xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 5b4cb650)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
58 
59 #include <trace/events/kvm.h>
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
73 
74 #define CREATE_TRACE_POINTS
75 #include "trace.h"
76 
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 
82 #define emul_to_vcpu(ctxt) \
83 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
84 
85 /* EFER defaults:
86  * - enable syscall per default because its emulated by KVM
87  * - enable LME and LMA per default on 64 bit KVM
88  */
89 #ifdef CONFIG_X86_64
90 static
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 #else
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
94 #endif
95 
96 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
97 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
108 
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32  __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64  __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
138 
139 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
140 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
141 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
143 
144 static bool __read_mostly vector_hashing = true;
145 module_param(vector_hashing, bool, S_IRUGO);
146 
147 bool __read_mostly enable_vmware_backdoor = false;
148 module_param(enable_vmware_backdoor, bool, S_IRUGO);
149 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
150 
151 static bool __read_mostly force_emulation_prefix = false;
152 module_param(force_emulation_prefix, bool, S_IRUGO);
153 
154 #define KVM_NR_SHARED_MSRS 16
155 
156 struct kvm_shared_msrs_global {
157 	int nr;
158 	u32 msrs[KVM_NR_SHARED_MSRS];
159 };
160 
161 struct kvm_shared_msrs {
162 	struct user_return_notifier urn;
163 	bool registered;
164 	struct kvm_shared_msr_values {
165 		u64 host;
166 		u64 curr;
167 	} values[KVM_NR_SHARED_MSRS];
168 };
169 
170 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
171 static struct kvm_shared_msrs __percpu *shared_msrs;
172 
173 struct kvm_stats_debugfs_item debugfs_entries[] = {
174 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
175 	{ "pf_guest", VCPU_STAT(pf_guest) },
176 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
177 	{ "invlpg", VCPU_STAT(invlpg) },
178 	{ "exits", VCPU_STAT(exits) },
179 	{ "io_exits", VCPU_STAT(io_exits) },
180 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
181 	{ "signal_exits", VCPU_STAT(signal_exits) },
182 	{ "irq_window", VCPU_STAT(irq_window_exits) },
183 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
184 	{ "halt_exits", VCPU_STAT(halt_exits) },
185 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
186 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
187 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
188 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
189 	{ "hypercalls", VCPU_STAT(hypercalls) },
190 	{ "request_irq", VCPU_STAT(request_irq_exits) },
191 	{ "irq_exits", VCPU_STAT(irq_exits) },
192 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
193 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
194 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
195 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
196 	{ "irq_injections", VCPU_STAT(irq_injections) },
197 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
198 	{ "req_event", VCPU_STAT(req_event) },
199 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
200 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
201 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
202 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
203 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
204 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
205 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
206 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
207 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
208 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
209 	{ "largepages", VM_STAT(lpages) },
210 	{ "max_mmu_page_hash_collisions",
211 		VM_STAT(max_mmu_page_hash_collisions) },
212 	{ NULL }
213 };
214 
215 u64 __read_mostly host_xcr0;
216 
217 struct kmem_cache *x86_fpu_cache;
218 EXPORT_SYMBOL_GPL(x86_fpu_cache);
219 
220 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
221 
222 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
223 {
224 	int i;
225 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
226 		vcpu->arch.apf.gfns[i] = ~0;
227 }
228 
229 static void kvm_on_user_return(struct user_return_notifier *urn)
230 {
231 	unsigned slot;
232 	struct kvm_shared_msrs *locals
233 		= container_of(urn, struct kvm_shared_msrs, urn);
234 	struct kvm_shared_msr_values *values;
235 	unsigned long flags;
236 
237 	/*
238 	 * Disabling irqs at this point since the following code could be
239 	 * interrupted and executed through kvm_arch_hardware_disable()
240 	 */
241 	local_irq_save(flags);
242 	if (locals->registered) {
243 		locals->registered = false;
244 		user_return_notifier_unregister(urn);
245 	}
246 	local_irq_restore(flags);
247 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
248 		values = &locals->values[slot];
249 		if (values->host != values->curr) {
250 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
251 			values->curr = values->host;
252 		}
253 	}
254 }
255 
256 static void shared_msr_update(unsigned slot, u32 msr)
257 {
258 	u64 value;
259 	unsigned int cpu = smp_processor_id();
260 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
261 
262 	/* only read, and nobody should modify it at this time,
263 	 * so don't need lock */
264 	if (slot >= shared_msrs_global.nr) {
265 		printk(KERN_ERR "kvm: invalid MSR slot!");
266 		return;
267 	}
268 	rdmsrl_safe(msr, &value);
269 	smsr->values[slot].host = value;
270 	smsr->values[slot].curr = value;
271 }
272 
273 void kvm_define_shared_msr(unsigned slot, u32 msr)
274 {
275 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
276 	shared_msrs_global.msrs[slot] = msr;
277 	if (slot >= shared_msrs_global.nr)
278 		shared_msrs_global.nr = slot + 1;
279 }
280 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
281 
282 static void kvm_shared_msr_cpu_online(void)
283 {
284 	unsigned i;
285 
286 	for (i = 0; i < shared_msrs_global.nr; ++i)
287 		shared_msr_update(i, shared_msrs_global.msrs[i]);
288 }
289 
290 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
291 {
292 	unsigned int cpu = smp_processor_id();
293 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
294 	int err;
295 
296 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
297 		return 0;
298 	smsr->values[slot].curr = value;
299 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
300 	if (err)
301 		return 1;
302 
303 	if (!smsr->registered) {
304 		smsr->urn.on_user_return = kvm_on_user_return;
305 		user_return_notifier_register(&smsr->urn);
306 		smsr->registered = true;
307 	}
308 	return 0;
309 }
310 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
311 
312 static void drop_user_return_notifiers(void)
313 {
314 	unsigned int cpu = smp_processor_id();
315 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
316 
317 	if (smsr->registered)
318 		kvm_on_user_return(&smsr->urn);
319 }
320 
321 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
322 {
323 	return vcpu->arch.apic_base;
324 }
325 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
326 
327 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
328 {
329 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
330 }
331 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
332 
333 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
334 {
335 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
336 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
337 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
338 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
339 
340 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
341 		return 1;
342 	if (!msr_info->host_initiated) {
343 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
344 			return 1;
345 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
346 			return 1;
347 	}
348 
349 	kvm_lapic_set_base(vcpu, msr_info->data);
350 	return 0;
351 }
352 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
353 
354 asmlinkage __visible void kvm_spurious_fault(void)
355 {
356 	/* Fault while not rebooting.  We want the trace. */
357 	BUG();
358 }
359 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
360 
361 #define EXCPT_BENIGN		0
362 #define EXCPT_CONTRIBUTORY	1
363 #define EXCPT_PF		2
364 
365 static int exception_class(int vector)
366 {
367 	switch (vector) {
368 	case PF_VECTOR:
369 		return EXCPT_PF;
370 	case DE_VECTOR:
371 	case TS_VECTOR:
372 	case NP_VECTOR:
373 	case SS_VECTOR:
374 	case GP_VECTOR:
375 		return EXCPT_CONTRIBUTORY;
376 	default:
377 		break;
378 	}
379 	return EXCPT_BENIGN;
380 }
381 
382 #define EXCPT_FAULT		0
383 #define EXCPT_TRAP		1
384 #define EXCPT_ABORT		2
385 #define EXCPT_INTERRUPT		3
386 
387 static int exception_type(int vector)
388 {
389 	unsigned int mask;
390 
391 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
392 		return EXCPT_INTERRUPT;
393 
394 	mask = 1 << vector;
395 
396 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
397 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
398 		return EXCPT_TRAP;
399 
400 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
401 		return EXCPT_ABORT;
402 
403 	/* Reserved exceptions will result in fault */
404 	return EXCPT_FAULT;
405 }
406 
407 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
408 {
409 	unsigned nr = vcpu->arch.exception.nr;
410 	bool has_payload = vcpu->arch.exception.has_payload;
411 	unsigned long payload = vcpu->arch.exception.payload;
412 
413 	if (!has_payload)
414 		return;
415 
416 	switch (nr) {
417 	case DB_VECTOR:
418 		/*
419 		 * "Certain debug exceptions may clear bit 0-3.  The
420 		 * remaining contents of the DR6 register are never
421 		 * cleared by the processor".
422 		 */
423 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
424 		/*
425 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
426 		 */
427 		vcpu->arch.dr6 |= DR6_RTM;
428 		vcpu->arch.dr6 |= payload;
429 		/*
430 		 * Bit 16 should be set in the payload whenever the #DB
431 		 * exception should clear DR6.RTM. This makes the payload
432 		 * compatible with the pending debug exceptions under VMX.
433 		 * Though not currently documented in the SDM, this also
434 		 * makes the payload compatible with the exit qualification
435 		 * for #DB exceptions under VMX.
436 		 */
437 		vcpu->arch.dr6 ^= payload & DR6_RTM;
438 		break;
439 	case PF_VECTOR:
440 		vcpu->arch.cr2 = payload;
441 		break;
442 	}
443 
444 	vcpu->arch.exception.has_payload = false;
445 	vcpu->arch.exception.payload = 0;
446 }
447 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
448 
449 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
450 		unsigned nr, bool has_error, u32 error_code,
451 	        bool has_payload, unsigned long payload, bool reinject)
452 {
453 	u32 prev_nr;
454 	int class1, class2;
455 
456 	kvm_make_request(KVM_REQ_EVENT, vcpu);
457 
458 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
459 	queue:
460 		if (has_error && !is_protmode(vcpu))
461 			has_error = false;
462 		if (reinject) {
463 			/*
464 			 * On vmentry, vcpu->arch.exception.pending is only
465 			 * true if an event injection was blocked by
466 			 * nested_run_pending.  In that case, however,
467 			 * vcpu_enter_guest requests an immediate exit,
468 			 * and the guest shouldn't proceed far enough to
469 			 * need reinjection.
470 			 */
471 			WARN_ON_ONCE(vcpu->arch.exception.pending);
472 			vcpu->arch.exception.injected = true;
473 			if (WARN_ON_ONCE(has_payload)) {
474 				/*
475 				 * A reinjected event has already
476 				 * delivered its payload.
477 				 */
478 				has_payload = false;
479 				payload = 0;
480 			}
481 		} else {
482 			vcpu->arch.exception.pending = true;
483 			vcpu->arch.exception.injected = false;
484 		}
485 		vcpu->arch.exception.has_error_code = has_error;
486 		vcpu->arch.exception.nr = nr;
487 		vcpu->arch.exception.error_code = error_code;
488 		vcpu->arch.exception.has_payload = has_payload;
489 		vcpu->arch.exception.payload = payload;
490 		/*
491 		 * In guest mode, payload delivery should be deferred,
492 		 * so that the L1 hypervisor can intercept #PF before
493 		 * CR2 is modified (or intercept #DB before DR6 is
494 		 * modified under nVMX).  However, for ABI
495 		 * compatibility with KVM_GET_VCPU_EVENTS and
496 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
497 		 * delivery unless userspace has enabled this
498 		 * functionality via the per-VM capability,
499 		 * KVM_CAP_EXCEPTION_PAYLOAD.
500 		 */
501 		if (!vcpu->kvm->arch.exception_payload_enabled ||
502 		    !is_guest_mode(vcpu))
503 			kvm_deliver_exception_payload(vcpu);
504 		return;
505 	}
506 
507 	/* to check exception */
508 	prev_nr = vcpu->arch.exception.nr;
509 	if (prev_nr == DF_VECTOR) {
510 		/* triple fault -> shutdown */
511 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
512 		return;
513 	}
514 	class1 = exception_class(prev_nr);
515 	class2 = exception_class(nr);
516 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
517 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
518 		/*
519 		 * Generate double fault per SDM Table 5-5.  Set
520 		 * exception.pending = true so that the double fault
521 		 * can trigger a nested vmexit.
522 		 */
523 		vcpu->arch.exception.pending = true;
524 		vcpu->arch.exception.injected = false;
525 		vcpu->arch.exception.has_error_code = true;
526 		vcpu->arch.exception.nr = DF_VECTOR;
527 		vcpu->arch.exception.error_code = 0;
528 		vcpu->arch.exception.has_payload = false;
529 		vcpu->arch.exception.payload = 0;
530 	} else
531 		/* replace previous exception with a new one in a hope
532 		   that instruction re-execution will regenerate lost
533 		   exception */
534 		goto queue;
535 }
536 
537 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
538 {
539 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
540 }
541 EXPORT_SYMBOL_GPL(kvm_queue_exception);
542 
543 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
544 {
545 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
546 }
547 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
548 
549 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
550 				  unsigned long payload)
551 {
552 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
553 }
554 
555 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
556 				    u32 error_code, unsigned long payload)
557 {
558 	kvm_multiple_exception(vcpu, nr, true, error_code,
559 			       true, payload, false);
560 }
561 
562 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
563 {
564 	if (err)
565 		kvm_inject_gp(vcpu, 0);
566 	else
567 		return kvm_skip_emulated_instruction(vcpu);
568 
569 	return 1;
570 }
571 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
572 
573 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
574 {
575 	++vcpu->stat.pf_guest;
576 	vcpu->arch.exception.nested_apf =
577 		is_guest_mode(vcpu) && fault->async_page_fault;
578 	if (vcpu->arch.exception.nested_apf) {
579 		vcpu->arch.apf.nested_apf_token = fault->address;
580 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
581 	} else {
582 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
583 					fault->address);
584 	}
585 }
586 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
587 
588 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
589 {
590 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
591 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
592 	else
593 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
594 
595 	return fault->nested_page_fault;
596 }
597 
598 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
599 {
600 	atomic_inc(&vcpu->arch.nmi_queued);
601 	kvm_make_request(KVM_REQ_NMI, vcpu);
602 }
603 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
604 
605 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
606 {
607 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
608 }
609 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
610 
611 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
612 {
613 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
614 }
615 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
616 
617 /*
618  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
619  * a #GP and return false.
620  */
621 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
622 {
623 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
624 		return true;
625 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
626 	return false;
627 }
628 EXPORT_SYMBOL_GPL(kvm_require_cpl);
629 
630 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
631 {
632 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
633 		return true;
634 
635 	kvm_queue_exception(vcpu, UD_VECTOR);
636 	return false;
637 }
638 EXPORT_SYMBOL_GPL(kvm_require_dr);
639 
640 /*
641  * This function will be used to read from the physical memory of the currently
642  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
643  * can read from guest physical or from the guest's guest physical memory.
644  */
645 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
646 			    gfn_t ngfn, void *data, int offset, int len,
647 			    u32 access)
648 {
649 	struct x86_exception exception;
650 	gfn_t real_gfn;
651 	gpa_t ngpa;
652 
653 	ngpa     = gfn_to_gpa(ngfn);
654 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
655 	if (real_gfn == UNMAPPED_GVA)
656 		return -EFAULT;
657 
658 	real_gfn = gpa_to_gfn(real_gfn);
659 
660 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
661 }
662 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
663 
664 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
665 			       void *data, int offset, int len, u32 access)
666 {
667 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
668 				       data, offset, len, access);
669 }
670 
671 /*
672  * Load the pae pdptrs.  Return true is they are all valid.
673  */
674 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
675 {
676 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
677 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
678 	int i;
679 	int ret;
680 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
681 
682 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
683 				      offset * sizeof(u64), sizeof(pdpte),
684 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
685 	if (ret < 0) {
686 		ret = 0;
687 		goto out;
688 	}
689 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
690 		if ((pdpte[i] & PT_PRESENT_MASK) &&
691 		    (pdpte[i] &
692 		     vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
693 			ret = 0;
694 			goto out;
695 		}
696 	}
697 	ret = 1;
698 
699 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
700 	__set_bit(VCPU_EXREG_PDPTR,
701 		  (unsigned long *)&vcpu->arch.regs_avail);
702 	__set_bit(VCPU_EXREG_PDPTR,
703 		  (unsigned long *)&vcpu->arch.regs_dirty);
704 out:
705 
706 	return ret;
707 }
708 EXPORT_SYMBOL_GPL(load_pdptrs);
709 
710 bool pdptrs_changed(struct kvm_vcpu *vcpu)
711 {
712 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
713 	bool changed = true;
714 	int offset;
715 	gfn_t gfn;
716 	int r;
717 
718 	if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
719 		return false;
720 
721 	if (!test_bit(VCPU_EXREG_PDPTR,
722 		      (unsigned long *)&vcpu->arch.regs_avail))
723 		return true;
724 
725 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
726 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
727 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
728 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
729 	if (r < 0)
730 		goto out;
731 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
732 out:
733 
734 	return changed;
735 }
736 EXPORT_SYMBOL_GPL(pdptrs_changed);
737 
738 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
739 {
740 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
741 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
742 
743 	cr0 |= X86_CR0_ET;
744 
745 #ifdef CONFIG_X86_64
746 	if (cr0 & 0xffffffff00000000UL)
747 		return 1;
748 #endif
749 
750 	cr0 &= ~CR0_RESERVED_BITS;
751 
752 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
753 		return 1;
754 
755 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
756 		return 1;
757 
758 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
759 #ifdef CONFIG_X86_64
760 		if ((vcpu->arch.efer & EFER_LME)) {
761 			int cs_db, cs_l;
762 
763 			if (!is_pae(vcpu))
764 				return 1;
765 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
766 			if (cs_l)
767 				return 1;
768 		} else
769 #endif
770 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
771 						 kvm_read_cr3(vcpu)))
772 			return 1;
773 	}
774 
775 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
776 		return 1;
777 
778 	kvm_x86_ops->set_cr0(vcpu, cr0);
779 
780 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
781 		kvm_clear_async_pf_completion_queue(vcpu);
782 		kvm_async_pf_hash_reset(vcpu);
783 	}
784 
785 	if ((cr0 ^ old_cr0) & update_bits)
786 		kvm_mmu_reset_context(vcpu);
787 
788 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
789 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
790 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
791 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
792 
793 	return 0;
794 }
795 EXPORT_SYMBOL_GPL(kvm_set_cr0);
796 
797 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
798 {
799 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
800 }
801 EXPORT_SYMBOL_GPL(kvm_lmsw);
802 
803 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
804 {
805 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
806 			!vcpu->guest_xcr0_loaded) {
807 		/* kvm_set_xcr() also depends on this */
808 		if (vcpu->arch.xcr0 != host_xcr0)
809 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
810 		vcpu->guest_xcr0_loaded = 1;
811 	}
812 }
813 
814 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
815 {
816 	if (vcpu->guest_xcr0_loaded) {
817 		if (vcpu->arch.xcr0 != host_xcr0)
818 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
819 		vcpu->guest_xcr0_loaded = 0;
820 	}
821 }
822 
823 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
824 {
825 	u64 xcr0 = xcr;
826 	u64 old_xcr0 = vcpu->arch.xcr0;
827 	u64 valid_bits;
828 
829 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
830 	if (index != XCR_XFEATURE_ENABLED_MASK)
831 		return 1;
832 	if (!(xcr0 & XFEATURE_MASK_FP))
833 		return 1;
834 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
835 		return 1;
836 
837 	/*
838 	 * Do not allow the guest to set bits that we do not support
839 	 * saving.  However, xcr0 bit 0 is always set, even if the
840 	 * emulated CPU does not support XSAVE (see fx_init).
841 	 */
842 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
843 	if (xcr0 & ~valid_bits)
844 		return 1;
845 
846 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
847 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
848 		return 1;
849 
850 	if (xcr0 & XFEATURE_MASK_AVX512) {
851 		if (!(xcr0 & XFEATURE_MASK_YMM))
852 			return 1;
853 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
854 			return 1;
855 	}
856 	vcpu->arch.xcr0 = xcr0;
857 
858 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
859 		kvm_update_cpuid(vcpu);
860 	return 0;
861 }
862 
863 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
864 {
865 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
866 	    __kvm_set_xcr(vcpu, index, xcr)) {
867 		kvm_inject_gp(vcpu, 0);
868 		return 1;
869 	}
870 	return 0;
871 }
872 EXPORT_SYMBOL_GPL(kvm_set_xcr);
873 
874 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
875 {
876 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
877 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
878 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
879 
880 	if (cr4 & CR4_RESERVED_BITS)
881 		return 1;
882 
883 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
884 		return 1;
885 
886 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
887 		return 1;
888 
889 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
890 		return 1;
891 
892 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
893 		return 1;
894 
895 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
896 		return 1;
897 
898 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
899 		return 1;
900 
901 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
902 		return 1;
903 
904 	if (is_long_mode(vcpu)) {
905 		if (!(cr4 & X86_CR4_PAE))
906 			return 1;
907 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
908 		   && ((cr4 ^ old_cr4) & pdptr_bits)
909 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
910 				   kvm_read_cr3(vcpu)))
911 		return 1;
912 
913 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
914 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
915 			return 1;
916 
917 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
918 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
919 			return 1;
920 	}
921 
922 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
923 		return 1;
924 
925 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
926 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
927 		kvm_mmu_reset_context(vcpu);
928 
929 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
930 		kvm_update_cpuid(vcpu);
931 
932 	return 0;
933 }
934 EXPORT_SYMBOL_GPL(kvm_set_cr4);
935 
936 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
937 {
938 	bool skip_tlb_flush = false;
939 #ifdef CONFIG_X86_64
940 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
941 
942 	if (pcid_enabled) {
943 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
944 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
945 	}
946 #endif
947 
948 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
949 		if (!skip_tlb_flush) {
950 			kvm_mmu_sync_roots(vcpu);
951 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
952 		}
953 		return 0;
954 	}
955 
956 	if (is_long_mode(vcpu) &&
957 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
958 		return 1;
959 	else if (is_pae(vcpu) && is_paging(vcpu) &&
960 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
961 		return 1;
962 
963 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
964 	vcpu->arch.cr3 = cr3;
965 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
966 
967 	return 0;
968 }
969 EXPORT_SYMBOL_GPL(kvm_set_cr3);
970 
971 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
972 {
973 	if (cr8 & CR8_RESERVED_BITS)
974 		return 1;
975 	if (lapic_in_kernel(vcpu))
976 		kvm_lapic_set_tpr(vcpu, cr8);
977 	else
978 		vcpu->arch.cr8 = cr8;
979 	return 0;
980 }
981 EXPORT_SYMBOL_GPL(kvm_set_cr8);
982 
983 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
984 {
985 	if (lapic_in_kernel(vcpu))
986 		return kvm_lapic_get_cr8(vcpu);
987 	else
988 		return vcpu->arch.cr8;
989 }
990 EXPORT_SYMBOL_GPL(kvm_get_cr8);
991 
992 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
993 {
994 	int i;
995 
996 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
997 		for (i = 0; i < KVM_NR_DB_REGS; i++)
998 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
999 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1000 	}
1001 }
1002 
1003 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1004 {
1005 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1006 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1007 }
1008 
1009 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1010 {
1011 	unsigned long dr7;
1012 
1013 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1014 		dr7 = vcpu->arch.guest_debug_dr7;
1015 	else
1016 		dr7 = vcpu->arch.dr7;
1017 	kvm_x86_ops->set_dr7(vcpu, dr7);
1018 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1019 	if (dr7 & DR7_BP_EN_MASK)
1020 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1021 }
1022 
1023 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1024 {
1025 	u64 fixed = DR6_FIXED_1;
1026 
1027 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1028 		fixed |= DR6_RTM;
1029 	return fixed;
1030 }
1031 
1032 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1033 {
1034 	switch (dr) {
1035 	case 0 ... 3:
1036 		vcpu->arch.db[dr] = val;
1037 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1038 			vcpu->arch.eff_db[dr] = val;
1039 		break;
1040 	case 4:
1041 		/* fall through */
1042 	case 6:
1043 		if (val & 0xffffffff00000000ULL)
1044 			return -1; /* #GP */
1045 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1046 		kvm_update_dr6(vcpu);
1047 		break;
1048 	case 5:
1049 		/* fall through */
1050 	default: /* 7 */
1051 		if (val & 0xffffffff00000000ULL)
1052 			return -1; /* #GP */
1053 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1054 		kvm_update_dr7(vcpu);
1055 		break;
1056 	}
1057 
1058 	return 0;
1059 }
1060 
1061 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1062 {
1063 	if (__kvm_set_dr(vcpu, dr, val)) {
1064 		kvm_inject_gp(vcpu, 0);
1065 		return 1;
1066 	}
1067 	return 0;
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_dr);
1070 
1071 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1072 {
1073 	switch (dr) {
1074 	case 0 ... 3:
1075 		*val = vcpu->arch.db[dr];
1076 		break;
1077 	case 4:
1078 		/* fall through */
1079 	case 6:
1080 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1081 			*val = vcpu->arch.dr6;
1082 		else
1083 			*val = kvm_x86_ops->get_dr6(vcpu);
1084 		break;
1085 	case 5:
1086 		/* fall through */
1087 	default: /* 7 */
1088 		*val = vcpu->arch.dr7;
1089 		break;
1090 	}
1091 	return 0;
1092 }
1093 EXPORT_SYMBOL_GPL(kvm_get_dr);
1094 
1095 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1096 {
1097 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1098 	u64 data;
1099 	int err;
1100 
1101 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1102 	if (err)
1103 		return err;
1104 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1105 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1106 	return err;
1107 }
1108 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1109 
1110 /*
1111  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1112  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1113  *
1114  * This list is modified at module load time to reflect the
1115  * capabilities of the host cpu. This capabilities test skips MSRs that are
1116  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1117  * may depend on host virtualization features rather than host cpu features.
1118  */
1119 
1120 static u32 msrs_to_save[] = {
1121 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1122 	MSR_STAR,
1123 #ifdef CONFIG_X86_64
1124 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1125 #endif
1126 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1127 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1128 	MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES,
1129 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1130 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1131 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1132 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1133 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1134 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1135 };
1136 
1137 static unsigned num_msrs_to_save;
1138 
1139 static u32 emulated_msrs[] = {
1140 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1141 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1142 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1143 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1144 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1145 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1146 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1147 	HV_X64_MSR_RESET,
1148 	HV_X64_MSR_VP_INDEX,
1149 	HV_X64_MSR_VP_RUNTIME,
1150 	HV_X64_MSR_SCONTROL,
1151 	HV_X64_MSR_STIMER0_CONFIG,
1152 	HV_X64_MSR_VP_ASSIST_PAGE,
1153 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1154 	HV_X64_MSR_TSC_EMULATION_STATUS,
1155 
1156 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1157 	MSR_KVM_PV_EOI_EN,
1158 
1159 	MSR_IA32_TSC_ADJUST,
1160 	MSR_IA32_TSCDEADLINE,
1161 	MSR_IA32_MISC_ENABLE,
1162 	MSR_IA32_MCG_STATUS,
1163 	MSR_IA32_MCG_CTL,
1164 	MSR_IA32_MCG_EXT_CTL,
1165 	MSR_IA32_SMBASE,
1166 	MSR_SMI_COUNT,
1167 	MSR_PLATFORM_INFO,
1168 	MSR_MISC_FEATURES_ENABLES,
1169 	MSR_AMD64_VIRT_SPEC_CTRL,
1170 };
1171 
1172 static unsigned num_emulated_msrs;
1173 
1174 /*
1175  * List of msr numbers which are used to expose MSR-based features that
1176  * can be used by a hypervisor to validate requested CPU features.
1177  */
1178 static u32 msr_based_features[] = {
1179 	MSR_IA32_VMX_BASIC,
1180 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1181 	MSR_IA32_VMX_PINBASED_CTLS,
1182 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1183 	MSR_IA32_VMX_PROCBASED_CTLS,
1184 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1185 	MSR_IA32_VMX_EXIT_CTLS,
1186 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1187 	MSR_IA32_VMX_ENTRY_CTLS,
1188 	MSR_IA32_VMX_MISC,
1189 	MSR_IA32_VMX_CR0_FIXED0,
1190 	MSR_IA32_VMX_CR0_FIXED1,
1191 	MSR_IA32_VMX_CR4_FIXED0,
1192 	MSR_IA32_VMX_CR4_FIXED1,
1193 	MSR_IA32_VMX_VMCS_ENUM,
1194 	MSR_IA32_VMX_PROCBASED_CTLS2,
1195 	MSR_IA32_VMX_EPT_VPID_CAP,
1196 	MSR_IA32_VMX_VMFUNC,
1197 
1198 	MSR_F10H_DECFG,
1199 	MSR_IA32_UCODE_REV,
1200 	MSR_IA32_ARCH_CAPABILITIES,
1201 };
1202 
1203 static unsigned int num_msr_based_features;
1204 
1205 u64 kvm_get_arch_capabilities(void)
1206 {
1207 	u64 data;
1208 
1209 	rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1210 
1211 	/*
1212 	 * If we're doing cache flushes (either "always" or "cond")
1213 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1214 	 * If an outer hypervisor is doing the cache flush for us
1215 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1216 	 * capability to the guest too, and if EPT is disabled we're not
1217 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1218 	 * require a nested hypervisor to do a flush of its own.
1219 	 */
1220 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1221 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1222 
1223 	return data;
1224 }
1225 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1226 
1227 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1228 {
1229 	switch (msr->index) {
1230 	case MSR_IA32_ARCH_CAPABILITIES:
1231 		msr->data = kvm_get_arch_capabilities();
1232 		break;
1233 	case MSR_IA32_UCODE_REV:
1234 		rdmsrl_safe(msr->index, &msr->data);
1235 		break;
1236 	default:
1237 		if (kvm_x86_ops->get_msr_feature(msr))
1238 			return 1;
1239 	}
1240 	return 0;
1241 }
1242 
1243 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1244 {
1245 	struct kvm_msr_entry msr;
1246 	int r;
1247 
1248 	msr.index = index;
1249 	r = kvm_get_msr_feature(&msr);
1250 	if (r)
1251 		return r;
1252 
1253 	*data = msr.data;
1254 
1255 	return 0;
1256 }
1257 
1258 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1259 {
1260 	if (efer & efer_reserved_bits)
1261 		return false;
1262 
1263 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1264 			return false;
1265 
1266 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1267 			return false;
1268 
1269 	return true;
1270 }
1271 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1272 
1273 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1274 {
1275 	u64 old_efer = vcpu->arch.efer;
1276 
1277 	if (!kvm_valid_efer(vcpu, efer))
1278 		return 1;
1279 
1280 	if (is_paging(vcpu)
1281 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1282 		return 1;
1283 
1284 	efer &= ~EFER_LMA;
1285 	efer |= vcpu->arch.efer & EFER_LMA;
1286 
1287 	kvm_x86_ops->set_efer(vcpu, efer);
1288 
1289 	/* Update reserved bits */
1290 	if ((efer ^ old_efer) & EFER_NX)
1291 		kvm_mmu_reset_context(vcpu);
1292 
1293 	return 0;
1294 }
1295 
1296 void kvm_enable_efer_bits(u64 mask)
1297 {
1298        efer_reserved_bits &= ~mask;
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1301 
1302 /*
1303  * Writes msr value into into the appropriate "register".
1304  * Returns 0 on success, non-0 otherwise.
1305  * Assumes vcpu_load() was already called.
1306  */
1307 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1308 {
1309 	switch (msr->index) {
1310 	case MSR_FS_BASE:
1311 	case MSR_GS_BASE:
1312 	case MSR_KERNEL_GS_BASE:
1313 	case MSR_CSTAR:
1314 	case MSR_LSTAR:
1315 		if (is_noncanonical_address(msr->data, vcpu))
1316 			return 1;
1317 		break;
1318 	case MSR_IA32_SYSENTER_EIP:
1319 	case MSR_IA32_SYSENTER_ESP:
1320 		/*
1321 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1322 		 * non-canonical address is written on Intel but not on
1323 		 * AMD (which ignores the top 32-bits, because it does
1324 		 * not implement 64-bit SYSENTER).
1325 		 *
1326 		 * 64-bit code should hence be able to write a non-canonical
1327 		 * value on AMD.  Making the address canonical ensures that
1328 		 * vmentry does not fail on Intel after writing a non-canonical
1329 		 * value, and that something deterministic happens if the guest
1330 		 * invokes 64-bit SYSENTER.
1331 		 */
1332 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1333 	}
1334 	return kvm_x86_ops->set_msr(vcpu, msr);
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_set_msr);
1337 
1338 /*
1339  * Adapt set_msr() to msr_io()'s calling convention
1340  */
1341 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1342 {
1343 	struct msr_data msr;
1344 	int r;
1345 
1346 	msr.index = index;
1347 	msr.host_initiated = true;
1348 	r = kvm_get_msr(vcpu, &msr);
1349 	if (r)
1350 		return r;
1351 
1352 	*data = msr.data;
1353 	return 0;
1354 }
1355 
1356 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1357 {
1358 	struct msr_data msr;
1359 
1360 	msr.data = *data;
1361 	msr.index = index;
1362 	msr.host_initiated = true;
1363 	return kvm_set_msr(vcpu, &msr);
1364 }
1365 
1366 #ifdef CONFIG_X86_64
1367 struct pvclock_gtod_data {
1368 	seqcount_t	seq;
1369 
1370 	struct { /* extract of a clocksource struct */
1371 		int vclock_mode;
1372 		u64	cycle_last;
1373 		u64	mask;
1374 		u32	mult;
1375 		u32	shift;
1376 	} clock;
1377 
1378 	u64		boot_ns;
1379 	u64		nsec_base;
1380 	u64		wall_time_sec;
1381 };
1382 
1383 static struct pvclock_gtod_data pvclock_gtod_data;
1384 
1385 static void update_pvclock_gtod(struct timekeeper *tk)
1386 {
1387 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1388 	u64 boot_ns;
1389 
1390 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1391 
1392 	write_seqcount_begin(&vdata->seq);
1393 
1394 	/* copy pvclock gtod data */
1395 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1396 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1397 	vdata->clock.mask		= tk->tkr_mono.mask;
1398 	vdata->clock.mult		= tk->tkr_mono.mult;
1399 	vdata->clock.shift		= tk->tkr_mono.shift;
1400 
1401 	vdata->boot_ns			= boot_ns;
1402 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1403 
1404 	vdata->wall_time_sec            = tk->xtime_sec;
1405 
1406 	write_seqcount_end(&vdata->seq);
1407 }
1408 #endif
1409 
1410 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1411 {
1412 	/*
1413 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1414 	 * vcpu_enter_guest.  This function is only called from
1415 	 * the physical CPU that is running vcpu.
1416 	 */
1417 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1418 }
1419 
1420 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1421 {
1422 	int version;
1423 	int r;
1424 	struct pvclock_wall_clock wc;
1425 	struct timespec64 boot;
1426 
1427 	if (!wall_clock)
1428 		return;
1429 
1430 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1431 	if (r)
1432 		return;
1433 
1434 	if (version & 1)
1435 		++version;  /* first time write, random junk */
1436 
1437 	++version;
1438 
1439 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1440 		return;
1441 
1442 	/*
1443 	 * The guest calculates current wall clock time by adding
1444 	 * system time (updated by kvm_guest_time_update below) to the
1445 	 * wall clock specified here.  guest system time equals host
1446 	 * system time for us, thus we must fill in host boot time here.
1447 	 */
1448 	getboottime64(&boot);
1449 
1450 	if (kvm->arch.kvmclock_offset) {
1451 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1452 		boot = timespec64_sub(boot, ts);
1453 	}
1454 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1455 	wc.nsec = boot.tv_nsec;
1456 	wc.version = version;
1457 
1458 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1459 
1460 	version++;
1461 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1462 }
1463 
1464 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1465 {
1466 	do_shl32_div32(dividend, divisor);
1467 	return dividend;
1468 }
1469 
1470 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1471 			       s8 *pshift, u32 *pmultiplier)
1472 {
1473 	uint64_t scaled64;
1474 	int32_t  shift = 0;
1475 	uint64_t tps64;
1476 	uint32_t tps32;
1477 
1478 	tps64 = base_hz;
1479 	scaled64 = scaled_hz;
1480 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1481 		tps64 >>= 1;
1482 		shift--;
1483 	}
1484 
1485 	tps32 = (uint32_t)tps64;
1486 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1487 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1488 			scaled64 >>= 1;
1489 		else
1490 			tps32 <<= 1;
1491 		shift++;
1492 	}
1493 
1494 	*pshift = shift;
1495 	*pmultiplier = div_frac(scaled64, tps32);
1496 
1497 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1498 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1499 }
1500 
1501 #ifdef CONFIG_X86_64
1502 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1503 #endif
1504 
1505 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1506 static unsigned long max_tsc_khz;
1507 
1508 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1509 {
1510 	u64 v = (u64)khz * (1000000 + ppm);
1511 	do_div(v, 1000000);
1512 	return v;
1513 }
1514 
1515 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1516 {
1517 	u64 ratio;
1518 
1519 	/* Guest TSC same frequency as host TSC? */
1520 	if (!scale) {
1521 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1522 		return 0;
1523 	}
1524 
1525 	/* TSC scaling supported? */
1526 	if (!kvm_has_tsc_control) {
1527 		if (user_tsc_khz > tsc_khz) {
1528 			vcpu->arch.tsc_catchup = 1;
1529 			vcpu->arch.tsc_always_catchup = 1;
1530 			return 0;
1531 		} else {
1532 			WARN(1, "user requested TSC rate below hardware speed\n");
1533 			return -1;
1534 		}
1535 	}
1536 
1537 	/* TSC scaling required  - calculate ratio */
1538 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1539 				user_tsc_khz, tsc_khz);
1540 
1541 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1542 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1543 			  user_tsc_khz);
1544 		return -1;
1545 	}
1546 
1547 	vcpu->arch.tsc_scaling_ratio = ratio;
1548 	return 0;
1549 }
1550 
1551 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1552 {
1553 	u32 thresh_lo, thresh_hi;
1554 	int use_scaling = 0;
1555 
1556 	/* tsc_khz can be zero if TSC calibration fails */
1557 	if (user_tsc_khz == 0) {
1558 		/* set tsc_scaling_ratio to a safe value */
1559 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1560 		return -1;
1561 	}
1562 
1563 	/* Compute a scale to convert nanoseconds in TSC cycles */
1564 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1565 			   &vcpu->arch.virtual_tsc_shift,
1566 			   &vcpu->arch.virtual_tsc_mult);
1567 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1568 
1569 	/*
1570 	 * Compute the variation in TSC rate which is acceptable
1571 	 * within the range of tolerance and decide if the
1572 	 * rate being applied is within that bounds of the hardware
1573 	 * rate.  If so, no scaling or compensation need be done.
1574 	 */
1575 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1576 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1577 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1578 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1579 		use_scaling = 1;
1580 	}
1581 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1582 }
1583 
1584 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1585 {
1586 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1587 				      vcpu->arch.virtual_tsc_mult,
1588 				      vcpu->arch.virtual_tsc_shift);
1589 	tsc += vcpu->arch.this_tsc_write;
1590 	return tsc;
1591 }
1592 
1593 static inline int gtod_is_based_on_tsc(int mode)
1594 {
1595 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1596 }
1597 
1598 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1599 {
1600 #ifdef CONFIG_X86_64
1601 	bool vcpus_matched;
1602 	struct kvm_arch *ka = &vcpu->kvm->arch;
1603 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1604 
1605 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1606 			 atomic_read(&vcpu->kvm->online_vcpus));
1607 
1608 	/*
1609 	 * Once the masterclock is enabled, always perform request in
1610 	 * order to update it.
1611 	 *
1612 	 * In order to enable masterclock, the host clocksource must be TSC
1613 	 * and the vcpus need to have matched TSCs.  When that happens,
1614 	 * perform request to enable masterclock.
1615 	 */
1616 	if (ka->use_master_clock ||
1617 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1618 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1619 
1620 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1621 			    atomic_read(&vcpu->kvm->online_vcpus),
1622 		            ka->use_master_clock, gtod->clock.vclock_mode);
1623 #endif
1624 }
1625 
1626 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1627 {
1628 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1629 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1630 }
1631 
1632 /*
1633  * Multiply tsc by a fixed point number represented by ratio.
1634  *
1635  * The most significant 64-N bits (mult) of ratio represent the
1636  * integral part of the fixed point number; the remaining N bits
1637  * (frac) represent the fractional part, ie. ratio represents a fixed
1638  * point number (mult + frac * 2^(-N)).
1639  *
1640  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1641  */
1642 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1643 {
1644 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1645 }
1646 
1647 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1648 {
1649 	u64 _tsc = tsc;
1650 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1651 
1652 	if (ratio != kvm_default_tsc_scaling_ratio)
1653 		_tsc = __scale_tsc(ratio, tsc);
1654 
1655 	return _tsc;
1656 }
1657 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1658 
1659 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1660 {
1661 	u64 tsc;
1662 
1663 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1664 
1665 	return target_tsc - tsc;
1666 }
1667 
1668 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1669 {
1670 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1671 
1672 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1673 }
1674 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1675 
1676 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1677 {
1678 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1679 }
1680 
1681 static inline bool kvm_check_tsc_unstable(void)
1682 {
1683 #ifdef CONFIG_X86_64
1684 	/*
1685 	 * TSC is marked unstable when we're running on Hyper-V,
1686 	 * 'TSC page' clocksource is good.
1687 	 */
1688 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1689 		return false;
1690 #endif
1691 	return check_tsc_unstable();
1692 }
1693 
1694 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1695 {
1696 	struct kvm *kvm = vcpu->kvm;
1697 	u64 offset, ns, elapsed;
1698 	unsigned long flags;
1699 	bool matched;
1700 	bool already_matched;
1701 	u64 data = msr->data;
1702 	bool synchronizing = false;
1703 
1704 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1705 	offset = kvm_compute_tsc_offset(vcpu, data);
1706 	ns = ktime_get_boot_ns();
1707 	elapsed = ns - kvm->arch.last_tsc_nsec;
1708 
1709 	if (vcpu->arch.virtual_tsc_khz) {
1710 		if (data == 0 && msr->host_initiated) {
1711 			/*
1712 			 * detection of vcpu initialization -- need to sync
1713 			 * with other vCPUs. This particularly helps to keep
1714 			 * kvm_clock stable after CPU hotplug
1715 			 */
1716 			synchronizing = true;
1717 		} else {
1718 			u64 tsc_exp = kvm->arch.last_tsc_write +
1719 						nsec_to_cycles(vcpu, elapsed);
1720 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1721 			/*
1722 			 * Special case: TSC write with a small delta (1 second)
1723 			 * of virtual cycle time against real time is
1724 			 * interpreted as an attempt to synchronize the CPU.
1725 			 */
1726 			synchronizing = data < tsc_exp + tsc_hz &&
1727 					data + tsc_hz > tsc_exp;
1728 		}
1729 	}
1730 
1731 	/*
1732 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1733 	 * TSC, we add elapsed time in this computation.  We could let the
1734 	 * compensation code attempt to catch up if we fall behind, but
1735 	 * it's better to try to match offsets from the beginning.
1736          */
1737 	if (synchronizing &&
1738 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1739 		if (!kvm_check_tsc_unstable()) {
1740 			offset = kvm->arch.cur_tsc_offset;
1741 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1742 		} else {
1743 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1744 			data += delta;
1745 			offset = kvm_compute_tsc_offset(vcpu, data);
1746 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1747 		}
1748 		matched = true;
1749 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1750 	} else {
1751 		/*
1752 		 * We split periods of matched TSC writes into generations.
1753 		 * For each generation, we track the original measured
1754 		 * nanosecond time, offset, and write, so if TSCs are in
1755 		 * sync, we can match exact offset, and if not, we can match
1756 		 * exact software computation in compute_guest_tsc()
1757 		 *
1758 		 * These values are tracked in kvm->arch.cur_xxx variables.
1759 		 */
1760 		kvm->arch.cur_tsc_generation++;
1761 		kvm->arch.cur_tsc_nsec = ns;
1762 		kvm->arch.cur_tsc_write = data;
1763 		kvm->arch.cur_tsc_offset = offset;
1764 		matched = false;
1765 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1766 			 kvm->arch.cur_tsc_generation, data);
1767 	}
1768 
1769 	/*
1770 	 * We also track th most recent recorded KHZ, write and time to
1771 	 * allow the matching interval to be extended at each write.
1772 	 */
1773 	kvm->arch.last_tsc_nsec = ns;
1774 	kvm->arch.last_tsc_write = data;
1775 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1776 
1777 	vcpu->arch.last_guest_tsc = data;
1778 
1779 	/* Keep track of which generation this VCPU has synchronized to */
1780 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1781 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1782 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1783 
1784 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1785 		update_ia32_tsc_adjust_msr(vcpu, offset);
1786 
1787 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1788 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1789 
1790 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1791 	if (!matched) {
1792 		kvm->arch.nr_vcpus_matched_tsc = 0;
1793 	} else if (!already_matched) {
1794 		kvm->arch.nr_vcpus_matched_tsc++;
1795 	}
1796 
1797 	kvm_track_tsc_matching(vcpu);
1798 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1799 }
1800 
1801 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1802 
1803 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1804 					   s64 adjustment)
1805 {
1806 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1807 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1808 }
1809 
1810 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1811 {
1812 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1813 		WARN_ON(adjustment < 0);
1814 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1815 	adjust_tsc_offset_guest(vcpu, adjustment);
1816 }
1817 
1818 #ifdef CONFIG_X86_64
1819 
1820 static u64 read_tsc(void)
1821 {
1822 	u64 ret = (u64)rdtsc_ordered();
1823 	u64 last = pvclock_gtod_data.clock.cycle_last;
1824 
1825 	if (likely(ret >= last))
1826 		return ret;
1827 
1828 	/*
1829 	 * GCC likes to generate cmov here, but this branch is extremely
1830 	 * predictable (it's just a function of time and the likely is
1831 	 * very likely) and there's a data dependence, so force GCC
1832 	 * to generate a branch instead.  I don't barrier() because
1833 	 * we don't actually need a barrier, and if this function
1834 	 * ever gets inlined it will generate worse code.
1835 	 */
1836 	asm volatile ("");
1837 	return last;
1838 }
1839 
1840 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1841 {
1842 	long v;
1843 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1844 	u64 tsc_pg_val;
1845 
1846 	switch (gtod->clock.vclock_mode) {
1847 	case VCLOCK_HVCLOCK:
1848 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1849 						  tsc_timestamp);
1850 		if (tsc_pg_val != U64_MAX) {
1851 			/* TSC page valid */
1852 			*mode = VCLOCK_HVCLOCK;
1853 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1854 				gtod->clock.mask;
1855 		} else {
1856 			/* TSC page invalid */
1857 			*mode = VCLOCK_NONE;
1858 		}
1859 		break;
1860 	case VCLOCK_TSC:
1861 		*mode = VCLOCK_TSC;
1862 		*tsc_timestamp = read_tsc();
1863 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1864 			gtod->clock.mask;
1865 		break;
1866 	default:
1867 		*mode = VCLOCK_NONE;
1868 	}
1869 
1870 	if (*mode == VCLOCK_NONE)
1871 		*tsc_timestamp = v = 0;
1872 
1873 	return v * gtod->clock.mult;
1874 }
1875 
1876 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1877 {
1878 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1879 	unsigned long seq;
1880 	int mode;
1881 	u64 ns;
1882 
1883 	do {
1884 		seq = read_seqcount_begin(&gtod->seq);
1885 		ns = gtod->nsec_base;
1886 		ns += vgettsc(tsc_timestamp, &mode);
1887 		ns >>= gtod->clock.shift;
1888 		ns += gtod->boot_ns;
1889 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1890 	*t = ns;
1891 
1892 	return mode;
1893 }
1894 
1895 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1896 {
1897 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1898 	unsigned long seq;
1899 	int mode;
1900 	u64 ns;
1901 
1902 	do {
1903 		seq = read_seqcount_begin(&gtod->seq);
1904 		ts->tv_sec = gtod->wall_time_sec;
1905 		ns = gtod->nsec_base;
1906 		ns += vgettsc(tsc_timestamp, &mode);
1907 		ns >>= gtod->clock.shift;
1908 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1909 
1910 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1911 	ts->tv_nsec = ns;
1912 
1913 	return mode;
1914 }
1915 
1916 /* returns true if host is using TSC based clocksource */
1917 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1918 {
1919 	/* checked again under seqlock below */
1920 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1921 		return false;
1922 
1923 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1924 						      tsc_timestamp));
1925 }
1926 
1927 /* returns true if host is using TSC based clocksource */
1928 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1929 					   u64 *tsc_timestamp)
1930 {
1931 	/* checked again under seqlock below */
1932 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1933 		return false;
1934 
1935 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1936 }
1937 #endif
1938 
1939 /*
1940  *
1941  * Assuming a stable TSC across physical CPUS, and a stable TSC
1942  * across virtual CPUs, the following condition is possible.
1943  * Each numbered line represents an event visible to both
1944  * CPUs at the next numbered event.
1945  *
1946  * "timespecX" represents host monotonic time. "tscX" represents
1947  * RDTSC value.
1948  *
1949  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1950  *
1951  * 1.  read timespec0,tsc0
1952  * 2.					| timespec1 = timespec0 + N
1953  * 					| tsc1 = tsc0 + M
1954  * 3. transition to guest		| transition to guest
1955  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1956  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1957  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1958  *
1959  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1960  *
1961  * 	- ret0 < ret1
1962  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1963  *		...
1964  *	- 0 < N - M => M < N
1965  *
1966  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1967  * always the case (the difference between two distinct xtime instances
1968  * might be smaller then the difference between corresponding TSC reads,
1969  * when updating guest vcpus pvclock areas).
1970  *
1971  * To avoid that problem, do not allow visibility of distinct
1972  * system_timestamp/tsc_timestamp values simultaneously: use a master
1973  * copy of host monotonic time values. Update that master copy
1974  * in lockstep.
1975  *
1976  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1977  *
1978  */
1979 
1980 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1981 {
1982 #ifdef CONFIG_X86_64
1983 	struct kvm_arch *ka = &kvm->arch;
1984 	int vclock_mode;
1985 	bool host_tsc_clocksource, vcpus_matched;
1986 
1987 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1988 			atomic_read(&kvm->online_vcpus));
1989 
1990 	/*
1991 	 * If the host uses TSC clock, then passthrough TSC as stable
1992 	 * to the guest.
1993 	 */
1994 	host_tsc_clocksource = kvm_get_time_and_clockread(
1995 					&ka->master_kernel_ns,
1996 					&ka->master_cycle_now);
1997 
1998 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1999 				&& !ka->backwards_tsc_observed
2000 				&& !ka->boot_vcpu_runs_old_kvmclock;
2001 
2002 	if (ka->use_master_clock)
2003 		atomic_set(&kvm_guest_has_master_clock, 1);
2004 
2005 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2006 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2007 					vcpus_matched);
2008 #endif
2009 }
2010 
2011 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2012 {
2013 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2014 }
2015 
2016 static void kvm_gen_update_masterclock(struct kvm *kvm)
2017 {
2018 #ifdef CONFIG_X86_64
2019 	int i;
2020 	struct kvm_vcpu *vcpu;
2021 	struct kvm_arch *ka = &kvm->arch;
2022 
2023 	spin_lock(&ka->pvclock_gtod_sync_lock);
2024 	kvm_make_mclock_inprogress_request(kvm);
2025 	/* no guest entries from this point */
2026 	pvclock_update_vm_gtod_copy(kvm);
2027 
2028 	kvm_for_each_vcpu(i, vcpu, kvm)
2029 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2030 
2031 	/* guest entries allowed */
2032 	kvm_for_each_vcpu(i, vcpu, kvm)
2033 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2034 
2035 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2036 #endif
2037 }
2038 
2039 u64 get_kvmclock_ns(struct kvm *kvm)
2040 {
2041 	struct kvm_arch *ka = &kvm->arch;
2042 	struct pvclock_vcpu_time_info hv_clock;
2043 	u64 ret;
2044 
2045 	spin_lock(&ka->pvclock_gtod_sync_lock);
2046 	if (!ka->use_master_clock) {
2047 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2048 		return ktime_get_boot_ns() + ka->kvmclock_offset;
2049 	}
2050 
2051 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2052 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2053 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2054 
2055 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2056 	get_cpu();
2057 
2058 	if (__this_cpu_read(cpu_tsc_khz)) {
2059 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2060 				   &hv_clock.tsc_shift,
2061 				   &hv_clock.tsc_to_system_mul);
2062 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2063 	} else
2064 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2065 
2066 	put_cpu();
2067 
2068 	return ret;
2069 }
2070 
2071 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2072 {
2073 	struct kvm_vcpu_arch *vcpu = &v->arch;
2074 	struct pvclock_vcpu_time_info guest_hv_clock;
2075 
2076 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2077 		&guest_hv_clock, sizeof(guest_hv_clock))))
2078 		return;
2079 
2080 	/* This VCPU is paused, but it's legal for a guest to read another
2081 	 * VCPU's kvmclock, so we really have to follow the specification where
2082 	 * it says that version is odd if data is being modified, and even after
2083 	 * it is consistent.
2084 	 *
2085 	 * Version field updates must be kept separate.  This is because
2086 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2087 	 * writes within a string instruction are weakly ordered.  So there
2088 	 * are three writes overall.
2089 	 *
2090 	 * As a small optimization, only write the version field in the first
2091 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2092 	 * version field is the first in the struct.
2093 	 */
2094 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2095 
2096 	if (guest_hv_clock.version & 1)
2097 		++guest_hv_clock.version;  /* first time write, random junk */
2098 
2099 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2100 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2101 				&vcpu->hv_clock,
2102 				sizeof(vcpu->hv_clock.version));
2103 
2104 	smp_wmb();
2105 
2106 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2107 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2108 
2109 	if (vcpu->pvclock_set_guest_stopped_request) {
2110 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2111 		vcpu->pvclock_set_guest_stopped_request = false;
2112 	}
2113 
2114 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2115 
2116 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2117 				&vcpu->hv_clock,
2118 				sizeof(vcpu->hv_clock));
2119 
2120 	smp_wmb();
2121 
2122 	vcpu->hv_clock.version++;
2123 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2124 				&vcpu->hv_clock,
2125 				sizeof(vcpu->hv_clock.version));
2126 }
2127 
2128 static int kvm_guest_time_update(struct kvm_vcpu *v)
2129 {
2130 	unsigned long flags, tgt_tsc_khz;
2131 	struct kvm_vcpu_arch *vcpu = &v->arch;
2132 	struct kvm_arch *ka = &v->kvm->arch;
2133 	s64 kernel_ns;
2134 	u64 tsc_timestamp, host_tsc;
2135 	u8 pvclock_flags;
2136 	bool use_master_clock;
2137 
2138 	kernel_ns = 0;
2139 	host_tsc = 0;
2140 
2141 	/*
2142 	 * If the host uses TSC clock, then passthrough TSC as stable
2143 	 * to the guest.
2144 	 */
2145 	spin_lock(&ka->pvclock_gtod_sync_lock);
2146 	use_master_clock = ka->use_master_clock;
2147 	if (use_master_clock) {
2148 		host_tsc = ka->master_cycle_now;
2149 		kernel_ns = ka->master_kernel_ns;
2150 	}
2151 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2152 
2153 	/* Keep irq disabled to prevent changes to the clock */
2154 	local_irq_save(flags);
2155 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2156 	if (unlikely(tgt_tsc_khz == 0)) {
2157 		local_irq_restore(flags);
2158 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2159 		return 1;
2160 	}
2161 	if (!use_master_clock) {
2162 		host_tsc = rdtsc();
2163 		kernel_ns = ktime_get_boot_ns();
2164 	}
2165 
2166 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2167 
2168 	/*
2169 	 * We may have to catch up the TSC to match elapsed wall clock
2170 	 * time for two reasons, even if kvmclock is used.
2171 	 *   1) CPU could have been running below the maximum TSC rate
2172 	 *   2) Broken TSC compensation resets the base at each VCPU
2173 	 *      entry to avoid unknown leaps of TSC even when running
2174 	 *      again on the same CPU.  This may cause apparent elapsed
2175 	 *      time to disappear, and the guest to stand still or run
2176 	 *	very slowly.
2177 	 */
2178 	if (vcpu->tsc_catchup) {
2179 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2180 		if (tsc > tsc_timestamp) {
2181 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2182 			tsc_timestamp = tsc;
2183 		}
2184 	}
2185 
2186 	local_irq_restore(flags);
2187 
2188 	/* With all the info we got, fill in the values */
2189 
2190 	if (kvm_has_tsc_control)
2191 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2192 
2193 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2194 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2195 				   &vcpu->hv_clock.tsc_shift,
2196 				   &vcpu->hv_clock.tsc_to_system_mul);
2197 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2198 	}
2199 
2200 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2201 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2202 	vcpu->last_guest_tsc = tsc_timestamp;
2203 
2204 	/* If the host uses TSC clocksource, then it is stable */
2205 	pvclock_flags = 0;
2206 	if (use_master_clock)
2207 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2208 
2209 	vcpu->hv_clock.flags = pvclock_flags;
2210 
2211 	if (vcpu->pv_time_enabled)
2212 		kvm_setup_pvclock_page(v);
2213 	if (v == kvm_get_vcpu(v->kvm, 0))
2214 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2215 	return 0;
2216 }
2217 
2218 /*
2219  * kvmclock updates which are isolated to a given vcpu, such as
2220  * vcpu->cpu migration, should not allow system_timestamp from
2221  * the rest of the vcpus to remain static. Otherwise ntp frequency
2222  * correction applies to one vcpu's system_timestamp but not
2223  * the others.
2224  *
2225  * So in those cases, request a kvmclock update for all vcpus.
2226  * We need to rate-limit these requests though, as they can
2227  * considerably slow guests that have a large number of vcpus.
2228  * The time for a remote vcpu to update its kvmclock is bound
2229  * by the delay we use to rate-limit the updates.
2230  */
2231 
2232 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2233 
2234 static void kvmclock_update_fn(struct work_struct *work)
2235 {
2236 	int i;
2237 	struct delayed_work *dwork = to_delayed_work(work);
2238 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2239 					   kvmclock_update_work);
2240 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2241 	struct kvm_vcpu *vcpu;
2242 
2243 	kvm_for_each_vcpu(i, vcpu, kvm) {
2244 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2245 		kvm_vcpu_kick(vcpu);
2246 	}
2247 }
2248 
2249 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2250 {
2251 	struct kvm *kvm = v->kvm;
2252 
2253 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2254 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2255 					KVMCLOCK_UPDATE_DELAY);
2256 }
2257 
2258 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2259 
2260 static void kvmclock_sync_fn(struct work_struct *work)
2261 {
2262 	struct delayed_work *dwork = to_delayed_work(work);
2263 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2264 					   kvmclock_sync_work);
2265 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2266 
2267 	if (!kvmclock_periodic_sync)
2268 		return;
2269 
2270 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2271 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2272 					KVMCLOCK_SYNC_PERIOD);
2273 }
2274 
2275 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2276 {
2277 	u64 mcg_cap = vcpu->arch.mcg_cap;
2278 	unsigned bank_num = mcg_cap & 0xff;
2279 	u32 msr = msr_info->index;
2280 	u64 data = msr_info->data;
2281 
2282 	switch (msr) {
2283 	case MSR_IA32_MCG_STATUS:
2284 		vcpu->arch.mcg_status = data;
2285 		break;
2286 	case MSR_IA32_MCG_CTL:
2287 		if (!(mcg_cap & MCG_CTL_P) &&
2288 		    (data || !msr_info->host_initiated))
2289 			return 1;
2290 		if (data != 0 && data != ~(u64)0)
2291 			return 1;
2292 		vcpu->arch.mcg_ctl = data;
2293 		break;
2294 	default:
2295 		if (msr >= MSR_IA32_MC0_CTL &&
2296 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2297 			u32 offset = msr - MSR_IA32_MC0_CTL;
2298 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2299 			 * some Linux kernels though clear bit 10 in bank 4 to
2300 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2301 			 * this to avoid an uncatched #GP in the guest
2302 			 */
2303 			if ((offset & 0x3) == 0 &&
2304 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2305 				return -1;
2306 			if (!msr_info->host_initiated &&
2307 				(offset & 0x3) == 1 && data != 0)
2308 				return -1;
2309 			vcpu->arch.mce_banks[offset] = data;
2310 			break;
2311 		}
2312 		return 1;
2313 	}
2314 	return 0;
2315 }
2316 
2317 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2318 {
2319 	struct kvm *kvm = vcpu->kvm;
2320 	int lm = is_long_mode(vcpu);
2321 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2322 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2323 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2324 		: kvm->arch.xen_hvm_config.blob_size_32;
2325 	u32 page_num = data & ~PAGE_MASK;
2326 	u64 page_addr = data & PAGE_MASK;
2327 	u8 *page;
2328 	int r;
2329 
2330 	r = -E2BIG;
2331 	if (page_num >= blob_size)
2332 		goto out;
2333 	r = -ENOMEM;
2334 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2335 	if (IS_ERR(page)) {
2336 		r = PTR_ERR(page);
2337 		goto out;
2338 	}
2339 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2340 		goto out_free;
2341 	r = 0;
2342 out_free:
2343 	kfree(page);
2344 out:
2345 	return r;
2346 }
2347 
2348 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2349 {
2350 	gpa_t gpa = data & ~0x3f;
2351 
2352 	/* Bits 3:5 are reserved, Should be zero */
2353 	if (data & 0x38)
2354 		return 1;
2355 
2356 	vcpu->arch.apf.msr_val = data;
2357 
2358 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2359 		kvm_clear_async_pf_completion_queue(vcpu);
2360 		kvm_async_pf_hash_reset(vcpu);
2361 		return 0;
2362 	}
2363 
2364 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2365 					sizeof(u32)))
2366 		return 1;
2367 
2368 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2369 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2370 	kvm_async_pf_wakeup_all(vcpu);
2371 	return 0;
2372 }
2373 
2374 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2375 {
2376 	vcpu->arch.pv_time_enabled = false;
2377 }
2378 
2379 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2380 {
2381 	++vcpu->stat.tlb_flush;
2382 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2383 }
2384 
2385 static void record_steal_time(struct kvm_vcpu *vcpu)
2386 {
2387 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2388 		return;
2389 
2390 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2391 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2392 		return;
2393 
2394 	/*
2395 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2396 	 * expensive IPIs.
2397 	 */
2398 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2399 		kvm_vcpu_flush_tlb(vcpu, false);
2400 
2401 	if (vcpu->arch.st.steal.version & 1)
2402 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2403 
2404 	vcpu->arch.st.steal.version += 1;
2405 
2406 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2407 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2408 
2409 	smp_wmb();
2410 
2411 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2412 		vcpu->arch.st.last_steal;
2413 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2414 
2415 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2416 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2417 
2418 	smp_wmb();
2419 
2420 	vcpu->arch.st.steal.version += 1;
2421 
2422 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2423 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2424 }
2425 
2426 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2427 {
2428 	bool pr = false;
2429 	u32 msr = msr_info->index;
2430 	u64 data = msr_info->data;
2431 
2432 	switch (msr) {
2433 	case MSR_AMD64_NB_CFG:
2434 	case MSR_IA32_UCODE_WRITE:
2435 	case MSR_VM_HSAVE_PA:
2436 	case MSR_AMD64_PATCH_LOADER:
2437 	case MSR_AMD64_BU_CFG2:
2438 	case MSR_AMD64_DC_CFG:
2439 	case MSR_F15H_EX_CFG:
2440 		break;
2441 
2442 	case MSR_IA32_UCODE_REV:
2443 		if (msr_info->host_initiated)
2444 			vcpu->arch.microcode_version = data;
2445 		break;
2446 	case MSR_EFER:
2447 		return set_efer(vcpu, data);
2448 	case MSR_K7_HWCR:
2449 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2450 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2451 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2452 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2453 		if (data != 0) {
2454 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2455 				    data);
2456 			return 1;
2457 		}
2458 		break;
2459 	case MSR_FAM10H_MMIO_CONF_BASE:
2460 		if (data != 0) {
2461 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2462 				    "0x%llx\n", data);
2463 			return 1;
2464 		}
2465 		break;
2466 	case MSR_IA32_DEBUGCTLMSR:
2467 		if (!data) {
2468 			/* We support the non-activated case already */
2469 			break;
2470 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2471 			/* Values other than LBR and BTF are vendor-specific,
2472 			   thus reserved and should throw a #GP */
2473 			return 1;
2474 		}
2475 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2476 			    __func__, data);
2477 		break;
2478 	case 0x200 ... 0x2ff:
2479 		return kvm_mtrr_set_msr(vcpu, msr, data);
2480 	case MSR_IA32_APICBASE:
2481 		return kvm_set_apic_base(vcpu, msr_info);
2482 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2483 		return kvm_x2apic_msr_write(vcpu, msr, data);
2484 	case MSR_IA32_TSCDEADLINE:
2485 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2486 		break;
2487 	case MSR_IA32_TSC_ADJUST:
2488 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2489 			if (!msr_info->host_initiated) {
2490 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2491 				adjust_tsc_offset_guest(vcpu, adj);
2492 			}
2493 			vcpu->arch.ia32_tsc_adjust_msr = data;
2494 		}
2495 		break;
2496 	case MSR_IA32_MISC_ENABLE:
2497 		vcpu->arch.ia32_misc_enable_msr = data;
2498 		break;
2499 	case MSR_IA32_SMBASE:
2500 		if (!msr_info->host_initiated)
2501 			return 1;
2502 		vcpu->arch.smbase = data;
2503 		break;
2504 	case MSR_IA32_TSC:
2505 		kvm_write_tsc(vcpu, msr_info);
2506 		break;
2507 	case MSR_SMI_COUNT:
2508 		if (!msr_info->host_initiated)
2509 			return 1;
2510 		vcpu->arch.smi_count = data;
2511 		break;
2512 	case MSR_KVM_WALL_CLOCK_NEW:
2513 	case MSR_KVM_WALL_CLOCK:
2514 		vcpu->kvm->arch.wall_clock = data;
2515 		kvm_write_wall_clock(vcpu->kvm, data);
2516 		break;
2517 	case MSR_KVM_SYSTEM_TIME_NEW:
2518 	case MSR_KVM_SYSTEM_TIME: {
2519 		struct kvm_arch *ka = &vcpu->kvm->arch;
2520 
2521 		kvmclock_reset(vcpu);
2522 
2523 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2524 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2525 
2526 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2527 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2528 
2529 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2530 		}
2531 
2532 		vcpu->arch.time = data;
2533 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2534 
2535 		/* we verify if the enable bit is set... */
2536 		if (!(data & 1))
2537 			break;
2538 
2539 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2540 		     &vcpu->arch.pv_time, data & ~1ULL,
2541 		     sizeof(struct pvclock_vcpu_time_info)))
2542 			vcpu->arch.pv_time_enabled = false;
2543 		else
2544 			vcpu->arch.pv_time_enabled = true;
2545 
2546 		break;
2547 	}
2548 	case MSR_KVM_ASYNC_PF_EN:
2549 		if (kvm_pv_enable_async_pf(vcpu, data))
2550 			return 1;
2551 		break;
2552 	case MSR_KVM_STEAL_TIME:
2553 
2554 		if (unlikely(!sched_info_on()))
2555 			return 1;
2556 
2557 		if (data & KVM_STEAL_RESERVED_MASK)
2558 			return 1;
2559 
2560 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2561 						data & KVM_STEAL_VALID_BITS,
2562 						sizeof(struct kvm_steal_time)))
2563 			return 1;
2564 
2565 		vcpu->arch.st.msr_val = data;
2566 
2567 		if (!(data & KVM_MSR_ENABLED))
2568 			break;
2569 
2570 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2571 
2572 		break;
2573 	case MSR_KVM_PV_EOI_EN:
2574 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2575 			return 1;
2576 		break;
2577 
2578 	case MSR_IA32_MCG_CTL:
2579 	case MSR_IA32_MCG_STATUS:
2580 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2581 		return set_msr_mce(vcpu, msr_info);
2582 
2583 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2584 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2585 		pr = true; /* fall through */
2586 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2587 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2588 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2589 			return kvm_pmu_set_msr(vcpu, msr_info);
2590 
2591 		if (pr || data != 0)
2592 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2593 				    "0x%x data 0x%llx\n", msr, data);
2594 		break;
2595 	case MSR_K7_CLK_CTL:
2596 		/*
2597 		 * Ignore all writes to this no longer documented MSR.
2598 		 * Writes are only relevant for old K7 processors,
2599 		 * all pre-dating SVM, but a recommended workaround from
2600 		 * AMD for these chips. It is possible to specify the
2601 		 * affected processor models on the command line, hence
2602 		 * the need to ignore the workaround.
2603 		 */
2604 		break;
2605 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2606 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2607 	case HV_X64_MSR_CRASH_CTL:
2608 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2609 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2610 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2611 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2612 		return kvm_hv_set_msr_common(vcpu, msr, data,
2613 					     msr_info->host_initiated);
2614 	case MSR_IA32_BBL_CR_CTL3:
2615 		/* Drop writes to this legacy MSR -- see rdmsr
2616 		 * counterpart for further detail.
2617 		 */
2618 		if (report_ignored_msrs)
2619 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2620 				msr, data);
2621 		break;
2622 	case MSR_AMD64_OSVW_ID_LENGTH:
2623 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2624 			return 1;
2625 		vcpu->arch.osvw.length = data;
2626 		break;
2627 	case MSR_AMD64_OSVW_STATUS:
2628 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2629 			return 1;
2630 		vcpu->arch.osvw.status = data;
2631 		break;
2632 	case MSR_PLATFORM_INFO:
2633 		if (!msr_info->host_initiated ||
2634 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2635 		     cpuid_fault_enabled(vcpu)))
2636 			return 1;
2637 		vcpu->arch.msr_platform_info = data;
2638 		break;
2639 	case MSR_MISC_FEATURES_ENABLES:
2640 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2641 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2642 		     !supports_cpuid_fault(vcpu)))
2643 			return 1;
2644 		vcpu->arch.msr_misc_features_enables = data;
2645 		break;
2646 	default:
2647 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2648 			return xen_hvm_config(vcpu, data);
2649 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2650 			return kvm_pmu_set_msr(vcpu, msr_info);
2651 		if (!ignore_msrs) {
2652 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2653 				    msr, data);
2654 			return 1;
2655 		} else {
2656 			if (report_ignored_msrs)
2657 				vcpu_unimpl(vcpu,
2658 					"ignored wrmsr: 0x%x data 0x%llx\n",
2659 					msr, data);
2660 			break;
2661 		}
2662 	}
2663 	return 0;
2664 }
2665 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2666 
2667 
2668 /*
2669  * Reads an msr value (of 'msr_index') into 'pdata'.
2670  * Returns 0 on success, non-0 otherwise.
2671  * Assumes vcpu_load() was already called.
2672  */
2673 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2674 {
2675 	return kvm_x86_ops->get_msr(vcpu, msr);
2676 }
2677 EXPORT_SYMBOL_GPL(kvm_get_msr);
2678 
2679 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2680 {
2681 	u64 data;
2682 	u64 mcg_cap = vcpu->arch.mcg_cap;
2683 	unsigned bank_num = mcg_cap & 0xff;
2684 
2685 	switch (msr) {
2686 	case MSR_IA32_P5_MC_ADDR:
2687 	case MSR_IA32_P5_MC_TYPE:
2688 		data = 0;
2689 		break;
2690 	case MSR_IA32_MCG_CAP:
2691 		data = vcpu->arch.mcg_cap;
2692 		break;
2693 	case MSR_IA32_MCG_CTL:
2694 		if (!(mcg_cap & MCG_CTL_P) && !host)
2695 			return 1;
2696 		data = vcpu->arch.mcg_ctl;
2697 		break;
2698 	case MSR_IA32_MCG_STATUS:
2699 		data = vcpu->arch.mcg_status;
2700 		break;
2701 	default:
2702 		if (msr >= MSR_IA32_MC0_CTL &&
2703 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2704 			u32 offset = msr - MSR_IA32_MC0_CTL;
2705 			data = vcpu->arch.mce_banks[offset];
2706 			break;
2707 		}
2708 		return 1;
2709 	}
2710 	*pdata = data;
2711 	return 0;
2712 }
2713 
2714 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2715 {
2716 	switch (msr_info->index) {
2717 	case MSR_IA32_PLATFORM_ID:
2718 	case MSR_IA32_EBL_CR_POWERON:
2719 	case MSR_IA32_DEBUGCTLMSR:
2720 	case MSR_IA32_LASTBRANCHFROMIP:
2721 	case MSR_IA32_LASTBRANCHTOIP:
2722 	case MSR_IA32_LASTINTFROMIP:
2723 	case MSR_IA32_LASTINTTOIP:
2724 	case MSR_K8_SYSCFG:
2725 	case MSR_K8_TSEG_ADDR:
2726 	case MSR_K8_TSEG_MASK:
2727 	case MSR_K7_HWCR:
2728 	case MSR_VM_HSAVE_PA:
2729 	case MSR_K8_INT_PENDING_MSG:
2730 	case MSR_AMD64_NB_CFG:
2731 	case MSR_FAM10H_MMIO_CONF_BASE:
2732 	case MSR_AMD64_BU_CFG2:
2733 	case MSR_IA32_PERF_CTL:
2734 	case MSR_AMD64_DC_CFG:
2735 	case MSR_F15H_EX_CFG:
2736 		msr_info->data = 0;
2737 		break;
2738 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2739 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2740 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2741 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2742 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2743 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2744 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2745 		msr_info->data = 0;
2746 		break;
2747 	case MSR_IA32_UCODE_REV:
2748 		msr_info->data = vcpu->arch.microcode_version;
2749 		break;
2750 	case MSR_IA32_TSC:
2751 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2752 		break;
2753 	case MSR_MTRRcap:
2754 	case 0x200 ... 0x2ff:
2755 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2756 	case 0xcd: /* fsb frequency */
2757 		msr_info->data = 3;
2758 		break;
2759 		/*
2760 		 * MSR_EBC_FREQUENCY_ID
2761 		 * Conservative value valid for even the basic CPU models.
2762 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2763 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2764 		 * and 266MHz for model 3, or 4. Set Core Clock
2765 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2766 		 * 31:24) even though these are only valid for CPU
2767 		 * models > 2, however guests may end up dividing or
2768 		 * multiplying by zero otherwise.
2769 		 */
2770 	case MSR_EBC_FREQUENCY_ID:
2771 		msr_info->data = 1 << 24;
2772 		break;
2773 	case MSR_IA32_APICBASE:
2774 		msr_info->data = kvm_get_apic_base(vcpu);
2775 		break;
2776 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2777 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2778 		break;
2779 	case MSR_IA32_TSCDEADLINE:
2780 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2781 		break;
2782 	case MSR_IA32_TSC_ADJUST:
2783 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2784 		break;
2785 	case MSR_IA32_MISC_ENABLE:
2786 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2787 		break;
2788 	case MSR_IA32_SMBASE:
2789 		if (!msr_info->host_initiated)
2790 			return 1;
2791 		msr_info->data = vcpu->arch.smbase;
2792 		break;
2793 	case MSR_SMI_COUNT:
2794 		msr_info->data = vcpu->arch.smi_count;
2795 		break;
2796 	case MSR_IA32_PERF_STATUS:
2797 		/* TSC increment by tick */
2798 		msr_info->data = 1000ULL;
2799 		/* CPU multiplier */
2800 		msr_info->data |= (((uint64_t)4ULL) << 40);
2801 		break;
2802 	case MSR_EFER:
2803 		msr_info->data = vcpu->arch.efer;
2804 		break;
2805 	case MSR_KVM_WALL_CLOCK:
2806 	case MSR_KVM_WALL_CLOCK_NEW:
2807 		msr_info->data = vcpu->kvm->arch.wall_clock;
2808 		break;
2809 	case MSR_KVM_SYSTEM_TIME:
2810 	case MSR_KVM_SYSTEM_TIME_NEW:
2811 		msr_info->data = vcpu->arch.time;
2812 		break;
2813 	case MSR_KVM_ASYNC_PF_EN:
2814 		msr_info->data = vcpu->arch.apf.msr_val;
2815 		break;
2816 	case MSR_KVM_STEAL_TIME:
2817 		msr_info->data = vcpu->arch.st.msr_val;
2818 		break;
2819 	case MSR_KVM_PV_EOI_EN:
2820 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2821 		break;
2822 	case MSR_IA32_P5_MC_ADDR:
2823 	case MSR_IA32_P5_MC_TYPE:
2824 	case MSR_IA32_MCG_CAP:
2825 	case MSR_IA32_MCG_CTL:
2826 	case MSR_IA32_MCG_STATUS:
2827 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2828 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2829 				   msr_info->host_initiated);
2830 	case MSR_K7_CLK_CTL:
2831 		/*
2832 		 * Provide expected ramp-up count for K7. All other
2833 		 * are set to zero, indicating minimum divisors for
2834 		 * every field.
2835 		 *
2836 		 * This prevents guest kernels on AMD host with CPU
2837 		 * type 6, model 8 and higher from exploding due to
2838 		 * the rdmsr failing.
2839 		 */
2840 		msr_info->data = 0x20000000;
2841 		break;
2842 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2843 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2844 	case HV_X64_MSR_CRASH_CTL:
2845 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2846 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2847 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2848 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2849 		return kvm_hv_get_msr_common(vcpu,
2850 					     msr_info->index, &msr_info->data,
2851 					     msr_info->host_initiated);
2852 		break;
2853 	case MSR_IA32_BBL_CR_CTL3:
2854 		/* This legacy MSR exists but isn't fully documented in current
2855 		 * silicon.  It is however accessed by winxp in very narrow
2856 		 * scenarios where it sets bit #19, itself documented as
2857 		 * a "reserved" bit.  Best effort attempt to source coherent
2858 		 * read data here should the balance of the register be
2859 		 * interpreted by the guest:
2860 		 *
2861 		 * L2 cache control register 3: 64GB range, 256KB size,
2862 		 * enabled, latency 0x1, configured
2863 		 */
2864 		msr_info->data = 0xbe702111;
2865 		break;
2866 	case MSR_AMD64_OSVW_ID_LENGTH:
2867 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2868 			return 1;
2869 		msr_info->data = vcpu->arch.osvw.length;
2870 		break;
2871 	case MSR_AMD64_OSVW_STATUS:
2872 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2873 			return 1;
2874 		msr_info->data = vcpu->arch.osvw.status;
2875 		break;
2876 	case MSR_PLATFORM_INFO:
2877 		if (!msr_info->host_initiated &&
2878 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2879 			return 1;
2880 		msr_info->data = vcpu->arch.msr_platform_info;
2881 		break;
2882 	case MSR_MISC_FEATURES_ENABLES:
2883 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2884 		break;
2885 	default:
2886 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2887 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2888 		if (!ignore_msrs) {
2889 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2890 					       msr_info->index);
2891 			return 1;
2892 		} else {
2893 			if (report_ignored_msrs)
2894 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2895 					msr_info->index);
2896 			msr_info->data = 0;
2897 		}
2898 		break;
2899 	}
2900 	return 0;
2901 }
2902 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2903 
2904 /*
2905  * Read or write a bunch of msrs. All parameters are kernel addresses.
2906  *
2907  * @return number of msrs set successfully.
2908  */
2909 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2910 		    struct kvm_msr_entry *entries,
2911 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2912 				  unsigned index, u64 *data))
2913 {
2914 	int i;
2915 
2916 	for (i = 0; i < msrs->nmsrs; ++i)
2917 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2918 			break;
2919 
2920 	return i;
2921 }
2922 
2923 /*
2924  * Read or write a bunch of msrs. Parameters are user addresses.
2925  *
2926  * @return number of msrs set successfully.
2927  */
2928 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2929 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2930 				unsigned index, u64 *data),
2931 		  int writeback)
2932 {
2933 	struct kvm_msrs msrs;
2934 	struct kvm_msr_entry *entries;
2935 	int r, n;
2936 	unsigned size;
2937 
2938 	r = -EFAULT;
2939 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2940 		goto out;
2941 
2942 	r = -E2BIG;
2943 	if (msrs.nmsrs >= MAX_IO_MSRS)
2944 		goto out;
2945 
2946 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2947 	entries = memdup_user(user_msrs->entries, size);
2948 	if (IS_ERR(entries)) {
2949 		r = PTR_ERR(entries);
2950 		goto out;
2951 	}
2952 
2953 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2954 	if (r < 0)
2955 		goto out_free;
2956 
2957 	r = -EFAULT;
2958 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2959 		goto out_free;
2960 
2961 	r = n;
2962 
2963 out_free:
2964 	kfree(entries);
2965 out:
2966 	return r;
2967 }
2968 
2969 static inline bool kvm_can_mwait_in_guest(void)
2970 {
2971 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2972 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2973 		boot_cpu_has(X86_FEATURE_ARAT);
2974 }
2975 
2976 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2977 {
2978 	int r = 0;
2979 
2980 	switch (ext) {
2981 	case KVM_CAP_IRQCHIP:
2982 	case KVM_CAP_HLT:
2983 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2984 	case KVM_CAP_SET_TSS_ADDR:
2985 	case KVM_CAP_EXT_CPUID:
2986 	case KVM_CAP_EXT_EMUL_CPUID:
2987 	case KVM_CAP_CLOCKSOURCE:
2988 	case KVM_CAP_PIT:
2989 	case KVM_CAP_NOP_IO_DELAY:
2990 	case KVM_CAP_MP_STATE:
2991 	case KVM_CAP_SYNC_MMU:
2992 	case KVM_CAP_USER_NMI:
2993 	case KVM_CAP_REINJECT_CONTROL:
2994 	case KVM_CAP_IRQ_INJECT_STATUS:
2995 	case KVM_CAP_IOEVENTFD:
2996 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2997 	case KVM_CAP_PIT2:
2998 	case KVM_CAP_PIT_STATE2:
2999 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3000 	case KVM_CAP_XEN_HVM:
3001 	case KVM_CAP_VCPU_EVENTS:
3002 	case KVM_CAP_HYPERV:
3003 	case KVM_CAP_HYPERV_VAPIC:
3004 	case KVM_CAP_HYPERV_SPIN:
3005 	case KVM_CAP_HYPERV_SYNIC:
3006 	case KVM_CAP_HYPERV_SYNIC2:
3007 	case KVM_CAP_HYPERV_VP_INDEX:
3008 	case KVM_CAP_HYPERV_EVENTFD:
3009 	case KVM_CAP_HYPERV_TLBFLUSH:
3010 	case KVM_CAP_HYPERV_SEND_IPI:
3011 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3012 	case KVM_CAP_HYPERV_CPUID:
3013 	case KVM_CAP_PCI_SEGMENT:
3014 	case KVM_CAP_DEBUGREGS:
3015 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3016 	case KVM_CAP_XSAVE:
3017 	case KVM_CAP_ASYNC_PF:
3018 	case KVM_CAP_GET_TSC_KHZ:
3019 	case KVM_CAP_KVMCLOCK_CTRL:
3020 	case KVM_CAP_READONLY_MEM:
3021 	case KVM_CAP_HYPERV_TIME:
3022 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3023 	case KVM_CAP_TSC_DEADLINE_TIMER:
3024 	case KVM_CAP_DISABLE_QUIRKS:
3025 	case KVM_CAP_SET_BOOT_CPU_ID:
3026  	case KVM_CAP_SPLIT_IRQCHIP:
3027 	case KVM_CAP_IMMEDIATE_EXIT:
3028 	case KVM_CAP_GET_MSR_FEATURES:
3029 	case KVM_CAP_MSR_PLATFORM_INFO:
3030 	case KVM_CAP_EXCEPTION_PAYLOAD:
3031 		r = 1;
3032 		break;
3033 	case KVM_CAP_SYNC_REGS:
3034 		r = KVM_SYNC_X86_VALID_FIELDS;
3035 		break;
3036 	case KVM_CAP_ADJUST_CLOCK:
3037 		r = KVM_CLOCK_TSC_STABLE;
3038 		break;
3039 	case KVM_CAP_X86_DISABLE_EXITS:
3040 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3041 		if(kvm_can_mwait_in_guest())
3042 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3043 		break;
3044 	case KVM_CAP_X86_SMM:
3045 		/* SMBASE is usually relocated above 1M on modern chipsets,
3046 		 * and SMM handlers might indeed rely on 4G segment limits,
3047 		 * so do not report SMM to be available if real mode is
3048 		 * emulated via vm86 mode.  Still, do not go to great lengths
3049 		 * to avoid userspace's usage of the feature, because it is a
3050 		 * fringe case that is not enabled except via specific settings
3051 		 * of the module parameters.
3052 		 */
3053 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3054 		break;
3055 	case KVM_CAP_VAPIC:
3056 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3057 		break;
3058 	case KVM_CAP_NR_VCPUS:
3059 		r = KVM_SOFT_MAX_VCPUS;
3060 		break;
3061 	case KVM_CAP_MAX_VCPUS:
3062 		r = KVM_MAX_VCPUS;
3063 		break;
3064 	case KVM_CAP_NR_MEMSLOTS:
3065 		r = KVM_USER_MEM_SLOTS;
3066 		break;
3067 	case KVM_CAP_PV_MMU:	/* obsolete */
3068 		r = 0;
3069 		break;
3070 	case KVM_CAP_MCE:
3071 		r = KVM_MAX_MCE_BANKS;
3072 		break;
3073 	case KVM_CAP_XCRS:
3074 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3075 		break;
3076 	case KVM_CAP_TSC_CONTROL:
3077 		r = kvm_has_tsc_control;
3078 		break;
3079 	case KVM_CAP_X2APIC_API:
3080 		r = KVM_X2APIC_API_VALID_FLAGS;
3081 		break;
3082 	case KVM_CAP_NESTED_STATE:
3083 		r = kvm_x86_ops->get_nested_state ?
3084 			kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3085 		break;
3086 	default:
3087 		break;
3088 	}
3089 	return r;
3090 
3091 }
3092 
3093 long kvm_arch_dev_ioctl(struct file *filp,
3094 			unsigned int ioctl, unsigned long arg)
3095 {
3096 	void __user *argp = (void __user *)arg;
3097 	long r;
3098 
3099 	switch (ioctl) {
3100 	case KVM_GET_MSR_INDEX_LIST: {
3101 		struct kvm_msr_list __user *user_msr_list = argp;
3102 		struct kvm_msr_list msr_list;
3103 		unsigned n;
3104 
3105 		r = -EFAULT;
3106 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3107 			goto out;
3108 		n = msr_list.nmsrs;
3109 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3110 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3111 			goto out;
3112 		r = -E2BIG;
3113 		if (n < msr_list.nmsrs)
3114 			goto out;
3115 		r = -EFAULT;
3116 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3117 				 num_msrs_to_save * sizeof(u32)))
3118 			goto out;
3119 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3120 				 &emulated_msrs,
3121 				 num_emulated_msrs * sizeof(u32)))
3122 			goto out;
3123 		r = 0;
3124 		break;
3125 	}
3126 	case KVM_GET_SUPPORTED_CPUID:
3127 	case KVM_GET_EMULATED_CPUID: {
3128 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3129 		struct kvm_cpuid2 cpuid;
3130 
3131 		r = -EFAULT;
3132 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3133 			goto out;
3134 
3135 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3136 					    ioctl);
3137 		if (r)
3138 			goto out;
3139 
3140 		r = -EFAULT;
3141 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3142 			goto out;
3143 		r = 0;
3144 		break;
3145 	}
3146 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3147 		r = -EFAULT;
3148 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3149 				 sizeof(kvm_mce_cap_supported)))
3150 			goto out;
3151 		r = 0;
3152 		break;
3153 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3154 		struct kvm_msr_list __user *user_msr_list = argp;
3155 		struct kvm_msr_list msr_list;
3156 		unsigned int n;
3157 
3158 		r = -EFAULT;
3159 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3160 			goto out;
3161 		n = msr_list.nmsrs;
3162 		msr_list.nmsrs = num_msr_based_features;
3163 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3164 			goto out;
3165 		r = -E2BIG;
3166 		if (n < msr_list.nmsrs)
3167 			goto out;
3168 		r = -EFAULT;
3169 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3170 				 num_msr_based_features * sizeof(u32)))
3171 			goto out;
3172 		r = 0;
3173 		break;
3174 	}
3175 	case KVM_GET_MSRS:
3176 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3177 		break;
3178 	}
3179 	default:
3180 		r = -EINVAL;
3181 	}
3182 out:
3183 	return r;
3184 }
3185 
3186 static void wbinvd_ipi(void *garbage)
3187 {
3188 	wbinvd();
3189 }
3190 
3191 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3192 {
3193 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3194 }
3195 
3196 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3197 {
3198 	/* Address WBINVD may be executed by guest */
3199 	if (need_emulate_wbinvd(vcpu)) {
3200 		if (kvm_x86_ops->has_wbinvd_exit())
3201 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3202 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3203 			smp_call_function_single(vcpu->cpu,
3204 					wbinvd_ipi, NULL, 1);
3205 	}
3206 
3207 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3208 
3209 	/* Apply any externally detected TSC adjustments (due to suspend) */
3210 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3211 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3212 		vcpu->arch.tsc_offset_adjustment = 0;
3213 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3214 	}
3215 
3216 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3217 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3218 				rdtsc() - vcpu->arch.last_host_tsc;
3219 		if (tsc_delta < 0)
3220 			mark_tsc_unstable("KVM discovered backwards TSC");
3221 
3222 		if (kvm_check_tsc_unstable()) {
3223 			u64 offset = kvm_compute_tsc_offset(vcpu,
3224 						vcpu->arch.last_guest_tsc);
3225 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3226 			vcpu->arch.tsc_catchup = 1;
3227 		}
3228 
3229 		if (kvm_lapic_hv_timer_in_use(vcpu))
3230 			kvm_lapic_restart_hv_timer(vcpu);
3231 
3232 		/*
3233 		 * On a host with synchronized TSC, there is no need to update
3234 		 * kvmclock on vcpu->cpu migration
3235 		 */
3236 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3237 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3238 		if (vcpu->cpu != cpu)
3239 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3240 		vcpu->cpu = cpu;
3241 	}
3242 
3243 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3244 }
3245 
3246 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3247 {
3248 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3249 		return;
3250 
3251 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3252 
3253 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3254 			&vcpu->arch.st.steal.preempted,
3255 			offsetof(struct kvm_steal_time, preempted),
3256 			sizeof(vcpu->arch.st.steal.preempted));
3257 }
3258 
3259 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3260 {
3261 	int idx;
3262 
3263 	if (vcpu->preempted)
3264 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3265 
3266 	/*
3267 	 * Disable page faults because we're in atomic context here.
3268 	 * kvm_write_guest_offset_cached() would call might_fault()
3269 	 * that relies on pagefault_disable() to tell if there's a
3270 	 * bug. NOTE: the write to guest memory may not go through if
3271 	 * during postcopy live migration or if there's heavy guest
3272 	 * paging.
3273 	 */
3274 	pagefault_disable();
3275 	/*
3276 	 * kvm_memslots() will be called by
3277 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3278 	 */
3279 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3280 	kvm_steal_time_set_preempted(vcpu);
3281 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3282 	pagefault_enable();
3283 	kvm_x86_ops->vcpu_put(vcpu);
3284 	vcpu->arch.last_host_tsc = rdtsc();
3285 	/*
3286 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3287 	 * on every vmexit, but if not, we might have a stale dr6 from the
3288 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3289 	 */
3290 	set_debugreg(0, 6);
3291 }
3292 
3293 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3294 				    struct kvm_lapic_state *s)
3295 {
3296 	if (vcpu->arch.apicv_active)
3297 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3298 
3299 	return kvm_apic_get_state(vcpu, s);
3300 }
3301 
3302 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3303 				    struct kvm_lapic_state *s)
3304 {
3305 	int r;
3306 
3307 	r = kvm_apic_set_state(vcpu, s);
3308 	if (r)
3309 		return r;
3310 	update_cr8_intercept(vcpu);
3311 
3312 	return 0;
3313 }
3314 
3315 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3316 {
3317 	return (!lapic_in_kernel(vcpu) ||
3318 		kvm_apic_accept_pic_intr(vcpu));
3319 }
3320 
3321 /*
3322  * if userspace requested an interrupt window, check that the
3323  * interrupt window is open.
3324  *
3325  * No need to exit to userspace if we already have an interrupt queued.
3326  */
3327 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3328 {
3329 	return kvm_arch_interrupt_allowed(vcpu) &&
3330 		!kvm_cpu_has_interrupt(vcpu) &&
3331 		!kvm_event_needs_reinjection(vcpu) &&
3332 		kvm_cpu_accept_dm_intr(vcpu);
3333 }
3334 
3335 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3336 				    struct kvm_interrupt *irq)
3337 {
3338 	if (irq->irq >= KVM_NR_INTERRUPTS)
3339 		return -EINVAL;
3340 
3341 	if (!irqchip_in_kernel(vcpu->kvm)) {
3342 		kvm_queue_interrupt(vcpu, irq->irq, false);
3343 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3344 		return 0;
3345 	}
3346 
3347 	/*
3348 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3349 	 * fail for in-kernel 8259.
3350 	 */
3351 	if (pic_in_kernel(vcpu->kvm))
3352 		return -ENXIO;
3353 
3354 	if (vcpu->arch.pending_external_vector != -1)
3355 		return -EEXIST;
3356 
3357 	vcpu->arch.pending_external_vector = irq->irq;
3358 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3359 	return 0;
3360 }
3361 
3362 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3363 {
3364 	kvm_inject_nmi(vcpu);
3365 
3366 	return 0;
3367 }
3368 
3369 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3370 {
3371 	kvm_make_request(KVM_REQ_SMI, vcpu);
3372 
3373 	return 0;
3374 }
3375 
3376 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3377 					   struct kvm_tpr_access_ctl *tac)
3378 {
3379 	if (tac->flags)
3380 		return -EINVAL;
3381 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3382 	return 0;
3383 }
3384 
3385 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3386 					u64 mcg_cap)
3387 {
3388 	int r;
3389 	unsigned bank_num = mcg_cap & 0xff, bank;
3390 
3391 	r = -EINVAL;
3392 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3393 		goto out;
3394 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3395 		goto out;
3396 	r = 0;
3397 	vcpu->arch.mcg_cap = mcg_cap;
3398 	/* Init IA32_MCG_CTL to all 1s */
3399 	if (mcg_cap & MCG_CTL_P)
3400 		vcpu->arch.mcg_ctl = ~(u64)0;
3401 	/* Init IA32_MCi_CTL to all 1s */
3402 	for (bank = 0; bank < bank_num; bank++)
3403 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3404 
3405 	if (kvm_x86_ops->setup_mce)
3406 		kvm_x86_ops->setup_mce(vcpu);
3407 out:
3408 	return r;
3409 }
3410 
3411 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3412 				      struct kvm_x86_mce *mce)
3413 {
3414 	u64 mcg_cap = vcpu->arch.mcg_cap;
3415 	unsigned bank_num = mcg_cap & 0xff;
3416 	u64 *banks = vcpu->arch.mce_banks;
3417 
3418 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3419 		return -EINVAL;
3420 	/*
3421 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3422 	 * reporting is disabled
3423 	 */
3424 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3425 	    vcpu->arch.mcg_ctl != ~(u64)0)
3426 		return 0;
3427 	banks += 4 * mce->bank;
3428 	/*
3429 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3430 	 * reporting is disabled for the bank
3431 	 */
3432 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3433 		return 0;
3434 	if (mce->status & MCI_STATUS_UC) {
3435 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3436 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3437 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3438 			return 0;
3439 		}
3440 		if (banks[1] & MCI_STATUS_VAL)
3441 			mce->status |= MCI_STATUS_OVER;
3442 		banks[2] = mce->addr;
3443 		banks[3] = mce->misc;
3444 		vcpu->arch.mcg_status = mce->mcg_status;
3445 		banks[1] = mce->status;
3446 		kvm_queue_exception(vcpu, MC_VECTOR);
3447 	} else if (!(banks[1] & MCI_STATUS_VAL)
3448 		   || !(banks[1] & MCI_STATUS_UC)) {
3449 		if (banks[1] & MCI_STATUS_VAL)
3450 			mce->status |= MCI_STATUS_OVER;
3451 		banks[2] = mce->addr;
3452 		banks[3] = mce->misc;
3453 		banks[1] = mce->status;
3454 	} else
3455 		banks[1] |= MCI_STATUS_OVER;
3456 	return 0;
3457 }
3458 
3459 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3460 					       struct kvm_vcpu_events *events)
3461 {
3462 	process_nmi(vcpu);
3463 
3464 	/*
3465 	 * The API doesn't provide the instruction length for software
3466 	 * exceptions, so don't report them. As long as the guest RIP
3467 	 * isn't advanced, we should expect to encounter the exception
3468 	 * again.
3469 	 */
3470 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3471 		events->exception.injected = 0;
3472 		events->exception.pending = 0;
3473 	} else {
3474 		events->exception.injected = vcpu->arch.exception.injected;
3475 		events->exception.pending = vcpu->arch.exception.pending;
3476 		/*
3477 		 * For ABI compatibility, deliberately conflate
3478 		 * pending and injected exceptions when
3479 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3480 		 */
3481 		if (!vcpu->kvm->arch.exception_payload_enabled)
3482 			events->exception.injected |=
3483 				vcpu->arch.exception.pending;
3484 	}
3485 	events->exception.nr = vcpu->arch.exception.nr;
3486 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3487 	events->exception.error_code = vcpu->arch.exception.error_code;
3488 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3489 	events->exception_payload = vcpu->arch.exception.payload;
3490 
3491 	events->interrupt.injected =
3492 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3493 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3494 	events->interrupt.soft = 0;
3495 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3496 
3497 	events->nmi.injected = vcpu->arch.nmi_injected;
3498 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3499 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3500 	events->nmi.pad = 0;
3501 
3502 	events->sipi_vector = 0; /* never valid when reporting to user space */
3503 
3504 	events->smi.smm = is_smm(vcpu);
3505 	events->smi.pending = vcpu->arch.smi_pending;
3506 	events->smi.smm_inside_nmi =
3507 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3508 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3509 
3510 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3511 			 | KVM_VCPUEVENT_VALID_SHADOW
3512 			 | KVM_VCPUEVENT_VALID_SMM);
3513 	if (vcpu->kvm->arch.exception_payload_enabled)
3514 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3515 
3516 	memset(&events->reserved, 0, sizeof(events->reserved));
3517 }
3518 
3519 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3520 
3521 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3522 					      struct kvm_vcpu_events *events)
3523 {
3524 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3525 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3526 			      | KVM_VCPUEVENT_VALID_SHADOW
3527 			      | KVM_VCPUEVENT_VALID_SMM
3528 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3529 		return -EINVAL;
3530 
3531 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3532 		if (!vcpu->kvm->arch.exception_payload_enabled)
3533 			return -EINVAL;
3534 		if (events->exception.pending)
3535 			events->exception.injected = 0;
3536 		else
3537 			events->exception_has_payload = 0;
3538 	} else {
3539 		events->exception.pending = 0;
3540 		events->exception_has_payload = 0;
3541 	}
3542 
3543 	if ((events->exception.injected || events->exception.pending) &&
3544 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3545 		return -EINVAL;
3546 
3547 	/* INITs are latched while in SMM */
3548 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3549 	    (events->smi.smm || events->smi.pending) &&
3550 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3551 		return -EINVAL;
3552 
3553 	process_nmi(vcpu);
3554 	vcpu->arch.exception.injected = events->exception.injected;
3555 	vcpu->arch.exception.pending = events->exception.pending;
3556 	vcpu->arch.exception.nr = events->exception.nr;
3557 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3558 	vcpu->arch.exception.error_code = events->exception.error_code;
3559 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3560 	vcpu->arch.exception.payload = events->exception_payload;
3561 
3562 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3563 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3564 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3565 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3566 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3567 						  events->interrupt.shadow);
3568 
3569 	vcpu->arch.nmi_injected = events->nmi.injected;
3570 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3571 		vcpu->arch.nmi_pending = events->nmi.pending;
3572 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3573 
3574 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3575 	    lapic_in_kernel(vcpu))
3576 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3577 
3578 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3579 		u32 hflags = vcpu->arch.hflags;
3580 		if (events->smi.smm)
3581 			hflags |= HF_SMM_MASK;
3582 		else
3583 			hflags &= ~HF_SMM_MASK;
3584 		kvm_set_hflags(vcpu, hflags);
3585 
3586 		vcpu->arch.smi_pending = events->smi.pending;
3587 
3588 		if (events->smi.smm) {
3589 			if (events->smi.smm_inside_nmi)
3590 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3591 			else
3592 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3593 			if (lapic_in_kernel(vcpu)) {
3594 				if (events->smi.latched_init)
3595 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3596 				else
3597 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3598 			}
3599 		}
3600 	}
3601 
3602 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3603 
3604 	return 0;
3605 }
3606 
3607 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3608 					     struct kvm_debugregs *dbgregs)
3609 {
3610 	unsigned long val;
3611 
3612 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3613 	kvm_get_dr(vcpu, 6, &val);
3614 	dbgregs->dr6 = val;
3615 	dbgregs->dr7 = vcpu->arch.dr7;
3616 	dbgregs->flags = 0;
3617 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3618 }
3619 
3620 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3621 					    struct kvm_debugregs *dbgregs)
3622 {
3623 	if (dbgregs->flags)
3624 		return -EINVAL;
3625 
3626 	if (dbgregs->dr6 & ~0xffffffffull)
3627 		return -EINVAL;
3628 	if (dbgregs->dr7 & ~0xffffffffull)
3629 		return -EINVAL;
3630 
3631 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3632 	kvm_update_dr0123(vcpu);
3633 	vcpu->arch.dr6 = dbgregs->dr6;
3634 	kvm_update_dr6(vcpu);
3635 	vcpu->arch.dr7 = dbgregs->dr7;
3636 	kvm_update_dr7(vcpu);
3637 
3638 	return 0;
3639 }
3640 
3641 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3642 
3643 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3644 {
3645 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3646 	u64 xstate_bv = xsave->header.xfeatures;
3647 	u64 valid;
3648 
3649 	/*
3650 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3651 	 * leaves 0 and 1 in the loop below.
3652 	 */
3653 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3654 
3655 	/* Set XSTATE_BV */
3656 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3657 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3658 
3659 	/*
3660 	 * Copy each region from the possibly compacted offset to the
3661 	 * non-compacted offset.
3662 	 */
3663 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3664 	while (valid) {
3665 		u64 feature = valid & -valid;
3666 		int index = fls64(feature) - 1;
3667 		void *src = get_xsave_addr(xsave, feature);
3668 
3669 		if (src) {
3670 			u32 size, offset, ecx, edx;
3671 			cpuid_count(XSTATE_CPUID, index,
3672 				    &size, &offset, &ecx, &edx);
3673 			if (feature == XFEATURE_MASK_PKRU)
3674 				memcpy(dest + offset, &vcpu->arch.pkru,
3675 				       sizeof(vcpu->arch.pkru));
3676 			else
3677 				memcpy(dest + offset, src, size);
3678 
3679 		}
3680 
3681 		valid -= feature;
3682 	}
3683 }
3684 
3685 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3686 {
3687 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3688 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3689 	u64 valid;
3690 
3691 	/*
3692 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3693 	 * leaves 0 and 1 in the loop below.
3694 	 */
3695 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3696 
3697 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3698 	xsave->header.xfeatures = xstate_bv;
3699 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3700 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3701 
3702 	/*
3703 	 * Copy each region from the non-compacted offset to the
3704 	 * possibly compacted offset.
3705 	 */
3706 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3707 	while (valid) {
3708 		u64 feature = valid & -valid;
3709 		int index = fls64(feature) - 1;
3710 		void *dest = get_xsave_addr(xsave, feature);
3711 
3712 		if (dest) {
3713 			u32 size, offset, ecx, edx;
3714 			cpuid_count(XSTATE_CPUID, index,
3715 				    &size, &offset, &ecx, &edx);
3716 			if (feature == XFEATURE_MASK_PKRU)
3717 				memcpy(&vcpu->arch.pkru, src + offset,
3718 				       sizeof(vcpu->arch.pkru));
3719 			else
3720 				memcpy(dest, src + offset, size);
3721 		}
3722 
3723 		valid -= feature;
3724 	}
3725 }
3726 
3727 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3728 					 struct kvm_xsave *guest_xsave)
3729 {
3730 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3731 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3732 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3733 	} else {
3734 		memcpy(guest_xsave->region,
3735 			&vcpu->arch.guest_fpu->state.fxsave,
3736 			sizeof(struct fxregs_state));
3737 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3738 			XFEATURE_MASK_FPSSE;
3739 	}
3740 }
3741 
3742 #define XSAVE_MXCSR_OFFSET 24
3743 
3744 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3745 					struct kvm_xsave *guest_xsave)
3746 {
3747 	u64 xstate_bv =
3748 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3749 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3750 
3751 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3752 		/*
3753 		 * Here we allow setting states that are not present in
3754 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3755 		 * with old userspace.
3756 		 */
3757 		if (xstate_bv & ~kvm_supported_xcr0() ||
3758 			mxcsr & ~mxcsr_feature_mask)
3759 			return -EINVAL;
3760 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3761 	} else {
3762 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3763 			mxcsr & ~mxcsr_feature_mask)
3764 			return -EINVAL;
3765 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3766 			guest_xsave->region, sizeof(struct fxregs_state));
3767 	}
3768 	return 0;
3769 }
3770 
3771 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3772 					struct kvm_xcrs *guest_xcrs)
3773 {
3774 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3775 		guest_xcrs->nr_xcrs = 0;
3776 		return;
3777 	}
3778 
3779 	guest_xcrs->nr_xcrs = 1;
3780 	guest_xcrs->flags = 0;
3781 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3782 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3783 }
3784 
3785 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3786 				       struct kvm_xcrs *guest_xcrs)
3787 {
3788 	int i, r = 0;
3789 
3790 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3791 		return -EINVAL;
3792 
3793 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3794 		return -EINVAL;
3795 
3796 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3797 		/* Only support XCR0 currently */
3798 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3799 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3800 				guest_xcrs->xcrs[i].value);
3801 			break;
3802 		}
3803 	if (r)
3804 		r = -EINVAL;
3805 	return r;
3806 }
3807 
3808 /*
3809  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3810  * stopped by the hypervisor.  This function will be called from the host only.
3811  * EINVAL is returned when the host attempts to set the flag for a guest that
3812  * does not support pv clocks.
3813  */
3814 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3815 {
3816 	if (!vcpu->arch.pv_time_enabled)
3817 		return -EINVAL;
3818 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3819 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3820 	return 0;
3821 }
3822 
3823 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3824 				     struct kvm_enable_cap *cap)
3825 {
3826 	int r;
3827 	uint16_t vmcs_version;
3828 	void __user *user_ptr;
3829 
3830 	if (cap->flags)
3831 		return -EINVAL;
3832 
3833 	switch (cap->cap) {
3834 	case KVM_CAP_HYPERV_SYNIC2:
3835 		if (cap->args[0])
3836 			return -EINVAL;
3837 	case KVM_CAP_HYPERV_SYNIC:
3838 		if (!irqchip_in_kernel(vcpu->kvm))
3839 			return -EINVAL;
3840 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3841 					     KVM_CAP_HYPERV_SYNIC2);
3842 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3843 		if (!kvm_x86_ops->nested_enable_evmcs)
3844 			return -ENOTTY;
3845 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3846 		if (!r) {
3847 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
3848 			if (copy_to_user(user_ptr, &vmcs_version,
3849 					 sizeof(vmcs_version)))
3850 				r = -EFAULT;
3851 		}
3852 		return r;
3853 
3854 	default:
3855 		return -EINVAL;
3856 	}
3857 }
3858 
3859 long kvm_arch_vcpu_ioctl(struct file *filp,
3860 			 unsigned int ioctl, unsigned long arg)
3861 {
3862 	struct kvm_vcpu *vcpu = filp->private_data;
3863 	void __user *argp = (void __user *)arg;
3864 	int r;
3865 	union {
3866 		struct kvm_lapic_state *lapic;
3867 		struct kvm_xsave *xsave;
3868 		struct kvm_xcrs *xcrs;
3869 		void *buffer;
3870 	} u;
3871 
3872 	vcpu_load(vcpu);
3873 
3874 	u.buffer = NULL;
3875 	switch (ioctl) {
3876 	case KVM_GET_LAPIC: {
3877 		r = -EINVAL;
3878 		if (!lapic_in_kernel(vcpu))
3879 			goto out;
3880 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3881 
3882 		r = -ENOMEM;
3883 		if (!u.lapic)
3884 			goto out;
3885 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3886 		if (r)
3887 			goto out;
3888 		r = -EFAULT;
3889 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3890 			goto out;
3891 		r = 0;
3892 		break;
3893 	}
3894 	case KVM_SET_LAPIC: {
3895 		r = -EINVAL;
3896 		if (!lapic_in_kernel(vcpu))
3897 			goto out;
3898 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3899 		if (IS_ERR(u.lapic)) {
3900 			r = PTR_ERR(u.lapic);
3901 			goto out_nofree;
3902 		}
3903 
3904 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3905 		break;
3906 	}
3907 	case KVM_INTERRUPT: {
3908 		struct kvm_interrupt irq;
3909 
3910 		r = -EFAULT;
3911 		if (copy_from_user(&irq, argp, sizeof(irq)))
3912 			goto out;
3913 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3914 		break;
3915 	}
3916 	case KVM_NMI: {
3917 		r = kvm_vcpu_ioctl_nmi(vcpu);
3918 		break;
3919 	}
3920 	case KVM_SMI: {
3921 		r = kvm_vcpu_ioctl_smi(vcpu);
3922 		break;
3923 	}
3924 	case KVM_SET_CPUID: {
3925 		struct kvm_cpuid __user *cpuid_arg = argp;
3926 		struct kvm_cpuid cpuid;
3927 
3928 		r = -EFAULT;
3929 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3930 			goto out;
3931 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3932 		break;
3933 	}
3934 	case KVM_SET_CPUID2: {
3935 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3936 		struct kvm_cpuid2 cpuid;
3937 
3938 		r = -EFAULT;
3939 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3940 			goto out;
3941 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3942 					      cpuid_arg->entries);
3943 		break;
3944 	}
3945 	case KVM_GET_CPUID2: {
3946 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3947 		struct kvm_cpuid2 cpuid;
3948 
3949 		r = -EFAULT;
3950 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3951 			goto out;
3952 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3953 					      cpuid_arg->entries);
3954 		if (r)
3955 			goto out;
3956 		r = -EFAULT;
3957 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3958 			goto out;
3959 		r = 0;
3960 		break;
3961 	}
3962 	case KVM_GET_MSRS: {
3963 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3964 		r = msr_io(vcpu, argp, do_get_msr, 1);
3965 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3966 		break;
3967 	}
3968 	case KVM_SET_MSRS: {
3969 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3970 		r = msr_io(vcpu, argp, do_set_msr, 0);
3971 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3972 		break;
3973 	}
3974 	case KVM_TPR_ACCESS_REPORTING: {
3975 		struct kvm_tpr_access_ctl tac;
3976 
3977 		r = -EFAULT;
3978 		if (copy_from_user(&tac, argp, sizeof(tac)))
3979 			goto out;
3980 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3981 		if (r)
3982 			goto out;
3983 		r = -EFAULT;
3984 		if (copy_to_user(argp, &tac, sizeof(tac)))
3985 			goto out;
3986 		r = 0;
3987 		break;
3988 	};
3989 	case KVM_SET_VAPIC_ADDR: {
3990 		struct kvm_vapic_addr va;
3991 		int idx;
3992 
3993 		r = -EINVAL;
3994 		if (!lapic_in_kernel(vcpu))
3995 			goto out;
3996 		r = -EFAULT;
3997 		if (copy_from_user(&va, argp, sizeof(va)))
3998 			goto out;
3999 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4000 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4001 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4002 		break;
4003 	}
4004 	case KVM_X86_SETUP_MCE: {
4005 		u64 mcg_cap;
4006 
4007 		r = -EFAULT;
4008 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4009 			goto out;
4010 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4011 		break;
4012 	}
4013 	case KVM_X86_SET_MCE: {
4014 		struct kvm_x86_mce mce;
4015 
4016 		r = -EFAULT;
4017 		if (copy_from_user(&mce, argp, sizeof(mce)))
4018 			goto out;
4019 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4020 		break;
4021 	}
4022 	case KVM_GET_VCPU_EVENTS: {
4023 		struct kvm_vcpu_events events;
4024 
4025 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4026 
4027 		r = -EFAULT;
4028 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4029 			break;
4030 		r = 0;
4031 		break;
4032 	}
4033 	case KVM_SET_VCPU_EVENTS: {
4034 		struct kvm_vcpu_events events;
4035 
4036 		r = -EFAULT;
4037 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4038 			break;
4039 
4040 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4041 		break;
4042 	}
4043 	case KVM_GET_DEBUGREGS: {
4044 		struct kvm_debugregs dbgregs;
4045 
4046 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4047 
4048 		r = -EFAULT;
4049 		if (copy_to_user(argp, &dbgregs,
4050 				 sizeof(struct kvm_debugregs)))
4051 			break;
4052 		r = 0;
4053 		break;
4054 	}
4055 	case KVM_SET_DEBUGREGS: {
4056 		struct kvm_debugregs dbgregs;
4057 
4058 		r = -EFAULT;
4059 		if (copy_from_user(&dbgregs, argp,
4060 				   sizeof(struct kvm_debugregs)))
4061 			break;
4062 
4063 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4064 		break;
4065 	}
4066 	case KVM_GET_XSAVE: {
4067 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
4068 		r = -ENOMEM;
4069 		if (!u.xsave)
4070 			break;
4071 
4072 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4073 
4074 		r = -EFAULT;
4075 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4076 			break;
4077 		r = 0;
4078 		break;
4079 	}
4080 	case KVM_SET_XSAVE: {
4081 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4082 		if (IS_ERR(u.xsave)) {
4083 			r = PTR_ERR(u.xsave);
4084 			goto out_nofree;
4085 		}
4086 
4087 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4088 		break;
4089 	}
4090 	case KVM_GET_XCRS: {
4091 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
4092 		r = -ENOMEM;
4093 		if (!u.xcrs)
4094 			break;
4095 
4096 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4097 
4098 		r = -EFAULT;
4099 		if (copy_to_user(argp, u.xcrs,
4100 				 sizeof(struct kvm_xcrs)))
4101 			break;
4102 		r = 0;
4103 		break;
4104 	}
4105 	case KVM_SET_XCRS: {
4106 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4107 		if (IS_ERR(u.xcrs)) {
4108 			r = PTR_ERR(u.xcrs);
4109 			goto out_nofree;
4110 		}
4111 
4112 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4113 		break;
4114 	}
4115 	case KVM_SET_TSC_KHZ: {
4116 		u32 user_tsc_khz;
4117 
4118 		r = -EINVAL;
4119 		user_tsc_khz = (u32)arg;
4120 
4121 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4122 			goto out;
4123 
4124 		if (user_tsc_khz == 0)
4125 			user_tsc_khz = tsc_khz;
4126 
4127 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4128 			r = 0;
4129 
4130 		goto out;
4131 	}
4132 	case KVM_GET_TSC_KHZ: {
4133 		r = vcpu->arch.virtual_tsc_khz;
4134 		goto out;
4135 	}
4136 	case KVM_KVMCLOCK_CTRL: {
4137 		r = kvm_set_guest_paused(vcpu);
4138 		goto out;
4139 	}
4140 	case KVM_ENABLE_CAP: {
4141 		struct kvm_enable_cap cap;
4142 
4143 		r = -EFAULT;
4144 		if (copy_from_user(&cap, argp, sizeof(cap)))
4145 			goto out;
4146 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4147 		break;
4148 	}
4149 	case KVM_GET_NESTED_STATE: {
4150 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4151 		u32 user_data_size;
4152 
4153 		r = -EINVAL;
4154 		if (!kvm_x86_ops->get_nested_state)
4155 			break;
4156 
4157 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4158 		r = -EFAULT;
4159 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4160 			break;
4161 
4162 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4163 						  user_data_size);
4164 		if (r < 0)
4165 			break;
4166 
4167 		if (r > user_data_size) {
4168 			if (put_user(r, &user_kvm_nested_state->size))
4169 				r = -EFAULT;
4170 			else
4171 				r = -E2BIG;
4172 			break;
4173 		}
4174 
4175 		r = 0;
4176 		break;
4177 	}
4178 	case KVM_SET_NESTED_STATE: {
4179 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4180 		struct kvm_nested_state kvm_state;
4181 
4182 		r = -EINVAL;
4183 		if (!kvm_x86_ops->set_nested_state)
4184 			break;
4185 
4186 		r = -EFAULT;
4187 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4188 			break;
4189 
4190 		r = -EINVAL;
4191 		if (kvm_state.size < sizeof(kvm_state))
4192 			break;
4193 
4194 		if (kvm_state.flags &
4195 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4196 		      | KVM_STATE_NESTED_EVMCS))
4197 			break;
4198 
4199 		/* nested_run_pending implies guest_mode.  */
4200 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4201 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4202 			break;
4203 
4204 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4205 		break;
4206 	}
4207 	case KVM_GET_SUPPORTED_HV_CPUID: {
4208 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4209 		struct kvm_cpuid2 cpuid;
4210 
4211 		r = -EFAULT;
4212 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4213 			goto out;
4214 
4215 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4216 						cpuid_arg->entries);
4217 		if (r)
4218 			goto out;
4219 
4220 		r = -EFAULT;
4221 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4222 			goto out;
4223 		r = 0;
4224 		break;
4225 	}
4226 	default:
4227 		r = -EINVAL;
4228 	}
4229 out:
4230 	kfree(u.buffer);
4231 out_nofree:
4232 	vcpu_put(vcpu);
4233 	return r;
4234 }
4235 
4236 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4237 {
4238 	return VM_FAULT_SIGBUS;
4239 }
4240 
4241 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4242 {
4243 	int ret;
4244 
4245 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4246 		return -EINVAL;
4247 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4248 	return ret;
4249 }
4250 
4251 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4252 					      u64 ident_addr)
4253 {
4254 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4255 }
4256 
4257 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4258 					  u32 kvm_nr_mmu_pages)
4259 {
4260 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4261 		return -EINVAL;
4262 
4263 	mutex_lock(&kvm->slots_lock);
4264 
4265 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4266 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4267 
4268 	mutex_unlock(&kvm->slots_lock);
4269 	return 0;
4270 }
4271 
4272 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4273 {
4274 	return kvm->arch.n_max_mmu_pages;
4275 }
4276 
4277 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4278 {
4279 	struct kvm_pic *pic = kvm->arch.vpic;
4280 	int r;
4281 
4282 	r = 0;
4283 	switch (chip->chip_id) {
4284 	case KVM_IRQCHIP_PIC_MASTER:
4285 		memcpy(&chip->chip.pic, &pic->pics[0],
4286 			sizeof(struct kvm_pic_state));
4287 		break;
4288 	case KVM_IRQCHIP_PIC_SLAVE:
4289 		memcpy(&chip->chip.pic, &pic->pics[1],
4290 			sizeof(struct kvm_pic_state));
4291 		break;
4292 	case KVM_IRQCHIP_IOAPIC:
4293 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4294 		break;
4295 	default:
4296 		r = -EINVAL;
4297 		break;
4298 	}
4299 	return r;
4300 }
4301 
4302 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4303 {
4304 	struct kvm_pic *pic = kvm->arch.vpic;
4305 	int r;
4306 
4307 	r = 0;
4308 	switch (chip->chip_id) {
4309 	case KVM_IRQCHIP_PIC_MASTER:
4310 		spin_lock(&pic->lock);
4311 		memcpy(&pic->pics[0], &chip->chip.pic,
4312 			sizeof(struct kvm_pic_state));
4313 		spin_unlock(&pic->lock);
4314 		break;
4315 	case KVM_IRQCHIP_PIC_SLAVE:
4316 		spin_lock(&pic->lock);
4317 		memcpy(&pic->pics[1], &chip->chip.pic,
4318 			sizeof(struct kvm_pic_state));
4319 		spin_unlock(&pic->lock);
4320 		break;
4321 	case KVM_IRQCHIP_IOAPIC:
4322 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4323 		break;
4324 	default:
4325 		r = -EINVAL;
4326 		break;
4327 	}
4328 	kvm_pic_update_irq(pic);
4329 	return r;
4330 }
4331 
4332 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4333 {
4334 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4335 
4336 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4337 
4338 	mutex_lock(&kps->lock);
4339 	memcpy(ps, &kps->channels, sizeof(*ps));
4340 	mutex_unlock(&kps->lock);
4341 	return 0;
4342 }
4343 
4344 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4345 {
4346 	int i;
4347 	struct kvm_pit *pit = kvm->arch.vpit;
4348 
4349 	mutex_lock(&pit->pit_state.lock);
4350 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4351 	for (i = 0; i < 3; i++)
4352 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4353 	mutex_unlock(&pit->pit_state.lock);
4354 	return 0;
4355 }
4356 
4357 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4358 {
4359 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4360 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4361 		sizeof(ps->channels));
4362 	ps->flags = kvm->arch.vpit->pit_state.flags;
4363 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4364 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4365 	return 0;
4366 }
4367 
4368 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4369 {
4370 	int start = 0;
4371 	int i;
4372 	u32 prev_legacy, cur_legacy;
4373 	struct kvm_pit *pit = kvm->arch.vpit;
4374 
4375 	mutex_lock(&pit->pit_state.lock);
4376 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4377 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4378 	if (!prev_legacy && cur_legacy)
4379 		start = 1;
4380 	memcpy(&pit->pit_state.channels, &ps->channels,
4381 	       sizeof(pit->pit_state.channels));
4382 	pit->pit_state.flags = ps->flags;
4383 	for (i = 0; i < 3; i++)
4384 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4385 				   start && i == 0);
4386 	mutex_unlock(&pit->pit_state.lock);
4387 	return 0;
4388 }
4389 
4390 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4391 				 struct kvm_reinject_control *control)
4392 {
4393 	struct kvm_pit *pit = kvm->arch.vpit;
4394 
4395 	if (!pit)
4396 		return -ENXIO;
4397 
4398 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4399 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4400 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4401 	 */
4402 	mutex_lock(&pit->pit_state.lock);
4403 	kvm_pit_set_reinject(pit, control->pit_reinject);
4404 	mutex_unlock(&pit->pit_state.lock);
4405 
4406 	return 0;
4407 }
4408 
4409 /**
4410  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4411  * @kvm: kvm instance
4412  * @log: slot id and address to which we copy the log
4413  *
4414  * Steps 1-4 below provide general overview of dirty page logging. See
4415  * kvm_get_dirty_log_protect() function description for additional details.
4416  *
4417  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4418  * always flush the TLB (step 4) even if previous step failed  and the dirty
4419  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4420  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4421  * writes will be marked dirty for next log read.
4422  *
4423  *   1. Take a snapshot of the bit and clear it if needed.
4424  *   2. Write protect the corresponding page.
4425  *   3. Copy the snapshot to the userspace.
4426  *   4. Flush TLB's if needed.
4427  */
4428 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4429 {
4430 	bool flush = false;
4431 	int r;
4432 
4433 	mutex_lock(&kvm->slots_lock);
4434 
4435 	/*
4436 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4437 	 */
4438 	if (kvm_x86_ops->flush_log_dirty)
4439 		kvm_x86_ops->flush_log_dirty(kvm);
4440 
4441 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
4442 
4443 	/*
4444 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4445 	 * kvm_mmu_slot_remove_write_access().
4446 	 */
4447 	lockdep_assert_held(&kvm->slots_lock);
4448 	if (flush)
4449 		kvm_flush_remote_tlbs(kvm);
4450 
4451 	mutex_unlock(&kvm->slots_lock);
4452 	return r;
4453 }
4454 
4455 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4456 {
4457 	bool flush = false;
4458 	int r;
4459 
4460 	mutex_lock(&kvm->slots_lock);
4461 
4462 	/*
4463 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4464 	 */
4465 	if (kvm_x86_ops->flush_log_dirty)
4466 		kvm_x86_ops->flush_log_dirty(kvm);
4467 
4468 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4469 
4470 	/*
4471 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4472 	 * kvm_mmu_slot_remove_write_access().
4473 	 */
4474 	lockdep_assert_held(&kvm->slots_lock);
4475 	if (flush)
4476 		kvm_flush_remote_tlbs(kvm);
4477 
4478 	mutex_unlock(&kvm->slots_lock);
4479 	return r;
4480 }
4481 
4482 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4483 			bool line_status)
4484 {
4485 	if (!irqchip_in_kernel(kvm))
4486 		return -ENXIO;
4487 
4488 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4489 					irq_event->irq, irq_event->level,
4490 					line_status);
4491 	return 0;
4492 }
4493 
4494 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4495 			    struct kvm_enable_cap *cap)
4496 {
4497 	int r;
4498 
4499 	if (cap->flags)
4500 		return -EINVAL;
4501 
4502 	switch (cap->cap) {
4503 	case KVM_CAP_DISABLE_QUIRKS:
4504 		kvm->arch.disabled_quirks = cap->args[0];
4505 		r = 0;
4506 		break;
4507 	case KVM_CAP_SPLIT_IRQCHIP: {
4508 		mutex_lock(&kvm->lock);
4509 		r = -EINVAL;
4510 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4511 			goto split_irqchip_unlock;
4512 		r = -EEXIST;
4513 		if (irqchip_in_kernel(kvm))
4514 			goto split_irqchip_unlock;
4515 		if (kvm->created_vcpus)
4516 			goto split_irqchip_unlock;
4517 		r = kvm_setup_empty_irq_routing(kvm);
4518 		if (r)
4519 			goto split_irqchip_unlock;
4520 		/* Pairs with irqchip_in_kernel. */
4521 		smp_wmb();
4522 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4523 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4524 		r = 0;
4525 split_irqchip_unlock:
4526 		mutex_unlock(&kvm->lock);
4527 		break;
4528 	}
4529 	case KVM_CAP_X2APIC_API:
4530 		r = -EINVAL;
4531 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4532 			break;
4533 
4534 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4535 			kvm->arch.x2apic_format = true;
4536 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4537 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4538 
4539 		r = 0;
4540 		break;
4541 	case KVM_CAP_X86_DISABLE_EXITS:
4542 		r = -EINVAL;
4543 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4544 			break;
4545 
4546 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4547 			kvm_can_mwait_in_guest())
4548 			kvm->arch.mwait_in_guest = true;
4549 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4550 			kvm->arch.hlt_in_guest = true;
4551 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4552 			kvm->arch.pause_in_guest = true;
4553 		r = 0;
4554 		break;
4555 	case KVM_CAP_MSR_PLATFORM_INFO:
4556 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4557 		r = 0;
4558 		break;
4559 	case KVM_CAP_EXCEPTION_PAYLOAD:
4560 		kvm->arch.exception_payload_enabled = cap->args[0];
4561 		r = 0;
4562 		break;
4563 	default:
4564 		r = -EINVAL;
4565 		break;
4566 	}
4567 	return r;
4568 }
4569 
4570 long kvm_arch_vm_ioctl(struct file *filp,
4571 		       unsigned int ioctl, unsigned long arg)
4572 {
4573 	struct kvm *kvm = filp->private_data;
4574 	void __user *argp = (void __user *)arg;
4575 	int r = -ENOTTY;
4576 	/*
4577 	 * This union makes it completely explicit to gcc-3.x
4578 	 * that these two variables' stack usage should be
4579 	 * combined, not added together.
4580 	 */
4581 	union {
4582 		struct kvm_pit_state ps;
4583 		struct kvm_pit_state2 ps2;
4584 		struct kvm_pit_config pit_config;
4585 	} u;
4586 
4587 	switch (ioctl) {
4588 	case KVM_SET_TSS_ADDR:
4589 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4590 		break;
4591 	case KVM_SET_IDENTITY_MAP_ADDR: {
4592 		u64 ident_addr;
4593 
4594 		mutex_lock(&kvm->lock);
4595 		r = -EINVAL;
4596 		if (kvm->created_vcpus)
4597 			goto set_identity_unlock;
4598 		r = -EFAULT;
4599 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4600 			goto set_identity_unlock;
4601 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4602 set_identity_unlock:
4603 		mutex_unlock(&kvm->lock);
4604 		break;
4605 	}
4606 	case KVM_SET_NR_MMU_PAGES:
4607 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4608 		break;
4609 	case KVM_GET_NR_MMU_PAGES:
4610 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4611 		break;
4612 	case KVM_CREATE_IRQCHIP: {
4613 		mutex_lock(&kvm->lock);
4614 
4615 		r = -EEXIST;
4616 		if (irqchip_in_kernel(kvm))
4617 			goto create_irqchip_unlock;
4618 
4619 		r = -EINVAL;
4620 		if (kvm->created_vcpus)
4621 			goto create_irqchip_unlock;
4622 
4623 		r = kvm_pic_init(kvm);
4624 		if (r)
4625 			goto create_irqchip_unlock;
4626 
4627 		r = kvm_ioapic_init(kvm);
4628 		if (r) {
4629 			kvm_pic_destroy(kvm);
4630 			goto create_irqchip_unlock;
4631 		}
4632 
4633 		r = kvm_setup_default_irq_routing(kvm);
4634 		if (r) {
4635 			kvm_ioapic_destroy(kvm);
4636 			kvm_pic_destroy(kvm);
4637 			goto create_irqchip_unlock;
4638 		}
4639 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4640 		smp_wmb();
4641 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4642 	create_irqchip_unlock:
4643 		mutex_unlock(&kvm->lock);
4644 		break;
4645 	}
4646 	case KVM_CREATE_PIT:
4647 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4648 		goto create_pit;
4649 	case KVM_CREATE_PIT2:
4650 		r = -EFAULT;
4651 		if (copy_from_user(&u.pit_config, argp,
4652 				   sizeof(struct kvm_pit_config)))
4653 			goto out;
4654 	create_pit:
4655 		mutex_lock(&kvm->lock);
4656 		r = -EEXIST;
4657 		if (kvm->arch.vpit)
4658 			goto create_pit_unlock;
4659 		r = -ENOMEM;
4660 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4661 		if (kvm->arch.vpit)
4662 			r = 0;
4663 	create_pit_unlock:
4664 		mutex_unlock(&kvm->lock);
4665 		break;
4666 	case KVM_GET_IRQCHIP: {
4667 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4668 		struct kvm_irqchip *chip;
4669 
4670 		chip = memdup_user(argp, sizeof(*chip));
4671 		if (IS_ERR(chip)) {
4672 			r = PTR_ERR(chip);
4673 			goto out;
4674 		}
4675 
4676 		r = -ENXIO;
4677 		if (!irqchip_kernel(kvm))
4678 			goto get_irqchip_out;
4679 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4680 		if (r)
4681 			goto get_irqchip_out;
4682 		r = -EFAULT;
4683 		if (copy_to_user(argp, chip, sizeof(*chip)))
4684 			goto get_irqchip_out;
4685 		r = 0;
4686 	get_irqchip_out:
4687 		kfree(chip);
4688 		break;
4689 	}
4690 	case KVM_SET_IRQCHIP: {
4691 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4692 		struct kvm_irqchip *chip;
4693 
4694 		chip = memdup_user(argp, sizeof(*chip));
4695 		if (IS_ERR(chip)) {
4696 			r = PTR_ERR(chip);
4697 			goto out;
4698 		}
4699 
4700 		r = -ENXIO;
4701 		if (!irqchip_kernel(kvm))
4702 			goto set_irqchip_out;
4703 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4704 		if (r)
4705 			goto set_irqchip_out;
4706 		r = 0;
4707 	set_irqchip_out:
4708 		kfree(chip);
4709 		break;
4710 	}
4711 	case KVM_GET_PIT: {
4712 		r = -EFAULT;
4713 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4714 			goto out;
4715 		r = -ENXIO;
4716 		if (!kvm->arch.vpit)
4717 			goto out;
4718 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4719 		if (r)
4720 			goto out;
4721 		r = -EFAULT;
4722 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4723 			goto out;
4724 		r = 0;
4725 		break;
4726 	}
4727 	case KVM_SET_PIT: {
4728 		r = -EFAULT;
4729 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4730 			goto out;
4731 		r = -ENXIO;
4732 		if (!kvm->arch.vpit)
4733 			goto out;
4734 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4735 		break;
4736 	}
4737 	case KVM_GET_PIT2: {
4738 		r = -ENXIO;
4739 		if (!kvm->arch.vpit)
4740 			goto out;
4741 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4742 		if (r)
4743 			goto out;
4744 		r = -EFAULT;
4745 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4746 			goto out;
4747 		r = 0;
4748 		break;
4749 	}
4750 	case KVM_SET_PIT2: {
4751 		r = -EFAULT;
4752 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4753 			goto out;
4754 		r = -ENXIO;
4755 		if (!kvm->arch.vpit)
4756 			goto out;
4757 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4758 		break;
4759 	}
4760 	case KVM_REINJECT_CONTROL: {
4761 		struct kvm_reinject_control control;
4762 		r =  -EFAULT;
4763 		if (copy_from_user(&control, argp, sizeof(control)))
4764 			goto out;
4765 		r = kvm_vm_ioctl_reinject(kvm, &control);
4766 		break;
4767 	}
4768 	case KVM_SET_BOOT_CPU_ID:
4769 		r = 0;
4770 		mutex_lock(&kvm->lock);
4771 		if (kvm->created_vcpus)
4772 			r = -EBUSY;
4773 		else
4774 			kvm->arch.bsp_vcpu_id = arg;
4775 		mutex_unlock(&kvm->lock);
4776 		break;
4777 	case KVM_XEN_HVM_CONFIG: {
4778 		struct kvm_xen_hvm_config xhc;
4779 		r = -EFAULT;
4780 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4781 			goto out;
4782 		r = -EINVAL;
4783 		if (xhc.flags)
4784 			goto out;
4785 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4786 		r = 0;
4787 		break;
4788 	}
4789 	case KVM_SET_CLOCK: {
4790 		struct kvm_clock_data user_ns;
4791 		u64 now_ns;
4792 
4793 		r = -EFAULT;
4794 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4795 			goto out;
4796 
4797 		r = -EINVAL;
4798 		if (user_ns.flags)
4799 			goto out;
4800 
4801 		r = 0;
4802 		/*
4803 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4804 		 * kvm_gen_update_masterclock() can be cut down to locked
4805 		 * pvclock_update_vm_gtod_copy().
4806 		 */
4807 		kvm_gen_update_masterclock(kvm);
4808 		now_ns = get_kvmclock_ns(kvm);
4809 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4810 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4811 		break;
4812 	}
4813 	case KVM_GET_CLOCK: {
4814 		struct kvm_clock_data user_ns;
4815 		u64 now_ns;
4816 
4817 		now_ns = get_kvmclock_ns(kvm);
4818 		user_ns.clock = now_ns;
4819 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4820 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4821 
4822 		r = -EFAULT;
4823 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4824 			goto out;
4825 		r = 0;
4826 		break;
4827 	}
4828 	case KVM_MEMORY_ENCRYPT_OP: {
4829 		r = -ENOTTY;
4830 		if (kvm_x86_ops->mem_enc_op)
4831 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4832 		break;
4833 	}
4834 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4835 		struct kvm_enc_region region;
4836 
4837 		r = -EFAULT;
4838 		if (copy_from_user(&region, argp, sizeof(region)))
4839 			goto out;
4840 
4841 		r = -ENOTTY;
4842 		if (kvm_x86_ops->mem_enc_reg_region)
4843 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4844 		break;
4845 	}
4846 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4847 		struct kvm_enc_region region;
4848 
4849 		r = -EFAULT;
4850 		if (copy_from_user(&region, argp, sizeof(region)))
4851 			goto out;
4852 
4853 		r = -ENOTTY;
4854 		if (kvm_x86_ops->mem_enc_unreg_region)
4855 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4856 		break;
4857 	}
4858 	case KVM_HYPERV_EVENTFD: {
4859 		struct kvm_hyperv_eventfd hvevfd;
4860 
4861 		r = -EFAULT;
4862 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4863 			goto out;
4864 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4865 		break;
4866 	}
4867 	default:
4868 		r = -ENOTTY;
4869 	}
4870 out:
4871 	return r;
4872 }
4873 
4874 static void kvm_init_msr_list(void)
4875 {
4876 	u32 dummy[2];
4877 	unsigned i, j;
4878 
4879 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4880 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4881 			continue;
4882 
4883 		/*
4884 		 * Even MSRs that are valid in the host may not be exposed
4885 		 * to the guests in some cases.
4886 		 */
4887 		switch (msrs_to_save[i]) {
4888 		case MSR_IA32_BNDCFGS:
4889 			if (!kvm_mpx_supported())
4890 				continue;
4891 			break;
4892 		case MSR_TSC_AUX:
4893 			if (!kvm_x86_ops->rdtscp_supported())
4894 				continue;
4895 			break;
4896 		case MSR_IA32_RTIT_CTL:
4897 		case MSR_IA32_RTIT_STATUS:
4898 			if (!kvm_x86_ops->pt_supported())
4899 				continue;
4900 			break;
4901 		case MSR_IA32_RTIT_CR3_MATCH:
4902 			if (!kvm_x86_ops->pt_supported() ||
4903 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
4904 				continue;
4905 			break;
4906 		case MSR_IA32_RTIT_OUTPUT_BASE:
4907 		case MSR_IA32_RTIT_OUTPUT_MASK:
4908 			if (!kvm_x86_ops->pt_supported() ||
4909 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
4910 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
4911 				continue;
4912 			break;
4913 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
4914 			if (!kvm_x86_ops->pt_supported() ||
4915 				msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
4916 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
4917 				continue;
4918 			break;
4919 		}
4920 		default:
4921 			break;
4922 		}
4923 
4924 		if (j < i)
4925 			msrs_to_save[j] = msrs_to_save[i];
4926 		j++;
4927 	}
4928 	num_msrs_to_save = j;
4929 
4930 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4931 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4932 			continue;
4933 
4934 		if (j < i)
4935 			emulated_msrs[j] = emulated_msrs[i];
4936 		j++;
4937 	}
4938 	num_emulated_msrs = j;
4939 
4940 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4941 		struct kvm_msr_entry msr;
4942 
4943 		msr.index = msr_based_features[i];
4944 		if (kvm_get_msr_feature(&msr))
4945 			continue;
4946 
4947 		if (j < i)
4948 			msr_based_features[j] = msr_based_features[i];
4949 		j++;
4950 	}
4951 	num_msr_based_features = j;
4952 }
4953 
4954 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4955 			   const void *v)
4956 {
4957 	int handled = 0;
4958 	int n;
4959 
4960 	do {
4961 		n = min(len, 8);
4962 		if (!(lapic_in_kernel(vcpu) &&
4963 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4964 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4965 			break;
4966 		handled += n;
4967 		addr += n;
4968 		len -= n;
4969 		v += n;
4970 	} while (len);
4971 
4972 	return handled;
4973 }
4974 
4975 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4976 {
4977 	int handled = 0;
4978 	int n;
4979 
4980 	do {
4981 		n = min(len, 8);
4982 		if (!(lapic_in_kernel(vcpu) &&
4983 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4984 					 addr, n, v))
4985 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4986 			break;
4987 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4988 		handled += n;
4989 		addr += n;
4990 		len -= n;
4991 		v += n;
4992 	} while (len);
4993 
4994 	return handled;
4995 }
4996 
4997 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4998 			struct kvm_segment *var, int seg)
4999 {
5000 	kvm_x86_ops->set_segment(vcpu, var, seg);
5001 }
5002 
5003 void kvm_get_segment(struct kvm_vcpu *vcpu,
5004 		     struct kvm_segment *var, int seg)
5005 {
5006 	kvm_x86_ops->get_segment(vcpu, var, seg);
5007 }
5008 
5009 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5010 			   struct x86_exception *exception)
5011 {
5012 	gpa_t t_gpa;
5013 
5014 	BUG_ON(!mmu_is_nested(vcpu));
5015 
5016 	/* NPT walks are always user-walks */
5017 	access |= PFERR_USER_MASK;
5018 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5019 
5020 	return t_gpa;
5021 }
5022 
5023 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5024 			      struct x86_exception *exception)
5025 {
5026 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5027 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5028 }
5029 
5030  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5031 				struct x86_exception *exception)
5032 {
5033 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5034 	access |= PFERR_FETCH_MASK;
5035 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5036 }
5037 
5038 gpa_t kvm_mmu_gva_to_gpa_write(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 	access |= PFERR_WRITE_MASK;
5043 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5044 }
5045 
5046 /* uses this to access any guest's mapped memory without checking CPL */
5047 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5048 				struct x86_exception *exception)
5049 {
5050 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5051 }
5052 
5053 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5054 				      struct kvm_vcpu *vcpu, u32 access,
5055 				      struct x86_exception *exception)
5056 {
5057 	void *data = val;
5058 	int r = X86EMUL_CONTINUE;
5059 
5060 	while (bytes) {
5061 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5062 							    exception);
5063 		unsigned offset = addr & (PAGE_SIZE-1);
5064 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5065 		int ret;
5066 
5067 		if (gpa == UNMAPPED_GVA)
5068 			return X86EMUL_PROPAGATE_FAULT;
5069 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5070 					       offset, toread);
5071 		if (ret < 0) {
5072 			r = X86EMUL_IO_NEEDED;
5073 			goto out;
5074 		}
5075 
5076 		bytes -= toread;
5077 		data += toread;
5078 		addr += toread;
5079 	}
5080 out:
5081 	return r;
5082 }
5083 
5084 /* used for instruction fetching */
5085 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5086 				gva_t addr, void *val, unsigned int bytes,
5087 				struct x86_exception *exception)
5088 {
5089 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5090 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5091 	unsigned offset;
5092 	int ret;
5093 
5094 	/* Inline kvm_read_guest_virt_helper for speed.  */
5095 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5096 						    exception);
5097 	if (unlikely(gpa == UNMAPPED_GVA))
5098 		return X86EMUL_PROPAGATE_FAULT;
5099 
5100 	offset = addr & (PAGE_SIZE-1);
5101 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5102 		bytes = (unsigned)PAGE_SIZE - offset;
5103 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5104 				       offset, bytes);
5105 	if (unlikely(ret < 0))
5106 		return X86EMUL_IO_NEEDED;
5107 
5108 	return X86EMUL_CONTINUE;
5109 }
5110 
5111 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5112 			       gva_t addr, void *val, unsigned int bytes,
5113 			       struct x86_exception *exception)
5114 {
5115 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5116 
5117 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5118 					  exception);
5119 }
5120 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5121 
5122 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5123 			     gva_t addr, void *val, unsigned int bytes,
5124 			     struct x86_exception *exception, bool system)
5125 {
5126 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5127 	u32 access = 0;
5128 
5129 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5130 		access |= PFERR_USER_MASK;
5131 
5132 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5133 }
5134 
5135 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5136 		unsigned long addr, void *val, unsigned int bytes)
5137 {
5138 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5139 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5140 
5141 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5142 }
5143 
5144 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5145 				      struct kvm_vcpu *vcpu, u32 access,
5146 				      struct x86_exception *exception)
5147 {
5148 	void *data = val;
5149 	int r = X86EMUL_CONTINUE;
5150 
5151 	while (bytes) {
5152 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5153 							     access,
5154 							     exception);
5155 		unsigned offset = addr & (PAGE_SIZE-1);
5156 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5157 		int ret;
5158 
5159 		if (gpa == UNMAPPED_GVA)
5160 			return X86EMUL_PROPAGATE_FAULT;
5161 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5162 		if (ret < 0) {
5163 			r = X86EMUL_IO_NEEDED;
5164 			goto out;
5165 		}
5166 
5167 		bytes -= towrite;
5168 		data += towrite;
5169 		addr += towrite;
5170 	}
5171 out:
5172 	return r;
5173 }
5174 
5175 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5176 			      unsigned int bytes, struct x86_exception *exception,
5177 			      bool system)
5178 {
5179 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5180 	u32 access = PFERR_WRITE_MASK;
5181 
5182 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5183 		access |= PFERR_USER_MASK;
5184 
5185 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5186 					   access, exception);
5187 }
5188 
5189 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5190 				unsigned int bytes, struct x86_exception *exception)
5191 {
5192 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5193 	vcpu->arch.l1tf_flush_l1d = true;
5194 
5195 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5196 					   PFERR_WRITE_MASK, exception);
5197 }
5198 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5199 
5200 int handle_ud(struct kvm_vcpu *vcpu)
5201 {
5202 	int emul_type = EMULTYPE_TRAP_UD;
5203 	enum emulation_result er;
5204 	char sig[5]; /* ud2; .ascii "kvm" */
5205 	struct x86_exception e;
5206 
5207 	if (force_emulation_prefix &&
5208 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5209 				sig, sizeof(sig), &e) == 0 &&
5210 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5211 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5212 		emul_type = 0;
5213 	}
5214 
5215 	er = kvm_emulate_instruction(vcpu, emul_type);
5216 	if (er == EMULATE_USER_EXIT)
5217 		return 0;
5218 	if (er != EMULATE_DONE)
5219 		kvm_queue_exception(vcpu, UD_VECTOR);
5220 	return 1;
5221 }
5222 EXPORT_SYMBOL_GPL(handle_ud);
5223 
5224 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5225 			    gpa_t gpa, bool write)
5226 {
5227 	/* For APIC access vmexit */
5228 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5229 		return 1;
5230 
5231 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5232 		trace_vcpu_match_mmio(gva, gpa, write, true);
5233 		return 1;
5234 	}
5235 
5236 	return 0;
5237 }
5238 
5239 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5240 				gpa_t *gpa, struct x86_exception *exception,
5241 				bool write)
5242 {
5243 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5244 		| (write ? PFERR_WRITE_MASK : 0);
5245 
5246 	/*
5247 	 * currently PKRU is only applied to ept enabled guest so
5248 	 * there is no pkey in EPT page table for L1 guest or EPT
5249 	 * shadow page table for L2 guest.
5250 	 */
5251 	if (vcpu_match_mmio_gva(vcpu, gva)
5252 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5253 				 vcpu->arch.access, 0, access)) {
5254 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5255 					(gva & (PAGE_SIZE - 1));
5256 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5257 		return 1;
5258 	}
5259 
5260 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5261 
5262 	if (*gpa == UNMAPPED_GVA)
5263 		return -1;
5264 
5265 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5266 }
5267 
5268 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5269 			const void *val, int bytes)
5270 {
5271 	int ret;
5272 
5273 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5274 	if (ret < 0)
5275 		return 0;
5276 	kvm_page_track_write(vcpu, gpa, val, bytes);
5277 	return 1;
5278 }
5279 
5280 struct read_write_emulator_ops {
5281 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5282 				  int bytes);
5283 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5284 				  void *val, int bytes);
5285 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5286 			       int bytes, void *val);
5287 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5288 				    void *val, int bytes);
5289 	bool write;
5290 };
5291 
5292 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5293 {
5294 	if (vcpu->mmio_read_completed) {
5295 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5296 			       vcpu->mmio_fragments[0].gpa, val);
5297 		vcpu->mmio_read_completed = 0;
5298 		return 1;
5299 	}
5300 
5301 	return 0;
5302 }
5303 
5304 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5305 			void *val, int bytes)
5306 {
5307 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5308 }
5309 
5310 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5311 			 void *val, int bytes)
5312 {
5313 	return emulator_write_phys(vcpu, gpa, val, bytes);
5314 }
5315 
5316 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5317 {
5318 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5319 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5320 }
5321 
5322 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5323 			  void *val, int bytes)
5324 {
5325 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5326 	return X86EMUL_IO_NEEDED;
5327 }
5328 
5329 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5330 			   void *val, int bytes)
5331 {
5332 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5333 
5334 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5335 	return X86EMUL_CONTINUE;
5336 }
5337 
5338 static const struct read_write_emulator_ops read_emultor = {
5339 	.read_write_prepare = read_prepare,
5340 	.read_write_emulate = read_emulate,
5341 	.read_write_mmio = vcpu_mmio_read,
5342 	.read_write_exit_mmio = read_exit_mmio,
5343 };
5344 
5345 static const struct read_write_emulator_ops write_emultor = {
5346 	.read_write_emulate = write_emulate,
5347 	.read_write_mmio = write_mmio,
5348 	.read_write_exit_mmio = write_exit_mmio,
5349 	.write = true,
5350 };
5351 
5352 static int emulator_read_write_onepage(unsigned long addr, void *val,
5353 				       unsigned int bytes,
5354 				       struct x86_exception *exception,
5355 				       struct kvm_vcpu *vcpu,
5356 				       const struct read_write_emulator_ops *ops)
5357 {
5358 	gpa_t gpa;
5359 	int handled, ret;
5360 	bool write = ops->write;
5361 	struct kvm_mmio_fragment *frag;
5362 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5363 
5364 	/*
5365 	 * If the exit was due to a NPF we may already have a GPA.
5366 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5367 	 * Note, this cannot be used on string operations since string
5368 	 * operation using rep will only have the initial GPA from the NPF
5369 	 * occurred.
5370 	 */
5371 	if (vcpu->arch.gpa_available &&
5372 	    emulator_can_use_gpa(ctxt) &&
5373 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5374 		gpa = vcpu->arch.gpa_val;
5375 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5376 	} else {
5377 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5378 		if (ret < 0)
5379 			return X86EMUL_PROPAGATE_FAULT;
5380 	}
5381 
5382 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5383 		return X86EMUL_CONTINUE;
5384 
5385 	/*
5386 	 * Is this MMIO handled locally?
5387 	 */
5388 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5389 	if (handled == bytes)
5390 		return X86EMUL_CONTINUE;
5391 
5392 	gpa += handled;
5393 	bytes -= handled;
5394 	val += handled;
5395 
5396 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5397 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5398 	frag->gpa = gpa;
5399 	frag->data = val;
5400 	frag->len = bytes;
5401 	return X86EMUL_CONTINUE;
5402 }
5403 
5404 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5405 			unsigned long addr,
5406 			void *val, unsigned int bytes,
5407 			struct x86_exception *exception,
5408 			const struct read_write_emulator_ops *ops)
5409 {
5410 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5411 	gpa_t gpa;
5412 	int rc;
5413 
5414 	if (ops->read_write_prepare &&
5415 		  ops->read_write_prepare(vcpu, val, bytes))
5416 		return X86EMUL_CONTINUE;
5417 
5418 	vcpu->mmio_nr_fragments = 0;
5419 
5420 	/* Crossing a page boundary? */
5421 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5422 		int now;
5423 
5424 		now = -addr & ~PAGE_MASK;
5425 		rc = emulator_read_write_onepage(addr, val, now, exception,
5426 						 vcpu, ops);
5427 
5428 		if (rc != X86EMUL_CONTINUE)
5429 			return rc;
5430 		addr += now;
5431 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5432 			addr = (u32)addr;
5433 		val += now;
5434 		bytes -= now;
5435 	}
5436 
5437 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5438 					 vcpu, ops);
5439 	if (rc != X86EMUL_CONTINUE)
5440 		return rc;
5441 
5442 	if (!vcpu->mmio_nr_fragments)
5443 		return rc;
5444 
5445 	gpa = vcpu->mmio_fragments[0].gpa;
5446 
5447 	vcpu->mmio_needed = 1;
5448 	vcpu->mmio_cur_fragment = 0;
5449 
5450 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5451 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5452 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5453 	vcpu->run->mmio.phys_addr = gpa;
5454 
5455 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5456 }
5457 
5458 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5459 				  unsigned long addr,
5460 				  void *val,
5461 				  unsigned int bytes,
5462 				  struct x86_exception *exception)
5463 {
5464 	return emulator_read_write(ctxt, addr, val, bytes,
5465 				   exception, &read_emultor);
5466 }
5467 
5468 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5469 			    unsigned long addr,
5470 			    const void *val,
5471 			    unsigned int bytes,
5472 			    struct x86_exception *exception)
5473 {
5474 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5475 				   exception, &write_emultor);
5476 }
5477 
5478 #define CMPXCHG_TYPE(t, ptr, old, new) \
5479 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5480 
5481 #ifdef CONFIG_X86_64
5482 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5483 #else
5484 #  define CMPXCHG64(ptr, old, new) \
5485 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5486 #endif
5487 
5488 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5489 				     unsigned long addr,
5490 				     const void *old,
5491 				     const void *new,
5492 				     unsigned int bytes,
5493 				     struct x86_exception *exception)
5494 {
5495 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5496 	gpa_t gpa;
5497 	struct page *page;
5498 	char *kaddr;
5499 	bool exchanged;
5500 
5501 	/* guests cmpxchg8b have to be emulated atomically */
5502 	if (bytes > 8 || (bytes & (bytes - 1)))
5503 		goto emul_write;
5504 
5505 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5506 
5507 	if (gpa == UNMAPPED_GVA ||
5508 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5509 		goto emul_write;
5510 
5511 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5512 		goto emul_write;
5513 
5514 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5515 	if (is_error_page(page))
5516 		goto emul_write;
5517 
5518 	kaddr = kmap_atomic(page);
5519 	kaddr += offset_in_page(gpa);
5520 	switch (bytes) {
5521 	case 1:
5522 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5523 		break;
5524 	case 2:
5525 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5526 		break;
5527 	case 4:
5528 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5529 		break;
5530 	case 8:
5531 		exchanged = CMPXCHG64(kaddr, old, new);
5532 		break;
5533 	default:
5534 		BUG();
5535 	}
5536 	kunmap_atomic(kaddr);
5537 	kvm_release_page_dirty(page);
5538 
5539 	if (!exchanged)
5540 		return X86EMUL_CMPXCHG_FAILED;
5541 
5542 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5543 	kvm_page_track_write(vcpu, gpa, new, bytes);
5544 
5545 	return X86EMUL_CONTINUE;
5546 
5547 emul_write:
5548 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5549 
5550 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5551 }
5552 
5553 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5554 {
5555 	int r = 0, i;
5556 
5557 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5558 		if (vcpu->arch.pio.in)
5559 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5560 					    vcpu->arch.pio.size, pd);
5561 		else
5562 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5563 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5564 					     pd);
5565 		if (r)
5566 			break;
5567 		pd += vcpu->arch.pio.size;
5568 	}
5569 	return r;
5570 }
5571 
5572 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5573 			       unsigned short port, void *val,
5574 			       unsigned int count, bool in)
5575 {
5576 	vcpu->arch.pio.port = port;
5577 	vcpu->arch.pio.in = in;
5578 	vcpu->arch.pio.count  = count;
5579 	vcpu->arch.pio.size = size;
5580 
5581 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5582 		vcpu->arch.pio.count = 0;
5583 		return 1;
5584 	}
5585 
5586 	vcpu->run->exit_reason = KVM_EXIT_IO;
5587 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5588 	vcpu->run->io.size = size;
5589 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5590 	vcpu->run->io.count = count;
5591 	vcpu->run->io.port = port;
5592 
5593 	return 0;
5594 }
5595 
5596 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5597 				    int size, unsigned short port, void *val,
5598 				    unsigned int count)
5599 {
5600 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5601 	int ret;
5602 
5603 	if (vcpu->arch.pio.count)
5604 		goto data_avail;
5605 
5606 	memset(vcpu->arch.pio_data, 0, size * count);
5607 
5608 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5609 	if (ret) {
5610 data_avail:
5611 		memcpy(val, vcpu->arch.pio_data, size * count);
5612 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5613 		vcpu->arch.pio.count = 0;
5614 		return 1;
5615 	}
5616 
5617 	return 0;
5618 }
5619 
5620 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5621 				     int size, unsigned short port,
5622 				     const void *val, unsigned int count)
5623 {
5624 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5625 
5626 	memcpy(vcpu->arch.pio_data, val, size * count);
5627 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5628 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5629 }
5630 
5631 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5632 {
5633 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5634 }
5635 
5636 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5637 {
5638 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5639 }
5640 
5641 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5642 {
5643 	if (!need_emulate_wbinvd(vcpu))
5644 		return X86EMUL_CONTINUE;
5645 
5646 	if (kvm_x86_ops->has_wbinvd_exit()) {
5647 		int cpu = get_cpu();
5648 
5649 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5650 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5651 				wbinvd_ipi, NULL, 1);
5652 		put_cpu();
5653 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5654 	} else
5655 		wbinvd();
5656 	return X86EMUL_CONTINUE;
5657 }
5658 
5659 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5660 {
5661 	kvm_emulate_wbinvd_noskip(vcpu);
5662 	return kvm_skip_emulated_instruction(vcpu);
5663 }
5664 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5665 
5666 
5667 
5668 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5669 {
5670 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5671 }
5672 
5673 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5674 			   unsigned long *dest)
5675 {
5676 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5677 }
5678 
5679 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5680 			   unsigned long value)
5681 {
5682 
5683 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5684 }
5685 
5686 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5687 {
5688 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5689 }
5690 
5691 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5692 {
5693 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5694 	unsigned long value;
5695 
5696 	switch (cr) {
5697 	case 0:
5698 		value = kvm_read_cr0(vcpu);
5699 		break;
5700 	case 2:
5701 		value = vcpu->arch.cr2;
5702 		break;
5703 	case 3:
5704 		value = kvm_read_cr3(vcpu);
5705 		break;
5706 	case 4:
5707 		value = kvm_read_cr4(vcpu);
5708 		break;
5709 	case 8:
5710 		value = kvm_get_cr8(vcpu);
5711 		break;
5712 	default:
5713 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5714 		return 0;
5715 	}
5716 
5717 	return value;
5718 }
5719 
5720 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5721 {
5722 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5723 	int res = 0;
5724 
5725 	switch (cr) {
5726 	case 0:
5727 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5728 		break;
5729 	case 2:
5730 		vcpu->arch.cr2 = val;
5731 		break;
5732 	case 3:
5733 		res = kvm_set_cr3(vcpu, val);
5734 		break;
5735 	case 4:
5736 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5737 		break;
5738 	case 8:
5739 		res = kvm_set_cr8(vcpu, val);
5740 		break;
5741 	default:
5742 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5743 		res = -1;
5744 	}
5745 
5746 	return res;
5747 }
5748 
5749 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5750 {
5751 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5752 }
5753 
5754 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5755 {
5756 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5757 }
5758 
5759 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5760 {
5761 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5762 }
5763 
5764 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5765 {
5766 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5767 }
5768 
5769 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5770 {
5771 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5772 }
5773 
5774 static unsigned long emulator_get_cached_segment_base(
5775 	struct x86_emulate_ctxt *ctxt, int seg)
5776 {
5777 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5778 }
5779 
5780 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5781 				 struct desc_struct *desc, u32 *base3,
5782 				 int seg)
5783 {
5784 	struct kvm_segment var;
5785 
5786 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5787 	*selector = var.selector;
5788 
5789 	if (var.unusable) {
5790 		memset(desc, 0, sizeof(*desc));
5791 		if (base3)
5792 			*base3 = 0;
5793 		return false;
5794 	}
5795 
5796 	if (var.g)
5797 		var.limit >>= 12;
5798 	set_desc_limit(desc, var.limit);
5799 	set_desc_base(desc, (unsigned long)var.base);
5800 #ifdef CONFIG_X86_64
5801 	if (base3)
5802 		*base3 = var.base >> 32;
5803 #endif
5804 	desc->type = var.type;
5805 	desc->s = var.s;
5806 	desc->dpl = var.dpl;
5807 	desc->p = var.present;
5808 	desc->avl = var.avl;
5809 	desc->l = var.l;
5810 	desc->d = var.db;
5811 	desc->g = var.g;
5812 
5813 	return true;
5814 }
5815 
5816 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5817 				 struct desc_struct *desc, u32 base3,
5818 				 int seg)
5819 {
5820 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5821 	struct kvm_segment var;
5822 
5823 	var.selector = selector;
5824 	var.base = get_desc_base(desc);
5825 #ifdef CONFIG_X86_64
5826 	var.base |= ((u64)base3) << 32;
5827 #endif
5828 	var.limit = get_desc_limit(desc);
5829 	if (desc->g)
5830 		var.limit = (var.limit << 12) | 0xfff;
5831 	var.type = desc->type;
5832 	var.dpl = desc->dpl;
5833 	var.db = desc->d;
5834 	var.s = desc->s;
5835 	var.l = desc->l;
5836 	var.g = desc->g;
5837 	var.avl = desc->avl;
5838 	var.present = desc->p;
5839 	var.unusable = !var.present;
5840 	var.padding = 0;
5841 
5842 	kvm_set_segment(vcpu, &var, seg);
5843 	return;
5844 }
5845 
5846 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5847 			    u32 msr_index, u64 *pdata)
5848 {
5849 	struct msr_data msr;
5850 	int r;
5851 
5852 	msr.index = msr_index;
5853 	msr.host_initiated = false;
5854 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5855 	if (r)
5856 		return r;
5857 
5858 	*pdata = msr.data;
5859 	return 0;
5860 }
5861 
5862 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5863 			    u32 msr_index, u64 data)
5864 {
5865 	struct msr_data msr;
5866 
5867 	msr.data = data;
5868 	msr.index = msr_index;
5869 	msr.host_initiated = false;
5870 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5871 }
5872 
5873 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5874 {
5875 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5876 
5877 	return vcpu->arch.smbase;
5878 }
5879 
5880 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5881 {
5882 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5883 
5884 	vcpu->arch.smbase = smbase;
5885 }
5886 
5887 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5888 			      u32 pmc)
5889 {
5890 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5891 }
5892 
5893 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5894 			     u32 pmc, u64 *pdata)
5895 {
5896 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5897 }
5898 
5899 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5900 {
5901 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5902 }
5903 
5904 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5905 			      struct x86_instruction_info *info,
5906 			      enum x86_intercept_stage stage)
5907 {
5908 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5909 }
5910 
5911 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5912 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5913 {
5914 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5915 }
5916 
5917 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5918 {
5919 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5920 }
5921 
5922 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5923 {
5924 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5925 }
5926 
5927 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5928 {
5929 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5930 }
5931 
5932 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5933 {
5934 	return emul_to_vcpu(ctxt)->arch.hflags;
5935 }
5936 
5937 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5938 {
5939 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5940 }
5941 
5942 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5943 {
5944 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5945 }
5946 
5947 static const struct x86_emulate_ops emulate_ops = {
5948 	.read_gpr            = emulator_read_gpr,
5949 	.write_gpr           = emulator_write_gpr,
5950 	.read_std            = emulator_read_std,
5951 	.write_std           = emulator_write_std,
5952 	.read_phys           = kvm_read_guest_phys_system,
5953 	.fetch               = kvm_fetch_guest_virt,
5954 	.read_emulated       = emulator_read_emulated,
5955 	.write_emulated      = emulator_write_emulated,
5956 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5957 	.invlpg              = emulator_invlpg,
5958 	.pio_in_emulated     = emulator_pio_in_emulated,
5959 	.pio_out_emulated    = emulator_pio_out_emulated,
5960 	.get_segment         = emulator_get_segment,
5961 	.set_segment         = emulator_set_segment,
5962 	.get_cached_segment_base = emulator_get_cached_segment_base,
5963 	.get_gdt             = emulator_get_gdt,
5964 	.get_idt	     = emulator_get_idt,
5965 	.set_gdt             = emulator_set_gdt,
5966 	.set_idt	     = emulator_set_idt,
5967 	.get_cr              = emulator_get_cr,
5968 	.set_cr              = emulator_set_cr,
5969 	.cpl                 = emulator_get_cpl,
5970 	.get_dr              = emulator_get_dr,
5971 	.set_dr              = emulator_set_dr,
5972 	.get_smbase          = emulator_get_smbase,
5973 	.set_smbase          = emulator_set_smbase,
5974 	.set_msr             = emulator_set_msr,
5975 	.get_msr             = emulator_get_msr,
5976 	.check_pmc	     = emulator_check_pmc,
5977 	.read_pmc            = emulator_read_pmc,
5978 	.halt                = emulator_halt,
5979 	.wbinvd              = emulator_wbinvd,
5980 	.fix_hypercall       = emulator_fix_hypercall,
5981 	.intercept           = emulator_intercept,
5982 	.get_cpuid           = emulator_get_cpuid,
5983 	.set_nmi_mask        = emulator_set_nmi_mask,
5984 	.get_hflags          = emulator_get_hflags,
5985 	.set_hflags          = emulator_set_hflags,
5986 	.pre_leave_smm       = emulator_pre_leave_smm,
5987 };
5988 
5989 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5990 {
5991 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5992 	/*
5993 	 * an sti; sti; sequence only disable interrupts for the first
5994 	 * instruction. So, if the last instruction, be it emulated or
5995 	 * not, left the system with the INT_STI flag enabled, it
5996 	 * means that the last instruction is an sti. We should not
5997 	 * leave the flag on in this case. The same goes for mov ss
5998 	 */
5999 	if (int_shadow & mask)
6000 		mask = 0;
6001 	if (unlikely(int_shadow || mask)) {
6002 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6003 		if (!mask)
6004 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6005 	}
6006 }
6007 
6008 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6009 {
6010 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6011 	if (ctxt->exception.vector == PF_VECTOR)
6012 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6013 
6014 	if (ctxt->exception.error_code_valid)
6015 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6016 				      ctxt->exception.error_code);
6017 	else
6018 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6019 	return false;
6020 }
6021 
6022 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6023 {
6024 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6025 	int cs_db, cs_l;
6026 
6027 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6028 
6029 	ctxt->eflags = kvm_get_rflags(vcpu);
6030 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6031 
6032 	ctxt->eip = kvm_rip_read(vcpu);
6033 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6034 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6035 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6036 		     cs_db				? X86EMUL_MODE_PROT32 :
6037 							  X86EMUL_MODE_PROT16;
6038 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6039 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6040 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6041 
6042 	init_decode_cache(ctxt);
6043 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6044 }
6045 
6046 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6047 {
6048 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6049 	int ret;
6050 
6051 	init_emulate_ctxt(vcpu);
6052 
6053 	ctxt->op_bytes = 2;
6054 	ctxt->ad_bytes = 2;
6055 	ctxt->_eip = ctxt->eip + inc_eip;
6056 	ret = emulate_int_real(ctxt, irq);
6057 
6058 	if (ret != X86EMUL_CONTINUE)
6059 		return EMULATE_FAIL;
6060 
6061 	ctxt->eip = ctxt->_eip;
6062 	kvm_rip_write(vcpu, ctxt->eip);
6063 	kvm_set_rflags(vcpu, ctxt->eflags);
6064 
6065 	return EMULATE_DONE;
6066 }
6067 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6068 
6069 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6070 {
6071 	int r = EMULATE_DONE;
6072 
6073 	++vcpu->stat.insn_emulation_fail;
6074 	trace_kvm_emulate_insn_failed(vcpu);
6075 
6076 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6077 		return EMULATE_FAIL;
6078 
6079 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6080 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6081 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6082 		vcpu->run->internal.ndata = 0;
6083 		r = EMULATE_USER_EXIT;
6084 	}
6085 
6086 	kvm_queue_exception(vcpu, UD_VECTOR);
6087 
6088 	return r;
6089 }
6090 
6091 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6092 				  bool write_fault_to_shadow_pgtable,
6093 				  int emulation_type)
6094 {
6095 	gpa_t gpa = cr2;
6096 	kvm_pfn_t pfn;
6097 
6098 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6099 		return false;
6100 
6101 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6102 		return false;
6103 
6104 	if (!vcpu->arch.mmu->direct_map) {
6105 		/*
6106 		 * Write permission should be allowed since only
6107 		 * write access need to be emulated.
6108 		 */
6109 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6110 
6111 		/*
6112 		 * If the mapping is invalid in guest, let cpu retry
6113 		 * it to generate fault.
6114 		 */
6115 		if (gpa == UNMAPPED_GVA)
6116 			return true;
6117 	}
6118 
6119 	/*
6120 	 * Do not retry the unhandleable instruction if it faults on the
6121 	 * readonly host memory, otherwise it will goto a infinite loop:
6122 	 * retry instruction -> write #PF -> emulation fail -> retry
6123 	 * instruction -> ...
6124 	 */
6125 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6126 
6127 	/*
6128 	 * If the instruction failed on the error pfn, it can not be fixed,
6129 	 * report the error to userspace.
6130 	 */
6131 	if (is_error_noslot_pfn(pfn))
6132 		return false;
6133 
6134 	kvm_release_pfn_clean(pfn);
6135 
6136 	/* The instructions are well-emulated on direct mmu. */
6137 	if (vcpu->arch.mmu->direct_map) {
6138 		unsigned int indirect_shadow_pages;
6139 
6140 		spin_lock(&vcpu->kvm->mmu_lock);
6141 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6142 		spin_unlock(&vcpu->kvm->mmu_lock);
6143 
6144 		if (indirect_shadow_pages)
6145 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6146 
6147 		return true;
6148 	}
6149 
6150 	/*
6151 	 * if emulation was due to access to shadowed page table
6152 	 * and it failed try to unshadow page and re-enter the
6153 	 * guest to let CPU execute the instruction.
6154 	 */
6155 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6156 
6157 	/*
6158 	 * If the access faults on its page table, it can not
6159 	 * be fixed by unprotecting shadow page and it should
6160 	 * be reported to userspace.
6161 	 */
6162 	return !write_fault_to_shadow_pgtable;
6163 }
6164 
6165 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6166 			      unsigned long cr2,  int emulation_type)
6167 {
6168 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6169 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6170 
6171 	last_retry_eip = vcpu->arch.last_retry_eip;
6172 	last_retry_addr = vcpu->arch.last_retry_addr;
6173 
6174 	/*
6175 	 * If the emulation is caused by #PF and it is non-page_table
6176 	 * writing instruction, it means the VM-EXIT is caused by shadow
6177 	 * page protected, we can zap the shadow page and retry this
6178 	 * instruction directly.
6179 	 *
6180 	 * Note: if the guest uses a non-page-table modifying instruction
6181 	 * on the PDE that points to the instruction, then we will unmap
6182 	 * the instruction and go to an infinite loop. So, we cache the
6183 	 * last retried eip and the last fault address, if we meet the eip
6184 	 * and the address again, we can break out of the potential infinite
6185 	 * loop.
6186 	 */
6187 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6188 
6189 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6190 		return false;
6191 
6192 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6193 		return false;
6194 
6195 	if (x86_page_table_writing_insn(ctxt))
6196 		return false;
6197 
6198 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6199 		return false;
6200 
6201 	vcpu->arch.last_retry_eip = ctxt->eip;
6202 	vcpu->arch.last_retry_addr = cr2;
6203 
6204 	if (!vcpu->arch.mmu->direct_map)
6205 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6206 
6207 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6208 
6209 	return true;
6210 }
6211 
6212 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6213 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6214 
6215 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6216 {
6217 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6218 		/* This is a good place to trace that we are exiting SMM.  */
6219 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6220 
6221 		/* Process a latched INIT or SMI, if any.  */
6222 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6223 	}
6224 
6225 	kvm_mmu_reset_context(vcpu);
6226 }
6227 
6228 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6229 {
6230 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
6231 
6232 	vcpu->arch.hflags = emul_flags;
6233 
6234 	if (changed & HF_SMM_MASK)
6235 		kvm_smm_changed(vcpu);
6236 }
6237 
6238 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6239 				unsigned long *db)
6240 {
6241 	u32 dr6 = 0;
6242 	int i;
6243 	u32 enable, rwlen;
6244 
6245 	enable = dr7;
6246 	rwlen = dr7 >> 16;
6247 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6248 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6249 			dr6 |= (1 << i);
6250 	return dr6;
6251 }
6252 
6253 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6254 {
6255 	struct kvm_run *kvm_run = vcpu->run;
6256 
6257 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6258 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6259 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6260 		kvm_run->debug.arch.exception = DB_VECTOR;
6261 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6262 		*r = EMULATE_USER_EXIT;
6263 	} else {
6264 		kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6265 	}
6266 }
6267 
6268 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6269 {
6270 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6271 	int r = EMULATE_DONE;
6272 
6273 	kvm_x86_ops->skip_emulated_instruction(vcpu);
6274 
6275 	/*
6276 	 * rflags is the old, "raw" value of the flags.  The new value has
6277 	 * not been saved yet.
6278 	 *
6279 	 * This is correct even for TF set by the guest, because "the
6280 	 * processor will not generate this exception after the instruction
6281 	 * that sets the TF flag".
6282 	 */
6283 	if (unlikely(rflags & X86_EFLAGS_TF))
6284 		kvm_vcpu_do_singlestep(vcpu, &r);
6285 	return r == EMULATE_DONE;
6286 }
6287 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6288 
6289 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6290 {
6291 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6292 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6293 		struct kvm_run *kvm_run = vcpu->run;
6294 		unsigned long eip = kvm_get_linear_rip(vcpu);
6295 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6296 					   vcpu->arch.guest_debug_dr7,
6297 					   vcpu->arch.eff_db);
6298 
6299 		if (dr6 != 0) {
6300 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6301 			kvm_run->debug.arch.pc = eip;
6302 			kvm_run->debug.arch.exception = DB_VECTOR;
6303 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6304 			*r = EMULATE_USER_EXIT;
6305 			return true;
6306 		}
6307 	}
6308 
6309 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6310 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6311 		unsigned long eip = kvm_get_linear_rip(vcpu);
6312 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6313 					   vcpu->arch.dr7,
6314 					   vcpu->arch.db);
6315 
6316 		if (dr6 != 0) {
6317 			vcpu->arch.dr6 &= ~15;
6318 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6319 			kvm_queue_exception(vcpu, DB_VECTOR);
6320 			*r = EMULATE_DONE;
6321 			return true;
6322 		}
6323 	}
6324 
6325 	return false;
6326 }
6327 
6328 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6329 {
6330 	switch (ctxt->opcode_len) {
6331 	case 1:
6332 		switch (ctxt->b) {
6333 		case 0xe4:	/* IN */
6334 		case 0xe5:
6335 		case 0xec:
6336 		case 0xed:
6337 		case 0xe6:	/* OUT */
6338 		case 0xe7:
6339 		case 0xee:
6340 		case 0xef:
6341 		case 0x6c:	/* INS */
6342 		case 0x6d:
6343 		case 0x6e:	/* OUTS */
6344 		case 0x6f:
6345 			return true;
6346 		}
6347 		break;
6348 	case 2:
6349 		switch (ctxt->b) {
6350 		case 0x33:	/* RDPMC */
6351 			return true;
6352 		}
6353 		break;
6354 	}
6355 
6356 	return false;
6357 }
6358 
6359 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6360 			    unsigned long cr2,
6361 			    int emulation_type,
6362 			    void *insn,
6363 			    int insn_len)
6364 {
6365 	int r;
6366 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6367 	bool writeback = true;
6368 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6369 
6370 	vcpu->arch.l1tf_flush_l1d = true;
6371 
6372 	/*
6373 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6374 	 * never reused.
6375 	 */
6376 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6377 	kvm_clear_exception_queue(vcpu);
6378 
6379 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6380 		init_emulate_ctxt(vcpu);
6381 
6382 		/*
6383 		 * We will reenter on the same instruction since
6384 		 * we do not set complete_userspace_io.  This does not
6385 		 * handle watchpoints yet, those would be handled in
6386 		 * the emulate_ops.
6387 		 */
6388 		if (!(emulation_type & EMULTYPE_SKIP) &&
6389 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6390 			return r;
6391 
6392 		ctxt->interruptibility = 0;
6393 		ctxt->have_exception = false;
6394 		ctxt->exception.vector = -1;
6395 		ctxt->perm_ok = false;
6396 
6397 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6398 
6399 		r = x86_decode_insn(ctxt, insn, insn_len);
6400 
6401 		trace_kvm_emulate_insn_start(vcpu);
6402 		++vcpu->stat.insn_emulation;
6403 		if (r != EMULATION_OK)  {
6404 			if (emulation_type & EMULTYPE_TRAP_UD)
6405 				return EMULATE_FAIL;
6406 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6407 						emulation_type))
6408 				return EMULATE_DONE;
6409 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6410 				return EMULATE_DONE;
6411 			if (emulation_type & EMULTYPE_SKIP)
6412 				return EMULATE_FAIL;
6413 			return handle_emulation_failure(vcpu, emulation_type);
6414 		}
6415 	}
6416 
6417 	if ((emulation_type & EMULTYPE_VMWARE) &&
6418 	    !is_vmware_backdoor_opcode(ctxt))
6419 		return EMULATE_FAIL;
6420 
6421 	if (emulation_type & EMULTYPE_SKIP) {
6422 		kvm_rip_write(vcpu, ctxt->_eip);
6423 		if (ctxt->eflags & X86_EFLAGS_RF)
6424 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6425 		return EMULATE_DONE;
6426 	}
6427 
6428 	if (retry_instruction(ctxt, cr2, emulation_type))
6429 		return EMULATE_DONE;
6430 
6431 	/* this is needed for vmware backdoor interface to work since it
6432 	   changes registers values  during IO operation */
6433 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6434 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6435 		emulator_invalidate_register_cache(ctxt);
6436 	}
6437 
6438 restart:
6439 	/* Save the faulting GPA (cr2) in the address field */
6440 	ctxt->exception.address = cr2;
6441 
6442 	r = x86_emulate_insn(ctxt);
6443 
6444 	if (r == EMULATION_INTERCEPTED)
6445 		return EMULATE_DONE;
6446 
6447 	if (r == EMULATION_FAILED) {
6448 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6449 					emulation_type))
6450 			return EMULATE_DONE;
6451 
6452 		return handle_emulation_failure(vcpu, emulation_type);
6453 	}
6454 
6455 	if (ctxt->have_exception) {
6456 		r = EMULATE_DONE;
6457 		if (inject_emulated_exception(vcpu))
6458 			return r;
6459 	} else if (vcpu->arch.pio.count) {
6460 		if (!vcpu->arch.pio.in) {
6461 			/* FIXME: return into emulator if single-stepping.  */
6462 			vcpu->arch.pio.count = 0;
6463 		} else {
6464 			writeback = false;
6465 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6466 		}
6467 		r = EMULATE_USER_EXIT;
6468 	} else if (vcpu->mmio_needed) {
6469 		if (!vcpu->mmio_is_write)
6470 			writeback = false;
6471 		r = EMULATE_USER_EXIT;
6472 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6473 	} else if (r == EMULATION_RESTART)
6474 		goto restart;
6475 	else
6476 		r = EMULATE_DONE;
6477 
6478 	if (writeback) {
6479 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6480 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6481 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6482 		kvm_rip_write(vcpu, ctxt->eip);
6483 		if (r == EMULATE_DONE &&
6484 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6485 			kvm_vcpu_do_singlestep(vcpu, &r);
6486 		if (!ctxt->have_exception ||
6487 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6488 			__kvm_set_rflags(vcpu, ctxt->eflags);
6489 
6490 		/*
6491 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6492 		 * do nothing, and it will be requested again as soon as
6493 		 * the shadow expires.  But we still need to check here,
6494 		 * because POPF has no interrupt shadow.
6495 		 */
6496 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6497 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6498 	} else
6499 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6500 
6501 	return r;
6502 }
6503 
6504 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6505 {
6506 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6507 }
6508 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6509 
6510 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6511 					void *insn, int insn_len)
6512 {
6513 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6514 }
6515 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6516 
6517 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6518 			    unsigned short port)
6519 {
6520 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6521 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6522 					    size, port, &val, 1);
6523 	/* do not return to emulator after return from userspace */
6524 	vcpu->arch.pio.count = 0;
6525 	return ret;
6526 }
6527 
6528 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6529 {
6530 	unsigned long val;
6531 
6532 	/* We should only ever be called with arch.pio.count equal to 1 */
6533 	BUG_ON(vcpu->arch.pio.count != 1);
6534 
6535 	/* For size less than 4 we merge, else we zero extend */
6536 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6537 					: 0;
6538 
6539 	/*
6540 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6541 	 * the copy and tracing
6542 	 */
6543 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6544 				 vcpu->arch.pio.port, &val, 1);
6545 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6546 
6547 	return 1;
6548 }
6549 
6550 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6551 			   unsigned short port)
6552 {
6553 	unsigned long val;
6554 	int ret;
6555 
6556 	/* For size less than 4 we merge, else we zero extend */
6557 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6558 
6559 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6560 				       &val, 1);
6561 	if (ret) {
6562 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6563 		return ret;
6564 	}
6565 
6566 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6567 
6568 	return 0;
6569 }
6570 
6571 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6572 {
6573 	int ret = kvm_skip_emulated_instruction(vcpu);
6574 
6575 	/*
6576 	 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6577 	 * KVM_EXIT_DEBUG here.
6578 	 */
6579 	if (in)
6580 		return kvm_fast_pio_in(vcpu, size, port) && ret;
6581 	else
6582 		return kvm_fast_pio_out(vcpu, size, port) && ret;
6583 }
6584 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6585 
6586 static int kvmclock_cpu_down_prep(unsigned int cpu)
6587 {
6588 	__this_cpu_write(cpu_tsc_khz, 0);
6589 	return 0;
6590 }
6591 
6592 static void tsc_khz_changed(void *data)
6593 {
6594 	struct cpufreq_freqs *freq = data;
6595 	unsigned long khz = 0;
6596 
6597 	if (data)
6598 		khz = freq->new;
6599 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6600 		khz = cpufreq_quick_get(raw_smp_processor_id());
6601 	if (!khz)
6602 		khz = tsc_khz;
6603 	__this_cpu_write(cpu_tsc_khz, khz);
6604 }
6605 
6606 #ifdef CONFIG_X86_64
6607 static void kvm_hyperv_tsc_notifier(void)
6608 {
6609 	struct kvm *kvm;
6610 	struct kvm_vcpu *vcpu;
6611 	int cpu;
6612 
6613 	spin_lock(&kvm_lock);
6614 	list_for_each_entry(kvm, &vm_list, vm_list)
6615 		kvm_make_mclock_inprogress_request(kvm);
6616 
6617 	hyperv_stop_tsc_emulation();
6618 
6619 	/* TSC frequency always matches when on Hyper-V */
6620 	for_each_present_cpu(cpu)
6621 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6622 	kvm_max_guest_tsc_khz = tsc_khz;
6623 
6624 	list_for_each_entry(kvm, &vm_list, vm_list) {
6625 		struct kvm_arch *ka = &kvm->arch;
6626 
6627 		spin_lock(&ka->pvclock_gtod_sync_lock);
6628 
6629 		pvclock_update_vm_gtod_copy(kvm);
6630 
6631 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6632 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6633 
6634 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6635 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6636 
6637 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6638 	}
6639 	spin_unlock(&kvm_lock);
6640 }
6641 #endif
6642 
6643 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6644 				     void *data)
6645 {
6646 	struct cpufreq_freqs *freq = data;
6647 	struct kvm *kvm;
6648 	struct kvm_vcpu *vcpu;
6649 	int i, send_ipi = 0;
6650 
6651 	/*
6652 	 * We allow guests to temporarily run on slowing clocks,
6653 	 * provided we notify them after, or to run on accelerating
6654 	 * clocks, provided we notify them before.  Thus time never
6655 	 * goes backwards.
6656 	 *
6657 	 * However, we have a problem.  We can't atomically update
6658 	 * the frequency of a given CPU from this function; it is
6659 	 * merely a notifier, which can be called from any CPU.
6660 	 * Changing the TSC frequency at arbitrary points in time
6661 	 * requires a recomputation of local variables related to
6662 	 * the TSC for each VCPU.  We must flag these local variables
6663 	 * to be updated and be sure the update takes place with the
6664 	 * new frequency before any guests proceed.
6665 	 *
6666 	 * Unfortunately, the combination of hotplug CPU and frequency
6667 	 * change creates an intractable locking scenario; the order
6668 	 * of when these callouts happen is undefined with respect to
6669 	 * CPU hotplug, and they can race with each other.  As such,
6670 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6671 	 * undefined; you can actually have a CPU frequency change take
6672 	 * place in between the computation of X and the setting of the
6673 	 * variable.  To protect against this problem, all updates of
6674 	 * the per_cpu tsc_khz variable are done in an interrupt
6675 	 * protected IPI, and all callers wishing to update the value
6676 	 * must wait for a synchronous IPI to complete (which is trivial
6677 	 * if the caller is on the CPU already).  This establishes the
6678 	 * necessary total order on variable updates.
6679 	 *
6680 	 * Note that because a guest time update may take place
6681 	 * anytime after the setting of the VCPU's request bit, the
6682 	 * correct TSC value must be set before the request.  However,
6683 	 * to ensure the update actually makes it to any guest which
6684 	 * starts running in hardware virtualization between the set
6685 	 * and the acquisition of the spinlock, we must also ping the
6686 	 * CPU after setting the request bit.
6687 	 *
6688 	 */
6689 
6690 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6691 		return 0;
6692 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6693 		return 0;
6694 
6695 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6696 
6697 	spin_lock(&kvm_lock);
6698 	list_for_each_entry(kvm, &vm_list, vm_list) {
6699 		kvm_for_each_vcpu(i, vcpu, kvm) {
6700 			if (vcpu->cpu != freq->cpu)
6701 				continue;
6702 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6703 			if (vcpu->cpu != smp_processor_id())
6704 				send_ipi = 1;
6705 		}
6706 	}
6707 	spin_unlock(&kvm_lock);
6708 
6709 	if (freq->old < freq->new && send_ipi) {
6710 		/*
6711 		 * We upscale the frequency.  Must make the guest
6712 		 * doesn't see old kvmclock values while running with
6713 		 * the new frequency, otherwise we risk the guest sees
6714 		 * time go backwards.
6715 		 *
6716 		 * In case we update the frequency for another cpu
6717 		 * (which might be in guest context) send an interrupt
6718 		 * to kick the cpu out of guest context.  Next time
6719 		 * guest context is entered kvmclock will be updated,
6720 		 * so the guest will not see stale values.
6721 		 */
6722 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6723 	}
6724 	return 0;
6725 }
6726 
6727 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6728 	.notifier_call  = kvmclock_cpufreq_notifier
6729 };
6730 
6731 static int kvmclock_cpu_online(unsigned int cpu)
6732 {
6733 	tsc_khz_changed(NULL);
6734 	return 0;
6735 }
6736 
6737 static void kvm_timer_init(void)
6738 {
6739 	max_tsc_khz = tsc_khz;
6740 
6741 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6742 #ifdef CONFIG_CPU_FREQ
6743 		struct cpufreq_policy policy;
6744 		int cpu;
6745 
6746 		memset(&policy, 0, sizeof(policy));
6747 		cpu = get_cpu();
6748 		cpufreq_get_policy(&policy, cpu);
6749 		if (policy.cpuinfo.max_freq)
6750 			max_tsc_khz = policy.cpuinfo.max_freq;
6751 		put_cpu();
6752 #endif
6753 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6754 					  CPUFREQ_TRANSITION_NOTIFIER);
6755 	}
6756 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6757 
6758 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6759 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6760 }
6761 
6762 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6763 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6764 
6765 int kvm_is_in_guest(void)
6766 {
6767 	return __this_cpu_read(current_vcpu) != NULL;
6768 }
6769 
6770 static int kvm_is_user_mode(void)
6771 {
6772 	int user_mode = 3;
6773 
6774 	if (__this_cpu_read(current_vcpu))
6775 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6776 
6777 	return user_mode != 0;
6778 }
6779 
6780 static unsigned long kvm_get_guest_ip(void)
6781 {
6782 	unsigned long ip = 0;
6783 
6784 	if (__this_cpu_read(current_vcpu))
6785 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6786 
6787 	return ip;
6788 }
6789 
6790 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6791 	.is_in_guest		= kvm_is_in_guest,
6792 	.is_user_mode		= kvm_is_user_mode,
6793 	.get_guest_ip		= kvm_get_guest_ip,
6794 };
6795 
6796 static void kvm_set_mmio_spte_mask(void)
6797 {
6798 	u64 mask;
6799 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6800 
6801 	/*
6802 	 * Set the reserved bits and the present bit of an paging-structure
6803 	 * entry to generate page fault with PFER.RSV = 1.
6804 	 */
6805 
6806 	/*
6807 	 * Mask the uppermost physical address bit, which would be reserved as
6808 	 * long as the supported physical address width is less than 52.
6809 	 */
6810 	mask = 1ull << 51;
6811 
6812 	/* Set the present bit. */
6813 	mask |= 1ull;
6814 
6815 	/*
6816 	 * If reserved bit is not supported, clear the present bit to disable
6817 	 * mmio page fault.
6818 	 */
6819 	if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6820 		mask &= ~1ull;
6821 
6822 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6823 }
6824 
6825 #ifdef CONFIG_X86_64
6826 static void pvclock_gtod_update_fn(struct work_struct *work)
6827 {
6828 	struct kvm *kvm;
6829 
6830 	struct kvm_vcpu *vcpu;
6831 	int i;
6832 
6833 	spin_lock(&kvm_lock);
6834 	list_for_each_entry(kvm, &vm_list, vm_list)
6835 		kvm_for_each_vcpu(i, vcpu, kvm)
6836 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6837 	atomic_set(&kvm_guest_has_master_clock, 0);
6838 	spin_unlock(&kvm_lock);
6839 }
6840 
6841 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6842 
6843 /*
6844  * Notification about pvclock gtod data update.
6845  */
6846 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6847 			       void *priv)
6848 {
6849 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6850 	struct timekeeper *tk = priv;
6851 
6852 	update_pvclock_gtod(tk);
6853 
6854 	/* disable master clock if host does not trust, or does not
6855 	 * use, TSC based clocksource.
6856 	 */
6857 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6858 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6859 		queue_work(system_long_wq, &pvclock_gtod_work);
6860 
6861 	return 0;
6862 }
6863 
6864 static struct notifier_block pvclock_gtod_notifier = {
6865 	.notifier_call = pvclock_gtod_notify,
6866 };
6867 #endif
6868 
6869 int kvm_arch_init(void *opaque)
6870 {
6871 	int r;
6872 	struct kvm_x86_ops *ops = opaque;
6873 
6874 	if (kvm_x86_ops) {
6875 		printk(KERN_ERR "kvm: already loaded the other module\n");
6876 		r = -EEXIST;
6877 		goto out;
6878 	}
6879 
6880 	if (!ops->cpu_has_kvm_support()) {
6881 		printk(KERN_ERR "kvm: no hardware support\n");
6882 		r = -EOPNOTSUPP;
6883 		goto out;
6884 	}
6885 	if (ops->disabled_by_bios()) {
6886 		printk(KERN_ERR "kvm: disabled by bios\n");
6887 		r = -EOPNOTSUPP;
6888 		goto out;
6889 	}
6890 
6891 	/*
6892 	 * KVM explicitly assumes that the guest has an FPU and
6893 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
6894 	 * vCPU's FPU state as a fxregs_state struct.
6895 	 */
6896 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
6897 		printk(KERN_ERR "kvm: inadequate fpu\n");
6898 		r = -EOPNOTSUPP;
6899 		goto out;
6900 	}
6901 
6902 	r = -ENOMEM;
6903 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
6904 					  __alignof__(struct fpu), SLAB_ACCOUNT,
6905 					  NULL);
6906 	if (!x86_fpu_cache) {
6907 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
6908 		goto out;
6909 	}
6910 
6911 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6912 	if (!shared_msrs) {
6913 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6914 		goto out_free_x86_fpu_cache;
6915 	}
6916 
6917 	r = kvm_mmu_module_init();
6918 	if (r)
6919 		goto out_free_percpu;
6920 
6921 	kvm_set_mmio_spte_mask();
6922 
6923 	kvm_x86_ops = ops;
6924 
6925 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6926 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6927 			PT_PRESENT_MASK, 0, sme_me_mask);
6928 	kvm_timer_init();
6929 
6930 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6931 
6932 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6933 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6934 
6935 	kvm_lapic_init();
6936 #ifdef CONFIG_X86_64
6937 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6938 
6939 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6940 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6941 #endif
6942 
6943 	return 0;
6944 
6945 out_free_percpu:
6946 	free_percpu(shared_msrs);
6947 out_free_x86_fpu_cache:
6948 	kmem_cache_destroy(x86_fpu_cache);
6949 out:
6950 	return r;
6951 }
6952 
6953 void kvm_arch_exit(void)
6954 {
6955 #ifdef CONFIG_X86_64
6956 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6957 		clear_hv_tscchange_cb();
6958 #endif
6959 	kvm_lapic_exit();
6960 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6961 
6962 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6963 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6964 					    CPUFREQ_TRANSITION_NOTIFIER);
6965 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6966 #ifdef CONFIG_X86_64
6967 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6968 #endif
6969 	kvm_x86_ops = NULL;
6970 	kvm_mmu_module_exit();
6971 	free_percpu(shared_msrs);
6972 	kmem_cache_destroy(x86_fpu_cache);
6973 }
6974 
6975 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6976 {
6977 	++vcpu->stat.halt_exits;
6978 	if (lapic_in_kernel(vcpu)) {
6979 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6980 		return 1;
6981 	} else {
6982 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6983 		return 0;
6984 	}
6985 }
6986 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6987 
6988 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6989 {
6990 	int ret = kvm_skip_emulated_instruction(vcpu);
6991 	/*
6992 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6993 	 * KVM_EXIT_DEBUG here.
6994 	 */
6995 	return kvm_vcpu_halt(vcpu) && ret;
6996 }
6997 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6998 
6999 #ifdef CONFIG_X86_64
7000 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7001 			        unsigned long clock_type)
7002 {
7003 	struct kvm_clock_pairing clock_pairing;
7004 	struct timespec64 ts;
7005 	u64 cycle;
7006 	int ret;
7007 
7008 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7009 		return -KVM_EOPNOTSUPP;
7010 
7011 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7012 		return -KVM_EOPNOTSUPP;
7013 
7014 	clock_pairing.sec = ts.tv_sec;
7015 	clock_pairing.nsec = ts.tv_nsec;
7016 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7017 	clock_pairing.flags = 0;
7018 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7019 
7020 	ret = 0;
7021 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7022 			    sizeof(struct kvm_clock_pairing)))
7023 		ret = -KVM_EFAULT;
7024 
7025 	return ret;
7026 }
7027 #endif
7028 
7029 /*
7030  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7031  *
7032  * @apicid - apicid of vcpu to be kicked.
7033  */
7034 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7035 {
7036 	struct kvm_lapic_irq lapic_irq;
7037 
7038 	lapic_irq.shorthand = 0;
7039 	lapic_irq.dest_mode = 0;
7040 	lapic_irq.level = 0;
7041 	lapic_irq.dest_id = apicid;
7042 	lapic_irq.msi_redir_hint = false;
7043 
7044 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7045 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7046 }
7047 
7048 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7049 {
7050 	vcpu->arch.apicv_active = false;
7051 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7052 }
7053 
7054 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7055 {
7056 	unsigned long nr, a0, a1, a2, a3, ret;
7057 	int op_64_bit;
7058 
7059 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7060 		return kvm_hv_hypercall(vcpu);
7061 
7062 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
7063 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
7064 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
7065 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
7066 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
7067 
7068 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7069 
7070 	op_64_bit = is_64_bit_mode(vcpu);
7071 	if (!op_64_bit) {
7072 		nr &= 0xFFFFFFFF;
7073 		a0 &= 0xFFFFFFFF;
7074 		a1 &= 0xFFFFFFFF;
7075 		a2 &= 0xFFFFFFFF;
7076 		a3 &= 0xFFFFFFFF;
7077 	}
7078 
7079 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7080 		ret = -KVM_EPERM;
7081 		goto out;
7082 	}
7083 
7084 	switch (nr) {
7085 	case KVM_HC_VAPIC_POLL_IRQ:
7086 		ret = 0;
7087 		break;
7088 	case KVM_HC_KICK_CPU:
7089 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7090 		ret = 0;
7091 		break;
7092 #ifdef CONFIG_X86_64
7093 	case KVM_HC_CLOCK_PAIRING:
7094 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7095 		break;
7096 	case KVM_HC_SEND_IPI:
7097 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7098 		break;
7099 #endif
7100 	default:
7101 		ret = -KVM_ENOSYS;
7102 		break;
7103 	}
7104 out:
7105 	if (!op_64_bit)
7106 		ret = (u32)ret;
7107 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7108 
7109 	++vcpu->stat.hypercalls;
7110 	return kvm_skip_emulated_instruction(vcpu);
7111 }
7112 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7113 
7114 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7115 {
7116 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7117 	char instruction[3];
7118 	unsigned long rip = kvm_rip_read(vcpu);
7119 
7120 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7121 
7122 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7123 		&ctxt->exception);
7124 }
7125 
7126 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7127 {
7128 	return vcpu->run->request_interrupt_window &&
7129 		likely(!pic_in_kernel(vcpu->kvm));
7130 }
7131 
7132 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7133 {
7134 	struct kvm_run *kvm_run = vcpu->run;
7135 
7136 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7137 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7138 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7139 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7140 	kvm_run->ready_for_interrupt_injection =
7141 		pic_in_kernel(vcpu->kvm) ||
7142 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7143 }
7144 
7145 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7146 {
7147 	int max_irr, tpr;
7148 
7149 	if (!kvm_x86_ops->update_cr8_intercept)
7150 		return;
7151 
7152 	if (!lapic_in_kernel(vcpu))
7153 		return;
7154 
7155 	if (vcpu->arch.apicv_active)
7156 		return;
7157 
7158 	if (!vcpu->arch.apic->vapic_addr)
7159 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7160 	else
7161 		max_irr = -1;
7162 
7163 	if (max_irr != -1)
7164 		max_irr >>= 4;
7165 
7166 	tpr = kvm_lapic_get_cr8(vcpu);
7167 
7168 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7169 }
7170 
7171 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7172 {
7173 	int r;
7174 
7175 	/* try to reinject previous events if any */
7176 
7177 	if (vcpu->arch.exception.injected)
7178 		kvm_x86_ops->queue_exception(vcpu);
7179 	/*
7180 	 * Do not inject an NMI or interrupt if there is a pending
7181 	 * exception.  Exceptions and interrupts are recognized at
7182 	 * instruction boundaries, i.e. the start of an instruction.
7183 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7184 	 * NMIs and interrupts, i.e. traps are recognized before an
7185 	 * NMI/interrupt that's pending on the same instruction.
7186 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7187 	 * priority, but are only generated (pended) during instruction
7188 	 * execution, i.e. a pending fault-like exception means the
7189 	 * fault occurred on the *previous* instruction and must be
7190 	 * serviced prior to recognizing any new events in order to
7191 	 * fully complete the previous instruction.
7192 	 */
7193 	else if (!vcpu->arch.exception.pending) {
7194 		if (vcpu->arch.nmi_injected)
7195 			kvm_x86_ops->set_nmi(vcpu);
7196 		else if (vcpu->arch.interrupt.injected)
7197 			kvm_x86_ops->set_irq(vcpu);
7198 	}
7199 
7200 	/*
7201 	 * Call check_nested_events() even if we reinjected a previous event
7202 	 * in order for caller to determine if it should require immediate-exit
7203 	 * from L2 to L1 due to pending L1 events which require exit
7204 	 * from L2 to L1.
7205 	 */
7206 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7207 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7208 		if (r != 0)
7209 			return r;
7210 	}
7211 
7212 	/* try to inject new event if pending */
7213 	if (vcpu->arch.exception.pending) {
7214 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7215 					vcpu->arch.exception.has_error_code,
7216 					vcpu->arch.exception.error_code);
7217 
7218 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7219 		vcpu->arch.exception.pending = false;
7220 		vcpu->arch.exception.injected = true;
7221 
7222 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7223 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7224 					     X86_EFLAGS_RF);
7225 
7226 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7227 			/*
7228 			 * This code assumes that nSVM doesn't use
7229 			 * check_nested_events(). If it does, the
7230 			 * DR6/DR7 changes should happen before L1
7231 			 * gets a #VMEXIT for an intercepted #DB in
7232 			 * L2.  (Under VMX, on the other hand, the
7233 			 * DR6/DR7 changes should not happen in the
7234 			 * event of a VM-exit to L1 for an intercepted
7235 			 * #DB in L2.)
7236 			 */
7237 			kvm_deliver_exception_payload(vcpu);
7238 			if (vcpu->arch.dr7 & DR7_GD) {
7239 				vcpu->arch.dr7 &= ~DR7_GD;
7240 				kvm_update_dr7(vcpu);
7241 			}
7242 		}
7243 
7244 		kvm_x86_ops->queue_exception(vcpu);
7245 	}
7246 
7247 	/* Don't consider new event if we re-injected an event */
7248 	if (kvm_event_needs_reinjection(vcpu))
7249 		return 0;
7250 
7251 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7252 	    kvm_x86_ops->smi_allowed(vcpu)) {
7253 		vcpu->arch.smi_pending = false;
7254 		++vcpu->arch.smi_count;
7255 		enter_smm(vcpu);
7256 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7257 		--vcpu->arch.nmi_pending;
7258 		vcpu->arch.nmi_injected = true;
7259 		kvm_x86_ops->set_nmi(vcpu);
7260 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7261 		/*
7262 		 * Because interrupts can be injected asynchronously, we are
7263 		 * calling check_nested_events again here to avoid a race condition.
7264 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7265 		 * proposal and current concerns.  Perhaps we should be setting
7266 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7267 		 */
7268 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7269 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7270 			if (r != 0)
7271 				return r;
7272 		}
7273 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7274 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7275 					    false);
7276 			kvm_x86_ops->set_irq(vcpu);
7277 		}
7278 	}
7279 
7280 	return 0;
7281 }
7282 
7283 static void process_nmi(struct kvm_vcpu *vcpu)
7284 {
7285 	unsigned limit = 2;
7286 
7287 	/*
7288 	 * x86 is limited to one NMI running, and one NMI pending after it.
7289 	 * If an NMI is already in progress, limit further NMIs to just one.
7290 	 * Otherwise, allow two (and we'll inject the first one immediately).
7291 	 */
7292 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7293 		limit = 1;
7294 
7295 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7296 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7297 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7298 }
7299 
7300 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7301 {
7302 	u32 flags = 0;
7303 	flags |= seg->g       << 23;
7304 	flags |= seg->db      << 22;
7305 	flags |= seg->l       << 21;
7306 	flags |= seg->avl     << 20;
7307 	flags |= seg->present << 15;
7308 	flags |= seg->dpl     << 13;
7309 	flags |= seg->s       << 12;
7310 	flags |= seg->type    << 8;
7311 	return flags;
7312 }
7313 
7314 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7315 {
7316 	struct kvm_segment seg;
7317 	int offset;
7318 
7319 	kvm_get_segment(vcpu, &seg, n);
7320 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7321 
7322 	if (n < 3)
7323 		offset = 0x7f84 + n * 12;
7324 	else
7325 		offset = 0x7f2c + (n - 3) * 12;
7326 
7327 	put_smstate(u32, buf, offset + 8, seg.base);
7328 	put_smstate(u32, buf, offset + 4, seg.limit);
7329 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7330 }
7331 
7332 #ifdef CONFIG_X86_64
7333 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7334 {
7335 	struct kvm_segment seg;
7336 	int offset;
7337 	u16 flags;
7338 
7339 	kvm_get_segment(vcpu, &seg, n);
7340 	offset = 0x7e00 + n * 16;
7341 
7342 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7343 	put_smstate(u16, buf, offset, seg.selector);
7344 	put_smstate(u16, buf, offset + 2, flags);
7345 	put_smstate(u32, buf, offset + 4, seg.limit);
7346 	put_smstate(u64, buf, offset + 8, seg.base);
7347 }
7348 #endif
7349 
7350 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7351 {
7352 	struct desc_ptr dt;
7353 	struct kvm_segment seg;
7354 	unsigned long val;
7355 	int i;
7356 
7357 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7358 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7359 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7360 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7361 
7362 	for (i = 0; i < 8; i++)
7363 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7364 
7365 	kvm_get_dr(vcpu, 6, &val);
7366 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7367 	kvm_get_dr(vcpu, 7, &val);
7368 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7369 
7370 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7371 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7372 	put_smstate(u32, buf, 0x7f64, seg.base);
7373 	put_smstate(u32, buf, 0x7f60, seg.limit);
7374 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7375 
7376 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7377 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7378 	put_smstate(u32, buf, 0x7f80, seg.base);
7379 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7380 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7381 
7382 	kvm_x86_ops->get_gdt(vcpu, &dt);
7383 	put_smstate(u32, buf, 0x7f74, dt.address);
7384 	put_smstate(u32, buf, 0x7f70, dt.size);
7385 
7386 	kvm_x86_ops->get_idt(vcpu, &dt);
7387 	put_smstate(u32, buf, 0x7f58, dt.address);
7388 	put_smstate(u32, buf, 0x7f54, dt.size);
7389 
7390 	for (i = 0; i < 6; i++)
7391 		enter_smm_save_seg_32(vcpu, buf, i);
7392 
7393 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7394 
7395 	/* revision id */
7396 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7397 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7398 }
7399 
7400 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7401 {
7402 #ifdef CONFIG_X86_64
7403 	struct desc_ptr dt;
7404 	struct kvm_segment seg;
7405 	unsigned long val;
7406 	int i;
7407 
7408 	for (i = 0; i < 16; i++)
7409 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7410 
7411 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7412 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7413 
7414 	kvm_get_dr(vcpu, 6, &val);
7415 	put_smstate(u64, buf, 0x7f68, val);
7416 	kvm_get_dr(vcpu, 7, &val);
7417 	put_smstate(u64, buf, 0x7f60, val);
7418 
7419 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7420 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7421 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7422 
7423 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7424 
7425 	/* revision id */
7426 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7427 
7428 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7429 
7430 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7431 	put_smstate(u16, buf, 0x7e90, seg.selector);
7432 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7433 	put_smstate(u32, buf, 0x7e94, seg.limit);
7434 	put_smstate(u64, buf, 0x7e98, seg.base);
7435 
7436 	kvm_x86_ops->get_idt(vcpu, &dt);
7437 	put_smstate(u32, buf, 0x7e84, dt.size);
7438 	put_smstate(u64, buf, 0x7e88, dt.address);
7439 
7440 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7441 	put_smstate(u16, buf, 0x7e70, seg.selector);
7442 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7443 	put_smstate(u32, buf, 0x7e74, seg.limit);
7444 	put_smstate(u64, buf, 0x7e78, seg.base);
7445 
7446 	kvm_x86_ops->get_gdt(vcpu, &dt);
7447 	put_smstate(u32, buf, 0x7e64, dt.size);
7448 	put_smstate(u64, buf, 0x7e68, dt.address);
7449 
7450 	for (i = 0; i < 6; i++)
7451 		enter_smm_save_seg_64(vcpu, buf, i);
7452 #else
7453 	WARN_ON_ONCE(1);
7454 #endif
7455 }
7456 
7457 static void enter_smm(struct kvm_vcpu *vcpu)
7458 {
7459 	struct kvm_segment cs, ds;
7460 	struct desc_ptr dt;
7461 	char buf[512];
7462 	u32 cr0;
7463 
7464 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7465 	memset(buf, 0, 512);
7466 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7467 		enter_smm_save_state_64(vcpu, buf);
7468 	else
7469 		enter_smm_save_state_32(vcpu, buf);
7470 
7471 	/*
7472 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7473 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7474 	 * the SMM state-save area.
7475 	 */
7476 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7477 
7478 	vcpu->arch.hflags |= HF_SMM_MASK;
7479 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7480 
7481 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7482 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7483 	else
7484 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7485 
7486 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7487 	kvm_rip_write(vcpu, 0x8000);
7488 
7489 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7490 	kvm_x86_ops->set_cr0(vcpu, cr0);
7491 	vcpu->arch.cr0 = cr0;
7492 
7493 	kvm_x86_ops->set_cr4(vcpu, 0);
7494 
7495 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7496 	dt.address = dt.size = 0;
7497 	kvm_x86_ops->set_idt(vcpu, &dt);
7498 
7499 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7500 
7501 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7502 	cs.base = vcpu->arch.smbase;
7503 
7504 	ds.selector = 0;
7505 	ds.base = 0;
7506 
7507 	cs.limit    = ds.limit = 0xffffffff;
7508 	cs.type     = ds.type = 0x3;
7509 	cs.dpl      = ds.dpl = 0;
7510 	cs.db       = ds.db = 0;
7511 	cs.s        = ds.s = 1;
7512 	cs.l        = ds.l = 0;
7513 	cs.g        = ds.g = 1;
7514 	cs.avl      = ds.avl = 0;
7515 	cs.present  = ds.present = 1;
7516 	cs.unusable = ds.unusable = 0;
7517 	cs.padding  = ds.padding = 0;
7518 
7519 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7520 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7521 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7522 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7523 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7524 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7525 
7526 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7527 		kvm_x86_ops->set_efer(vcpu, 0);
7528 
7529 	kvm_update_cpuid(vcpu);
7530 	kvm_mmu_reset_context(vcpu);
7531 }
7532 
7533 static void process_smi(struct kvm_vcpu *vcpu)
7534 {
7535 	vcpu->arch.smi_pending = true;
7536 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7537 }
7538 
7539 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7540 {
7541 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7542 }
7543 
7544 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7545 {
7546 	if (!kvm_apic_present(vcpu))
7547 		return;
7548 
7549 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7550 
7551 	if (irqchip_split(vcpu->kvm))
7552 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7553 	else {
7554 		if (vcpu->arch.apicv_active)
7555 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7556 		if (ioapic_in_kernel(vcpu->kvm))
7557 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7558 	}
7559 
7560 	if (is_guest_mode(vcpu))
7561 		vcpu->arch.load_eoi_exitmap_pending = true;
7562 	else
7563 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7564 }
7565 
7566 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7567 {
7568 	u64 eoi_exit_bitmap[4];
7569 
7570 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7571 		return;
7572 
7573 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7574 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7575 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7576 }
7577 
7578 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7579 		unsigned long start, unsigned long end,
7580 		bool blockable)
7581 {
7582 	unsigned long apic_address;
7583 
7584 	/*
7585 	 * The physical address of apic access page is stored in the VMCS.
7586 	 * Update it when it becomes invalid.
7587 	 */
7588 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7589 	if (start <= apic_address && apic_address < end)
7590 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7591 
7592 	return 0;
7593 }
7594 
7595 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7596 {
7597 	struct page *page = NULL;
7598 
7599 	if (!lapic_in_kernel(vcpu))
7600 		return;
7601 
7602 	if (!kvm_x86_ops->set_apic_access_page_addr)
7603 		return;
7604 
7605 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7606 	if (is_error_page(page))
7607 		return;
7608 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7609 
7610 	/*
7611 	 * Do not pin apic access page in memory, the MMU notifier
7612 	 * will call us again if it is migrated or swapped out.
7613 	 */
7614 	put_page(page);
7615 }
7616 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7617 
7618 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7619 {
7620 	smp_send_reschedule(vcpu->cpu);
7621 }
7622 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7623 
7624 /*
7625  * Returns 1 to let vcpu_run() continue the guest execution loop without
7626  * exiting to the userspace.  Otherwise, the value will be returned to the
7627  * userspace.
7628  */
7629 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7630 {
7631 	int r;
7632 	bool req_int_win =
7633 		dm_request_for_irq_injection(vcpu) &&
7634 		kvm_cpu_accept_dm_intr(vcpu);
7635 
7636 	bool req_immediate_exit = false;
7637 
7638 	if (kvm_request_pending(vcpu)) {
7639 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7640 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7641 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7642 			kvm_mmu_unload(vcpu);
7643 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7644 			__kvm_migrate_timers(vcpu);
7645 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7646 			kvm_gen_update_masterclock(vcpu->kvm);
7647 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7648 			kvm_gen_kvmclock_update(vcpu);
7649 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7650 			r = kvm_guest_time_update(vcpu);
7651 			if (unlikely(r))
7652 				goto out;
7653 		}
7654 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7655 			kvm_mmu_sync_roots(vcpu);
7656 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7657 			kvm_mmu_load_cr3(vcpu);
7658 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7659 			kvm_vcpu_flush_tlb(vcpu, true);
7660 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7661 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7662 			r = 0;
7663 			goto out;
7664 		}
7665 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7666 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7667 			vcpu->mmio_needed = 0;
7668 			r = 0;
7669 			goto out;
7670 		}
7671 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7672 			/* Page is swapped out. Do synthetic halt */
7673 			vcpu->arch.apf.halted = true;
7674 			r = 1;
7675 			goto out;
7676 		}
7677 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7678 			record_steal_time(vcpu);
7679 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7680 			process_smi(vcpu);
7681 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7682 			process_nmi(vcpu);
7683 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7684 			kvm_pmu_handle_event(vcpu);
7685 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7686 			kvm_pmu_deliver_pmi(vcpu);
7687 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7688 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7689 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7690 				     vcpu->arch.ioapic_handled_vectors)) {
7691 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7692 				vcpu->run->eoi.vector =
7693 						vcpu->arch.pending_ioapic_eoi;
7694 				r = 0;
7695 				goto out;
7696 			}
7697 		}
7698 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7699 			vcpu_scan_ioapic(vcpu);
7700 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7701 			vcpu_load_eoi_exitmap(vcpu);
7702 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7703 			kvm_vcpu_reload_apic_access_page(vcpu);
7704 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7705 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7706 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7707 			r = 0;
7708 			goto out;
7709 		}
7710 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7711 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7712 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7713 			r = 0;
7714 			goto out;
7715 		}
7716 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7717 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7718 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7719 			r = 0;
7720 			goto out;
7721 		}
7722 
7723 		/*
7724 		 * KVM_REQ_HV_STIMER has to be processed after
7725 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7726 		 * depend on the guest clock being up-to-date
7727 		 */
7728 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7729 			kvm_hv_process_stimers(vcpu);
7730 	}
7731 
7732 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7733 		++vcpu->stat.req_event;
7734 		kvm_apic_accept_events(vcpu);
7735 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7736 			r = 1;
7737 			goto out;
7738 		}
7739 
7740 		if (inject_pending_event(vcpu, req_int_win) != 0)
7741 			req_immediate_exit = true;
7742 		else {
7743 			/* Enable SMI/NMI/IRQ window open exits if needed.
7744 			 *
7745 			 * SMIs have three cases:
7746 			 * 1) They can be nested, and then there is nothing to
7747 			 *    do here because RSM will cause a vmexit anyway.
7748 			 * 2) There is an ISA-specific reason why SMI cannot be
7749 			 *    injected, and the moment when this changes can be
7750 			 *    intercepted.
7751 			 * 3) Or the SMI can be pending because
7752 			 *    inject_pending_event has completed the injection
7753 			 *    of an IRQ or NMI from the previous vmexit, and
7754 			 *    then we request an immediate exit to inject the
7755 			 *    SMI.
7756 			 */
7757 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7758 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7759 					req_immediate_exit = true;
7760 			if (vcpu->arch.nmi_pending)
7761 				kvm_x86_ops->enable_nmi_window(vcpu);
7762 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7763 				kvm_x86_ops->enable_irq_window(vcpu);
7764 			WARN_ON(vcpu->arch.exception.pending);
7765 		}
7766 
7767 		if (kvm_lapic_enabled(vcpu)) {
7768 			update_cr8_intercept(vcpu);
7769 			kvm_lapic_sync_to_vapic(vcpu);
7770 		}
7771 	}
7772 
7773 	r = kvm_mmu_reload(vcpu);
7774 	if (unlikely(r)) {
7775 		goto cancel_injection;
7776 	}
7777 
7778 	preempt_disable();
7779 
7780 	kvm_x86_ops->prepare_guest_switch(vcpu);
7781 
7782 	/*
7783 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7784 	 * IPI are then delayed after guest entry, which ensures that they
7785 	 * result in virtual interrupt delivery.
7786 	 */
7787 	local_irq_disable();
7788 	vcpu->mode = IN_GUEST_MODE;
7789 
7790 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7791 
7792 	/*
7793 	 * 1) We should set ->mode before checking ->requests.  Please see
7794 	 * the comment in kvm_vcpu_exiting_guest_mode().
7795 	 *
7796 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
7797 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7798 	 * (see vmx_deliver_posted_interrupt).
7799 	 *
7800 	 * 3) This also orders the write to mode from any reads to the page
7801 	 * tables done while the VCPU is running.  Please see the comment
7802 	 * in kvm_flush_remote_tlbs.
7803 	 */
7804 	smp_mb__after_srcu_read_unlock();
7805 
7806 	/*
7807 	 * This handles the case where a posted interrupt was
7808 	 * notified with kvm_vcpu_kick.
7809 	 */
7810 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7811 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7812 
7813 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7814 	    || need_resched() || signal_pending(current)) {
7815 		vcpu->mode = OUTSIDE_GUEST_MODE;
7816 		smp_wmb();
7817 		local_irq_enable();
7818 		preempt_enable();
7819 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7820 		r = 1;
7821 		goto cancel_injection;
7822 	}
7823 
7824 	kvm_load_guest_xcr0(vcpu);
7825 
7826 	if (req_immediate_exit) {
7827 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7828 		kvm_x86_ops->request_immediate_exit(vcpu);
7829 	}
7830 
7831 	trace_kvm_entry(vcpu->vcpu_id);
7832 	if (lapic_timer_advance_ns)
7833 		wait_lapic_expire(vcpu);
7834 	guest_enter_irqoff();
7835 
7836 	if (unlikely(vcpu->arch.switch_db_regs)) {
7837 		set_debugreg(0, 7);
7838 		set_debugreg(vcpu->arch.eff_db[0], 0);
7839 		set_debugreg(vcpu->arch.eff_db[1], 1);
7840 		set_debugreg(vcpu->arch.eff_db[2], 2);
7841 		set_debugreg(vcpu->arch.eff_db[3], 3);
7842 		set_debugreg(vcpu->arch.dr6, 6);
7843 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7844 	}
7845 
7846 	kvm_x86_ops->run(vcpu);
7847 
7848 	/*
7849 	 * Do this here before restoring debug registers on the host.  And
7850 	 * since we do this before handling the vmexit, a DR access vmexit
7851 	 * can (a) read the correct value of the debug registers, (b) set
7852 	 * KVM_DEBUGREG_WONT_EXIT again.
7853 	 */
7854 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7855 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7856 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7857 		kvm_update_dr0123(vcpu);
7858 		kvm_update_dr6(vcpu);
7859 		kvm_update_dr7(vcpu);
7860 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7861 	}
7862 
7863 	/*
7864 	 * If the guest has used debug registers, at least dr7
7865 	 * will be disabled while returning to the host.
7866 	 * If we don't have active breakpoints in the host, we don't
7867 	 * care about the messed up debug address registers. But if
7868 	 * we have some of them active, restore the old state.
7869 	 */
7870 	if (hw_breakpoint_active())
7871 		hw_breakpoint_restore();
7872 
7873 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7874 
7875 	vcpu->mode = OUTSIDE_GUEST_MODE;
7876 	smp_wmb();
7877 
7878 	kvm_put_guest_xcr0(vcpu);
7879 
7880 	kvm_before_interrupt(vcpu);
7881 	kvm_x86_ops->handle_external_intr(vcpu);
7882 	kvm_after_interrupt(vcpu);
7883 
7884 	++vcpu->stat.exits;
7885 
7886 	guest_exit_irqoff();
7887 
7888 	local_irq_enable();
7889 	preempt_enable();
7890 
7891 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7892 
7893 	/*
7894 	 * Profile KVM exit RIPs:
7895 	 */
7896 	if (unlikely(prof_on == KVM_PROFILING)) {
7897 		unsigned long rip = kvm_rip_read(vcpu);
7898 		profile_hit(KVM_PROFILING, (void *)rip);
7899 	}
7900 
7901 	if (unlikely(vcpu->arch.tsc_always_catchup))
7902 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7903 
7904 	if (vcpu->arch.apic_attention)
7905 		kvm_lapic_sync_from_vapic(vcpu);
7906 
7907 	vcpu->arch.gpa_available = false;
7908 	r = kvm_x86_ops->handle_exit(vcpu);
7909 	return r;
7910 
7911 cancel_injection:
7912 	kvm_x86_ops->cancel_injection(vcpu);
7913 	if (unlikely(vcpu->arch.apic_attention))
7914 		kvm_lapic_sync_from_vapic(vcpu);
7915 out:
7916 	return r;
7917 }
7918 
7919 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7920 {
7921 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7922 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7923 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7924 		kvm_vcpu_block(vcpu);
7925 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7926 
7927 		if (kvm_x86_ops->post_block)
7928 			kvm_x86_ops->post_block(vcpu);
7929 
7930 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7931 			return 1;
7932 	}
7933 
7934 	kvm_apic_accept_events(vcpu);
7935 	switch(vcpu->arch.mp_state) {
7936 	case KVM_MP_STATE_HALTED:
7937 		vcpu->arch.pv.pv_unhalted = false;
7938 		vcpu->arch.mp_state =
7939 			KVM_MP_STATE_RUNNABLE;
7940 	case KVM_MP_STATE_RUNNABLE:
7941 		vcpu->arch.apf.halted = false;
7942 		break;
7943 	case KVM_MP_STATE_INIT_RECEIVED:
7944 		break;
7945 	default:
7946 		return -EINTR;
7947 		break;
7948 	}
7949 	return 1;
7950 }
7951 
7952 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7953 {
7954 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7955 		kvm_x86_ops->check_nested_events(vcpu, false);
7956 
7957 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7958 		!vcpu->arch.apf.halted);
7959 }
7960 
7961 static int vcpu_run(struct kvm_vcpu *vcpu)
7962 {
7963 	int r;
7964 	struct kvm *kvm = vcpu->kvm;
7965 
7966 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7967 	vcpu->arch.l1tf_flush_l1d = true;
7968 
7969 	for (;;) {
7970 		if (kvm_vcpu_running(vcpu)) {
7971 			r = vcpu_enter_guest(vcpu);
7972 		} else {
7973 			r = vcpu_block(kvm, vcpu);
7974 		}
7975 
7976 		if (r <= 0)
7977 			break;
7978 
7979 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7980 		if (kvm_cpu_has_pending_timer(vcpu))
7981 			kvm_inject_pending_timer_irqs(vcpu);
7982 
7983 		if (dm_request_for_irq_injection(vcpu) &&
7984 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7985 			r = 0;
7986 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7987 			++vcpu->stat.request_irq_exits;
7988 			break;
7989 		}
7990 
7991 		kvm_check_async_pf_completion(vcpu);
7992 
7993 		if (signal_pending(current)) {
7994 			r = -EINTR;
7995 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7996 			++vcpu->stat.signal_exits;
7997 			break;
7998 		}
7999 		if (need_resched()) {
8000 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8001 			cond_resched();
8002 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8003 		}
8004 	}
8005 
8006 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8007 
8008 	return r;
8009 }
8010 
8011 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8012 {
8013 	int r;
8014 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8015 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8016 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8017 	if (r != EMULATE_DONE)
8018 		return 0;
8019 	return 1;
8020 }
8021 
8022 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8023 {
8024 	BUG_ON(!vcpu->arch.pio.count);
8025 
8026 	return complete_emulated_io(vcpu);
8027 }
8028 
8029 /*
8030  * Implements the following, as a state machine:
8031  *
8032  * read:
8033  *   for each fragment
8034  *     for each mmio piece in the fragment
8035  *       write gpa, len
8036  *       exit
8037  *       copy data
8038  *   execute insn
8039  *
8040  * write:
8041  *   for each fragment
8042  *     for each mmio piece in the fragment
8043  *       write gpa, len
8044  *       copy data
8045  *       exit
8046  */
8047 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8048 {
8049 	struct kvm_run *run = vcpu->run;
8050 	struct kvm_mmio_fragment *frag;
8051 	unsigned len;
8052 
8053 	BUG_ON(!vcpu->mmio_needed);
8054 
8055 	/* Complete previous fragment */
8056 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8057 	len = min(8u, frag->len);
8058 	if (!vcpu->mmio_is_write)
8059 		memcpy(frag->data, run->mmio.data, len);
8060 
8061 	if (frag->len <= 8) {
8062 		/* Switch to the next fragment. */
8063 		frag++;
8064 		vcpu->mmio_cur_fragment++;
8065 	} else {
8066 		/* Go forward to the next mmio piece. */
8067 		frag->data += len;
8068 		frag->gpa += len;
8069 		frag->len -= len;
8070 	}
8071 
8072 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8073 		vcpu->mmio_needed = 0;
8074 
8075 		/* FIXME: return into emulator if single-stepping.  */
8076 		if (vcpu->mmio_is_write)
8077 			return 1;
8078 		vcpu->mmio_read_completed = 1;
8079 		return complete_emulated_io(vcpu);
8080 	}
8081 
8082 	run->exit_reason = KVM_EXIT_MMIO;
8083 	run->mmio.phys_addr = frag->gpa;
8084 	if (vcpu->mmio_is_write)
8085 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8086 	run->mmio.len = min(8u, frag->len);
8087 	run->mmio.is_write = vcpu->mmio_is_write;
8088 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8089 	return 0;
8090 }
8091 
8092 /* Swap (qemu) user FPU context for the guest FPU context. */
8093 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8094 {
8095 	preempt_disable();
8096 	copy_fpregs_to_fpstate(&current->thread.fpu);
8097 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8098 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8099 				~XFEATURE_MASK_PKRU);
8100 	preempt_enable();
8101 	trace_kvm_fpu(1);
8102 }
8103 
8104 /* When vcpu_run ends, restore user space FPU context. */
8105 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8106 {
8107 	preempt_disable();
8108 	copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8109 	copy_kernel_to_fpregs(&current->thread.fpu.state);
8110 	preempt_enable();
8111 	++vcpu->stat.fpu_reload;
8112 	trace_kvm_fpu(0);
8113 }
8114 
8115 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8116 {
8117 	int r;
8118 
8119 	vcpu_load(vcpu);
8120 	kvm_sigset_activate(vcpu);
8121 	kvm_load_guest_fpu(vcpu);
8122 
8123 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8124 		if (kvm_run->immediate_exit) {
8125 			r = -EINTR;
8126 			goto out;
8127 		}
8128 		kvm_vcpu_block(vcpu);
8129 		kvm_apic_accept_events(vcpu);
8130 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8131 		r = -EAGAIN;
8132 		if (signal_pending(current)) {
8133 			r = -EINTR;
8134 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8135 			++vcpu->stat.signal_exits;
8136 		}
8137 		goto out;
8138 	}
8139 
8140 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8141 		r = -EINVAL;
8142 		goto out;
8143 	}
8144 
8145 	if (vcpu->run->kvm_dirty_regs) {
8146 		r = sync_regs(vcpu);
8147 		if (r != 0)
8148 			goto out;
8149 	}
8150 
8151 	/* re-sync apic's tpr */
8152 	if (!lapic_in_kernel(vcpu)) {
8153 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8154 			r = -EINVAL;
8155 			goto out;
8156 		}
8157 	}
8158 
8159 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8160 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8161 		vcpu->arch.complete_userspace_io = NULL;
8162 		r = cui(vcpu);
8163 		if (r <= 0)
8164 			goto out;
8165 	} else
8166 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8167 
8168 	if (kvm_run->immediate_exit)
8169 		r = -EINTR;
8170 	else
8171 		r = vcpu_run(vcpu);
8172 
8173 out:
8174 	kvm_put_guest_fpu(vcpu);
8175 	if (vcpu->run->kvm_valid_regs)
8176 		store_regs(vcpu);
8177 	post_kvm_run_save(vcpu);
8178 	kvm_sigset_deactivate(vcpu);
8179 
8180 	vcpu_put(vcpu);
8181 	return r;
8182 }
8183 
8184 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8185 {
8186 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8187 		/*
8188 		 * We are here if userspace calls get_regs() in the middle of
8189 		 * instruction emulation. Registers state needs to be copied
8190 		 * back from emulation context to vcpu. Userspace shouldn't do
8191 		 * that usually, but some bad designed PV devices (vmware
8192 		 * backdoor interface) need this to work
8193 		 */
8194 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8195 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8196 	}
8197 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8198 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8199 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8200 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8201 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8202 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8203 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8204 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8205 #ifdef CONFIG_X86_64
8206 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8207 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8208 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8209 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8210 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8211 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8212 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8213 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8214 #endif
8215 
8216 	regs->rip = kvm_rip_read(vcpu);
8217 	regs->rflags = kvm_get_rflags(vcpu);
8218 }
8219 
8220 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8221 {
8222 	vcpu_load(vcpu);
8223 	__get_regs(vcpu, regs);
8224 	vcpu_put(vcpu);
8225 	return 0;
8226 }
8227 
8228 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8229 {
8230 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8231 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8232 
8233 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8234 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8235 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8236 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8237 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8238 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8239 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8240 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8241 #ifdef CONFIG_X86_64
8242 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8243 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8244 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8245 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8246 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8247 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8248 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8249 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8250 #endif
8251 
8252 	kvm_rip_write(vcpu, regs->rip);
8253 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8254 
8255 	vcpu->arch.exception.pending = false;
8256 
8257 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8258 }
8259 
8260 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8261 {
8262 	vcpu_load(vcpu);
8263 	__set_regs(vcpu, regs);
8264 	vcpu_put(vcpu);
8265 	return 0;
8266 }
8267 
8268 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8269 {
8270 	struct kvm_segment cs;
8271 
8272 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8273 	*db = cs.db;
8274 	*l = cs.l;
8275 }
8276 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8277 
8278 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8279 {
8280 	struct desc_ptr dt;
8281 
8282 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8283 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8284 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8285 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8286 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8287 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8288 
8289 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8290 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8291 
8292 	kvm_x86_ops->get_idt(vcpu, &dt);
8293 	sregs->idt.limit = dt.size;
8294 	sregs->idt.base = dt.address;
8295 	kvm_x86_ops->get_gdt(vcpu, &dt);
8296 	sregs->gdt.limit = dt.size;
8297 	sregs->gdt.base = dt.address;
8298 
8299 	sregs->cr0 = kvm_read_cr0(vcpu);
8300 	sregs->cr2 = vcpu->arch.cr2;
8301 	sregs->cr3 = kvm_read_cr3(vcpu);
8302 	sregs->cr4 = kvm_read_cr4(vcpu);
8303 	sregs->cr8 = kvm_get_cr8(vcpu);
8304 	sregs->efer = vcpu->arch.efer;
8305 	sregs->apic_base = kvm_get_apic_base(vcpu);
8306 
8307 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8308 
8309 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8310 		set_bit(vcpu->arch.interrupt.nr,
8311 			(unsigned long *)sregs->interrupt_bitmap);
8312 }
8313 
8314 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8315 				  struct kvm_sregs *sregs)
8316 {
8317 	vcpu_load(vcpu);
8318 	__get_sregs(vcpu, sregs);
8319 	vcpu_put(vcpu);
8320 	return 0;
8321 }
8322 
8323 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8324 				    struct kvm_mp_state *mp_state)
8325 {
8326 	vcpu_load(vcpu);
8327 
8328 	kvm_apic_accept_events(vcpu);
8329 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8330 					vcpu->arch.pv.pv_unhalted)
8331 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8332 	else
8333 		mp_state->mp_state = vcpu->arch.mp_state;
8334 
8335 	vcpu_put(vcpu);
8336 	return 0;
8337 }
8338 
8339 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8340 				    struct kvm_mp_state *mp_state)
8341 {
8342 	int ret = -EINVAL;
8343 
8344 	vcpu_load(vcpu);
8345 
8346 	if (!lapic_in_kernel(vcpu) &&
8347 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8348 		goto out;
8349 
8350 	/* INITs are latched while in SMM */
8351 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8352 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8353 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8354 		goto out;
8355 
8356 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8357 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8358 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8359 	} else
8360 		vcpu->arch.mp_state = mp_state->mp_state;
8361 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8362 
8363 	ret = 0;
8364 out:
8365 	vcpu_put(vcpu);
8366 	return ret;
8367 }
8368 
8369 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8370 		    int reason, bool has_error_code, u32 error_code)
8371 {
8372 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8373 	int ret;
8374 
8375 	init_emulate_ctxt(vcpu);
8376 
8377 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8378 				   has_error_code, error_code);
8379 
8380 	if (ret)
8381 		return EMULATE_FAIL;
8382 
8383 	kvm_rip_write(vcpu, ctxt->eip);
8384 	kvm_set_rflags(vcpu, ctxt->eflags);
8385 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8386 	return EMULATE_DONE;
8387 }
8388 EXPORT_SYMBOL_GPL(kvm_task_switch);
8389 
8390 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8391 {
8392 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8393 			(sregs->cr4 & X86_CR4_OSXSAVE))
8394 		return  -EINVAL;
8395 
8396 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8397 		/*
8398 		 * When EFER.LME and CR0.PG are set, the processor is in
8399 		 * 64-bit mode (though maybe in a 32-bit code segment).
8400 		 * CR4.PAE and EFER.LMA must be set.
8401 		 */
8402 		if (!(sregs->cr4 & X86_CR4_PAE)
8403 		    || !(sregs->efer & EFER_LMA))
8404 			return -EINVAL;
8405 	} else {
8406 		/*
8407 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8408 		 * segment cannot be 64-bit.
8409 		 */
8410 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8411 			return -EINVAL;
8412 	}
8413 
8414 	return 0;
8415 }
8416 
8417 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8418 {
8419 	struct msr_data apic_base_msr;
8420 	int mmu_reset_needed = 0;
8421 	int cpuid_update_needed = 0;
8422 	int pending_vec, max_bits, idx;
8423 	struct desc_ptr dt;
8424 	int ret = -EINVAL;
8425 
8426 	if (kvm_valid_sregs(vcpu, sregs))
8427 		goto out;
8428 
8429 	apic_base_msr.data = sregs->apic_base;
8430 	apic_base_msr.host_initiated = true;
8431 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8432 		goto out;
8433 
8434 	dt.size = sregs->idt.limit;
8435 	dt.address = sregs->idt.base;
8436 	kvm_x86_ops->set_idt(vcpu, &dt);
8437 	dt.size = sregs->gdt.limit;
8438 	dt.address = sregs->gdt.base;
8439 	kvm_x86_ops->set_gdt(vcpu, &dt);
8440 
8441 	vcpu->arch.cr2 = sregs->cr2;
8442 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8443 	vcpu->arch.cr3 = sregs->cr3;
8444 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8445 
8446 	kvm_set_cr8(vcpu, sregs->cr8);
8447 
8448 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8449 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8450 
8451 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8452 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8453 	vcpu->arch.cr0 = sregs->cr0;
8454 
8455 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8456 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8457 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8458 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8459 	if (cpuid_update_needed)
8460 		kvm_update_cpuid(vcpu);
8461 
8462 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8463 	if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8464 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8465 		mmu_reset_needed = 1;
8466 	}
8467 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8468 
8469 	if (mmu_reset_needed)
8470 		kvm_mmu_reset_context(vcpu);
8471 
8472 	max_bits = KVM_NR_INTERRUPTS;
8473 	pending_vec = find_first_bit(
8474 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8475 	if (pending_vec < max_bits) {
8476 		kvm_queue_interrupt(vcpu, pending_vec, false);
8477 		pr_debug("Set back pending irq %d\n", pending_vec);
8478 	}
8479 
8480 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8481 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8482 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8483 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8484 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8485 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8486 
8487 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8488 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8489 
8490 	update_cr8_intercept(vcpu);
8491 
8492 	/* Older userspace won't unhalt the vcpu on reset. */
8493 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8494 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8495 	    !is_protmode(vcpu))
8496 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8497 
8498 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8499 
8500 	ret = 0;
8501 out:
8502 	return ret;
8503 }
8504 
8505 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8506 				  struct kvm_sregs *sregs)
8507 {
8508 	int ret;
8509 
8510 	vcpu_load(vcpu);
8511 	ret = __set_sregs(vcpu, sregs);
8512 	vcpu_put(vcpu);
8513 	return ret;
8514 }
8515 
8516 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8517 					struct kvm_guest_debug *dbg)
8518 {
8519 	unsigned long rflags;
8520 	int i, r;
8521 
8522 	vcpu_load(vcpu);
8523 
8524 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8525 		r = -EBUSY;
8526 		if (vcpu->arch.exception.pending)
8527 			goto out;
8528 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8529 			kvm_queue_exception(vcpu, DB_VECTOR);
8530 		else
8531 			kvm_queue_exception(vcpu, BP_VECTOR);
8532 	}
8533 
8534 	/*
8535 	 * Read rflags as long as potentially injected trace flags are still
8536 	 * filtered out.
8537 	 */
8538 	rflags = kvm_get_rflags(vcpu);
8539 
8540 	vcpu->guest_debug = dbg->control;
8541 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8542 		vcpu->guest_debug = 0;
8543 
8544 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8545 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8546 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8547 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8548 	} else {
8549 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8550 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8551 	}
8552 	kvm_update_dr7(vcpu);
8553 
8554 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8555 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8556 			get_segment_base(vcpu, VCPU_SREG_CS);
8557 
8558 	/*
8559 	 * Trigger an rflags update that will inject or remove the trace
8560 	 * flags.
8561 	 */
8562 	kvm_set_rflags(vcpu, rflags);
8563 
8564 	kvm_x86_ops->update_bp_intercept(vcpu);
8565 
8566 	r = 0;
8567 
8568 out:
8569 	vcpu_put(vcpu);
8570 	return r;
8571 }
8572 
8573 /*
8574  * Translate a guest virtual address to a guest physical address.
8575  */
8576 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8577 				    struct kvm_translation *tr)
8578 {
8579 	unsigned long vaddr = tr->linear_address;
8580 	gpa_t gpa;
8581 	int idx;
8582 
8583 	vcpu_load(vcpu);
8584 
8585 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8586 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8587 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8588 	tr->physical_address = gpa;
8589 	tr->valid = gpa != UNMAPPED_GVA;
8590 	tr->writeable = 1;
8591 	tr->usermode = 0;
8592 
8593 	vcpu_put(vcpu);
8594 	return 0;
8595 }
8596 
8597 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8598 {
8599 	struct fxregs_state *fxsave;
8600 
8601 	vcpu_load(vcpu);
8602 
8603 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8604 	memcpy(fpu->fpr, fxsave->st_space, 128);
8605 	fpu->fcw = fxsave->cwd;
8606 	fpu->fsw = fxsave->swd;
8607 	fpu->ftwx = fxsave->twd;
8608 	fpu->last_opcode = fxsave->fop;
8609 	fpu->last_ip = fxsave->rip;
8610 	fpu->last_dp = fxsave->rdp;
8611 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8612 
8613 	vcpu_put(vcpu);
8614 	return 0;
8615 }
8616 
8617 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8618 {
8619 	struct fxregs_state *fxsave;
8620 
8621 	vcpu_load(vcpu);
8622 
8623 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8624 
8625 	memcpy(fxsave->st_space, fpu->fpr, 128);
8626 	fxsave->cwd = fpu->fcw;
8627 	fxsave->swd = fpu->fsw;
8628 	fxsave->twd = fpu->ftwx;
8629 	fxsave->fop = fpu->last_opcode;
8630 	fxsave->rip = fpu->last_ip;
8631 	fxsave->rdp = fpu->last_dp;
8632 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8633 
8634 	vcpu_put(vcpu);
8635 	return 0;
8636 }
8637 
8638 static void store_regs(struct kvm_vcpu *vcpu)
8639 {
8640 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8641 
8642 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8643 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8644 
8645 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8646 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8647 
8648 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8649 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8650 				vcpu, &vcpu->run->s.regs.events);
8651 }
8652 
8653 static int sync_regs(struct kvm_vcpu *vcpu)
8654 {
8655 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8656 		return -EINVAL;
8657 
8658 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8659 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8660 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8661 	}
8662 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8663 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8664 			return -EINVAL;
8665 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8666 	}
8667 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8668 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8669 				vcpu, &vcpu->run->s.regs.events))
8670 			return -EINVAL;
8671 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8672 	}
8673 
8674 	return 0;
8675 }
8676 
8677 static void fx_init(struct kvm_vcpu *vcpu)
8678 {
8679 	fpstate_init(&vcpu->arch.guest_fpu->state);
8680 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8681 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8682 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8683 
8684 	/*
8685 	 * Ensure guest xcr0 is valid for loading
8686 	 */
8687 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8688 
8689 	vcpu->arch.cr0 |= X86_CR0_ET;
8690 }
8691 
8692 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8693 {
8694 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8695 
8696 	kvmclock_reset(vcpu);
8697 
8698 	kvm_x86_ops->vcpu_free(vcpu);
8699 	free_cpumask_var(wbinvd_dirty_mask);
8700 }
8701 
8702 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8703 						unsigned int id)
8704 {
8705 	struct kvm_vcpu *vcpu;
8706 
8707 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8708 		printk_once(KERN_WARNING
8709 		"kvm: SMP vm created on host with unstable TSC; "
8710 		"guest TSC will not be reliable\n");
8711 
8712 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8713 
8714 	return vcpu;
8715 }
8716 
8717 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8718 {
8719 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8720 	kvm_vcpu_mtrr_init(vcpu);
8721 	vcpu_load(vcpu);
8722 	kvm_vcpu_reset(vcpu, false);
8723 	kvm_init_mmu(vcpu, false);
8724 	vcpu_put(vcpu);
8725 	return 0;
8726 }
8727 
8728 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8729 {
8730 	struct msr_data msr;
8731 	struct kvm *kvm = vcpu->kvm;
8732 
8733 	kvm_hv_vcpu_postcreate(vcpu);
8734 
8735 	if (mutex_lock_killable(&vcpu->mutex))
8736 		return;
8737 	vcpu_load(vcpu);
8738 	msr.data = 0x0;
8739 	msr.index = MSR_IA32_TSC;
8740 	msr.host_initiated = true;
8741 	kvm_write_tsc(vcpu, &msr);
8742 	vcpu_put(vcpu);
8743 	mutex_unlock(&vcpu->mutex);
8744 
8745 	if (!kvmclock_periodic_sync)
8746 		return;
8747 
8748 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8749 					KVMCLOCK_SYNC_PERIOD);
8750 }
8751 
8752 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8753 {
8754 	vcpu->arch.apf.msr_val = 0;
8755 
8756 	vcpu_load(vcpu);
8757 	kvm_mmu_unload(vcpu);
8758 	vcpu_put(vcpu);
8759 
8760 	kvm_x86_ops->vcpu_free(vcpu);
8761 }
8762 
8763 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8764 {
8765 	kvm_lapic_reset(vcpu, init_event);
8766 
8767 	vcpu->arch.hflags = 0;
8768 
8769 	vcpu->arch.smi_pending = 0;
8770 	vcpu->arch.smi_count = 0;
8771 	atomic_set(&vcpu->arch.nmi_queued, 0);
8772 	vcpu->arch.nmi_pending = 0;
8773 	vcpu->arch.nmi_injected = false;
8774 	kvm_clear_interrupt_queue(vcpu);
8775 	kvm_clear_exception_queue(vcpu);
8776 	vcpu->arch.exception.pending = false;
8777 
8778 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8779 	kvm_update_dr0123(vcpu);
8780 	vcpu->arch.dr6 = DR6_INIT;
8781 	kvm_update_dr6(vcpu);
8782 	vcpu->arch.dr7 = DR7_FIXED_1;
8783 	kvm_update_dr7(vcpu);
8784 
8785 	vcpu->arch.cr2 = 0;
8786 
8787 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8788 	vcpu->arch.apf.msr_val = 0;
8789 	vcpu->arch.st.msr_val = 0;
8790 
8791 	kvmclock_reset(vcpu);
8792 
8793 	kvm_clear_async_pf_completion_queue(vcpu);
8794 	kvm_async_pf_hash_reset(vcpu);
8795 	vcpu->arch.apf.halted = false;
8796 
8797 	if (kvm_mpx_supported()) {
8798 		void *mpx_state_buffer;
8799 
8800 		/*
8801 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8802 		 * called with loaded FPU and does not let userspace fix the state.
8803 		 */
8804 		if (init_event)
8805 			kvm_put_guest_fpu(vcpu);
8806 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8807 					XFEATURE_MASK_BNDREGS);
8808 		if (mpx_state_buffer)
8809 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8810 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8811 					XFEATURE_MASK_BNDCSR);
8812 		if (mpx_state_buffer)
8813 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8814 		if (init_event)
8815 			kvm_load_guest_fpu(vcpu);
8816 	}
8817 
8818 	if (!init_event) {
8819 		kvm_pmu_reset(vcpu);
8820 		vcpu->arch.smbase = 0x30000;
8821 
8822 		vcpu->arch.msr_misc_features_enables = 0;
8823 
8824 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8825 	}
8826 
8827 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8828 	vcpu->arch.regs_avail = ~0;
8829 	vcpu->arch.regs_dirty = ~0;
8830 
8831 	vcpu->arch.ia32_xss = 0;
8832 
8833 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8834 }
8835 
8836 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8837 {
8838 	struct kvm_segment cs;
8839 
8840 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8841 	cs.selector = vector << 8;
8842 	cs.base = vector << 12;
8843 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8844 	kvm_rip_write(vcpu, 0);
8845 }
8846 
8847 int kvm_arch_hardware_enable(void)
8848 {
8849 	struct kvm *kvm;
8850 	struct kvm_vcpu *vcpu;
8851 	int i;
8852 	int ret;
8853 	u64 local_tsc;
8854 	u64 max_tsc = 0;
8855 	bool stable, backwards_tsc = false;
8856 
8857 	kvm_shared_msr_cpu_online();
8858 	ret = kvm_x86_ops->hardware_enable();
8859 	if (ret != 0)
8860 		return ret;
8861 
8862 	local_tsc = rdtsc();
8863 	stable = !kvm_check_tsc_unstable();
8864 	list_for_each_entry(kvm, &vm_list, vm_list) {
8865 		kvm_for_each_vcpu(i, vcpu, kvm) {
8866 			if (!stable && vcpu->cpu == smp_processor_id())
8867 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8868 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8869 				backwards_tsc = true;
8870 				if (vcpu->arch.last_host_tsc > max_tsc)
8871 					max_tsc = vcpu->arch.last_host_tsc;
8872 			}
8873 		}
8874 	}
8875 
8876 	/*
8877 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8878 	 * platforms that reset TSC during suspend or hibernate actions, but
8879 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8880 	 * detect that condition here, which happens early in CPU bringup,
8881 	 * before any KVM threads can be running.  Unfortunately, we can't
8882 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8883 	 * enough into CPU bringup that we know how much real time has actually
8884 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8885 	 * variables that haven't been updated yet.
8886 	 *
8887 	 * So we simply find the maximum observed TSC above, then record the
8888 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8889 	 * the adjustment will be applied.  Note that we accumulate
8890 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8891 	 * gets a chance to run again.  In the event that no KVM threads get a
8892 	 * chance to run, we will miss the entire elapsed period, as we'll have
8893 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8894 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8895 	 * uniform across all VCPUs (not to mention the scenario is extremely
8896 	 * unlikely). It is possible that a second hibernate recovery happens
8897 	 * much faster than a first, causing the observed TSC here to be
8898 	 * smaller; this would require additional padding adjustment, which is
8899 	 * why we set last_host_tsc to the local tsc observed here.
8900 	 *
8901 	 * N.B. - this code below runs only on platforms with reliable TSC,
8902 	 * as that is the only way backwards_tsc is set above.  Also note
8903 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8904 	 * have the same delta_cyc adjustment applied if backwards_tsc
8905 	 * is detected.  Note further, this adjustment is only done once,
8906 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8907 	 * called multiple times (one for each physical CPU bringup).
8908 	 *
8909 	 * Platforms with unreliable TSCs don't have to deal with this, they
8910 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8911 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8912 	 * guarantee that they stay in perfect synchronization.
8913 	 */
8914 	if (backwards_tsc) {
8915 		u64 delta_cyc = max_tsc - local_tsc;
8916 		list_for_each_entry(kvm, &vm_list, vm_list) {
8917 			kvm->arch.backwards_tsc_observed = true;
8918 			kvm_for_each_vcpu(i, vcpu, kvm) {
8919 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8920 				vcpu->arch.last_host_tsc = local_tsc;
8921 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8922 			}
8923 
8924 			/*
8925 			 * We have to disable TSC offset matching.. if you were
8926 			 * booting a VM while issuing an S4 host suspend....
8927 			 * you may have some problem.  Solving this issue is
8928 			 * left as an exercise to the reader.
8929 			 */
8930 			kvm->arch.last_tsc_nsec = 0;
8931 			kvm->arch.last_tsc_write = 0;
8932 		}
8933 
8934 	}
8935 	return 0;
8936 }
8937 
8938 void kvm_arch_hardware_disable(void)
8939 {
8940 	kvm_x86_ops->hardware_disable();
8941 	drop_user_return_notifiers();
8942 }
8943 
8944 int kvm_arch_hardware_setup(void)
8945 {
8946 	int r;
8947 
8948 	r = kvm_x86_ops->hardware_setup();
8949 	if (r != 0)
8950 		return r;
8951 
8952 	if (kvm_has_tsc_control) {
8953 		/*
8954 		 * Make sure the user can only configure tsc_khz values that
8955 		 * fit into a signed integer.
8956 		 * A min value is not calculated because it will always
8957 		 * be 1 on all machines.
8958 		 */
8959 		u64 max = min(0x7fffffffULL,
8960 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8961 		kvm_max_guest_tsc_khz = max;
8962 
8963 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8964 	}
8965 
8966 	kvm_init_msr_list();
8967 	return 0;
8968 }
8969 
8970 void kvm_arch_hardware_unsetup(void)
8971 {
8972 	kvm_x86_ops->hardware_unsetup();
8973 }
8974 
8975 void kvm_arch_check_processor_compat(void *rtn)
8976 {
8977 	kvm_x86_ops->check_processor_compatibility(rtn);
8978 }
8979 
8980 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8981 {
8982 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8983 }
8984 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8985 
8986 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8987 {
8988 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8989 }
8990 
8991 struct static_key kvm_no_apic_vcpu __read_mostly;
8992 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8993 
8994 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8995 {
8996 	struct page *page;
8997 	int r;
8998 
8999 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9000 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9001 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9002 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9003 	else
9004 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9005 
9006 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9007 	if (!page) {
9008 		r = -ENOMEM;
9009 		goto fail;
9010 	}
9011 	vcpu->arch.pio_data = page_address(page);
9012 
9013 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9014 
9015 	r = kvm_mmu_create(vcpu);
9016 	if (r < 0)
9017 		goto fail_free_pio_data;
9018 
9019 	if (irqchip_in_kernel(vcpu->kvm)) {
9020 		r = kvm_create_lapic(vcpu);
9021 		if (r < 0)
9022 			goto fail_mmu_destroy;
9023 	} else
9024 		static_key_slow_inc(&kvm_no_apic_vcpu);
9025 
9026 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9027 				       GFP_KERNEL);
9028 	if (!vcpu->arch.mce_banks) {
9029 		r = -ENOMEM;
9030 		goto fail_free_lapic;
9031 	}
9032 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9033 
9034 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
9035 		r = -ENOMEM;
9036 		goto fail_free_mce_banks;
9037 	}
9038 
9039 	fx_init(vcpu);
9040 
9041 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9042 
9043 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9044 
9045 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9046 
9047 	kvm_async_pf_hash_reset(vcpu);
9048 	kvm_pmu_init(vcpu);
9049 
9050 	vcpu->arch.pending_external_vector = -1;
9051 	vcpu->arch.preempted_in_kernel = false;
9052 
9053 	kvm_hv_vcpu_init(vcpu);
9054 
9055 	return 0;
9056 
9057 fail_free_mce_banks:
9058 	kfree(vcpu->arch.mce_banks);
9059 fail_free_lapic:
9060 	kvm_free_lapic(vcpu);
9061 fail_mmu_destroy:
9062 	kvm_mmu_destroy(vcpu);
9063 fail_free_pio_data:
9064 	free_page((unsigned long)vcpu->arch.pio_data);
9065 fail:
9066 	return r;
9067 }
9068 
9069 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9070 {
9071 	int idx;
9072 
9073 	kvm_hv_vcpu_uninit(vcpu);
9074 	kvm_pmu_destroy(vcpu);
9075 	kfree(vcpu->arch.mce_banks);
9076 	kvm_free_lapic(vcpu);
9077 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9078 	kvm_mmu_destroy(vcpu);
9079 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9080 	free_page((unsigned long)vcpu->arch.pio_data);
9081 	if (!lapic_in_kernel(vcpu))
9082 		static_key_slow_dec(&kvm_no_apic_vcpu);
9083 }
9084 
9085 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9086 {
9087 	vcpu->arch.l1tf_flush_l1d = true;
9088 	kvm_x86_ops->sched_in(vcpu, cpu);
9089 }
9090 
9091 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9092 {
9093 	if (type)
9094 		return -EINVAL;
9095 
9096 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9097 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9098 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9099 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9100 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9101 
9102 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9103 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9104 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9105 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9106 		&kvm->arch.irq_sources_bitmap);
9107 
9108 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9109 	mutex_init(&kvm->arch.apic_map_lock);
9110 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9111 
9112 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9113 	pvclock_update_vm_gtod_copy(kvm);
9114 
9115 	kvm->arch.guest_can_read_msr_platform_info = true;
9116 
9117 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9118 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9119 
9120 	kvm_hv_init_vm(kvm);
9121 	kvm_page_track_init(kvm);
9122 	kvm_mmu_init_vm(kvm);
9123 
9124 	if (kvm_x86_ops->vm_init)
9125 		return kvm_x86_ops->vm_init(kvm);
9126 
9127 	return 0;
9128 }
9129 
9130 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9131 {
9132 	vcpu_load(vcpu);
9133 	kvm_mmu_unload(vcpu);
9134 	vcpu_put(vcpu);
9135 }
9136 
9137 static void kvm_free_vcpus(struct kvm *kvm)
9138 {
9139 	unsigned int i;
9140 	struct kvm_vcpu *vcpu;
9141 
9142 	/*
9143 	 * Unpin any mmu pages first.
9144 	 */
9145 	kvm_for_each_vcpu(i, vcpu, kvm) {
9146 		kvm_clear_async_pf_completion_queue(vcpu);
9147 		kvm_unload_vcpu_mmu(vcpu);
9148 	}
9149 	kvm_for_each_vcpu(i, vcpu, kvm)
9150 		kvm_arch_vcpu_free(vcpu);
9151 
9152 	mutex_lock(&kvm->lock);
9153 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9154 		kvm->vcpus[i] = NULL;
9155 
9156 	atomic_set(&kvm->online_vcpus, 0);
9157 	mutex_unlock(&kvm->lock);
9158 }
9159 
9160 void kvm_arch_sync_events(struct kvm *kvm)
9161 {
9162 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9163 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9164 	kvm_free_pit(kvm);
9165 }
9166 
9167 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9168 {
9169 	int i, r;
9170 	unsigned long hva;
9171 	struct kvm_memslots *slots = kvm_memslots(kvm);
9172 	struct kvm_memory_slot *slot, old;
9173 
9174 	/* Called with kvm->slots_lock held.  */
9175 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9176 		return -EINVAL;
9177 
9178 	slot = id_to_memslot(slots, id);
9179 	if (size) {
9180 		if (slot->npages)
9181 			return -EEXIST;
9182 
9183 		/*
9184 		 * MAP_SHARED to prevent internal slot pages from being moved
9185 		 * by fork()/COW.
9186 		 */
9187 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9188 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9189 		if (IS_ERR((void *)hva))
9190 			return PTR_ERR((void *)hva);
9191 	} else {
9192 		if (!slot->npages)
9193 			return 0;
9194 
9195 		hva = 0;
9196 	}
9197 
9198 	old = *slot;
9199 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9200 		struct kvm_userspace_memory_region m;
9201 
9202 		m.slot = id | (i << 16);
9203 		m.flags = 0;
9204 		m.guest_phys_addr = gpa;
9205 		m.userspace_addr = hva;
9206 		m.memory_size = size;
9207 		r = __kvm_set_memory_region(kvm, &m);
9208 		if (r < 0)
9209 			return r;
9210 	}
9211 
9212 	if (!size)
9213 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9214 
9215 	return 0;
9216 }
9217 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9218 
9219 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9220 {
9221 	int r;
9222 
9223 	mutex_lock(&kvm->slots_lock);
9224 	r = __x86_set_memory_region(kvm, id, gpa, size);
9225 	mutex_unlock(&kvm->slots_lock);
9226 
9227 	return r;
9228 }
9229 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9230 
9231 void kvm_arch_destroy_vm(struct kvm *kvm)
9232 {
9233 	if (current->mm == kvm->mm) {
9234 		/*
9235 		 * Free memory regions allocated on behalf of userspace,
9236 		 * unless the the memory map has changed due to process exit
9237 		 * or fd copying.
9238 		 */
9239 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9240 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9241 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9242 	}
9243 	if (kvm_x86_ops->vm_destroy)
9244 		kvm_x86_ops->vm_destroy(kvm);
9245 	kvm_pic_destroy(kvm);
9246 	kvm_ioapic_destroy(kvm);
9247 	kvm_free_vcpus(kvm);
9248 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9249 	kvm_mmu_uninit_vm(kvm);
9250 	kvm_page_track_cleanup(kvm);
9251 	kvm_hv_destroy_vm(kvm);
9252 }
9253 
9254 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9255 			   struct kvm_memory_slot *dont)
9256 {
9257 	int i;
9258 
9259 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9260 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9261 			kvfree(free->arch.rmap[i]);
9262 			free->arch.rmap[i] = NULL;
9263 		}
9264 		if (i == 0)
9265 			continue;
9266 
9267 		if (!dont || free->arch.lpage_info[i - 1] !=
9268 			     dont->arch.lpage_info[i - 1]) {
9269 			kvfree(free->arch.lpage_info[i - 1]);
9270 			free->arch.lpage_info[i - 1] = NULL;
9271 		}
9272 	}
9273 
9274 	kvm_page_track_free_memslot(free, dont);
9275 }
9276 
9277 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9278 			    unsigned long npages)
9279 {
9280 	int i;
9281 
9282 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9283 		struct kvm_lpage_info *linfo;
9284 		unsigned long ugfn;
9285 		int lpages;
9286 		int level = i + 1;
9287 
9288 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9289 				      slot->base_gfn, level) + 1;
9290 
9291 		slot->arch.rmap[i] =
9292 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9293 				 GFP_KERNEL);
9294 		if (!slot->arch.rmap[i])
9295 			goto out_free;
9296 		if (i == 0)
9297 			continue;
9298 
9299 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9300 		if (!linfo)
9301 			goto out_free;
9302 
9303 		slot->arch.lpage_info[i - 1] = linfo;
9304 
9305 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9306 			linfo[0].disallow_lpage = 1;
9307 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9308 			linfo[lpages - 1].disallow_lpage = 1;
9309 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9310 		/*
9311 		 * If the gfn and userspace address are not aligned wrt each
9312 		 * other, or if explicitly asked to, disable large page
9313 		 * support for this slot
9314 		 */
9315 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9316 		    !kvm_largepages_enabled()) {
9317 			unsigned long j;
9318 
9319 			for (j = 0; j < lpages; ++j)
9320 				linfo[j].disallow_lpage = 1;
9321 		}
9322 	}
9323 
9324 	if (kvm_page_track_create_memslot(slot, npages))
9325 		goto out_free;
9326 
9327 	return 0;
9328 
9329 out_free:
9330 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9331 		kvfree(slot->arch.rmap[i]);
9332 		slot->arch.rmap[i] = NULL;
9333 		if (i == 0)
9334 			continue;
9335 
9336 		kvfree(slot->arch.lpage_info[i - 1]);
9337 		slot->arch.lpage_info[i - 1] = NULL;
9338 	}
9339 	return -ENOMEM;
9340 }
9341 
9342 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
9343 {
9344 	/*
9345 	 * memslots->generation has been incremented.
9346 	 * mmio generation may have reached its maximum value.
9347 	 */
9348 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
9349 }
9350 
9351 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9352 				struct kvm_memory_slot *memslot,
9353 				const struct kvm_userspace_memory_region *mem,
9354 				enum kvm_mr_change change)
9355 {
9356 	return 0;
9357 }
9358 
9359 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9360 				     struct kvm_memory_slot *new)
9361 {
9362 	/* Still write protect RO slot */
9363 	if (new->flags & KVM_MEM_READONLY) {
9364 		kvm_mmu_slot_remove_write_access(kvm, new);
9365 		return;
9366 	}
9367 
9368 	/*
9369 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9370 	 *
9371 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9372 	 *
9373 	 *  - KVM_MR_CREATE with dirty logging is disabled
9374 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9375 	 *
9376 	 * The reason is, in case of PML, we need to set D-bit for any slots
9377 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9378 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9379 	 * guarantees leaving PML enabled during guest's lifetime won't have
9380 	 * any additional overhead from PML when guest is running with dirty
9381 	 * logging disabled for memory slots.
9382 	 *
9383 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9384 	 * to dirty logging mode.
9385 	 *
9386 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9387 	 *
9388 	 * In case of write protect:
9389 	 *
9390 	 * Write protect all pages for dirty logging.
9391 	 *
9392 	 * All the sptes including the large sptes which point to this
9393 	 * slot are set to readonly. We can not create any new large
9394 	 * spte on this slot until the end of the logging.
9395 	 *
9396 	 * See the comments in fast_page_fault().
9397 	 */
9398 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9399 		if (kvm_x86_ops->slot_enable_log_dirty)
9400 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9401 		else
9402 			kvm_mmu_slot_remove_write_access(kvm, new);
9403 	} else {
9404 		if (kvm_x86_ops->slot_disable_log_dirty)
9405 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9406 	}
9407 }
9408 
9409 void kvm_arch_commit_memory_region(struct kvm *kvm,
9410 				const struct kvm_userspace_memory_region *mem,
9411 				const struct kvm_memory_slot *old,
9412 				const struct kvm_memory_slot *new,
9413 				enum kvm_mr_change change)
9414 {
9415 	int nr_mmu_pages = 0;
9416 
9417 	if (!kvm->arch.n_requested_mmu_pages)
9418 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9419 
9420 	if (nr_mmu_pages)
9421 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9422 
9423 	/*
9424 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9425 	 * sptes have to be split.  If live migration is successful, the guest
9426 	 * in the source machine will be destroyed and large sptes will be
9427 	 * created in the destination. However, if the guest continues to run
9428 	 * in the source machine (for example if live migration fails), small
9429 	 * sptes will remain around and cause bad performance.
9430 	 *
9431 	 * Scan sptes if dirty logging has been stopped, dropping those
9432 	 * which can be collapsed into a single large-page spte.  Later
9433 	 * page faults will create the large-page sptes.
9434 	 */
9435 	if ((change != KVM_MR_DELETE) &&
9436 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9437 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9438 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9439 
9440 	/*
9441 	 * Set up write protection and/or dirty logging for the new slot.
9442 	 *
9443 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9444 	 * been zapped so no dirty logging staff is needed for old slot. For
9445 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9446 	 * new and it's also covered when dealing with the new slot.
9447 	 *
9448 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9449 	 */
9450 	if (change != KVM_MR_DELETE)
9451 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9452 }
9453 
9454 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9455 {
9456 	kvm_mmu_invalidate_zap_all_pages(kvm);
9457 }
9458 
9459 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9460 				   struct kvm_memory_slot *slot)
9461 {
9462 	kvm_page_track_flush_slot(kvm, slot);
9463 }
9464 
9465 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9466 {
9467 	return (is_guest_mode(vcpu) &&
9468 			kvm_x86_ops->guest_apic_has_interrupt &&
9469 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9470 }
9471 
9472 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9473 {
9474 	if (!list_empty_careful(&vcpu->async_pf.done))
9475 		return true;
9476 
9477 	if (kvm_apic_has_events(vcpu))
9478 		return true;
9479 
9480 	if (vcpu->arch.pv.pv_unhalted)
9481 		return true;
9482 
9483 	if (vcpu->arch.exception.pending)
9484 		return true;
9485 
9486 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9487 	    (vcpu->arch.nmi_pending &&
9488 	     kvm_x86_ops->nmi_allowed(vcpu)))
9489 		return true;
9490 
9491 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9492 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9493 		return true;
9494 
9495 	if (kvm_arch_interrupt_allowed(vcpu) &&
9496 	    (kvm_cpu_has_interrupt(vcpu) ||
9497 	    kvm_guest_apic_has_interrupt(vcpu)))
9498 		return true;
9499 
9500 	if (kvm_hv_has_stimer_pending(vcpu))
9501 		return true;
9502 
9503 	return false;
9504 }
9505 
9506 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9507 {
9508 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9509 }
9510 
9511 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9512 {
9513 	return vcpu->arch.preempted_in_kernel;
9514 }
9515 
9516 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9517 {
9518 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9519 }
9520 
9521 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9522 {
9523 	return kvm_x86_ops->interrupt_allowed(vcpu);
9524 }
9525 
9526 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9527 {
9528 	if (is_64_bit_mode(vcpu))
9529 		return kvm_rip_read(vcpu);
9530 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9531 		     kvm_rip_read(vcpu));
9532 }
9533 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9534 
9535 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9536 {
9537 	return kvm_get_linear_rip(vcpu) == linear_rip;
9538 }
9539 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9540 
9541 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9542 {
9543 	unsigned long rflags;
9544 
9545 	rflags = kvm_x86_ops->get_rflags(vcpu);
9546 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9547 		rflags &= ~X86_EFLAGS_TF;
9548 	return rflags;
9549 }
9550 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9551 
9552 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9553 {
9554 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9555 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9556 		rflags |= X86_EFLAGS_TF;
9557 	kvm_x86_ops->set_rflags(vcpu, rflags);
9558 }
9559 
9560 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9561 {
9562 	__kvm_set_rflags(vcpu, rflags);
9563 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9564 }
9565 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9566 
9567 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9568 {
9569 	int r;
9570 
9571 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9572 	      work->wakeup_all)
9573 		return;
9574 
9575 	r = kvm_mmu_reload(vcpu);
9576 	if (unlikely(r))
9577 		return;
9578 
9579 	if (!vcpu->arch.mmu->direct_map &&
9580 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9581 		return;
9582 
9583 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9584 }
9585 
9586 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9587 {
9588 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9589 }
9590 
9591 static inline u32 kvm_async_pf_next_probe(u32 key)
9592 {
9593 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9594 }
9595 
9596 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9597 {
9598 	u32 key = kvm_async_pf_hash_fn(gfn);
9599 
9600 	while (vcpu->arch.apf.gfns[key] != ~0)
9601 		key = kvm_async_pf_next_probe(key);
9602 
9603 	vcpu->arch.apf.gfns[key] = gfn;
9604 }
9605 
9606 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9607 {
9608 	int i;
9609 	u32 key = kvm_async_pf_hash_fn(gfn);
9610 
9611 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9612 		     (vcpu->arch.apf.gfns[key] != gfn &&
9613 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9614 		key = kvm_async_pf_next_probe(key);
9615 
9616 	return key;
9617 }
9618 
9619 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9620 {
9621 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9622 }
9623 
9624 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9625 {
9626 	u32 i, j, k;
9627 
9628 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9629 	while (true) {
9630 		vcpu->arch.apf.gfns[i] = ~0;
9631 		do {
9632 			j = kvm_async_pf_next_probe(j);
9633 			if (vcpu->arch.apf.gfns[j] == ~0)
9634 				return;
9635 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9636 			/*
9637 			 * k lies cyclically in ]i,j]
9638 			 * |    i.k.j |
9639 			 * |....j i.k.| or  |.k..j i...|
9640 			 */
9641 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9642 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9643 		i = j;
9644 	}
9645 }
9646 
9647 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9648 {
9649 
9650 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9651 				      sizeof(val));
9652 }
9653 
9654 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9655 {
9656 
9657 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9658 				      sizeof(u32));
9659 }
9660 
9661 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9662 				     struct kvm_async_pf *work)
9663 {
9664 	struct x86_exception fault;
9665 
9666 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9667 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9668 
9669 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9670 	    (vcpu->arch.apf.send_user_only &&
9671 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9672 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9673 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9674 		fault.vector = PF_VECTOR;
9675 		fault.error_code_valid = true;
9676 		fault.error_code = 0;
9677 		fault.nested_page_fault = false;
9678 		fault.address = work->arch.token;
9679 		fault.async_page_fault = true;
9680 		kvm_inject_page_fault(vcpu, &fault);
9681 	}
9682 }
9683 
9684 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9685 				 struct kvm_async_pf *work)
9686 {
9687 	struct x86_exception fault;
9688 	u32 val;
9689 
9690 	if (work->wakeup_all)
9691 		work->arch.token = ~0; /* broadcast wakeup */
9692 	else
9693 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9694 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9695 
9696 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9697 	    !apf_get_user(vcpu, &val)) {
9698 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9699 		    vcpu->arch.exception.pending &&
9700 		    vcpu->arch.exception.nr == PF_VECTOR &&
9701 		    !apf_put_user(vcpu, 0)) {
9702 			vcpu->arch.exception.injected = false;
9703 			vcpu->arch.exception.pending = false;
9704 			vcpu->arch.exception.nr = 0;
9705 			vcpu->arch.exception.has_error_code = false;
9706 			vcpu->arch.exception.error_code = 0;
9707 			vcpu->arch.exception.has_payload = false;
9708 			vcpu->arch.exception.payload = 0;
9709 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9710 			fault.vector = PF_VECTOR;
9711 			fault.error_code_valid = true;
9712 			fault.error_code = 0;
9713 			fault.nested_page_fault = false;
9714 			fault.address = work->arch.token;
9715 			fault.async_page_fault = true;
9716 			kvm_inject_page_fault(vcpu, &fault);
9717 		}
9718 	}
9719 	vcpu->arch.apf.halted = false;
9720 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9721 }
9722 
9723 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9724 {
9725 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9726 		return true;
9727 	else
9728 		return kvm_can_do_async_pf(vcpu);
9729 }
9730 
9731 void kvm_arch_start_assignment(struct kvm *kvm)
9732 {
9733 	atomic_inc(&kvm->arch.assigned_device_count);
9734 }
9735 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9736 
9737 void kvm_arch_end_assignment(struct kvm *kvm)
9738 {
9739 	atomic_dec(&kvm->arch.assigned_device_count);
9740 }
9741 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9742 
9743 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9744 {
9745 	return atomic_read(&kvm->arch.assigned_device_count);
9746 }
9747 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9748 
9749 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9750 {
9751 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9752 }
9753 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9754 
9755 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9756 {
9757 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9758 }
9759 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9760 
9761 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9762 {
9763 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9764 }
9765 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9766 
9767 bool kvm_arch_has_irq_bypass(void)
9768 {
9769 	return kvm_x86_ops->update_pi_irte != NULL;
9770 }
9771 
9772 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9773 				      struct irq_bypass_producer *prod)
9774 {
9775 	struct kvm_kernel_irqfd *irqfd =
9776 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9777 
9778 	irqfd->producer = prod;
9779 
9780 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9781 					   prod->irq, irqfd->gsi, 1);
9782 }
9783 
9784 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9785 				      struct irq_bypass_producer *prod)
9786 {
9787 	int ret;
9788 	struct kvm_kernel_irqfd *irqfd =
9789 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9790 
9791 	WARN_ON(irqfd->producer != prod);
9792 	irqfd->producer = NULL;
9793 
9794 	/*
9795 	 * When producer of consumer is unregistered, we change back to
9796 	 * remapped mode, so we can re-use the current implementation
9797 	 * when the irq is masked/disabled or the consumer side (KVM
9798 	 * int this case doesn't want to receive the interrupts.
9799 	*/
9800 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9801 	if (ret)
9802 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9803 		       " fails: %d\n", irqfd->consumer.token, ret);
9804 }
9805 
9806 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9807 				   uint32_t guest_irq, bool set)
9808 {
9809 	if (!kvm_x86_ops->update_pi_irte)
9810 		return -EINVAL;
9811 
9812 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9813 }
9814 
9815 bool kvm_vector_hashing_enabled(void)
9816 {
9817 	return vector_hashing;
9818 }
9819 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9820 
9821 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9822 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9823 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9824 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9825 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9826 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9827 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9828 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9829 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9830 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9831 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9832 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9833 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9834 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9835 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9836 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9837 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9838 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9839 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9840