.pkru,
3687 				       sizeof(vcpu->arch.pkru));
3688 			else
3689 				memcpy(dest + offset, src, size);
3690 
3691 		}
3692 
3693 		valid -= feature;
3694 	}
3695 }
3696 
3697 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3698 {
3699 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3700 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3701 	u64 valid;
3702 
3703 	/*
3704 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3705 	 * leaves 0 and 1 in the loop below.
3706 	 */
3707 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3708 
3709 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3710 	xsave->header.xfeatures = xstate_bv;
3711 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3712 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3713 
3714 	/*
3715 	 * Copy each region from the non-compacted offset to the
3716 	 * possibly compacted offset.
3717 	 */
3718 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3719 	while (valid) {
3720 		u64 feature = valid & -valid;
3721 		int index = fls64(feature) - 1;
3722 		void *dest = get_xsave_addr(xsave, feature);
3723 
3724 		if (dest) {
3725 			u32 size, offset, ecx, edx;
3726 			cpuid_count(XSTATE_CPUID, index,
3727 				    &size, &offset, &ecx, &edx);
3728 			if (feature == XFEATURE_MASK_PKRU)
3729 				memcpy(&vcpu->arch.pkru, src + offset,
3730 				       sizeof(vcpu->arch.pkru));
3731 			else
3732 				memcpy(dest, src + offset, size);
3733 		}
3734 
3735 		valid -= feature;
3736 	}
3737 }
3738 
3739 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3740 					 struct kvm_xsave *guest_xsave)
3741 {
3742 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3743 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3744 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3745 	} else {
3746 		memcpy(guest_xsave->region,
3747 			&vcpu->arch.guest_fpu->state.fxsave,
3748 			sizeof(struct fxregs_state));
3749 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3750 			XFEATURE_MASK_FPSSE;
3751 	}
3752 }
3753 
3754 #define XSAVE_MXCSR_OFFSET 24
3755 
3756 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3757 					struct kvm_xsave *guest_xsave)
3758 {
3759 	u64 xstate_bv =
3760 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3761 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3762 
3763 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3764 		/*
3765 		 * Here we allow setting states that are not present in
3766 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3767 		 * with old userspace.
3768 		 */
3769 		if (xstate_bv & ~kvm_supported_xcr0() ||
3770 			mxcsr & ~mxcsr_feature_mask)
3771 			return -EINVAL;
3772 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3773 	} else {
3774 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3775 			mxcsr & ~mxcsr_feature_mask)
3776 			return -EINVAL;
3777 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3778 			guest_xsave->region, sizeof(struct fxregs_state));
3779 	}
3780 	return 0;
3781 }
3782 
3783 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3784 					struct kvm_xcrs *guest_xcrs)
3785 {
3786 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3787 		guest_xcrs->nr_xcrs = 0;
3788 		return;
3789 	}
3790 
3791 	guest_xcrs->nr_xcrs = 1;
3792 	guest_xcrs->flags = 0;
3793 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3794 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3795 }
3796 
3797 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3798 				       struct kvm_xcrs *guest_xcrs)
3799 {
3800 	int i, r = 0;
3801 
3802 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3803 		return -EINVAL;
3804 
3805 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3806 		return -EINVAL;
3807 
3808 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3809 		/* Only support XCR0 currently */
3810 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3811 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3812 				guest_xcrs->xcrs[i].value);
3813 			break;
3814 		}
3815 	if (r)
3816 		r = -EINVAL;
3817 	return r;
3818 }
3819 
3820 /*
3821  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3822  * stopped by the hypervisor.  This function will be called from the host only.
3823  * EINVAL is returned when the host attempts to set the flag for a guest that
3824  * does not support pv clocks.
3825  */
3826 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3827 {
3828 	if (!vcpu->arch.pv_time_enabled)
3829 		return -EINVAL;
3830 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3831 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3832 	return 0;
3833 }
3834 
3835 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3836 				     struct kvm_enable_cap *cap)
3837 {
3838 	int r;
3839 	uint16_t vmcs_version;
3840 	void __user *user_ptr;
3841 
3842 	if (cap->flags)
3843 		return -EINVAL;
3844 
3845 	switch (cap->cap) {
3846 	case KVM_CAP_HYPERV_SYNIC2:
3847 		if (cap->args[0])
3848 			return -EINVAL;
3849 		/* fall through */
3850 
3851 	case KVM_CAP_HYPERV_SYNIC:
3852 		if (!irqchip_in_kernel(vcpu->kvm))
3853 			return -EINVAL;
3854 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3855 					     KVM_CAP_HYPERV_SYNIC2);
3856 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3857 		if (!kvm_x86_ops->nested_enable_evmcs)
3858 			return -ENOTTY;
3859 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3860 		if (!r) {
3861 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
3862 			if (copy_to_user(user_ptr, &vmcs_version,
3863 					 sizeof(vmcs_version)))
3864 				r = -EFAULT;
3865 		}
3866 		return r;
3867 
3868 	default:
3869 		return -EINVAL;
3870 	}
3871 }
3872 
3873 long kvm_arch_vcpu_ioctl(struct file *filp,
3874 			 unsigned int ioctl, unsigned long arg)
3875 {
3876 	struct kvm_vcpu *vcpu = filp->private_data;
3877 	void __user *argp = (void __user *)arg;
3878 	int r;
3879 	union {
3880 		struct kvm_lapic_state *lapic;
3881 		struct kvm_xsave *xsave;
3882 		struct kvm_xcrs *xcrs;
3883 		void *buffer;
3884 	} u;
3885 
3886 	vcpu_load(vcpu);
3887 
3888 	u.buffer = NULL;
3889 	switch (ioctl) {
3890 	case KVM_GET_LAPIC: {
3891 		r = -EINVAL;
3892 		if (!lapic_in_kernel(vcpu))
3893 			goto out;
3894 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
3895 				GFP_KERNEL_ACCOUNT);
3896 
3897 		r = -ENOMEM;
3898 		if (!u.lapic)
3899 			goto out;
3900 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3901 		if (r)
3902 			goto out;
3903 		r = -EFAULT;
3904 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3905 			goto out;
3906 		r = 0;
3907 		break;
3908 	}
3909 	case KVM_SET_LAPIC: {
3910 		r = -EINVAL;
3911 		if (!lapic_in_kernel(vcpu))
3912 			goto out;
3913 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3914 		if (IS_ERR(u.lapic)) {
3915 			r = PTR_ERR(u.lapic);
3916 			goto out_nofree;
3917 		}
3918 
3919 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3920 		break;
3921 	}
3922 	case KVM_INTERRUPT: {
3923 		struct kvm_interrupt irq;
3924 
3925 		r = -EFAULT;
3926 		if (copy_from_user(&irq, argp, sizeof(irq)))
3927 			goto out;
3928 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3929 		break;
3930 	}
3931 	case KVM_NMI: {
3932 		r = kvm_vcpu_ioctl_nmi(vcpu);
3933 		break;
3934 	}
3935 	case KVM_SMI: {
3936 		r = kvm_vcpu_ioctl_smi(vcpu);
3937 		break;
3938 	}
3939 	case KVM_SET_CPUID: {
3940 		struct kvm_cpuid __user *cpuid_arg = argp;
3941 		struct kvm_cpuid cpuid;
3942 
3943 		r = -EFAULT;
3944 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3945 			goto out;
3946 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3947 		break;
3948 	}
3949 	case KVM_SET_CPUID2: {
3950 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3951 		struct kvm_cpuid2 cpuid;
3952 
3953 		r = -EFAULT;
3954 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3955 			goto out;
3956 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3957 					      cpuid_arg->entries);
3958 		break;
3959 	}
3960 	case KVM_GET_CPUID2: {
3961 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3962 		struct kvm_cpuid2 cpuid;
3963 
3964 		r = -EFAULT;
3965 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3966 			goto out;
3967 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3968 					      cpuid_arg->entries);
3969 		if (r)
3970 			goto out;
3971 		r = -EFAULT;
3972 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3973 			goto out;
3974 		r = 0;
3975 		break;
3976 	}
3977 	case KVM_GET_MSRS: {
3978 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3979 		r = msr_io(vcpu, argp, do_get_msr, 1);
3980 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3981 		break;
3982 	}
3983 	case KVM_SET_MSRS: {
3984 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3985 		r = msr_io(vcpu, argp, do_set_msr, 0);
3986 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3987 		break;
3988 	}
3989 	case KVM_TPR_ACCESS_REPORTING: {
3990 		struct kvm_tpr_access_ctl tac;
3991 
3992 		r = -EFAULT;
3993 		if (copy_from_user(&tac, argp, sizeof(tac)))
3994 			goto out;
3995 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3996 		if (r)
3997 			goto out;
3998 		r = -EFAULT;
3999 		if (copy_to_user(argp, &tac, sizeof(tac)))
4000 			goto out;
4001 		r = 0;
4002 		break;
4003 	};
4004 	case KVM_SET_VAPIC_ADDR: {
4005 		struct kvm_vapic_addr va;
4006 		int idx;
4007 
4008 		r = -EINVAL;
4009 		if (!lapic_in_kernel(vcpu))
4010 			goto out;
4011 		r = -EFAULT;
4012 		if (copy_from_user(&va, argp, sizeof(va)))
4013 			goto out;
4014 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4015 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4016 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4017 		break;
4018 	}
4019 	case KVM_X86_SETUP_MCE: {
4020 		u64 mcg_cap;
4021 
4022 		r = -EFAULT;
4023 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4024 			goto out;
4025 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4026 		break;
4027 	}
4028 	case KVM_X86_SET_MCE: {
4029 		struct kvm_x86_mce mce;
4030 
4031 		r = -EFAULT;
4032 		if (copy_from_user(&mce, argp, sizeof(mce)))
4033 			goto out;
4034 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4035 		break;
4036 	}
4037 	case KVM_GET_VCPU_EVENTS: {
4038 		struct kvm_vcpu_events events;
4039 
4040 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4041 
4042 		r = -EFAULT;
4043 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4044 			break;
4045 		r = 0;
4046 		break;
4047 	}
4048 	case KVM_SET_VCPU_EVENTS: {
4049 		struct kvm_vcpu_events events;
4050 
4051 		r = -EFAULT;
4052 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4053 			break;
4054 
4055 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4056 		break;
4057 	}
4058 	case KVM_GET_DEBUGREGS: {
4059 		struct kvm_debugregs dbgregs;
4060 
4061 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4062 
4063 		r = -EFAULT;
4064 		if (copy_to_user(argp, &dbgregs,
4065 				 sizeof(struct kvm_debugregs)))
4066 			break;
4067 		r = 0;
4068 		break;
4069 	}
4070 	case KVM_SET_DEBUGREGS: {
4071 		struct kvm_debugregs dbgregs;
4072 
4073 		r = -EFAULT;
4074 		if (copy_from_user(&dbgregs, argp,
4075 				   sizeof(struct kvm_debugregs)))
4076 			break;
4077 
4078 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4079 		break;
4080 	}
4081 	case KVM_GET_XSAVE: {
4082 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4083 		r = -ENOMEM;
4084 		if (!u.xsave)
4085 			break;
4086 
4087 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4088 
4089 		r = -EFAULT;
4090 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4091 			break;
4092 		r = 0;
4093 		break;
4094 	}
4095 	case KVM_SET_XSAVE: {
4096 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4097 		if (IS_ERR(u.xsave)) {
4098 			r = PTR_ERR(u.xsave);
4099 			goto out_nofree;
4100 		}
4101 
4102 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4103 		break;
4104 	}
4105 	case KVM_GET_XCRS: {
4106 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4107 		r = -ENOMEM;
4108 		if (!u.xcrs)
4109 			break;
4110 
4111 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4112 
4113 		r = -EFAULT;
4114 		if (copy_to_user(argp, u.xcrs,
4115 				 sizeof(struct kvm_xcrs)))
4116 			break;
4117 		r = 0;
4118 		break;
4119 	}
4120 	case KVM_SET_XCRS: {
4121 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4122 		if (IS_ERR(u.xcrs)) {
4123 			r = PTR_ERR(u.xcrs);
4124 			goto out_nofree;
4125 		}
4126 
4127 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4128 		break;
4129 	}
4130 	case KVM_SET_TSC_KHZ: {
4131 		u32 user_tsc_khz;
4132 
4133 		r = -EINVAL;
4134 		user_tsc_khz = (u32)arg;
4135 
4136 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4137 			goto out;
4138 
4139 		if (user_tsc_khz == 0)
4140 			user_tsc_khz = tsc_khz;
4141 
4142 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4143 			r = 0;
4144 
4145 		goto out;
4146 	}
4147 	case KVM_GET_TSC_KHZ: {
4148 		r = vcpu->arch.virtual_tsc_khz;
4149 		goto out;
4150 	}
4151 	case KVM_KVMCLOCK_CTRL: {
4152 		r = kvm_set_guest_paused(vcpu);
4153 		goto out;
4154 	}
4155 	case KVM_ENABLE_CAP: {
4156 		struct kvm_enable_cap cap;
4157 
4158 		r = -EFAULT;
4159 		if (copy_from_user(&cap, argp, sizeof(cap)))
4160 			goto out;
4161 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4162 		break;
4163 	}
4164 	case KVM_GET_NESTED_STATE: {
4165 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4166 		u32 user_data_size;
4167 
4168 		r = -EINVAL;
4169 		if (!kvm_x86_ops->get_nested_state)
4170 			break;
4171 
4172 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4173 		r = -EFAULT;
4174 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4175 			break;
4176 
4177 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4178 						  user_data_size);
4179 		if (r < 0)
4180 			break;
4181 
4182 		if (r > user_data_size) {
4183 			if (put_user(r, &user_kvm_nested_state->size))
4184 				r = -EFAULT;
4185 			else
4186 				r = -E2BIG;
4187 			break;
4188 		}
4189 
4190 		r = 0;
4191 		break;
4192 	}
4193 	case KVM_SET_NESTED_STATE: {
4194 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4195 		struct kvm_nested_state kvm_state;
4196 
4197 		r = -EINVAL;
4198 		if (!kvm_x86_ops->set_nested_state)
4199 			break;
4200 
4201 		r = -EFAULT;
4202 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4203 			break;
4204 
4205 		r = -EINVAL;
4206 		if (kvm_state.size < sizeof(kvm_state))
4207 			break;
4208 
4209 		if (kvm_state.flags &
4210 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4211 		      | KVM_STATE_NESTED_EVMCS))
4212 			break;
4213 
4214 		/* nested_run_pending implies guest_mode.  */
4215 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4216 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4217 			break;
4218 
4219 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4220 		break;
4221 	}
4222 	case KVM_GET_SUPPORTED_HV_CPUID: {
4223 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4224 		struct kvm_cpuid2 cpuid;
4225 
4226 		r = -EFAULT;
4227 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4228 			goto out;
4229 
4230 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4231 						cpuid_arg->entries);
4232 		if (r)
4233 			goto out;
4234 
4235 		r = -EFAULT;
4236 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4237 			goto out;
4238 		r = 0;
4239 		break;
4240 	}
4241 	default:
4242 		r = -EINVAL;
4243 	}
4244 out:
4245 	kfree(u.buffer);
4246 out_nofree:
4247 	vcpu_put(vcpu);
4248 	return r;
4249 }
4250 
4251 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4252 {
4253 	return VM_FAULT_SIGBUS;
4254 }
4255 
4256 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4257 {
4258 	int ret;
4259 
4260 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4261 		return -EINVAL;
4262 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4263 	return ret;
4264 }
4265 
4266 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4267 					      u64 ident_addr)
4268 {
4269 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4270 }
4271 
4272 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4273 					  u32 kvm_nr_mmu_pages)
4274 {
4275 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4276 		return -EINVAL;
4277 
4278 	mutex_lock(&kvm->slots_lock);
4279 
4280 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4281 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4282 
4283 	mutex_unlock(&kvm->slots_lock);
4284 	return 0;
4285 }
4286 
4287 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4288 {
4289 	return kvm->arch.n_max_mmu_pages;
4290 }
4291 
4292 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4293 {
4294 	struct kvm_pic *pic = kvm->arch.vpic;
4295 	int r;
4296 
4297 	r = 0;
4298 	switch (chip->chip_id) {
4299 	case KVM_IRQCHIP_PIC_MASTER:
4300 		memcpy(&chip->chip.pic, &pic->pics[0],
4301 			sizeof(struct kvm_pic_state));
4302 		break;
4303 	case KVM_IRQCHIP_PIC_SLAVE:
4304 		memcpy(&chip->chip.pic, &pic->pics[1],
4305 			sizeof(struct kvm_pic_state));
4306 		break;
4307 	case KVM_IRQCHIP_IOAPIC:
4308 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4309 		break;
4310 	default:
4311 		r = -EINVAL;
4312 		break;
4313 	}
4314 	return r;
4315 }
4316 
4317 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4318 {
4319 	struct kvm_pic *pic = kvm->arch.vpic;
4320 	int r;
4321 
4322 	r = 0;
4323 	switch (chip->chip_id) {
4324 	case KVM_IRQCHIP_PIC_MASTER:
4325 		spin_lock(&pic->lock);
4326 		memcpy(&pic->pics[0], &chip->chip.pic,
4327 			sizeof(struct kvm_pic_state));
4328 		spin_unlock(&pic->lock);
4329 		break;
4330 	case KVM_IRQCHIP_PIC_SLAVE:
4331 		spin_lock(&pic->lock);
4332 		memcpy(&pic->pics[1], &chip->chip.pic,
4333 			sizeof(struct kvm_pic_state));
4334 		spin_unlock(&pic->lock);
4335 		break;
4336 	case KVM_IRQCHIP_IOAPIC:
4337 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4338 		break;
4339 	default:
4340 		r = -EINVAL;
4341 		break;
4342 	}
4343 	kvm_pic_update_irq(pic);
4344 	return r;
4345 }
4346 
4347 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4348 {
4349 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4350 
4351 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4352 
4353 	mutex_lock(&kps->lock);
4354 	memcpy(ps, &kps->channels, sizeof(*ps));
4355 	mutex_unlock(&kps->lock);
4356 	return 0;
4357 }
4358 
4359 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4360 {
4361 	int i;
4362 	struct kvm_pit *pit = kvm->arch.vpit;
4363 
4364 	mutex_lock(&pit->pit_state.lock);
4365 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4366 	for (i = 0; i < 3; i++)
4367 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4368 	mutex_unlock(&pit->pit_state.lock);
4369 	return 0;
4370 }
4371 
4372 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4373 {
4374 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4375 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4376 		sizeof(ps->channels));
4377 	ps->flags = kvm->arch.vpit->pit_state.flags;
4378 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4379 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4380 	return 0;
4381 }
4382 
4383 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4384 {
4385 	int start = 0;
4386 	int i;
4387 	u32 prev_legacy, cur_legacy;
4388 	struct kvm_pit *pit = kvm->arch.vpit;
4389 
4390 	mutex_lock(&pit->pit_state.lock);
4391 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4392 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4393 	if (!prev_legacy && cur_legacy)
4394 		start = 1;
4395 	memcpy(&pit->pit_state.channels, &ps->channels,
4396 	       sizeof(pit->pit_state.channels));
4397 	pit->pit_state.flags = ps->flags;
4398 	for (i = 0; i < 3; i++)
4399 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4400 				   start && i == 0);
4401 	mutex_unlock(&pit->pit_state.lock);
4402 	return 0;
4403 }
4404 
4405 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4406 				 struct kvm_reinject_control *control)
4407 {
4408 	struct kvm_pit *pit = kvm->arch.vpit;
4409 
4410 	if (!pit)
4411 		return -ENXIO;
4412 
4413 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4414 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4415 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4416 	 */
4417 	mutex_lock(&pit->pit_state.lock);
4418 	kvm_pit_set_reinject(pit, control->pit_reinject);
4419 	mutex_unlock(&pit->pit_state.lock);
4420 
4421 	return 0;
4422 }
4423 
4424 /**
4425  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4426  * @kvm: kvm instance
4427  * @log: slot id and address to which we copy the log
4428  *
4429  * Steps 1-4 below provide general overview of dirty page logging. See
4430  * kvm_get_dirty_log_protect() function description for additional details.
4431  *
4432  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4433  * always flush the TLB (step 4) even if previous step failed  and the dirty
4434  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4435  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4436  * writes will be marked dirty for next log read.
4437  *
4438  *   1. Take a snapshot of the bit and clear it if needed.
4439  *   2. Write protect the corresponding page.
4440  *   3. Copy the snapshot to the userspace.
4441  *   4. Flush TLB's if needed.
4442  */
4443 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4444 {
4445 	bool flush = false;
4446 	int r;
4447 
4448 	mutex_lock(&kvm->slots_lock);
4449 
4450 	/*
4451 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4452 	 */
4453 	if (kvm_x86_ops->flush_log_dirty)
4454 		kvm_x86_ops->flush_log_dirty(kvm);
4455 
4456 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4457 
4458 	/*
4459 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4460 	 * kvm_mmu_slot_remove_write_access().
4461 	 */
4462 	lockdep_assert_held(&kvm->slots_lock);
4463 	if (flush)
4464 		kvm_flush_remote_tlbs(kvm);
4465 
4466 	mutex_unlock(&kvm->slots_lock);
4467 	return r;
4468 }
4469 
4470 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4471 {
4472 	bool flush = false;
4473 	int r;
4474 
4475 	mutex_lock(&kvm->slots_lock);
4476 
4477 	/*
4478 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4479 	 */
4480 	if (kvm_x86_ops->flush_log_dirty)
4481 		kvm_x86_ops->flush_log_dirty(kvm);
4482 
4483 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4484 
4485 	/*
4486 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4487 	 * kvm_mmu_slot_remove_write_access().
4488 	 */
4489 	lockdep_assert_held(&kvm->slots_lock);
4490 	if (flush)
4491 		kvm_flush_remote_tlbs(kvm);
4492 
4493 	mutex_unlock(&kvm->slots_lock);
4494 	return r;
4495 }
4496 
4497 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4498 			bool line_status)
4499 {
4500 	if (!irqchip_in_kernel(kvm))
4501 		return -ENXIO;
4502 
4503 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4504 					irq_event->irq, irq_event->level,
4505 					line_status);
4506 	return 0;
4507 }
4508 
4509 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4510 			    struct kvm_enable_cap *cap)
4511 {
4512 	int r;
4513 
4514 	if (cap->flags)
4515 		return -EINVAL;
4516 
4517 	switch (cap->cap) {
4518 	case KVM_CAP_DISABLE_QUIRKS:
4519 		kvm->arch.disabled_quirks = cap->args[0];
4520 		r = 0;
4521 		break;
4522 	case KVM_CAP_SPLIT_IRQCHIP: {
4523 		mutex_lock(&kvm->lock);
4524 		r = -EINVAL;
4525 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4526 			goto split_irqchip_unlock;
4527 		r = -EEXIST;
4528 		if (irqchip_in_kernel(kvm))
4529 			goto split_irqchip_unlock;
4530 		if (kvm->created_vcpus)
4531 			goto split_irqchip_unlock;
4532 		r = kvm_setup_empty_irq_routing(kvm);
4533 		if (r)
4534 			goto split_irqchip_unlock;
4535 		/* Pairs with irqchip_in_kernel. */
4536 		smp_wmb();
4537 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4538 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4539 		r = 0;
4540 split_irqchip_unlock:
4541 		mutex_unlock(&kvm->lock);
4542 		break;
4543 	}
4544 	case KVM_CAP_X2APIC_API:
4545 		r = -EINVAL;
4546 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4547 			break;
4548 
4549 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4550 			kvm->arch.x2apic_format = true;
4551 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4552 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4553 
4554 		r = 0;
4555 		break;
4556 	case KVM_CAP_X86_DISABLE_EXITS:
4557 		r = -EINVAL;
4558 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4559 			break;
4560 
4561 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4562 			kvm_can_mwait_in_guest())
4563 			kvm->arch.mwait_in_guest = true;
4564 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4565 			kvm->arch.hlt_in_guest = true;
4566 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4567 			kvm->arch.pause_in_guest = true;
4568 		r = 0;
4569 		break;
4570 	case KVM_CAP_MSR_PLATFORM_INFO:
4571 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4572 		r = 0;
4573 		break;
4574 	case KVM_CAP_EXCEPTION_PAYLOAD:
4575 		kvm->arch.exception_payload_enabled = cap->args[0];
4576 		r = 0;
4577 		break;
4578 	default:
4579 		r = -EINVAL;
4580 		break;
4581 	}
4582 	return r;
4583 }
4584 
4585 long kvm_arch_vm_ioctl(struct file *filp,
4586 		       unsigned int ioctl, unsigned long arg)
4587 {
4588 	struct kvm *kvm = filp->private_data;
4589 	void __user *argp = (void __user *)arg;
4590 	int r = -ENOTTY;
4591 	/*
4592 	 * This union makes it completely explicit to gcc-3.x
4593 	 * that these two variables' stack usage should be
4594 	 * combined, not added together.
4595 	 */
4596 	union {
4597 		struct kvm_pit_state ps;
4598 		struct kvm_pit_state2 ps2;
4599 		struct kvm_pit_config pit_config;
4600 	} u;
4601 
4602 	switch (ioctl) {
4603 	case KVM_SET_TSS_ADDR:
4604 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4605 		break;
4606 	case KVM_SET_IDENTITY_MAP_ADDR: {
4607 		u64 ident_addr;
4608 
4609 		mutex_lock(&kvm->lock);
4610 		r = -EINVAL;
4611 		if (kvm->created_vcpus)
4612 			goto set_identity_unlock;
4613 		r = -EFAULT;
4614 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4615 			goto set_identity_unlock;
4616 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4617 set_identity_unlock:
4618 		mutex_unlock(&kvm->lock);
4619 		break;
4620 	}
4621 	case KVM_SET_NR_MMU_PAGES:
4622 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4623 		break;
4624 	case KVM_GET_NR_MMU_PAGES:
4625 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4626 		break;
4627 	case KVM_CREATE_IRQCHIP: {
4628 		mutex_lock(&kvm->lock);
4629 
4630 		r = -EEXIST;
4631 		if (irqchip_in_kernel(kvm))
4632 			goto create_irqchip_unlock;
4633 
4634 		r = -EINVAL;
4635 		if (kvm->created_vcpus)
4636 			goto create_irqchip_unlock;
4637 
4638 		r = kvm_pic_init(kvm);
4639 		if (r)
4640 			goto create_irqchip_unlock;
4641 
4642 		r = kvm_ioapic_init(kvm);
4643 		if (r) {
4644 			kvm_pic_destroy(kvm);
4645 			goto create_irqchip_unlock;
4646 		}
4647 
4648 		r = kvm_setup_default_irq_routing(kvm);
4649 		if (r) {
4650 			kvm_ioapic_destroy(kvm);
4651 			kvm_pic_destroy(kvm);
4652 			goto create_irqchip_unlock;
4653 		}
4654 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4655 		smp_wmb();
4656 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4657 	create_irqchip_unlock:
4658 		mutex_unlock(&kvm->lock);
4659 		break;
4660 	}
4661 	case KVM_CREATE_PIT:
4662 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4663 		goto create_pit;
4664 	case KVM_CREATE_PIT2:
4665 		r = -EFAULT;
4666 		if (copy_from_user(&u.pit_config, argp,
4667 				   sizeof(struct kvm_pit_config)))
4668 			goto out;
4669 	create_pit:
4670 		mutex_lock(&kvm->lock);
4671 		r = -EEXIST;
4672 		if (kvm->arch.vpit)
4673 			goto create_pit_unlock;
4674 		r = -ENOMEM;
4675 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4676 		if (kvm->arch.vpit)
4677 			r = 0;
4678 	create_pit_unlock:
4679 		mutex_unlock(&kvm->lock);
4680 		break;
4681 	case KVM_GET_IRQCHIP: {
4682 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4683 		struct kvm_irqchip *chip;
4684 
4685 		chip = memdup_user(argp, sizeof(*chip));
4686 		if (IS_ERR(chip)) {
4687 			r = PTR_ERR(chip);
4688 			goto out;
4689 		}
4690 
4691 		r = -ENXIO;
4692 		if (!irqchip_kernel(kvm))
4693 			goto get_irqchip_out;
4694 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4695 		if (r)
4696 			goto get_irqchip_out;
4697 		r = -EFAULT;
4698 		if (copy_to_user(argp, chip, sizeof(*chip)))
4699 			goto get_irqchip_out;
4700 		r = 0;
4701 	get_irqchip_out:
4702 		kfree(chip);
4703 		break;
4704 	}
4705 	case KVM_SET_IRQCHIP: {
4706 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4707 		struct kvm_irqchip *chip;
4708 
4709 		chip = memdup_user(argp, sizeof(*chip));
4710 		if (IS_ERR(chip)) {
4711 			r = PTR_ERR(chip);
4712 			goto out;
4713 		}
4714 
4715 		r = -ENXIO;
4716 		if (!irqchip_kernel(kvm))
4717 			goto set_irqchip_out;
4718 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4719 		if (r)
4720 			goto set_irqchip_out;
4721 		r = 0;
4722 	set_irqchip_out:
4723 		kfree(chip);
4724 		break;
4725 	}
4726 	case KVM_GET_PIT: {
4727 		r = -EFAULT;
4728 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4729 			goto out;
4730 		r = -ENXIO;
4731 		if (!kvm->arch.vpit)
4732 			goto out;
4733 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4734 		if (r)
4735 			goto out;
4736 		r = -EFAULT;
4737 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4738 			goto out;
4739 		r = 0;
4740 		break;
4741 	}
4742 	case KVM_SET_PIT: {
4743 		r = -EFAULT;
4744 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4745 			goto out;
4746 		r = -ENXIO;
4747 		if (!kvm->arch.vpit)
4748 			goto out;
4749 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4750 		break;
4751 	}
4752 	case KVM_GET_PIT2: {
4753 		r = -ENXIO;
4754 		if (!kvm->arch.vpit)
4755 			goto out;
4756 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4757 		if (r)
4758 			goto out;
4759 		r = -EFAULT;
4760 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4761 			goto out;
4762 		r = 0;
4763 		break;
4764 	}
4765 	case KVM_SET_PIT2: {
4766 		r = -EFAULT;
4767 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4768 			goto out;
4769 		r = -ENXIO;
4770 		if (!kvm->arch.vpit)
4771 			goto out;
4772 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4773 		break;
4774 	}
4775 	case KVM_REINJECT_CONTROL: {
4776 		struct kvm_reinject_control control;
4777 		r =  -EFAULT;
4778 		if (copy_from_user(&control, argp, sizeof(control)))
4779 			goto out;
4780 		r = kvm_vm_ioctl_reinject(kvm, &control);
4781 		break;
4782 	}
4783 	case KVM_SET_BOOT_CPU_ID:
4784 		r = 0;
4785 		mutex_lock(&kvm->lock);
4786 		if (kvm->created_vcpus)
4787 			r = -EBUSY;
4788 		else
4789 			kvm->arch.bsp_vcpu_id = arg;
4790 		mutex_unlock(&kvm->lock);
4791 		break;
4792 	case KVM_XEN_HVM_CONFIG: {
4793 		struct kvm_xen_hvm_config xhc;
4794 		r = -EFAULT;
4795 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4796 			goto out;
4797 		r = -EINVAL;
4798 		if (xhc.flags)
4799 			goto out;
4800 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4801 		r = 0;
4802 		break;
4803 	}
4804 	case KVM_SET_CLOCK: {
4805 		struct kvm_clock_data user_ns;
4806 		u64 now_ns;
4807 
4808 		r = -EFAULT;
4809 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4810 			goto out;
4811 
4812 		r = -EINVAL;
4813 		if (user_ns.flags)
4814 			goto out;
4815 
4816 		r = 0;
4817 		/*
4818 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4819 		 * kvm_gen_update_masterclock() can be cut down to locked
4820 		 * pvclock_update_vm_gtod_copy().
4821 		 */
4822 		kvm_gen_update_masterclock(kvm);
4823 		now_ns = get_kvmclock_ns(kvm);
4824 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4825 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4826 		break;
4827 	}
4828 	case KVM_GET_CLOCK: {
4829 		struct kvm_clock_data user_ns;
4830 		u64 now_ns;
4831 
4832 		now_ns = get_kvmclock_ns(kvm);
4833 		user_ns.clock = now_ns;
4834 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4835 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4836 
4837 		r = -EFAULT;
4838 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4839 			goto out;
4840 		r = 0;
4841 		break;
4842 	}
4843 	case KVM_MEMORY_ENCRYPT_OP: {
4844 		r = -ENOTTY;
4845 		if (kvm_x86_ops->mem_enc_op)
4846 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4847 		break;
4848 	}
4849 	case KVM_MEMORY_ENCRYPT_REG_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_reg_region)
4858 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4859 		break;
4860 	}
4861 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4862 		struct kvm_enc_region region;
4863 
4864 		r = -EFAULT;
4865 		if (copy_from_user(&region, argp, sizeof(region)))
4866 			goto out;
4867 
4868 		r = -ENOTTY;
4869 		if (kvm_x86_ops->mem_enc_unreg_region)
4870 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4871 		break;
4872 	}
4873 	case KVM_HYPERV_EVENTFD: {
4874 		struct kvm_hyperv_eventfd hvevfd;
4875 
4876 		r = -EFAULT;
4877 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4878 			goto out;
4879 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4880 		break;
4881 	}
4882 	default:
4883 		r = -ENOTTY;
4884 	}
4885 out:
4886 	return r;
4887 }
4888 
4889 static void kvm_init_msr_list(void)
4890 {
4891 	u32 dummy[2];
4892 	unsigned i, j;
4893 
4894 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4895 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4896 			continue;
4897 
4898 		/*
4899 		 * Even MSRs that are valid in the host may not be exposed
4900 		 * to the guests in some cases.
4901 		 */
4902 		switch (msrs_to_save[i]) {
4903 		case MSR_IA32_BNDCFGS:
4904 			if (!kvm_mpx_supported())
4905 				continue;
4906 			break;
4907 		case MSR_TSC_AUX:
4908 			if (!kvm_x86_ops->rdtscp_supported())
4909 				continue;
4910 			break;
4911 		case MSR_IA32_RTIT_CTL:
4912 		case MSR_IA32_RTIT_STATUS:
4913 			if (!kvm_x86_ops->pt_supported())
4914 				continue;
4915 			break;
4916 		case MSR_IA32_RTIT_CR3_MATCH:
4917 			if (!kvm_x86_ops->pt_supported() ||
4918 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
4919 				continue;
4920 			break;
4921 		case MSR_IA32_RTIT_OUTPUT_BASE:
4922 		case MSR_IA32_RTIT_OUTPUT_MASK:
4923 			if (!kvm_x86_ops->pt_supported() ||
4924 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
4925 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
4926 				continue;
4927 			break;
4928 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
4929 			if (!kvm_x86_ops->pt_supported() ||
4930 				msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
4931 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
4932 				continue;
4933 			break;
4934 		}
4935 		default:
4936 			break;
4937 		}
4938 
4939 		if (j < i)
4940 			msrs_to_save[j] = msrs_to_save[i];
4941 		j++;
4942 	}
4943 	num_msrs_to_save = j;
4944 
4945 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4946 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4947 			continue;
4948 
4949 		if (j < i)
4950 			emulated_msrs[j] = emulated_msrs[i];
4951 		j++;
4952 	}
4953 	num_emulated_msrs = j;
4954 
4955 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4956 		struct kvm_msr_entry msr;
4957 
4958 		msr.index = msr_based_features[i];
4959 		if (kvm_get_msr_feature(&msr))
4960 			continue;
4961 
4962 		if (j < i)
4963 			msr_based_features[j] = msr_based_features[i];
4964 		j++;
4965 	}
4966 	num_msr_based_features = j;
4967 }
4968 
4969 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4970 			   const void *v)
4971 {
4972 	int handled = 0;
4973 	int n;
4974 
4975 	do {
4976 		n = min(len, 8);
4977 		if (!(lapic_in_kernel(vcpu) &&
4978 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4979 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4980 			break;
4981 		handled += n;
4982 		addr += n;
4983 		len -= n;
4984 		v += n;
4985 	} while (len);
4986 
4987 	return handled;
4988 }
4989 
4990 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4991 {
4992 	int handled = 0;
4993 	int n;
4994 
4995 	do {
4996 		n = min(len, 8);
4997 		if (!(lapic_in_kernel(vcpu) &&
4998 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4999 					 addr, n, v))
5000 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5001 			break;
5002 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5003 		handled += n;
5004 		addr += n;
5005 		len -= n;
5006 		v += n;
5007 	} while (len);
5008 
5009 	return handled;
5010 }
5011 
5012 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5013 			struct kvm_segment *var, int seg)
5014 {
5015 	kvm_x86_ops->set_segment(vcpu, var, seg);
5016 }
5017 
5018 void kvm_get_segment(struct kvm_vcpu *vcpu,
5019 		     struct kvm_segment *var, int seg)
5020 {
5021 	kvm_x86_ops->get_segment(vcpu, var, seg);
5022 }
5023 
5024 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5025 			   struct x86_exception *exception)
5026 {
5027 	gpa_t t_gpa;
5028 
5029 	BUG_ON(!mmu_is_nested(vcpu));
5030 
5031 	/* NPT walks are always user-walks */
5032 	access |= PFERR_USER_MASK;
5033 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5034 
5035 	return t_gpa;
5036 }
5037 
5038 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5039 			      struct x86_exception *exception)
5040 {
5041 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5042 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5043 }
5044 
5045  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5046 				struct x86_exception *exception)
5047 {
5048 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5049 	access |= PFERR_FETCH_MASK;
5050 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5051 }
5052 
5053 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5054 			       struct x86_exception *exception)
5055 {
5056 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5057 	access |= PFERR_WRITE_MASK;
5058 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5059 }
5060 
5061 /* uses this to access any guest's mapped memory without checking CPL */
5062 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5063 				struct x86_exception *exception)
5064 {
5065 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5066 }
5067 
5068 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5069 				      struct kvm_vcpu *vcpu, u32 access,
5070 				      struct x86_exception *exception)
5071 {
5072 	void *data = val;
5073 	int r = X86EMUL_CONTINUE;
5074 
5075 	while (bytes) {
5076 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5077 							    exception);
5078 		unsigned offset = addr & (PAGE_SIZE-1);
5079 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5080 		int ret;
5081 
5082 		if (gpa == UNMAPPED_GVA)
5083 			return X86EMUL_PROPAGATE_FAULT;
5084 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5085 					       offset, toread);
5086 		if (ret < 0) {
5087 			r = X86EMUL_IO_NEEDED;
5088 			goto out;
5089 		}
5090 
5091 		bytes -= toread;
5092 		data += toread;
5093 		addr += toread;
5094 	}
5095 out:
5096 	return r;
5097 }
5098 
5099 /* used for instruction fetching */
5100 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5101 				gva_t addr, void *val, unsigned int bytes,
5102 				struct x86_exception *exception)
5103 {
5104 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5105 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5106 	unsigned offset;
5107 	int ret;
5108 
5109 	/* Inline kvm_read_guest_virt_helper for speed.  */
5110 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5111 						    exception);
5112 	if (unlikely(gpa == UNMAPPED_GVA))
5113 		return X86EMUL_PROPAGATE_FAULT;
5114 
5115 	offset = addr & (PAGE_SIZE-1);
5116 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5117 		bytes = (unsigned)PAGE_SIZE - offset;
5118 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5119 				       offset, bytes);
5120 	if (unlikely(ret < 0))
5121 		return X86EMUL_IO_NEEDED;
5122 
5123 	return X86EMUL_CONTINUE;
5124 }
5125 
5126 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5127 			       gva_t addr, void *val, unsigned int bytes,
5128 			       struct x86_exception *exception)
5129 {
5130 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5131 
5132 	/*
5133 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5134 	 * is returned, but our callers are not ready for that and they blindly
5135 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5136 	 * uninitialized kernel stack memory into cr2 and error code.
5137 	 */
5138 	memset(exception, 0, sizeof(*exception));
5139 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5140 					  exception);
5141 }
5142 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5143 
5144 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5145 			     gva_t addr, void *val, unsigned int bytes,
5146 			     struct x86_exception *exception, bool system)
5147 {
5148 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5149 	u32 access = 0;
5150 
5151 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5152 		access |= PFERR_USER_MASK;
5153 
5154 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5155 }
5156 
5157 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5158 		unsigned long addr, void *val, unsigned int bytes)
5159 {
5160 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5161 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5162 
5163 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5164 }
5165 
5166 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5167 				      struct kvm_vcpu *vcpu, u32 access,
5168 				      struct x86_exception *exception)
5169 {
5170 	void *data = val;
5171 	int r = X86EMUL_CONTINUE;
5172 
5173 	while (bytes) {
5174 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5175 							     access,
5176 							     exception);
5177 		unsigned offset = addr & (PAGE_SIZE-1);
5178 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5179 		int ret;
5180 
5181 		if (gpa == UNMAPPED_GVA)
5182 			return X86EMUL_PROPAGATE_FAULT;
5183 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5184 		if (ret < 0) {
5185 			r = X86EMUL_IO_NEEDED;
5186 			goto out;
5187 		}
5188 
5189 		bytes -= towrite;
5190 		data += towrite;
5191 		addr += towrite;
5192 	}
5193 out:
5194 	return r;
5195 }
5196 
5197 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5198 			      unsigned int bytes, struct x86_exception *exception,
5199 			      bool system)
5200 {
5201 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5202 	u32 access = PFERR_WRITE_MASK;
5203 
5204 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5205 		access |= PFERR_USER_MASK;
5206 
5207 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5208 					   access, exception);
5209 }
5210 
5211 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5212 				unsigned int bytes, struct x86_exception *exception)
5213 {
5214 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5215 	vcpu->arch.l1tf_flush_l1d = true;
5216 
5217 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5218 					   PFERR_WRITE_MASK, exception);
5219 }
5220 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5221 
5222 int handle_ud(struct kvm_vcpu *vcpu)
5223 {
5224 	int emul_type = EMULTYPE_TRAP_UD;
5225 	enum emulation_result er;
5226 	char sig[5]; /* ud2; .ascii "kvm" */
5227 	struct x86_exception e;
5228 
5229 	if (force_emulation_prefix &&
5230 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5231 				sig, sizeof(sig), &e) == 0 &&
5232 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5233 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5234 		emul_type = 0;
5235 	}
5236 
5237 	er = kvm_emulate_instruction(vcpu, emul_type);
5238 	if (er == EMULATE_USER_EXIT)
5239 		return 0;
5240 	if (er != EMULATE_DONE)
5241 		kvm_queue_exception(vcpu, UD_VECTOR);
5242 	return 1;
5243 }
5244 EXPORT_SYMBOL_GPL(handle_ud);
5245 
5246 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5247 			    gpa_t gpa, bool write)
5248 {
5249 	/* For APIC access vmexit */
5250 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5251 		return 1;
5252 
5253 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5254 		trace_vcpu_match_mmio(gva, gpa, write, true);
5255 		return 1;
5256 	}
5257 
5258 	return 0;
5259 }
5260 
5261 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5262 				gpa_t *gpa, struct x86_exception *exception,
5263 				bool write)
5264 {
5265 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5266 		| (write ? PFERR_WRITE_MASK : 0);
5267 
5268 	/*
5269 	 * currently PKRU is only applied to ept enabled guest so
5270 	 * there is no pkey in EPT page table for L1 guest or EPT
5271 	 * shadow page table for L2 guest.
5272 	 */
5273 	if (vcpu_match_mmio_gva(vcpu, gva)
5274 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5275 				 vcpu->arch.access, 0, access)) {
5276 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5277 					(gva & (PAGE_SIZE - 1));
5278 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5279 		return 1;
5280 	}
5281 
5282 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5283 
5284 	if (*gpa == UNMAPPED_GVA)
5285 		return -1;
5286 
5287 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5288 }
5289 
5290 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5291 			const void *val, int bytes)
5292 {
5293 	int ret;
5294 
5295 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5296 	if (ret < 0)
5297 		return 0;
5298 	kvm_page_track_write(vcpu, gpa, val, bytes);
5299 	return 1;
5300 }
5301 
5302 struct read_write_emulator_ops {
5303 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5304 				  int bytes);
5305 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5306 				  void *val, int bytes);
5307 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5308 			       int bytes, void *val);
5309 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5310 				    void *val, int bytes);
5311 	bool write;
5312 };
5313 
5314 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5315 {
5316 	if (vcpu->mmio_read_completed) {
5317 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5318 			       vcpu->mmio_fragments[0].gpa, val);
5319 		vcpu->mmio_read_completed = 0;
5320 		return 1;
5321 	}
5322 
5323 	return 0;
5324 }
5325 
5326 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5327 			void *val, int bytes)
5328 {
5329 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5330 }
5331 
5332 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5333 			 void *val, int bytes)
5334 {
5335 	return emulator_write_phys(vcpu, gpa, val, bytes);
5336 }
5337 
5338 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5339 {
5340 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5341 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5342 }
5343 
5344 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5345 			  void *val, int bytes)
5346 {
5347 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5348 	return X86EMUL_IO_NEEDED;
5349 }
5350 
5351 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5352 			   void *val, int bytes)
5353 {
5354 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5355 
5356 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5357 	return X86EMUL_CONTINUE;
5358 }
5359 
5360 static const struct read_write_emulator_ops read_emultor = {
5361 	.read_write_prepare = read_prepare,
5362 	.read_write_emulate = read_emulate,
5363 	.read_write_mmio = vcpu_mmio_read,
5364 	.read_write_exit_mmio = read_exit_mmio,
5365 };
5366 
5367 static const struct read_write_emulator_ops write_emultor = {
5368 	.read_write_emulate = write_emulate,
5369 	.read_write_mmio = write_mmio,
5370 	.read_write_exit_mmio = write_exit_mmio,
5371 	.write = true,
5372 };
5373 
5374 static int emulator_read_write_onepage(unsigned long addr, void *val,
5375 				       unsigned int bytes,
5376 				       struct x86_exception *exception,
5377 				       struct kvm_vcpu *vcpu,
5378 				       const struct read_write_emulator_ops *ops)
5379 {
5380 	gpa_t gpa;
5381 	int handled, ret;
5382 	bool write = ops->write;
5383 	struct kvm_mmio_fragment *frag;
5384 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5385 
5386 	/*
5387 	 * If the exit was due to a NPF we may already have a GPA.
5388 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5389 	 * Note, this cannot be used on string operations since string
5390 	 * operation using rep will only have the initial GPA from the NPF
5391 	 * occurred.
5392 	 */
5393 	if (vcpu->arch.gpa_available &&
5394 	    emulator_can_use_gpa(ctxt) &&
5395 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5396 		gpa = vcpu->arch.gpa_val;
5397 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5398 	} else {
5399 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5400 		if (ret < 0)
5401 			return X86EMUL_PROPAGATE_FAULT;
5402 	}
5403 
5404 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5405 		return X86EMUL_CONTINUE;
5406 
5407 	/*
5408 	 * Is this MMIO handled locally?
5409 	 */
5410 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5411 	if (handled == bytes)
5412 		return X86EMUL_CONTINUE;
5413 
5414 	gpa += handled;
5415 	bytes -= handled;
5416 	val += handled;
5417 
5418 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5419 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5420 	frag->gpa = gpa;
5421 	frag->data = val;
5422 	frag->len = bytes;
5423 	return X86EMUL_CONTINUE;
5424 }
5425 
5426 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5427 			unsigned long addr,
5428 			void *val, unsigned int bytes,
5429 			struct x86_exception *exception,
5430 			const struct read_write_emulator_ops *ops)
5431 {
5432 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5433 	gpa_t gpa;
5434 	int rc;
5435 
5436 	if (ops->read_write_prepare &&
5437 		  ops->read_write_prepare(vcpu, val, bytes))
5438 		return X86EMUL_CONTINUE;
5439 
5440 	vcpu->mmio_nr_fragments = 0;
5441 
5442 	/* Crossing a page boundary? */
5443 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5444 		int now;
5445 
5446 		now = -addr & ~PAGE_MASK;
5447 		rc = emulator_read_write_onepage(addr, val, now, exception,
5448 						 vcpu, ops);
5449 
5450 		if (rc != X86EMUL_CONTINUE)
5451 			return rc;
5452 		addr += now;
5453 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5454 			addr = (u32)addr;
5455 		val += now;
5456 		bytes -= now;
5457 	}
5458 
5459 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5460 					 vcpu, ops);
5461 	if (rc != X86EMUL_CONTINUE)
5462 		return rc;
5463 
5464 	if (!vcpu->mmio_nr_fragments)
5465 		return rc;
5466 
5467 	gpa = vcpu->mmio_fragments[0].gpa;
5468 
5469 	vcpu->mmio_needed = 1;
5470 	vcpu->mmio_cur_fragment = 0;
5471 
5472 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5473 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5474 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5475 	vcpu->run->mmio.phys_addr = gpa;
5476 
5477 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5478 }
5479 
5480 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5481 				  unsigned long addr,
5482 				  void *val,
5483 				  unsigned int bytes,
5484 				  struct x86_exception *exception)
5485 {
5486 	return emulator_read_write(ctxt, addr, val, bytes,
5487 				   exception, &read_emultor);
5488 }
5489 
5490 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5491 			    unsigned long addr,
5492 			    const void *val,
5493 			    unsigned int bytes,
5494 			    struct x86_exception *exception)
5495 {
5496 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5497 				   exception, &write_emultor);
5498 }
5499 
5500 #define CMPXCHG_TYPE(t, ptr, old, new) \
5501 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5502 
5503 #ifdef CONFIG_X86_64
5504 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5505 #else
5506 #  define CMPXCHG64(ptr, old, new) \
5507 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5508 #endif
5509 
5510 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5511 				     unsigned long addr,
5512 				     const void *old,
5513 				     const void *new,
5514 				     unsigned int bytes,
5515 				     struct x86_exception *exception)
5516 {
5517 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5518 	gpa_t gpa;
5519 	struct page *page;
5520 	char *kaddr;
5521 	bool exchanged;
5522 
5523 	/* guests cmpxchg8b have to be emulated atomically */
5524 	if (bytes > 8 || (bytes & (bytes - 1)))
5525 		goto emul_write;
5526 
5527 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5528 
5529 	if (gpa == UNMAPPED_GVA ||
5530 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5531 		goto emul_write;
5532 
5533 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5534 		goto emul_write;
5535 
5536 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5537 	if (is_error_page(page))
5538 		goto emul_write;
5539 
5540 	kaddr = kmap_atomic(page);
5541 	kaddr += offset_in_page(gpa);
5542 	switch (bytes) {
5543 	case 1:
5544 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5545 		break;
5546 	case 2:
5547 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5548 		break;
5549 	case 4:
5550 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5551 		break;
5552 	case 8:
5553 		exchanged = CMPXCHG64(kaddr, old, new);
5554 		break;
5555 	default:
5556 		BUG();
5557 	}
5558 	kunmap_atomic(kaddr);
5559 	kvm_release_page_dirty(page);
5560 
5561 	if (!exchanged)
5562 		return X86EMUL_CMPXCHG_FAILED;
5563 
5564 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5565 	kvm_page_track_write(vcpu, gpa, new, bytes);
5566 
5567 	return X86EMUL_CONTINUE;
5568 
5569 emul_write:
5570 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5571 
5572 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5573 }
5574 
5575 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5576 {
5577 	int r = 0, i;
5578 
5579 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5580 		if (vcpu->arch.pio.in)
5581 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5582 					    vcpu->arch.pio.size, pd);
5583 		else
5584 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5585 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5586 					     pd);
5587 		if (r)
5588 			break;
5589 		pd += vcpu->arch.pio.size;
5590 	}
5591 	return r;
5592 }
5593 
5594 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5595 			       unsigned short port, void *val,
5596 			       unsigned int count, bool in)
5597 {
5598 	vcpu->arch.pio.port = port;
5599 	vcpu->arch.pio.in = in;
5600 	vcpu->arch.pio.count  = count;
5601 	vcpu->arch.pio.size = size;
5602 
5603 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5604 		vcpu->arch.pio.count = 0;
5605 		return 1;
5606 	}
5607 
5608 	vcpu->run->exit_reason = KVM_EXIT_IO;
5609 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5610 	vcpu->run->io.size = size;
5611 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5612 	vcpu->run->io.count = count;
5613 	vcpu->run->io.port = port;
5614 
5615 	return 0;
5616 }
5617 
5618 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5619 				    int size, unsigned short port, void *val,
5620 				    unsigned int count)
5621 {
5622 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5623 	int ret;
5624 
5625 	if (vcpu->arch.pio.count)
5626 		goto data_avail;
5627 
5628 	memset(vcpu->arch.pio_data, 0, size * count);
5629 
5630 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5631 	if (ret) {
5632 data_avail:
5633 		memcpy(val, vcpu->arch.pio_data, size * count);
5634 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5635 		vcpu->arch.pio.count = 0;
5636 		return 1;
5637 	}
5638 
5639 	return 0;
5640 }
5641 
5642 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5643 				     int size, unsigned short port,
5644 				     const void *val, unsigned int count)
5645 {
5646 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5647 
5648 	memcpy(vcpu->arch.pio_data, val, size * count);
5649 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5650 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5651 }
5652 
5653 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5654 {
5655 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5656 }
5657 
5658 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5659 {
5660 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5661 }
5662 
5663 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5664 {
5665 	if (!need_emulate_wbinvd(vcpu))
5666 		return X86EMUL_CONTINUE;
5667 
5668 	if (kvm_x86_ops->has_wbinvd_exit()) {
5669 		int cpu = get_cpu();
5670 
5671 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5672 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5673 				wbinvd_ipi, NULL, 1);
5674 		put_cpu();
5675 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5676 	} else
5677 		wbinvd();
5678 	return X86EMUL_CONTINUE;
5679 }
5680 
5681 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5682 {
5683 	kvm_emulate_wbinvd_noskip(vcpu);
5684 	return kvm_skip_emulated_instruction(vcpu);
5685 }
5686 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5687 
5688 
5689 
5690 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5691 {
5692 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5693 }
5694 
5695 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5696 			   unsigned long *dest)
5697 {
5698 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5699 }
5700 
5701 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5702 			   unsigned long value)
5703 {
5704 
5705 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5706 }
5707 
5708 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5709 {
5710 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5711 }
5712 
5713 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5714 {
5715 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5716 	unsigned long value;
5717 
5718 	switch (cr) {
5719 	case 0:
5720 		value = kvm_read_cr0(vcpu);
5721 		break;
5722 	case 2:
5723 		value = vcpu->arch.cr2;
5724 		break;
5725 	case 3:
5726 		value = kvm_read_cr3(vcpu);
5727 		break;
5728 	case 4:
5729 		value = kvm_read_cr4(vcpu);
5730 		break;
5731 	case 8:
5732 		value = kvm_get_cr8(vcpu);
5733 		break;
5734 	default:
5735 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5736 		return 0;
5737 	}
5738 
5739 	return value;
5740 }
5741 
5742 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5743 {
5744 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5745 	int res = 0;
5746 
5747 	switch (cr) {
5748 	case 0:
5749 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5750 		break;
5751 	case 2:
5752 		vcpu->arch.cr2 = val;
5753 		break;
5754 	case 3:
5755 		res = kvm_set_cr3(vcpu, val);
5756 		break;
5757 	case 4:
5758 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5759 		break;
5760 	case 8:
5761 		res = kvm_set_cr8(vcpu, val);
5762 		break;
5763 	default:
5764 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5765 		res = -1;
5766 	}
5767 
5768 	return res;
5769 }
5770 
5771 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5772 {
5773 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5774 }
5775 
5776 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5777 {
5778 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5779 }
5780 
5781 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5782 {
5783 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5784 }
5785 
5786 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5787 {
5788 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5789 }
5790 
5791 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5792 {
5793 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5794 }
5795 
5796 static unsigned long emulator_get_cached_segment_base(
5797 	struct x86_emulate_ctxt *ctxt, int seg)
5798 {
5799 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5800 }
5801 
5802 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5803 				 struct desc_struct *desc, u32 *base3,
5804 				 int seg)
5805 {
5806 	struct kvm_segment var;
5807 
5808 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5809 	*selector = var.selector;
5810 
5811 	if (var.unusable) {
5812 		memset(desc, 0, sizeof(*desc));
5813 		if (base3)
5814 			*base3 = 0;
5815 		return false;
5816 	}
5817 
5818 	if (var.g)
5819 		var.limit >>= 12;
5820 	set_desc_limit(desc, var.limit);
5821 	set_desc_base(desc, (unsigned long)var.base);
5822 #ifdef CONFIG_X86_64
5823 	if (base3)
5824 		*base3 = var.base >> 32;
5825 #endif
5826 	desc->type = var.type;
5827 	desc->s = var.s;
5828 	desc->dpl = var.dpl;
5829 	desc->p = var.present;
5830 	desc->avl = var.avl;
5831 	desc->l = var.l;
5832 	desc->d = var.db;
5833 	desc->g = var.g;
5834 
5835 	return true;
5836 }
5837 
5838 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5839 				 struct desc_struct *desc, u32 base3,
5840 				 int seg)
5841 {
5842 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5843 	struct kvm_segment var;
5844 
5845 	var.selector = selector;
5846 	var.base = get_desc_base(desc);
5847 #ifdef CONFIG_X86_64
5848 	var.base |= ((u64)base3) << 32;
5849 #endif
5850 	var.limit = get_desc_limit(desc);
5851 	if (desc->g)
5852 		var.limit = (var.limit << 12) | 0xfff;
5853 	var.type = desc->type;
5854 	var.dpl = desc->dpl;
5855 	var.db = desc->d;
5856 	var.s = desc->s;
5857 	var.l = desc->l;
5858 	var.g = desc->g;
5859 	var.avl = desc->avl;
5860 	var.present = desc->p;
5861 	var.unusable = !var.present;
5862 	var.padding = 0;
5863 
5864 	kvm_set_segment(vcpu, &var, seg);
5865 	return;
5866 }
5867 
5868 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5869 			    u32 msr_index, u64 *pdata)
5870 {
5871 	struct msr_data msr;
5872 	int r;
5873 
5874 	msr.index = msr_index;
5875 	msr.host_initiated = false;
5876 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5877 	if (r)
5878 		return r;
5879 
5880 	*pdata = msr.data;
5881 	return 0;
5882 }
5883 
5884 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5885 			    u32 msr_index, u64 data)
5886 {
5887 	struct msr_data msr;
5888 
5889 	msr.data = data;
5890 	msr.index = msr_index;
5891 	msr.host_initiated = false;
5892 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5893 }
5894 
5895 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5896 {
5897 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5898 
5899 	return vcpu->arch.smbase;
5900 }
5901 
5902 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5903 {
5904 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5905 
5906 	vcpu->arch.smbase = smbase;
5907 }
5908 
5909 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5910 			      u32 pmc)
5911 {
5912 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5913 }
5914 
5915 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5916 			     u32 pmc, u64 *pdata)
5917 {
5918 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5919 }
5920 
5921 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5922 {
5923 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5924 }
5925 
5926 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5927 			      struct x86_instruction_info *info,
5928 			      enum x86_intercept_stage stage)
5929 {
5930 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5931 }
5932 
5933 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5934 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5935 {
5936 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5937 }
5938 
5939 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5940 {
5941 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5942 }
5943 
5944 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5945 {
5946 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5947 }
5948 
5949 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5950 {
5951 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5952 }
5953 
5954 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5955 {
5956 	return emul_to_vcpu(ctxt)->arch.hflags;
5957 }
5958 
5959 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5960 {
5961 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5962 }
5963 
5964 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5965 {
5966 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5967 }
5968 
5969 static const struct x86_emulate_ops emulate_ops = {
5970 	.read_gpr            = emulator_read_gpr,
5971 	.write_gpr           = emulator_write_gpr,
5972 	.read_std            = emulator_read_std,
5973 	.write_std           = emulator_write_std,
5974 	.read_phys           = kvm_read_guest_phys_system,
5975 	.fetch               = kvm_fetch_guest_virt,
5976 	.read_emulated       = emulator_read_emulated,
5977 	.write_emulated      = emulator_write_emulated,
5978 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5979 	.invlpg              = emulator_invlpg,
5980 	.pio_in_emulated     = emulator_pio_in_emulated,
5981 	.pio_out_emulated    = emulator_pio_out_emulated,
5982 	.get_segment         = emulator_get_segment,
5983 	.set_segment         = emulator_set_segment,
5984 	.get_cached_segment_base = emulator_get_cached_segment_base,
5985 	.get_gdt             = emulator_get_gdt,
5986 	.get_idt	     = emulator_get_idt,
5987 	.set_gdt             = emulator_set_gdt,
5988 	.set_idt	     = emulator_set_idt,
5989 	.get_cr              = emulator_get_cr,
5990 	.set_cr              = emulator_set_cr,
5991 	.cpl                 = emulator_get_cpl,
5992 	.get_dr              = emulator_get_dr,
5993 	.set_dr              = emulator_set_dr,
5994 	.get_smbase          = emulator_get_smbase,
5995 	.set_smbase          = emulator_set_smbase,
5996 	.set_msr             = emulator_set_msr,
5997 	.get_msr             = emulator_get_msr,
5998 	.check_pmc	     = emulator_check_pmc,
5999 	.read_pmc            = emulator_read_pmc,
6000 	.halt                = emulator_halt,
6001 	.wbinvd              = emulator_wbinvd,
6002 	.fix_hypercall       = emulator_fix_hypercall,
6003 	.intercept           = emulator_intercept,
6004 	.get_cpuid           = emulator_get_cpuid,
6005 	.set_nmi_mask        = emulator_set_nmi_mask,
6006 	.get_hflags          = emulator_get_hflags,
6007 	.set_hflags          = emulator_set_hflags,
6008 	.pre_leave_smm       = emulator_pre_leave_smm,
6009 };
6010 
6011 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6012 {
6013 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6014 	/*
6015 	 * an sti; sti; sequence only disable interrupts for the first
6016 	 * instruction. So, if the last instruction, be it emulated or
6017 	 * not, left the system with the INT_STI flag enabled, it
6018 	 * means that the last instruction is an sti. We should not
6019 	 * leave the flag on in this case. The same goes for mov ss
6020 	 */
6021 	if (int_shadow & mask)
6022 		mask = 0;
6023 	if (unlikely(int_shadow || mask)) {
6024 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6025 		if (!mask)
6026 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6027 	}
6028 }
6029 
6030 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6031 {
6032 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6033 	if (ctxt->exception.vector == PF_VECTOR)
6034 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6035 
6036 	if (ctxt->exception.error_code_valid)
6037 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6038 				      ctxt->exception.error_code);
6039 	else
6040 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6041 	return false;
6042 }
6043 
6044 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6045 {
6046 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6047 	int cs_db, cs_l;
6048 
6049 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6050 
6051 	ctxt->eflags = kvm_get_rflags(vcpu);
6052 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6053 
6054 	ctxt->eip = kvm_rip_read(vcpu);
6055 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6056 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6057 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6058 		     cs_db				? X86EMUL_MODE_PROT32 :
6059 							  X86EMUL_MODE_PROT16;
6060 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6061 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6062 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6063 
6064 	init_decode_cache(ctxt);
6065 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6066 }
6067 
6068 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6069 {
6070 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6071 	int ret;
6072 
6073 	init_emulate_ctxt(vcpu);
6074 
6075 	ctxt->op_bytes = 2;
6076 	ctxt->ad_bytes = 2;
6077 	ctxt->_eip = ctxt->eip + inc_eip;
6078 	ret = emulate_int_real(ctxt, irq);
6079 
6080 	if (ret != X86EMUL_CONTINUE)
6081 		return EMULATE_FAIL;
6082 
6083 	ctxt->eip = ctxt->_eip;
6084 	kvm_rip_write(vcpu, ctxt->eip);
6085 	kvm_set_rflags(vcpu, ctxt->eflags);
6086 
6087 	return EMULATE_DONE;
6088 }
6089 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6090 
6091 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6092 {
6093 	int r = EMULATE_DONE;
6094 
6095 	++vcpu->stat.insn_emulation_fail;
6096 	trace_kvm_emulate_insn_failed(vcpu);
6097 
6098 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6099 		return EMULATE_FAIL;
6100 
6101 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6102 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6103 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6104 		vcpu->run->internal.ndata = 0;
6105 		r = EMULATE_USER_EXIT;
6106 	}
6107 
6108 	kvm_queue_exception(vcpu, UD_VECTOR);
6109 
6110 	return r;
6111 }
6112 
6113 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6114 				  bool write_fault_to_shadow_pgtable,
6115 				  int emulation_type)
6116 {
6117 	gpa_t gpa = cr2;
6118 	kvm_pfn_t pfn;
6119 
6120 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6121 		return false;
6122 
6123 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6124 		return false;
6125 
6126 	if (!vcpu->arch.mmu->direct_map) {
6127 		/*
6128 		 * Write permission should be allowed since only
6129 		 * write access need to be emulated.
6130 		 */
6131 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6132 
6133 		/*
6134 		 * If the mapping is invalid in guest, let cpu retry
6135 		 * it to generate fault.
6136 		 */
6137 		if (gpa == UNMAPPED_GVA)
6138 			return true;
6139 	}
6140 
6141 	/*
6142 	 * Do not retry the unhandleable instruction if it faults on the
6143 	 * readonly host memory, otherwise it will goto a infinite loop:
6144 	 * retry instruction -> write #PF -> emulation fail -> retry
6145 	 * instruction -> ...
6146 	 */
6147 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6148 
6149 	/*
6150 	 * If the instruction failed on the error pfn, it can not be fixed,
6151 	 * report the error to userspace.
6152 	 */
6153 	if (is_error_noslot_pfn(pfn))
6154 		return false;
6155 
6156 	kvm_release_pfn_clean(pfn);
6157 
6158 	/* The instructions are well-emulated on direct mmu. */
6159 	if (vcpu->arch.mmu->direct_map) {
6160 		unsigned int indirect_shadow_pages;
6161 
6162 		spin_lock(&vcpu->kvm->mmu_lock);
6163 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6164 		spin_unlock(&vcpu->kvm->mmu_lock);
6165 
6166 		if (indirect_shadow_pages)
6167 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6168 
6169 		return true;
6170 	}
6171 
6172 	/*
6173 	 * if emulation was due to access to shadowed page table
6174 	 * and it failed try to unshadow page and re-enter the
6175 	 * guest to let CPU execute the instruction.
6176 	 */
6177 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6178 
6179 	/*
6180 	 * If the access faults on its page table, it can not
6181 	 * be fixed by unprotecting shadow page and it should
6182 	 * be reported to userspace.
6183 	 */
6184 	return !write_fault_to_shadow_pgtable;
6185 }
6186 
6187 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6188 			      unsigned long cr2,  int emulation_type)
6189 {
6190 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6191 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6192 
6193 	last_retry_eip = vcpu->arch.last_retry_eip;
6194 	last_retry_addr = vcpu->arch.last_retry_addr;
6195 
6196 	/*
6197 	 * If the emulation is caused by #PF and it is non-page_table
6198 	 * writing instruction, it means the VM-EXIT is caused by shadow
6199 	 * page protected, we can zap the shadow page and retry this
6200 	 * instruction directly.
6201 	 *
6202 	 * Note: if the guest uses a non-page-table modifying instruction
6203 	 * on the PDE that points to the instruction, then we will unmap
6204 	 * the instruction and go to an infinite loop. So, we cache the
6205 	 * last retried eip and the last fault address, if we meet the eip
6206 	 * and the address again, we can break out of the potential infinite
6207 	 * loop.
6208 	 */
6209 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6210 
6211 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6212 		return false;
6213 
6214 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6215 		return false;
6216 
6217 	if (x86_page_table_writing_insn(ctxt))
6218 		return false;
6219 
6220 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6221 		return false;
6222 
6223 	vcpu->arch.last_retry_eip = ctxt->eip;
6224 	vcpu->arch.last_retry_addr = cr2;
6225 
6226 	if (!vcpu->arch.mmu->direct_map)
6227 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6228 
6229 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6230 
6231 	return true;
6232 }
6233 
6234 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6235 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6236 
6237 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6238 {
6239 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6240 		/* This is a good place to trace that we are exiting SMM.  */
6241 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6242 
6243 		/* Process a latched INIT or SMI, if any.  */
6244 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6245 	}
6246 
6247 	kvm_mmu_reset_context(vcpu);
6248 }
6249 
6250 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6251 {
6252 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
6253 
6254 	vcpu->arch.hflags = emul_flags;
6255 
6256 	if (changed & HF_SMM_MASK)
6257 		kvm_smm_changed(vcpu);
6258 }
6259 
6260 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6261 				unsigned long *db)
6262 {
6263 	u32 dr6 = 0;
6264 	int i;
6265 	u32 enable, rwlen;
6266 
6267 	enable = dr7;
6268 	rwlen = dr7 >> 16;
6269 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6270 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6271 			dr6 |= (1 << i);
6272 	return dr6;
6273 }
6274 
6275 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6276 {
6277 	struct kvm_run *kvm_run = vcpu->run;
6278 
6279 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6280 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6281 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6282 		kvm_run->debug.arch.exception = DB_VECTOR;
6283 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6284 		*r = EMULATE_USER_EXIT;
6285 	} else {
6286 		kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6287 	}
6288 }
6289 
6290 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6291 {
6292 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6293 	int r = EMULATE_DONE;
6294 
6295 	kvm_x86_ops->skip_emulated_instruction(vcpu);
6296 
6297 	/*
6298 	 * rflags is the old, "raw" value of the flags.  The new value has
6299 	 * not been saved yet.
6300 	 *
6301 	 * This is correct even for TF set by the guest, because "the
6302 	 * processor will not generate this exception after the instruction
6303 	 * that sets the TF flag".
6304 	 */
6305 	if (unlikely(rflags & X86_EFLAGS_TF))
6306 		kvm_vcpu_do_singlestep(vcpu, &r);
6307 	return r == EMULATE_DONE;
6308 }
6309 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6310 
6311 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6312 {
6313 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6314 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6315 		struct kvm_run *kvm_run = vcpu->run;
6316 		unsigned long eip = kvm_get_linear_rip(vcpu);
6317 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6318 					   vcpu->arch.guest_debug_dr7,
6319 					   vcpu->arch.eff_db);
6320 
6321 		if (dr6 != 0) {
6322 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6323 			kvm_run->debug.arch.pc = eip;
6324 			kvm_run->debug.arch.exception = DB_VECTOR;
6325 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6326 			*r = EMULATE_USER_EXIT;
6327 			return true;
6328 		}
6329 	}
6330 
6331 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6332 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6333 		unsigned long eip = kvm_get_linear_rip(vcpu);
6334 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6335 					   vcpu->arch.dr7,
6336 					   vcpu->arch.db);
6337 
6338 		if (dr6 != 0) {
6339 			vcpu->arch.dr6 &= ~15;
6340 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6341 			kvm_queue_exception(vcpu, DB_VECTOR);
6342 			*r = EMULATE_DONE;
6343 			return true;
6344 		}
6345 	}
6346 
6347 	return false;
6348 }
6349 
6350 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6351 {
6352 	switch (ctxt->opcode_len) {
6353 	case 1:
6354 		switch (ctxt->b) {
6355 		case 0xe4:	/* IN */
6356 		case 0xe5:
6357 		case 0xec:
6358 		case 0xed:
6359 		case 0xe6:	/* OUT */
6360 		case 0xe7:
6361 		case 0xee:
6362 		case 0xef:
6363 		case 0x6c:	/* INS */
6364 		case 0x6d:
6365 		case 0x6e:	/* OUTS */
6366 		case 0x6f:
6367 			return true;
6368 		}
6369 		break;
6370 	case 2:
6371 		switch (ctxt->b) {
6372 		case 0x33:	/* RDPMC */
6373 			return true;
6374 		}
6375 		break;
6376 	}
6377 
6378 	return false;
6379 }
6380 
6381 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6382 			    unsigned long cr2,
6383 			    int emulation_type,
6384 			    void *insn,
6385 			    int insn_len)
6386 {
6387 	int r;
6388 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6389 	bool writeback = true;
6390 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6391 
6392 	vcpu->arch.l1tf_flush_l1d = true;
6393 
6394 	/*
6395 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6396 	 * never reused.
6397 	 */
6398 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6399 	kvm_clear_exception_queue(vcpu);
6400 
6401 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6402 		init_emulate_ctxt(vcpu);
6403 
6404 		/*
6405 		 * We will reenter on the same instruction since
6406 		 * we do not set complete_userspace_io.  This does not
6407 		 * handle watchpoints yet, those would be handled in
6408 		 * the emulate_ops.
6409 		 */
6410 		if (!(emulation_type & EMULTYPE_SKIP) &&
6411 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6412 			return r;
6413 
6414 		ctxt->interruptibility = 0;
6415 		ctxt->have_exception = false;
6416 		ctxt->exception.vector = -1;
6417 		ctxt->perm_ok = false;
6418 
6419 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6420 
6421 		r = x86_decode_insn(ctxt, insn, insn_len);
6422 
6423 		trace_kvm_emulate_insn_start(vcpu);
6424 		++vcpu->stat.insn_emulation;
6425 		if (r != EMULATION_OK)  {
6426 			if (emulation_type & EMULTYPE_TRAP_UD)
6427 				return EMULATE_FAIL;
6428 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6429 						emulation_type))
6430 				return EMULATE_DONE;
6431 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6432 				return EMULATE_DONE;
6433 			if (emulation_type & EMULTYPE_SKIP)
6434 				return EMULATE_FAIL;
6435 			return handle_emulation_failure(vcpu, emulation_type);
6436 		}
6437 	}
6438 
6439 	if ((emulation_type & EMULTYPE_VMWARE) &&
6440 	    !is_vmware_backdoor_opcode(ctxt))
6441 		return EMULATE_FAIL;
6442 
6443 	if (emulation_type & EMULTYPE_SKIP) {
6444 		kvm_rip_write(vcpu, ctxt->_eip);
6445 		if (ctxt->eflags & X86_EFLAGS_RF)
6446 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6447 		return EMULATE_DONE;
6448 	}
6449 
6450 	if (retry_instruction(ctxt, cr2, emulation_type))
6451 		return EMULATE_DONE;
6452 
6453 	/* this is needed for vmware backdoor interface to work since it
6454 	   changes registers values  during IO operation */
6455 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6456 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6457 		emulator_invalidate_register_cache(ctxt);
6458 	}
6459 
6460 restart:
6461 	/* Save the faulting GPA (cr2) in the address field */
6462 	ctxt->exception.address = cr2;
6463 
6464 	r = x86_emulate_insn(ctxt);
6465 
6466 	if (r == EMULATION_INTERCEPTED)
6467 		return EMULATE_DONE;
6468 
6469 	if (r == EMULATION_FAILED) {
6470 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6471 					emulation_type))
6472 			return EMULATE_DONE;
6473 
6474 		return handle_emulation_failure(vcpu, emulation_type);
6475 	}
6476 
6477 	if (ctxt->have_exception) {
6478 		r = EMULATE_DONE;
6479 		if (inject_emulated_exception(vcpu))
6480 			return r;
6481 	} else if (vcpu->arch.pio.count) {
6482 		if (!vcpu->arch.pio.in) {
6483 			/* FIXME: return into emulator if single-stepping.  */
6484 			vcpu->arch.pio.count = 0;
6485 		} else {
6486 			writeback = false;
6487 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6488 		}
6489 		r = EMULATE_USER_EXIT;
6490 	} else if (vcpu->mmio_needed) {
6491 		if (!vcpu->mmio_is_write)
6492 			writeback = false;
6493 		r = EMULATE_USER_EXIT;
6494 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6495 	} else if (r == EMULATION_RESTART)
6496 		goto restart;
6497 	else
6498 		r = EMULATE_DONE;
6499 
6500 	if (writeback) {
6501 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6502 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6503 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6504 		kvm_rip_write(vcpu, ctxt->eip);
6505 		if (r == EMULATE_DONE && ctxt->tf)
6506 			kvm_vcpu_do_singlestep(vcpu, &r);
6507 		if (!ctxt->have_exception ||
6508 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6509 			__kvm_set_rflags(vcpu, ctxt->eflags);
6510 
6511 		/*
6512 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6513 		 * do nothing, and it will be requested again as soon as
6514 		 * the shadow expires.  But we still need to check here,
6515 		 * because POPF has no interrupt shadow.
6516 		 */
6517 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6518 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6519 	} else
6520 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6521 
6522 	return r;
6523 }
6524 
6525 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6526 {
6527 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6528 }
6529 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6530 
6531 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6532 					void *insn, int insn_len)
6533 {
6534 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6535 }
6536 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6537 
6538 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6539 {
6540 	vcpu->arch.pio.count = 0;
6541 
6542 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6543 		return 1;
6544 
6545 	return kvm_skip_emulated_instruction(vcpu);
6546 }
6547 
6548 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6549 			    unsigned short port)
6550 {
6551 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6552 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6553 					    size, port, &val, 1);
6554 
6555 	if (!ret) {
6556 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6557 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6558 	}
6559 	return ret;
6560 }
6561 
6562 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6563 {
6564 	unsigned long val;
6565 
6566 	/* We should only ever be called with arch.pio.count equal to 1 */
6567 	BUG_ON(vcpu->arch.pio.count != 1);
6568 
6569 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6570 		vcpu->arch.pio.count = 0;
6571 		return 1;
6572 	}
6573 
6574 	/* For size less than 4 we merge, else we zero extend */
6575 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6576 					: 0;
6577 
6578 	/*
6579 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6580 	 * the copy and tracing
6581 	 */
6582 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6583 				 vcpu->arch.pio.port, &val, 1);
6584 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6585 
6586 	return kvm_skip_emulated_instruction(vcpu);
6587 }
6588 
6589 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6590 			   unsigned short port)
6591 {
6592 	unsigned long val;
6593 	int ret;
6594 
6595 	/* For size less than 4 we merge, else we zero extend */
6596 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6597 
6598 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6599 				       &val, 1);
6600 	if (ret) {
6601 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6602 		return ret;
6603 	}
6604 
6605 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6606 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6607 
6608 	return 0;
6609 }
6610 
6611 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6612 {
6613 	int ret;
6614 
6615 	if (in)
6616 		ret = kvm_fast_pio_in(vcpu, size, port);
6617 	else
6618 		ret = kvm_fast_pio_out(vcpu, size, port);
6619 	return ret && kvm_skip_emulated_instruction(vcpu);
6620 }
6621 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6622 
6623 static int kvmclock_cpu_down_prep(unsigned int cpu)
6624 {
6625 	__this_cpu_write(cpu_tsc_khz, 0);
6626 	return 0;
6627 }
6628 
6629 static void tsc_khz_changed(void *data)
6630 {
6631 	struct cpufreq_freqs *freq = data;
6632 	unsigned long khz = 0;
6633 
6634 	if (data)
6635 		khz = freq->new;
6636 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6637 		khz = cpufreq_quick_get(raw_smp_processor_id());
6638 	if (!khz)
6639 		khz = tsc_khz;
6640 	__this_cpu_write(cpu_tsc_khz, khz);
6641 }
6642 
6643 #ifdef CONFIG_X86_64
6644 static void kvm_hyperv_tsc_notifier(void)
6645 {
6646 	struct kvm *kvm;
6647 	struct kvm_vcpu *vcpu;
6648 	int cpu;
6649 
6650 	spin_lock(&kvm_lock);
6651 	list_for_each_entry(kvm, &vm_list, vm_list)
6652 		kvm_make_mclock_inprogress_request(kvm);
6653 
6654 	hyperv_stop_tsc_emulation();
6655 
6656 	/* TSC frequency always matches when on Hyper-V */
6657 	for_each_present_cpu(cpu)
6658 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6659 	kvm_max_guest_tsc_khz = tsc_khz;
6660 
6661 	list_for_each_entry(kvm, &vm_list, vm_list) {
6662 		struct kvm_arch *ka = &kvm->arch;
6663 
6664 		spin_lock(&ka->pvclock_gtod_sync_lock);
6665 
6666 		pvclock_update_vm_gtod_copy(kvm);
6667 
6668 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6669 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6670 
6671 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6672 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6673 
6674 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6675 	}
6676 	spin_unlock(&kvm_lock);
6677 }
6678 #endif
6679 
6680 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6681 				     void *data)
6682 {
6683 	struct cpufreq_freqs *freq = data;
6684 	struct kvm *kvm;
6685 	struct kvm_vcpu *vcpu;
6686 	int i, send_ipi = 0;
6687 
6688 	/*
6689 	 * We allow guests to temporarily run on slowing clocks,
6690 	 * provided we notify them after, or to run on accelerating
6691 	 * clocks, provided we notify them before.  Thus time never
6692 	 * goes backwards.
6693 	 *
6694 	 * However, we have a problem.  We can't atomically update
6695 	 * the frequency of a given CPU from this function; it is
6696 	 * merely a notifier, which can be called from any CPU.
6697 	 * Changing the TSC frequency at arbitrary points in time
6698 	 * requires a recomputation of local variables related to
6699 	 * the TSC for each VCPU.  We must flag these local variables
6700 	 * to be updated and be sure the update takes place with the
6701 	 * new frequency before any guests proceed.
6702 	 *
6703 	 * Unfortunately, the combination of hotplug CPU and frequency
6704 	 * change creates an intractable locking scenario; the order
6705 	 * of when these callouts happen is undefined with respect to
6706 	 * CPU hotplug, and they can race with each other.  As such,
6707 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6708 	 * undefined; you can actually have a CPU frequency change take
6709 	 * place in between the computation of X and the setting of the
6710 	 * variable.  To protect against this problem, all updates of
6711 	 * the per_cpu tsc_khz variable are done in an interrupt
6712 	 * protected IPI, and all callers wishing to update the value
6713 	 * must wait for a synchronous IPI to complete (which is trivial
6714 	 * if the caller is on the CPU already).  This establishes the
6715 	 * necessary total order on variable updates.
6716 	 *
6717 	 * Note that because a guest time update may take place
6718 	 * anytime after the setting of the VCPU's request bit, the
6719 	 * correct TSC value must be set before the request.  However,
6720 	 * to ensure the update actually makes it to any guest which
6721 	 * starts running in hardware virtualization between the set
6722 	 * and the acquisition of the spinlock, we must also ping the
6723 	 * CPU after setting the request bit.
6724 	 *
6725 	 */
6726 
6727 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6728 		return 0;
6729 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6730 		return 0;
6731 
6732 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6733 
6734 	spin_lock(&kvm_lock);
6735 	list_for_each_entry(kvm, &vm_list, vm_list) {
6736 		kvm_for_each_vcpu(i, vcpu, kvm) {
6737 			if (vcpu->cpu != freq->cpu)
6738 				continue;
6739 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6740 			if (vcpu->cpu != smp_processor_id())
6741 				send_ipi = 1;
6742 		}
6743 	}
6744 	spin_unlock(&kvm_lock);
6745 
6746 	if (freq->old < freq->new && send_ipi) {
6747 		/*
6748 		 * We upscale the frequency.  Must make the guest
6749 		 * doesn't see old kvmclock values while running with
6750 		 * the new frequency, otherwise we risk the guest sees
6751 		 * time go backwards.
6752 		 *
6753 		 * In case we update the frequency for another cpu
6754 		 * (which might be in guest context) send an interrupt
6755 		 * to kick the cpu out of guest context.  Next time
6756 		 * guest context is entered kvmclock will be updated,
6757 		 * so the guest will not see stale values.
6758 		 */
6759 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6760 	}
6761 	return 0;
6762 }
6763 
6764 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6765 	.notifier_call  = kvmclock_cpufreq_notifier
6766 };
6767 
6768 static int kvmclock_cpu_online(unsigned int cpu)
6769 {
6770 	tsc_khz_changed(NULL);
6771 	return 0;
6772 }
6773 
6774 static void kvm_timer_init(void)
6775 {
6776 	max_tsc_khz = tsc_khz;
6777 
6778 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6779 #ifdef CONFIG_CPU_FREQ
6780 		struct cpufreq_policy policy;
6781 		int cpu;
6782 
6783 		memset(&policy, 0, sizeof(policy));
6784 		cpu = get_cpu();
6785 		cpufreq_get_policy(&policy, cpu);
6786 		if (policy.cpuinfo.max_freq)
6787 			max_tsc_khz = policy.cpuinfo.max_freq;
6788 		put_cpu();
6789 #endif
6790 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6791 					  CPUFREQ_TRANSITION_NOTIFIER);
6792 	}
6793 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6794 
6795 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6796 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6797 }
6798 
6799 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6800 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6801 
6802 int kvm_is_in_guest(void)
6803 {
6804 	return __this_cpu_read(current_vcpu) != NULL;
6805 }
6806 
6807 static int kvm_is_user_mode(void)
6808 {
6809 	int user_mode = 3;
6810 
6811 	if (__this_cpu_read(current_vcpu))
6812 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6813 
6814 	return user_mode != 0;
6815 }
6816 
6817 static unsigned long kvm_get_guest_ip(void)
6818 {
6819 	unsigned long ip = 0;
6820 
6821 	if (__this_cpu_read(current_vcpu))
6822 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6823 
6824 	return ip;
6825 }
6826 
6827 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6828 	.is_in_guest		= kvm_is_in_guest,
6829 	.is_user_mode		= kvm_is_user_mode,
6830 	.get_guest_ip		= kvm_get_guest_ip,
6831 };
6832 
6833 static void kvm_set_mmio_spte_mask(void)
6834 {
6835 	u64 mask;
6836 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6837 
6838 	/*
6839 	 * Set the reserved bits and the present bit of an paging-structure
6840 	 * entry to generate page fault with PFER.RSV = 1.
6841 	 */
6842 
6843 	/*
6844 	 * Mask the uppermost physical address bit, which would be reserved as
6845 	 * long as the supported physical address width is less than 52.
6846 	 */
6847 	mask = 1ull << 51;
6848 
6849 	/* Set the present bit. */
6850 	mask |= 1ull;
6851 
6852 	/*
6853 	 * If reserved bit is not supported, clear the present bit to disable
6854 	 * mmio page fault.
6855 	 */
6856 	if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6857 		mask &= ~1ull;
6858 
6859 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6860 }
6861 
6862 #ifdef CONFIG_X86_64
6863 static void pvclock_gtod_update_fn(struct work_struct *work)
6864 {
6865 	struct kvm *kvm;
6866 
6867 	struct kvm_vcpu *vcpu;
6868 	int i;
6869 
6870 	spin_lock(&kvm_lock);
6871 	list_for_each_entry(kvm, &vm_list, vm_list)
6872 		kvm_for_each_vcpu(i, vcpu, kvm)
6873 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6874 	atomic_set(&kvm_guest_has_master_clock, 0);
6875 	spin_unlock(&kvm_lock);
6876 }
6877 
6878 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6879 
6880 /*
6881  * Notification about pvclock gtod data update.
6882  */
6883 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6884 			       void *priv)
6885 {
6886 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6887 	struct timekeeper *tk = priv;
6888 
6889 	update_pvclock_gtod(tk);
6890 
6891 	/* disable master clock if host does not trust, or does not
6892 	 * use, TSC based clocksource.
6893 	 */
6894 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6895 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6896 		queue_work(system_long_wq, &pvclock_gtod_work);
6897 
6898 	return 0;
6899 }
6900 
6901 static struct notifier_block pvclock_gtod_notifier = {
6902 	.notifier_call = pvclock_gtod_notify,
6903 };
6904 #endif
6905 
6906 int kvm_arch_init(void *opaque)
6907 {
6908 	int r;
6909 	struct kvm_x86_ops *ops = opaque;
6910 
6911 	if (kvm_x86_ops) {
6912 		printk(KERN_ERR "kvm: already loaded the other module\n");
6913 		r = -EEXIST;
6914 		goto out;
6915 	}
6916 
6917 	if (!ops->cpu_has_kvm_support()) {
6918 		printk(KERN_ERR "kvm: no hardware support\n");
6919 		r = -EOPNOTSUPP;
6920 		goto out;
6921 	}
6922 	if (ops->disabled_by_bios()) {
6923 		printk(KERN_ERR "kvm: disabled by bios\n");
6924 		r = -EOPNOTSUPP;
6925 		goto out;
6926 	}
6927 
6928 	/*
6929 	 * KVM explicitly assumes that the guest has an FPU and
6930 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
6931 	 * vCPU's FPU state as a fxregs_state struct.
6932 	 */
6933 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
6934 		printk(KERN_ERR "kvm: inadequate fpu\n");
6935 		r = -EOPNOTSUPP;
6936 		goto out;
6937 	}
6938 
6939 	r = -ENOMEM;
6940 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
6941 					  __alignof__(struct fpu), SLAB_ACCOUNT,
6942 					  NULL);
6943 	if (!x86_fpu_cache) {
6944 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
6945 		goto out;
6946 	}
6947 
6948 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6949 	if (!shared_msrs) {
6950 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6951 		goto out_free_x86_fpu_cache;
6952 	}
6953 
6954 	r = kvm_mmu_module_init();
6955 	if (r)
6956 		goto out_free_percpu;
6957 
6958 	kvm_set_mmio_spte_mask();
6959 
6960 	kvm_x86_ops = ops;
6961 
6962 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6963 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6964 			PT_PRESENT_MASK, 0, sme_me_mask);
6965 	kvm_timer_init();
6966 
6967 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6968 
6969 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6970 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6971 
6972 	kvm_lapic_init();
6973 #ifdef CONFIG_X86_64
6974 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6975 
6976 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6977 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6978 #endif
6979 
6980 	return 0;
6981 
6982 out_free_percpu:
6983 	free_percpu(shared_msrs);
6984 out_free_x86_fpu_cache:
6985 	kmem_cache_destroy(x86_fpu_cache);
6986 out:
6987 	return r;
6988 }
6989 
6990 void kvm_arch_exit(void)
6991 {
6992 #ifdef CONFIG_X86_64
6993 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6994 		clear_hv_tscchange_cb();
6995 #endif
6996 	kvm_lapic_exit();
6997 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6998 
6999 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7000 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7001 					    CPUFREQ_TRANSITION_NOTIFIER);
7002 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7003 #ifdef CONFIG_X86_64
7004 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7005 #endif
7006 	kvm_x86_ops = NULL;
7007 	kvm_mmu_module_exit();
7008 	free_percpu(shared_msrs);
7009 	kmem_cache_destroy(x86_fpu_cache);
7010 }
7011 
7012 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7013 {
7014 	++vcpu->stat.halt_exits;
7015 	if (lapic_in_kernel(vcpu)) {
7016 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7017 		return 1;
7018 	} else {
7019 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7020 		return 0;
7021 	}
7022 }
7023 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7024 
7025 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7026 {
7027 	int ret = kvm_skip_emulated_instruction(vcpu);
7028 	/*
7029 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7030 	 * KVM_EXIT_DEBUG here.
7031 	 */
7032 	return kvm_vcpu_halt(vcpu) && ret;
7033 }
7034 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7035 
7036 #ifdef CONFIG_X86_64
7037 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7038 			        unsigned long clock_type)
7039 {
7040 	struct kvm_clock_pairing clock_pairing;
7041 	struct timespec64 ts;
7042 	u64 cycle;
7043 	int ret;
7044 
7045 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7046 		return -KVM_EOPNOTSUPP;
7047 
7048 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7049 		return -KVM_EOPNOTSUPP;
7050 
7051 	clock_pairing.sec = ts.tv_sec;
7052 	clock_pairing.nsec = ts.tv_nsec;
7053 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7054 	clock_pairing.flags = 0;
7055 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7056 
7057 	ret = 0;
7058 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7059 			    sizeof(struct kvm_clock_pairing)))
7060 		ret = -KVM_EFAULT;
7061 
7062 	return ret;
7063 }
7064 #endif
7065 
7066 /*
7067  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7068  *
7069  * @apicid - apicid of vcpu to be kicked.
7070  */
7071 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7072 {
7073 	struct kvm_lapic_irq lapic_irq;
7074 
7075 	lapic_irq.shorthand = 0;
7076 	lapic_irq.dest_mode = 0;
7077 	lapic_irq.level = 0;
7078 	lapic_irq.dest_id = apicid;
7079 	lapic_irq.msi_redir_hint = false;
7080 
7081 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7082 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7083 }
7084 
7085 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7086 {
7087 	if (!lapic_in_kernel(vcpu)) {
7088 		WARN_ON_ONCE(vcpu->arch.apicv_active);
7089 		return;
7090 	}
7091 	if (!vcpu->arch.apicv_active)
7092 		return;
7093 
7094 	vcpu->arch.apicv_active = false;
7095 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7096 }
7097 
7098 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7099 {
7100 	unsigned long nr, a0, a1, a2, a3, ret;
7101 	int op_64_bit;
7102 
7103 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7104 		return kvm_hv_hypercall(vcpu);
7105 
7106 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
7107 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
7108 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
7109 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
7110 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
7111 
7112 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7113 
7114 	op_64_bit = is_64_bit_mode(vcpu);
7115 	if (!op_64_bit) {
7116 		nr &= 0xFFFFFFFF;
7117 		a0 &= 0xFFFFFFFF;
7118 		a1 &= 0xFFFFFFFF;
7119 		a2 &= 0xFFFFFFFF;
7120 		a3 &= 0xFFFFFFFF;
7121 	}
7122 
7123 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7124 		ret = -KVM_EPERM;
7125 		goto out;
7126 	}
7127 
7128 	switch (nr) {
7129 	case KVM_HC_VAPIC_POLL_IRQ:
7130 		ret = 0;
7131 		break;
7132 	case KVM_HC_KICK_CPU:
7133 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7134 		ret = 0;
7135 		break;
7136 #ifdef CONFIG_X86_64
7137 	case KVM_HC_CLOCK_PAIRING:
7138 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7139 		break;
7140 #endif
7141 	case KVM_HC_SEND_IPI:
7142 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7143 		break;
7144 	default:
7145 		ret = -KVM_ENOSYS;
7146 		break;
7147 	}
7148 out:
7149 	if (!op_64_bit)
7150 		ret = (u32)ret;
7151 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7152 
7153 	++vcpu->stat.hypercalls;
7154 	return kvm_skip_emulated_instruction(vcpu);
7155 }
7156 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7157 
7158 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7159 {
7160 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7161 	char instruction[3];
7162 	unsigned long rip = kvm_rip_read(vcpu);
7163 
7164 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7165 
7166 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7167 		&ctxt->exception);
7168 }
7169 
7170 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7171 {
7172 	return vcpu->run->request_interrupt_window &&
7173 		likely(!pic_in_kernel(vcpu->kvm));
7174 }
7175 
7176 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7177 {
7178 	struct kvm_run *kvm_run = vcpu->run;
7179 
7180 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7181 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7182 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7183 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7184 	kvm_run->ready_for_interrupt_injection =
7185 		pic_in_kernel(vcpu->kvm) ||
7186 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7187 }
7188 
7189 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7190 {
7191 	int max_irr, tpr;
7192 
7193 	if (!kvm_x86_ops->update_cr8_intercept)
7194 		return;
7195 
7196 	if (!lapic_in_kernel(vcpu))
7197 		return;
7198 
7199 	if (vcpu->arch.apicv_active)
7200 		return;
7201 
7202 	if (!vcpu->arch.apic->vapic_addr)
7203 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7204 	else
7205 		max_irr = -1;
7206 
7207 	if (max_irr != -1)
7208 		max_irr >>= 4;
7209 
7210 	tpr = kvm_lapic_get_cr8(vcpu);
7211 
7212 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7213 }
7214 
7215 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7216 {
7217 	int r;
7218 
7219 	/* try to reinject previous events if any */
7220 
7221 	if (vcpu->arch.exception.injected)
7222 		kvm_x86_ops->queue_exception(vcpu);
7223 	/*
7224 	 * Do not inject an NMI or interrupt if there is a pending
7225 	 * exception.  Exceptions and interrupts are recognized at
7226 	 * instruction boundaries, i.e. the start of an instruction.
7227 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7228 	 * NMIs and interrupts, i.e. traps are recognized before an
7229 	 * NMI/interrupt that's pending on the same instruction.
7230 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7231 	 * priority, but are only generated (pended) during instruction
7232 	 * execution, i.e. a pending fault-like exception means the
7233 	 * fault occurred on the *previous* instruction and must be
7234 	 * serviced prior to recognizing any new events in order to
7235 	 * fully complete the previous instruction.
7236 	 */
7237 	else if (!vcpu->arch.exception.pending) {
7238 		if (vcpu->arch.nmi_injected)
7239 			kvm_x86_ops->set_nmi(vcpu);
7240 		else if (vcpu->arch.interrupt.injected)
7241 			kvm_x86_ops->set_irq(vcpu);
7242 	}
7243 
7244 	/*
7245 	 * Call check_nested_events() even if we reinjected a previous event
7246 	 * in order for caller to determine if it should require immediate-exit
7247 	 * from L2 to L1 due to pending L1 events which require exit
7248 	 * from L2 to L1.
7249 	 */
7250 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7251 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7252 		if (r != 0)
7253 			return r;
7254 	}
7255 
7256 	/* try to inject new event if pending */
7257 	if (vcpu->arch.exception.pending) {
7258 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7259 					vcpu->arch.exception.has_error_code,
7260 					vcpu->arch.exception.error_code);
7261 
7262 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7263 		vcpu->arch.exception.pending = false;
7264 		vcpu->arch.exception.injected = true;
7265 
7266 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7267 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7268 					     X86_EFLAGS_RF);
7269 
7270 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7271 			/*
7272 			 * This code assumes that nSVM doesn't use
7273 			 * check_nested_events(). If it does, the
7274 			 * DR6/DR7 changes should happen before L1
7275 			 * gets a #VMEXIT for an intercepted #DB in
7276 			 * L2.  (Under VMX, on the other hand, the
7277 			 * DR6/DR7 changes should not happen in the
7278 			 * event of a VM-exit to L1 for an intercepted
7279 			 * #DB in L2.)
7280 			 */
7281 			kvm_deliver_exception_payload(vcpu);
7282 			if (vcpu->arch.dr7 & DR7_GD) {
7283 				vcpu->arch.dr7 &= ~DR7_GD;
7284 				kvm_update_dr7(vcpu);
7285 			}
7286 		}
7287 
7288 		kvm_x86_ops->queue_exception(vcpu);
7289 	}
7290 
7291 	/* Don't consider new event if we re-injected an event */
7292 	if (kvm_event_needs_reinjection(vcpu))
7293 		return 0;
7294 
7295 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7296 	    kvm_x86_ops->smi_allowed(vcpu)) {
7297 		vcpu->arch.smi_pending = false;
7298 		++vcpu->arch.smi_count;
7299 		enter_smm(vcpu);
7300 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7301 		--vcpu->arch.nmi_pending;
7302 		vcpu->arch.nmi_injected = true;
7303 		kvm_x86_ops->set_nmi(vcpu);
7304 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7305 		/*
7306 		 * Because interrupts can be injected asynchronously, we are
7307 		 * calling check_nested_events again here to avoid a race condition.
7308 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7309 		 * proposal and current concerns.  Perhaps we should be setting
7310 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7311 		 */
7312 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7313 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7314 			if (r != 0)
7315 				return r;
7316 		}
7317 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7318 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7319 					    false);
7320 			kvm_x86_ops->set_irq(vcpu);
7321 		}
7322 	}
7323 
7324 	return 0;
7325 }
7326 
7327 static void process_nmi(struct kvm_vcpu *vcpu)
7328 {
7329 	unsigned limit = 2;
7330 
7331 	/*
7332 	 * x86 is limited to one NMI running, and one NMI pending after it.
7333 	 * If an NMI is already in progress, limit further NMIs to just one.
7334 	 * Otherwise, allow two (and we'll inject the first one immediately).
7335 	 */
7336 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7337 		limit = 1;
7338 
7339 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7340 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7341 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7342 }
7343 
7344 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7345 {
7346 	u32 flags = 0;
7347 	flags |= seg->g       << 23;
7348 	flags |= seg->db      << 22;
7349 	flags |= seg->l       << 21;
7350 	flags |= seg->avl     << 20;
7351 	flags |= seg->present << 15;
7352 	flags |= seg->dpl     << 13;
7353 	flags |= seg->s       << 12;
7354 	flags |= seg->type    << 8;
7355 	return flags;
7356 }
7357 
7358 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7359 {
7360 	struct kvm_segment seg;
7361 	int offset;
7362 
7363 	kvm_get_segment(vcpu, &seg, n);
7364 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7365 
7366 	if (n < 3)
7367 		offset = 0x7f84 + n * 12;
7368 	else
7369 		offset = 0x7f2c + (n - 3) * 12;
7370 
7371 	put_smstate(u32, buf, offset + 8, seg.base);
7372 	put_smstate(u32, buf, offset + 4, seg.limit);
7373 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7374 }
7375 
7376 #ifdef CONFIG_X86_64
7377 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7378 {
7379 	struct kvm_segment seg;
7380 	int offset;
7381 	u16 flags;
7382 
7383 	kvm_get_segment(vcpu, &seg, n);
7384 	offset = 0x7e00 + n * 16;
7385 
7386 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7387 	put_smstate(u16, buf, offset, seg.selector);
7388 	put_smstate(u16, buf, offset + 2, flags);
7389 	put_smstate(u32, buf, offset + 4, seg.limit);
7390 	put_smstate(u64, buf, offset + 8, seg.base);
7391 }
7392 #endif
7393 
7394 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7395 {
7396 	struct desc_ptr dt;
7397 	struct kvm_segment seg;
7398 	unsigned long val;
7399 	int i;
7400 
7401 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7402 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7403 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7404 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7405 
7406 	for (i = 0; i < 8; i++)
7407 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7408 
7409 	kvm_get_dr(vcpu, 6, &val);
7410 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7411 	kvm_get_dr(vcpu, 7, &val);
7412 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7413 
7414 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7415 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7416 	put_smstate(u32, buf, 0x7f64, seg.base);
7417 	put_smstate(u32, buf, 0x7f60, seg.limit);
7418 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7419 
7420 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7421 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7422 	put_smstate(u32, buf, 0x7f80, seg.base);
7423 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7424 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7425 
7426 	kvm_x86_ops->get_gdt(vcpu, &dt);
7427 	put_smstate(u32, buf, 0x7f74, dt.address);
7428 	put_smstate(u32, buf, 0x7f70, dt.size);
7429 
7430 	kvm_x86_ops->get_idt(vcpu, &dt);
7431 	put_smstate(u32, buf, 0x7f58, dt.address);
7432 	put_smstate(u32, buf, 0x7f54, dt.size);
7433 
7434 	for (i = 0; i < 6; i++)
7435 		enter_smm_save_seg_32(vcpu, buf, i);
7436 
7437 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7438 
7439 	/* revision id */
7440 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7441 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7442 }
7443 
7444 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7445 {
7446 #ifdef CONFIG_X86_64
7447 	struct desc_ptr dt;
7448 	struct kvm_segment seg;
7449 	unsigned long val;
7450 	int i;
7451 
7452 	for (i = 0; i < 16; i++)
7453 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7454 
7455 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7456 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7457 
7458 	kvm_get_dr(vcpu, 6, &val);
7459 	put_smstate(u64, buf, 0x7f68, val);
7460 	kvm_get_dr(vcpu, 7, &val);
7461 	put_smstate(u64, buf, 0x7f60, val);
7462 
7463 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7464 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7465 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7466 
7467 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7468 
7469 	/* revision id */
7470 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7471 
7472 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7473 
7474 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7475 	put_smstate(u16, buf, 0x7e90, seg.selector);
7476 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7477 	put_smstate(u32, buf, 0x7e94, seg.limit);
7478 	put_smstate(u64, buf, 0x7e98, seg.base);
7479 
7480 	kvm_x86_ops->get_idt(vcpu, &dt);
7481 	put_smstate(u32, buf, 0x7e84, dt.size);
7482 	put_smstate(u64, buf, 0x7e88, dt.address);
7483 
7484 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7485 	put_smstate(u16, buf, 0x7e70, seg.selector);
7486 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7487 	put_smstate(u32, buf, 0x7e74, seg.limit);
7488 	put_smstate(u64, buf, 0x7e78, seg.base);
7489 
7490 	kvm_x86_ops->get_gdt(vcpu, &dt);
7491 	put_smstate(u32, buf, 0x7e64, dt.size);
7492 	put_smstate(u64, buf, 0x7e68, dt.address);
7493 
7494 	for (i = 0; i < 6; i++)
7495 		enter_smm_save_seg_64(vcpu, buf, i);
7496 #else
7497 	WARN_ON_ONCE(1);
7498 #endif
7499 }
7500 
7501 static void enter_smm(struct kvm_vcpu *vcpu)
7502 {
7503 	struct kvm_segment cs, ds;
7504 	struct desc_ptr dt;
7505 	char buf[512];
7506 	u32 cr0;
7507 
7508 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7509 	memset(buf, 0, 512);
7510 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7511 		enter_smm_save_state_64(vcpu, buf);
7512 	else
7513 		enter_smm_save_state_32(vcpu, buf);
7514 
7515 	/*
7516 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7517 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7518 	 * the SMM state-save area.
7519 	 */
7520 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7521 
7522 	vcpu->arch.hflags |= HF_SMM_MASK;
7523 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7524 
7525 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7526 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7527 	else
7528 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7529 
7530 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7531 	kvm_rip_write(vcpu, 0x8000);
7532 
7533 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7534 	kvm_x86_ops->set_cr0(vcpu, cr0);
7535 	vcpu->arch.cr0 = cr0;
7536 
7537 	kvm_x86_ops->set_cr4(vcpu, 0);
7538 
7539 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7540 	dt.address = dt.size = 0;
7541 	kvm_x86_ops->set_idt(vcpu, &dt);
7542 
7543 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7544 
7545 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7546 	cs.base = vcpu->arch.smbase;
7547 
7548 	ds.selector = 0;
7549 	ds.base = 0;
7550 
7551 	cs.limit    = ds.limit = 0xffffffff;
7552 	cs.type     = ds.type = 0x3;
7553 	cs.dpl      = ds.dpl = 0;
7554 	cs.db       = ds.db = 0;
7555 	cs.s        = ds.s = 1;
7556 	cs.l        = ds.l = 0;
7557 	cs.g        = ds.g = 1;
7558 	cs.avl      = ds.avl = 0;
7559 	cs.present  = ds.present = 1;
7560 	cs.unusable = ds.unusable = 0;
7561 	cs.padding  = ds.padding = 0;
7562 
7563 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7564 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7565 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7566 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7567 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7568 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7569 
7570 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7571 		kvm_x86_ops->set_efer(vcpu, 0);
7572 
7573 	kvm_update_cpuid(vcpu);
7574 	kvm_mmu_reset_context(vcpu);
7575 }
7576 
7577 static void process_smi(struct kvm_vcpu *vcpu)
7578 {
7579 	vcpu->arch.smi_pending = true;
7580 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7581 }
7582 
7583 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7584 {
7585 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7586 }
7587 
7588 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7589 {
7590 	if (!kvm_apic_present(vcpu))
7591 		return;
7592 
7593 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7594 
7595 	if (irqchip_split(vcpu->kvm))
7596 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7597 	else {
7598 		if (vcpu->arch.apicv_active)
7599 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7600 		if (ioapic_in_kernel(vcpu->kvm))
7601 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7602 	}
7603 
7604 	if (is_guest_mode(vcpu))
7605 		vcpu->arch.load_eoi_exitmap_pending = true;
7606 	else
7607 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7608 }
7609 
7610 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7611 {
7612 	u64 eoi_exit_bitmap[4];
7613 
7614 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7615 		return;
7616 
7617 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7618 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7619 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7620 }
7621 
7622 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7623 		unsigned long start, unsigned long end,
7624 		bool blockable)
7625 {
7626 	unsigned long apic_address;
7627 
7628 	/*
7629 	 * The physical address of apic access page is stored in the VMCS.
7630 	 * Update it when it becomes invalid.
7631 	 */
7632 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7633 	if (start <= apic_address && apic_address < end)
7634 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7635 
7636 	return 0;
7637 }
7638 
7639 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7640 {
7641 	struct page *page = NULL;
7642 
7643 	if (!lapic_in_kernel(vcpu))
7644 		return;
7645 
7646 	if (!kvm_x86_ops->set_apic_access_page_addr)
7647 		return;
7648 
7649 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7650 	if (is_error_page(page))
7651 		return;
7652 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7653 
7654 	/*
7655 	 * Do not pin apic access page in memory, the MMU notifier
7656 	 * will call us again if it is migrated or swapped out.
7657 	 */
7658 	put_page(page);
7659 }
7660 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7661 
7662 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7663 {
7664 	smp_send_reschedule(vcpu->cpu);
7665 }
7666 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7667 
7668 /*
7669  * Returns 1 to let vcpu_run() continue the guest execution loop without
7670  * exiting to the userspace.  Otherwise, the value will be returned to the
7671  * userspace.
7672  */
7673 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7674 {
7675 	int r;
7676 	bool req_int_win =
7677 		dm_request_for_irq_injection(vcpu) &&
7678 		kvm_cpu_accept_dm_intr(vcpu);
7679 
7680 	bool req_immediate_exit = false;
7681 
7682 	if (kvm_request_pending(vcpu)) {
7683 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7684 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7685 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7686 			kvm_mmu_unload(vcpu);
7687 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7688 			__kvm_migrate_timers(vcpu);
7689 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7690 			kvm_gen_update_masterclock(vcpu->kvm);
7691 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7692 			kvm_gen_kvmclock_update(vcpu);
7693 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7694 			r = kvm_guest_time_update(vcpu);
7695 			if (unlikely(r))
7696 				goto out;
7697 		}
7698 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7699 			kvm_mmu_sync_roots(vcpu);
7700 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7701 			kvm_mmu_load_cr3(vcpu);
7702 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7703 			kvm_vcpu_flush_tlb(vcpu, true);
7704 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7705 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7706 			r = 0;
7707 			goto out;
7708 		}
7709 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7710 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7711 			vcpu->mmio_needed = 0;
7712 			r = 0;
7713 			goto out;
7714 		}
7715 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7716 			/* Page is swapped out. Do synthetic halt */
7717 			vcpu->arch.apf.halted = true;
7718 			r = 1;
7719 			goto out;
7720 		}
7721 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7722 			record_steal_time(vcpu);
7723 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7724 			process_smi(vcpu);
7725 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7726 			process_nmi(vcpu);
7727 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7728 			kvm_pmu_handle_event(vcpu);
7729 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7730 			kvm_pmu_deliver_pmi(vcpu);
7731 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7732 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7733 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7734 				     vcpu->arch.ioapic_handled_vectors)) {
7735 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7736 				vcpu->run->eoi.vector =
7737 						vcpu->arch.pending_ioapic_eoi;
7738 				r = 0;
7739 				goto out;
7740 			}
7741 		}
7742 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7743 			vcpu_scan_ioapic(vcpu);
7744 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7745 			vcpu_load_eoi_exitmap(vcpu);
7746 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7747 			kvm_vcpu_reload_apic_access_page(vcpu);
7748 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7749 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7750 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7751 			r = 0;
7752 			goto out;
7753 		}
7754 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7755 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7756 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7757 			r = 0;
7758 			goto out;
7759 		}
7760 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7761 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7762 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7763 			r = 0;
7764 			goto out;
7765 		}
7766 
7767 		/*
7768 		 * KVM_REQ_HV_STIMER has to be processed after
7769 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7770 		 * depend on the guest clock being up-to-date
7771 		 */
7772 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7773 			kvm_hv_process_stimers(vcpu);
7774 	}
7775 
7776 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7777 		++vcpu->stat.req_event;
7778 		kvm_apic_accept_events(vcpu);
7779 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7780 			r = 1;
7781 			goto out;
7782 		}
7783 
7784 		if (inject_pending_event(vcpu, req_int_win) != 0)
7785 			req_immediate_exit = true;
7786 		else {
7787 			/* Enable SMI/NMI/IRQ window open exits if needed.
7788 			 *
7789 			 * SMIs have three cases:
7790 			 * 1) They can be nested, and then there is nothing to
7791 			 *    do here because RSM will cause a vmexit anyway.
7792 			 * 2) There is an ISA-specific reason why SMI cannot be
7793 			 *    injected, and the moment when this changes can be
7794 			 *    intercepted.
7795 			 * 3) Or the SMI can be pending because
7796 			 *    inject_pending_event has completed the injection
7797 			 *    of an IRQ or NMI from the previous vmexit, and
7798 			 *    then we request an immediate exit to inject the
7799 			 *    SMI.
7800 			 */
7801 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7802 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7803 					req_immediate_exit = true;
7804 			if (vcpu->arch.nmi_pending)
7805 				kvm_x86_ops->enable_nmi_window(vcpu);
7806 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7807 				kvm_x86_ops->enable_irq_window(vcpu);
7808 			WARN_ON(vcpu->arch.exception.pending);
7809 		}
7810 
7811 		if (kvm_lapic_enabled(vcpu)) {
7812 			update_cr8_intercept(vcpu);
7813 			kvm_lapic_sync_to_vapic(vcpu);
7814 		}
7815 	}
7816 
7817 	r = kvm_mmu_reload(vcpu);
7818 	if (unlikely(r)) {
7819 		goto cancel_injection;
7820 	}
7821 
7822 	preempt_disable();
7823 
7824 	kvm_x86_ops->prepare_guest_switch(vcpu);
7825 
7826 	/*
7827 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7828 	 * IPI are then delayed after guest entry, which ensures that they
7829 	 * result in virtual interrupt delivery.
7830 	 */
7831 	local_irq_disable();
7832 	vcpu->mode = IN_GUEST_MODE;
7833 
7834 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7835 
7836 	/*
7837 	 * 1) We should set ->mode before checking ->requests.  Please see
7838 	 * the comment in kvm_vcpu_exiting_guest_mode().
7839 	 *
7840 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
7841 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7842 	 * (see vmx_deliver_posted_interrupt).
7843 	 *
7844 	 * 3) This also orders the write to mode from any reads to the page
7845 	 * tables done while the VCPU is running.  Please see the comment
7846 	 * in kvm_flush_remote_tlbs.
7847 	 */
7848 	smp_mb__after_srcu_read_unlock();
7849 
7850 	/*
7851 	 * This handles the case where a posted interrupt was
7852 	 * notified with kvm_vcpu_kick.
7853 	 */
7854 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7855 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7856 
7857 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7858 	    || need_resched() || signal_pending(current)) {
7859 		vcpu->mode = OUTSIDE_GUEST_MODE;
7860 		smp_wmb();
7861 		local_irq_enable();
7862 		preempt_enable();
7863 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7864 		r = 1;
7865 		goto cancel_injection;
7866 	}
7867 
7868 	kvm_load_guest_xcr0(vcpu);
7869 
7870 	if (req_immediate_exit) {
7871 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7872 		kvm_x86_ops->request_immediate_exit(vcpu);
7873 	}
7874 
7875 	trace_kvm_entry(vcpu->vcpu_id);
7876 	if (lapic_timer_advance_ns)
7877 		wait_lapic_expire(vcpu);
7878 	guest_enter_irqoff();
7879 
7880 	if (unlikely(vcpu->arch.switch_db_regs)) {
7881 		set_debugreg(0, 7);
7882 		set_debugreg(vcpu->arch.eff_db[0], 0);
7883 		set_debugreg(vcpu->arch.eff_db[1], 1);
7884 		set_debugreg(vcpu->arch.eff_db[2], 2);
7885 		set_debugreg(vcpu->arch.eff_db[3], 3);
7886 		set_debugreg(vcpu->arch.dr6, 6);
7887 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7888 	}
7889 
7890 	kvm_x86_ops->run(vcpu);
7891 
7892 	/*
7893 	 * Do this here before restoring debug registers on the host.  And
7894 	 * since we do this before handling the vmexit, a DR access vmexit
7895 	 * can (a) read the correct value of the debug registers, (b) set
7896 	 * KVM_DEBUGREG_WONT_EXIT again.
7897 	 */
7898 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7899 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7900 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7901 		kvm_update_dr0123(vcpu);
7902 		kvm_update_dr6(vcpu);
7903 		kvm_update_dr7(vcpu);
7904 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7905 	}
7906 
7907 	/*
7908 	 * If the guest has used debug registers, at least dr7
7909 	 * will be disabled while returning to the host.
7910 	 * If we don't have active breakpoints in the host, we don't
7911 	 * care about the messed up debug address registers. But if
7912 	 * we have some of them active, restore the old state.
7913 	 */
7914 	if (hw_breakpoint_active())
7915 		hw_breakpoint_restore();
7916 
7917 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7918 
7919 	vcpu->mode = OUTSIDE_GUEST_MODE;
7920 	smp_wmb();
7921 
7922 	kvm_put_guest_xcr0(vcpu);
7923 
7924 	kvm_before_interrupt(vcpu);
7925 	kvm_x86_ops->handle_external_intr(vcpu);
7926 	kvm_after_interrupt(vcpu);
7927 
7928 	++vcpu->stat.exits;
7929 
7930 	guest_exit_irqoff();
7931 
7932 	local_irq_enable();
7933 	preempt_enable();
7934 
7935 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7936 
7937 	/*
7938 	 * Profile KVM exit RIPs:
7939 	 */
7940 	if (unlikely(prof_on == KVM_PROFILING)) {
7941 		unsigned long rip = kvm_rip_read(vcpu);
7942 		profile_hit(KVM_PROFILING, (void *)rip);
7943 	}
7944 
7945 	if (unlikely(vcpu->arch.tsc_always_catchup))
7946 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7947 
7948 	if (vcpu->arch.apic_attention)
7949 		kvm_lapic_sync_from_vapic(vcpu);
7950 
7951 	vcpu->arch.gpa_available = false;
7952 	r = kvm_x86_ops->handle_exit(vcpu);
7953 	return r;
7954 
7955 cancel_injection:
7956 	kvm_x86_ops->cancel_injection(vcpu);
7957 	if (unlikely(vcpu->arch.apic_attention))
7958 		kvm_lapic_sync_from_vapic(vcpu);
7959 out:
7960 	return r;
7961 }
7962 
7963 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7964 {
7965 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7966 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7967 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7968 		kvm_vcpu_block(vcpu);
7969 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7970 
7971 		if (kvm_x86_ops->post_block)
7972 			kvm_x86_ops->post_block(vcpu);
7973 
7974 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7975 			return 1;
7976 	}
7977 
7978 	kvm_apic_accept_events(vcpu);
7979 	switch(vcpu->arch.mp_state) {
7980 	case KVM_MP_STATE_HALTED:
7981 		vcpu->arch.pv.pv_unhalted = false;
7982 		vcpu->arch.mp_state =
7983 			KVM_MP_STATE_RUNNABLE;
7984 		/* fall through */
7985 	case KVM_MP_STATE_RUNNABLE:
7986 		vcpu->arch.apf.halted = false;
7987 		break;
7988 	case KVM_MP_STATE_INIT_RECEIVED:
7989 		break;
7990 	default:
7991 		return -EINTR;
7992 		break;
7993 	}
7994 	return 1;
7995 }
7996 
7997 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7998 {
7999 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8000 		kvm_x86_ops->check_nested_events(vcpu, false);
8001 
8002 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8003 		!vcpu->arch.apf.halted);
8004 }
8005 
8006 static int vcpu_run(struct kvm_vcpu *vcpu)
8007 {
8008 	int r;
8009 	struct kvm *kvm = vcpu->kvm;
8010 
8011 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8012 	vcpu->arch.l1tf_flush_l1d = true;
8013 
8014 	for (;;) {
8015 		if (kvm_vcpu_running(vcpu)) {
8016 			r = vcpu_enter_guest(vcpu);
8017 		} else {
8018 			r = vcpu_block(kvm, vcpu);
8019 		}
8020 
8021 		if (r <= 0)
8022 			break;
8023 
8024 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8025 		if (kvm_cpu_has_pending_timer(vcpu))
8026 			kvm_inject_pending_timer_irqs(vcpu);
8027 
8028 		if (dm_request_for_irq_injection(vcpu) &&
8029 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8030 			r = 0;
8031 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8032 			++vcpu->stat.request_irq_exits;
8033 			break;
8034 		}
8035 
8036 		kvm_check_async_pf_completion(vcpu);
8037 
8038 		if (signal_pending(current)) {
8039 			r = -EINTR;
8040 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8041 			++vcpu->stat.signal_exits;
8042 			break;
8043 		}
8044 		if (need_resched()) {
8045 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8046 			cond_resched();
8047 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8048 		}
8049 	}
8050 
8051 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8052 
8053 	return r;
8054 }
8055 
8056 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8057 {
8058 	int r;
8059 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8060 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8061 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8062 	if (r != EMULATE_DONE)
8063 		return 0;
8064 	return 1;
8065 }
8066 
8067 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8068 {
8069 	BUG_ON(!vcpu->arch.pio.count);
8070 
8071 	return complete_emulated_io(vcpu);
8072 }
8073 
8074 /*
8075  * Implements the following, as a state machine:
8076  *
8077  * read:
8078  *   for each fragment
8079  *     for each mmio piece in the fragment
8080  *       write gpa, len
8081  *       exit
8082  *       copy data
8083  *   execute insn
8084  *
8085  * write:
8086  *   for each fragment
8087  *     for each mmio piece in the fragment
8088  *       write gpa, len
8089  *       copy data
8090  *       exit
8091  */
8092 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8093 {
8094 	struct kvm_run *run = vcpu->run;
8095 	struct kvm_mmio_fragment *frag;
8096 	unsigned len;
8097 
8098 	BUG_ON(!vcpu->mmio_needed);
8099 
8100 	/* Complete previous fragment */
8101 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8102 	len = min(8u, frag->len);
8103 	if (!vcpu->mmio_is_write)
8104 		memcpy(frag->data, run->mmio.data, len);
8105 
8106 	if (frag->len <= 8) {
8107 		/* Switch to the next fragment. */
8108 		frag++;
8109 		vcpu->mmio_cur_fragment++;
8110 	} else {
8111 		/* Go forward to the next mmio piece. */
8112 		frag->data += len;
8113 		frag->gpa += len;
8114 		frag->len -= len;
8115 	}
8116 
8117 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8118 		vcpu->mmio_needed = 0;
8119 
8120 		/* FIXME: return into emulator if single-stepping.  */
8121 		if (vcpu->mmio_is_write)
8122 			return 1;
8123 		vcpu->mmio_read_completed = 1;
8124 		return complete_emulated_io(vcpu);
8125 	}
8126 
8127 	run->exit_reason = KVM_EXIT_MMIO;
8128 	run->mmio.phys_addr = frag->gpa;
8129 	if (vcpu->mmio_is_write)
8130 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8131 	run->mmio.len = min(8u, frag->len);
8132 	run->mmio.is_write = vcpu->mmio_is_write;
8133 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8134 	return 0;
8135 }
8136 
8137 /* Swap (qemu) user FPU context for the guest FPU context. */
8138 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8139 {
8140 	preempt_disable();
8141 	copy_fpregs_to_fpstate(&current->thread.fpu);
8142 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8143 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8144 				~XFEATURE_MASK_PKRU);
8145 	preempt_enable();
8146 	trace_kvm_fpu(1);
8147 }
8148 
8149 /* When vcpu_run ends, restore user space FPU context. */
8150 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8151 {
8152 	preempt_disable();
8153 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8154 	copy_kernel_to_fpregs(&current->thread.fpu.state);
8155 	preempt_enable();
8156 	++vcpu->stat.fpu_reload;
8157 	trace_kvm_fpu(0);
8158 }
8159 
8160 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8161 {
8162 	int r;
8163 
8164 	vcpu_load(vcpu);
8165 	kvm_sigset_activate(vcpu);
8166 	kvm_load_guest_fpu(vcpu);
8167 
8168 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8169 		if (kvm_run->immediate_exit) {
8170 			r = -EINTR;
8171 			goto out;
8172 		}
8173 		kvm_vcpu_block(vcpu);
8174 		kvm_apic_accept_events(vcpu);
8175 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8176 		r = -EAGAIN;
8177 		if (signal_pending(current)) {
8178 			r = -EINTR;
8179 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8180 			++vcpu->stat.signal_exits;
8181 		}
8182 		goto out;
8183 	}
8184 
8185 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8186 		r = -EINVAL;
8187 		goto out;
8188 	}
8189 
8190 	if (vcpu->run->kvm_dirty_regs) {
8191 		r = sync_regs(vcpu);
8192 		if (r != 0)
8193 			goto out;
8194 	}
8195 
8196 	/* re-sync apic's tpr */
8197 	if (!lapic_in_kernel(vcpu)) {
8198 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8199 			r = -EINVAL;
8200 			goto out;
8201 		}
8202 	}
8203 
8204 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8205 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8206 		vcpu->arch.complete_userspace_io = NULL;
8207 		r = cui(vcpu);
8208 		if (r <= 0)
8209 			goto out;
8210 	} else
8211 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8212 
8213 	if (kvm_run->immediate_exit)
8214 		r = -EINTR;
8215 	else
8216 		r = vcpu_run(vcpu);
8217 
8218 out:
8219 	kvm_put_guest_fpu(vcpu);
8220 	if (vcpu->run->kvm_valid_regs)
8221 		store_regs(vcpu);
8222 	post_kvm_run_save(vcpu);
8223 	kvm_sigset_deactivate(vcpu);
8224 
8225 	vcpu_put(vcpu);
8226 	return r;
8227 }
8228 
8229 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8230 {
8231 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8232 		/*
8233 		 * We are here if userspace calls get_regs() in the middle of
8234 		 * instruction emulation. Registers state needs to be copied
8235 		 * back from emulation context to vcpu. Userspace shouldn't do
8236 		 * that usually, but some bad designed PV devices (vmware
8237 		 * backdoor interface) need this to work
8238 		 */
8239 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8240 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8241 	}
8242 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8243 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8244 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8245 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8246 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8247 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8248 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8249 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8250 #ifdef CONFIG_X86_64
8251 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8252 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8253 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8254 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8255 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8256 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8257 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8258 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8259 #endif
8260 
8261 	regs->rip = kvm_rip_read(vcpu);
8262 	regs->rflags = kvm_get_rflags(vcpu);
8263 }
8264 
8265 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8266 {
8267 	vcpu_load(vcpu);
8268 	__get_regs(vcpu, regs);
8269 	vcpu_put(vcpu);
8270 	return 0;
8271 }
8272 
8273 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8274 {
8275 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8276 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8277 
8278 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8279 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8280 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8281 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8282 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8283 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8284 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8285 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8286 #ifdef CONFIG_X86_64
8287 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8288 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8289 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8290 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8291 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8292 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8293 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8294 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8295 #endif
8296 
8297 	kvm_rip_write(vcpu, regs->rip);
8298 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8299 
8300 	vcpu->arch.exception.pending = false;
8301 
8302 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8303 }
8304 
8305 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8306 {
8307 	vcpu_load(vcpu);
8308 	__set_regs(vcpu, regs);
8309 	vcpu_put(vcpu);
8310 	return 0;
8311 }
8312 
8313 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8314 {
8315 	struct kvm_segment cs;
8316 
8317 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8318 	*db = cs.db;
8319 	*l = cs.l;
8320 }
8321 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8322 
8323 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8324 {
8325 	struct desc_ptr dt;
8326 
8327 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8328 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8329 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8330 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8331 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8332 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8333 
8334 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8335 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8336 
8337 	kvm_x86_ops->get_idt(vcpu, &dt);
8338 	sregs->idt.limit = dt.size;
8339 	sregs->idt.base = dt.address;
8340 	kvm_x86_ops->get_gdt(vcpu, &dt);
8341 	sregs->gdt.limit = dt.size;
8342 	sregs->gdt.base = dt.address;
8343 
8344 	sregs->cr0 = kvm_read_cr0(vcpu);
8345 	sregs->cr2 = vcpu->arch.cr2;
8346 	sregs->cr3 = kvm_read_cr3(vcpu);
8347 	sregs->cr4 = kvm_read_cr4(vcpu);
8348 	sregs->cr8 = kvm_get_cr8(vcpu);
8349 	sregs->efer = vcpu->arch.efer;
8350 	sregs->apic_base = kvm_get_apic_base(vcpu);
8351 
8352 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8353 
8354 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8355 		set_bit(vcpu->arch.interrupt.nr,
8356 			(unsigned long *)sregs->interrupt_bitmap);
8357 }
8358 
8359 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8360 				  struct kvm_sregs *sregs)
8361 {
8362 	vcpu_load(vcpu);
8363 	__get_sregs(vcpu, sregs);
8364 	vcpu_put(vcpu);
8365 	return 0;
8366 }
8367 
8368 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8369 				    struct kvm_mp_state *mp_state)
8370 {
8371 	vcpu_load(vcpu);
8372 
8373 	kvm_apic_accept_events(vcpu);
8374 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8375 					vcpu->arch.pv.pv_unhalted)
8376 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8377 	else
8378 		mp_state->mp_state = vcpu->arch.mp_state;
8379 
8380 	vcpu_put(vcpu);
8381 	return 0;
8382 }
8383 
8384 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8385 				    struct kvm_mp_state *mp_state)
8386 {
8387 	int ret = -EINVAL;
8388 
8389 	vcpu_load(vcpu);
8390 
8391 	if (!lapic_in_kernel(vcpu) &&
8392 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8393 		goto out;
8394 
8395 	/* INITs are latched while in SMM */
8396 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8397 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8398 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8399 		goto out;
8400 
8401 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8402 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8403 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8404 	} else
8405 		vcpu->arch.mp_state = mp_state->mp_state;
8406 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8407 
8408 	ret = 0;
8409 out:
8410 	vcpu_put(vcpu);
8411 	return ret;
8412 }
8413 
8414 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8415 		    int reason, bool has_error_code, u32 error_code)
8416 {
8417 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8418 	int ret;
8419 
8420 	init_emulate_ctxt(vcpu);
8421 
8422 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8423 				   has_error_code, error_code);
8424 
8425 	if (ret)
8426 		return EMULATE_FAIL;
8427 
8428 	kvm_rip_write(vcpu, ctxt->eip);
8429 	kvm_set_rflags(vcpu, ctxt->eflags);
8430 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8431 	return EMULATE_DONE;
8432 }
8433 EXPORT_SYMBOL_GPL(kvm_task_switch);
8434 
8435 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8436 {
8437 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8438 			(sregs->cr4 & X86_CR4_OSXSAVE))
8439 		return  -EINVAL;
8440 
8441 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8442 		/*
8443 		 * When EFER.LME and CR0.PG are set, the processor is in
8444 		 * 64-bit mode (though maybe in a 32-bit code segment).
8445 		 * CR4.PAE and EFER.LMA must be set.
8446 		 */
8447 		if (!(sregs->cr4 & X86_CR4_PAE)
8448 		    || !(sregs->efer & EFER_LMA))
8449 			return -EINVAL;
8450 	} else {
8451 		/*
8452 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8453 		 * segment cannot be 64-bit.
8454 		 */
8455 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8456 			return -EINVAL;
8457 	}
8458 
8459 	return 0;
8460 }
8461 
8462 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8463 {
8464 	struct msr_data apic_base_msr;
8465 	int mmu_reset_needed = 0;
8466 	int cpuid_update_needed = 0;
8467 	int pending_vec, max_bits, idx;
8468 	struct desc_ptr dt;
8469 	int ret = -EINVAL;
8470 
8471 	if (kvm_valid_sregs(vcpu, sregs))
8472 		goto out;
8473 
8474 	apic_base_msr.data = sregs->apic_base;
8475 	apic_base_msr.host_initiated = true;
8476 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8477 		goto out;
8478 
8479 	dt.size = sregs->idt.limit;
8480 	dt.address = sregs->idt.base;
8481 	kvm_x86_ops->set_idt(vcpu, &dt);
8482 	dt.size = sregs->gdt.limit;
8483 	dt.address = sregs->gdt.base;
8484 	kvm_x86_ops->set_gdt(vcpu, &dt);
8485 
8486 	vcpu->arch.cr2 = sregs->cr2;
8487 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8488 	vcpu->arch.cr3 = sregs->cr3;
8489 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8490 
8491 	kvm_set_cr8(vcpu, sregs->cr8);
8492 
8493 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8494 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8495 
8496 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8497 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8498 	vcpu->arch.cr0 = sregs->cr0;
8499 
8500 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8501 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8502 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8503 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8504 	if (cpuid_update_needed)
8505 		kvm_update_cpuid(vcpu);
8506 
8507 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8508 	if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8509 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8510 		mmu_reset_needed = 1;
8511 	}
8512 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8513 
8514 	if (mmu_reset_needed)
8515 		kvm_mmu_reset_context(vcpu);
8516 
8517 	max_bits = KVM_NR_INTERRUPTS;
8518 	pending_vec = find_first_bit(
8519 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8520 	if (pending_vec < max_bits) {
8521 		kvm_queue_interrupt(vcpu, pending_vec, false);
8522 		pr_debug("Set back pending irq %d\n", pending_vec);
8523 	}
8524 
8525 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8526 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8527 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8528 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8529 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8530 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8531 
8532 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8533 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8534 
8535 	update_cr8_intercept(vcpu);
8536 
8537 	/* Older userspace won't unhalt the vcpu on reset. */
8538 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8539 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8540 	    !is_protmode(vcpu))
8541 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8542 
8543 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8544 
8545 	ret = 0;
8546 out:
8547 	return ret;
8548 }
8549 
8550 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8551 				  struct kvm_sregs *sregs)
8552 {
8553 	int ret;
8554 
8555 	vcpu_load(vcpu);
8556 	ret = __set_sregs(vcpu, sregs);
8557 	vcpu_put(vcpu);
8558 	return ret;
8559 }
8560 
8561 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8562 					struct kvm_guest_debug *dbg)
8563 {
8564 	unsigned long rflags;
8565 	int i, r;
8566 
8567 	vcpu_load(vcpu);
8568 
8569 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8570 		r = -EBUSY;
8571 		if (vcpu->arch.exception.pending)
8572 			goto out;
8573 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8574 			kvm_queue_exception(vcpu, DB_VECTOR);
8575 		else
8576 			kvm_queue_exception(vcpu, BP_VECTOR);
8577 	}
8578 
8579 	/*
8580 	 * Read rflags as long as potentially injected trace flags are still
8581 	 * filtered out.
8582 	 */
8583 	rflags = kvm_get_rflags(vcpu);
8584 
8585 	vcpu->guest_debug = dbg->control;
8586 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8587 		vcpu->guest_debug = 0;
8588 
8589 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8590 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8591 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8592 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8593 	} else {
8594 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8595 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8596 	}
8597 	kvm_update_dr7(vcpu);
8598 
8599 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8600 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8601 			get_segment_base(vcpu, VCPU_SREG_CS);
8602 
8603 	/*
8604 	 * Trigger an rflags update that will inject or remove the trace
8605 	 * flags.
8606 	 */
8607 	kvm_set_rflags(vcpu, rflags);
8608 
8609 	kvm_x86_ops->update_bp_intercept(vcpu);
8610 
8611 	r = 0;
8612 
8613 out:
8614 	vcpu_put(vcpu);
8615 	return r;
8616 }
8617 
8618 /*
8619  * Translate a guest virtual address to a guest physical address.
8620  */
8621 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8622 				    struct kvm_translation *tr)
8623 {
8624 	unsigned long vaddr = tr->linear_address;
8625 	gpa_t gpa;
8626 	int idx;
8627 
8628 	vcpu_load(vcpu);
8629 
8630 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8631 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8632 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8633 	tr->physical_address = gpa;
8634 	tr->valid = gpa != UNMAPPED_GVA;
8635 	tr->writeable = 1;
8636 	tr->usermode = 0;
8637 
8638 	vcpu_put(vcpu);
8639 	return 0;
8640 }
8641 
8642 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8643 {
8644 	struct fxregs_state *fxsave;
8645 
8646 	vcpu_load(vcpu);
8647 
8648 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8649 	memcpy(fpu->fpr, fxsave->st_space, 128);
8650 	fpu->fcw = fxsave->cwd;
8651 	fpu->fsw = fxsave->swd;
8652 	fpu->ftwx = fxsave->twd;
8653 	fpu->last_opcode = fxsave->fop;
8654 	fpu->last_ip = fxsave->rip;
8655 	fpu->last_dp = fxsave->rdp;
8656 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8657 
8658 	vcpu_put(vcpu);
8659 	return 0;
8660 }
8661 
8662 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8663 {
8664 	struct fxregs_state *fxsave;
8665 
8666 	vcpu_load(vcpu);
8667 
8668 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8669 
8670 	memcpy(fxsave->st_space, fpu->fpr, 128);
8671 	fxsave->cwd = fpu->fcw;
8672 	fxsave->swd = fpu->fsw;
8673 	fxsave->twd = fpu->ftwx;
8674 	fxsave->fop = fpu->last_opcode;
8675 	fxsave->rip = fpu->last_ip;
8676 	fxsave->rdp = fpu->last_dp;
8677 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8678 
8679 	vcpu_put(vcpu);
8680 	return 0;
8681 }
8682 
8683 static void store_regs(struct kvm_vcpu *vcpu)
8684 {
8685 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8686 
8687 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8688 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8689 
8690 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8691 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8692 
8693 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8694 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8695 				vcpu, &vcpu->run->s.regs.events);
8696 }
8697 
8698 static int sync_regs(struct kvm_vcpu *vcpu)
8699 {
8700 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8701 		return -EINVAL;
8702 
8703 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8704 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8705 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8706 	}
8707 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8708 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8709 			return -EINVAL;
8710 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8711 	}
8712 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8713 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8714 				vcpu, &vcpu->run->s.regs.events))
8715 			return -EINVAL;
8716 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8717 	}
8718 
8719 	return 0;
8720 }
8721 
8722 static void fx_init(struct kvm_vcpu *vcpu)
8723 {
8724 	fpstate_init(&vcpu->arch.guest_fpu->state);
8725 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8726 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8727 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8728 
8729 	/*
8730 	 * Ensure guest xcr0 is valid for loading
8731 	 */
8732 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8733 
8734 	vcpu->arch.cr0 |= X86_CR0_ET;
8735 }
8736 
8737 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8738 {
8739 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8740 
8741 	kvmclock_reset(vcpu);
8742 
8743 	kvm_x86_ops->vcpu_free(vcpu);
8744 	free_cpumask_var(wbinvd_dirty_mask);
8745 }
8746 
8747 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8748 						unsigned int id)
8749 {
8750 	struct kvm_vcpu *vcpu;
8751 
8752 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8753 		printk_once(KERN_WARNING
8754 		"kvm: SMP vm created on host with unstable TSC; "
8755 		"guest TSC will not be reliable\n");
8756 
8757 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8758 
8759 	return vcpu;
8760 }
8761 
8762 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8763 {
8764 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
8765 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8766 	kvm_vcpu_mtrr_init(vcpu);
8767 	vcpu_load(vcpu);
8768 	kvm_vcpu_reset(vcpu, false);
8769 	kvm_init_mmu(vcpu, false);
8770 	vcpu_put(vcpu);
8771 	return 0;
8772 }
8773 
8774 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8775 {
8776 	struct msr_data msr;
8777 	struct kvm *kvm = vcpu->kvm;
8778 
8779 	kvm_hv_vcpu_postcreate(vcpu);
8780 
8781 	if (mutex_lock_killable(&vcpu->mutex))
8782 		return;
8783 	vcpu_load(vcpu);
8784 	msr.data = 0x0;
8785 	msr.index = MSR_IA32_TSC;
8786 	msr.host_initiated = true;
8787 	kvm_write_tsc(vcpu, &msr);
8788 	vcpu_put(vcpu);
8789 	mutex_unlock(&vcpu->mutex);
8790 
8791 	if (!kvmclock_periodic_sync)
8792 		return;
8793 
8794 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8795 					KVMCLOCK_SYNC_PERIOD);
8796 }
8797 
8798 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8799 {
8800 	vcpu->arch.apf.msr_val = 0;
8801 
8802 	vcpu_load(vcpu);
8803 	kvm_mmu_unload(vcpu);
8804 	vcpu_put(vcpu);
8805 
8806 	kvm_x86_ops->vcpu_free(vcpu);
8807 }
8808 
8809 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8810 {
8811 	kvm_lapic_reset(vcpu, init_event);
8812 
8813 	vcpu->arch.hflags = 0;
8814 
8815 	vcpu->arch.smi_pending = 0;
8816 	vcpu->arch.smi_count = 0;
8817 	atomic_set(&vcpu->arch.nmi_queued, 0);
8818 	vcpu->arch.nmi_pending = 0;
8819 	vcpu->arch.nmi_injected = false;
8820 	kvm_clear_interrupt_queue(vcpu);
8821 	kvm_clear_exception_queue(vcpu);
8822 	vcpu->arch.exception.pending = false;
8823 
8824 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8825 	kvm_update_dr0123(vcpu);
8826 	vcpu->arch.dr6 = DR6_INIT;
8827 	kvm_update_dr6(vcpu);
8828 	vcpu->arch.dr7 = DR7_FIXED_1;
8829 	kvm_update_dr7(vcpu);
8830 
8831 	vcpu->arch.cr2 = 0;
8832 
8833 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8834 	vcpu->arch.apf.msr_val = 0;
8835 	vcpu->arch.st.msr_val = 0;
8836 
8837 	kvmclock_reset(vcpu);
8838 
8839 	kvm_clear_async_pf_completion_queue(vcpu);
8840 	kvm_async_pf_hash_reset(vcpu);
8841 	vcpu->arch.apf.halted = false;
8842 
8843 	if (kvm_mpx_supported()) {
8844 		void *mpx_state_buffer;
8845 
8846 		/*
8847 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8848 		 * called with loaded FPU and does not let userspace fix the state.
8849 		 */
8850 		if (init_event)
8851 			kvm_put_guest_fpu(vcpu);
8852 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8853 					XFEATURE_MASK_BNDREGS);
8854 		if (mpx_state_buffer)
8855 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8856 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8857 					XFEATURE_MASK_BNDCSR);
8858 		if (mpx_state_buffer)
8859 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8860 		if (init_event)
8861 			kvm_load_guest_fpu(vcpu);
8862 	}
8863 
8864 	if (!init_event) {
8865 		kvm_pmu_reset(vcpu);
8866 		vcpu->arch.smbase = 0x30000;
8867 
8868 		vcpu->arch.msr_misc_features_enables = 0;
8869 
8870 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8871 	}
8872 
8873 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8874 	vcpu->arch.regs_avail = ~0;
8875 	vcpu->arch.regs_dirty = ~0;
8876 
8877 	vcpu->arch.ia32_xss = 0;
8878 
8879 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8880 }
8881 
8882 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8883 {
8884 	struct kvm_segment cs;
8885 
8886 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8887 	cs.selector = vector << 8;
8888 	cs.base = vector << 12;
8889 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8890 	kvm_rip_write(vcpu, 0);
8891 }
8892 
8893 int kvm_arch_hardware_enable(void)
8894 {
8895 	struct kvm *kvm;
8896 	struct kvm_vcpu *vcpu;
8897 	int i;
8898 	int ret;
8899 	u64 local_tsc;
8900 	u64 max_tsc = 0;
8901 	bool stable, backwards_tsc = false;
8902 
8903 	kvm_shared_msr_cpu_online();
8904 	ret = kvm_x86_ops->hardware_enable();
8905 	if (ret != 0)
8906 		return ret;
8907 
8908 	local_tsc = rdtsc();
8909 	stable = !kvm_check_tsc_unstable();
8910 	list_for_each_entry(kvm, &vm_list, vm_list) {
8911 		kvm_for_each_vcpu(i, vcpu, kvm) {
8912 			if (!stable && vcpu->cpu == smp_processor_id())
8913 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8914 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8915 				backwards_tsc = true;
8916 				if (vcpu->arch.last_host_tsc > max_tsc)
8917 					max_tsc = vcpu->arch.last_host_tsc;
8918 			}
8919 		}
8920 	}
8921 
8922 	/*
8923 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8924 	 * platforms that reset TSC during suspend or hibernate actions, but
8925 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8926 	 * detect that condition here, which happens early in CPU bringup,
8927 	 * before any KVM threads can be running.  Unfortunately, we can't
8928 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8929 	 * enough into CPU bringup that we know how much real time has actually
8930 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8931 	 * variables that haven't been updated yet.
8932 	 *
8933 	 * So we simply find the maximum observed TSC above, then record the
8934 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8935 	 * the adjustment will be applied.  Note that we accumulate
8936 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8937 	 * gets a chance to run again.  In the event that no KVM threads get a
8938 	 * chance to run, we will miss the entire elapsed period, as we'll have
8939 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8940 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8941 	 * uniform across all VCPUs (not to mention the scenario is extremely
8942 	 * unlikely). It is possible that a second hibernate recovery happens
8943 	 * much faster than a first, causing the observed TSC here to be
8944 	 * smaller; this would require additional padding adjustment, which is
8945 	 * why we set last_host_tsc to the local tsc observed here.
8946 	 *
8947 	 * N.B. - this code below runs only on platforms with reliable TSC,
8948 	 * as that is the only way backwards_tsc is set above.  Also note
8949 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8950 	 * have the same delta_cyc adjustment applied if backwards_tsc
8951 	 * is detected.  Note further, this adjustment is only done once,
8952 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8953 	 * called multiple times (one for each physical CPU bringup).
8954 	 *
8955 	 * Platforms with unreliable TSCs don't have to deal with this, they
8956 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8957 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8958 	 * guarantee that they stay in perfect synchronization.
8959 	 */
8960 	if (backwards_tsc) {
8961 		u64 delta_cyc = max_tsc - local_tsc;
8962 		list_for_each_entry(kvm, &vm_list, vm_list) {
8963 			kvm->arch.backwards_tsc_observed = true;
8964 			kvm_for_each_vcpu(i, vcpu, kvm) {
8965 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8966 				vcpu->arch.last_host_tsc = local_tsc;
8967 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8968 			}
8969 
8970 			/*
8971 			 * We have to disable TSC offset matching.. if you were
8972 			 * booting a VM while issuing an S4 host suspend....
8973 			 * you may have some problem.  Solving this issue is
8974 			 * left as an exercise to the reader.
8975 			 */
8976 			kvm->arch.last_tsc_nsec = 0;
8977 			kvm->arch.last_tsc_write = 0;
8978 		}
8979 
8980 	}
8981 	return 0;
8982 }
8983 
8984 void kvm_arch_hardware_disable(void)
8985 {
8986 	kvm_x86_ops->hardware_disable();
8987 	drop_user_return_notifiers();
8988 }
8989 
8990 int kvm_arch_hardware_setup(void)
8991 {
8992 	int r;
8993 
8994 	r = kvm_x86_ops->hardware_setup();
8995 	if (r != 0)
8996 		return r;
8997 
8998 	if (kvm_has_tsc_control) {
8999 		/*
9000 		 * Make sure the user can only configure tsc_khz values that
9001 		 * fit into a signed integer.
9002 		 * A min value is not calculated because it will always
9003 		 * be 1 on all machines.
9004 		 */
9005 		u64 max = min(0x7fffffffULL,
9006 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9007 		kvm_max_guest_tsc_khz = max;
9008 
9009 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9010 	}
9011 
9012 	kvm_init_msr_list();
9013 	return 0;
9014 }
9015 
9016 void kvm_arch_hardware_unsetup(void)
9017 {
9018 	kvm_x86_ops->hardware_unsetup();
9019 }
9020 
9021 void kvm_arch_check_processor_compat(void *rtn)
9022 {
9023 	kvm_x86_ops->check_processor_compatibility(rtn);
9024 }
9025 
9026 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9027 {
9028 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9029 }
9030 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9031 
9032 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9033 {
9034 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9035 }
9036 
9037 struct static_key kvm_no_apic_vcpu __read_mostly;
9038 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9039 
9040 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9041 {
9042 	struct page *page;
9043 	int r;
9044 
9045 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9046 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9047 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9048 	else
9049 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9050 
9051 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9052 	if (!page) {
9053 		r = -ENOMEM;
9054 		goto fail;
9055 	}
9056 	vcpu->arch.pio_data = page_address(page);
9057 
9058 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9059 
9060 	r = kvm_mmu_create(vcpu);
9061 	if (r < 0)
9062 		goto fail_free_pio_data;
9063 
9064 	if (irqchip_in_kernel(vcpu->kvm)) {
9065 		vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9066 		r = kvm_create_lapic(vcpu);
9067 		if (r < 0)
9068 			goto fail_mmu_destroy;
9069 	} else
9070 		static_key_slow_inc(&kvm_no_apic_vcpu);
9071 
9072 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9073 				       GFP_KERNEL_ACCOUNT);
9074 	if (!vcpu->arch.mce_banks) {
9075 		r = -ENOMEM;
9076 		goto fail_free_lapic;
9077 	}
9078 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9079 
9080 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9081 				GFP_KERNEL_ACCOUNT)) {
9082 		r = -ENOMEM;
9083 		goto fail_free_mce_banks;
9084 	}
9085 
9086 	fx_init(vcpu);
9087 
9088 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9089 
9090 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9091 
9092 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9093 
9094 	kvm_async_pf_hash_reset(vcpu);
9095 	kvm_pmu_init(vcpu);
9096 
9097 	vcpu->arch.pending_external_vector = -1;
9098 	vcpu->arch.preempted_in_kernel = false;
9099 
9100 	kvm_hv_vcpu_init(vcpu);
9101 
9102 	return 0;
9103 
9104 fail_free_mce_banks:
9105 	kfree(vcpu->arch.mce_banks);
9106 fail_free_lapic:
9107 	kvm_free_lapic(vcpu);
9108 fail_mmu_destroy:
9109 	kvm_mmu_destroy(vcpu);
9110 fail_free_pio_data:
9111 	free_page((unsigned long)vcpu->arch.pio_data);
9112 fail:
9113 	return r;
9114 }
9115 
9116 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9117 {
9118 	int idx;
9119 
9120 	kvm_hv_vcpu_uninit(vcpu);
9121 	kvm_pmu_destroy(vcpu);
9122 	kfree(vcpu->arch.mce_banks);
9123 	kvm_free_lapic(vcpu);
9124 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9125 	kvm_mmu_destroy(vcpu);
9126 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9127 	free_page((unsigned long)vcpu->arch.pio_data);
9128 	if (!lapic_in_kernel(vcpu))
9129 		static_key_slow_dec(&kvm_no_apic_vcpu);
9130 }
9131 
9132 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9133 {
9134 	vcpu->arch.l1tf_flush_l1d = true;
9135 	kvm_x86_ops->sched_in(vcpu, cpu);
9136 }
9137 
9138 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9139 {
9140 	if (type)
9141 		return -EINVAL;
9142 
9143 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9144 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9145 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9146 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9147 
9148 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9149 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9150 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9151 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9152 		&kvm->arch.irq_sources_bitmap);
9153 
9154 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9155 	mutex_init(&kvm->arch.apic_map_lock);
9156 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9157 
9158 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9159 	pvclock_update_vm_gtod_copy(kvm);
9160 
9161 	kvm->arch.guest_can_read_msr_platform_info = true;
9162 
9163 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9164 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9165 
9166 	kvm_hv_init_vm(kvm);
9167 	kvm_page_track_init(kvm);
9168 	kvm_mmu_init_vm(kvm);
9169 
9170 	if (kvm_x86_ops->vm_init)
9171 		return kvm_x86_ops->vm_init(kvm);
9172 
9173 	return 0;
9174 }
9175 
9176 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9177 {
9178 	vcpu_load(vcpu);
9179 	kvm_mmu_unload(vcpu);
9180 	vcpu_put(vcpu);
9181 }
9182 
9183 static void kvm_free_vcpus(struct kvm *kvm)
9184 {
9185 	unsigned int i;
9186 	struct kvm_vcpu *vcpu;
9187 
9188 	/*
9189 	 * Unpin any mmu pages first.
9190 	 */
9191 	kvm_for_each_vcpu(i, vcpu, kvm) {
9192 		kvm_clear_async_pf_completion_queue(vcpu);
9193 		kvm_unload_vcpu_mmu(vcpu);
9194 	}
9195 	kvm_for_each_vcpu(i, vcpu, kvm)
9196 		kvm_arch_vcpu_free(vcpu);
9197 
9198 	mutex_lock(&kvm->lock);
9199 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9200 		kvm->vcpus[i] = NULL;
9201 
9202 	atomic_set(&kvm->online_vcpus, 0);
9203 	mutex_unlock(&kvm->lock);
9204 }
9205 
9206 void kvm_arch_sync_events(struct kvm *kvm)
9207 {
9208 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9209 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9210 	kvm_free_pit(kvm);
9211 }
9212 
9213 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9214 {
9215 	int i, r;
9216 	unsigned long hva;
9217 	struct kvm_memslots *slots = kvm_memslots(kvm);
9218 	struct kvm_memory_slot *slot, old;
9219 
9220 	/* Called with kvm->slots_lock held.  */
9221 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9222 		return -EINVAL;
9223 
9224 	slot = id_to_memslot(slots, id);
9225 	if (size) {
9226 		if (slot->npages)
9227 			return -EEXIST;
9228 
9229 		/*
9230 		 * MAP_SHARED to prevent internal slot pages from being moved
9231 		 * by fork()/COW.
9232 		 */
9233 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9234 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9235 		if (IS_ERR((void *)hva))
9236 			return PTR_ERR((void *)hva);
9237 	} else {
9238 		if (!slot->npages)
9239 			return 0;
9240 
9241 		hva = 0;
9242 	}
9243 
9244 	old = *slot;
9245 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9246 		struct kvm_userspace_memory_region m;
9247 
9248 		m.slot = id | (i << 16);
9249 		m.flags = 0;
9250 		m.guest_phys_addr = gpa;
9251 		m.userspace_addr = hva;
9252 		m.memory_size = size;
9253 		r = __kvm_set_memory_region(kvm, &m);
9254 		if (r < 0)
9255 			return r;
9256 	}
9257 
9258 	if (!size)
9259 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9260 
9261 	return 0;
9262 }
9263 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9264 
9265 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9266 {
9267 	int r;
9268 
9269 	mutex_lock(&kvm->slots_lock);
9270 	r = __x86_set_memory_region(kvm, id, gpa, size);
9271 	mutex_unlock(&kvm->slots_lock);
9272 
9273 	return r;
9274 }
9275 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9276 
9277 void kvm_arch_destroy_vm(struct kvm *kvm)
9278 {
9279 	if (current->mm == kvm->mm) {
9280 		/*
9281 		 * Free memory regions allocated on behalf of userspace,
9282 		 * unless the the memory map has changed due to process exit
9283 		 * or fd copying.
9284 		 */
9285 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9286 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9287 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9288 	}
9289 	if (kvm_x86_ops->vm_destroy)
9290 		kvm_x86_ops->vm_destroy(kvm);
9291 	kvm_pic_destroy(kvm);
9292 	kvm_ioapic_destroy(kvm);
9293 	kvm_free_vcpus(kvm);
9294 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9295 	kvm_mmu_uninit_vm(kvm);
9296 	kvm_page_track_cleanup(kvm);
9297 	kvm_hv_destroy_vm(kvm);
9298 }
9299 
9300 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9301 			   struct kvm_memory_slot *dont)
9302 {
9303 	int i;
9304 
9305 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9306 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9307 			kvfree(free->arch.rmap[i]);
9308 			free->arch.rmap[i] = NULL;
9309 		}
9310 		if (i == 0)
9311 			continue;
9312 
9313 		if (!dont || free->arch.lpage_info[i - 1] !=
9314 			     dont->arch.lpage_info[i - 1]) {
9315 			kvfree(free->arch.lpage_info[i - 1]);
9316 			free->arch.lpage_info[i - 1] = NULL;
9317 		}
9318 	}
9319 
9320 	kvm_page_track_free_memslot(free, dont);
9321 }
9322 
9323 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9324 			    unsigned long npages)
9325 {
9326 	int i;
9327 
9328 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9329 		struct kvm_lpage_info *linfo;
9330 		unsigned long ugfn;
9331 		int lpages;
9332 		int level = i + 1;
9333 
9334 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9335 				      slot->base_gfn, level) + 1;
9336 
9337 		slot->arch.rmap[i] =
9338 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9339 				 GFP_KERNEL_ACCOUNT);
9340 		if (!slot->arch.rmap[i])
9341 			goto out_free;
9342 		if (i == 0)
9343 			continue;
9344 
9345 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9346 		if (!linfo)
9347 			goto out_free;
9348 
9349 		slot->arch.lpage_info[i - 1] = linfo;
9350 
9351 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9352 			linfo[0].disallow_lpage = 1;
9353 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9354 			linfo[lpages - 1].disallow_lpage = 1;
9355 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9356 		/*
9357 		 * If the gfn and userspace address are not aligned wrt each
9358 		 * other, or if explicitly asked to, disable large page
9359 		 * support for this slot
9360 		 */
9361 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9362 		    !kvm_largepages_enabled()) {
9363 			unsigned long j;
9364 
9365 			for (j = 0; j < lpages; ++j)
9366 				linfo[j].disallow_lpage = 1;
9367 		}
9368 	}
9369 
9370 	if (kvm_page_track_create_memslot(slot, npages))
9371 		goto out_free;
9372 
9373 	return 0;
9374 
9375 out_free:
9376 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9377 		kvfree(slot->arch.rmap[i]);
9378 		slot->arch.rmap[i] = NULL;
9379 		if (i == 0)
9380 			continue;
9381 
9382 		kvfree(slot->arch.lpage_info[i - 1]);
9383 		slot->arch.lpage_info[i - 1] = NULL;
9384 	}
9385 	return -ENOMEM;
9386 }
9387 
9388 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9389 {
9390 	/*
9391 	 * memslots->generation has been incremented.
9392 	 * mmio generation may have reached its maximum value.
9393 	 */
9394 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9395 }
9396 
9397 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9398 				struct kvm_memory_slot *memslot,
9399 				const struct kvm_userspace_memory_region *mem,
9400 				enum kvm_mr_change change)
9401 {
9402 	return 0;
9403 }
9404 
9405 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9406 				     struct kvm_memory_slot *new)
9407 {
9408 	/* Still write protect RO slot */
9409 	if (new->flags & KVM_MEM_READONLY) {
9410 		kvm_mmu_slot_remove_write_access(kvm, new);
9411 		return;
9412 	}
9413 
9414 	/*
9415 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9416 	 *
9417 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9418 	 *
9419 	 *  - KVM_MR_CREATE with dirty logging is disabled
9420 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9421 	 *
9422 	 * The reason is, in case of PML, we need to set D-bit for any slots
9423 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9424 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9425 	 * guarantees leaving PML enabled during guest's lifetime won't have
9426 	 * any additional overhead from PML when guest is running with dirty
9427 	 * logging disabled for memory slots.
9428 	 *
9429 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9430 	 * to dirty logging mode.
9431 	 *
9432 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9433 	 *
9434 	 * In case of write protect:
9435 	 *
9436 	 * Write protect all pages for dirty logging.
9437 	 *
9438 	 * All the sptes including the large sptes which point to this
9439 	 * slot are set to readonly. We can not create any new large
9440 	 * spte on this slot until the end of the logging.
9441 	 *
9442 	 * See the comments in fast_page_fault().
9443 	 */
9444 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9445 		if (kvm_x86_ops->slot_enable_log_dirty)
9446 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9447 		else
9448 			kvm_mmu_slot_remove_write_access(kvm, new);
9449 	} else {
9450 		if (kvm_x86_ops->slot_disable_log_dirty)
9451 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9452 	}
9453 }
9454 
9455 void kvm_arch_commit_memory_region(struct kvm *kvm,
9456 				const struct kvm_userspace_memory_region *mem,
9457 				const struct kvm_memory_slot *old,
9458 				const struct kvm_memory_slot *new,
9459 				enum kvm_mr_change change)
9460 {
9461 	if (!kvm->arch.n_requested_mmu_pages)
9462 		kvm_mmu_change_mmu_pages(kvm,
9463 				kvm_mmu_calculate_default_mmu_pages(kvm));
9464 
9465 	/*
9466 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9467 	 * sptes have to be split.  If live migration is successful, the guest
9468 	 * in the source machine will be destroyed and large sptes will be
9469 	 * created in the destination. However, if the guest continues to run
9470 	 * in the source machine (for example if live migration fails), small
9471 	 * sptes will remain around and cause bad performance.
9472 	 *
9473 	 * Scan sptes if dirty logging has been stopped, dropping those
9474 	 * which can be collapsed into a single large-page spte.  Later
9475 	 * page faults will create the large-page sptes.
9476 	 */
9477 	if ((change != KVM_MR_DELETE) &&
9478 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9479 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9480 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9481 
9482 	/*
9483 	 * Set up write protection and/or dirty logging for the new slot.
9484 	 *
9485 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9486 	 * been zapped so no dirty logging staff is needed for old slot. For
9487 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9488 	 * new and it's also covered when dealing with the new slot.
9489 	 *
9490 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9491 	 */
9492 	if (change != KVM_MR_DELETE)
9493 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9494 }
9495 
9496 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9497 {
9498 	kvm_mmu_zap_all(kvm);
9499 }
9500 
9501 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9502 				   struct kvm_memory_slot *slot)
9503 {
9504 	kvm_page_track_flush_slot(kvm, slot);
9505 }
9506 
9507 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9508 {
9509 	return (is_guest_mode(vcpu) &&
9510 			kvm_x86_ops->guest_apic_has_interrupt &&
9511 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9512 }
9513 
9514 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9515 {
9516 	if (!list_empty_careful(&vcpu->async_pf.done))
9517 		return true;
9518 
9519 	if (kvm_apic_has_events(vcpu))
9520 		return true;
9521 
9522 	if (vcpu->arch.pv.pv_unhalted)
9523 		return true;
9524 
9525 	if (vcpu->arch.exception.pending)
9526 		return true;
9527 
9528 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9529 	    (vcpu->arch.nmi_pending &&
9530 	     kvm_x86_ops->nmi_allowed(vcpu)))
9531 		return true;
9532 
9533 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9534 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9535 		return true;
9536 
9537 	if (kvm_arch_interrupt_allowed(vcpu) &&
9538 	    (kvm_cpu_has_interrupt(vcpu) ||
9539 	    kvm_guest_apic_has_interrupt(vcpu)))
9540 		return true;
9541 
9542 	if (kvm_hv_has_stimer_pending(vcpu))
9543 		return true;
9544 
9545 	return false;
9546 }
9547 
9548 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9549 {
9550 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9551 }
9552 
9553 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9554 {
9555 	return vcpu->arch.preempted_in_kernel;
9556 }
9557 
9558 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9559 {
9560 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9561 }
9562 
9563 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9564 {
9565 	return kvm_x86_ops->interrupt_allowed(vcpu);
9566 }
9567 
9568 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9569 {
9570 	if (is_64_bit_mode(vcpu))
9571 		return kvm_rip_read(vcpu);
9572 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9573 		     kvm_rip_read(vcpu));
9574 }
9575 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9576 
9577 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9578 {
9579 	return kvm_get_linear_rip(vcpu) == linear_rip;
9580 }
9581 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9582 
9583 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9584 {
9585 	unsigned long rflags;
9586 
9587 	rflags = kvm_x86_ops->get_rflags(vcpu);
9588 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9589 		rflags &= ~X86_EFLAGS_TF;
9590 	return rflags;
9591 }
9592 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9593 
9594 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9595 {
9596 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9597 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9598 		rflags |= X86_EFLAGS_TF;
9599 	kvm_x86_ops->set_rflags(vcpu, rflags);
9600 }
9601 
9602 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9603 {
9604 	__kvm_set_rflags(vcpu, rflags);
9605 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9606 }
9607 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9608 
9609 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9610 {
9611 	int r;
9612 
9613 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9614 	      work->wakeup_all)
9615 		return;
9616 
9617 	r = kvm_mmu_reload(vcpu);
9618 	if (unlikely(r))
9619 		return;
9620 
9621 	if (!vcpu->arch.mmu->direct_map &&
9622 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9623 		return;
9624 
9625 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9626 }
9627 
9628 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9629 {
9630 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9631 }
9632 
9633 static inline u32 kvm_async_pf_next_probe(u32 key)
9634 {
9635 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9636 }
9637 
9638 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9639 {
9640 	u32 key = kvm_async_pf_hash_fn(gfn);
9641 
9642 	while (vcpu->arch.apf.gfns[key] != ~0)
9643 		key = kvm_async_pf_next_probe(key);
9644 
9645 	vcpu->arch.apf.gfns[key] = gfn;
9646 }
9647 
9648 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9649 {
9650 	int i;
9651 	u32 key = kvm_async_pf_hash_fn(gfn);
9652 
9653 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9654 		     (vcpu->arch.apf.gfns[key] != gfn &&
9655 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9656 		key = kvm_async_pf_next_probe(key);
9657 
9658 	return key;
9659 }
9660 
9661 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9662 {
9663 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9664 }
9665 
9666 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9667 {
9668 	u32 i, j, k;
9669 
9670 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9671 	while (true) {
9672 		vcpu->arch.apf.gfns[i] = ~0;
9673 		do {
9674 			j = kvm_async_pf_next_probe(j);
9675 			if (vcpu->arch.apf.gfns[j] == ~0)
9676 				return;
9677 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9678 			/*
9679 			 * k lies cyclically in ]i,j]
9680 			 * |    i.k.j |
9681 			 * |....j i.k.| or  |.k..j i...|
9682 			 */
9683 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9684 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9685 		i = j;
9686 	}
9687 }
9688 
9689 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9690 {
9691 
9692 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9693 				      sizeof(val));
9694 }
9695 
9696 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9697 {
9698 
9699 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9700 				      sizeof(u32));
9701 }
9702 
9703 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9704 				     struct kvm_async_pf *work)
9705 {
9706 	struct x86_exception fault;
9707 
9708 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9709 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9710 
9711 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9712 	    (vcpu->arch.apf.send_user_only &&
9713 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9714 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9715 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9716 		fault.vector = PF_VECTOR;
9717 		fault.error_code_valid = true;
9718 		fault.error_code = 0;
9719 		fault.nested_page_fault = false;
9720 		fault.address = work->arch.token;
9721 		fault.async_page_fault = true;
9722 		kvm_inject_page_fault(vcpu, &fault);
9723 	}
9724 }
9725 
9726 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9727 				 struct kvm_async_pf *work)
9728 {
9729 	struct x86_exception fault;
9730 	u32 val;
9731 
9732 	if (work->wakeup_all)
9733 		work->arch.token = ~0; /* broadcast wakeup */
9734 	else
9735 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9736 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9737 
9738 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9739 	    !apf_get_user(vcpu, &val)) {
9740 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9741 		    vcpu->arch.exception.pending &&
9742 		    vcpu->arch.exception.nr == PF_VECTOR &&
9743 		    !apf_put_user(vcpu, 0)) {
9744 			vcpu->arch.exception.injected = false;
9745 			vcpu->arch.exception.pending = false;
9746 			vcpu->arch.exception.nr = 0;
9747 			vcpu->arch.exception.has_error_code = false;
9748 			vcpu->arch.exception.error_code = 0;
9749 			vcpu->arch.exception.has_payload = false;
9750 			vcpu->arch.exception.payload = 0;
9751 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9752 			fault.vector = PF_VECTOR;
9753 			fault.error_code_valid = true;
9754 			fault.error_code = 0;
9755 			fault.nested_page_fault = false;
9756 			fault.address = work->arch.token;
9757 			fault.async_page_fault = true;
9758 			kvm_inject_page_fault(vcpu, &fault);
9759 		}
9760 	}
9761 	vcpu->arch.apf.halted = false;
9762 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9763 }
9764 
9765 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9766 {
9767 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9768 		return true;
9769 	else
9770 		return kvm_can_do_async_pf(vcpu);
9771 }
9772 
9773 void kvm_arch_start_assignment(struct kvm *kvm)
9774 {
9775 	atomic_inc(&kvm->arch.assigned_device_count);
9776 }
9777 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9778 
9779 void kvm_arch_end_assignment(struct kvm *kvm)
9780 {
9781 	atomic_dec(&kvm->arch.assigned_device_count);
9782 }
9783 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9784 
9785 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9786 {
9787 	return atomic_read(&kvm->arch.assigned_device_count);
9788 }
9789 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9790 
9791 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9792 {
9793 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9794 }
9795 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9796 
9797 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9798 {
9799 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9800 }
9801 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9802 
9803 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9804 {
9805 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9806 }
9807 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9808 
9809 bool kvm_arch_has_irq_bypass(void)
9810 {
9811 	return kvm_x86_ops->update_pi_irte != NULL;
9812 }
9813 
9814 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9815 				      struct irq_bypass_producer *prod)
9816 {
9817 	struct kvm_kernel_irqfd *irqfd =
9818 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9819 
9820 	irqfd->producer = prod;
9821 
9822 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9823 					   prod->irq, irqfd->gsi, 1);
9824 }
9825 
9826 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9827 				      struct irq_bypass_producer *prod)
9828 {
9829 	int ret;
9830 	struct kvm_kernel_irqfd *irqfd =
9831 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9832 
9833 	WARN_ON(irqfd->producer != prod);
9834 	irqfd->producer = NULL;
9835 
9836 	/*
9837 	 * When producer of consumer is unregistered, we change back to
9838 	 * remapped mode, so we can re-use the current implementation
9839 	 * when the irq is masked/disabled or the consumer side (KVM
9840 	 * int this case doesn't want to receive the interrupts.
9841 	*/
9842 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9843 	if (ret)
9844 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9845 		       " fails: %d\n", irqfd->consumer.token, ret);
9846 }
9847 
9848 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9849 				   uint32_t guest_irq, bool set)
9850 {
9851 	if (!kvm_x86_ops->update_pi_irte)
9852 		return -EINVAL;
9853 
9854 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9855 }
9856 
9857 bool kvm_vector_hashing_enabled(void)
9858 {
9859 	return vector_hashing;
9860 }
9861 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9862 
9863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9875 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9876 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9877 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9878 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9879 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9880 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9881 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9882