xref: /openbmc/linux/arch/x86/kvm/x86.c (revision a01822e94ee53e8ebc9632fe2764048b81921254)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "mmu.h"
22 #include "i8254.h"
23 #include "tss.h"
24 #include "kvm_cache_regs.h"
25 #include "kvm_emulate.h"
26 #include "x86.h"
27 #include "cpuid.h"
28 #include "pmu.h"
29 #include "hyperv.h"
30 #include "lapic.h"
31 
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
35 #include <linux/fs.h>
36 #include <linux/vmalloc.h>
37 #include <linux/export.h>
38 #include <linux/moduleparam.h>
39 #include <linux/mman.h>
40 #include <linux/highmem.h>
41 #include <linux/iommu.h>
42 #include <linux/intel-iommu.h>
43 #include <linux/cpufreq.h>
44 #include <linux/user-return-notifier.h>
45 #include <linux/srcu.h>
46 #include <linux/slab.h>
47 #include <linux/perf_event.h>
48 #include <linux/uaccess.h>
49 #include <linux/hash.h>
50 #include <linux/pci.h>
51 #include <linux/timekeeper_internal.h>
52 #include <linux/pvclock_gtod.h>
53 #include <linux/kvm_irqfd.h>
54 #include <linux/irqbypass.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/isolation.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 #include <asm/emulate_prefix.h>
74 #include <clocksource/hyperv_timer.h>
75 
76 #define CREATE_TRACE_POINTS
77 #include "trace.h"
78 
79 #define MAX_IO_MSRS 256
80 #define KVM_MAX_MCE_BANKS 32
81 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
82 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
83 
84 #define emul_to_vcpu(ctxt) \
85 	((struct kvm_vcpu *)(ctxt)->vcpu)
86 
87 /* EFER defaults:
88  * - enable syscall per default because its emulated by KVM
89  * - enable LME and LMA per default on 64 bit KVM
90  */
91 #ifdef CONFIG_X86_64
92 static
93 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
94 #else
95 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 #endif
97 
98 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
99 
100 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
101 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
102 
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
105 
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void enter_smm(struct kvm_vcpu *vcpu);
109 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
110 static void store_regs(struct kvm_vcpu *vcpu);
111 static int sync_regs(struct kvm_vcpu *vcpu);
112 
113 struct kvm_x86_ops kvm_x86_ops __read_mostly;
114 EXPORT_SYMBOL_GPL(kvm_x86_ops);
115 
116 static bool __read_mostly ignore_msrs = 0;
117 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
118 
119 static bool __read_mostly report_ignored_msrs = true;
120 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
121 
122 unsigned int min_timer_period_us = 200;
123 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
124 
125 static bool __read_mostly kvmclock_periodic_sync = true;
126 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
127 
128 bool __read_mostly kvm_has_tsc_control;
129 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
130 u32  __read_mostly kvm_max_guest_tsc_khz;
131 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
132 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
133 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
134 u64  __read_mostly kvm_max_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
136 u64 __read_mostly kvm_default_tsc_scaling_ratio;
137 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
138 
139 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
140 static u32 __read_mostly tsc_tolerance_ppm = 250;
141 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
142 
143 /*
144  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
145  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
146  * advancement entirely.  Any other value is used as-is and disables adaptive
147  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
148  */
149 static int __read_mostly lapic_timer_advance_ns = -1;
150 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
151 
152 static bool __read_mostly vector_hashing = true;
153 module_param(vector_hashing, bool, S_IRUGO);
154 
155 bool __read_mostly enable_vmware_backdoor = false;
156 module_param(enable_vmware_backdoor, bool, S_IRUGO);
157 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
158 
159 static bool __read_mostly force_emulation_prefix = false;
160 module_param(force_emulation_prefix, bool, S_IRUGO);
161 
162 int __read_mostly pi_inject_timer = -1;
163 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
164 
165 #define KVM_NR_SHARED_MSRS 16
166 
167 struct kvm_shared_msrs_global {
168 	int nr;
169 	u32 msrs[KVM_NR_SHARED_MSRS];
170 };
171 
172 struct kvm_shared_msrs {
173 	struct user_return_notifier urn;
174 	bool registered;
175 	struct kvm_shared_msr_values {
176 		u64 host;
177 		u64 curr;
178 	} values[KVM_NR_SHARED_MSRS];
179 };
180 
181 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
182 static struct kvm_shared_msrs __percpu *shared_msrs;
183 
184 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
185 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
186 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
187 				| XFEATURE_MASK_PKRU)
188 
189 u64 __read_mostly host_efer;
190 EXPORT_SYMBOL_GPL(host_efer);
191 
192 static u64 __read_mostly host_xss;
193 u64 __read_mostly supported_xss;
194 EXPORT_SYMBOL_GPL(supported_xss);
195 
196 struct kvm_stats_debugfs_item debugfs_entries[] = {
197 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
198 	{ "pf_guest", VCPU_STAT(pf_guest) },
199 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
200 	{ "invlpg", VCPU_STAT(invlpg) },
201 	{ "exits", VCPU_STAT(exits) },
202 	{ "io_exits", VCPU_STAT(io_exits) },
203 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
204 	{ "signal_exits", VCPU_STAT(signal_exits) },
205 	{ "irq_window", VCPU_STAT(irq_window_exits) },
206 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
207 	{ "halt_exits", VCPU_STAT(halt_exits) },
208 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
209 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
210 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
211 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
212 	{ "hypercalls", VCPU_STAT(hypercalls) },
213 	{ "request_irq", VCPU_STAT(request_irq_exits) },
214 	{ "irq_exits", VCPU_STAT(irq_exits) },
215 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
216 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
217 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
218 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
219 	{ "irq_injections", VCPU_STAT(irq_injections) },
220 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
221 	{ "req_event", VCPU_STAT(req_event) },
222 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
223 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
224 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
225 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
226 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
227 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
228 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
229 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
230 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
231 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
232 	{ "largepages", VM_STAT(lpages, .mode = 0444) },
233 	{ "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
234 	{ "max_mmu_page_hash_collisions",
235 		VM_STAT(max_mmu_page_hash_collisions) },
236 	{ NULL }
237 };
238 
239 u64 __read_mostly host_xcr0;
240 u64 __read_mostly supported_xcr0;
241 EXPORT_SYMBOL_GPL(supported_xcr0);
242 
243 struct kmem_cache *x86_fpu_cache;
244 EXPORT_SYMBOL_GPL(x86_fpu_cache);
245 
246 static struct kmem_cache *x86_emulator_cache;
247 
248 static struct kmem_cache *kvm_alloc_emulator_cache(void)
249 {
250 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
251 	unsigned int size = sizeof(struct x86_emulate_ctxt);
252 
253 	return kmem_cache_create_usercopy("x86_emulator", size,
254 					  __alignof__(struct x86_emulate_ctxt),
255 					  SLAB_ACCOUNT, useroffset,
256 					  size - useroffset, NULL);
257 }
258 
259 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
260 
261 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
262 {
263 	int i;
264 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
265 		vcpu->arch.apf.gfns[i] = ~0;
266 }
267 
268 static void kvm_on_user_return(struct user_return_notifier *urn)
269 {
270 	unsigned slot;
271 	struct kvm_shared_msrs *locals
272 		= container_of(urn, struct kvm_shared_msrs, urn);
273 	struct kvm_shared_msr_values *values;
274 	unsigned long flags;
275 
276 	/*
277 	 * Disabling irqs at this point since the following code could be
278 	 * interrupted and executed through kvm_arch_hardware_disable()
279 	 */
280 	local_irq_save(flags);
281 	if (locals->registered) {
282 		locals->registered = false;
283 		user_return_notifier_unregister(urn);
284 	}
285 	local_irq_restore(flags);
286 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
287 		values = &locals->values[slot];
288 		if (values->host != values->curr) {
289 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
290 			values->curr = values->host;
291 		}
292 	}
293 }
294 
295 void kvm_define_shared_msr(unsigned slot, u32 msr)
296 {
297 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
298 	shared_msrs_global.msrs[slot] = msr;
299 	if (slot >= shared_msrs_global.nr)
300 		shared_msrs_global.nr = slot + 1;
301 }
302 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
303 
304 static void kvm_shared_msr_cpu_online(void)
305 {
306 	unsigned int cpu = smp_processor_id();
307 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
308 	u64 value;
309 	int i;
310 
311 	for (i = 0; i < shared_msrs_global.nr; ++i) {
312 		rdmsrl_safe(shared_msrs_global.msrs[i], &value);
313 		smsr->values[i].host = value;
314 		smsr->values[i].curr = value;
315 	}
316 }
317 
318 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
319 {
320 	unsigned int cpu = smp_processor_id();
321 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
322 	int err;
323 
324 	value = (value & mask) | (smsr->values[slot].host & ~mask);
325 	if (value == smsr->values[slot].curr)
326 		return 0;
327 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
328 	if (err)
329 		return 1;
330 
331 	smsr->values[slot].curr = value;
332 	if (!smsr->registered) {
333 		smsr->urn.on_user_return = kvm_on_user_return;
334 		user_return_notifier_register(&smsr->urn);
335 		smsr->registered = true;
336 	}
337 	return 0;
338 }
339 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
340 
341 static void drop_user_return_notifiers(void)
342 {
343 	unsigned int cpu = smp_processor_id();
344 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
345 
346 	if (smsr->registered)
347 		kvm_on_user_return(&smsr->urn);
348 }
349 
350 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
351 {
352 	return vcpu->arch.apic_base;
353 }
354 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
355 
356 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
357 {
358 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
359 }
360 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
361 
362 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
363 {
364 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
365 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
366 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
367 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
368 
369 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
370 		return 1;
371 	if (!msr_info->host_initiated) {
372 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
373 			return 1;
374 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
375 			return 1;
376 	}
377 
378 	kvm_lapic_set_base(vcpu, msr_info->data);
379 	kvm_recalculate_apic_map(vcpu->kvm);
380 	return 0;
381 }
382 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
383 
384 asmlinkage __visible void kvm_spurious_fault(void)
385 {
386 	/* Fault while not rebooting.  We want the trace. */
387 	BUG_ON(!kvm_rebooting);
388 }
389 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
390 
391 #define EXCPT_BENIGN		0
392 #define EXCPT_CONTRIBUTORY	1
393 #define EXCPT_PF		2
394 
395 static int exception_class(int vector)
396 {
397 	switch (vector) {
398 	case PF_VECTOR:
399 		return EXCPT_PF;
400 	case DE_VECTOR:
401 	case TS_VECTOR:
402 	case NP_VECTOR:
403 	case SS_VECTOR:
404 	case GP_VECTOR:
405 		return EXCPT_CONTRIBUTORY;
406 	default:
407 		break;
408 	}
409 	return EXCPT_BENIGN;
410 }
411 
412 #define EXCPT_FAULT		0
413 #define EXCPT_TRAP		1
414 #define EXCPT_ABORT		2
415 #define EXCPT_INTERRUPT		3
416 
417 static int exception_type(int vector)
418 {
419 	unsigned int mask;
420 
421 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
422 		return EXCPT_INTERRUPT;
423 
424 	mask = 1 << vector;
425 
426 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
427 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
428 		return EXCPT_TRAP;
429 
430 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
431 		return EXCPT_ABORT;
432 
433 	/* Reserved exceptions will result in fault */
434 	return EXCPT_FAULT;
435 }
436 
437 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
438 {
439 	unsigned nr = vcpu->arch.exception.nr;
440 	bool has_payload = vcpu->arch.exception.has_payload;
441 	unsigned long payload = vcpu->arch.exception.payload;
442 
443 	if (!has_payload)
444 		return;
445 
446 	switch (nr) {
447 	case DB_VECTOR:
448 		/*
449 		 * "Certain debug exceptions may clear bit 0-3.  The
450 		 * remaining contents of the DR6 register are never
451 		 * cleared by the processor".
452 		 */
453 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
454 		/*
455 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
456 		 */
457 		vcpu->arch.dr6 |= DR6_RTM;
458 		vcpu->arch.dr6 |= payload;
459 		/*
460 		 * Bit 16 should be set in the payload whenever the #DB
461 		 * exception should clear DR6.RTM. This makes the payload
462 		 * compatible with the pending debug exceptions under VMX.
463 		 * Though not currently documented in the SDM, this also
464 		 * makes the payload compatible with the exit qualification
465 		 * for #DB exceptions under VMX.
466 		 */
467 		vcpu->arch.dr6 ^= payload & DR6_RTM;
468 
469 		/*
470 		 * The #DB payload is defined as compatible with the 'pending
471 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
472 		 * defined in the 'pending debug exceptions' field (enabled
473 		 * breakpoint), it is reserved and must be zero in DR6.
474 		 */
475 		vcpu->arch.dr6 &= ~BIT(12);
476 		break;
477 	case PF_VECTOR:
478 		vcpu->arch.cr2 = payload;
479 		break;
480 	}
481 
482 	vcpu->arch.exception.has_payload = false;
483 	vcpu->arch.exception.payload = 0;
484 }
485 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
486 
487 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
488 		unsigned nr, bool has_error, u32 error_code,
489 	        bool has_payload, unsigned long payload, bool reinject)
490 {
491 	u32 prev_nr;
492 	int class1, class2;
493 
494 	kvm_make_request(KVM_REQ_EVENT, vcpu);
495 
496 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
497 	queue:
498 		if (has_error && !is_protmode(vcpu))
499 			has_error = false;
500 		if (reinject) {
501 			/*
502 			 * On vmentry, vcpu->arch.exception.pending is only
503 			 * true if an event injection was blocked by
504 			 * nested_run_pending.  In that case, however,
505 			 * vcpu_enter_guest requests an immediate exit,
506 			 * and the guest shouldn't proceed far enough to
507 			 * need reinjection.
508 			 */
509 			WARN_ON_ONCE(vcpu->arch.exception.pending);
510 			vcpu->arch.exception.injected = true;
511 			if (WARN_ON_ONCE(has_payload)) {
512 				/*
513 				 * A reinjected event has already
514 				 * delivered its payload.
515 				 */
516 				has_payload = false;
517 				payload = 0;
518 			}
519 		} else {
520 			vcpu->arch.exception.pending = true;
521 			vcpu->arch.exception.injected = false;
522 		}
523 		vcpu->arch.exception.has_error_code = has_error;
524 		vcpu->arch.exception.nr = nr;
525 		vcpu->arch.exception.error_code = error_code;
526 		vcpu->arch.exception.has_payload = has_payload;
527 		vcpu->arch.exception.payload = payload;
528 		if (!is_guest_mode(vcpu))
529 			kvm_deliver_exception_payload(vcpu);
530 		return;
531 	}
532 
533 	/* to check exception */
534 	prev_nr = vcpu->arch.exception.nr;
535 	if (prev_nr == DF_VECTOR) {
536 		/* triple fault -> shutdown */
537 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
538 		return;
539 	}
540 	class1 = exception_class(prev_nr);
541 	class2 = exception_class(nr);
542 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
543 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
544 		/*
545 		 * Generate double fault per SDM Table 5-5.  Set
546 		 * exception.pending = true so that the double fault
547 		 * can trigger a nested vmexit.
548 		 */
549 		vcpu->arch.exception.pending = true;
550 		vcpu->arch.exception.injected = false;
551 		vcpu->arch.exception.has_error_code = true;
552 		vcpu->arch.exception.nr = DF_VECTOR;
553 		vcpu->arch.exception.error_code = 0;
554 		vcpu->arch.exception.has_payload = false;
555 		vcpu->arch.exception.payload = 0;
556 	} else
557 		/* replace previous exception with a new one in a hope
558 		   that instruction re-execution will regenerate lost
559 		   exception */
560 		goto queue;
561 }
562 
563 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
564 {
565 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
566 }
567 EXPORT_SYMBOL_GPL(kvm_queue_exception);
568 
569 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
570 {
571 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
572 }
573 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
574 
575 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
576 			   unsigned long payload)
577 {
578 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
579 }
580 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
581 
582 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
583 				    u32 error_code, unsigned long payload)
584 {
585 	kvm_multiple_exception(vcpu, nr, true, error_code,
586 			       true, payload, false);
587 }
588 
589 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
590 {
591 	if (err)
592 		kvm_inject_gp(vcpu, 0);
593 	else
594 		return kvm_skip_emulated_instruction(vcpu);
595 
596 	return 1;
597 }
598 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
599 
600 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
601 {
602 	++vcpu->stat.pf_guest;
603 	vcpu->arch.exception.nested_apf =
604 		is_guest_mode(vcpu) && fault->async_page_fault;
605 	if (vcpu->arch.exception.nested_apf) {
606 		vcpu->arch.apf.nested_apf_token = fault->address;
607 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
608 	} else {
609 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
610 					fault->address);
611 	}
612 }
613 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
614 
615 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
616 {
617 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
618 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
619 	else
620 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
621 
622 	return fault->nested_page_fault;
623 }
624 
625 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
626 {
627 	atomic_inc(&vcpu->arch.nmi_queued);
628 	kvm_make_request(KVM_REQ_NMI, vcpu);
629 }
630 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
631 
632 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
633 {
634 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
635 }
636 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
637 
638 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
639 {
640 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
641 }
642 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
643 
644 /*
645  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
646  * a #GP and return false.
647  */
648 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
649 {
650 	if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
651 		return true;
652 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
653 	return false;
654 }
655 EXPORT_SYMBOL_GPL(kvm_require_cpl);
656 
657 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
658 {
659 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
660 		return true;
661 
662 	kvm_queue_exception(vcpu, UD_VECTOR);
663 	return false;
664 }
665 EXPORT_SYMBOL_GPL(kvm_require_dr);
666 
667 /*
668  * This function will be used to read from the physical memory of the currently
669  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
670  * can read from guest physical or from the guest's guest physical memory.
671  */
672 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
673 			    gfn_t ngfn, void *data, int offset, int len,
674 			    u32 access)
675 {
676 	struct x86_exception exception;
677 	gfn_t real_gfn;
678 	gpa_t ngpa;
679 
680 	ngpa     = gfn_to_gpa(ngfn);
681 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
682 	if (real_gfn == UNMAPPED_GVA)
683 		return -EFAULT;
684 
685 	real_gfn = gpa_to_gfn(real_gfn);
686 
687 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
688 }
689 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
690 
691 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
692 			       void *data, int offset, int len, u32 access)
693 {
694 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
695 				       data, offset, len, access);
696 }
697 
698 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
699 {
700 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
701 	       rsvd_bits(1, 2);
702 }
703 
704 /*
705  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
706  */
707 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
708 {
709 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
710 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
711 	int i;
712 	int ret;
713 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
714 
715 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
716 				      offset * sizeof(u64), sizeof(pdpte),
717 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
718 	if (ret < 0) {
719 		ret = 0;
720 		goto out;
721 	}
722 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
723 		if ((pdpte[i] & PT_PRESENT_MASK) &&
724 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
725 			ret = 0;
726 			goto out;
727 		}
728 	}
729 	ret = 1;
730 
731 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
732 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
733 
734 out:
735 
736 	return ret;
737 }
738 EXPORT_SYMBOL_GPL(load_pdptrs);
739 
740 bool pdptrs_changed(struct kvm_vcpu *vcpu)
741 {
742 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
743 	int offset;
744 	gfn_t gfn;
745 	int r;
746 
747 	if (!is_pae_paging(vcpu))
748 		return false;
749 
750 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
751 		return true;
752 
753 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
754 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
755 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
756 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
757 	if (r < 0)
758 		return true;
759 
760 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
761 }
762 EXPORT_SYMBOL_GPL(pdptrs_changed);
763 
764 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
765 {
766 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
767 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
768 
769 	cr0 |= X86_CR0_ET;
770 
771 #ifdef CONFIG_X86_64
772 	if (cr0 & 0xffffffff00000000UL)
773 		return 1;
774 #endif
775 
776 	cr0 &= ~CR0_RESERVED_BITS;
777 
778 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
779 		return 1;
780 
781 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
782 		return 1;
783 
784 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
785 #ifdef CONFIG_X86_64
786 		if ((vcpu->arch.efer & EFER_LME)) {
787 			int cs_db, cs_l;
788 
789 			if (!is_pae(vcpu))
790 				return 1;
791 			kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
792 			if (cs_l)
793 				return 1;
794 		} else
795 #endif
796 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
797 						 kvm_read_cr3(vcpu)))
798 			return 1;
799 	}
800 
801 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
802 		return 1;
803 
804 	kvm_x86_ops.set_cr0(vcpu, cr0);
805 
806 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
807 		kvm_clear_async_pf_completion_queue(vcpu);
808 		kvm_async_pf_hash_reset(vcpu);
809 	}
810 
811 	if ((cr0 ^ old_cr0) & update_bits)
812 		kvm_mmu_reset_context(vcpu);
813 
814 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
815 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
816 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
817 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
818 
819 	return 0;
820 }
821 EXPORT_SYMBOL_GPL(kvm_set_cr0);
822 
823 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
824 {
825 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
826 }
827 EXPORT_SYMBOL_GPL(kvm_lmsw);
828 
829 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
830 {
831 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
832 
833 		if (vcpu->arch.xcr0 != host_xcr0)
834 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
835 
836 		if (vcpu->arch.xsaves_enabled &&
837 		    vcpu->arch.ia32_xss != host_xss)
838 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
839 	}
840 
841 	if (static_cpu_has(X86_FEATURE_PKU) &&
842 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
843 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
844 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
845 		__write_pkru(vcpu->arch.pkru);
846 }
847 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
848 
849 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
850 {
851 	if (static_cpu_has(X86_FEATURE_PKU) &&
852 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
853 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
854 		vcpu->arch.pkru = rdpkru();
855 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
856 			__write_pkru(vcpu->arch.host_pkru);
857 	}
858 
859 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
860 
861 		if (vcpu->arch.xcr0 != host_xcr0)
862 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
863 
864 		if (vcpu->arch.xsaves_enabled &&
865 		    vcpu->arch.ia32_xss != host_xss)
866 			wrmsrl(MSR_IA32_XSS, host_xss);
867 	}
868 
869 }
870 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
871 
872 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
873 {
874 	u64 xcr0 = xcr;
875 	u64 old_xcr0 = vcpu->arch.xcr0;
876 	u64 valid_bits;
877 
878 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
879 	if (index != XCR_XFEATURE_ENABLED_MASK)
880 		return 1;
881 	if (!(xcr0 & XFEATURE_MASK_FP))
882 		return 1;
883 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
884 		return 1;
885 
886 	/*
887 	 * Do not allow the guest to set bits that we do not support
888 	 * saving.  However, xcr0 bit 0 is always set, even if the
889 	 * emulated CPU does not support XSAVE (see fx_init).
890 	 */
891 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
892 	if (xcr0 & ~valid_bits)
893 		return 1;
894 
895 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
896 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
897 		return 1;
898 
899 	if (xcr0 & XFEATURE_MASK_AVX512) {
900 		if (!(xcr0 & XFEATURE_MASK_YMM))
901 			return 1;
902 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
903 			return 1;
904 	}
905 	vcpu->arch.xcr0 = xcr0;
906 
907 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
908 		kvm_update_cpuid(vcpu);
909 	return 0;
910 }
911 
912 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
913 {
914 	if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
915 	    __kvm_set_xcr(vcpu, index, xcr)) {
916 		kvm_inject_gp(vcpu, 0);
917 		return 1;
918 	}
919 	return 0;
920 }
921 EXPORT_SYMBOL_GPL(kvm_set_xcr);
922 
923 #define __cr4_reserved_bits(__cpu_has, __c)		\
924 ({							\
925 	u64 __reserved_bits = CR4_RESERVED_BITS;	\
926 							\
927 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))		\
928 		__reserved_bits |= X86_CR4_OSXSAVE;	\
929 	if (!__cpu_has(__c, X86_FEATURE_SMEP))		\
930 		__reserved_bits |= X86_CR4_SMEP;	\
931 	if (!__cpu_has(__c, X86_FEATURE_SMAP))		\
932 		__reserved_bits |= X86_CR4_SMAP;	\
933 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))	\
934 		__reserved_bits |= X86_CR4_FSGSBASE;	\
935 	if (!__cpu_has(__c, X86_FEATURE_PKU))		\
936 		__reserved_bits |= X86_CR4_PKE;		\
937 	if (!__cpu_has(__c, X86_FEATURE_LA57))		\
938 		__reserved_bits |= X86_CR4_LA57;	\
939 	if (!__cpu_has(__c, X86_FEATURE_UMIP))		\
940 		__reserved_bits |= X86_CR4_UMIP;	\
941 	__reserved_bits;				\
942 })
943 
944 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
945 {
946 	if (cr4 & cr4_reserved_bits)
947 		return -EINVAL;
948 
949 	if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
950 		return -EINVAL;
951 
952 	return 0;
953 }
954 
955 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
956 {
957 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
958 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
959 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
960 
961 	if (kvm_valid_cr4(vcpu, cr4))
962 		return 1;
963 
964 	if (is_long_mode(vcpu)) {
965 		if (!(cr4 & X86_CR4_PAE))
966 			return 1;
967 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
968 		   && ((cr4 ^ old_cr4) & pdptr_bits)
969 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
970 				   kvm_read_cr3(vcpu)))
971 		return 1;
972 
973 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
974 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
975 			return 1;
976 
977 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
978 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
979 			return 1;
980 	}
981 
982 	if (kvm_x86_ops.set_cr4(vcpu, cr4))
983 		return 1;
984 
985 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
986 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
987 		kvm_mmu_reset_context(vcpu);
988 
989 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
990 		kvm_update_cpuid(vcpu);
991 
992 	return 0;
993 }
994 EXPORT_SYMBOL_GPL(kvm_set_cr4);
995 
996 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
997 {
998 	bool skip_tlb_flush = false;
999 #ifdef CONFIG_X86_64
1000 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1001 
1002 	if (pcid_enabled) {
1003 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1004 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1005 	}
1006 #endif
1007 
1008 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1009 		if (!skip_tlb_flush) {
1010 			kvm_mmu_sync_roots(vcpu);
1011 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1012 		}
1013 		return 0;
1014 	}
1015 
1016 	if (is_long_mode(vcpu) &&
1017 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1018 		return 1;
1019 	else if (is_pae_paging(vcpu) &&
1020 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1021 		return 1;
1022 
1023 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
1024 	vcpu->arch.cr3 = cr3;
1025 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1026 
1027 	return 0;
1028 }
1029 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1030 
1031 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1032 {
1033 	if (cr8 & CR8_RESERVED_BITS)
1034 		return 1;
1035 	if (lapic_in_kernel(vcpu))
1036 		kvm_lapic_set_tpr(vcpu, cr8);
1037 	else
1038 		vcpu->arch.cr8 = cr8;
1039 	return 0;
1040 }
1041 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1042 
1043 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1044 {
1045 	if (lapic_in_kernel(vcpu))
1046 		return kvm_lapic_get_cr8(vcpu);
1047 	else
1048 		return vcpu->arch.cr8;
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1051 
1052 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1053 {
1054 	int i;
1055 
1056 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1057 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1058 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1059 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1060 	}
1061 }
1062 
1063 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1064 {
1065 	unsigned long dr7;
1066 
1067 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1068 		dr7 = vcpu->arch.guest_debug_dr7;
1069 	else
1070 		dr7 = vcpu->arch.dr7;
1071 	kvm_x86_ops.set_dr7(vcpu, dr7);
1072 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1073 	if (dr7 & DR7_BP_EN_MASK)
1074 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1075 }
1076 
1077 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1078 {
1079 	u64 fixed = DR6_FIXED_1;
1080 
1081 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1082 		fixed |= DR6_RTM;
1083 	return fixed;
1084 }
1085 
1086 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1087 {
1088 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1089 
1090 	switch (dr) {
1091 	case 0 ... 3:
1092 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1093 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1094 			vcpu->arch.eff_db[dr] = val;
1095 		break;
1096 	case 4:
1097 		/* fall through */
1098 	case 6:
1099 		if (val & 0xffffffff00000000ULL)
1100 			return -1; /* #GP */
1101 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1102 		break;
1103 	case 5:
1104 		/* fall through */
1105 	default: /* 7 */
1106 		if (!kvm_dr7_valid(val))
1107 			return -1; /* #GP */
1108 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1109 		kvm_update_dr7(vcpu);
1110 		break;
1111 	}
1112 
1113 	return 0;
1114 }
1115 
1116 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1117 {
1118 	if (__kvm_set_dr(vcpu, dr, val)) {
1119 		kvm_inject_gp(vcpu, 0);
1120 		return 1;
1121 	}
1122 	return 0;
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_set_dr);
1125 
1126 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1127 {
1128 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1129 
1130 	switch (dr) {
1131 	case 0 ... 3:
1132 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1133 		break;
1134 	case 4:
1135 		/* fall through */
1136 	case 6:
1137 		*val = vcpu->arch.dr6;
1138 		break;
1139 	case 5:
1140 		/* fall through */
1141 	default: /* 7 */
1142 		*val = vcpu->arch.dr7;
1143 		break;
1144 	}
1145 	return 0;
1146 }
1147 EXPORT_SYMBOL_GPL(kvm_get_dr);
1148 
1149 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1150 {
1151 	u32 ecx = kvm_rcx_read(vcpu);
1152 	u64 data;
1153 	int err;
1154 
1155 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1156 	if (err)
1157 		return err;
1158 	kvm_rax_write(vcpu, (u32)data);
1159 	kvm_rdx_write(vcpu, data >> 32);
1160 	return err;
1161 }
1162 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1163 
1164 /*
1165  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1166  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1167  *
1168  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1169  * extract the supported MSRs from the related const lists.
1170  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1171  * capabilities of the host cpu. This capabilities test skips MSRs that are
1172  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1173  * may depend on host virtualization features rather than host cpu features.
1174  */
1175 
1176 static const u32 msrs_to_save_all[] = {
1177 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1178 	MSR_STAR,
1179 #ifdef CONFIG_X86_64
1180 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1181 #endif
1182 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1183 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1184 	MSR_IA32_SPEC_CTRL,
1185 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1186 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1187 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1188 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1189 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1190 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1191 	MSR_IA32_UMWAIT_CONTROL,
1192 
1193 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1194 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1195 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1196 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1197 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1198 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1199 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1200 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1201 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1202 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1203 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1204 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1205 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1206 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1207 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1208 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1209 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1210 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1211 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1212 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1213 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1214 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1215 };
1216 
1217 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1218 static unsigned num_msrs_to_save;
1219 
1220 static const u32 emulated_msrs_all[] = {
1221 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1222 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1223 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1224 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1225 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1226 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1227 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1228 	HV_X64_MSR_RESET,
1229 	HV_X64_MSR_VP_INDEX,
1230 	HV_X64_MSR_VP_RUNTIME,
1231 	HV_X64_MSR_SCONTROL,
1232 	HV_X64_MSR_STIMER0_CONFIG,
1233 	HV_X64_MSR_VP_ASSIST_PAGE,
1234 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1235 	HV_X64_MSR_TSC_EMULATION_STATUS,
1236 
1237 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1238 	MSR_KVM_PV_EOI_EN,
1239 
1240 	MSR_IA32_TSC_ADJUST,
1241 	MSR_IA32_TSCDEADLINE,
1242 	MSR_IA32_ARCH_CAPABILITIES,
1243 	MSR_IA32_MISC_ENABLE,
1244 	MSR_IA32_MCG_STATUS,
1245 	MSR_IA32_MCG_CTL,
1246 	MSR_IA32_MCG_EXT_CTL,
1247 	MSR_IA32_SMBASE,
1248 	MSR_SMI_COUNT,
1249 	MSR_PLATFORM_INFO,
1250 	MSR_MISC_FEATURES_ENABLES,
1251 	MSR_AMD64_VIRT_SPEC_CTRL,
1252 	MSR_IA32_POWER_CTL,
1253 	MSR_IA32_UCODE_REV,
1254 
1255 	/*
1256 	 * The following list leaves out MSRs whose values are determined
1257 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1258 	 * We always support the "true" VMX control MSRs, even if the host
1259 	 * processor does not, so I am putting these registers here rather
1260 	 * than in msrs_to_save_all.
1261 	 */
1262 	MSR_IA32_VMX_BASIC,
1263 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1264 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1265 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1266 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1267 	MSR_IA32_VMX_MISC,
1268 	MSR_IA32_VMX_CR0_FIXED0,
1269 	MSR_IA32_VMX_CR4_FIXED0,
1270 	MSR_IA32_VMX_VMCS_ENUM,
1271 	MSR_IA32_VMX_PROCBASED_CTLS2,
1272 	MSR_IA32_VMX_EPT_VPID_CAP,
1273 	MSR_IA32_VMX_VMFUNC,
1274 
1275 	MSR_K7_HWCR,
1276 	MSR_KVM_POLL_CONTROL,
1277 };
1278 
1279 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1280 static unsigned num_emulated_msrs;
1281 
1282 /*
1283  * List of msr numbers which are used to expose MSR-based features that
1284  * can be used by a hypervisor to validate requested CPU features.
1285  */
1286 static const u32 msr_based_features_all[] = {
1287 	MSR_IA32_VMX_BASIC,
1288 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1289 	MSR_IA32_VMX_PINBASED_CTLS,
1290 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1291 	MSR_IA32_VMX_PROCBASED_CTLS,
1292 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1293 	MSR_IA32_VMX_EXIT_CTLS,
1294 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1295 	MSR_IA32_VMX_ENTRY_CTLS,
1296 	MSR_IA32_VMX_MISC,
1297 	MSR_IA32_VMX_CR0_FIXED0,
1298 	MSR_IA32_VMX_CR0_FIXED1,
1299 	MSR_IA32_VMX_CR4_FIXED0,
1300 	MSR_IA32_VMX_CR4_FIXED1,
1301 	MSR_IA32_VMX_VMCS_ENUM,
1302 	MSR_IA32_VMX_PROCBASED_CTLS2,
1303 	MSR_IA32_VMX_EPT_VPID_CAP,
1304 	MSR_IA32_VMX_VMFUNC,
1305 
1306 	MSR_F10H_DECFG,
1307 	MSR_IA32_UCODE_REV,
1308 	MSR_IA32_ARCH_CAPABILITIES,
1309 };
1310 
1311 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1312 static unsigned int num_msr_based_features;
1313 
1314 static u64 kvm_get_arch_capabilities(void)
1315 {
1316 	u64 data = 0;
1317 
1318 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1319 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1320 
1321 	/*
1322 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1323 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1324 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1325 	 * L1 guests, so it need not worry about its own (L2) guests.
1326 	 */
1327 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1328 
1329 	/*
1330 	 * If we're doing cache flushes (either "always" or "cond")
1331 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1332 	 * If an outer hypervisor is doing the cache flush for us
1333 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1334 	 * capability to the guest too, and if EPT is disabled we're not
1335 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1336 	 * require a nested hypervisor to do a flush of its own.
1337 	 */
1338 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1339 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1340 
1341 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1342 		data |= ARCH_CAP_RDCL_NO;
1343 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1344 		data |= ARCH_CAP_SSB_NO;
1345 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1346 		data |= ARCH_CAP_MDS_NO;
1347 
1348 	/*
1349 	 * On TAA affected systems:
1350 	 *      - nothing to do if TSX is disabled on the host.
1351 	 *      - we emulate TSX_CTRL if present on the host.
1352 	 *	  This lets the guest use VERW to clear CPU buffers.
1353 	 */
1354 	if (!boot_cpu_has(X86_FEATURE_RTM))
1355 		data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1356 	else if (!boot_cpu_has_bug(X86_BUG_TAA))
1357 		data |= ARCH_CAP_TAA_NO;
1358 
1359 	return data;
1360 }
1361 
1362 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1363 {
1364 	switch (msr->index) {
1365 	case MSR_IA32_ARCH_CAPABILITIES:
1366 		msr->data = kvm_get_arch_capabilities();
1367 		break;
1368 	case MSR_IA32_UCODE_REV:
1369 		rdmsrl_safe(msr->index, &msr->data);
1370 		break;
1371 	default:
1372 		if (kvm_x86_ops.get_msr_feature(msr))
1373 			return 1;
1374 	}
1375 	return 0;
1376 }
1377 
1378 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1379 {
1380 	struct kvm_msr_entry msr;
1381 	int r;
1382 
1383 	msr.index = index;
1384 	r = kvm_get_msr_feature(&msr);
1385 	if (r)
1386 		return r;
1387 
1388 	*data = msr.data;
1389 
1390 	return 0;
1391 }
1392 
1393 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1394 {
1395 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1396 		return false;
1397 
1398 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1399 		return false;
1400 
1401 	if (efer & (EFER_LME | EFER_LMA) &&
1402 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1403 		return false;
1404 
1405 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1406 		return false;
1407 
1408 	return true;
1409 
1410 }
1411 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1412 {
1413 	if (efer & efer_reserved_bits)
1414 		return false;
1415 
1416 	return __kvm_valid_efer(vcpu, efer);
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1419 
1420 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1421 {
1422 	u64 old_efer = vcpu->arch.efer;
1423 	u64 efer = msr_info->data;
1424 
1425 	if (efer & efer_reserved_bits)
1426 		return 1;
1427 
1428 	if (!msr_info->host_initiated) {
1429 		if (!__kvm_valid_efer(vcpu, efer))
1430 			return 1;
1431 
1432 		if (is_paging(vcpu) &&
1433 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1434 			return 1;
1435 	}
1436 
1437 	efer &= ~EFER_LMA;
1438 	efer |= vcpu->arch.efer & EFER_LMA;
1439 
1440 	kvm_x86_ops.set_efer(vcpu, efer);
1441 
1442 	/* Update reserved bits */
1443 	if ((efer ^ old_efer) & EFER_NX)
1444 		kvm_mmu_reset_context(vcpu);
1445 
1446 	return 0;
1447 }
1448 
1449 void kvm_enable_efer_bits(u64 mask)
1450 {
1451        efer_reserved_bits &= ~mask;
1452 }
1453 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1454 
1455 /*
1456  * Write @data into the MSR specified by @index.  Select MSR specific fault
1457  * checks are bypassed if @host_initiated is %true.
1458  * Returns 0 on success, non-0 otherwise.
1459  * Assumes vcpu_load() was already called.
1460  */
1461 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1462 			 bool host_initiated)
1463 {
1464 	struct msr_data msr;
1465 
1466 	switch (index) {
1467 	case MSR_FS_BASE:
1468 	case MSR_GS_BASE:
1469 	case MSR_KERNEL_GS_BASE:
1470 	case MSR_CSTAR:
1471 	case MSR_LSTAR:
1472 		if (is_noncanonical_address(data, vcpu))
1473 			return 1;
1474 		break;
1475 	case MSR_IA32_SYSENTER_EIP:
1476 	case MSR_IA32_SYSENTER_ESP:
1477 		/*
1478 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1479 		 * non-canonical address is written on Intel but not on
1480 		 * AMD (which ignores the top 32-bits, because it does
1481 		 * not implement 64-bit SYSENTER).
1482 		 *
1483 		 * 64-bit code should hence be able to write a non-canonical
1484 		 * value on AMD.  Making the address canonical ensures that
1485 		 * vmentry does not fail on Intel after writing a non-canonical
1486 		 * value, and that something deterministic happens if the guest
1487 		 * invokes 64-bit SYSENTER.
1488 		 */
1489 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1490 	}
1491 
1492 	msr.data = data;
1493 	msr.index = index;
1494 	msr.host_initiated = host_initiated;
1495 
1496 	return kvm_x86_ops.set_msr(vcpu, &msr);
1497 }
1498 
1499 /*
1500  * Read the MSR specified by @index into @data.  Select MSR specific fault
1501  * checks are bypassed if @host_initiated is %true.
1502  * Returns 0 on success, non-0 otherwise.
1503  * Assumes vcpu_load() was already called.
1504  */
1505 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1506 		  bool host_initiated)
1507 {
1508 	struct msr_data msr;
1509 	int ret;
1510 
1511 	msr.index = index;
1512 	msr.host_initiated = host_initiated;
1513 
1514 	ret = kvm_x86_ops.get_msr(vcpu, &msr);
1515 	if (!ret)
1516 		*data = msr.data;
1517 	return ret;
1518 }
1519 
1520 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1521 {
1522 	return __kvm_get_msr(vcpu, index, data, false);
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_get_msr);
1525 
1526 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1527 {
1528 	return __kvm_set_msr(vcpu, index, data, false);
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_set_msr);
1531 
1532 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1533 {
1534 	u32 ecx = kvm_rcx_read(vcpu);
1535 	u64 data;
1536 
1537 	if (kvm_get_msr(vcpu, ecx, &data)) {
1538 		trace_kvm_msr_read_ex(ecx);
1539 		kvm_inject_gp(vcpu, 0);
1540 		return 1;
1541 	}
1542 
1543 	trace_kvm_msr_read(ecx, data);
1544 
1545 	kvm_rax_write(vcpu, data & -1u);
1546 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1547 	return kvm_skip_emulated_instruction(vcpu);
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1550 
1551 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1552 {
1553 	u32 ecx = kvm_rcx_read(vcpu);
1554 	u64 data = kvm_read_edx_eax(vcpu);
1555 
1556 	if (kvm_set_msr(vcpu, ecx, data)) {
1557 		trace_kvm_msr_write_ex(ecx, data);
1558 		kvm_inject_gp(vcpu, 0);
1559 		return 1;
1560 	}
1561 
1562 	trace_kvm_msr_write(ecx, data);
1563 	return kvm_skip_emulated_instruction(vcpu);
1564 }
1565 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1566 
1567 /*
1568  * The fast path for frequent and performance sensitive wrmsr emulation,
1569  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1570  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1571  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1572  * other cases which must be called after interrupts are enabled on the host.
1573  */
1574 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1575 {
1576 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1577 		return 1;
1578 
1579 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1580 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1581 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1582 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1583 
1584 		data &= ~(1 << 12);
1585 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1586 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1587 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1588 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1589 		return 0;
1590 	}
1591 
1592 	return 1;
1593 }
1594 
1595 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1596 {
1597 	u32 msr = kvm_rcx_read(vcpu);
1598 	u64 data;
1599 	int ret = 0;
1600 
1601 	switch (msr) {
1602 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1603 		data = kvm_read_edx_eax(vcpu);
1604 		ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1605 		break;
1606 	default:
1607 		return EXIT_FASTPATH_NONE;
1608 	}
1609 
1610 	if (!ret) {
1611 		trace_kvm_msr_write(msr, data);
1612 		return EXIT_FASTPATH_SKIP_EMUL_INS;
1613 	}
1614 
1615 	return EXIT_FASTPATH_NONE;
1616 }
1617 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1618 
1619 /*
1620  * Adapt set_msr() to msr_io()'s calling convention
1621  */
1622 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1623 {
1624 	return __kvm_get_msr(vcpu, index, data, true);
1625 }
1626 
1627 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1628 {
1629 	return __kvm_set_msr(vcpu, index, *data, true);
1630 }
1631 
1632 #ifdef CONFIG_X86_64
1633 struct pvclock_clock {
1634 	int vclock_mode;
1635 	u64 cycle_last;
1636 	u64 mask;
1637 	u32 mult;
1638 	u32 shift;
1639 	u64 base_cycles;
1640 	u64 offset;
1641 };
1642 
1643 struct pvclock_gtod_data {
1644 	seqcount_t	seq;
1645 
1646 	struct pvclock_clock clock; /* extract of a clocksource struct */
1647 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1648 
1649 	ktime_t		offs_boot;
1650 	u64		wall_time_sec;
1651 };
1652 
1653 static struct pvclock_gtod_data pvclock_gtod_data;
1654 
1655 static void update_pvclock_gtod(struct timekeeper *tk)
1656 {
1657 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1658 
1659 	write_seqcount_begin(&vdata->seq);
1660 
1661 	/* copy pvclock gtod data */
1662 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
1663 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1664 	vdata->clock.mask		= tk->tkr_mono.mask;
1665 	vdata->clock.mult		= tk->tkr_mono.mult;
1666 	vdata->clock.shift		= tk->tkr_mono.shift;
1667 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
1668 	vdata->clock.offset		= tk->tkr_mono.base;
1669 
1670 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
1671 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1672 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1673 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1674 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1675 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
1676 	vdata->raw_clock.offset		= tk->tkr_raw.base;
1677 
1678 	vdata->wall_time_sec            = tk->xtime_sec;
1679 
1680 	vdata->offs_boot		= tk->offs_boot;
1681 
1682 	write_seqcount_end(&vdata->seq);
1683 }
1684 
1685 static s64 get_kvmclock_base_ns(void)
1686 {
1687 	/* Count up from boot time, but with the frequency of the raw clock.  */
1688 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1689 }
1690 #else
1691 static s64 get_kvmclock_base_ns(void)
1692 {
1693 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
1694 	return ktime_get_boottime_ns();
1695 }
1696 #endif
1697 
1698 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1699 {
1700 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1701 	kvm_vcpu_kick(vcpu);
1702 }
1703 
1704 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1705 {
1706 	int version;
1707 	int r;
1708 	struct pvclock_wall_clock wc;
1709 	u64 wall_nsec;
1710 
1711 	if (!wall_clock)
1712 		return;
1713 
1714 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1715 	if (r)
1716 		return;
1717 
1718 	if (version & 1)
1719 		++version;  /* first time write, random junk */
1720 
1721 	++version;
1722 
1723 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1724 		return;
1725 
1726 	/*
1727 	 * The guest calculates current wall clock time by adding
1728 	 * system time (updated by kvm_guest_time_update below) to the
1729 	 * wall clock specified here.  We do the reverse here.
1730 	 */
1731 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1732 
1733 	wc.nsec = do_div(wall_nsec, 1000000000);
1734 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1735 	wc.version = version;
1736 
1737 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1738 
1739 	version++;
1740 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1741 }
1742 
1743 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1744 {
1745 	do_shl32_div32(dividend, divisor);
1746 	return dividend;
1747 }
1748 
1749 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1750 			       s8 *pshift, u32 *pmultiplier)
1751 {
1752 	uint64_t scaled64;
1753 	int32_t  shift = 0;
1754 	uint64_t tps64;
1755 	uint32_t tps32;
1756 
1757 	tps64 = base_hz;
1758 	scaled64 = scaled_hz;
1759 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1760 		tps64 >>= 1;
1761 		shift--;
1762 	}
1763 
1764 	tps32 = (uint32_t)tps64;
1765 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1766 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1767 			scaled64 >>= 1;
1768 		else
1769 			tps32 <<= 1;
1770 		shift++;
1771 	}
1772 
1773 	*pshift = shift;
1774 	*pmultiplier = div_frac(scaled64, tps32);
1775 }
1776 
1777 #ifdef CONFIG_X86_64
1778 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1779 #endif
1780 
1781 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1782 static unsigned long max_tsc_khz;
1783 
1784 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1785 {
1786 	u64 v = (u64)khz * (1000000 + ppm);
1787 	do_div(v, 1000000);
1788 	return v;
1789 }
1790 
1791 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1792 {
1793 	u64 ratio;
1794 
1795 	/* Guest TSC same frequency as host TSC? */
1796 	if (!scale) {
1797 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1798 		return 0;
1799 	}
1800 
1801 	/* TSC scaling supported? */
1802 	if (!kvm_has_tsc_control) {
1803 		if (user_tsc_khz > tsc_khz) {
1804 			vcpu->arch.tsc_catchup = 1;
1805 			vcpu->arch.tsc_always_catchup = 1;
1806 			return 0;
1807 		} else {
1808 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1809 			return -1;
1810 		}
1811 	}
1812 
1813 	/* TSC scaling required  - calculate ratio */
1814 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1815 				user_tsc_khz, tsc_khz);
1816 
1817 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1818 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1819 			            user_tsc_khz);
1820 		return -1;
1821 	}
1822 
1823 	vcpu->arch.tsc_scaling_ratio = ratio;
1824 	return 0;
1825 }
1826 
1827 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1828 {
1829 	u32 thresh_lo, thresh_hi;
1830 	int use_scaling = 0;
1831 
1832 	/* tsc_khz can be zero if TSC calibration fails */
1833 	if (user_tsc_khz == 0) {
1834 		/* set tsc_scaling_ratio to a safe value */
1835 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1836 		return -1;
1837 	}
1838 
1839 	/* Compute a scale to convert nanoseconds in TSC cycles */
1840 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1841 			   &vcpu->arch.virtual_tsc_shift,
1842 			   &vcpu->arch.virtual_tsc_mult);
1843 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1844 
1845 	/*
1846 	 * Compute the variation in TSC rate which is acceptable
1847 	 * within the range of tolerance and decide if the
1848 	 * rate being applied is within that bounds of the hardware
1849 	 * rate.  If so, no scaling or compensation need be done.
1850 	 */
1851 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1852 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1853 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1854 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1855 		use_scaling = 1;
1856 	}
1857 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1858 }
1859 
1860 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1861 {
1862 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1863 				      vcpu->arch.virtual_tsc_mult,
1864 				      vcpu->arch.virtual_tsc_shift);
1865 	tsc += vcpu->arch.this_tsc_write;
1866 	return tsc;
1867 }
1868 
1869 static inline int gtod_is_based_on_tsc(int mode)
1870 {
1871 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1872 }
1873 
1874 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1875 {
1876 #ifdef CONFIG_X86_64
1877 	bool vcpus_matched;
1878 	struct kvm_arch *ka = &vcpu->kvm->arch;
1879 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1880 
1881 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1882 			 atomic_read(&vcpu->kvm->online_vcpus));
1883 
1884 	/*
1885 	 * Once the masterclock is enabled, always perform request in
1886 	 * order to update it.
1887 	 *
1888 	 * In order to enable masterclock, the host clocksource must be TSC
1889 	 * and the vcpus need to have matched TSCs.  When that happens,
1890 	 * perform request to enable masterclock.
1891 	 */
1892 	if (ka->use_master_clock ||
1893 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1894 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1895 
1896 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1897 			    atomic_read(&vcpu->kvm->online_vcpus),
1898 		            ka->use_master_clock, gtod->clock.vclock_mode);
1899 #endif
1900 }
1901 
1902 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1903 {
1904 	u64 curr_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1905 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1906 }
1907 
1908 /*
1909  * Multiply tsc by a fixed point number represented by ratio.
1910  *
1911  * The most significant 64-N bits (mult) of ratio represent the
1912  * integral part of the fixed point number; the remaining N bits
1913  * (frac) represent the fractional part, ie. ratio represents a fixed
1914  * point number (mult + frac * 2^(-N)).
1915  *
1916  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1917  */
1918 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1919 {
1920 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1921 }
1922 
1923 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1924 {
1925 	u64 _tsc = tsc;
1926 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1927 
1928 	if (ratio != kvm_default_tsc_scaling_ratio)
1929 		_tsc = __scale_tsc(ratio, tsc);
1930 
1931 	return _tsc;
1932 }
1933 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1934 
1935 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1936 {
1937 	u64 tsc;
1938 
1939 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1940 
1941 	return target_tsc - tsc;
1942 }
1943 
1944 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1945 {
1946 	u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1947 
1948 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1949 }
1950 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1951 
1952 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1953 {
1954 	vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
1955 }
1956 
1957 static inline bool kvm_check_tsc_unstable(void)
1958 {
1959 #ifdef CONFIG_X86_64
1960 	/*
1961 	 * TSC is marked unstable when we're running on Hyper-V,
1962 	 * 'TSC page' clocksource is good.
1963 	 */
1964 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
1965 		return false;
1966 #endif
1967 	return check_tsc_unstable();
1968 }
1969 
1970 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1971 {
1972 	struct kvm *kvm = vcpu->kvm;
1973 	u64 offset, ns, elapsed;
1974 	unsigned long flags;
1975 	bool matched;
1976 	bool already_matched;
1977 	u64 data = msr->data;
1978 	bool synchronizing = false;
1979 
1980 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1981 	offset = kvm_compute_tsc_offset(vcpu, data);
1982 	ns = get_kvmclock_base_ns();
1983 	elapsed = ns - kvm->arch.last_tsc_nsec;
1984 
1985 	if (vcpu->arch.virtual_tsc_khz) {
1986 		if (data == 0 && msr->host_initiated) {
1987 			/*
1988 			 * detection of vcpu initialization -- need to sync
1989 			 * with other vCPUs. This particularly helps to keep
1990 			 * kvm_clock stable after CPU hotplug
1991 			 */
1992 			synchronizing = true;
1993 		} else {
1994 			u64 tsc_exp = kvm->arch.last_tsc_write +
1995 						nsec_to_cycles(vcpu, elapsed);
1996 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1997 			/*
1998 			 * Special case: TSC write with a small delta (1 second)
1999 			 * of virtual cycle time against real time is
2000 			 * interpreted as an attempt to synchronize the CPU.
2001 			 */
2002 			synchronizing = data < tsc_exp + tsc_hz &&
2003 					data + tsc_hz > tsc_exp;
2004 		}
2005 	}
2006 
2007 	/*
2008 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2009 	 * TSC, we add elapsed time in this computation.  We could let the
2010 	 * compensation code attempt to catch up if we fall behind, but
2011 	 * it's better to try to match offsets from the beginning.
2012          */
2013 	if (synchronizing &&
2014 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2015 		if (!kvm_check_tsc_unstable()) {
2016 			offset = kvm->arch.cur_tsc_offset;
2017 		} else {
2018 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2019 			data += delta;
2020 			offset = kvm_compute_tsc_offset(vcpu, data);
2021 		}
2022 		matched = true;
2023 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2024 	} else {
2025 		/*
2026 		 * We split periods of matched TSC writes into generations.
2027 		 * For each generation, we track the original measured
2028 		 * nanosecond time, offset, and write, so if TSCs are in
2029 		 * sync, we can match exact offset, and if not, we can match
2030 		 * exact software computation in compute_guest_tsc()
2031 		 *
2032 		 * These values are tracked in kvm->arch.cur_xxx variables.
2033 		 */
2034 		kvm->arch.cur_tsc_generation++;
2035 		kvm->arch.cur_tsc_nsec = ns;
2036 		kvm->arch.cur_tsc_write = data;
2037 		kvm->arch.cur_tsc_offset = offset;
2038 		matched = false;
2039 	}
2040 
2041 	/*
2042 	 * We also track th most recent recorded KHZ, write and time to
2043 	 * allow the matching interval to be extended at each write.
2044 	 */
2045 	kvm->arch.last_tsc_nsec = ns;
2046 	kvm->arch.last_tsc_write = data;
2047 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2048 
2049 	vcpu->arch.last_guest_tsc = data;
2050 
2051 	/* Keep track of which generation this VCPU has synchronized to */
2052 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2053 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2054 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2055 
2056 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2057 		update_ia32_tsc_adjust_msr(vcpu, offset);
2058 
2059 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2060 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2061 
2062 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2063 	if (!matched) {
2064 		kvm->arch.nr_vcpus_matched_tsc = 0;
2065 	} else if (!already_matched) {
2066 		kvm->arch.nr_vcpus_matched_tsc++;
2067 	}
2068 
2069 	kvm_track_tsc_matching(vcpu);
2070 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2071 }
2072 
2073 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2074 
2075 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2076 					   s64 adjustment)
2077 {
2078 	u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
2079 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2080 }
2081 
2082 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2083 {
2084 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2085 		WARN_ON(adjustment < 0);
2086 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2087 	adjust_tsc_offset_guest(vcpu, adjustment);
2088 }
2089 
2090 #ifdef CONFIG_X86_64
2091 
2092 static u64 read_tsc(void)
2093 {
2094 	u64 ret = (u64)rdtsc_ordered();
2095 	u64 last = pvclock_gtod_data.clock.cycle_last;
2096 
2097 	if (likely(ret >= last))
2098 		return ret;
2099 
2100 	/*
2101 	 * GCC likes to generate cmov here, but this branch is extremely
2102 	 * predictable (it's just a function of time and the likely is
2103 	 * very likely) and there's a data dependence, so force GCC
2104 	 * to generate a branch instead.  I don't barrier() because
2105 	 * we don't actually need a barrier, and if this function
2106 	 * ever gets inlined it will generate worse code.
2107 	 */
2108 	asm volatile ("");
2109 	return last;
2110 }
2111 
2112 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2113 			  int *mode)
2114 {
2115 	long v;
2116 	u64 tsc_pg_val;
2117 
2118 	switch (clock->vclock_mode) {
2119 	case VDSO_CLOCKMODE_HVCLOCK:
2120 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2121 						  tsc_timestamp);
2122 		if (tsc_pg_val != U64_MAX) {
2123 			/* TSC page valid */
2124 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2125 			v = (tsc_pg_val - clock->cycle_last) &
2126 				clock->mask;
2127 		} else {
2128 			/* TSC page invalid */
2129 			*mode = VDSO_CLOCKMODE_NONE;
2130 		}
2131 		break;
2132 	case VDSO_CLOCKMODE_TSC:
2133 		*mode = VDSO_CLOCKMODE_TSC;
2134 		*tsc_timestamp = read_tsc();
2135 		v = (*tsc_timestamp - clock->cycle_last) &
2136 			clock->mask;
2137 		break;
2138 	default:
2139 		*mode = VDSO_CLOCKMODE_NONE;
2140 	}
2141 
2142 	if (*mode == VDSO_CLOCKMODE_NONE)
2143 		*tsc_timestamp = v = 0;
2144 
2145 	return v * clock->mult;
2146 }
2147 
2148 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2149 {
2150 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2151 	unsigned long seq;
2152 	int mode;
2153 	u64 ns;
2154 
2155 	do {
2156 		seq = read_seqcount_begin(&gtod->seq);
2157 		ns = gtod->raw_clock.base_cycles;
2158 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2159 		ns >>= gtod->raw_clock.shift;
2160 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2161 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2162 	*t = ns;
2163 
2164 	return mode;
2165 }
2166 
2167 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2168 {
2169 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2170 	unsigned long seq;
2171 	int mode;
2172 	u64 ns;
2173 
2174 	do {
2175 		seq = read_seqcount_begin(&gtod->seq);
2176 		ts->tv_sec = gtod->wall_time_sec;
2177 		ns = gtod->clock.base_cycles;
2178 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2179 		ns >>= gtod->clock.shift;
2180 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2181 
2182 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2183 	ts->tv_nsec = ns;
2184 
2185 	return mode;
2186 }
2187 
2188 /* returns true if host is using TSC based clocksource */
2189 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2190 {
2191 	/* checked again under seqlock below */
2192 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2193 		return false;
2194 
2195 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2196 						      tsc_timestamp));
2197 }
2198 
2199 /* returns true if host is using TSC based clocksource */
2200 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2201 					   u64 *tsc_timestamp)
2202 {
2203 	/* checked again under seqlock below */
2204 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2205 		return false;
2206 
2207 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2208 }
2209 #endif
2210 
2211 /*
2212  *
2213  * Assuming a stable TSC across physical CPUS, and a stable TSC
2214  * across virtual CPUs, the following condition is possible.
2215  * Each numbered line represents an event visible to both
2216  * CPUs at the next numbered event.
2217  *
2218  * "timespecX" represents host monotonic time. "tscX" represents
2219  * RDTSC value.
2220  *
2221  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2222  *
2223  * 1.  read timespec0,tsc0
2224  * 2.					| timespec1 = timespec0 + N
2225  * 					| tsc1 = tsc0 + M
2226  * 3. transition to guest		| transition to guest
2227  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2228  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2229  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2230  *
2231  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2232  *
2233  * 	- ret0 < ret1
2234  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2235  *		...
2236  *	- 0 < N - M => M < N
2237  *
2238  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2239  * always the case (the difference between two distinct xtime instances
2240  * might be smaller then the difference between corresponding TSC reads,
2241  * when updating guest vcpus pvclock areas).
2242  *
2243  * To avoid that problem, do not allow visibility of distinct
2244  * system_timestamp/tsc_timestamp values simultaneously: use a master
2245  * copy of host monotonic time values. Update that master copy
2246  * in lockstep.
2247  *
2248  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2249  *
2250  */
2251 
2252 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2253 {
2254 #ifdef CONFIG_X86_64
2255 	struct kvm_arch *ka = &kvm->arch;
2256 	int vclock_mode;
2257 	bool host_tsc_clocksource, vcpus_matched;
2258 
2259 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2260 			atomic_read(&kvm->online_vcpus));
2261 
2262 	/*
2263 	 * If the host uses TSC clock, then passthrough TSC as stable
2264 	 * to the guest.
2265 	 */
2266 	host_tsc_clocksource = kvm_get_time_and_clockread(
2267 					&ka->master_kernel_ns,
2268 					&ka->master_cycle_now);
2269 
2270 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2271 				&& !ka->backwards_tsc_observed
2272 				&& !ka->boot_vcpu_runs_old_kvmclock;
2273 
2274 	if (ka->use_master_clock)
2275 		atomic_set(&kvm_guest_has_master_clock, 1);
2276 
2277 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2278 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2279 					vcpus_matched);
2280 #endif
2281 }
2282 
2283 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2284 {
2285 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2286 }
2287 
2288 static void kvm_gen_update_masterclock(struct kvm *kvm)
2289 {
2290 #ifdef CONFIG_X86_64
2291 	int i;
2292 	struct kvm_vcpu *vcpu;
2293 	struct kvm_arch *ka = &kvm->arch;
2294 
2295 	spin_lock(&ka->pvclock_gtod_sync_lock);
2296 	kvm_make_mclock_inprogress_request(kvm);
2297 	/* no guest entries from this point */
2298 	pvclock_update_vm_gtod_copy(kvm);
2299 
2300 	kvm_for_each_vcpu(i, vcpu, kvm)
2301 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2302 
2303 	/* guest entries allowed */
2304 	kvm_for_each_vcpu(i, vcpu, kvm)
2305 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2306 
2307 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2308 #endif
2309 }
2310 
2311 u64 get_kvmclock_ns(struct kvm *kvm)
2312 {
2313 	struct kvm_arch *ka = &kvm->arch;
2314 	struct pvclock_vcpu_time_info hv_clock;
2315 	u64 ret;
2316 
2317 	spin_lock(&ka->pvclock_gtod_sync_lock);
2318 	if (!ka->use_master_clock) {
2319 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2320 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2321 	}
2322 
2323 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2324 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2325 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2326 
2327 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2328 	get_cpu();
2329 
2330 	if (__this_cpu_read(cpu_tsc_khz)) {
2331 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2332 				   &hv_clock.tsc_shift,
2333 				   &hv_clock.tsc_to_system_mul);
2334 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2335 	} else
2336 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2337 
2338 	put_cpu();
2339 
2340 	return ret;
2341 }
2342 
2343 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2344 {
2345 	struct kvm_vcpu_arch *vcpu = &v->arch;
2346 	struct pvclock_vcpu_time_info guest_hv_clock;
2347 
2348 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2349 		&guest_hv_clock, sizeof(guest_hv_clock))))
2350 		return;
2351 
2352 	/* This VCPU is paused, but it's legal for a guest to read another
2353 	 * VCPU's kvmclock, so we really have to follow the specification where
2354 	 * it says that version is odd if data is being modified, and even after
2355 	 * it is consistent.
2356 	 *
2357 	 * Version field updates must be kept separate.  This is because
2358 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2359 	 * writes within a string instruction are weakly ordered.  So there
2360 	 * are three writes overall.
2361 	 *
2362 	 * As a small optimization, only write the version field in the first
2363 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2364 	 * version field is the first in the struct.
2365 	 */
2366 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2367 
2368 	if (guest_hv_clock.version & 1)
2369 		++guest_hv_clock.version;  /* first time write, random junk */
2370 
2371 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2372 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2373 				&vcpu->hv_clock,
2374 				sizeof(vcpu->hv_clock.version));
2375 
2376 	smp_wmb();
2377 
2378 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2379 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2380 
2381 	if (vcpu->pvclock_set_guest_stopped_request) {
2382 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2383 		vcpu->pvclock_set_guest_stopped_request = false;
2384 	}
2385 
2386 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2387 
2388 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2389 				&vcpu->hv_clock,
2390 				sizeof(vcpu->hv_clock));
2391 
2392 	smp_wmb();
2393 
2394 	vcpu->hv_clock.version++;
2395 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2396 				&vcpu->hv_clock,
2397 				sizeof(vcpu->hv_clock.version));
2398 }
2399 
2400 static int kvm_guest_time_update(struct kvm_vcpu *v)
2401 {
2402 	unsigned long flags, tgt_tsc_khz;
2403 	struct kvm_vcpu_arch *vcpu = &v->arch;
2404 	struct kvm_arch *ka = &v->kvm->arch;
2405 	s64 kernel_ns;
2406 	u64 tsc_timestamp, host_tsc;
2407 	u8 pvclock_flags;
2408 	bool use_master_clock;
2409 
2410 	kernel_ns = 0;
2411 	host_tsc = 0;
2412 
2413 	/*
2414 	 * If the host uses TSC clock, then passthrough TSC as stable
2415 	 * to the guest.
2416 	 */
2417 	spin_lock(&ka->pvclock_gtod_sync_lock);
2418 	use_master_clock = ka->use_master_clock;
2419 	if (use_master_clock) {
2420 		host_tsc = ka->master_cycle_now;
2421 		kernel_ns = ka->master_kernel_ns;
2422 	}
2423 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2424 
2425 	/* Keep irq disabled to prevent changes to the clock */
2426 	local_irq_save(flags);
2427 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2428 	if (unlikely(tgt_tsc_khz == 0)) {
2429 		local_irq_restore(flags);
2430 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2431 		return 1;
2432 	}
2433 	if (!use_master_clock) {
2434 		host_tsc = rdtsc();
2435 		kernel_ns = get_kvmclock_base_ns();
2436 	}
2437 
2438 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2439 
2440 	/*
2441 	 * We may have to catch up the TSC to match elapsed wall clock
2442 	 * time for two reasons, even if kvmclock is used.
2443 	 *   1) CPU could have been running below the maximum TSC rate
2444 	 *   2) Broken TSC compensation resets the base at each VCPU
2445 	 *      entry to avoid unknown leaps of TSC even when running
2446 	 *      again on the same CPU.  This may cause apparent elapsed
2447 	 *      time to disappear, and the guest to stand still or run
2448 	 *	very slowly.
2449 	 */
2450 	if (vcpu->tsc_catchup) {
2451 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2452 		if (tsc > tsc_timestamp) {
2453 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2454 			tsc_timestamp = tsc;
2455 		}
2456 	}
2457 
2458 	local_irq_restore(flags);
2459 
2460 	/* With all the info we got, fill in the values */
2461 
2462 	if (kvm_has_tsc_control)
2463 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2464 
2465 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2466 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2467 				   &vcpu->hv_clock.tsc_shift,
2468 				   &vcpu->hv_clock.tsc_to_system_mul);
2469 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2470 	}
2471 
2472 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2473 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2474 	vcpu->last_guest_tsc = tsc_timestamp;
2475 
2476 	/* If the host uses TSC clocksource, then it is stable */
2477 	pvclock_flags = 0;
2478 	if (use_master_clock)
2479 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2480 
2481 	vcpu->hv_clock.flags = pvclock_flags;
2482 
2483 	if (vcpu->pv_time_enabled)
2484 		kvm_setup_pvclock_page(v);
2485 	if (v == kvm_get_vcpu(v->kvm, 0))
2486 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2487 	return 0;
2488 }
2489 
2490 /*
2491  * kvmclock updates which are isolated to a given vcpu, such as
2492  * vcpu->cpu migration, should not allow system_timestamp from
2493  * the rest of the vcpus to remain static. Otherwise ntp frequency
2494  * correction applies to one vcpu's system_timestamp but not
2495  * the others.
2496  *
2497  * So in those cases, request a kvmclock update for all vcpus.
2498  * We need to rate-limit these requests though, as they can
2499  * considerably slow guests that have a large number of vcpus.
2500  * The time for a remote vcpu to update its kvmclock is bound
2501  * by the delay we use to rate-limit the updates.
2502  */
2503 
2504 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2505 
2506 static void kvmclock_update_fn(struct work_struct *work)
2507 {
2508 	int i;
2509 	struct delayed_work *dwork = to_delayed_work(work);
2510 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2511 					   kvmclock_update_work);
2512 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2513 	struct kvm_vcpu *vcpu;
2514 
2515 	kvm_for_each_vcpu(i, vcpu, kvm) {
2516 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2517 		kvm_vcpu_kick(vcpu);
2518 	}
2519 }
2520 
2521 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2522 {
2523 	struct kvm *kvm = v->kvm;
2524 
2525 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2526 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2527 					KVMCLOCK_UPDATE_DELAY);
2528 }
2529 
2530 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2531 
2532 static void kvmclock_sync_fn(struct work_struct *work)
2533 {
2534 	struct delayed_work *dwork = to_delayed_work(work);
2535 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2536 					   kvmclock_sync_work);
2537 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2538 
2539 	if (!kvmclock_periodic_sync)
2540 		return;
2541 
2542 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2543 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2544 					KVMCLOCK_SYNC_PERIOD);
2545 }
2546 
2547 /*
2548  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2549  */
2550 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2551 {
2552 	/* McStatusWrEn enabled? */
2553 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2554 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2555 
2556 	return false;
2557 }
2558 
2559 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2560 {
2561 	u64 mcg_cap = vcpu->arch.mcg_cap;
2562 	unsigned bank_num = mcg_cap & 0xff;
2563 	u32 msr = msr_info->index;
2564 	u64 data = msr_info->data;
2565 
2566 	switch (msr) {
2567 	case MSR_IA32_MCG_STATUS:
2568 		vcpu->arch.mcg_status = data;
2569 		break;
2570 	case MSR_IA32_MCG_CTL:
2571 		if (!(mcg_cap & MCG_CTL_P) &&
2572 		    (data || !msr_info->host_initiated))
2573 			return 1;
2574 		if (data != 0 && data != ~(u64)0)
2575 			return 1;
2576 		vcpu->arch.mcg_ctl = data;
2577 		break;
2578 	default:
2579 		if (msr >= MSR_IA32_MC0_CTL &&
2580 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2581 			u32 offset = array_index_nospec(
2582 				msr - MSR_IA32_MC0_CTL,
2583 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2584 
2585 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2586 			 * some Linux kernels though clear bit 10 in bank 4 to
2587 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2588 			 * this to avoid an uncatched #GP in the guest
2589 			 */
2590 			if ((offset & 0x3) == 0 &&
2591 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2592 				return -1;
2593 
2594 			/* MCi_STATUS */
2595 			if (!msr_info->host_initiated &&
2596 			    (offset & 0x3) == 1 && data != 0) {
2597 				if (!can_set_mci_status(vcpu))
2598 					return -1;
2599 			}
2600 
2601 			vcpu->arch.mce_banks[offset] = data;
2602 			break;
2603 		}
2604 		return 1;
2605 	}
2606 	return 0;
2607 }
2608 
2609 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2610 {
2611 	struct kvm *kvm = vcpu->kvm;
2612 	int lm = is_long_mode(vcpu);
2613 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2614 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2615 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2616 		: kvm->arch.xen_hvm_config.blob_size_32;
2617 	u32 page_num = data & ~PAGE_MASK;
2618 	u64 page_addr = data & PAGE_MASK;
2619 	u8 *page;
2620 	int r;
2621 
2622 	r = -E2BIG;
2623 	if (page_num >= blob_size)
2624 		goto out;
2625 	r = -ENOMEM;
2626 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2627 	if (IS_ERR(page)) {
2628 		r = PTR_ERR(page);
2629 		goto out;
2630 	}
2631 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2632 		goto out_free;
2633 	r = 0;
2634 out_free:
2635 	kfree(page);
2636 out:
2637 	return r;
2638 }
2639 
2640 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2641 {
2642 	gpa_t gpa = data & ~0x3f;
2643 
2644 	/* Bits 3:5 are reserved, Should be zero */
2645 	if (data & 0x38)
2646 		return 1;
2647 
2648 	vcpu->arch.apf.msr_val = data;
2649 
2650 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2651 		kvm_clear_async_pf_completion_queue(vcpu);
2652 		kvm_async_pf_hash_reset(vcpu);
2653 		return 0;
2654 	}
2655 
2656 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2657 					sizeof(u32)))
2658 		return 1;
2659 
2660 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2661 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2662 	kvm_async_pf_wakeup_all(vcpu);
2663 	return 0;
2664 }
2665 
2666 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2667 {
2668 	vcpu->arch.pv_time_enabled = false;
2669 	vcpu->arch.time = 0;
2670 }
2671 
2672 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2673 {
2674 	++vcpu->stat.tlb_flush;
2675 	kvm_x86_ops.tlb_flush(vcpu, invalidate_gpa);
2676 }
2677 
2678 static void record_steal_time(struct kvm_vcpu *vcpu)
2679 {
2680 	struct kvm_host_map map;
2681 	struct kvm_steal_time *st;
2682 
2683 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2684 		return;
2685 
2686 	/* -EAGAIN is returned in atomic context so we can just return. */
2687 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2688 			&map, &vcpu->arch.st.cache, false))
2689 		return;
2690 
2691 	st = map.hva +
2692 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2693 
2694 	/*
2695 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2696 	 * expensive IPIs.
2697 	 */
2698 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2699 		st->preempted & KVM_VCPU_FLUSH_TLB);
2700 	if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2701 		kvm_vcpu_flush_tlb(vcpu, false);
2702 
2703 	vcpu->arch.st.preempted = 0;
2704 
2705 	if (st->version & 1)
2706 		st->version += 1;  /* first time write, random junk */
2707 
2708 	st->version += 1;
2709 
2710 	smp_wmb();
2711 
2712 	st->steal += current->sched_info.run_delay -
2713 		vcpu->arch.st.last_steal;
2714 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2715 
2716 	smp_wmb();
2717 
2718 	st->version += 1;
2719 
2720 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2721 }
2722 
2723 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2724 {
2725 	bool pr = false;
2726 	u32 msr = msr_info->index;
2727 	u64 data = msr_info->data;
2728 
2729 	switch (msr) {
2730 	case MSR_AMD64_NB_CFG:
2731 	case MSR_IA32_UCODE_WRITE:
2732 	case MSR_VM_HSAVE_PA:
2733 	case MSR_AMD64_PATCH_LOADER:
2734 	case MSR_AMD64_BU_CFG2:
2735 	case MSR_AMD64_DC_CFG:
2736 	case MSR_F15H_EX_CFG:
2737 		break;
2738 
2739 	case MSR_IA32_UCODE_REV:
2740 		if (msr_info->host_initiated)
2741 			vcpu->arch.microcode_version = data;
2742 		break;
2743 	case MSR_IA32_ARCH_CAPABILITIES:
2744 		if (!msr_info->host_initiated)
2745 			return 1;
2746 		vcpu->arch.arch_capabilities = data;
2747 		break;
2748 	case MSR_EFER:
2749 		return set_efer(vcpu, msr_info);
2750 	case MSR_K7_HWCR:
2751 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2752 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2753 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2754 
2755 		/* Handle McStatusWrEn */
2756 		if (data == BIT_ULL(18)) {
2757 			vcpu->arch.msr_hwcr = data;
2758 		} else if (data != 0) {
2759 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2760 				    data);
2761 			return 1;
2762 		}
2763 		break;
2764 	case MSR_FAM10H_MMIO_CONF_BASE:
2765 		if (data != 0) {
2766 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2767 				    "0x%llx\n", data);
2768 			return 1;
2769 		}
2770 		break;
2771 	case MSR_IA32_DEBUGCTLMSR:
2772 		if (!data) {
2773 			/* We support the non-activated case already */
2774 			break;
2775 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2776 			/* Values other than LBR and BTF are vendor-specific,
2777 			   thus reserved and should throw a #GP */
2778 			return 1;
2779 		}
2780 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2781 			    __func__, data);
2782 		break;
2783 	case 0x200 ... 0x2ff:
2784 		return kvm_mtrr_set_msr(vcpu, msr, data);
2785 	case MSR_IA32_APICBASE:
2786 		return kvm_set_apic_base(vcpu, msr_info);
2787 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2788 		return kvm_x2apic_msr_write(vcpu, msr, data);
2789 	case MSR_IA32_TSCDEADLINE:
2790 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2791 		break;
2792 	case MSR_IA32_TSC_ADJUST:
2793 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2794 			if (!msr_info->host_initiated) {
2795 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2796 				adjust_tsc_offset_guest(vcpu, adj);
2797 			}
2798 			vcpu->arch.ia32_tsc_adjust_msr = data;
2799 		}
2800 		break;
2801 	case MSR_IA32_MISC_ENABLE:
2802 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2803 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2804 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2805 				return 1;
2806 			vcpu->arch.ia32_misc_enable_msr = data;
2807 			kvm_update_cpuid(vcpu);
2808 		} else {
2809 			vcpu->arch.ia32_misc_enable_msr = data;
2810 		}
2811 		break;
2812 	case MSR_IA32_SMBASE:
2813 		if (!msr_info->host_initiated)
2814 			return 1;
2815 		vcpu->arch.smbase = data;
2816 		break;
2817 	case MSR_IA32_POWER_CTL:
2818 		vcpu->arch.msr_ia32_power_ctl = data;
2819 		break;
2820 	case MSR_IA32_TSC:
2821 		kvm_write_tsc(vcpu, msr_info);
2822 		break;
2823 	case MSR_IA32_XSS:
2824 		if (!msr_info->host_initiated &&
2825 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2826 			return 1;
2827 		/*
2828 		 * KVM supports exposing PT to the guest, but does not support
2829 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2830 		 * XSAVES/XRSTORS to save/restore PT MSRs.
2831 		 */
2832 		if (data & ~supported_xss)
2833 			return 1;
2834 		vcpu->arch.ia32_xss = data;
2835 		break;
2836 	case MSR_SMI_COUNT:
2837 		if (!msr_info->host_initiated)
2838 			return 1;
2839 		vcpu->arch.smi_count = data;
2840 		break;
2841 	case MSR_KVM_WALL_CLOCK_NEW:
2842 	case MSR_KVM_WALL_CLOCK:
2843 		vcpu->kvm->arch.wall_clock = data;
2844 		kvm_write_wall_clock(vcpu->kvm, data);
2845 		break;
2846 	case MSR_KVM_SYSTEM_TIME_NEW:
2847 	case MSR_KVM_SYSTEM_TIME: {
2848 		struct kvm_arch *ka = &vcpu->kvm->arch;
2849 
2850 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2851 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2852 
2853 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2854 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2855 
2856 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2857 		}
2858 
2859 		vcpu->arch.time = data;
2860 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2861 
2862 		/* we verify if the enable bit is set... */
2863 		vcpu->arch.pv_time_enabled = false;
2864 		if (!(data & 1))
2865 			break;
2866 
2867 		if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2868 		     &vcpu->arch.pv_time, data & ~1ULL,
2869 		     sizeof(struct pvclock_vcpu_time_info)))
2870 			vcpu->arch.pv_time_enabled = true;
2871 
2872 		break;
2873 	}
2874 	case MSR_KVM_ASYNC_PF_EN:
2875 		if (kvm_pv_enable_async_pf(vcpu, data))
2876 			return 1;
2877 		break;
2878 	case MSR_KVM_STEAL_TIME:
2879 
2880 		if (unlikely(!sched_info_on()))
2881 			return 1;
2882 
2883 		if (data & KVM_STEAL_RESERVED_MASK)
2884 			return 1;
2885 
2886 		vcpu->arch.st.msr_val = data;
2887 
2888 		if (!(data & KVM_MSR_ENABLED))
2889 			break;
2890 
2891 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2892 
2893 		break;
2894 	case MSR_KVM_PV_EOI_EN:
2895 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2896 			return 1;
2897 		break;
2898 
2899 	case MSR_KVM_POLL_CONTROL:
2900 		/* only enable bit supported */
2901 		if (data & (-1ULL << 1))
2902 			return 1;
2903 
2904 		vcpu->arch.msr_kvm_poll_control = data;
2905 		break;
2906 
2907 	case MSR_IA32_MCG_CTL:
2908 	case MSR_IA32_MCG_STATUS:
2909 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2910 		return set_msr_mce(vcpu, msr_info);
2911 
2912 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2913 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2914 		pr = true; /* fall through */
2915 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2916 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2917 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2918 			return kvm_pmu_set_msr(vcpu, msr_info);
2919 
2920 		if (pr || data != 0)
2921 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2922 				    "0x%x data 0x%llx\n", msr, data);
2923 		break;
2924 	case MSR_K7_CLK_CTL:
2925 		/*
2926 		 * Ignore all writes to this no longer documented MSR.
2927 		 * Writes are only relevant for old K7 processors,
2928 		 * all pre-dating SVM, but a recommended workaround from
2929 		 * AMD for these chips. It is possible to specify the
2930 		 * affected processor models on the command line, hence
2931 		 * the need to ignore the workaround.
2932 		 */
2933 		break;
2934 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2935 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2936 	case HV_X64_MSR_CRASH_CTL:
2937 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2938 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2939 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2940 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2941 		return kvm_hv_set_msr_common(vcpu, msr, data,
2942 					     msr_info->host_initiated);
2943 	case MSR_IA32_BBL_CR_CTL3:
2944 		/* Drop writes to this legacy MSR -- see rdmsr
2945 		 * counterpart for further detail.
2946 		 */
2947 		if (report_ignored_msrs)
2948 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2949 				msr, data);
2950 		break;
2951 	case MSR_AMD64_OSVW_ID_LENGTH:
2952 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2953 			return 1;
2954 		vcpu->arch.osvw.length = data;
2955 		break;
2956 	case MSR_AMD64_OSVW_STATUS:
2957 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2958 			return 1;
2959 		vcpu->arch.osvw.status = data;
2960 		break;
2961 	case MSR_PLATFORM_INFO:
2962 		if (!msr_info->host_initiated ||
2963 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2964 		     cpuid_fault_enabled(vcpu)))
2965 			return 1;
2966 		vcpu->arch.msr_platform_info = data;
2967 		break;
2968 	case MSR_MISC_FEATURES_ENABLES:
2969 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2970 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2971 		     !supports_cpuid_fault(vcpu)))
2972 			return 1;
2973 		vcpu->arch.msr_misc_features_enables = data;
2974 		break;
2975 	default:
2976 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2977 			return xen_hvm_config(vcpu, data);
2978 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2979 			return kvm_pmu_set_msr(vcpu, msr_info);
2980 		if (!ignore_msrs) {
2981 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2982 				    msr, data);
2983 			return 1;
2984 		} else {
2985 			if (report_ignored_msrs)
2986 				vcpu_unimpl(vcpu,
2987 					"ignored wrmsr: 0x%x data 0x%llx\n",
2988 					msr, data);
2989 			break;
2990 		}
2991 	}
2992 	return 0;
2993 }
2994 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2995 
2996 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2997 {
2998 	u64 data;
2999 	u64 mcg_cap = vcpu->arch.mcg_cap;
3000 	unsigned bank_num = mcg_cap & 0xff;
3001 
3002 	switch (msr) {
3003 	case MSR_IA32_P5_MC_ADDR:
3004 	case MSR_IA32_P5_MC_TYPE:
3005 		data = 0;
3006 		break;
3007 	case MSR_IA32_MCG_CAP:
3008 		data = vcpu->arch.mcg_cap;
3009 		break;
3010 	case MSR_IA32_MCG_CTL:
3011 		if (!(mcg_cap & MCG_CTL_P) && !host)
3012 			return 1;
3013 		data = vcpu->arch.mcg_ctl;
3014 		break;
3015 	case MSR_IA32_MCG_STATUS:
3016 		data = vcpu->arch.mcg_status;
3017 		break;
3018 	default:
3019 		if (msr >= MSR_IA32_MC0_CTL &&
3020 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3021 			u32 offset = array_index_nospec(
3022 				msr - MSR_IA32_MC0_CTL,
3023 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3024 
3025 			data = vcpu->arch.mce_banks[offset];
3026 			break;
3027 		}
3028 		return 1;
3029 	}
3030 	*pdata = data;
3031 	return 0;
3032 }
3033 
3034 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3035 {
3036 	switch (msr_info->index) {
3037 	case MSR_IA32_PLATFORM_ID:
3038 	case MSR_IA32_EBL_CR_POWERON:
3039 	case MSR_IA32_DEBUGCTLMSR:
3040 	case MSR_IA32_LASTBRANCHFROMIP:
3041 	case MSR_IA32_LASTBRANCHTOIP:
3042 	case MSR_IA32_LASTINTFROMIP:
3043 	case MSR_IA32_LASTINTTOIP:
3044 	case MSR_K8_SYSCFG:
3045 	case MSR_K8_TSEG_ADDR:
3046 	case MSR_K8_TSEG_MASK:
3047 	case MSR_VM_HSAVE_PA:
3048 	case MSR_K8_INT_PENDING_MSG:
3049 	case MSR_AMD64_NB_CFG:
3050 	case MSR_FAM10H_MMIO_CONF_BASE:
3051 	case MSR_AMD64_BU_CFG2:
3052 	case MSR_IA32_PERF_CTL:
3053 	case MSR_AMD64_DC_CFG:
3054 	case MSR_F15H_EX_CFG:
3055 	/*
3056 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3057 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3058 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3059 	 * so for existing CPU-specific MSRs.
3060 	 */
3061 	case MSR_RAPL_POWER_UNIT:
3062 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3063 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3064 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3065 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3066 		msr_info->data = 0;
3067 		break;
3068 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3069 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3070 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3071 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3072 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3073 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3074 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3075 		msr_info->data = 0;
3076 		break;
3077 	case MSR_IA32_UCODE_REV:
3078 		msr_info->data = vcpu->arch.microcode_version;
3079 		break;
3080 	case MSR_IA32_ARCH_CAPABILITIES:
3081 		if (!msr_info->host_initiated &&
3082 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3083 			return 1;
3084 		msr_info->data = vcpu->arch.arch_capabilities;
3085 		break;
3086 	case MSR_IA32_POWER_CTL:
3087 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3088 		break;
3089 	case MSR_IA32_TSC:
3090 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3091 		break;
3092 	case MSR_MTRRcap:
3093 	case 0x200 ... 0x2ff:
3094 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3095 	case 0xcd: /* fsb frequency */
3096 		msr_info->data = 3;
3097 		break;
3098 		/*
3099 		 * MSR_EBC_FREQUENCY_ID
3100 		 * Conservative value valid for even the basic CPU models.
3101 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3102 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3103 		 * and 266MHz for model 3, or 4. Set Core Clock
3104 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3105 		 * 31:24) even though these are only valid for CPU
3106 		 * models > 2, however guests may end up dividing or
3107 		 * multiplying by zero otherwise.
3108 		 */
3109 	case MSR_EBC_FREQUENCY_ID:
3110 		msr_info->data = 1 << 24;
3111 		break;
3112 	case MSR_IA32_APICBASE:
3113 		msr_info->data = kvm_get_apic_base(vcpu);
3114 		break;
3115 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3116 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3117 	case MSR_IA32_TSCDEADLINE:
3118 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3119 		break;
3120 	case MSR_IA32_TSC_ADJUST:
3121 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3122 		break;
3123 	case MSR_IA32_MISC_ENABLE:
3124 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3125 		break;
3126 	case MSR_IA32_SMBASE:
3127 		if (!msr_info->host_initiated)
3128 			return 1;
3129 		msr_info->data = vcpu->arch.smbase;
3130 		break;
3131 	case MSR_SMI_COUNT:
3132 		msr_info->data = vcpu->arch.smi_count;
3133 		break;
3134 	case MSR_IA32_PERF_STATUS:
3135 		/* TSC increment by tick */
3136 		msr_info->data = 1000ULL;
3137 		/* CPU multiplier */
3138 		msr_info->data |= (((uint64_t)4ULL) << 40);
3139 		break;
3140 	case MSR_EFER:
3141 		msr_info->data = vcpu->arch.efer;
3142 		break;
3143 	case MSR_KVM_WALL_CLOCK:
3144 	case MSR_KVM_WALL_CLOCK_NEW:
3145 		msr_info->data = vcpu->kvm->arch.wall_clock;
3146 		break;
3147 	case MSR_KVM_SYSTEM_TIME:
3148 	case MSR_KVM_SYSTEM_TIME_NEW:
3149 		msr_info->data = vcpu->arch.time;
3150 		break;
3151 	case MSR_KVM_ASYNC_PF_EN:
3152 		msr_info->data = vcpu->arch.apf.msr_val;
3153 		break;
3154 	case MSR_KVM_STEAL_TIME:
3155 		msr_info->data = vcpu->arch.st.msr_val;
3156 		break;
3157 	case MSR_KVM_PV_EOI_EN:
3158 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3159 		break;
3160 	case MSR_KVM_POLL_CONTROL:
3161 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3162 		break;
3163 	case MSR_IA32_P5_MC_ADDR:
3164 	case MSR_IA32_P5_MC_TYPE:
3165 	case MSR_IA32_MCG_CAP:
3166 	case MSR_IA32_MCG_CTL:
3167 	case MSR_IA32_MCG_STATUS:
3168 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3169 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3170 				   msr_info->host_initiated);
3171 	case MSR_IA32_XSS:
3172 		if (!msr_info->host_initiated &&
3173 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3174 			return 1;
3175 		msr_info->data = vcpu->arch.ia32_xss;
3176 		break;
3177 	case MSR_K7_CLK_CTL:
3178 		/*
3179 		 * Provide expected ramp-up count for K7. All other
3180 		 * are set to zero, indicating minimum divisors for
3181 		 * every field.
3182 		 *
3183 		 * This prevents guest kernels on AMD host with CPU
3184 		 * type 6, model 8 and higher from exploding due to
3185 		 * the rdmsr failing.
3186 		 */
3187 		msr_info->data = 0x20000000;
3188 		break;
3189 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3190 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3191 	case HV_X64_MSR_CRASH_CTL:
3192 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3193 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3194 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3195 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3196 		return kvm_hv_get_msr_common(vcpu,
3197 					     msr_info->index, &msr_info->data,
3198 					     msr_info->host_initiated);
3199 	case MSR_IA32_BBL_CR_CTL3:
3200 		/* This legacy MSR exists but isn't fully documented in current
3201 		 * silicon.  It is however accessed by winxp in very narrow
3202 		 * scenarios where it sets bit #19, itself documented as
3203 		 * a "reserved" bit.  Best effort attempt to source coherent
3204 		 * read data here should the balance of the register be
3205 		 * interpreted by the guest:
3206 		 *
3207 		 * L2 cache control register 3: 64GB range, 256KB size,
3208 		 * enabled, latency 0x1, configured
3209 		 */
3210 		msr_info->data = 0xbe702111;
3211 		break;
3212 	case MSR_AMD64_OSVW_ID_LENGTH:
3213 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3214 			return 1;
3215 		msr_info->data = vcpu->arch.osvw.length;
3216 		break;
3217 	case MSR_AMD64_OSVW_STATUS:
3218 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3219 			return 1;
3220 		msr_info->data = vcpu->arch.osvw.status;
3221 		break;
3222 	case MSR_PLATFORM_INFO:
3223 		if (!msr_info->host_initiated &&
3224 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3225 			return 1;
3226 		msr_info->data = vcpu->arch.msr_platform_info;
3227 		break;
3228 	case MSR_MISC_FEATURES_ENABLES:
3229 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3230 		break;
3231 	case MSR_K7_HWCR:
3232 		msr_info->data = vcpu->arch.msr_hwcr;
3233 		break;
3234 	default:
3235 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3236 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3237 		if (!ignore_msrs) {
3238 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3239 					       msr_info->index);
3240 			return 1;
3241 		} else {
3242 			if (report_ignored_msrs)
3243 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3244 					msr_info->index);
3245 			msr_info->data = 0;
3246 		}
3247 		break;
3248 	}
3249 	return 0;
3250 }
3251 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3252 
3253 /*
3254  * Read or write a bunch of msrs. All parameters are kernel addresses.
3255  *
3256  * @return number of msrs set successfully.
3257  */
3258 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3259 		    struct kvm_msr_entry *entries,
3260 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3261 				  unsigned index, u64 *data))
3262 {
3263 	int i;
3264 
3265 	for (i = 0; i < msrs->nmsrs; ++i)
3266 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3267 			break;
3268 
3269 	return i;
3270 }
3271 
3272 /*
3273  * Read or write a bunch of msrs. Parameters are user addresses.
3274  *
3275  * @return number of msrs set successfully.
3276  */
3277 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3278 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3279 				unsigned index, u64 *data),
3280 		  int writeback)
3281 {
3282 	struct kvm_msrs msrs;
3283 	struct kvm_msr_entry *entries;
3284 	int r, n;
3285 	unsigned size;
3286 
3287 	r = -EFAULT;
3288 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3289 		goto out;
3290 
3291 	r = -E2BIG;
3292 	if (msrs.nmsrs >= MAX_IO_MSRS)
3293 		goto out;
3294 
3295 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3296 	entries = memdup_user(user_msrs->entries, size);
3297 	if (IS_ERR(entries)) {
3298 		r = PTR_ERR(entries);
3299 		goto out;
3300 	}
3301 
3302 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3303 	if (r < 0)
3304 		goto out_free;
3305 
3306 	r = -EFAULT;
3307 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3308 		goto out_free;
3309 
3310 	r = n;
3311 
3312 out_free:
3313 	kfree(entries);
3314 out:
3315 	return r;
3316 }
3317 
3318 static inline bool kvm_can_mwait_in_guest(void)
3319 {
3320 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3321 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3322 		boot_cpu_has(X86_FEATURE_ARAT);
3323 }
3324 
3325 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3326 {
3327 	int r = 0;
3328 
3329 	switch (ext) {
3330 	case KVM_CAP_IRQCHIP:
3331 	case KVM_CAP_HLT:
3332 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3333 	case KVM_CAP_SET_TSS_ADDR:
3334 	case KVM_CAP_EXT_CPUID:
3335 	case KVM_CAP_EXT_EMUL_CPUID:
3336 	case KVM_CAP_CLOCKSOURCE:
3337 	case KVM_CAP_PIT:
3338 	case KVM_CAP_NOP_IO_DELAY:
3339 	case KVM_CAP_MP_STATE:
3340 	case KVM_CAP_SYNC_MMU:
3341 	case KVM_CAP_USER_NMI:
3342 	case KVM_CAP_REINJECT_CONTROL:
3343 	case KVM_CAP_IRQ_INJECT_STATUS:
3344 	case KVM_CAP_IOEVENTFD:
3345 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3346 	case KVM_CAP_PIT2:
3347 	case KVM_CAP_PIT_STATE2:
3348 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3349 	case KVM_CAP_XEN_HVM:
3350 	case KVM_CAP_VCPU_EVENTS:
3351 	case KVM_CAP_HYPERV:
3352 	case KVM_CAP_HYPERV_VAPIC:
3353 	case KVM_CAP_HYPERV_SPIN:
3354 	case KVM_CAP_HYPERV_SYNIC:
3355 	case KVM_CAP_HYPERV_SYNIC2:
3356 	case KVM_CAP_HYPERV_VP_INDEX:
3357 	case KVM_CAP_HYPERV_EVENTFD:
3358 	case KVM_CAP_HYPERV_TLBFLUSH:
3359 	case KVM_CAP_HYPERV_SEND_IPI:
3360 	case KVM_CAP_HYPERV_CPUID:
3361 	case KVM_CAP_PCI_SEGMENT:
3362 	case KVM_CAP_DEBUGREGS:
3363 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3364 	case KVM_CAP_XSAVE:
3365 	case KVM_CAP_ASYNC_PF:
3366 	case KVM_CAP_GET_TSC_KHZ:
3367 	case KVM_CAP_KVMCLOCK_CTRL:
3368 	case KVM_CAP_READONLY_MEM:
3369 	case KVM_CAP_HYPERV_TIME:
3370 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3371 	case KVM_CAP_TSC_DEADLINE_TIMER:
3372 	case KVM_CAP_DISABLE_QUIRKS:
3373 	case KVM_CAP_SET_BOOT_CPU_ID:
3374  	case KVM_CAP_SPLIT_IRQCHIP:
3375 	case KVM_CAP_IMMEDIATE_EXIT:
3376 	case KVM_CAP_PMU_EVENT_FILTER:
3377 	case KVM_CAP_GET_MSR_FEATURES:
3378 	case KVM_CAP_MSR_PLATFORM_INFO:
3379 	case KVM_CAP_EXCEPTION_PAYLOAD:
3380 	case KVM_CAP_SET_GUEST_DEBUG:
3381 		r = 1;
3382 		break;
3383 	case KVM_CAP_SYNC_REGS:
3384 		r = KVM_SYNC_X86_VALID_FIELDS;
3385 		break;
3386 	case KVM_CAP_ADJUST_CLOCK:
3387 		r = KVM_CLOCK_TSC_STABLE;
3388 		break;
3389 	case KVM_CAP_X86_DISABLE_EXITS:
3390 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3391 		      KVM_X86_DISABLE_EXITS_CSTATE;
3392 		if(kvm_can_mwait_in_guest())
3393 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3394 		break;
3395 	case KVM_CAP_X86_SMM:
3396 		/* SMBASE is usually relocated above 1M on modern chipsets,
3397 		 * and SMM handlers might indeed rely on 4G segment limits,
3398 		 * so do not report SMM to be available if real mode is
3399 		 * emulated via vm86 mode.  Still, do not go to great lengths
3400 		 * to avoid userspace's usage of the feature, because it is a
3401 		 * fringe case that is not enabled except via specific settings
3402 		 * of the module parameters.
3403 		 */
3404 		r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3405 		break;
3406 	case KVM_CAP_VAPIC:
3407 		r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3408 		break;
3409 	case KVM_CAP_NR_VCPUS:
3410 		r = KVM_SOFT_MAX_VCPUS;
3411 		break;
3412 	case KVM_CAP_MAX_VCPUS:
3413 		r = KVM_MAX_VCPUS;
3414 		break;
3415 	case KVM_CAP_MAX_VCPU_ID:
3416 		r = KVM_MAX_VCPU_ID;
3417 		break;
3418 	case KVM_CAP_PV_MMU:	/* obsolete */
3419 		r = 0;
3420 		break;
3421 	case KVM_CAP_MCE:
3422 		r = KVM_MAX_MCE_BANKS;
3423 		break;
3424 	case KVM_CAP_XCRS:
3425 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3426 		break;
3427 	case KVM_CAP_TSC_CONTROL:
3428 		r = kvm_has_tsc_control;
3429 		break;
3430 	case KVM_CAP_X2APIC_API:
3431 		r = KVM_X2APIC_API_VALID_FLAGS;
3432 		break;
3433 	case KVM_CAP_NESTED_STATE:
3434 		r = kvm_x86_ops.get_nested_state ?
3435 			kvm_x86_ops.get_nested_state(NULL, NULL, 0) : 0;
3436 		break;
3437 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3438 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3439 		break;
3440 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3441 		r = kvm_x86_ops.nested_enable_evmcs != NULL;
3442 		break;
3443 	default:
3444 		break;
3445 	}
3446 	return r;
3447 
3448 }
3449 
3450 long kvm_arch_dev_ioctl(struct file *filp,
3451 			unsigned int ioctl, unsigned long arg)
3452 {
3453 	void __user *argp = (void __user *)arg;
3454 	long r;
3455 
3456 	switch (ioctl) {
3457 	case KVM_GET_MSR_INDEX_LIST: {
3458 		struct kvm_msr_list __user *user_msr_list = argp;
3459 		struct kvm_msr_list msr_list;
3460 		unsigned n;
3461 
3462 		r = -EFAULT;
3463 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3464 			goto out;
3465 		n = msr_list.nmsrs;
3466 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3467 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3468 			goto out;
3469 		r = -E2BIG;
3470 		if (n < msr_list.nmsrs)
3471 			goto out;
3472 		r = -EFAULT;
3473 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3474 				 num_msrs_to_save * sizeof(u32)))
3475 			goto out;
3476 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3477 				 &emulated_msrs,
3478 				 num_emulated_msrs * sizeof(u32)))
3479 			goto out;
3480 		r = 0;
3481 		break;
3482 	}
3483 	case KVM_GET_SUPPORTED_CPUID:
3484 	case KVM_GET_EMULATED_CPUID: {
3485 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3486 		struct kvm_cpuid2 cpuid;
3487 
3488 		r = -EFAULT;
3489 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3490 			goto out;
3491 
3492 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3493 					    ioctl);
3494 		if (r)
3495 			goto out;
3496 
3497 		r = -EFAULT;
3498 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3499 			goto out;
3500 		r = 0;
3501 		break;
3502 	}
3503 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
3504 		r = -EFAULT;
3505 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3506 				 sizeof(kvm_mce_cap_supported)))
3507 			goto out;
3508 		r = 0;
3509 		break;
3510 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3511 		struct kvm_msr_list __user *user_msr_list = argp;
3512 		struct kvm_msr_list msr_list;
3513 		unsigned int n;
3514 
3515 		r = -EFAULT;
3516 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3517 			goto out;
3518 		n = msr_list.nmsrs;
3519 		msr_list.nmsrs = num_msr_based_features;
3520 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3521 			goto out;
3522 		r = -E2BIG;
3523 		if (n < msr_list.nmsrs)
3524 			goto out;
3525 		r = -EFAULT;
3526 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3527 				 num_msr_based_features * sizeof(u32)))
3528 			goto out;
3529 		r = 0;
3530 		break;
3531 	}
3532 	case KVM_GET_MSRS:
3533 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3534 		break;
3535 	default:
3536 		r = -EINVAL;
3537 		break;
3538 	}
3539 out:
3540 	return r;
3541 }
3542 
3543 static void wbinvd_ipi(void *garbage)
3544 {
3545 	wbinvd();
3546 }
3547 
3548 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3549 {
3550 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3551 }
3552 
3553 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3554 {
3555 	/* Address WBINVD may be executed by guest */
3556 	if (need_emulate_wbinvd(vcpu)) {
3557 		if (kvm_x86_ops.has_wbinvd_exit())
3558 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3559 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3560 			smp_call_function_single(vcpu->cpu,
3561 					wbinvd_ipi, NULL, 1);
3562 	}
3563 
3564 	kvm_x86_ops.vcpu_load(vcpu, cpu);
3565 
3566 	/* Save host pkru register if supported */
3567 	vcpu->arch.host_pkru = read_pkru();
3568 
3569 	/* Apply any externally detected TSC adjustments (due to suspend) */
3570 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3571 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3572 		vcpu->arch.tsc_offset_adjustment = 0;
3573 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3574 	}
3575 
3576 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3577 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3578 				rdtsc() - vcpu->arch.last_host_tsc;
3579 		if (tsc_delta < 0)
3580 			mark_tsc_unstable("KVM discovered backwards TSC");
3581 
3582 		if (kvm_check_tsc_unstable()) {
3583 			u64 offset = kvm_compute_tsc_offset(vcpu,
3584 						vcpu->arch.last_guest_tsc);
3585 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3586 			vcpu->arch.tsc_catchup = 1;
3587 		}
3588 
3589 		if (kvm_lapic_hv_timer_in_use(vcpu))
3590 			kvm_lapic_restart_hv_timer(vcpu);
3591 
3592 		/*
3593 		 * On a host with synchronized TSC, there is no need to update
3594 		 * kvmclock on vcpu->cpu migration
3595 		 */
3596 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3597 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3598 		if (vcpu->cpu != cpu)
3599 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3600 		vcpu->cpu = cpu;
3601 	}
3602 
3603 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3604 }
3605 
3606 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3607 {
3608 	struct kvm_host_map map;
3609 	struct kvm_steal_time *st;
3610 
3611 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3612 		return;
3613 
3614 	if (vcpu->arch.st.preempted)
3615 		return;
3616 
3617 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3618 			&vcpu->arch.st.cache, true))
3619 		return;
3620 
3621 	st = map.hva +
3622 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3623 
3624 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3625 
3626 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3627 }
3628 
3629 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3630 {
3631 	int idx;
3632 
3633 	if (vcpu->preempted)
3634 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3635 
3636 	/*
3637 	 * Disable page faults because we're in atomic context here.
3638 	 * kvm_write_guest_offset_cached() would call might_fault()
3639 	 * that relies on pagefault_disable() to tell if there's a
3640 	 * bug. NOTE: the write to guest memory may not go through if
3641 	 * during postcopy live migration or if there's heavy guest
3642 	 * paging.
3643 	 */
3644 	pagefault_disable();
3645 	/*
3646 	 * kvm_memslots() will be called by
3647 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3648 	 */
3649 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3650 	kvm_steal_time_set_preempted(vcpu);
3651 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3652 	pagefault_enable();
3653 	kvm_x86_ops.vcpu_put(vcpu);
3654 	vcpu->arch.last_host_tsc = rdtsc();
3655 	/*
3656 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3657 	 * on every vmexit, but if not, we might have a stale dr6 from the
3658 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3659 	 */
3660 	set_debugreg(0, 6);
3661 }
3662 
3663 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3664 				    struct kvm_lapic_state *s)
3665 {
3666 	if (vcpu->arch.apicv_active)
3667 		kvm_x86_ops.sync_pir_to_irr(vcpu);
3668 
3669 	return kvm_apic_get_state(vcpu, s);
3670 }
3671 
3672 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3673 				    struct kvm_lapic_state *s)
3674 {
3675 	int r;
3676 
3677 	r = kvm_apic_set_state(vcpu, s);
3678 	if (r)
3679 		return r;
3680 	update_cr8_intercept(vcpu);
3681 
3682 	return 0;
3683 }
3684 
3685 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3686 {
3687 	return (!lapic_in_kernel(vcpu) ||
3688 		kvm_apic_accept_pic_intr(vcpu));
3689 }
3690 
3691 /*
3692  * if userspace requested an interrupt window, check that the
3693  * interrupt window is open.
3694  *
3695  * No need to exit to userspace if we already have an interrupt queued.
3696  */
3697 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3698 {
3699 	return kvm_arch_interrupt_allowed(vcpu) &&
3700 		!kvm_cpu_has_interrupt(vcpu) &&
3701 		!kvm_event_needs_reinjection(vcpu) &&
3702 		kvm_cpu_accept_dm_intr(vcpu);
3703 }
3704 
3705 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3706 				    struct kvm_interrupt *irq)
3707 {
3708 	if (irq->irq >= KVM_NR_INTERRUPTS)
3709 		return -EINVAL;
3710 
3711 	if (!irqchip_in_kernel(vcpu->kvm)) {
3712 		kvm_queue_interrupt(vcpu, irq->irq, false);
3713 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3714 		return 0;
3715 	}
3716 
3717 	/*
3718 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3719 	 * fail for in-kernel 8259.
3720 	 */
3721 	if (pic_in_kernel(vcpu->kvm))
3722 		return -ENXIO;
3723 
3724 	if (vcpu->arch.pending_external_vector != -1)
3725 		return -EEXIST;
3726 
3727 	vcpu->arch.pending_external_vector = irq->irq;
3728 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3729 	return 0;
3730 }
3731 
3732 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3733 {
3734 	kvm_inject_nmi(vcpu);
3735 
3736 	return 0;
3737 }
3738 
3739 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3740 {
3741 	kvm_make_request(KVM_REQ_SMI, vcpu);
3742 
3743 	return 0;
3744 }
3745 
3746 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3747 					   struct kvm_tpr_access_ctl *tac)
3748 {
3749 	if (tac->flags)
3750 		return -EINVAL;
3751 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3752 	return 0;
3753 }
3754 
3755 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3756 					u64 mcg_cap)
3757 {
3758 	int r;
3759 	unsigned bank_num = mcg_cap & 0xff, bank;
3760 
3761 	r = -EINVAL;
3762 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
3763 		goto out;
3764 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3765 		goto out;
3766 	r = 0;
3767 	vcpu->arch.mcg_cap = mcg_cap;
3768 	/* Init IA32_MCG_CTL to all 1s */
3769 	if (mcg_cap & MCG_CTL_P)
3770 		vcpu->arch.mcg_ctl = ~(u64)0;
3771 	/* Init IA32_MCi_CTL to all 1s */
3772 	for (bank = 0; bank < bank_num; bank++)
3773 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3774 
3775 	kvm_x86_ops.setup_mce(vcpu);
3776 out:
3777 	return r;
3778 }
3779 
3780 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3781 				      struct kvm_x86_mce *mce)
3782 {
3783 	u64 mcg_cap = vcpu->arch.mcg_cap;
3784 	unsigned bank_num = mcg_cap & 0xff;
3785 	u64 *banks = vcpu->arch.mce_banks;
3786 
3787 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3788 		return -EINVAL;
3789 	/*
3790 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3791 	 * reporting is disabled
3792 	 */
3793 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3794 	    vcpu->arch.mcg_ctl != ~(u64)0)
3795 		return 0;
3796 	banks += 4 * mce->bank;
3797 	/*
3798 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3799 	 * reporting is disabled for the bank
3800 	 */
3801 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3802 		return 0;
3803 	if (mce->status & MCI_STATUS_UC) {
3804 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3805 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3806 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3807 			return 0;
3808 		}
3809 		if (banks[1] & MCI_STATUS_VAL)
3810 			mce->status |= MCI_STATUS_OVER;
3811 		banks[2] = mce->addr;
3812 		banks[3] = mce->misc;
3813 		vcpu->arch.mcg_status = mce->mcg_status;
3814 		banks[1] = mce->status;
3815 		kvm_queue_exception(vcpu, MC_VECTOR);
3816 	} else if (!(banks[1] & MCI_STATUS_VAL)
3817 		   || !(banks[1] & MCI_STATUS_UC)) {
3818 		if (banks[1] & MCI_STATUS_VAL)
3819 			mce->status |= MCI_STATUS_OVER;
3820 		banks[2] = mce->addr;
3821 		banks[3] = mce->misc;
3822 		banks[1] = mce->status;
3823 	} else
3824 		banks[1] |= MCI_STATUS_OVER;
3825 	return 0;
3826 }
3827 
3828 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3829 					       struct kvm_vcpu_events *events)
3830 {
3831 	process_nmi(vcpu);
3832 
3833 	/*
3834 	 * In guest mode, payload delivery should be deferred,
3835 	 * so that the L1 hypervisor can intercept #PF before
3836 	 * CR2 is modified (or intercept #DB before DR6 is
3837 	 * modified under nVMX). Unless the per-VM capability,
3838 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3839 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3840 	 * opportunistically defer the exception payload, deliver it if the
3841 	 * capability hasn't been requested before processing a
3842 	 * KVM_GET_VCPU_EVENTS.
3843 	 */
3844 	if (!vcpu->kvm->arch.exception_payload_enabled &&
3845 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3846 		kvm_deliver_exception_payload(vcpu);
3847 
3848 	/*
3849 	 * The API doesn't provide the instruction length for software
3850 	 * exceptions, so don't report them. As long as the guest RIP
3851 	 * isn't advanced, we should expect to encounter the exception
3852 	 * again.
3853 	 */
3854 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3855 		events->exception.injected = 0;
3856 		events->exception.pending = 0;
3857 	} else {
3858 		events->exception.injected = vcpu->arch.exception.injected;
3859 		events->exception.pending = vcpu->arch.exception.pending;
3860 		/*
3861 		 * For ABI compatibility, deliberately conflate
3862 		 * pending and injected exceptions when
3863 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3864 		 */
3865 		if (!vcpu->kvm->arch.exception_payload_enabled)
3866 			events->exception.injected |=
3867 				vcpu->arch.exception.pending;
3868 	}
3869 	events->exception.nr = vcpu->arch.exception.nr;
3870 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3871 	events->exception.error_code = vcpu->arch.exception.error_code;
3872 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3873 	events->exception_payload = vcpu->arch.exception.payload;
3874 
3875 	events->interrupt.injected =
3876 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3877 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3878 	events->interrupt.soft = 0;
3879 	events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3880 
3881 	events->nmi.injected = vcpu->arch.nmi_injected;
3882 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3883 	events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3884 	events->nmi.pad = 0;
3885 
3886 	events->sipi_vector = 0; /* never valid when reporting to user space */
3887 
3888 	events->smi.smm = is_smm(vcpu);
3889 	events->smi.pending = vcpu->arch.smi_pending;
3890 	events->smi.smm_inside_nmi =
3891 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3892 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3893 
3894 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3895 			 | KVM_VCPUEVENT_VALID_SHADOW
3896 			 | KVM_VCPUEVENT_VALID_SMM);
3897 	if (vcpu->kvm->arch.exception_payload_enabled)
3898 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3899 
3900 	memset(&events->reserved, 0, sizeof(events->reserved));
3901 }
3902 
3903 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3904 
3905 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3906 					      struct kvm_vcpu_events *events)
3907 {
3908 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3909 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3910 			      | KVM_VCPUEVENT_VALID_SHADOW
3911 			      | KVM_VCPUEVENT_VALID_SMM
3912 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3913 		return -EINVAL;
3914 
3915 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3916 		if (!vcpu->kvm->arch.exception_payload_enabled)
3917 			return -EINVAL;
3918 		if (events->exception.pending)
3919 			events->exception.injected = 0;
3920 		else
3921 			events->exception_has_payload = 0;
3922 	} else {
3923 		events->exception.pending = 0;
3924 		events->exception_has_payload = 0;
3925 	}
3926 
3927 	if ((events->exception.injected || events->exception.pending) &&
3928 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3929 		return -EINVAL;
3930 
3931 	/* INITs are latched while in SMM */
3932 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3933 	    (events->smi.smm || events->smi.pending) &&
3934 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3935 		return -EINVAL;
3936 
3937 	process_nmi(vcpu);
3938 	vcpu->arch.exception.injected = events->exception.injected;
3939 	vcpu->arch.exception.pending = events->exception.pending;
3940 	vcpu->arch.exception.nr = events->exception.nr;
3941 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3942 	vcpu->arch.exception.error_code = events->exception.error_code;
3943 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3944 	vcpu->arch.exception.payload = events->exception_payload;
3945 
3946 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3947 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3948 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3949 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3950 		kvm_x86_ops.set_interrupt_shadow(vcpu,
3951 						  events->interrupt.shadow);
3952 
3953 	vcpu->arch.nmi_injected = events->nmi.injected;
3954 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3955 		vcpu->arch.nmi_pending = events->nmi.pending;
3956 	kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
3957 
3958 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3959 	    lapic_in_kernel(vcpu))
3960 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3961 
3962 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3963 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3964 			if (events->smi.smm)
3965 				vcpu->arch.hflags |= HF_SMM_MASK;
3966 			else
3967 				vcpu->arch.hflags &= ~HF_SMM_MASK;
3968 			kvm_smm_changed(vcpu);
3969 		}
3970 
3971 		vcpu->arch.smi_pending = events->smi.pending;
3972 
3973 		if (events->smi.smm) {
3974 			if (events->smi.smm_inside_nmi)
3975 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3976 			else
3977 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3978 		}
3979 
3980 		if (lapic_in_kernel(vcpu)) {
3981 			if (events->smi.latched_init)
3982 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3983 			else
3984 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3985 		}
3986 	}
3987 
3988 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3989 
3990 	return 0;
3991 }
3992 
3993 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3994 					     struct kvm_debugregs *dbgregs)
3995 {
3996 	unsigned long val;
3997 
3998 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3999 	kvm_get_dr(vcpu, 6, &val);
4000 	dbgregs->dr6 = val;
4001 	dbgregs->dr7 = vcpu->arch.dr7;
4002 	dbgregs->flags = 0;
4003 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4004 }
4005 
4006 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4007 					    struct kvm_debugregs *dbgregs)
4008 {
4009 	if (dbgregs->flags)
4010 		return -EINVAL;
4011 
4012 	if (dbgregs->dr6 & ~0xffffffffull)
4013 		return -EINVAL;
4014 	if (dbgregs->dr7 & ~0xffffffffull)
4015 		return -EINVAL;
4016 
4017 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4018 	kvm_update_dr0123(vcpu);
4019 	vcpu->arch.dr6 = dbgregs->dr6;
4020 	vcpu->arch.dr7 = dbgregs->dr7;
4021 	kvm_update_dr7(vcpu);
4022 
4023 	return 0;
4024 }
4025 
4026 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4027 
4028 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4029 {
4030 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4031 	u64 xstate_bv = xsave->header.xfeatures;
4032 	u64 valid;
4033 
4034 	/*
4035 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4036 	 * leaves 0 and 1 in the loop below.
4037 	 */
4038 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4039 
4040 	/* Set XSTATE_BV */
4041 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4042 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4043 
4044 	/*
4045 	 * Copy each region from the possibly compacted offset to the
4046 	 * non-compacted offset.
4047 	 */
4048 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4049 	while (valid) {
4050 		u64 xfeature_mask = valid & -valid;
4051 		int xfeature_nr = fls64(xfeature_mask) - 1;
4052 		void *src = get_xsave_addr(xsave, xfeature_nr);
4053 
4054 		if (src) {
4055 			u32 size, offset, ecx, edx;
4056 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4057 				    &size, &offset, &ecx, &edx);
4058 			if (xfeature_nr == XFEATURE_PKRU)
4059 				memcpy(dest + offset, &vcpu->arch.pkru,
4060 				       sizeof(vcpu->arch.pkru));
4061 			else
4062 				memcpy(dest + offset, src, size);
4063 
4064 		}
4065 
4066 		valid -= xfeature_mask;
4067 	}
4068 }
4069 
4070 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4071 {
4072 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4073 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4074 	u64 valid;
4075 
4076 	/*
4077 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4078 	 * leaves 0 and 1 in the loop below.
4079 	 */
4080 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4081 
4082 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4083 	xsave->header.xfeatures = xstate_bv;
4084 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4085 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4086 
4087 	/*
4088 	 * Copy each region from the non-compacted offset to the
4089 	 * possibly compacted offset.
4090 	 */
4091 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4092 	while (valid) {
4093 		u64 xfeature_mask = valid & -valid;
4094 		int xfeature_nr = fls64(xfeature_mask) - 1;
4095 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4096 
4097 		if (dest) {
4098 			u32 size, offset, ecx, edx;
4099 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4100 				    &size, &offset, &ecx, &edx);
4101 			if (xfeature_nr == XFEATURE_PKRU)
4102 				memcpy(&vcpu->arch.pkru, src + offset,
4103 				       sizeof(vcpu->arch.pkru));
4104 			else
4105 				memcpy(dest, src + offset, size);
4106 		}
4107 
4108 		valid -= xfeature_mask;
4109 	}
4110 }
4111 
4112 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4113 					 struct kvm_xsave *guest_xsave)
4114 {
4115 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4116 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4117 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4118 	} else {
4119 		memcpy(guest_xsave->region,
4120 			&vcpu->arch.guest_fpu->state.fxsave,
4121 			sizeof(struct fxregs_state));
4122 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4123 			XFEATURE_MASK_FPSSE;
4124 	}
4125 }
4126 
4127 #define XSAVE_MXCSR_OFFSET 24
4128 
4129 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4130 					struct kvm_xsave *guest_xsave)
4131 {
4132 	u64 xstate_bv =
4133 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4134 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4135 
4136 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4137 		/*
4138 		 * Here we allow setting states that are not present in
4139 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4140 		 * with old userspace.
4141 		 */
4142 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4143 			return -EINVAL;
4144 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4145 	} else {
4146 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4147 			mxcsr & ~mxcsr_feature_mask)
4148 			return -EINVAL;
4149 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4150 			guest_xsave->region, sizeof(struct fxregs_state));
4151 	}
4152 	return 0;
4153 }
4154 
4155 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4156 					struct kvm_xcrs *guest_xcrs)
4157 {
4158 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4159 		guest_xcrs->nr_xcrs = 0;
4160 		return;
4161 	}
4162 
4163 	guest_xcrs->nr_xcrs = 1;
4164 	guest_xcrs->flags = 0;
4165 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4166 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4167 }
4168 
4169 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4170 				       struct kvm_xcrs *guest_xcrs)
4171 {
4172 	int i, r = 0;
4173 
4174 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4175 		return -EINVAL;
4176 
4177 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4178 		return -EINVAL;
4179 
4180 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4181 		/* Only support XCR0 currently */
4182 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4183 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4184 				guest_xcrs->xcrs[i].value);
4185 			break;
4186 		}
4187 	if (r)
4188 		r = -EINVAL;
4189 	return r;
4190 }
4191 
4192 /*
4193  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4194  * stopped by the hypervisor.  This function will be called from the host only.
4195  * EINVAL is returned when the host attempts to set the flag for a guest that
4196  * does not support pv clocks.
4197  */
4198 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4199 {
4200 	if (!vcpu->arch.pv_time_enabled)
4201 		return -EINVAL;
4202 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4203 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4204 	return 0;
4205 }
4206 
4207 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4208 				     struct kvm_enable_cap *cap)
4209 {
4210 	int r;
4211 	uint16_t vmcs_version;
4212 	void __user *user_ptr;
4213 
4214 	if (cap->flags)
4215 		return -EINVAL;
4216 
4217 	switch (cap->cap) {
4218 	case KVM_CAP_HYPERV_SYNIC2:
4219 		if (cap->args[0])
4220 			return -EINVAL;
4221 		/* fall through */
4222 
4223 	case KVM_CAP_HYPERV_SYNIC:
4224 		if (!irqchip_in_kernel(vcpu->kvm))
4225 			return -EINVAL;
4226 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4227 					     KVM_CAP_HYPERV_SYNIC2);
4228 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4229 		if (!kvm_x86_ops.nested_enable_evmcs)
4230 			return -ENOTTY;
4231 		r = kvm_x86_ops.nested_enable_evmcs(vcpu, &vmcs_version);
4232 		if (!r) {
4233 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4234 			if (copy_to_user(user_ptr, &vmcs_version,
4235 					 sizeof(vmcs_version)))
4236 				r = -EFAULT;
4237 		}
4238 		return r;
4239 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4240 		if (!kvm_x86_ops.enable_direct_tlbflush)
4241 			return -ENOTTY;
4242 
4243 		return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4244 
4245 	default:
4246 		return -EINVAL;
4247 	}
4248 }
4249 
4250 long kvm_arch_vcpu_ioctl(struct file *filp,
4251 			 unsigned int ioctl, unsigned long arg)
4252 {
4253 	struct kvm_vcpu *vcpu = filp->private_data;
4254 	void __user *argp = (void __user *)arg;
4255 	int r;
4256 	union {
4257 		struct kvm_lapic_state *lapic;
4258 		struct kvm_xsave *xsave;
4259 		struct kvm_xcrs *xcrs;
4260 		void *buffer;
4261 	} u;
4262 
4263 	vcpu_load(vcpu);
4264 
4265 	u.buffer = NULL;
4266 	switch (ioctl) {
4267 	case KVM_GET_LAPIC: {
4268 		r = -EINVAL;
4269 		if (!lapic_in_kernel(vcpu))
4270 			goto out;
4271 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4272 				GFP_KERNEL_ACCOUNT);
4273 
4274 		r = -ENOMEM;
4275 		if (!u.lapic)
4276 			goto out;
4277 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4278 		if (r)
4279 			goto out;
4280 		r = -EFAULT;
4281 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4282 			goto out;
4283 		r = 0;
4284 		break;
4285 	}
4286 	case KVM_SET_LAPIC: {
4287 		r = -EINVAL;
4288 		if (!lapic_in_kernel(vcpu))
4289 			goto out;
4290 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4291 		if (IS_ERR(u.lapic)) {
4292 			r = PTR_ERR(u.lapic);
4293 			goto out_nofree;
4294 		}
4295 
4296 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4297 		break;
4298 	}
4299 	case KVM_INTERRUPT: {
4300 		struct kvm_interrupt irq;
4301 
4302 		r = -EFAULT;
4303 		if (copy_from_user(&irq, argp, sizeof(irq)))
4304 			goto out;
4305 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4306 		break;
4307 	}
4308 	case KVM_NMI: {
4309 		r = kvm_vcpu_ioctl_nmi(vcpu);
4310 		break;
4311 	}
4312 	case KVM_SMI: {
4313 		r = kvm_vcpu_ioctl_smi(vcpu);
4314 		break;
4315 	}
4316 	case KVM_SET_CPUID: {
4317 		struct kvm_cpuid __user *cpuid_arg = argp;
4318 		struct kvm_cpuid cpuid;
4319 
4320 		r = -EFAULT;
4321 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4322 			goto out;
4323 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4324 		break;
4325 	}
4326 	case KVM_SET_CPUID2: {
4327 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4328 		struct kvm_cpuid2 cpuid;
4329 
4330 		r = -EFAULT;
4331 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4332 			goto out;
4333 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4334 					      cpuid_arg->entries);
4335 		break;
4336 	}
4337 	case KVM_GET_CPUID2: {
4338 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4339 		struct kvm_cpuid2 cpuid;
4340 
4341 		r = -EFAULT;
4342 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4343 			goto out;
4344 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4345 					      cpuid_arg->entries);
4346 		if (r)
4347 			goto out;
4348 		r = -EFAULT;
4349 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4350 			goto out;
4351 		r = 0;
4352 		break;
4353 	}
4354 	case KVM_GET_MSRS: {
4355 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4356 		r = msr_io(vcpu, argp, do_get_msr, 1);
4357 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4358 		break;
4359 	}
4360 	case KVM_SET_MSRS: {
4361 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4362 		r = msr_io(vcpu, argp, do_set_msr, 0);
4363 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4364 		break;
4365 	}
4366 	case KVM_TPR_ACCESS_REPORTING: {
4367 		struct kvm_tpr_access_ctl tac;
4368 
4369 		r = -EFAULT;
4370 		if (copy_from_user(&tac, argp, sizeof(tac)))
4371 			goto out;
4372 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4373 		if (r)
4374 			goto out;
4375 		r = -EFAULT;
4376 		if (copy_to_user(argp, &tac, sizeof(tac)))
4377 			goto out;
4378 		r = 0;
4379 		break;
4380 	};
4381 	case KVM_SET_VAPIC_ADDR: {
4382 		struct kvm_vapic_addr va;
4383 		int idx;
4384 
4385 		r = -EINVAL;
4386 		if (!lapic_in_kernel(vcpu))
4387 			goto out;
4388 		r = -EFAULT;
4389 		if (copy_from_user(&va, argp, sizeof(va)))
4390 			goto out;
4391 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4392 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4393 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4394 		break;
4395 	}
4396 	case KVM_X86_SETUP_MCE: {
4397 		u64 mcg_cap;
4398 
4399 		r = -EFAULT;
4400 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4401 			goto out;
4402 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4403 		break;
4404 	}
4405 	case KVM_X86_SET_MCE: {
4406 		struct kvm_x86_mce mce;
4407 
4408 		r = -EFAULT;
4409 		if (copy_from_user(&mce, argp, sizeof(mce)))
4410 			goto out;
4411 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4412 		break;
4413 	}
4414 	case KVM_GET_VCPU_EVENTS: {
4415 		struct kvm_vcpu_events events;
4416 
4417 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4418 
4419 		r = -EFAULT;
4420 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4421 			break;
4422 		r = 0;
4423 		break;
4424 	}
4425 	case KVM_SET_VCPU_EVENTS: {
4426 		struct kvm_vcpu_events events;
4427 
4428 		r = -EFAULT;
4429 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4430 			break;
4431 
4432 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4433 		break;
4434 	}
4435 	case KVM_GET_DEBUGREGS: {
4436 		struct kvm_debugregs dbgregs;
4437 
4438 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4439 
4440 		r = -EFAULT;
4441 		if (copy_to_user(argp, &dbgregs,
4442 				 sizeof(struct kvm_debugregs)))
4443 			break;
4444 		r = 0;
4445 		break;
4446 	}
4447 	case KVM_SET_DEBUGREGS: {
4448 		struct kvm_debugregs dbgregs;
4449 
4450 		r = -EFAULT;
4451 		if (copy_from_user(&dbgregs, argp,
4452 				   sizeof(struct kvm_debugregs)))
4453 			break;
4454 
4455 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4456 		break;
4457 	}
4458 	case KVM_GET_XSAVE: {
4459 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4460 		r = -ENOMEM;
4461 		if (!u.xsave)
4462 			break;
4463 
4464 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4465 
4466 		r = -EFAULT;
4467 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4468 			break;
4469 		r = 0;
4470 		break;
4471 	}
4472 	case KVM_SET_XSAVE: {
4473 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4474 		if (IS_ERR(u.xsave)) {
4475 			r = PTR_ERR(u.xsave);
4476 			goto out_nofree;
4477 		}
4478 
4479 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4480 		break;
4481 	}
4482 	case KVM_GET_XCRS: {
4483 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4484 		r = -ENOMEM;
4485 		if (!u.xcrs)
4486 			break;
4487 
4488 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4489 
4490 		r = -EFAULT;
4491 		if (copy_to_user(argp, u.xcrs,
4492 				 sizeof(struct kvm_xcrs)))
4493 			break;
4494 		r = 0;
4495 		break;
4496 	}
4497 	case KVM_SET_XCRS: {
4498 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4499 		if (IS_ERR(u.xcrs)) {
4500 			r = PTR_ERR(u.xcrs);
4501 			goto out_nofree;
4502 		}
4503 
4504 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4505 		break;
4506 	}
4507 	case KVM_SET_TSC_KHZ: {
4508 		u32 user_tsc_khz;
4509 
4510 		r = -EINVAL;
4511 		user_tsc_khz = (u32)arg;
4512 
4513 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4514 			goto out;
4515 
4516 		if (user_tsc_khz == 0)
4517 			user_tsc_khz = tsc_khz;
4518 
4519 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4520 			r = 0;
4521 
4522 		goto out;
4523 	}
4524 	case KVM_GET_TSC_KHZ: {
4525 		r = vcpu->arch.virtual_tsc_khz;
4526 		goto out;
4527 	}
4528 	case KVM_KVMCLOCK_CTRL: {
4529 		r = kvm_set_guest_paused(vcpu);
4530 		goto out;
4531 	}
4532 	case KVM_ENABLE_CAP: {
4533 		struct kvm_enable_cap cap;
4534 
4535 		r = -EFAULT;
4536 		if (copy_from_user(&cap, argp, sizeof(cap)))
4537 			goto out;
4538 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4539 		break;
4540 	}
4541 	case KVM_GET_NESTED_STATE: {
4542 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4543 		u32 user_data_size;
4544 
4545 		r = -EINVAL;
4546 		if (!kvm_x86_ops.get_nested_state)
4547 			break;
4548 
4549 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4550 		r = -EFAULT;
4551 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4552 			break;
4553 
4554 		r = kvm_x86_ops.get_nested_state(vcpu, user_kvm_nested_state,
4555 						  user_data_size);
4556 		if (r < 0)
4557 			break;
4558 
4559 		if (r > user_data_size) {
4560 			if (put_user(r, &user_kvm_nested_state->size))
4561 				r = -EFAULT;
4562 			else
4563 				r = -E2BIG;
4564 			break;
4565 		}
4566 
4567 		r = 0;
4568 		break;
4569 	}
4570 	case KVM_SET_NESTED_STATE: {
4571 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4572 		struct kvm_nested_state kvm_state;
4573 		int idx;
4574 
4575 		r = -EINVAL;
4576 		if (!kvm_x86_ops.set_nested_state)
4577 			break;
4578 
4579 		r = -EFAULT;
4580 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4581 			break;
4582 
4583 		r = -EINVAL;
4584 		if (kvm_state.size < sizeof(kvm_state))
4585 			break;
4586 
4587 		if (kvm_state.flags &
4588 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4589 		      | KVM_STATE_NESTED_EVMCS))
4590 			break;
4591 
4592 		/* nested_run_pending implies guest_mode.  */
4593 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4594 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4595 			break;
4596 
4597 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4598 		r = kvm_x86_ops.set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4599 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4600 		break;
4601 	}
4602 	case KVM_GET_SUPPORTED_HV_CPUID: {
4603 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4604 		struct kvm_cpuid2 cpuid;
4605 
4606 		r = -EFAULT;
4607 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4608 			goto out;
4609 
4610 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4611 						cpuid_arg->entries);
4612 		if (r)
4613 			goto out;
4614 
4615 		r = -EFAULT;
4616 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4617 			goto out;
4618 		r = 0;
4619 		break;
4620 	}
4621 	default:
4622 		r = -EINVAL;
4623 	}
4624 out:
4625 	kfree(u.buffer);
4626 out_nofree:
4627 	vcpu_put(vcpu);
4628 	return r;
4629 }
4630 
4631 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4632 {
4633 	return VM_FAULT_SIGBUS;
4634 }
4635 
4636 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4637 {
4638 	int ret;
4639 
4640 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4641 		return -EINVAL;
4642 	ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4643 	return ret;
4644 }
4645 
4646 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4647 					      u64 ident_addr)
4648 {
4649 	return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4650 }
4651 
4652 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4653 					 unsigned long kvm_nr_mmu_pages)
4654 {
4655 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4656 		return -EINVAL;
4657 
4658 	mutex_lock(&kvm->slots_lock);
4659 
4660 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4661 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4662 
4663 	mutex_unlock(&kvm->slots_lock);
4664 	return 0;
4665 }
4666 
4667 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4668 {
4669 	return kvm->arch.n_max_mmu_pages;
4670 }
4671 
4672 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4673 {
4674 	struct kvm_pic *pic = kvm->arch.vpic;
4675 	int r;
4676 
4677 	r = 0;
4678 	switch (chip->chip_id) {
4679 	case KVM_IRQCHIP_PIC_MASTER:
4680 		memcpy(&chip->chip.pic, &pic->pics[0],
4681 			sizeof(struct kvm_pic_state));
4682 		break;
4683 	case KVM_IRQCHIP_PIC_SLAVE:
4684 		memcpy(&chip->chip.pic, &pic->pics[1],
4685 			sizeof(struct kvm_pic_state));
4686 		break;
4687 	case KVM_IRQCHIP_IOAPIC:
4688 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4689 		break;
4690 	default:
4691 		r = -EINVAL;
4692 		break;
4693 	}
4694 	return r;
4695 }
4696 
4697 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4698 {
4699 	struct kvm_pic *pic = kvm->arch.vpic;
4700 	int r;
4701 
4702 	r = 0;
4703 	switch (chip->chip_id) {
4704 	case KVM_IRQCHIP_PIC_MASTER:
4705 		spin_lock(&pic->lock);
4706 		memcpy(&pic->pics[0], &chip->chip.pic,
4707 			sizeof(struct kvm_pic_state));
4708 		spin_unlock(&pic->lock);
4709 		break;
4710 	case KVM_IRQCHIP_PIC_SLAVE:
4711 		spin_lock(&pic->lock);
4712 		memcpy(&pic->pics[1], &chip->chip.pic,
4713 			sizeof(struct kvm_pic_state));
4714 		spin_unlock(&pic->lock);
4715 		break;
4716 	case KVM_IRQCHIP_IOAPIC:
4717 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4718 		break;
4719 	default:
4720 		r = -EINVAL;
4721 		break;
4722 	}
4723 	kvm_pic_update_irq(pic);
4724 	return r;
4725 }
4726 
4727 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4728 {
4729 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4730 
4731 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4732 
4733 	mutex_lock(&kps->lock);
4734 	memcpy(ps, &kps->channels, sizeof(*ps));
4735 	mutex_unlock(&kps->lock);
4736 	return 0;
4737 }
4738 
4739 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4740 {
4741 	int i;
4742 	struct kvm_pit *pit = kvm->arch.vpit;
4743 
4744 	mutex_lock(&pit->pit_state.lock);
4745 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4746 	for (i = 0; i < 3; i++)
4747 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4748 	mutex_unlock(&pit->pit_state.lock);
4749 	return 0;
4750 }
4751 
4752 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4753 {
4754 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4755 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4756 		sizeof(ps->channels));
4757 	ps->flags = kvm->arch.vpit->pit_state.flags;
4758 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4759 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4760 	return 0;
4761 }
4762 
4763 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4764 {
4765 	int start = 0;
4766 	int i;
4767 	u32 prev_legacy, cur_legacy;
4768 	struct kvm_pit *pit = kvm->arch.vpit;
4769 
4770 	mutex_lock(&pit->pit_state.lock);
4771 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4772 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4773 	if (!prev_legacy && cur_legacy)
4774 		start = 1;
4775 	memcpy(&pit->pit_state.channels, &ps->channels,
4776 	       sizeof(pit->pit_state.channels));
4777 	pit->pit_state.flags = ps->flags;
4778 	for (i = 0; i < 3; i++)
4779 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4780 				   start && i == 0);
4781 	mutex_unlock(&pit->pit_state.lock);
4782 	return 0;
4783 }
4784 
4785 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4786 				 struct kvm_reinject_control *control)
4787 {
4788 	struct kvm_pit *pit = kvm->arch.vpit;
4789 
4790 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4791 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4792 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4793 	 */
4794 	mutex_lock(&pit->pit_state.lock);
4795 	kvm_pit_set_reinject(pit, control->pit_reinject);
4796 	mutex_unlock(&pit->pit_state.lock);
4797 
4798 	return 0;
4799 }
4800 
4801 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4802 {
4803 	/*
4804 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4805 	 */
4806 	if (kvm_x86_ops.flush_log_dirty)
4807 		kvm_x86_ops.flush_log_dirty(kvm);
4808 }
4809 
4810 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4811 			bool line_status)
4812 {
4813 	if (!irqchip_in_kernel(kvm))
4814 		return -ENXIO;
4815 
4816 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4817 					irq_event->irq, irq_event->level,
4818 					line_status);
4819 	return 0;
4820 }
4821 
4822 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4823 			    struct kvm_enable_cap *cap)
4824 {
4825 	int r;
4826 
4827 	if (cap->flags)
4828 		return -EINVAL;
4829 
4830 	switch (cap->cap) {
4831 	case KVM_CAP_DISABLE_QUIRKS:
4832 		kvm->arch.disabled_quirks = cap->args[0];
4833 		r = 0;
4834 		break;
4835 	case KVM_CAP_SPLIT_IRQCHIP: {
4836 		mutex_lock(&kvm->lock);
4837 		r = -EINVAL;
4838 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4839 			goto split_irqchip_unlock;
4840 		r = -EEXIST;
4841 		if (irqchip_in_kernel(kvm))
4842 			goto split_irqchip_unlock;
4843 		if (kvm->created_vcpus)
4844 			goto split_irqchip_unlock;
4845 		r = kvm_setup_empty_irq_routing(kvm);
4846 		if (r)
4847 			goto split_irqchip_unlock;
4848 		/* Pairs with irqchip_in_kernel. */
4849 		smp_wmb();
4850 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4851 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4852 		r = 0;
4853 split_irqchip_unlock:
4854 		mutex_unlock(&kvm->lock);
4855 		break;
4856 	}
4857 	case KVM_CAP_X2APIC_API:
4858 		r = -EINVAL;
4859 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4860 			break;
4861 
4862 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4863 			kvm->arch.x2apic_format = true;
4864 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4865 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4866 
4867 		r = 0;
4868 		break;
4869 	case KVM_CAP_X86_DISABLE_EXITS:
4870 		r = -EINVAL;
4871 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4872 			break;
4873 
4874 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4875 			kvm_can_mwait_in_guest())
4876 			kvm->arch.mwait_in_guest = true;
4877 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4878 			kvm->arch.hlt_in_guest = true;
4879 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4880 			kvm->arch.pause_in_guest = true;
4881 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4882 			kvm->arch.cstate_in_guest = true;
4883 		r = 0;
4884 		break;
4885 	case KVM_CAP_MSR_PLATFORM_INFO:
4886 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4887 		r = 0;
4888 		break;
4889 	case KVM_CAP_EXCEPTION_PAYLOAD:
4890 		kvm->arch.exception_payload_enabled = cap->args[0];
4891 		r = 0;
4892 		break;
4893 	default:
4894 		r = -EINVAL;
4895 		break;
4896 	}
4897 	return r;
4898 }
4899 
4900 long kvm_arch_vm_ioctl(struct file *filp,
4901 		       unsigned int ioctl, unsigned long arg)
4902 {
4903 	struct kvm *kvm = filp->private_data;
4904 	void __user *argp = (void __user *)arg;
4905 	int r = -ENOTTY;
4906 	/*
4907 	 * This union makes it completely explicit to gcc-3.x
4908 	 * that these two variables' stack usage should be
4909 	 * combined, not added together.
4910 	 */
4911 	union {
4912 		struct kvm_pit_state ps;
4913 		struct kvm_pit_state2 ps2;
4914 		struct kvm_pit_config pit_config;
4915 	} u;
4916 
4917 	switch (ioctl) {
4918 	case KVM_SET_TSS_ADDR:
4919 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4920 		break;
4921 	case KVM_SET_IDENTITY_MAP_ADDR: {
4922 		u64 ident_addr;
4923 
4924 		mutex_lock(&kvm->lock);
4925 		r = -EINVAL;
4926 		if (kvm->created_vcpus)
4927 			goto set_identity_unlock;
4928 		r = -EFAULT;
4929 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4930 			goto set_identity_unlock;
4931 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4932 set_identity_unlock:
4933 		mutex_unlock(&kvm->lock);
4934 		break;
4935 	}
4936 	case KVM_SET_NR_MMU_PAGES:
4937 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4938 		break;
4939 	case KVM_GET_NR_MMU_PAGES:
4940 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4941 		break;
4942 	case KVM_CREATE_IRQCHIP: {
4943 		mutex_lock(&kvm->lock);
4944 
4945 		r = -EEXIST;
4946 		if (irqchip_in_kernel(kvm))
4947 			goto create_irqchip_unlock;
4948 
4949 		r = -EINVAL;
4950 		if (kvm->created_vcpus)
4951 			goto create_irqchip_unlock;
4952 
4953 		r = kvm_pic_init(kvm);
4954 		if (r)
4955 			goto create_irqchip_unlock;
4956 
4957 		r = kvm_ioapic_init(kvm);
4958 		if (r) {
4959 			kvm_pic_destroy(kvm);
4960 			goto create_irqchip_unlock;
4961 		}
4962 
4963 		r = kvm_setup_default_irq_routing(kvm);
4964 		if (r) {
4965 			kvm_ioapic_destroy(kvm);
4966 			kvm_pic_destroy(kvm);
4967 			goto create_irqchip_unlock;
4968 		}
4969 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4970 		smp_wmb();
4971 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4972 	create_irqchip_unlock:
4973 		mutex_unlock(&kvm->lock);
4974 		break;
4975 	}
4976 	case KVM_CREATE_PIT:
4977 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4978 		goto create_pit;
4979 	case KVM_CREATE_PIT2:
4980 		r = -EFAULT;
4981 		if (copy_from_user(&u.pit_config, argp,
4982 				   sizeof(struct kvm_pit_config)))
4983 			goto out;
4984 	create_pit:
4985 		mutex_lock(&kvm->lock);
4986 		r = -EEXIST;
4987 		if (kvm->arch.vpit)
4988 			goto create_pit_unlock;
4989 		r = -ENOMEM;
4990 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4991 		if (kvm->arch.vpit)
4992 			r = 0;
4993 	create_pit_unlock:
4994 		mutex_unlock(&kvm->lock);
4995 		break;
4996 	case KVM_GET_IRQCHIP: {
4997 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4998 		struct kvm_irqchip *chip;
4999 
5000 		chip = memdup_user(argp, sizeof(*chip));
5001 		if (IS_ERR(chip)) {
5002 			r = PTR_ERR(chip);
5003 			goto out;
5004 		}
5005 
5006 		r = -ENXIO;
5007 		if (!irqchip_kernel(kvm))
5008 			goto get_irqchip_out;
5009 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5010 		if (r)
5011 			goto get_irqchip_out;
5012 		r = -EFAULT;
5013 		if (copy_to_user(argp, chip, sizeof(*chip)))
5014 			goto get_irqchip_out;
5015 		r = 0;
5016 	get_irqchip_out:
5017 		kfree(chip);
5018 		break;
5019 	}
5020 	case KVM_SET_IRQCHIP: {
5021 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5022 		struct kvm_irqchip *chip;
5023 
5024 		chip = memdup_user(argp, sizeof(*chip));
5025 		if (IS_ERR(chip)) {
5026 			r = PTR_ERR(chip);
5027 			goto out;
5028 		}
5029 
5030 		r = -ENXIO;
5031 		if (!irqchip_kernel(kvm))
5032 			goto set_irqchip_out;
5033 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5034 	set_irqchip_out:
5035 		kfree(chip);
5036 		break;
5037 	}
5038 	case KVM_GET_PIT: {
5039 		r = -EFAULT;
5040 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5041 			goto out;
5042 		r = -ENXIO;
5043 		if (!kvm->arch.vpit)
5044 			goto out;
5045 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5046 		if (r)
5047 			goto out;
5048 		r = -EFAULT;
5049 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5050 			goto out;
5051 		r = 0;
5052 		break;
5053 	}
5054 	case KVM_SET_PIT: {
5055 		r = -EFAULT;
5056 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5057 			goto out;
5058 		mutex_lock(&kvm->lock);
5059 		r = -ENXIO;
5060 		if (!kvm->arch.vpit)
5061 			goto set_pit_out;
5062 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5063 set_pit_out:
5064 		mutex_unlock(&kvm->lock);
5065 		break;
5066 	}
5067 	case KVM_GET_PIT2: {
5068 		r = -ENXIO;
5069 		if (!kvm->arch.vpit)
5070 			goto out;
5071 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5072 		if (r)
5073 			goto out;
5074 		r = -EFAULT;
5075 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5076 			goto out;
5077 		r = 0;
5078 		break;
5079 	}
5080 	case KVM_SET_PIT2: {
5081 		r = -EFAULT;
5082 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5083 			goto out;
5084 		mutex_lock(&kvm->lock);
5085 		r = -ENXIO;
5086 		if (!kvm->arch.vpit)
5087 			goto set_pit2_out;
5088 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5089 set_pit2_out:
5090 		mutex_unlock(&kvm->lock);
5091 		break;
5092 	}
5093 	case KVM_REINJECT_CONTROL: {
5094 		struct kvm_reinject_control control;
5095 		r =  -EFAULT;
5096 		if (copy_from_user(&control, argp, sizeof(control)))
5097 			goto out;
5098 		r = -ENXIO;
5099 		if (!kvm->arch.vpit)
5100 			goto out;
5101 		r = kvm_vm_ioctl_reinject(kvm, &control);
5102 		break;
5103 	}
5104 	case KVM_SET_BOOT_CPU_ID:
5105 		r = 0;
5106 		mutex_lock(&kvm->lock);
5107 		if (kvm->created_vcpus)
5108 			r = -EBUSY;
5109 		else
5110 			kvm->arch.bsp_vcpu_id = arg;
5111 		mutex_unlock(&kvm->lock);
5112 		break;
5113 	case KVM_XEN_HVM_CONFIG: {
5114 		struct kvm_xen_hvm_config xhc;
5115 		r = -EFAULT;
5116 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5117 			goto out;
5118 		r = -EINVAL;
5119 		if (xhc.flags)
5120 			goto out;
5121 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5122 		r = 0;
5123 		break;
5124 	}
5125 	case KVM_SET_CLOCK: {
5126 		struct kvm_clock_data user_ns;
5127 		u64 now_ns;
5128 
5129 		r = -EFAULT;
5130 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5131 			goto out;
5132 
5133 		r = -EINVAL;
5134 		if (user_ns.flags)
5135 			goto out;
5136 
5137 		r = 0;
5138 		/*
5139 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5140 		 * kvm_gen_update_masterclock() can be cut down to locked
5141 		 * pvclock_update_vm_gtod_copy().
5142 		 */
5143 		kvm_gen_update_masterclock(kvm);
5144 		now_ns = get_kvmclock_ns(kvm);
5145 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5146 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5147 		break;
5148 	}
5149 	case KVM_GET_CLOCK: {
5150 		struct kvm_clock_data user_ns;
5151 		u64 now_ns;
5152 
5153 		now_ns = get_kvmclock_ns(kvm);
5154 		user_ns.clock = now_ns;
5155 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5156 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5157 
5158 		r = -EFAULT;
5159 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5160 			goto out;
5161 		r = 0;
5162 		break;
5163 	}
5164 	case KVM_MEMORY_ENCRYPT_OP: {
5165 		r = -ENOTTY;
5166 		if (kvm_x86_ops.mem_enc_op)
5167 			r = kvm_x86_ops.mem_enc_op(kvm, argp);
5168 		break;
5169 	}
5170 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5171 		struct kvm_enc_region region;
5172 
5173 		r = -EFAULT;
5174 		if (copy_from_user(&region, argp, sizeof(region)))
5175 			goto out;
5176 
5177 		r = -ENOTTY;
5178 		if (kvm_x86_ops.mem_enc_reg_region)
5179 			r = kvm_x86_ops.mem_enc_reg_region(kvm, &region);
5180 		break;
5181 	}
5182 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5183 		struct kvm_enc_region region;
5184 
5185 		r = -EFAULT;
5186 		if (copy_from_user(&region, argp, sizeof(region)))
5187 			goto out;
5188 
5189 		r = -ENOTTY;
5190 		if (kvm_x86_ops.mem_enc_unreg_region)
5191 			r = kvm_x86_ops.mem_enc_unreg_region(kvm, &region);
5192 		break;
5193 	}
5194 	case KVM_HYPERV_EVENTFD: {
5195 		struct kvm_hyperv_eventfd hvevfd;
5196 
5197 		r = -EFAULT;
5198 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5199 			goto out;
5200 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5201 		break;
5202 	}
5203 	case KVM_SET_PMU_EVENT_FILTER:
5204 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5205 		break;
5206 	default:
5207 		r = -ENOTTY;
5208 	}
5209 out:
5210 	return r;
5211 }
5212 
5213 static void kvm_init_msr_list(void)
5214 {
5215 	struct x86_pmu_capability x86_pmu;
5216 	u32 dummy[2];
5217 	unsigned i;
5218 
5219 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5220 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5221 
5222 	perf_get_x86_pmu_capability(&x86_pmu);
5223 
5224 	num_msrs_to_save = 0;
5225 	num_emulated_msrs = 0;
5226 	num_msr_based_features = 0;
5227 
5228 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5229 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5230 			continue;
5231 
5232 		/*
5233 		 * Even MSRs that are valid in the host may not be exposed
5234 		 * to the guests in some cases.
5235 		 */
5236 		switch (msrs_to_save_all[i]) {
5237 		case MSR_IA32_BNDCFGS:
5238 			if (!kvm_mpx_supported())
5239 				continue;
5240 			break;
5241 		case MSR_TSC_AUX:
5242 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5243 				continue;
5244 			break;
5245 		case MSR_IA32_RTIT_CTL:
5246 		case MSR_IA32_RTIT_STATUS:
5247 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5248 				continue;
5249 			break;
5250 		case MSR_IA32_RTIT_CR3_MATCH:
5251 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5252 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5253 				continue;
5254 			break;
5255 		case MSR_IA32_RTIT_OUTPUT_BASE:
5256 		case MSR_IA32_RTIT_OUTPUT_MASK:
5257 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5258 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5259 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5260 				continue;
5261 			break;
5262 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5263 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5264 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5265 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5266 				continue;
5267 			break;
5268 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5269 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5270 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5271 				continue;
5272 			break;
5273 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5274 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5275 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5276 				continue;
5277 		}
5278 		default:
5279 			break;
5280 		}
5281 
5282 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5283 	}
5284 
5285 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5286 		if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5287 			continue;
5288 
5289 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5290 	}
5291 
5292 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5293 		struct kvm_msr_entry msr;
5294 
5295 		msr.index = msr_based_features_all[i];
5296 		if (kvm_get_msr_feature(&msr))
5297 			continue;
5298 
5299 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5300 	}
5301 }
5302 
5303 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5304 			   const void *v)
5305 {
5306 	int handled = 0;
5307 	int n;
5308 
5309 	do {
5310 		n = min(len, 8);
5311 		if (!(lapic_in_kernel(vcpu) &&
5312 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5313 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5314 			break;
5315 		handled += n;
5316 		addr += n;
5317 		len -= n;
5318 		v += n;
5319 	} while (len);
5320 
5321 	return handled;
5322 }
5323 
5324 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5325 {
5326 	int handled = 0;
5327 	int n;
5328 
5329 	do {
5330 		n = min(len, 8);
5331 		if (!(lapic_in_kernel(vcpu) &&
5332 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5333 					 addr, n, v))
5334 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5335 			break;
5336 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5337 		handled += n;
5338 		addr += n;
5339 		len -= n;
5340 		v += n;
5341 	} while (len);
5342 
5343 	return handled;
5344 }
5345 
5346 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5347 			struct kvm_segment *var, int seg)
5348 {
5349 	kvm_x86_ops.set_segment(vcpu, var, seg);
5350 }
5351 
5352 void kvm_get_segment(struct kvm_vcpu *vcpu,
5353 		     struct kvm_segment *var, int seg)
5354 {
5355 	kvm_x86_ops.get_segment(vcpu, var, seg);
5356 }
5357 
5358 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5359 			   struct x86_exception *exception)
5360 {
5361 	gpa_t t_gpa;
5362 
5363 	BUG_ON(!mmu_is_nested(vcpu));
5364 
5365 	/* NPT walks are always user-walks */
5366 	access |= PFERR_USER_MASK;
5367 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5368 
5369 	return t_gpa;
5370 }
5371 
5372 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5373 			      struct x86_exception *exception)
5374 {
5375 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5376 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5377 }
5378 
5379  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5380 				struct x86_exception *exception)
5381 {
5382 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5383 	access |= PFERR_FETCH_MASK;
5384 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5385 }
5386 
5387 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5388 			       struct x86_exception *exception)
5389 {
5390 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5391 	access |= PFERR_WRITE_MASK;
5392 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5393 }
5394 
5395 /* uses this to access any guest's mapped memory without checking CPL */
5396 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5397 				struct x86_exception *exception)
5398 {
5399 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5400 }
5401 
5402 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5403 				      struct kvm_vcpu *vcpu, u32 access,
5404 				      struct x86_exception *exception)
5405 {
5406 	void *data = val;
5407 	int r = X86EMUL_CONTINUE;
5408 
5409 	while (bytes) {
5410 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5411 							    exception);
5412 		unsigned offset = addr & (PAGE_SIZE-1);
5413 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5414 		int ret;
5415 
5416 		if (gpa == UNMAPPED_GVA)
5417 			return X86EMUL_PROPAGATE_FAULT;
5418 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5419 					       offset, toread);
5420 		if (ret < 0) {
5421 			r = X86EMUL_IO_NEEDED;
5422 			goto out;
5423 		}
5424 
5425 		bytes -= toread;
5426 		data += toread;
5427 		addr += toread;
5428 	}
5429 out:
5430 	return r;
5431 }
5432 
5433 /* used for instruction fetching */
5434 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5435 				gva_t addr, void *val, unsigned int bytes,
5436 				struct x86_exception *exception)
5437 {
5438 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5439 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5440 	unsigned offset;
5441 	int ret;
5442 
5443 	/* Inline kvm_read_guest_virt_helper for speed.  */
5444 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5445 						    exception);
5446 	if (unlikely(gpa == UNMAPPED_GVA))
5447 		return X86EMUL_PROPAGATE_FAULT;
5448 
5449 	offset = addr & (PAGE_SIZE-1);
5450 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5451 		bytes = (unsigned)PAGE_SIZE - offset;
5452 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5453 				       offset, bytes);
5454 	if (unlikely(ret < 0))
5455 		return X86EMUL_IO_NEEDED;
5456 
5457 	return X86EMUL_CONTINUE;
5458 }
5459 
5460 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5461 			       gva_t addr, void *val, unsigned int bytes,
5462 			       struct x86_exception *exception)
5463 {
5464 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5465 
5466 	/*
5467 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5468 	 * is returned, but our callers are not ready for that and they blindly
5469 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5470 	 * uninitialized kernel stack memory into cr2 and error code.
5471 	 */
5472 	memset(exception, 0, sizeof(*exception));
5473 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5474 					  exception);
5475 }
5476 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5477 
5478 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5479 			     gva_t addr, void *val, unsigned int bytes,
5480 			     struct x86_exception *exception, bool system)
5481 {
5482 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5483 	u32 access = 0;
5484 
5485 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5486 		access |= PFERR_USER_MASK;
5487 
5488 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5489 }
5490 
5491 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5492 		unsigned long addr, void *val, unsigned int bytes)
5493 {
5494 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5495 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5496 
5497 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5498 }
5499 
5500 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5501 				      struct kvm_vcpu *vcpu, u32 access,
5502 				      struct x86_exception *exception)
5503 {
5504 	void *data = val;
5505 	int r = X86EMUL_CONTINUE;
5506 
5507 	while (bytes) {
5508 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5509 							     access,
5510 							     exception);
5511 		unsigned offset = addr & (PAGE_SIZE-1);
5512 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5513 		int ret;
5514 
5515 		if (gpa == UNMAPPED_GVA)
5516 			return X86EMUL_PROPAGATE_FAULT;
5517 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5518 		if (ret < 0) {
5519 			r = X86EMUL_IO_NEEDED;
5520 			goto out;
5521 		}
5522 
5523 		bytes -= towrite;
5524 		data += towrite;
5525 		addr += towrite;
5526 	}
5527 out:
5528 	return r;
5529 }
5530 
5531 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5532 			      unsigned int bytes, struct x86_exception *exception,
5533 			      bool system)
5534 {
5535 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5536 	u32 access = PFERR_WRITE_MASK;
5537 
5538 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5539 		access |= PFERR_USER_MASK;
5540 
5541 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5542 					   access, exception);
5543 }
5544 
5545 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5546 				unsigned int bytes, struct x86_exception *exception)
5547 {
5548 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5549 	vcpu->arch.l1tf_flush_l1d = true;
5550 
5551 	/*
5552 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5553 	 * is returned, but our callers are not ready for that and they blindly
5554 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5555 	 * uninitialized kernel stack memory into cr2 and error code.
5556 	 */
5557 	memset(exception, 0, sizeof(*exception));
5558 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5559 					   PFERR_WRITE_MASK, exception);
5560 }
5561 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5562 
5563 int handle_ud(struct kvm_vcpu *vcpu)
5564 {
5565 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5566 	int emul_type = EMULTYPE_TRAP_UD;
5567 	char sig[5]; /* ud2; .ascii "kvm" */
5568 	struct x86_exception e;
5569 
5570 	if (force_emulation_prefix &&
5571 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5572 				sig, sizeof(sig), &e) == 0 &&
5573 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5574 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5575 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5576 	}
5577 
5578 	return kvm_emulate_instruction(vcpu, emul_type);
5579 }
5580 EXPORT_SYMBOL_GPL(handle_ud);
5581 
5582 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5583 			    gpa_t gpa, bool write)
5584 {
5585 	/* For APIC access vmexit */
5586 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5587 		return 1;
5588 
5589 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5590 		trace_vcpu_match_mmio(gva, gpa, write, true);
5591 		return 1;
5592 	}
5593 
5594 	return 0;
5595 }
5596 
5597 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5598 				gpa_t *gpa, struct x86_exception *exception,
5599 				bool write)
5600 {
5601 	u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5602 		| (write ? PFERR_WRITE_MASK : 0);
5603 
5604 	/*
5605 	 * currently PKRU is only applied to ept enabled guest so
5606 	 * there is no pkey in EPT page table for L1 guest or EPT
5607 	 * shadow page table for L2 guest.
5608 	 */
5609 	if (vcpu_match_mmio_gva(vcpu, gva)
5610 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5611 				 vcpu->arch.mmio_access, 0, access)) {
5612 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5613 					(gva & (PAGE_SIZE - 1));
5614 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5615 		return 1;
5616 	}
5617 
5618 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5619 
5620 	if (*gpa == UNMAPPED_GVA)
5621 		return -1;
5622 
5623 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5624 }
5625 
5626 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5627 			const void *val, int bytes)
5628 {
5629 	int ret;
5630 
5631 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5632 	if (ret < 0)
5633 		return 0;
5634 	kvm_page_track_write(vcpu, gpa, val, bytes);
5635 	return 1;
5636 }
5637 
5638 struct read_write_emulator_ops {
5639 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5640 				  int bytes);
5641 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5642 				  void *val, int bytes);
5643 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5644 			       int bytes, void *val);
5645 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5646 				    void *val, int bytes);
5647 	bool write;
5648 };
5649 
5650 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5651 {
5652 	if (vcpu->mmio_read_completed) {
5653 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5654 			       vcpu->mmio_fragments[0].gpa, val);
5655 		vcpu->mmio_read_completed = 0;
5656 		return 1;
5657 	}
5658 
5659 	return 0;
5660 }
5661 
5662 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5663 			void *val, int bytes)
5664 {
5665 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5666 }
5667 
5668 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5669 			 void *val, int bytes)
5670 {
5671 	return emulator_write_phys(vcpu, gpa, val, bytes);
5672 }
5673 
5674 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5675 {
5676 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5677 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5678 }
5679 
5680 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5681 			  void *val, int bytes)
5682 {
5683 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5684 	return X86EMUL_IO_NEEDED;
5685 }
5686 
5687 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5688 			   void *val, int bytes)
5689 {
5690 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5691 
5692 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5693 	return X86EMUL_CONTINUE;
5694 }
5695 
5696 static const struct read_write_emulator_ops read_emultor = {
5697 	.read_write_prepare = read_prepare,
5698 	.read_write_emulate = read_emulate,
5699 	.read_write_mmio = vcpu_mmio_read,
5700 	.read_write_exit_mmio = read_exit_mmio,
5701 };
5702 
5703 static const struct read_write_emulator_ops write_emultor = {
5704 	.read_write_emulate = write_emulate,
5705 	.read_write_mmio = write_mmio,
5706 	.read_write_exit_mmio = write_exit_mmio,
5707 	.write = true,
5708 };
5709 
5710 static int emulator_read_write_onepage(unsigned long addr, void *val,
5711 				       unsigned int bytes,
5712 				       struct x86_exception *exception,
5713 				       struct kvm_vcpu *vcpu,
5714 				       const struct read_write_emulator_ops *ops)
5715 {
5716 	gpa_t gpa;
5717 	int handled, ret;
5718 	bool write = ops->write;
5719 	struct kvm_mmio_fragment *frag;
5720 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5721 
5722 	/*
5723 	 * If the exit was due to a NPF we may already have a GPA.
5724 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5725 	 * Note, this cannot be used on string operations since string
5726 	 * operation using rep will only have the initial GPA from the NPF
5727 	 * occurred.
5728 	 */
5729 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5730 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5731 		gpa = ctxt->gpa_val;
5732 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5733 	} else {
5734 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5735 		if (ret < 0)
5736 			return X86EMUL_PROPAGATE_FAULT;
5737 	}
5738 
5739 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5740 		return X86EMUL_CONTINUE;
5741 
5742 	/*
5743 	 * Is this MMIO handled locally?
5744 	 */
5745 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5746 	if (handled == bytes)
5747 		return X86EMUL_CONTINUE;
5748 
5749 	gpa += handled;
5750 	bytes -= handled;
5751 	val += handled;
5752 
5753 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5754 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5755 	frag->gpa = gpa;
5756 	frag->data = val;
5757 	frag->len = bytes;
5758 	return X86EMUL_CONTINUE;
5759 }
5760 
5761 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5762 			unsigned long addr,
5763 			void *val, unsigned int bytes,
5764 			struct x86_exception *exception,
5765 			const struct read_write_emulator_ops *ops)
5766 {
5767 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5768 	gpa_t gpa;
5769 	int rc;
5770 
5771 	if (ops->read_write_prepare &&
5772 		  ops->read_write_prepare(vcpu, val, bytes))
5773 		return X86EMUL_CONTINUE;
5774 
5775 	vcpu->mmio_nr_fragments = 0;
5776 
5777 	/* Crossing a page boundary? */
5778 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5779 		int now;
5780 
5781 		now = -addr & ~PAGE_MASK;
5782 		rc = emulator_read_write_onepage(addr, val, now, exception,
5783 						 vcpu, ops);
5784 
5785 		if (rc != X86EMUL_CONTINUE)
5786 			return rc;
5787 		addr += now;
5788 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5789 			addr = (u32)addr;
5790 		val += now;
5791 		bytes -= now;
5792 	}
5793 
5794 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5795 					 vcpu, ops);
5796 	if (rc != X86EMUL_CONTINUE)
5797 		return rc;
5798 
5799 	if (!vcpu->mmio_nr_fragments)
5800 		return rc;
5801 
5802 	gpa = vcpu->mmio_fragments[0].gpa;
5803 
5804 	vcpu->mmio_needed = 1;
5805 	vcpu->mmio_cur_fragment = 0;
5806 
5807 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5808 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5809 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5810 	vcpu->run->mmio.phys_addr = gpa;
5811 
5812 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5813 }
5814 
5815 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5816 				  unsigned long addr,
5817 				  void *val,
5818 				  unsigned int bytes,
5819 				  struct x86_exception *exception)
5820 {
5821 	return emulator_read_write(ctxt, addr, val, bytes,
5822 				   exception, &read_emultor);
5823 }
5824 
5825 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5826 			    unsigned long addr,
5827 			    const void *val,
5828 			    unsigned int bytes,
5829 			    struct x86_exception *exception)
5830 {
5831 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5832 				   exception, &write_emultor);
5833 }
5834 
5835 #define CMPXCHG_TYPE(t, ptr, old, new) \
5836 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5837 
5838 #ifdef CONFIG_X86_64
5839 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5840 #else
5841 #  define CMPXCHG64(ptr, old, new) \
5842 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5843 #endif
5844 
5845 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5846 				     unsigned long addr,
5847 				     const void *old,
5848 				     const void *new,
5849 				     unsigned int bytes,
5850 				     struct x86_exception *exception)
5851 {
5852 	struct kvm_host_map map;
5853 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5854 	u64 page_line_mask;
5855 	gpa_t gpa;
5856 	char *kaddr;
5857 	bool exchanged;
5858 
5859 	/* guests cmpxchg8b have to be emulated atomically */
5860 	if (bytes > 8 || (bytes & (bytes - 1)))
5861 		goto emul_write;
5862 
5863 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5864 
5865 	if (gpa == UNMAPPED_GVA ||
5866 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5867 		goto emul_write;
5868 
5869 	/*
5870 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
5871 	 * enabled in the host and the access splits a cache line.
5872 	 */
5873 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5874 		page_line_mask = ~(cache_line_size() - 1);
5875 	else
5876 		page_line_mask = PAGE_MASK;
5877 
5878 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
5879 		goto emul_write;
5880 
5881 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5882 		goto emul_write;
5883 
5884 	kaddr = map.hva + offset_in_page(gpa);
5885 
5886 	switch (bytes) {
5887 	case 1:
5888 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5889 		break;
5890 	case 2:
5891 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5892 		break;
5893 	case 4:
5894 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5895 		break;
5896 	case 8:
5897 		exchanged = CMPXCHG64(kaddr, old, new);
5898 		break;
5899 	default:
5900 		BUG();
5901 	}
5902 
5903 	kvm_vcpu_unmap(vcpu, &map, true);
5904 
5905 	if (!exchanged)
5906 		return X86EMUL_CMPXCHG_FAILED;
5907 
5908 	kvm_page_track_write(vcpu, gpa, new, bytes);
5909 
5910 	return X86EMUL_CONTINUE;
5911 
5912 emul_write:
5913 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5914 
5915 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5916 }
5917 
5918 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5919 {
5920 	int r = 0, i;
5921 
5922 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5923 		if (vcpu->arch.pio.in)
5924 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5925 					    vcpu->arch.pio.size, pd);
5926 		else
5927 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5928 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5929 					     pd);
5930 		if (r)
5931 			break;
5932 		pd += vcpu->arch.pio.size;
5933 	}
5934 	return r;
5935 }
5936 
5937 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5938 			       unsigned short port, void *val,
5939 			       unsigned int count, bool in)
5940 {
5941 	vcpu->arch.pio.port = port;
5942 	vcpu->arch.pio.in = in;
5943 	vcpu->arch.pio.count  = count;
5944 	vcpu->arch.pio.size = size;
5945 
5946 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5947 		vcpu->arch.pio.count = 0;
5948 		return 1;
5949 	}
5950 
5951 	vcpu->run->exit_reason = KVM_EXIT_IO;
5952 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5953 	vcpu->run->io.size = size;
5954 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5955 	vcpu->run->io.count = count;
5956 	vcpu->run->io.port = port;
5957 
5958 	return 0;
5959 }
5960 
5961 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
5962 			   unsigned short port, void *val, unsigned int count)
5963 {
5964 	int ret;
5965 
5966 	if (vcpu->arch.pio.count)
5967 		goto data_avail;
5968 
5969 	memset(vcpu->arch.pio_data, 0, size * count);
5970 
5971 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5972 	if (ret) {
5973 data_avail:
5974 		memcpy(val, vcpu->arch.pio_data, size * count);
5975 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5976 		vcpu->arch.pio.count = 0;
5977 		return 1;
5978 	}
5979 
5980 	return 0;
5981 }
5982 
5983 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5984 				    int size, unsigned short port, void *val,
5985 				    unsigned int count)
5986 {
5987 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
5988 
5989 }
5990 
5991 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
5992 			    unsigned short port, const void *val,
5993 			    unsigned int count)
5994 {
5995 	memcpy(vcpu->arch.pio_data, val, size * count);
5996 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5997 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5998 }
5999 
6000 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6001 				     int size, unsigned short port,
6002 				     const void *val, unsigned int count)
6003 {
6004 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6005 }
6006 
6007 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6008 {
6009 	return kvm_x86_ops.get_segment_base(vcpu, seg);
6010 }
6011 
6012 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6013 {
6014 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6015 }
6016 
6017 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6018 {
6019 	if (!need_emulate_wbinvd(vcpu))
6020 		return X86EMUL_CONTINUE;
6021 
6022 	if (kvm_x86_ops.has_wbinvd_exit()) {
6023 		int cpu = get_cpu();
6024 
6025 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6026 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6027 				wbinvd_ipi, NULL, 1);
6028 		put_cpu();
6029 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6030 	} else
6031 		wbinvd();
6032 	return X86EMUL_CONTINUE;
6033 }
6034 
6035 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6036 {
6037 	kvm_emulate_wbinvd_noskip(vcpu);
6038 	return kvm_skip_emulated_instruction(vcpu);
6039 }
6040 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6041 
6042 
6043 
6044 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6045 {
6046 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6047 }
6048 
6049 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6050 			   unsigned long *dest)
6051 {
6052 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6053 }
6054 
6055 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6056 			   unsigned long value)
6057 {
6058 
6059 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6060 }
6061 
6062 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6063 {
6064 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6065 }
6066 
6067 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6068 {
6069 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6070 	unsigned long value;
6071 
6072 	switch (cr) {
6073 	case 0:
6074 		value = kvm_read_cr0(vcpu);
6075 		break;
6076 	case 2:
6077 		value = vcpu->arch.cr2;
6078 		break;
6079 	case 3:
6080 		value = kvm_read_cr3(vcpu);
6081 		break;
6082 	case 4:
6083 		value = kvm_read_cr4(vcpu);
6084 		break;
6085 	case 8:
6086 		value = kvm_get_cr8(vcpu);
6087 		break;
6088 	default:
6089 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6090 		return 0;
6091 	}
6092 
6093 	return value;
6094 }
6095 
6096 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6097 {
6098 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6099 	int res = 0;
6100 
6101 	switch (cr) {
6102 	case 0:
6103 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6104 		break;
6105 	case 2:
6106 		vcpu->arch.cr2 = val;
6107 		break;
6108 	case 3:
6109 		res = kvm_set_cr3(vcpu, val);
6110 		break;
6111 	case 4:
6112 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6113 		break;
6114 	case 8:
6115 		res = kvm_set_cr8(vcpu, val);
6116 		break;
6117 	default:
6118 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6119 		res = -1;
6120 	}
6121 
6122 	return res;
6123 }
6124 
6125 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6126 {
6127 	return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6128 }
6129 
6130 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6131 {
6132 	kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6133 }
6134 
6135 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6136 {
6137 	kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6138 }
6139 
6140 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6141 {
6142 	kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6143 }
6144 
6145 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6146 {
6147 	kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6148 }
6149 
6150 static unsigned long emulator_get_cached_segment_base(
6151 	struct x86_emulate_ctxt *ctxt, int seg)
6152 {
6153 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6154 }
6155 
6156 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6157 				 struct desc_struct *desc, u32 *base3,
6158 				 int seg)
6159 {
6160 	struct kvm_segment var;
6161 
6162 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6163 	*selector = var.selector;
6164 
6165 	if (var.unusable) {
6166 		memset(desc, 0, sizeof(*desc));
6167 		if (base3)
6168 			*base3 = 0;
6169 		return false;
6170 	}
6171 
6172 	if (var.g)
6173 		var.limit >>= 12;
6174 	set_desc_limit(desc, var.limit);
6175 	set_desc_base(desc, (unsigned long)var.base);
6176 #ifdef CONFIG_X86_64
6177 	if (base3)
6178 		*base3 = var.base >> 32;
6179 #endif
6180 	desc->type = var.type;
6181 	desc->s = var.s;
6182 	desc->dpl = var.dpl;
6183 	desc->p = var.present;
6184 	desc->avl = var.avl;
6185 	desc->l = var.l;
6186 	desc->d = var.db;
6187 	desc->g = var.g;
6188 
6189 	return true;
6190 }
6191 
6192 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6193 				 struct desc_struct *desc, u32 base3,
6194 				 int seg)
6195 {
6196 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6197 	struct kvm_segment var;
6198 
6199 	var.selector = selector;
6200 	var.base = get_desc_base(desc);
6201 #ifdef CONFIG_X86_64
6202 	var.base |= ((u64)base3) << 32;
6203 #endif
6204 	var.limit = get_desc_limit(desc);
6205 	if (desc->g)
6206 		var.limit = (var.limit << 12) | 0xfff;
6207 	var.type = desc->type;
6208 	var.dpl = desc->dpl;
6209 	var.db = desc->d;
6210 	var.s = desc->s;
6211 	var.l = desc->l;
6212 	var.g = desc->g;
6213 	var.avl = desc->avl;
6214 	var.present = desc->p;
6215 	var.unusable = !var.present;
6216 	var.padding = 0;
6217 
6218 	kvm_set_segment(vcpu, &var, seg);
6219 	return;
6220 }
6221 
6222 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6223 			    u32 msr_index, u64 *pdata)
6224 {
6225 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6226 }
6227 
6228 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6229 			    u32 msr_index, u64 data)
6230 {
6231 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6232 }
6233 
6234 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6235 {
6236 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6237 
6238 	return vcpu->arch.smbase;
6239 }
6240 
6241 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6242 {
6243 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6244 
6245 	vcpu->arch.smbase = smbase;
6246 }
6247 
6248 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6249 			      u32 pmc)
6250 {
6251 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6252 }
6253 
6254 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6255 			     u32 pmc, u64 *pdata)
6256 {
6257 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6258 }
6259 
6260 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6261 {
6262 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6263 }
6264 
6265 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6266 			      struct x86_instruction_info *info,
6267 			      enum x86_intercept_stage stage)
6268 {
6269 	return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6270 					    &ctxt->exception);
6271 }
6272 
6273 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6274 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6275 			      bool exact_only)
6276 {
6277 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6278 }
6279 
6280 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6281 {
6282 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6283 }
6284 
6285 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6286 {
6287 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6288 }
6289 
6290 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6291 {
6292 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6293 }
6294 
6295 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6296 {
6297 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6298 }
6299 
6300 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6301 {
6302 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6303 }
6304 
6305 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6306 {
6307 	kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6308 }
6309 
6310 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6311 {
6312 	return emul_to_vcpu(ctxt)->arch.hflags;
6313 }
6314 
6315 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6316 {
6317 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6318 }
6319 
6320 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6321 				  const char *smstate)
6322 {
6323 	return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6324 }
6325 
6326 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6327 {
6328 	kvm_smm_changed(emul_to_vcpu(ctxt));
6329 }
6330 
6331 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6332 {
6333 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6334 }
6335 
6336 static const struct x86_emulate_ops emulate_ops = {
6337 	.read_gpr            = emulator_read_gpr,
6338 	.write_gpr           = emulator_write_gpr,
6339 	.read_std            = emulator_read_std,
6340 	.write_std           = emulator_write_std,
6341 	.read_phys           = kvm_read_guest_phys_system,
6342 	.fetch               = kvm_fetch_guest_virt,
6343 	.read_emulated       = emulator_read_emulated,
6344 	.write_emulated      = emulator_write_emulated,
6345 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6346 	.invlpg              = emulator_invlpg,
6347 	.pio_in_emulated     = emulator_pio_in_emulated,
6348 	.pio_out_emulated    = emulator_pio_out_emulated,
6349 	.get_segment         = emulator_get_segment,
6350 	.set_segment         = emulator_set_segment,
6351 	.get_cached_segment_base = emulator_get_cached_segment_base,
6352 	.get_gdt             = emulator_get_gdt,
6353 	.get_idt	     = emulator_get_idt,
6354 	.set_gdt             = emulator_set_gdt,
6355 	.set_idt	     = emulator_set_idt,
6356 	.get_cr              = emulator_get_cr,
6357 	.set_cr              = emulator_set_cr,
6358 	.cpl                 = emulator_get_cpl,
6359 	.get_dr              = emulator_get_dr,
6360 	.set_dr              = emulator_set_dr,
6361 	.get_smbase          = emulator_get_smbase,
6362 	.set_smbase          = emulator_set_smbase,
6363 	.set_msr             = emulator_set_msr,
6364 	.get_msr             = emulator_get_msr,
6365 	.check_pmc	     = emulator_check_pmc,
6366 	.read_pmc            = emulator_read_pmc,
6367 	.halt                = emulator_halt,
6368 	.wbinvd              = emulator_wbinvd,
6369 	.fix_hypercall       = emulator_fix_hypercall,
6370 	.intercept           = emulator_intercept,
6371 	.get_cpuid           = emulator_get_cpuid,
6372 	.guest_has_long_mode = emulator_guest_has_long_mode,
6373 	.guest_has_movbe     = emulator_guest_has_movbe,
6374 	.guest_has_fxsr      = emulator_guest_has_fxsr,
6375 	.set_nmi_mask        = emulator_set_nmi_mask,
6376 	.get_hflags          = emulator_get_hflags,
6377 	.set_hflags          = emulator_set_hflags,
6378 	.pre_leave_smm       = emulator_pre_leave_smm,
6379 	.post_leave_smm      = emulator_post_leave_smm,
6380 	.set_xcr             = emulator_set_xcr,
6381 };
6382 
6383 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6384 {
6385 	u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6386 	/*
6387 	 * an sti; sti; sequence only disable interrupts for the first
6388 	 * instruction. So, if the last instruction, be it emulated or
6389 	 * not, left the system with the INT_STI flag enabled, it
6390 	 * means that the last instruction is an sti. We should not
6391 	 * leave the flag on in this case. The same goes for mov ss
6392 	 */
6393 	if (int_shadow & mask)
6394 		mask = 0;
6395 	if (unlikely(int_shadow || mask)) {
6396 		kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6397 		if (!mask)
6398 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6399 	}
6400 }
6401 
6402 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6403 {
6404 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6405 	if (ctxt->exception.vector == PF_VECTOR)
6406 		return kvm_propagate_fault(vcpu, &ctxt->exception);
6407 
6408 	if (ctxt->exception.error_code_valid)
6409 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6410 				      ctxt->exception.error_code);
6411 	else
6412 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6413 	return false;
6414 }
6415 
6416 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6417 {
6418 	struct x86_emulate_ctxt *ctxt;
6419 
6420 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6421 	if (!ctxt) {
6422 		pr_err("kvm: failed to allocate vcpu's emulator\n");
6423 		return NULL;
6424 	}
6425 
6426 	ctxt->vcpu = vcpu;
6427 	ctxt->ops = &emulate_ops;
6428 	vcpu->arch.emulate_ctxt = ctxt;
6429 
6430 	return ctxt;
6431 }
6432 
6433 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6434 {
6435 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6436 	int cs_db, cs_l;
6437 
6438 	kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6439 
6440 	ctxt->gpa_available = false;
6441 	ctxt->eflags = kvm_get_rflags(vcpu);
6442 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6443 
6444 	ctxt->eip = kvm_rip_read(vcpu);
6445 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6446 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6447 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6448 		     cs_db				? X86EMUL_MODE_PROT32 :
6449 							  X86EMUL_MODE_PROT16;
6450 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6451 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6452 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6453 
6454 	init_decode_cache(ctxt);
6455 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6456 }
6457 
6458 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6459 {
6460 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6461 	int ret;
6462 
6463 	init_emulate_ctxt(vcpu);
6464 
6465 	ctxt->op_bytes = 2;
6466 	ctxt->ad_bytes = 2;
6467 	ctxt->_eip = ctxt->eip + inc_eip;
6468 	ret = emulate_int_real(ctxt, irq);
6469 
6470 	if (ret != X86EMUL_CONTINUE) {
6471 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6472 	} else {
6473 		ctxt->eip = ctxt->_eip;
6474 		kvm_rip_write(vcpu, ctxt->eip);
6475 		kvm_set_rflags(vcpu, ctxt->eflags);
6476 	}
6477 }
6478 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6479 
6480 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6481 {
6482 	++vcpu->stat.insn_emulation_fail;
6483 	trace_kvm_emulate_insn_failed(vcpu);
6484 
6485 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6486 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6487 		return 1;
6488 	}
6489 
6490 	if (emulation_type & EMULTYPE_SKIP) {
6491 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6492 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6493 		vcpu->run->internal.ndata = 0;
6494 		return 0;
6495 	}
6496 
6497 	kvm_queue_exception(vcpu, UD_VECTOR);
6498 
6499 	if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6500 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6501 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6502 		vcpu->run->internal.ndata = 0;
6503 		return 0;
6504 	}
6505 
6506 	return 1;
6507 }
6508 
6509 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6510 				  bool write_fault_to_shadow_pgtable,
6511 				  int emulation_type)
6512 {
6513 	gpa_t gpa = cr2_or_gpa;
6514 	kvm_pfn_t pfn;
6515 
6516 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6517 		return false;
6518 
6519 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6520 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6521 		return false;
6522 
6523 	if (!vcpu->arch.mmu->direct_map) {
6524 		/*
6525 		 * Write permission should be allowed since only
6526 		 * write access need to be emulated.
6527 		 */
6528 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6529 
6530 		/*
6531 		 * If the mapping is invalid in guest, let cpu retry
6532 		 * it to generate fault.
6533 		 */
6534 		if (gpa == UNMAPPED_GVA)
6535 			return true;
6536 	}
6537 
6538 	/*
6539 	 * Do not retry the unhandleable instruction if it faults on the
6540 	 * readonly host memory, otherwise it will goto a infinite loop:
6541 	 * retry instruction -> write #PF -> emulation fail -> retry
6542 	 * instruction -> ...
6543 	 */
6544 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6545 
6546 	/*
6547 	 * If the instruction failed on the error pfn, it can not be fixed,
6548 	 * report the error to userspace.
6549 	 */
6550 	if (is_error_noslot_pfn(pfn))
6551 		return false;
6552 
6553 	kvm_release_pfn_clean(pfn);
6554 
6555 	/* The instructions are well-emulated on direct mmu. */
6556 	if (vcpu->arch.mmu->direct_map) {
6557 		unsigned int indirect_shadow_pages;
6558 
6559 		spin_lock(&vcpu->kvm->mmu_lock);
6560 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6561 		spin_unlock(&vcpu->kvm->mmu_lock);
6562 
6563 		if (indirect_shadow_pages)
6564 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6565 
6566 		return true;
6567 	}
6568 
6569 	/*
6570 	 * if emulation was due to access to shadowed page table
6571 	 * and it failed try to unshadow page and re-enter the
6572 	 * guest to let CPU execute the instruction.
6573 	 */
6574 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6575 
6576 	/*
6577 	 * If the access faults on its page table, it can not
6578 	 * be fixed by unprotecting shadow page and it should
6579 	 * be reported to userspace.
6580 	 */
6581 	return !write_fault_to_shadow_pgtable;
6582 }
6583 
6584 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6585 			      gpa_t cr2_or_gpa,  int emulation_type)
6586 {
6587 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6588 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6589 
6590 	last_retry_eip = vcpu->arch.last_retry_eip;
6591 	last_retry_addr = vcpu->arch.last_retry_addr;
6592 
6593 	/*
6594 	 * If the emulation is caused by #PF and it is non-page_table
6595 	 * writing instruction, it means the VM-EXIT is caused by shadow
6596 	 * page protected, we can zap the shadow page and retry this
6597 	 * instruction directly.
6598 	 *
6599 	 * Note: if the guest uses a non-page-table modifying instruction
6600 	 * on the PDE that points to the instruction, then we will unmap
6601 	 * the instruction and go to an infinite loop. So, we cache the
6602 	 * last retried eip and the last fault address, if we meet the eip
6603 	 * and the address again, we can break out of the potential infinite
6604 	 * loop.
6605 	 */
6606 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6607 
6608 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6609 		return false;
6610 
6611 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6612 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6613 		return false;
6614 
6615 	if (x86_page_table_writing_insn(ctxt))
6616 		return false;
6617 
6618 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6619 		return false;
6620 
6621 	vcpu->arch.last_retry_eip = ctxt->eip;
6622 	vcpu->arch.last_retry_addr = cr2_or_gpa;
6623 
6624 	if (!vcpu->arch.mmu->direct_map)
6625 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6626 
6627 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6628 
6629 	return true;
6630 }
6631 
6632 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6633 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6634 
6635 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6636 {
6637 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6638 		/* This is a good place to trace that we are exiting SMM.  */
6639 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6640 
6641 		/* Process a latched INIT or SMI, if any.  */
6642 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6643 	}
6644 
6645 	kvm_mmu_reset_context(vcpu);
6646 }
6647 
6648 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6649 				unsigned long *db)
6650 {
6651 	u32 dr6 = 0;
6652 	int i;
6653 	u32 enable, rwlen;
6654 
6655 	enable = dr7;
6656 	rwlen = dr7 >> 16;
6657 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6658 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6659 			dr6 |= (1 << i);
6660 	return dr6;
6661 }
6662 
6663 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6664 {
6665 	struct kvm_run *kvm_run = vcpu->run;
6666 
6667 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6668 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6669 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6670 		kvm_run->debug.arch.exception = DB_VECTOR;
6671 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6672 		return 0;
6673 	}
6674 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6675 	return 1;
6676 }
6677 
6678 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6679 {
6680 	unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6681 	int r;
6682 
6683 	r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6684 	if (unlikely(!r))
6685 		return 0;
6686 
6687 	/*
6688 	 * rflags is the old, "raw" value of the flags.  The new value has
6689 	 * not been saved yet.
6690 	 *
6691 	 * This is correct even for TF set by the guest, because "the
6692 	 * processor will not generate this exception after the instruction
6693 	 * that sets the TF flag".
6694 	 */
6695 	if (unlikely(rflags & X86_EFLAGS_TF))
6696 		r = kvm_vcpu_do_singlestep(vcpu);
6697 	return r;
6698 }
6699 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6700 
6701 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6702 {
6703 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6704 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6705 		struct kvm_run *kvm_run = vcpu->run;
6706 		unsigned long eip = kvm_get_linear_rip(vcpu);
6707 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6708 					   vcpu->arch.guest_debug_dr7,
6709 					   vcpu->arch.eff_db);
6710 
6711 		if (dr6 != 0) {
6712 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6713 			kvm_run->debug.arch.pc = eip;
6714 			kvm_run->debug.arch.exception = DB_VECTOR;
6715 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6716 			*r = 0;
6717 			return true;
6718 		}
6719 	}
6720 
6721 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6722 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6723 		unsigned long eip = kvm_get_linear_rip(vcpu);
6724 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6725 					   vcpu->arch.dr7,
6726 					   vcpu->arch.db);
6727 
6728 		if (dr6 != 0) {
6729 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
6730 			*r = 1;
6731 			return true;
6732 		}
6733 	}
6734 
6735 	return false;
6736 }
6737 
6738 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6739 {
6740 	switch (ctxt->opcode_len) {
6741 	case 1:
6742 		switch (ctxt->b) {
6743 		case 0xe4:	/* IN */
6744 		case 0xe5:
6745 		case 0xec:
6746 		case 0xed:
6747 		case 0xe6:	/* OUT */
6748 		case 0xe7:
6749 		case 0xee:
6750 		case 0xef:
6751 		case 0x6c:	/* INS */
6752 		case 0x6d:
6753 		case 0x6e:	/* OUTS */
6754 		case 0x6f:
6755 			return true;
6756 		}
6757 		break;
6758 	case 2:
6759 		switch (ctxt->b) {
6760 		case 0x33:	/* RDPMC */
6761 			return true;
6762 		}
6763 		break;
6764 	}
6765 
6766 	return false;
6767 }
6768 
6769 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6770 			    int emulation_type, void *insn, int insn_len)
6771 {
6772 	int r;
6773 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6774 	bool writeback = true;
6775 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6776 
6777 	vcpu->arch.l1tf_flush_l1d = true;
6778 
6779 	/*
6780 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6781 	 * never reused.
6782 	 */
6783 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6784 	kvm_clear_exception_queue(vcpu);
6785 
6786 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6787 		init_emulate_ctxt(vcpu);
6788 
6789 		/*
6790 		 * We will reenter on the same instruction since
6791 		 * we do not set complete_userspace_io.  This does not
6792 		 * handle watchpoints yet, those would be handled in
6793 		 * the emulate_ops.
6794 		 */
6795 		if (!(emulation_type & EMULTYPE_SKIP) &&
6796 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6797 			return r;
6798 
6799 		ctxt->interruptibility = 0;
6800 		ctxt->have_exception = false;
6801 		ctxt->exception.vector = -1;
6802 		ctxt->perm_ok = false;
6803 
6804 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6805 
6806 		r = x86_decode_insn(ctxt, insn, insn_len);
6807 
6808 		trace_kvm_emulate_insn_start(vcpu);
6809 		++vcpu->stat.insn_emulation;
6810 		if (r != EMULATION_OK)  {
6811 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6812 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6813 				kvm_queue_exception(vcpu, UD_VECTOR);
6814 				return 1;
6815 			}
6816 			if (reexecute_instruction(vcpu, cr2_or_gpa,
6817 						  write_fault_to_spt,
6818 						  emulation_type))
6819 				return 1;
6820 			if (ctxt->have_exception) {
6821 				/*
6822 				 * #UD should result in just EMULATION_FAILED, and trap-like
6823 				 * exception should not be encountered during decode.
6824 				 */
6825 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6826 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6827 				inject_emulated_exception(vcpu);
6828 				return 1;
6829 			}
6830 			return handle_emulation_failure(vcpu, emulation_type);
6831 		}
6832 	}
6833 
6834 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6835 	    !is_vmware_backdoor_opcode(ctxt)) {
6836 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6837 		return 1;
6838 	}
6839 
6840 	/*
6841 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6842 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6843 	 * updating interruptibility state and injecting single-step #DBs.
6844 	 */
6845 	if (emulation_type & EMULTYPE_SKIP) {
6846 		kvm_rip_write(vcpu, ctxt->_eip);
6847 		if (ctxt->eflags & X86_EFLAGS_RF)
6848 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6849 		return 1;
6850 	}
6851 
6852 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6853 		return 1;
6854 
6855 	/* this is needed for vmware backdoor interface to work since it
6856 	   changes registers values  during IO operation */
6857 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6858 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6859 		emulator_invalidate_register_cache(ctxt);
6860 	}
6861 
6862 restart:
6863 	if (emulation_type & EMULTYPE_PF) {
6864 		/* Save the faulting GPA (cr2) in the address field */
6865 		ctxt->exception.address = cr2_or_gpa;
6866 
6867 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
6868 		if (vcpu->arch.mmu->direct_map) {
6869 			ctxt->gpa_available = true;
6870 			ctxt->gpa_val = cr2_or_gpa;
6871 		}
6872 	} else {
6873 		/* Sanitize the address out of an abundance of paranoia. */
6874 		ctxt->exception.address = 0;
6875 	}
6876 
6877 	r = x86_emulate_insn(ctxt);
6878 
6879 	if (r == EMULATION_INTERCEPTED)
6880 		return 1;
6881 
6882 	if (r == EMULATION_FAILED) {
6883 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6884 					emulation_type))
6885 			return 1;
6886 
6887 		return handle_emulation_failure(vcpu, emulation_type);
6888 	}
6889 
6890 	if (ctxt->have_exception) {
6891 		r = 1;
6892 		if (inject_emulated_exception(vcpu))
6893 			return r;
6894 	} else if (vcpu->arch.pio.count) {
6895 		if (!vcpu->arch.pio.in) {
6896 			/* FIXME: return into emulator if single-stepping.  */
6897 			vcpu->arch.pio.count = 0;
6898 		} else {
6899 			writeback = false;
6900 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6901 		}
6902 		r = 0;
6903 	} else if (vcpu->mmio_needed) {
6904 		++vcpu->stat.mmio_exits;
6905 
6906 		if (!vcpu->mmio_is_write)
6907 			writeback = false;
6908 		r = 0;
6909 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6910 	} else if (r == EMULATION_RESTART)
6911 		goto restart;
6912 	else
6913 		r = 1;
6914 
6915 	if (writeback) {
6916 		unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6917 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6918 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6919 		if (!ctxt->have_exception ||
6920 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6921 			kvm_rip_write(vcpu, ctxt->eip);
6922 			if (r && ctxt->tf)
6923 				r = kvm_vcpu_do_singlestep(vcpu);
6924 			if (kvm_x86_ops.update_emulated_instruction)
6925 				kvm_x86_ops.update_emulated_instruction(vcpu);
6926 			__kvm_set_rflags(vcpu, ctxt->eflags);
6927 		}
6928 
6929 		/*
6930 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6931 		 * do nothing, and it will be requested again as soon as
6932 		 * the shadow expires.  But we still need to check here,
6933 		 * because POPF has no interrupt shadow.
6934 		 */
6935 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6936 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6937 	} else
6938 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6939 
6940 	return r;
6941 }
6942 
6943 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6944 {
6945 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6946 }
6947 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6948 
6949 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6950 					void *insn, int insn_len)
6951 {
6952 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6953 }
6954 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6955 
6956 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6957 {
6958 	vcpu->arch.pio.count = 0;
6959 	return 1;
6960 }
6961 
6962 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6963 {
6964 	vcpu->arch.pio.count = 0;
6965 
6966 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6967 		return 1;
6968 
6969 	return kvm_skip_emulated_instruction(vcpu);
6970 }
6971 
6972 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6973 			    unsigned short port)
6974 {
6975 	unsigned long val = kvm_rax_read(vcpu);
6976 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
6977 
6978 	if (ret)
6979 		return ret;
6980 
6981 	/*
6982 	 * Workaround userspace that relies on old KVM behavior of %rip being
6983 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6984 	 */
6985 	if (port == 0x7e &&
6986 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6987 		vcpu->arch.complete_userspace_io =
6988 			complete_fast_pio_out_port_0x7e;
6989 		kvm_skip_emulated_instruction(vcpu);
6990 	} else {
6991 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6992 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6993 	}
6994 	return 0;
6995 }
6996 
6997 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6998 {
6999 	unsigned long val;
7000 
7001 	/* We should only ever be called with arch.pio.count equal to 1 */
7002 	BUG_ON(vcpu->arch.pio.count != 1);
7003 
7004 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7005 		vcpu->arch.pio.count = 0;
7006 		return 1;
7007 	}
7008 
7009 	/* For size less than 4 we merge, else we zero extend */
7010 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7011 
7012 	/*
7013 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7014 	 * the copy and tracing
7015 	 */
7016 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7017 	kvm_rax_write(vcpu, val);
7018 
7019 	return kvm_skip_emulated_instruction(vcpu);
7020 }
7021 
7022 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7023 			   unsigned short port)
7024 {
7025 	unsigned long val;
7026 	int ret;
7027 
7028 	/* For size less than 4 we merge, else we zero extend */
7029 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7030 
7031 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7032 	if (ret) {
7033 		kvm_rax_write(vcpu, val);
7034 		return ret;
7035 	}
7036 
7037 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7038 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7039 
7040 	return 0;
7041 }
7042 
7043 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7044 {
7045 	int ret;
7046 
7047 	if (in)
7048 		ret = kvm_fast_pio_in(vcpu, size, port);
7049 	else
7050 		ret = kvm_fast_pio_out(vcpu, size, port);
7051 	return ret && kvm_skip_emulated_instruction(vcpu);
7052 }
7053 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7054 
7055 static int kvmclock_cpu_down_prep(unsigned int cpu)
7056 {
7057 	__this_cpu_write(cpu_tsc_khz, 0);
7058 	return 0;
7059 }
7060 
7061 static void tsc_khz_changed(void *data)
7062 {
7063 	struct cpufreq_freqs *freq = data;
7064 	unsigned long khz = 0;
7065 
7066 	if (data)
7067 		khz = freq->new;
7068 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7069 		khz = cpufreq_quick_get(raw_smp_processor_id());
7070 	if (!khz)
7071 		khz = tsc_khz;
7072 	__this_cpu_write(cpu_tsc_khz, khz);
7073 }
7074 
7075 #ifdef CONFIG_X86_64
7076 static void kvm_hyperv_tsc_notifier(void)
7077 {
7078 	struct kvm *kvm;
7079 	struct kvm_vcpu *vcpu;
7080 	int cpu;
7081 
7082 	mutex_lock(&kvm_lock);
7083 	list_for_each_entry(kvm, &vm_list, vm_list)
7084 		kvm_make_mclock_inprogress_request(kvm);
7085 
7086 	hyperv_stop_tsc_emulation();
7087 
7088 	/* TSC frequency always matches when on Hyper-V */
7089 	for_each_present_cpu(cpu)
7090 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7091 	kvm_max_guest_tsc_khz = tsc_khz;
7092 
7093 	list_for_each_entry(kvm, &vm_list, vm_list) {
7094 		struct kvm_arch *ka = &kvm->arch;
7095 
7096 		spin_lock(&ka->pvclock_gtod_sync_lock);
7097 
7098 		pvclock_update_vm_gtod_copy(kvm);
7099 
7100 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7101 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7102 
7103 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7104 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7105 
7106 		spin_unlock(&ka->pvclock_gtod_sync_lock);
7107 	}
7108 	mutex_unlock(&kvm_lock);
7109 }
7110 #endif
7111 
7112 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7113 {
7114 	struct kvm *kvm;
7115 	struct kvm_vcpu *vcpu;
7116 	int i, send_ipi = 0;
7117 
7118 	/*
7119 	 * We allow guests to temporarily run on slowing clocks,
7120 	 * provided we notify them after, or to run on accelerating
7121 	 * clocks, provided we notify them before.  Thus time never
7122 	 * goes backwards.
7123 	 *
7124 	 * However, we have a problem.  We can't atomically update
7125 	 * the frequency of a given CPU from this function; it is
7126 	 * merely a notifier, which can be called from any CPU.
7127 	 * Changing the TSC frequency at arbitrary points in time
7128 	 * requires a recomputation of local variables related to
7129 	 * the TSC for each VCPU.  We must flag these local variables
7130 	 * to be updated and be sure the update takes place with the
7131 	 * new frequency before any guests proceed.
7132 	 *
7133 	 * Unfortunately, the combination of hotplug CPU and frequency
7134 	 * change creates an intractable locking scenario; the order
7135 	 * of when these callouts happen is undefined with respect to
7136 	 * CPU hotplug, and they can race with each other.  As such,
7137 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7138 	 * undefined; you can actually have a CPU frequency change take
7139 	 * place in between the computation of X and the setting of the
7140 	 * variable.  To protect against this problem, all updates of
7141 	 * the per_cpu tsc_khz variable are done in an interrupt
7142 	 * protected IPI, and all callers wishing to update the value
7143 	 * must wait for a synchronous IPI to complete (which is trivial
7144 	 * if the caller is on the CPU already).  This establishes the
7145 	 * necessary total order on variable updates.
7146 	 *
7147 	 * Note that because a guest time update may take place
7148 	 * anytime after the setting of the VCPU's request bit, the
7149 	 * correct TSC value must be set before the request.  However,
7150 	 * to ensure the update actually makes it to any guest which
7151 	 * starts running in hardware virtualization between the set
7152 	 * and the acquisition of the spinlock, we must also ping the
7153 	 * CPU after setting the request bit.
7154 	 *
7155 	 */
7156 
7157 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7158 
7159 	mutex_lock(&kvm_lock);
7160 	list_for_each_entry(kvm, &vm_list, vm_list) {
7161 		kvm_for_each_vcpu(i, vcpu, kvm) {
7162 			if (vcpu->cpu != cpu)
7163 				continue;
7164 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7165 			if (vcpu->cpu != raw_smp_processor_id())
7166 				send_ipi = 1;
7167 		}
7168 	}
7169 	mutex_unlock(&kvm_lock);
7170 
7171 	if (freq->old < freq->new && send_ipi) {
7172 		/*
7173 		 * We upscale the frequency.  Must make the guest
7174 		 * doesn't see old kvmclock values while running with
7175 		 * the new frequency, otherwise we risk the guest sees
7176 		 * time go backwards.
7177 		 *
7178 		 * In case we update the frequency for another cpu
7179 		 * (which might be in guest context) send an interrupt
7180 		 * to kick the cpu out of guest context.  Next time
7181 		 * guest context is entered kvmclock will be updated,
7182 		 * so the guest will not see stale values.
7183 		 */
7184 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7185 	}
7186 }
7187 
7188 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7189 				     void *data)
7190 {
7191 	struct cpufreq_freqs *freq = data;
7192 	int cpu;
7193 
7194 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7195 		return 0;
7196 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7197 		return 0;
7198 
7199 	for_each_cpu(cpu, freq->policy->cpus)
7200 		__kvmclock_cpufreq_notifier(freq, cpu);
7201 
7202 	return 0;
7203 }
7204 
7205 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7206 	.notifier_call  = kvmclock_cpufreq_notifier
7207 };
7208 
7209 static int kvmclock_cpu_online(unsigned int cpu)
7210 {
7211 	tsc_khz_changed(NULL);
7212 	return 0;
7213 }
7214 
7215 static void kvm_timer_init(void)
7216 {
7217 	max_tsc_khz = tsc_khz;
7218 
7219 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7220 #ifdef CONFIG_CPU_FREQ
7221 		struct cpufreq_policy *policy;
7222 		int cpu;
7223 
7224 		cpu = get_cpu();
7225 		policy = cpufreq_cpu_get(cpu);
7226 		if (policy) {
7227 			if (policy->cpuinfo.max_freq)
7228 				max_tsc_khz = policy->cpuinfo.max_freq;
7229 			cpufreq_cpu_put(policy);
7230 		}
7231 		put_cpu();
7232 #endif
7233 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7234 					  CPUFREQ_TRANSITION_NOTIFIER);
7235 	}
7236 
7237 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7238 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7239 }
7240 
7241 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7242 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7243 
7244 int kvm_is_in_guest(void)
7245 {
7246 	return __this_cpu_read(current_vcpu) != NULL;
7247 }
7248 
7249 static int kvm_is_user_mode(void)
7250 {
7251 	int user_mode = 3;
7252 
7253 	if (__this_cpu_read(current_vcpu))
7254 		user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7255 
7256 	return user_mode != 0;
7257 }
7258 
7259 static unsigned long kvm_get_guest_ip(void)
7260 {
7261 	unsigned long ip = 0;
7262 
7263 	if (__this_cpu_read(current_vcpu))
7264 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7265 
7266 	return ip;
7267 }
7268 
7269 static void kvm_handle_intel_pt_intr(void)
7270 {
7271 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7272 
7273 	kvm_make_request(KVM_REQ_PMI, vcpu);
7274 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7275 			(unsigned long *)&vcpu->arch.pmu.global_status);
7276 }
7277 
7278 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7279 	.is_in_guest		= kvm_is_in_guest,
7280 	.is_user_mode		= kvm_is_user_mode,
7281 	.get_guest_ip		= kvm_get_guest_ip,
7282 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7283 };
7284 
7285 #ifdef CONFIG_X86_64
7286 static void pvclock_gtod_update_fn(struct work_struct *work)
7287 {
7288 	struct kvm *kvm;
7289 
7290 	struct kvm_vcpu *vcpu;
7291 	int i;
7292 
7293 	mutex_lock(&kvm_lock);
7294 	list_for_each_entry(kvm, &vm_list, vm_list)
7295 		kvm_for_each_vcpu(i, vcpu, kvm)
7296 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7297 	atomic_set(&kvm_guest_has_master_clock, 0);
7298 	mutex_unlock(&kvm_lock);
7299 }
7300 
7301 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7302 
7303 /*
7304  * Notification about pvclock gtod data update.
7305  */
7306 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7307 			       void *priv)
7308 {
7309 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7310 	struct timekeeper *tk = priv;
7311 
7312 	update_pvclock_gtod(tk);
7313 
7314 	/* disable master clock if host does not trust, or does not
7315 	 * use, TSC based clocksource.
7316 	 */
7317 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7318 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7319 		queue_work(system_long_wq, &pvclock_gtod_work);
7320 
7321 	return 0;
7322 }
7323 
7324 static struct notifier_block pvclock_gtod_notifier = {
7325 	.notifier_call = pvclock_gtod_notify,
7326 };
7327 #endif
7328 
7329 int kvm_arch_init(void *opaque)
7330 {
7331 	struct kvm_x86_init_ops *ops = opaque;
7332 	int r;
7333 
7334 	if (kvm_x86_ops.hardware_enable) {
7335 		printk(KERN_ERR "kvm: already loaded the other module\n");
7336 		r = -EEXIST;
7337 		goto out;
7338 	}
7339 
7340 	if (!ops->cpu_has_kvm_support()) {
7341 		pr_err_ratelimited("kvm: no hardware support\n");
7342 		r = -EOPNOTSUPP;
7343 		goto out;
7344 	}
7345 	if (ops->disabled_by_bios()) {
7346 		pr_err_ratelimited("kvm: disabled by bios\n");
7347 		r = -EOPNOTSUPP;
7348 		goto out;
7349 	}
7350 
7351 	/*
7352 	 * KVM explicitly assumes that the guest has an FPU and
7353 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7354 	 * vCPU's FPU state as a fxregs_state struct.
7355 	 */
7356 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7357 		printk(KERN_ERR "kvm: inadequate fpu\n");
7358 		r = -EOPNOTSUPP;
7359 		goto out;
7360 	}
7361 
7362 	r = -ENOMEM;
7363 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7364 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7365 					  NULL);
7366 	if (!x86_fpu_cache) {
7367 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7368 		goto out;
7369 	}
7370 
7371 	x86_emulator_cache = kvm_alloc_emulator_cache();
7372 	if (!x86_emulator_cache) {
7373 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
7374 		goto out_free_x86_fpu_cache;
7375 	}
7376 
7377 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7378 	if (!shared_msrs) {
7379 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7380 		goto out_free_x86_emulator_cache;
7381 	}
7382 
7383 	r = kvm_mmu_module_init();
7384 	if (r)
7385 		goto out_free_percpu;
7386 
7387 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7388 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7389 			PT_PRESENT_MASK, 0, sme_me_mask);
7390 	kvm_timer_init();
7391 
7392 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7393 
7394 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7395 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7396 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7397 	}
7398 
7399 	kvm_lapic_init();
7400 	if (pi_inject_timer == -1)
7401 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7402 #ifdef CONFIG_X86_64
7403 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7404 
7405 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7406 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7407 #endif
7408 
7409 	return 0;
7410 
7411 out_free_percpu:
7412 	free_percpu(shared_msrs);
7413 out_free_x86_emulator_cache:
7414 	kmem_cache_destroy(x86_emulator_cache);
7415 out_free_x86_fpu_cache:
7416 	kmem_cache_destroy(x86_fpu_cache);
7417 out:
7418 	return r;
7419 }
7420 
7421 void kvm_arch_exit(void)
7422 {
7423 #ifdef CONFIG_X86_64
7424 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7425 		clear_hv_tscchange_cb();
7426 #endif
7427 	kvm_lapic_exit();
7428 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7429 
7430 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7431 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7432 					    CPUFREQ_TRANSITION_NOTIFIER);
7433 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7434 #ifdef CONFIG_X86_64
7435 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7436 #endif
7437 	kvm_x86_ops.hardware_enable = NULL;
7438 	kvm_mmu_module_exit();
7439 	free_percpu(shared_msrs);
7440 	kmem_cache_destroy(x86_fpu_cache);
7441 }
7442 
7443 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7444 {
7445 	++vcpu->stat.halt_exits;
7446 	if (lapic_in_kernel(vcpu)) {
7447 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7448 		return 1;
7449 	} else {
7450 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7451 		return 0;
7452 	}
7453 }
7454 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7455 
7456 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7457 {
7458 	int ret = kvm_skip_emulated_instruction(vcpu);
7459 	/*
7460 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7461 	 * KVM_EXIT_DEBUG here.
7462 	 */
7463 	return kvm_vcpu_halt(vcpu) && ret;
7464 }
7465 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7466 
7467 #ifdef CONFIG_X86_64
7468 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7469 			        unsigned long clock_type)
7470 {
7471 	struct kvm_clock_pairing clock_pairing;
7472 	struct timespec64 ts;
7473 	u64 cycle;
7474 	int ret;
7475 
7476 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7477 		return -KVM_EOPNOTSUPP;
7478 
7479 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7480 		return -KVM_EOPNOTSUPP;
7481 
7482 	clock_pairing.sec = ts.tv_sec;
7483 	clock_pairing.nsec = ts.tv_nsec;
7484 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7485 	clock_pairing.flags = 0;
7486 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7487 
7488 	ret = 0;
7489 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7490 			    sizeof(struct kvm_clock_pairing)))
7491 		ret = -KVM_EFAULT;
7492 
7493 	return ret;
7494 }
7495 #endif
7496 
7497 /*
7498  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7499  *
7500  * @apicid - apicid of vcpu to be kicked.
7501  */
7502 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7503 {
7504 	struct kvm_lapic_irq lapic_irq;
7505 
7506 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
7507 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7508 	lapic_irq.level = 0;
7509 	lapic_irq.dest_id = apicid;
7510 	lapic_irq.msi_redir_hint = false;
7511 
7512 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7513 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7514 }
7515 
7516 bool kvm_apicv_activated(struct kvm *kvm)
7517 {
7518 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7519 }
7520 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7521 
7522 void kvm_apicv_init(struct kvm *kvm, bool enable)
7523 {
7524 	if (enable)
7525 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
7526 			  &kvm->arch.apicv_inhibit_reasons);
7527 	else
7528 		set_bit(APICV_INHIBIT_REASON_DISABLE,
7529 			&kvm->arch.apicv_inhibit_reasons);
7530 }
7531 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7532 
7533 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7534 {
7535 	struct kvm_vcpu *target = NULL;
7536 	struct kvm_apic_map *map;
7537 
7538 	rcu_read_lock();
7539 	map = rcu_dereference(kvm->arch.apic_map);
7540 
7541 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7542 		target = map->phys_map[dest_id]->vcpu;
7543 
7544 	rcu_read_unlock();
7545 
7546 	if (target && READ_ONCE(target->ready))
7547 		kvm_vcpu_yield_to(target);
7548 }
7549 
7550 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7551 {
7552 	unsigned long nr, a0, a1, a2, a3, ret;
7553 	int op_64_bit;
7554 
7555 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7556 		return kvm_hv_hypercall(vcpu);
7557 
7558 	nr = kvm_rax_read(vcpu);
7559 	a0 = kvm_rbx_read(vcpu);
7560 	a1 = kvm_rcx_read(vcpu);
7561 	a2 = kvm_rdx_read(vcpu);
7562 	a3 = kvm_rsi_read(vcpu);
7563 
7564 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7565 
7566 	op_64_bit = is_64_bit_mode(vcpu);
7567 	if (!op_64_bit) {
7568 		nr &= 0xFFFFFFFF;
7569 		a0 &= 0xFFFFFFFF;
7570 		a1 &= 0xFFFFFFFF;
7571 		a2 &= 0xFFFFFFFF;
7572 		a3 &= 0xFFFFFFFF;
7573 	}
7574 
7575 	if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7576 		ret = -KVM_EPERM;
7577 		goto out;
7578 	}
7579 
7580 	switch (nr) {
7581 	case KVM_HC_VAPIC_POLL_IRQ:
7582 		ret = 0;
7583 		break;
7584 	case KVM_HC_KICK_CPU:
7585 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7586 		kvm_sched_yield(vcpu->kvm, a1);
7587 		ret = 0;
7588 		break;
7589 #ifdef CONFIG_X86_64
7590 	case KVM_HC_CLOCK_PAIRING:
7591 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7592 		break;
7593 #endif
7594 	case KVM_HC_SEND_IPI:
7595 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7596 		break;
7597 	case KVM_HC_SCHED_YIELD:
7598 		kvm_sched_yield(vcpu->kvm, a0);
7599 		ret = 0;
7600 		break;
7601 	default:
7602 		ret = -KVM_ENOSYS;
7603 		break;
7604 	}
7605 out:
7606 	if (!op_64_bit)
7607 		ret = (u32)ret;
7608 	kvm_rax_write(vcpu, ret);
7609 
7610 	++vcpu->stat.hypercalls;
7611 	return kvm_skip_emulated_instruction(vcpu);
7612 }
7613 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7614 
7615 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7616 {
7617 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7618 	char instruction[3];
7619 	unsigned long rip = kvm_rip_read(vcpu);
7620 
7621 	kvm_x86_ops.patch_hypercall(vcpu, instruction);
7622 
7623 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7624 		&ctxt->exception);
7625 }
7626 
7627 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7628 {
7629 	return vcpu->run->request_interrupt_window &&
7630 		likely(!pic_in_kernel(vcpu->kvm));
7631 }
7632 
7633 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7634 {
7635 	struct kvm_run *kvm_run = vcpu->run;
7636 
7637 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7638 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7639 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7640 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7641 	kvm_run->ready_for_interrupt_injection =
7642 		pic_in_kernel(vcpu->kvm) ||
7643 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7644 }
7645 
7646 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7647 {
7648 	int max_irr, tpr;
7649 
7650 	if (!kvm_x86_ops.update_cr8_intercept)
7651 		return;
7652 
7653 	if (!lapic_in_kernel(vcpu))
7654 		return;
7655 
7656 	if (vcpu->arch.apicv_active)
7657 		return;
7658 
7659 	if (!vcpu->arch.apic->vapic_addr)
7660 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7661 	else
7662 		max_irr = -1;
7663 
7664 	if (max_irr != -1)
7665 		max_irr >>= 4;
7666 
7667 	tpr = kvm_lapic_get_cr8(vcpu);
7668 
7669 	kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7670 }
7671 
7672 static int inject_pending_event(struct kvm_vcpu *vcpu)
7673 {
7674 	int r;
7675 
7676 	/* try to reinject previous events if any */
7677 
7678 	if (vcpu->arch.exception.injected)
7679 		kvm_x86_ops.queue_exception(vcpu);
7680 	/*
7681 	 * Do not inject an NMI or interrupt if there is a pending
7682 	 * exception.  Exceptions and interrupts are recognized at
7683 	 * instruction boundaries, i.e. the start of an instruction.
7684 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7685 	 * NMIs and interrupts, i.e. traps are recognized before an
7686 	 * NMI/interrupt that's pending on the same instruction.
7687 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7688 	 * priority, but are only generated (pended) during instruction
7689 	 * execution, i.e. a pending fault-like exception means the
7690 	 * fault occurred on the *previous* instruction and must be
7691 	 * serviced prior to recognizing any new events in order to
7692 	 * fully complete the previous instruction.
7693 	 */
7694 	else if (!vcpu->arch.exception.pending) {
7695 		if (vcpu->arch.nmi_injected)
7696 			kvm_x86_ops.set_nmi(vcpu);
7697 		else if (vcpu->arch.interrupt.injected)
7698 			kvm_x86_ops.set_irq(vcpu);
7699 	}
7700 
7701 	/*
7702 	 * Call check_nested_events() even if we reinjected a previous event
7703 	 * in order for caller to determine if it should require immediate-exit
7704 	 * from L2 to L1 due to pending L1 events which require exit
7705 	 * from L2 to L1.
7706 	 */
7707 	if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7708 		r = kvm_x86_ops.check_nested_events(vcpu);
7709 		if (r != 0)
7710 			return r;
7711 	}
7712 
7713 	/* try to inject new event if pending */
7714 	if (vcpu->arch.exception.pending) {
7715 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7716 					vcpu->arch.exception.has_error_code,
7717 					vcpu->arch.exception.error_code);
7718 
7719 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7720 		vcpu->arch.exception.pending = false;
7721 		vcpu->arch.exception.injected = true;
7722 
7723 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7724 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7725 					     X86_EFLAGS_RF);
7726 
7727 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7728 			/*
7729 			 * This code assumes that nSVM doesn't use
7730 			 * check_nested_events(). If it does, the
7731 			 * DR6/DR7 changes should happen before L1
7732 			 * gets a #VMEXIT for an intercepted #DB in
7733 			 * L2.  (Under VMX, on the other hand, the
7734 			 * DR6/DR7 changes should not happen in the
7735 			 * event of a VM-exit to L1 for an intercepted
7736 			 * #DB in L2.)
7737 			 */
7738 			kvm_deliver_exception_payload(vcpu);
7739 			if (vcpu->arch.dr7 & DR7_GD) {
7740 				vcpu->arch.dr7 &= ~DR7_GD;
7741 				kvm_update_dr7(vcpu);
7742 			}
7743 		}
7744 
7745 		kvm_x86_ops.queue_exception(vcpu);
7746 	}
7747 
7748 	/* Don't consider new event if we re-injected an event */
7749 	if (kvm_event_needs_reinjection(vcpu))
7750 		return 0;
7751 
7752 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7753 	    kvm_x86_ops.smi_allowed(vcpu)) {
7754 		vcpu->arch.smi_pending = false;
7755 		++vcpu->arch.smi_count;
7756 		enter_smm(vcpu);
7757 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops.nmi_allowed(vcpu)) {
7758 		--vcpu->arch.nmi_pending;
7759 		vcpu->arch.nmi_injected = true;
7760 		kvm_x86_ops.set_nmi(vcpu);
7761 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7762 		/*
7763 		 * Because interrupts can be injected asynchronously, we are
7764 		 * calling check_nested_events again here to avoid a race condition.
7765 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7766 		 * proposal and current concerns.  Perhaps we should be setting
7767 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7768 		 */
7769 		if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7770 			r = kvm_x86_ops.check_nested_events(vcpu);
7771 			if (r != 0)
7772 				return r;
7773 		}
7774 		if (kvm_x86_ops.interrupt_allowed(vcpu)) {
7775 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7776 					    false);
7777 			kvm_x86_ops.set_irq(vcpu);
7778 		}
7779 	}
7780 
7781 	return 0;
7782 }
7783 
7784 static void process_nmi(struct kvm_vcpu *vcpu)
7785 {
7786 	unsigned limit = 2;
7787 
7788 	/*
7789 	 * x86 is limited to one NMI running, and one NMI pending after it.
7790 	 * If an NMI is already in progress, limit further NMIs to just one.
7791 	 * Otherwise, allow two (and we'll inject the first one immediately).
7792 	 */
7793 	if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7794 		limit = 1;
7795 
7796 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7797 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7798 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7799 }
7800 
7801 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7802 {
7803 	u32 flags = 0;
7804 	flags |= seg->g       << 23;
7805 	flags |= seg->db      << 22;
7806 	flags |= seg->l       << 21;
7807 	flags |= seg->avl     << 20;
7808 	flags |= seg->present << 15;
7809 	flags |= seg->dpl     << 13;
7810 	flags |= seg->s       << 12;
7811 	flags |= seg->type    << 8;
7812 	return flags;
7813 }
7814 
7815 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7816 {
7817 	struct kvm_segment seg;
7818 	int offset;
7819 
7820 	kvm_get_segment(vcpu, &seg, n);
7821 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7822 
7823 	if (n < 3)
7824 		offset = 0x7f84 + n * 12;
7825 	else
7826 		offset = 0x7f2c + (n - 3) * 12;
7827 
7828 	put_smstate(u32, buf, offset + 8, seg.base);
7829 	put_smstate(u32, buf, offset + 4, seg.limit);
7830 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7831 }
7832 
7833 #ifdef CONFIG_X86_64
7834 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7835 {
7836 	struct kvm_segment seg;
7837 	int offset;
7838 	u16 flags;
7839 
7840 	kvm_get_segment(vcpu, &seg, n);
7841 	offset = 0x7e00 + n * 16;
7842 
7843 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7844 	put_smstate(u16, buf, offset, seg.selector);
7845 	put_smstate(u16, buf, offset + 2, flags);
7846 	put_smstate(u32, buf, offset + 4, seg.limit);
7847 	put_smstate(u64, buf, offset + 8, seg.base);
7848 }
7849 #endif
7850 
7851 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7852 {
7853 	struct desc_ptr dt;
7854 	struct kvm_segment seg;
7855 	unsigned long val;
7856 	int i;
7857 
7858 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7859 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7860 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7861 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7862 
7863 	for (i = 0; i < 8; i++)
7864 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7865 
7866 	kvm_get_dr(vcpu, 6, &val);
7867 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7868 	kvm_get_dr(vcpu, 7, &val);
7869 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7870 
7871 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7872 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7873 	put_smstate(u32, buf, 0x7f64, seg.base);
7874 	put_smstate(u32, buf, 0x7f60, seg.limit);
7875 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7876 
7877 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7878 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7879 	put_smstate(u32, buf, 0x7f80, seg.base);
7880 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7881 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7882 
7883 	kvm_x86_ops.get_gdt(vcpu, &dt);
7884 	put_smstate(u32, buf, 0x7f74, dt.address);
7885 	put_smstate(u32, buf, 0x7f70, dt.size);
7886 
7887 	kvm_x86_ops.get_idt(vcpu, &dt);
7888 	put_smstate(u32, buf, 0x7f58, dt.address);
7889 	put_smstate(u32, buf, 0x7f54, dt.size);
7890 
7891 	for (i = 0; i < 6; i++)
7892 		enter_smm_save_seg_32(vcpu, buf, i);
7893 
7894 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7895 
7896 	/* revision id */
7897 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7898 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7899 }
7900 
7901 #ifdef CONFIG_X86_64
7902 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7903 {
7904 	struct desc_ptr dt;
7905 	struct kvm_segment seg;
7906 	unsigned long val;
7907 	int i;
7908 
7909 	for (i = 0; i < 16; i++)
7910 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7911 
7912 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7913 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7914 
7915 	kvm_get_dr(vcpu, 6, &val);
7916 	put_smstate(u64, buf, 0x7f68, val);
7917 	kvm_get_dr(vcpu, 7, &val);
7918 	put_smstate(u64, buf, 0x7f60, val);
7919 
7920 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7921 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7922 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7923 
7924 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7925 
7926 	/* revision id */
7927 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7928 
7929 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7930 
7931 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7932 	put_smstate(u16, buf, 0x7e90, seg.selector);
7933 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7934 	put_smstate(u32, buf, 0x7e94, seg.limit);
7935 	put_smstate(u64, buf, 0x7e98, seg.base);
7936 
7937 	kvm_x86_ops.get_idt(vcpu, &dt);
7938 	put_smstate(u32, buf, 0x7e84, dt.size);
7939 	put_smstate(u64, buf, 0x7e88, dt.address);
7940 
7941 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7942 	put_smstate(u16, buf, 0x7e70, seg.selector);
7943 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7944 	put_smstate(u32, buf, 0x7e74, seg.limit);
7945 	put_smstate(u64, buf, 0x7e78, seg.base);
7946 
7947 	kvm_x86_ops.get_gdt(vcpu, &dt);
7948 	put_smstate(u32, buf, 0x7e64, dt.size);
7949 	put_smstate(u64, buf, 0x7e68, dt.address);
7950 
7951 	for (i = 0; i < 6; i++)
7952 		enter_smm_save_seg_64(vcpu, buf, i);
7953 }
7954 #endif
7955 
7956 static void enter_smm(struct kvm_vcpu *vcpu)
7957 {
7958 	struct kvm_segment cs, ds;
7959 	struct desc_ptr dt;
7960 	char buf[512];
7961 	u32 cr0;
7962 
7963 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7964 	memset(buf, 0, 512);
7965 #ifdef CONFIG_X86_64
7966 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7967 		enter_smm_save_state_64(vcpu, buf);
7968 	else
7969 #endif
7970 		enter_smm_save_state_32(vcpu, buf);
7971 
7972 	/*
7973 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7974 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7975 	 * the SMM state-save area.
7976 	 */
7977 	kvm_x86_ops.pre_enter_smm(vcpu, buf);
7978 
7979 	vcpu->arch.hflags |= HF_SMM_MASK;
7980 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7981 
7982 	if (kvm_x86_ops.get_nmi_mask(vcpu))
7983 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7984 	else
7985 		kvm_x86_ops.set_nmi_mask(vcpu, true);
7986 
7987 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7988 	kvm_rip_write(vcpu, 0x8000);
7989 
7990 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7991 	kvm_x86_ops.set_cr0(vcpu, cr0);
7992 	vcpu->arch.cr0 = cr0;
7993 
7994 	kvm_x86_ops.set_cr4(vcpu, 0);
7995 
7996 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7997 	dt.address = dt.size = 0;
7998 	kvm_x86_ops.set_idt(vcpu, &dt);
7999 
8000 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8001 
8002 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8003 	cs.base = vcpu->arch.smbase;
8004 
8005 	ds.selector = 0;
8006 	ds.base = 0;
8007 
8008 	cs.limit    = ds.limit = 0xffffffff;
8009 	cs.type     = ds.type = 0x3;
8010 	cs.dpl      = ds.dpl = 0;
8011 	cs.db       = ds.db = 0;
8012 	cs.s        = ds.s = 1;
8013 	cs.l        = ds.l = 0;
8014 	cs.g        = ds.g = 1;
8015 	cs.avl      = ds.avl = 0;
8016 	cs.present  = ds.present = 1;
8017 	cs.unusable = ds.unusable = 0;
8018 	cs.padding  = ds.padding = 0;
8019 
8020 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8021 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8022 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8023 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8024 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8025 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8026 
8027 #ifdef CONFIG_X86_64
8028 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8029 		kvm_x86_ops.set_efer(vcpu, 0);
8030 #endif
8031 
8032 	kvm_update_cpuid(vcpu);
8033 	kvm_mmu_reset_context(vcpu);
8034 }
8035 
8036 static void process_smi(struct kvm_vcpu *vcpu)
8037 {
8038 	vcpu->arch.smi_pending = true;
8039 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8040 }
8041 
8042 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8043 				       unsigned long *vcpu_bitmap)
8044 {
8045 	cpumask_var_t cpus;
8046 
8047 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8048 
8049 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8050 				    NULL, vcpu_bitmap, cpus);
8051 
8052 	free_cpumask_var(cpus);
8053 }
8054 
8055 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8056 {
8057 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8058 }
8059 
8060 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8061 {
8062 	if (!lapic_in_kernel(vcpu))
8063 		return;
8064 
8065 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8066 	kvm_apic_update_apicv(vcpu);
8067 	kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8068 }
8069 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8070 
8071 /*
8072  * NOTE: Do not hold any lock prior to calling this.
8073  *
8074  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8075  * locked, because it calls __x86_set_memory_region() which does
8076  * synchronize_srcu(&kvm->srcu).
8077  */
8078 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8079 {
8080 	struct kvm_vcpu *except;
8081 	unsigned long old, new, expected;
8082 
8083 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8084 	    !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8085 		return;
8086 
8087 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8088 	do {
8089 		expected = new = old;
8090 		if (activate)
8091 			__clear_bit(bit, &new);
8092 		else
8093 			__set_bit(bit, &new);
8094 		if (new == old)
8095 			break;
8096 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8097 	} while (old != expected);
8098 
8099 	if (!!old == !!new)
8100 		return;
8101 
8102 	trace_kvm_apicv_update_request(activate, bit);
8103 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8104 		kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8105 
8106 	/*
8107 	 * Sending request to update APICV for all other vcpus,
8108 	 * while update the calling vcpu immediately instead of
8109 	 * waiting for another #VMEXIT to handle the request.
8110 	 */
8111 	except = kvm_get_running_vcpu();
8112 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8113 					 except);
8114 	if (except)
8115 		kvm_vcpu_update_apicv(except);
8116 }
8117 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8118 
8119 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8120 {
8121 	if (!kvm_apic_present(vcpu))
8122 		return;
8123 
8124 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8125 
8126 	if (irqchip_split(vcpu->kvm))
8127 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8128 	else {
8129 		if (vcpu->arch.apicv_active)
8130 			kvm_x86_ops.sync_pir_to_irr(vcpu);
8131 		if (ioapic_in_kernel(vcpu->kvm))
8132 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8133 	}
8134 
8135 	if (is_guest_mode(vcpu))
8136 		vcpu->arch.load_eoi_exitmap_pending = true;
8137 	else
8138 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8139 }
8140 
8141 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8142 {
8143 	u64 eoi_exit_bitmap[4];
8144 
8145 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8146 		return;
8147 
8148 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8149 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
8150 	kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8151 }
8152 
8153 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8154 		unsigned long start, unsigned long end,
8155 		bool blockable)
8156 {
8157 	unsigned long apic_address;
8158 
8159 	/*
8160 	 * The physical address of apic access page is stored in the VMCS.
8161 	 * Update it when it becomes invalid.
8162 	 */
8163 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8164 	if (start <= apic_address && apic_address < end)
8165 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8166 
8167 	return 0;
8168 }
8169 
8170 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8171 {
8172 	struct page *page = NULL;
8173 
8174 	if (!lapic_in_kernel(vcpu))
8175 		return;
8176 
8177 	if (!kvm_x86_ops.set_apic_access_page_addr)
8178 		return;
8179 
8180 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8181 	if (is_error_page(page))
8182 		return;
8183 	kvm_x86_ops.set_apic_access_page_addr(vcpu, page_to_phys(page));
8184 
8185 	/*
8186 	 * Do not pin apic access page in memory, the MMU notifier
8187 	 * will call us again if it is migrated or swapped out.
8188 	 */
8189 	put_page(page);
8190 }
8191 
8192 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8193 {
8194 	smp_send_reschedule(vcpu->cpu);
8195 }
8196 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8197 
8198 /*
8199  * Returns 1 to let vcpu_run() continue the guest execution loop without
8200  * exiting to the userspace.  Otherwise, the value will be returned to the
8201  * userspace.
8202  */
8203 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8204 {
8205 	int r;
8206 	bool req_int_win =
8207 		dm_request_for_irq_injection(vcpu) &&
8208 		kvm_cpu_accept_dm_intr(vcpu);
8209 	enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8210 
8211 	bool req_immediate_exit = false;
8212 
8213 	if (kvm_request_pending(vcpu)) {
8214 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8215 			if (unlikely(!kvm_x86_ops.get_vmcs12_pages(vcpu))) {
8216 				r = 0;
8217 				goto out;
8218 			}
8219 		}
8220 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8221 			kvm_mmu_unload(vcpu);
8222 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8223 			__kvm_migrate_timers(vcpu);
8224 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8225 			kvm_gen_update_masterclock(vcpu->kvm);
8226 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8227 			kvm_gen_kvmclock_update(vcpu);
8228 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8229 			r = kvm_guest_time_update(vcpu);
8230 			if (unlikely(r))
8231 				goto out;
8232 		}
8233 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8234 			kvm_mmu_sync_roots(vcpu);
8235 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8236 			kvm_mmu_load_pgd(vcpu);
8237 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8238 			kvm_vcpu_flush_tlb(vcpu, true);
8239 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8240 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8241 			r = 0;
8242 			goto out;
8243 		}
8244 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8245 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8246 			vcpu->mmio_needed = 0;
8247 			r = 0;
8248 			goto out;
8249 		}
8250 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8251 			/* Page is swapped out. Do synthetic halt */
8252 			vcpu->arch.apf.halted = true;
8253 			r = 1;
8254 			goto out;
8255 		}
8256 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8257 			record_steal_time(vcpu);
8258 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8259 			process_smi(vcpu);
8260 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8261 			process_nmi(vcpu);
8262 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8263 			kvm_pmu_handle_event(vcpu);
8264 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8265 			kvm_pmu_deliver_pmi(vcpu);
8266 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8267 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8268 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8269 				     vcpu->arch.ioapic_handled_vectors)) {
8270 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8271 				vcpu->run->eoi.vector =
8272 						vcpu->arch.pending_ioapic_eoi;
8273 				r = 0;
8274 				goto out;
8275 			}
8276 		}
8277 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8278 			vcpu_scan_ioapic(vcpu);
8279 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8280 			vcpu_load_eoi_exitmap(vcpu);
8281 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8282 			kvm_vcpu_reload_apic_access_page(vcpu);
8283 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8284 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8285 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8286 			r = 0;
8287 			goto out;
8288 		}
8289 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8290 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8291 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8292 			r = 0;
8293 			goto out;
8294 		}
8295 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8296 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8297 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8298 			r = 0;
8299 			goto out;
8300 		}
8301 
8302 		/*
8303 		 * KVM_REQ_HV_STIMER has to be processed after
8304 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8305 		 * depend on the guest clock being up-to-date
8306 		 */
8307 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8308 			kvm_hv_process_stimers(vcpu);
8309 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8310 			kvm_vcpu_update_apicv(vcpu);
8311 	}
8312 
8313 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8314 		++vcpu->stat.req_event;
8315 		kvm_apic_accept_events(vcpu);
8316 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8317 			r = 1;
8318 			goto out;
8319 		}
8320 
8321 		if (inject_pending_event(vcpu) != 0)
8322 			req_immediate_exit = true;
8323 		else {
8324 			/* Enable SMI/NMI/IRQ window open exits if needed.
8325 			 *
8326 			 * SMIs have three cases:
8327 			 * 1) They can be nested, and then there is nothing to
8328 			 *    do here because RSM will cause a vmexit anyway.
8329 			 * 2) There is an ISA-specific reason why SMI cannot be
8330 			 *    injected, and the moment when this changes can be
8331 			 *    intercepted.
8332 			 * 3) Or the SMI can be pending because
8333 			 *    inject_pending_event has completed the injection
8334 			 *    of an IRQ or NMI from the previous vmexit, and
8335 			 *    then we request an immediate exit to inject the
8336 			 *    SMI.
8337 			 */
8338 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
8339 				if (!kvm_x86_ops.enable_smi_window(vcpu))
8340 					req_immediate_exit = true;
8341 			if (vcpu->arch.nmi_pending)
8342 				kvm_x86_ops.enable_nmi_window(vcpu);
8343 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8344 				kvm_x86_ops.enable_irq_window(vcpu);
8345 			WARN_ON(vcpu->arch.exception.pending);
8346 		}
8347 
8348 		if (kvm_lapic_enabled(vcpu)) {
8349 			update_cr8_intercept(vcpu);
8350 			kvm_lapic_sync_to_vapic(vcpu);
8351 		}
8352 	}
8353 
8354 	r = kvm_mmu_reload(vcpu);
8355 	if (unlikely(r)) {
8356 		goto cancel_injection;
8357 	}
8358 
8359 	preempt_disable();
8360 
8361 	kvm_x86_ops.prepare_guest_switch(vcpu);
8362 
8363 	/*
8364 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8365 	 * IPI are then delayed after guest entry, which ensures that they
8366 	 * result in virtual interrupt delivery.
8367 	 */
8368 	local_irq_disable();
8369 	vcpu->mode = IN_GUEST_MODE;
8370 
8371 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8372 
8373 	/*
8374 	 * 1) We should set ->mode before checking ->requests.  Please see
8375 	 * the comment in kvm_vcpu_exiting_guest_mode().
8376 	 *
8377 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8378 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8379 	 * (see vmx_deliver_posted_interrupt).
8380 	 *
8381 	 * 3) This also orders the write to mode from any reads to the page
8382 	 * tables done while the VCPU is running.  Please see the comment
8383 	 * in kvm_flush_remote_tlbs.
8384 	 */
8385 	smp_mb__after_srcu_read_unlock();
8386 
8387 	/*
8388 	 * This handles the case where a posted interrupt was
8389 	 * notified with kvm_vcpu_kick.
8390 	 */
8391 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8392 		kvm_x86_ops.sync_pir_to_irr(vcpu);
8393 
8394 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8395 	    || need_resched() || signal_pending(current)) {
8396 		vcpu->mode = OUTSIDE_GUEST_MODE;
8397 		smp_wmb();
8398 		local_irq_enable();
8399 		preempt_enable();
8400 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8401 		r = 1;
8402 		goto cancel_injection;
8403 	}
8404 
8405 	if (req_immediate_exit) {
8406 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8407 		kvm_x86_ops.request_immediate_exit(vcpu);
8408 	}
8409 
8410 	trace_kvm_entry(vcpu->vcpu_id);
8411 	guest_enter_irqoff();
8412 
8413 	fpregs_assert_state_consistent();
8414 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8415 		switch_fpu_return();
8416 
8417 	if (unlikely(vcpu->arch.switch_db_regs)) {
8418 		set_debugreg(0, 7);
8419 		set_debugreg(vcpu->arch.eff_db[0], 0);
8420 		set_debugreg(vcpu->arch.eff_db[1], 1);
8421 		set_debugreg(vcpu->arch.eff_db[2], 2);
8422 		set_debugreg(vcpu->arch.eff_db[3], 3);
8423 		set_debugreg(vcpu->arch.dr6, 6);
8424 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8425 	}
8426 
8427 	kvm_x86_ops.run(vcpu);
8428 
8429 	/*
8430 	 * Do this here before restoring debug registers on the host.  And
8431 	 * since we do this before handling the vmexit, a DR access vmexit
8432 	 * can (a) read the correct value of the debug registers, (b) set
8433 	 * KVM_DEBUGREG_WONT_EXIT again.
8434 	 */
8435 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8436 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8437 		kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8438 		kvm_update_dr0123(vcpu);
8439 		kvm_update_dr7(vcpu);
8440 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8441 	}
8442 
8443 	/*
8444 	 * If the guest has used debug registers, at least dr7
8445 	 * will be disabled while returning to the host.
8446 	 * If we don't have active breakpoints in the host, we don't
8447 	 * care about the messed up debug address registers. But if
8448 	 * we have some of them active, restore the old state.
8449 	 */
8450 	if (hw_breakpoint_active())
8451 		hw_breakpoint_restore();
8452 
8453 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8454 
8455 	vcpu->mode = OUTSIDE_GUEST_MODE;
8456 	smp_wmb();
8457 
8458 	kvm_x86_ops.handle_exit_irqoff(vcpu, &exit_fastpath);
8459 
8460 	/*
8461 	 * Consume any pending interrupts, including the possible source of
8462 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8463 	 * An instruction is required after local_irq_enable() to fully unblock
8464 	 * interrupts on processors that implement an interrupt shadow, the
8465 	 * stat.exits increment will do nicely.
8466 	 */
8467 	kvm_before_interrupt(vcpu);
8468 	local_irq_enable();
8469 	++vcpu->stat.exits;
8470 	local_irq_disable();
8471 	kvm_after_interrupt(vcpu);
8472 
8473 	guest_exit_irqoff();
8474 	if (lapic_in_kernel(vcpu)) {
8475 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8476 		if (delta != S64_MIN) {
8477 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8478 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8479 		}
8480 	}
8481 
8482 	local_irq_enable();
8483 	preempt_enable();
8484 
8485 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8486 
8487 	/*
8488 	 * Profile KVM exit RIPs:
8489 	 */
8490 	if (unlikely(prof_on == KVM_PROFILING)) {
8491 		unsigned long rip = kvm_rip_read(vcpu);
8492 		profile_hit(KVM_PROFILING, (void *)rip);
8493 	}
8494 
8495 	if (unlikely(vcpu->arch.tsc_always_catchup))
8496 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8497 
8498 	if (vcpu->arch.apic_attention)
8499 		kvm_lapic_sync_from_vapic(vcpu);
8500 
8501 	r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8502 	return r;
8503 
8504 cancel_injection:
8505 	kvm_x86_ops.cancel_injection(vcpu);
8506 	if (unlikely(vcpu->arch.apic_attention))
8507 		kvm_lapic_sync_from_vapic(vcpu);
8508 out:
8509 	return r;
8510 }
8511 
8512 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8513 {
8514 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8515 	    (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8516 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8517 		kvm_vcpu_block(vcpu);
8518 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8519 
8520 		if (kvm_x86_ops.post_block)
8521 			kvm_x86_ops.post_block(vcpu);
8522 
8523 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8524 			return 1;
8525 	}
8526 
8527 	kvm_apic_accept_events(vcpu);
8528 	switch(vcpu->arch.mp_state) {
8529 	case KVM_MP_STATE_HALTED:
8530 		vcpu->arch.pv.pv_unhalted = false;
8531 		vcpu->arch.mp_state =
8532 			KVM_MP_STATE_RUNNABLE;
8533 		/* fall through */
8534 	case KVM_MP_STATE_RUNNABLE:
8535 		vcpu->arch.apf.halted = false;
8536 		break;
8537 	case KVM_MP_STATE_INIT_RECEIVED:
8538 		break;
8539 	default:
8540 		return -EINTR;
8541 	}
8542 	return 1;
8543 }
8544 
8545 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8546 {
8547 	if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events)
8548 		kvm_x86_ops.check_nested_events(vcpu);
8549 
8550 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8551 		!vcpu->arch.apf.halted);
8552 }
8553 
8554 static int vcpu_run(struct kvm_vcpu *vcpu)
8555 {
8556 	int r;
8557 	struct kvm *kvm = vcpu->kvm;
8558 
8559 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8560 	vcpu->arch.l1tf_flush_l1d = true;
8561 
8562 	for (;;) {
8563 		if (kvm_vcpu_running(vcpu)) {
8564 			r = vcpu_enter_guest(vcpu);
8565 		} else {
8566 			r = vcpu_block(kvm, vcpu);
8567 		}
8568 
8569 		if (r <= 0)
8570 			break;
8571 
8572 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8573 		if (kvm_cpu_has_pending_timer(vcpu))
8574 			kvm_inject_pending_timer_irqs(vcpu);
8575 
8576 		if (dm_request_for_irq_injection(vcpu) &&
8577 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8578 			r = 0;
8579 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8580 			++vcpu->stat.request_irq_exits;
8581 			break;
8582 		}
8583 
8584 		kvm_check_async_pf_completion(vcpu);
8585 
8586 		if (signal_pending(current)) {
8587 			r = -EINTR;
8588 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8589 			++vcpu->stat.signal_exits;
8590 			break;
8591 		}
8592 		if (need_resched()) {
8593 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8594 			cond_resched();
8595 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8596 		}
8597 	}
8598 
8599 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8600 
8601 	return r;
8602 }
8603 
8604 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8605 {
8606 	int r;
8607 
8608 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8609 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8610 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8611 	return r;
8612 }
8613 
8614 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8615 {
8616 	BUG_ON(!vcpu->arch.pio.count);
8617 
8618 	return complete_emulated_io(vcpu);
8619 }
8620 
8621 /*
8622  * Implements the following, as a state machine:
8623  *
8624  * read:
8625  *   for each fragment
8626  *     for each mmio piece in the fragment
8627  *       write gpa, len
8628  *       exit
8629  *       copy data
8630  *   execute insn
8631  *
8632  * write:
8633  *   for each fragment
8634  *     for each mmio piece in the fragment
8635  *       write gpa, len
8636  *       copy data
8637  *       exit
8638  */
8639 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8640 {
8641 	struct kvm_run *run = vcpu->run;
8642 	struct kvm_mmio_fragment *frag;
8643 	unsigned len;
8644 
8645 	BUG_ON(!vcpu->mmio_needed);
8646 
8647 	/* Complete previous fragment */
8648 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8649 	len = min(8u, frag->len);
8650 	if (!vcpu->mmio_is_write)
8651 		memcpy(frag->data, run->mmio.data, len);
8652 
8653 	if (frag->len <= 8) {
8654 		/* Switch to the next fragment. */
8655 		frag++;
8656 		vcpu->mmio_cur_fragment++;
8657 	} else {
8658 		/* Go forward to the next mmio piece. */
8659 		frag->data += len;
8660 		frag->gpa += len;
8661 		frag->len -= len;
8662 	}
8663 
8664 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8665 		vcpu->mmio_needed = 0;
8666 
8667 		/* FIXME: return into emulator if single-stepping.  */
8668 		if (vcpu->mmio_is_write)
8669 			return 1;
8670 		vcpu->mmio_read_completed = 1;
8671 		return complete_emulated_io(vcpu);
8672 	}
8673 
8674 	run->exit_reason = KVM_EXIT_MMIO;
8675 	run->mmio.phys_addr = frag->gpa;
8676 	if (vcpu->mmio_is_write)
8677 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8678 	run->mmio.len = min(8u, frag->len);
8679 	run->mmio.is_write = vcpu->mmio_is_write;
8680 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8681 	return 0;
8682 }
8683 
8684 static void kvm_save_current_fpu(struct fpu *fpu)
8685 {
8686 	/*
8687 	 * If the target FPU state is not resident in the CPU registers, just
8688 	 * memcpy() from current, else save CPU state directly to the target.
8689 	 */
8690 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8691 		memcpy(&fpu->state, &current->thread.fpu.state,
8692 		       fpu_kernel_xstate_size);
8693 	else
8694 		copy_fpregs_to_fpstate(fpu);
8695 }
8696 
8697 /* Swap (qemu) user FPU context for the guest FPU context. */
8698 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8699 {
8700 	fpregs_lock();
8701 
8702 	kvm_save_current_fpu(vcpu->arch.user_fpu);
8703 
8704 	/* PKRU is separately restored in kvm_x86_ops.run.  */
8705 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8706 				~XFEATURE_MASK_PKRU);
8707 
8708 	fpregs_mark_activate();
8709 	fpregs_unlock();
8710 
8711 	trace_kvm_fpu(1);
8712 }
8713 
8714 /* When vcpu_run ends, restore user space FPU context. */
8715 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8716 {
8717 	fpregs_lock();
8718 
8719 	kvm_save_current_fpu(vcpu->arch.guest_fpu);
8720 
8721 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8722 
8723 	fpregs_mark_activate();
8724 	fpregs_unlock();
8725 
8726 	++vcpu->stat.fpu_reload;
8727 	trace_kvm_fpu(0);
8728 }
8729 
8730 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8731 {
8732 	int r;
8733 
8734 	vcpu_load(vcpu);
8735 	kvm_sigset_activate(vcpu);
8736 	kvm_load_guest_fpu(vcpu);
8737 
8738 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8739 		if (kvm_run->immediate_exit) {
8740 			r = -EINTR;
8741 			goto out;
8742 		}
8743 		kvm_vcpu_block(vcpu);
8744 		kvm_apic_accept_events(vcpu);
8745 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8746 		r = -EAGAIN;
8747 		if (signal_pending(current)) {
8748 			r = -EINTR;
8749 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8750 			++vcpu->stat.signal_exits;
8751 		}
8752 		goto out;
8753 	}
8754 
8755 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8756 		r = -EINVAL;
8757 		goto out;
8758 	}
8759 
8760 	if (vcpu->run->kvm_dirty_regs) {
8761 		r = sync_regs(vcpu);
8762 		if (r != 0)
8763 			goto out;
8764 	}
8765 
8766 	/* re-sync apic's tpr */
8767 	if (!lapic_in_kernel(vcpu)) {
8768 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8769 			r = -EINVAL;
8770 			goto out;
8771 		}
8772 	}
8773 
8774 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8775 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8776 		vcpu->arch.complete_userspace_io = NULL;
8777 		r = cui(vcpu);
8778 		if (r <= 0)
8779 			goto out;
8780 	} else
8781 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8782 
8783 	if (kvm_run->immediate_exit)
8784 		r = -EINTR;
8785 	else
8786 		r = vcpu_run(vcpu);
8787 
8788 out:
8789 	kvm_put_guest_fpu(vcpu);
8790 	if (vcpu->run->kvm_valid_regs)
8791 		store_regs(vcpu);
8792 	post_kvm_run_save(vcpu);
8793 	kvm_sigset_deactivate(vcpu);
8794 
8795 	vcpu_put(vcpu);
8796 	return r;
8797 }
8798 
8799 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8800 {
8801 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8802 		/*
8803 		 * We are here if userspace calls get_regs() in the middle of
8804 		 * instruction emulation. Registers state needs to be copied
8805 		 * back from emulation context to vcpu. Userspace shouldn't do
8806 		 * that usually, but some bad designed PV devices (vmware
8807 		 * backdoor interface) need this to work
8808 		 */
8809 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8810 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8811 	}
8812 	regs->rax = kvm_rax_read(vcpu);
8813 	regs->rbx = kvm_rbx_read(vcpu);
8814 	regs->rcx = kvm_rcx_read(vcpu);
8815 	regs->rdx = kvm_rdx_read(vcpu);
8816 	regs->rsi = kvm_rsi_read(vcpu);
8817 	regs->rdi = kvm_rdi_read(vcpu);
8818 	regs->rsp = kvm_rsp_read(vcpu);
8819 	regs->rbp = kvm_rbp_read(vcpu);
8820 #ifdef CONFIG_X86_64
8821 	regs->r8 = kvm_r8_read(vcpu);
8822 	regs->r9 = kvm_r9_read(vcpu);
8823 	regs->r10 = kvm_r10_read(vcpu);
8824 	regs->r11 = kvm_r11_read(vcpu);
8825 	regs->r12 = kvm_r12_read(vcpu);
8826 	regs->r13 = kvm_r13_read(vcpu);
8827 	regs->r14 = kvm_r14_read(vcpu);
8828 	regs->r15 = kvm_r15_read(vcpu);
8829 #endif
8830 
8831 	regs->rip = kvm_rip_read(vcpu);
8832 	regs->rflags = kvm_get_rflags(vcpu);
8833 }
8834 
8835 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8836 {
8837 	vcpu_load(vcpu);
8838 	__get_regs(vcpu, regs);
8839 	vcpu_put(vcpu);
8840 	return 0;
8841 }
8842 
8843 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8844 {
8845 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8846 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8847 
8848 	kvm_rax_write(vcpu, regs->rax);
8849 	kvm_rbx_write(vcpu, regs->rbx);
8850 	kvm_rcx_write(vcpu, regs->rcx);
8851 	kvm_rdx_write(vcpu, regs->rdx);
8852 	kvm_rsi_write(vcpu, regs->rsi);
8853 	kvm_rdi_write(vcpu, regs->rdi);
8854 	kvm_rsp_write(vcpu, regs->rsp);
8855 	kvm_rbp_write(vcpu, regs->rbp);
8856 #ifdef CONFIG_X86_64
8857 	kvm_r8_write(vcpu, regs->r8);
8858 	kvm_r9_write(vcpu, regs->r9);
8859 	kvm_r10_write(vcpu, regs->r10);
8860 	kvm_r11_write(vcpu, regs->r11);
8861 	kvm_r12_write(vcpu, regs->r12);
8862 	kvm_r13_write(vcpu, regs->r13);
8863 	kvm_r14_write(vcpu, regs->r14);
8864 	kvm_r15_write(vcpu, regs->r15);
8865 #endif
8866 
8867 	kvm_rip_write(vcpu, regs->rip);
8868 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8869 
8870 	vcpu->arch.exception.pending = false;
8871 
8872 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8873 }
8874 
8875 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8876 {
8877 	vcpu_load(vcpu);
8878 	__set_regs(vcpu, regs);
8879 	vcpu_put(vcpu);
8880 	return 0;
8881 }
8882 
8883 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8884 {
8885 	struct kvm_segment cs;
8886 
8887 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8888 	*db = cs.db;
8889 	*l = cs.l;
8890 }
8891 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8892 
8893 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8894 {
8895 	struct desc_ptr dt;
8896 
8897 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8898 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8899 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8900 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8901 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8902 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8903 
8904 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8905 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8906 
8907 	kvm_x86_ops.get_idt(vcpu, &dt);
8908 	sregs->idt.limit = dt.size;
8909 	sregs->idt.base = dt.address;
8910 	kvm_x86_ops.get_gdt(vcpu, &dt);
8911 	sregs->gdt.limit = dt.size;
8912 	sregs->gdt.base = dt.address;
8913 
8914 	sregs->cr0 = kvm_read_cr0(vcpu);
8915 	sregs->cr2 = vcpu->arch.cr2;
8916 	sregs->cr3 = kvm_read_cr3(vcpu);
8917 	sregs->cr4 = kvm_read_cr4(vcpu);
8918 	sregs->cr8 = kvm_get_cr8(vcpu);
8919 	sregs->efer = vcpu->arch.efer;
8920 	sregs->apic_base = kvm_get_apic_base(vcpu);
8921 
8922 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8923 
8924 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8925 		set_bit(vcpu->arch.interrupt.nr,
8926 			(unsigned long *)sregs->interrupt_bitmap);
8927 }
8928 
8929 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8930 				  struct kvm_sregs *sregs)
8931 {
8932 	vcpu_load(vcpu);
8933 	__get_sregs(vcpu, sregs);
8934 	vcpu_put(vcpu);
8935 	return 0;
8936 }
8937 
8938 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8939 				    struct kvm_mp_state *mp_state)
8940 {
8941 	vcpu_load(vcpu);
8942 	if (kvm_mpx_supported())
8943 		kvm_load_guest_fpu(vcpu);
8944 
8945 	kvm_apic_accept_events(vcpu);
8946 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8947 					vcpu->arch.pv.pv_unhalted)
8948 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8949 	else
8950 		mp_state->mp_state = vcpu->arch.mp_state;
8951 
8952 	if (kvm_mpx_supported())
8953 		kvm_put_guest_fpu(vcpu);
8954 	vcpu_put(vcpu);
8955 	return 0;
8956 }
8957 
8958 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8959 				    struct kvm_mp_state *mp_state)
8960 {
8961 	int ret = -EINVAL;
8962 
8963 	vcpu_load(vcpu);
8964 
8965 	if (!lapic_in_kernel(vcpu) &&
8966 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8967 		goto out;
8968 
8969 	/*
8970 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8971 	 * INIT state; latched init should be reported using
8972 	 * KVM_SET_VCPU_EVENTS, so reject it here.
8973 	 */
8974 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8975 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8976 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8977 		goto out;
8978 
8979 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8980 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8981 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8982 	} else
8983 		vcpu->arch.mp_state = mp_state->mp_state;
8984 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8985 
8986 	ret = 0;
8987 out:
8988 	vcpu_put(vcpu);
8989 	return ret;
8990 }
8991 
8992 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8993 		    int reason, bool has_error_code, u32 error_code)
8994 {
8995 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8996 	int ret;
8997 
8998 	init_emulate_ctxt(vcpu);
8999 
9000 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9001 				   has_error_code, error_code);
9002 	if (ret) {
9003 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9004 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9005 		vcpu->run->internal.ndata = 0;
9006 		return 0;
9007 	}
9008 
9009 	kvm_rip_write(vcpu, ctxt->eip);
9010 	kvm_set_rflags(vcpu, ctxt->eflags);
9011 	return 1;
9012 }
9013 EXPORT_SYMBOL_GPL(kvm_task_switch);
9014 
9015 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9016 {
9017 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9018 		/*
9019 		 * When EFER.LME and CR0.PG are set, the processor is in
9020 		 * 64-bit mode (though maybe in a 32-bit code segment).
9021 		 * CR4.PAE and EFER.LMA must be set.
9022 		 */
9023 		if (!(sregs->cr4 & X86_CR4_PAE)
9024 		    || !(sregs->efer & EFER_LMA))
9025 			return -EINVAL;
9026 	} else {
9027 		/*
9028 		 * Not in 64-bit mode: EFER.LMA is clear and the code
9029 		 * segment cannot be 64-bit.
9030 		 */
9031 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9032 			return -EINVAL;
9033 	}
9034 
9035 	return kvm_valid_cr4(vcpu, sregs->cr4);
9036 }
9037 
9038 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9039 {
9040 	struct msr_data apic_base_msr;
9041 	int mmu_reset_needed = 0;
9042 	int cpuid_update_needed = 0;
9043 	int pending_vec, max_bits, idx;
9044 	struct desc_ptr dt;
9045 	int ret = -EINVAL;
9046 
9047 	if (kvm_valid_sregs(vcpu, sregs))
9048 		goto out;
9049 
9050 	apic_base_msr.data = sregs->apic_base;
9051 	apic_base_msr.host_initiated = true;
9052 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
9053 		goto out;
9054 
9055 	dt.size = sregs->idt.limit;
9056 	dt.address = sregs->idt.base;
9057 	kvm_x86_ops.set_idt(vcpu, &dt);
9058 	dt.size = sregs->gdt.limit;
9059 	dt.address = sregs->gdt.base;
9060 	kvm_x86_ops.set_gdt(vcpu, &dt);
9061 
9062 	vcpu->arch.cr2 = sregs->cr2;
9063 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9064 	vcpu->arch.cr3 = sregs->cr3;
9065 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9066 
9067 	kvm_set_cr8(vcpu, sregs->cr8);
9068 
9069 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9070 	kvm_x86_ops.set_efer(vcpu, sregs->efer);
9071 
9072 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9073 	kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9074 	vcpu->arch.cr0 = sregs->cr0;
9075 
9076 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9077 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9078 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
9079 	kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9080 	if (cpuid_update_needed)
9081 		kvm_update_cpuid(vcpu);
9082 
9083 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9084 	if (is_pae_paging(vcpu)) {
9085 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9086 		mmu_reset_needed = 1;
9087 	}
9088 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9089 
9090 	if (mmu_reset_needed)
9091 		kvm_mmu_reset_context(vcpu);
9092 
9093 	max_bits = KVM_NR_INTERRUPTS;
9094 	pending_vec = find_first_bit(
9095 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
9096 	if (pending_vec < max_bits) {
9097 		kvm_queue_interrupt(vcpu, pending_vec, false);
9098 		pr_debug("Set back pending irq %d\n", pending_vec);
9099 	}
9100 
9101 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9102 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9103 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9104 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9105 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9106 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9107 
9108 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9109 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9110 
9111 	update_cr8_intercept(vcpu);
9112 
9113 	/* Older userspace won't unhalt the vcpu on reset. */
9114 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9115 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9116 	    !is_protmode(vcpu))
9117 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9118 
9119 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9120 
9121 	ret = 0;
9122 out:
9123 	return ret;
9124 }
9125 
9126 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9127 				  struct kvm_sregs *sregs)
9128 {
9129 	int ret;
9130 
9131 	vcpu_load(vcpu);
9132 	ret = __set_sregs(vcpu, sregs);
9133 	vcpu_put(vcpu);
9134 	return ret;
9135 }
9136 
9137 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9138 					struct kvm_guest_debug *dbg)
9139 {
9140 	unsigned long rflags;
9141 	int i, r;
9142 
9143 	vcpu_load(vcpu);
9144 
9145 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9146 		r = -EBUSY;
9147 		if (vcpu->arch.exception.pending)
9148 			goto out;
9149 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9150 			kvm_queue_exception(vcpu, DB_VECTOR);
9151 		else
9152 			kvm_queue_exception(vcpu, BP_VECTOR);
9153 	}
9154 
9155 	/*
9156 	 * Read rflags as long as potentially injected trace flags are still
9157 	 * filtered out.
9158 	 */
9159 	rflags = kvm_get_rflags(vcpu);
9160 
9161 	vcpu->guest_debug = dbg->control;
9162 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9163 		vcpu->guest_debug = 0;
9164 
9165 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9166 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
9167 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9168 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9169 	} else {
9170 		for (i = 0; i < KVM_NR_DB_REGS; i++)
9171 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9172 	}
9173 	kvm_update_dr7(vcpu);
9174 
9175 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9176 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9177 			get_segment_base(vcpu, VCPU_SREG_CS);
9178 
9179 	/*
9180 	 * Trigger an rflags update that will inject or remove the trace
9181 	 * flags.
9182 	 */
9183 	kvm_set_rflags(vcpu, rflags);
9184 
9185 	kvm_x86_ops.update_bp_intercept(vcpu);
9186 
9187 	r = 0;
9188 
9189 out:
9190 	vcpu_put(vcpu);
9191 	return r;
9192 }
9193 
9194 /*
9195  * Translate a guest virtual address to a guest physical address.
9196  */
9197 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9198 				    struct kvm_translation *tr)
9199 {
9200 	unsigned long vaddr = tr->linear_address;
9201 	gpa_t gpa;
9202 	int idx;
9203 
9204 	vcpu_load(vcpu);
9205 
9206 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9207 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9208 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9209 	tr->physical_address = gpa;
9210 	tr->valid = gpa != UNMAPPED_GVA;
9211 	tr->writeable = 1;
9212 	tr->usermode = 0;
9213 
9214 	vcpu_put(vcpu);
9215 	return 0;
9216 }
9217 
9218 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9219 {
9220 	struct fxregs_state *fxsave;
9221 
9222 	vcpu_load(vcpu);
9223 
9224 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9225 	memcpy(fpu->fpr, fxsave->st_space, 128);
9226 	fpu->fcw = fxsave->cwd;
9227 	fpu->fsw = fxsave->swd;
9228 	fpu->ftwx = fxsave->twd;
9229 	fpu->last_opcode = fxsave->fop;
9230 	fpu->last_ip = fxsave->rip;
9231 	fpu->last_dp = fxsave->rdp;
9232 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9233 
9234 	vcpu_put(vcpu);
9235 	return 0;
9236 }
9237 
9238 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9239 {
9240 	struct fxregs_state *fxsave;
9241 
9242 	vcpu_load(vcpu);
9243 
9244 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9245 
9246 	memcpy(fxsave->st_space, fpu->fpr, 128);
9247 	fxsave->cwd = fpu->fcw;
9248 	fxsave->swd = fpu->fsw;
9249 	fxsave->twd = fpu->ftwx;
9250 	fxsave->fop = fpu->last_opcode;
9251 	fxsave->rip = fpu->last_ip;
9252 	fxsave->rdp = fpu->last_dp;
9253 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9254 
9255 	vcpu_put(vcpu);
9256 	return 0;
9257 }
9258 
9259 static void store_regs(struct kvm_vcpu *vcpu)
9260 {
9261 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9262 
9263 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9264 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9265 
9266 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9267 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9268 
9269 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9270 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9271 				vcpu, &vcpu->run->s.regs.events);
9272 }
9273 
9274 static int sync_regs(struct kvm_vcpu *vcpu)
9275 {
9276 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9277 		return -EINVAL;
9278 
9279 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9280 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9281 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9282 	}
9283 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9284 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9285 			return -EINVAL;
9286 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9287 	}
9288 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9289 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9290 				vcpu, &vcpu->run->s.regs.events))
9291 			return -EINVAL;
9292 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9293 	}
9294 
9295 	return 0;
9296 }
9297 
9298 static void fx_init(struct kvm_vcpu *vcpu)
9299 {
9300 	fpstate_init(&vcpu->arch.guest_fpu->state);
9301 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9302 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9303 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9304 
9305 	/*
9306 	 * Ensure guest xcr0 is valid for loading
9307 	 */
9308 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9309 
9310 	vcpu->arch.cr0 |= X86_CR0_ET;
9311 }
9312 
9313 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9314 {
9315 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9316 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9317 			     "guest TSC will not be reliable\n");
9318 
9319 	return 0;
9320 }
9321 
9322 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9323 {
9324 	struct page *page;
9325 	int r;
9326 
9327 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9328 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9329 	else
9330 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9331 
9332 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9333 
9334 	r = kvm_mmu_create(vcpu);
9335 	if (r < 0)
9336 		return r;
9337 
9338 	if (irqchip_in_kernel(vcpu->kvm)) {
9339 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9340 		if (r < 0)
9341 			goto fail_mmu_destroy;
9342 		if (kvm_apicv_activated(vcpu->kvm))
9343 			vcpu->arch.apicv_active = true;
9344 	} else
9345 		static_key_slow_inc(&kvm_no_apic_vcpu);
9346 
9347 	r = -ENOMEM;
9348 
9349 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9350 	if (!page)
9351 		goto fail_free_lapic;
9352 	vcpu->arch.pio_data = page_address(page);
9353 
9354 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9355 				       GFP_KERNEL_ACCOUNT);
9356 	if (!vcpu->arch.mce_banks)
9357 		goto fail_free_pio_data;
9358 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9359 
9360 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9361 				GFP_KERNEL_ACCOUNT))
9362 		goto fail_free_mce_banks;
9363 
9364 	if (!alloc_emulate_ctxt(vcpu))
9365 		goto free_wbinvd_dirty_mask;
9366 
9367 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9368 						GFP_KERNEL_ACCOUNT);
9369 	if (!vcpu->arch.user_fpu) {
9370 		pr_err("kvm: failed to allocate userspace's fpu\n");
9371 		goto free_emulate_ctxt;
9372 	}
9373 
9374 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9375 						 GFP_KERNEL_ACCOUNT);
9376 	if (!vcpu->arch.guest_fpu) {
9377 		pr_err("kvm: failed to allocate vcpu's fpu\n");
9378 		goto free_user_fpu;
9379 	}
9380 	fx_init(vcpu);
9381 
9382 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9383 
9384 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9385 
9386 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9387 
9388 	kvm_async_pf_hash_reset(vcpu);
9389 	kvm_pmu_init(vcpu);
9390 
9391 	vcpu->arch.pending_external_vector = -1;
9392 	vcpu->arch.preempted_in_kernel = false;
9393 
9394 	kvm_hv_vcpu_init(vcpu);
9395 
9396 	r = kvm_x86_ops.vcpu_create(vcpu);
9397 	if (r)
9398 		goto free_guest_fpu;
9399 
9400 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9401 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9402 	kvm_vcpu_mtrr_init(vcpu);
9403 	vcpu_load(vcpu);
9404 	kvm_vcpu_reset(vcpu, false);
9405 	kvm_init_mmu(vcpu, false);
9406 	vcpu_put(vcpu);
9407 	return 0;
9408 
9409 free_guest_fpu:
9410 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9411 free_user_fpu:
9412 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9413 free_emulate_ctxt:
9414 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9415 free_wbinvd_dirty_mask:
9416 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9417 fail_free_mce_banks:
9418 	kfree(vcpu->arch.mce_banks);
9419 fail_free_pio_data:
9420 	free_page((unsigned long)vcpu->arch.pio_data);
9421 fail_free_lapic:
9422 	kvm_free_lapic(vcpu);
9423 fail_mmu_destroy:
9424 	kvm_mmu_destroy(vcpu);
9425 	return r;
9426 }
9427 
9428 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9429 {
9430 	struct msr_data msr;
9431 	struct kvm *kvm = vcpu->kvm;
9432 
9433 	kvm_hv_vcpu_postcreate(vcpu);
9434 
9435 	if (mutex_lock_killable(&vcpu->mutex))
9436 		return;
9437 	vcpu_load(vcpu);
9438 	msr.data = 0x0;
9439 	msr.index = MSR_IA32_TSC;
9440 	msr.host_initiated = true;
9441 	kvm_write_tsc(vcpu, &msr);
9442 	vcpu_put(vcpu);
9443 
9444 	/* poll control enabled by default */
9445 	vcpu->arch.msr_kvm_poll_control = 1;
9446 
9447 	mutex_unlock(&vcpu->mutex);
9448 
9449 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9450 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9451 						KVMCLOCK_SYNC_PERIOD);
9452 }
9453 
9454 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9455 {
9456 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9457 	int idx;
9458 
9459 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
9460 
9461 	kvmclock_reset(vcpu);
9462 
9463 	kvm_x86_ops.vcpu_free(vcpu);
9464 
9465 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9466 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9467 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9468 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9469 
9470 	kvm_hv_vcpu_uninit(vcpu);
9471 	kvm_pmu_destroy(vcpu);
9472 	kfree(vcpu->arch.mce_banks);
9473 	kvm_free_lapic(vcpu);
9474 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9475 	kvm_mmu_destroy(vcpu);
9476 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9477 	free_page((unsigned long)vcpu->arch.pio_data);
9478 	if (!lapic_in_kernel(vcpu))
9479 		static_key_slow_dec(&kvm_no_apic_vcpu);
9480 }
9481 
9482 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9483 {
9484 	kvm_lapic_reset(vcpu, init_event);
9485 
9486 	vcpu->arch.hflags = 0;
9487 
9488 	vcpu->arch.smi_pending = 0;
9489 	vcpu->arch.smi_count = 0;
9490 	atomic_set(&vcpu->arch.nmi_queued, 0);
9491 	vcpu->arch.nmi_pending = 0;
9492 	vcpu->arch.nmi_injected = false;
9493 	kvm_clear_interrupt_queue(vcpu);
9494 	kvm_clear_exception_queue(vcpu);
9495 
9496 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9497 	kvm_update_dr0123(vcpu);
9498 	vcpu->arch.dr6 = DR6_INIT;
9499 	vcpu->arch.dr7 = DR7_FIXED_1;
9500 	kvm_update_dr7(vcpu);
9501 
9502 	vcpu->arch.cr2 = 0;
9503 
9504 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9505 	vcpu->arch.apf.msr_val = 0;
9506 	vcpu->arch.st.msr_val = 0;
9507 
9508 	kvmclock_reset(vcpu);
9509 
9510 	kvm_clear_async_pf_completion_queue(vcpu);
9511 	kvm_async_pf_hash_reset(vcpu);
9512 	vcpu->arch.apf.halted = false;
9513 
9514 	if (kvm_mpx_supported()) {
9515 		void *mpx_state_buffer;
9516 
9517 		/*
9518 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9519 		 * called with loaded FPU and does not let userspace fix the state.
9520 		 */
9521 		if (init_event)
9522 			kvm_put_guest_fpu(vcpu);
9523 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9524 					XFEATURE_BNDREGS);
9525 		if (mpx_state_buffer)
9526 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9527 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9528 					XFEATURE_BNDCSR);
9529 		if (mpx_state_buffer)
9530 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9531 		if (init_event)
9532 			kvm_load_guest_fpu(vcpu);
9533 	}
9534 
9535 	if (!init_event) {
9536 		kvm_pmu_reset(vcpu);
9537 		vcpu->arch.smbase = 0x30000;
9538 
9539 		vcpu->arch.msr_misc_features_enables = 0;
9540 
9541 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9542 	}
9543 
9544 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9545 	vcpu->arch.regs_avail = ~0;
9546 	vcpu->arch.regs_dirty = ~0;
9547 
9548 	vcpu->arch.ia32_xss = 0;
9549 
9550 	kvm_x86_ops.vcpu_reset(vcpu, init_event);
9551 }
9552 
9553 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9554 {
9555 	struct kvm_segment cs;
9556 
9557 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9558 	cs.selector = vector << 8;
9559 	cs.base = vector << 12;
9560 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9561 	kvm_rip_write(vcpu, 0);
9562 }
9563 
9564 int kvm_arch_hardware_enable(void)
9565 {
9566 	struct kvm *kvm;
9567 	struct kvm_vcpu *vcpu;
9568 	int i;
9569 	int ret;
9570 	u64 local_tsc;
9571 	u64 max_tsc = 0;
9572 	bool stable, backwards_tsc = false;
9573 
9574 	kvm_shared_msr_cpu_online();
9575 	ret = kvm_x86_ops.hardware_enable();
9576 	if (ret != 0)
9577 		return ret;
9578 
9579 	local_tsc = rdtsc();
9580 	stable = !kvm_check_tsc_unstable();
9581 	list_for_each_entry(kvm, &vm_list, vm_list) {
9582 		kvm_for_each_vcpu(i, vcpu, kvm) {
9583 			if (!stable && vcpu->cpu == smp_processor_id())
9584 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9585 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9586 				backwards_tsc = true;
9587 				if (vcpu->arch.last_host_tsc > max_tsc)
9588 					max_tsc = vcpu->arch.last_host_tsc;
9589 			}
9590 		}
9591 	}
9592 
9593 	/*
9594 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9595 	 * platforms that reset TSC during suspend or hibernate actions, but
9596 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9597 	 * detect that condition here, which happens early in CPU bringup,
9598 	 * before any KVM threads can be running.  Unfortunately, we can't
9599 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9600 	 * enough into CPU bringup that we know how much real time has actually
9601 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9602 	 * variables that haven't been updated yet.
9603 	 *
9604 	 * So we simply find the maximum observed TSC above, then record the
9605 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9606 	 * the adjustment will be applied.  Note that we accumulate
9607 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9608 	 * gets a chance to run again.  In the event that no KVM threads get a
9609 	 * chance to run, we will miss the entire elapsed period, as we'll have
9610 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9611 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9612 	 * uniform across all VCPUs (not to mention the scenario is extremely
9613 	 * unlikely). It is possible that a second hibernate recovery happens
9614 	 * much faster than a first, causing the observed TSC here to be
9615 	 * smaller; this would require additional padding adjustment, which is
9616 	 * why we set last_host_tsc to the local tsc observed here.
9617 	 *
9618 	 * N.B. - this code below runs only on platforms with reliable TSC,
9619 	 * as that is the only way backwards_tsc is set above.  Also note
9620 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9621 	 * have the same delta_cyc adjustment applied if backwards_tsc
9622 	 * is detected.  Note further, this adjustment is only done once,
9623 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9624 	 * called multiple times (one for each physical CPU bringup).
9625 	 *
9626 	 * Platforms with unreliable TSCs don't have to deal with this, they
9627 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9628 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9629 	 * guarantee that they stay in perfect synchronization.
9630 	 */
9631 	if (backwards_tsc) {
9632 		u64 delta_cyc = max_tsc - local_tsc;
9633 		list_for_each_entry(kvm, &vm_list, vm_list) {
9634 			kvm->arch.backwards_tsc_observed = true;
9635 			kvm_for_each_vcpu(i, vcpu, kvm) {
9636 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9637 				vcpu->arch.last_host_tsc = local_tsc;
9638 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9639 			}
9640 
9641 			/*
9642 			 * We have to disable TSC offset matching.. if you were
9643 			 * booting a VM while issuing an S4 host suspend....
9644 			 * you may have some problem.  Solving this issue is
9645 			 * left as an exercise to the reader.
9646 			 */
9647 			kvm->arch.last_tsc_nsec = 0;
9648 			kvm->arch.last_tsc_write = 0;
9649 		}
9650 
9651 	}
9652 	return 0;
9653 }
9654 
9655 void kvm_arch_hardware_disable(void)
9656 {
9657 	kvm_x86_ops.hardware_disable();
9658 	drop_user_return_notifiers();
9659 }
9660 
9661 int kvm_arch_hardware_setup(void *opaque)
9662 {
9663 	struct kvm_x86_init_ops *ops = opaque;
9664 	int r;
9665 
9666 	rdmsrl_safe(MSR_EFER, &host_efer);
9667 
9668 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9669 		rdmsrl(MSR_IA32_XSS, host_xss);
9670 
9671 	r = ops->hardware_setup();
9672 	if (r != 0)
9673 		return r;
9674 
9675 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9676 
9677 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9678 		supported_xss = 0;
9679 
9680 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9681 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9682 #undef __kvm_cpu_cap_has
9683 
9684 	if (kvm_has_tsc_control) {
9685 		/*
9686 		 * Make sure the user can only configure tsc_khz values that
9687 		 * fit into a signed integer.
9688 		 * A min value is not calculated because it will always
9689 		 * be 1 on all machines.
9690 		 */
9691 		u64 max = min(0x7fffffffULL,
9692 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9693 		kvm_max_guest_tsc_khz = max;
9694 
9695 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9696 	}
9697 
9698 	kvm_init_msr_list();
9699 	return 0;
9700 }
9701 
9702 void kvm_arch_hardware_unsetup(void)
9703 {
9704 	kvm_x86_ops.hardware_unsetup();
9705 }
9706 
9707 int kvm_arch_check_processor_compat(void *opaque)
9708 {
9709 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9710 	struct kvm_x86_init_ops *ops = opaque;
9711 
9712 	WARN_ON(!irqs_disabled());
9713 
9714 	if (__cr4_reserved_bits(cpu_has, c) !=
9715 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9716 		return -EIO;
9717 
9718 	return ops->check_processor_compatibility();
9719 }
9720 
9721 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9722 {
9723 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9724 }
9725 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9726 
9727 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9728 {
9729 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9730 }
9731 
9732 struct static_key kvm_no_apic_vcpu __read_mostly;
9733 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9734 
9735 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9736 {
9737 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9738 
9739 	vcpu->arch.l1tf_flush_l1d = true;
9740 	if (pmu->version && unlikely(pmu->event_count)) {
9741 		pmu->need_cleanup = true;
9742 		kvm_make_request(KVM_REQ_PMU, vcpu);
9743 	}
9744 	kvm_x86_ops.sched_in(vcpu, cpu);
9745 }
9746 
9747 void kvm_arch_free_vm(struct kvm *kvm)
9748 {
9749 	kfree(kvm->arch.hyperv.hv_pa_pg);
9750 	vfree(kvm);
9751 }
9752 
9753 
9754 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9755 {
9756 	if (type)
9757 		return -EINVAL;
9758 
9759 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9760 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9761 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9762 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9763 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9764 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9765 
9766 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9767 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9768 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9769 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9770 		&kvm->arch.irq_sources_bitmap);
9771 
9772 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9773 	mutex_init(&kvm->arch.apic_map_lock);
9774 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9775 
9776 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9777 	pvclock_update_vm_gtod_copy(kvm);
9778 
9779 	kvm->arch.guest_can_read_msr_platform_info = true;
9780 
9781 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9782 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9783 
9784 	kvm_hv_init_vm(kvm);
9785 	kvm_page_track_init(kvm);
9786 	kvm_mmu_init_vm(kvm);
9787 
9788 	return kvm_x86_ops.vm_init(kvm);
9789 }
9790 
9791 int kvm_arch_post_init_vm(struct kvm *kvm)
9792 {
9793 	return kvm_mmu_post_init_vm(kvm);
9794 }
9795 
9796 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9797 {
9798 	vcpu_load(vcpu);
9799 	kvm_mmu_unload(vcpu);
9800 	vcpu_put(vcpu);
9801 }
9802 
9803 static void kvm_free_vcpus(struct kvm *kvm)
9804 {
9805 	unsigned int i;
9806 	struct kvm_vcpu *vcpu;
9807 
9808 	/*
9809 	 * Unpin any mmu pages first.
9810 	 */
9811 	kvm_for_each_vcpu(i, vcpu, kvm) {
9812 		kvm_clear_async_pf_completion_queue(vcpu);
9813 		kvm_unload_vcpu_mmu(vcpu);
9814 	}
9815 	kvm_for_each_vcpu(i, vcpu, kvm)
9816 		kvm_vcpu_destroy(vcpu);
9817 
9818 	mutex_lock(&kvm->lock);
9819 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9820 		kvm->vcpus[i] = NULL;
9821 
9822 	atomic_set(&kvm->online_vcpus, 0);
9823 	mutex_unlock(&kvm->lock);
9824 }
9825 
9826 void kvm_arch_sync_events(struct kvm *kvm)
9827 {
9828 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9829 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9830 	kvm_free_pit(kvm);
9831 }
9832 
9833 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9834 {
9835 	int i, r;
9836 	unsigned long hva, uninitialized_var(old_npages);
9837 	struct kvm_memslots *slots = kvm_memslots(kvm);
9838 	struct kvm_memory_slot *slot;
9839 
9840 	/* Called with kvm->slots_lock held.  */
9841 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9842 		return -EINVAL;
9843 
9844 	slot = id_to_memslot(slots, id);
9845 	if (size) {
9846 		if (slot && slot->npages)
9847 			return -EEXIST;
9848 
9849 		/*
9850 		 * MAP_SHARED to prevent internal slot pages from being moved
9851 		 * by fork()/COW.
9852 		 */
9853 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9854 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9855 		if (IS_ERR((void *)hva))
9856 			return PTR_ERR((void *)hva);
9857 	} else {
9858 		if (!slot || !slot->npages)
9859 			return 0;
9860 
9861 		/*
9862 		 * Stuff a non-canonical value to catch use-after-delete.  This
9863 		 * ends up being 0 on 32-bit KVM, but there's no better
9864 		 * alternative.
9865 		 */
9866 		hva = (unsigned long)(0xdeadull << 48);
9867 		old_npages = slot->npages;
9868 	}
9869 
9870 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9871 		struct kvm_userspace_memory_region m;
9872 
9873 		m.slot = id | (i << 16);
9874 		m.flags = 0;
9875 		m.guest_phys_addr = gpa;
9876 		m.userspace_addr = hva;
9877 		m.memory_size = size;
9878 		r = __kvm_set_memory_region(kvm, &m);
9879 		if (r < 0)
9880 			return r;
9881 	}
9882 
9883 	if (!size)
9884 		vm_munmap(hva, old_npages * PAGE_SIZE);
9885 
9886 	return 0;
9887 }
9888 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9889 
9890 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9891 {
9892 	kvm_mmu_pre_destroy_vm(kvm);
9893 }
9894 
9895 void kvm_arch_destroy_vm(struct kvm *kvm)
9896 {
9897 	if (current->mm == kvm->mm) {
9898 		/*
9899 		 * Free memory regions allocated on behalf of userspace,
9900 		 * unless the the memory map has changed due to process exit
9901 		 * or fd copying.
9902 		 */
9903 		mutex_lock(&kvm->slots_lock);
9904 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9905 					0, 0);
9906 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9907 					0, 0);
9908 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9909 		mutex_unlock(&kvm->slots_lock);
9910 	}
9911 	if (kvm_x86_ops.vm_destroy)
9912 		kvm_x86_ops.vm_destroy(kvm);
9913 	kvm_pic_destroy(kvm);
9914 	kvm_ioapic_destroy(kvm);
9915 	kvm_free_vcpus(kvm);
9916 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9917 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9918 	kvm_mmu_uninit_vm(kvm);
9919 	kvm_page_track_cleanup(kvm);
9920 	kvm_hv_destroy_vm(kvm);
9921 }
9922 
9923 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
9924 {
9925 	int i;
9926 
9927 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9928 		kvfree(slot->arch.rmap[i]);
9929 		slot->arch.rmap[i] = NULL;
9930 
9931 		if (i == 0)
9932 			continue;
9933 
9934 		kvfree(slot->arch.lpage_info[i - 1]);
9935 		slot->arch.lpage_info[i - 1] = NULL;
9936 	}
9937 
9938 	kvm_page_track_free_memslot(slot);
9939 }
9940 
9941 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
9942 				      unsigned long npages)
9943 {
9944 	int i;
9945 
9946 	/*
9947 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
9948 	 * old arrays will be freed by __kvm_set_memory_region() if installing
9949 	 * the new memslot is successful.
9950 	 */
9951 	memset(&slot->arch, 0, sizeof(slot->arch));
9952 
9953 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9954 		struct kvm_lpage_info *linfo;
9955 		unsigned long ugfn;
9956 		int lpages;
9957 		int level = i + 1;
9958 
9959 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9960 				      slot->base_gfn, level) + 1;
9961 
9962 		slot->arch.rmap[i] =
9963 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9964 				 GFP_KERNEL_ACCOUNT);
9965 		if (!slot->arch.rmap[i])
9966 			goto out_free;
9967 		if (i == 0)
9968 			continue;
9969 
9970 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9971 		if (!linfo)
9972 			goto out_free;
9973 
9974 		slot->arch.lpage_info[i - 1] = linfo;
9975 
9976 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9977 			linfo[0].disallow_lpage = 1;
9978 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9979 			linfo[lpages - 1].disallow_lpage = 1;
9980 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9981 		/*
9982 		 * If the gfn and userspace address are not aligned wrt each
9983 		 * other, disable large page support for this slot.
9984 		 */
9985 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
9986 			unsigned long j;
9987 
9988 			for (j = 0; j < lpages; ++j)
9989 				linfo[j].disallow_lpage = 1;
9990 		}
9991 	}
9992 
9993 	if (kvm_page_track_create_memslot(slot, npages))
9994 		goto out_free;
9995 
9996 	return 0;
9997 
9998 out_free:
9999 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10000 		kvfree(slot->arch.rmap[i]);
10001 		slot->arch.rmap[i] = NULL;
10002 		if (i == 0)
10003 			continue;
10004 
10005 		kvfree(slot->arch.lpage_info[i - 1]);
10006 		slot->arch.lpage_info[i - 1] = NULL;
10007 	}
10008 	return -ENOMEM;
10009 }
10010 
10011 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10012 {
10013 	struct kvm_vcpu *vcpu;
10014 	int i;
10015 
10016 	/*
10017 	 * memslots->generation has been incremented.
10018 	 * mmio generation may have reached its maximum value.
10019 	 */
10020 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10021 
10022 	/* Force re-initialization of steal_time cache */
10023 	kvm_for_each_vcpu(i, vcpu, kvm)
10024 		kvm_vcpu_kick(vcpu);
10025 }
10026 
10027 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10028 				struct kvm_memory_slot *memslot,
10029 				const struct kvm_userspace_memory_region *mem,
10030 				enum kvm_mr_change change)
10031 {
10032 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10033 		return kvm_alloc_memslot_metadata(memslot,
10034 						  mem->memory_size >> PAGE_SHIFT);
10035 	return 0;
10036 }
10037 
10038 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10039 				     struct kvm_memory_slot *new)
10040 {
10041 	/* Still write protect RO slot */
10042 	if (new->flags & KVM_MEM_READONLY) {
10043 		kvm_mmu_slot_remove_write_access(kvm, new, PT_PAGE_TABLE_LEVEL);
10044 		return;
10045 	}
10046 
10047 	/*
10048 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
10049 	 *
10050 	 * kvm_x86_ops.slot_disable_log_dirty is called when:
10051 	 *
10052 	 *  - KVM_MR_CREATE with dirty logging is disabled
10053 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
10054 	 *
10055 	 * The reason is, in case of PML, we need to set D-bit for any slots
10056 	 * with dirty logging disabled in order to eliminate unnecessary GPA
10057 	 * logging in PML buffer (and potential PML buffer full VMEXIT). This
10058 	 * guarantees leaving PML enabled during guest's lifetime won't have
10059 	 * any additional overhead from PML when guest is running with dirty
10060 	 * logging disabled for memory slots.
10061 	 *
10062 	 * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot
10063 	 * to dirty logging mode.
10064 	 *
10065 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
10066 	 *
10067 	 * In case of write protect:
10068 	 *
10069 	 * Write protect all pages for dirty logging.
10070 	 *
10071 	 * All the sptes including the large sptes which point to this
10072 	 * slot are set to readonly. We can not create any new large
10073 	 * spte on this slot until the end of the logging.
10074 	 *
10075 	 * See the comments in fast_page_fault().
10076 	 */
10077 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10078 		if (kvm_x86_ops.slot_enable_log_dirty) {
10079 			kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10080 		} else {
10081 			int level =
10082 				kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10083 				PT_DIRECTORY_LEVEL : PT_PAGE_TABLE_LEVEL;
10084 
10085 			/*
10086 			 * If we're with initial-all-set, we don't need
10087 			 * to write protect any small page because
10088 			 * they're reported as dirty already.  However
10089 			 * we still need to write-protect huge pages
10090 			 * so that the page split can happen lazily on
10091 			 * the first write to the huge page.
10092 			 */
10093 			kvm_mmu_slot_remove_write_access(kvm, new, level);
10094 		}
10095 	} else {
10096 		if (kvm_x86_ops.slot_disable_log_dirty)
10097 			kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10098 	}
10099 }
10100 
10101 void kvm_arch_commit_memory_region(struct kvm *kvm,
10102 				const struct kvm_userspace_memory_region *mem,
10103 				struct kvm_memory_slot *old,
10104 				const struct kvm_memory_slot *new,
10105 				enum kvm_mr_change change)
10106 {
10107 	if (!kvm->arch.n_requested_mmu_pages)
10108 		kvm_mmu_change_mmu_pages(kvm,
10109 				kvm_mmu_calculate_default_mmu_pages(kvm));
10110 
10111 	/*
10112 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
10113 	 * sptes have to be split.  If live migration is successful, the guest
10114 	 * in the source machine will be destroyed and large sptes will be
10115 	 * created in the destination. However, if the guest continues to run
10116 	 * in the source machine (for example if live migration fails), small
10117 	 * sptes will remain around and cause bad performance.
10118 	 *
10119 	 * Scan sptes if dirty logging has been stopped, dropping those
10120 	 * which can be collapsed into a single large-page spte.  Later
10121 	 * page faults will create the large-page sptes.
10122 	 *
10123 	 * There is no need to do this in any of the following cases:
10124 	 * CREATE:	No dirty mappings will already exist.
10125 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
10126 	 *		kvm_arch_flush_shadow_memslot()
10127 	 */
10128 	if (change == KVM_MR_FLAGS_ONLY &&
10129 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10130 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10131 		kvm_mmu_zap_collapsible_sptes(kvm, new);
10132 
10133 	/*
10134 	 * Set up write protection and/or dirty logging for the new slot.
10135 	 *
10136 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
10137 	 * been zapped so no dirty logging staff is needed for old slot. For
10138 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
10139 	 * new and it's also covered when dealing with the new slot.
10140 	 *
10141 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
10142 	 */
10143 	if (change != KVM_MR_DELETE)
10144 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
10145 
10146 	/* Free the arrays associated with the old memslot. */
10147 	if (change == KVM_MR_MOVE)
10148 		kvm_arch_free_memslot(kvm, old);
10149 }
10150 
10151 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10152 {
10153 	kvm_mmu_zap_all(kvm);
10154 }
10155 
10156 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10157 				   struct kvm_memory_slot *slot)
10158 {
10159 	kvm_page_track_flush_slot(kvm, slot);
10160 }
10161 
10162 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10163 {
10164 	return (is_guest_mode(vcpu) &&
10165 			kvm_x86_ops.guest_apic_has_interrupt &&
10166 			kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10167 }
10168 
10169 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10170 {
10171 	if (!list_empty_careful(&vcpu->async_pf.done))
10172 		return true;
10173 
10174 	if (kvm_apic_has_events(vcpu))
10175 		return true;
10176 
10177 	if (vcpu->arch.pv.pv_unhalted)
10178 		return true;
10179 
10180 	if (vcpu->arch.exception.pending)
10181 		return true;
10182 
10183 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10184 	    (vcpu->arch.nmi_pending &&
10185 	     kvm_x86_ops.nmi_allowed(vcpu)))
10186 		return true;
10187 
10188 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10189 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
10190 		return true;
10191 
10192 	if (kvm_arch_interrupt_allowed(vcpu) &&
10193 	    (kvm_cpu_has_interrupt(vcpu) ||
10194 	    kvm_guest_apic_has_interrupt(vcpu)))
10195 		return true;
10196 
10197 	if (kvm_hv_has_stimer_pending(vcpu))
10198 		return true;
10199 
10200 	return false;
10201 }
10202 
10203 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10204 {
10205 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10206 }
10207 
10208 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10209 {
10210 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10211 		return true;
10212 
10213 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10214 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
10215 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
10216 		return true;
10217 
10218 	if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10219 		return true;
10220 
10221 	return false;
10222 }
10223 
10224 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10225 {
10226 	return vcpu->arch.preempted_in_kernel;
10227 }
10228 
10229 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10230 {
10231 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10232 }
10233 
10234 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10235 {
10236 	return kvm_x86_ops.interrupt_allowed(vcpu);
10237 }
10238 
10239 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10240 {
10241 	if (is_64_bit_mode(vcpu))
10242 		return kvm_rip_read(vcpu);
10243 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10244 		     kvm_rip_read(vcpu));
10245 }
10246 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10247 
10248 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10249 {
10250 	return kvm_get_linear_rip(vcpu) == linear_rip;
10251 }
10252 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10253 
10254 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10255 {
10256 	unsigned long rflags;
10257 
10258 	rflags = kvm_x86_ops.get_rflags(vcpu);
10259 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10260 		rflags &= ~X86_EFLAGS_TF;
10261 	return rflags;
10262 }
10263 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10264 
10265 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10266 {
10267 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10268 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10269 		rflags |= X86_EFLAGS_TF;
10270 	kvm_x86_ops.set_rflags(vcpu, rflags);
10271 }
10272 
10273 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10274 {
10275 	__kvm_set_rflags(vcpu, rflags);
10276 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10277 }
10278 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10279 
10280 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10281 {
10282 	int r;
10283 
10284 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10285 	      work->wakeup_all)
10286 		return;
10287 
10288 	r = kvm_mmu_reload(vcpu);
10289 	if (unlikely(r))
10290 		return;
10291 
10292 	if (!vcpu->arch.mmu->direct_map &&
10293 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10294 		return;
10295 
10296 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10297 }
10298 
10299 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10300 {
10301 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10302 }
10303 
10304 static inline u32 kvm_async_pf_next_probe(u32 key)
10305 {
10306 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10307 }
10308 
10309 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10310 {
10311 	u32 key = kvm_async_pf_hash_fn(gfn);
10312 
10313 	while (vcpu->arch.apf.gfns[key] != ~0)
10314 		key = kvm_async_pf_next_probe(key);
10315 
10316 	vcpu->arch.apf.gfns[key] = gfn;
10317 }
10318 
10319 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10320 {
10321 	int i;
10322 	u32 key = kvm_async_pf_hash_fn(gfn);
10323 
10324 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10325 		     (vcpu->arch.apf.gfns[key] != gfn &&
10326 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10327 		key = kvm_async_pf_next_probe(key);
10328 
10329 	return key;
10330 }
10331 
10332 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10333 {
10334 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10335 }
10336 
10337 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10338 {
10339 	u32 i, j, k;
10340 
10341 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10342 	while (true) {
10343 		vcpu->arch.apf.gfns[i] = ~0;
10344 		do {
10345 			j = kvm_async_pf_next_probe(j);
10346 			if (vcpu->arch.apf.gfns[j] == ~0)
10347 				return;
10348 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10349 			/*
10350 			 * k lies cyclically in ]i,j]
10351 			 * |    i.k.j |
10352 			 * |....j i.k.| or  |.k..j i...|
10353 			 */
10354 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10355 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10356 		i = j;
10357 	}
10358 }
10359 
10360 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10361 {
10362 
10363 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10364 				      sizeof(val));
10365 }
10366 
10367 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10368 {
10369 
10370 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10371 				      sizeof(u32));
10372 }
10373 
10374 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10375 {
10376 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10377 		return false;
10378 
10379 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10380 	    (vcpu->arch.apf.send_user_only &&
10381 	     kvm_x86_ops.get_cpl(vcpu) == 0))
10382 		return false;
10383 
10384 	return true;
10385 }
10386 
10387 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10388 {
10389 	if (unlikely(!lapic_in_kernel(vcpu) ||
10390 		     kvm_event_needs_reinjection(vcpu) ||
10391 		     vcpu->arch.exception.pending))
10392 		return false;
10393 
10394 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10395 		return false;
10396 
10397 	/*
10398 	 * If interrupts are off we cannot even use an artificial
10399 	 * halt state.
10400 	 */
10401 	return kvm_x86_ops.interrupt_allowed(vcpu);
10402 }
10403 
10404 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10405 				     struct kvm_async_pf *work)
10406 {
10407 	struct x86_exception fault;
10408 
10409 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10410 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10411 
10412 	if (kvm_can_deliver_async_pf(vcpu) &&
10413 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10414 		fault.vector = PF_VECTOR;
10415 		fault.error_code_valid = true;
10416 		fault.error_code = 0;
10417 		fault.nested_page_fault = false;
10418 		fault.address = work->arch.token;
10419 		fault.async_page_fault = true;
10420 		kvm_inject_page_fault(vcpu, &fault);
10421 	} else {
10422 		/*
10423 		 * It is not possible to deliver a paravirtualized asynchronous
10424 		 * page fault, but putting the guest in an artificial halt state
10425 		 * can be beneficial nevertheless: if an interrupt arrives, we
10426 		 * can deliver it timely and perhaps the guest will schedule
10427 		 * another process.  When the instruction that triggered a page
10428 		 * fault is retried, hopefully the page will be ready in the host.
10429 		 */
10430 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10431 	}
10432 }
10433 
10434 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10435 				 struct kvm_async_pf *work)
10436 {
10437 	struct x86_exception fault;
10438 	u32 val;
10439 
10440 	if (work->wakeup_all)
10441 		work->arch.token = ~0; /* broadcast wakeup */
10442 	else
10443 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10444 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10445 
10446 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10447 	    !apf_get_user(vcpu, &val)) {
10448 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10449 		    vcpu->arch.exception.pending &&
10450 		    vcpu->arch.exception.nr == PF_VECTOR &&
10451 		    !apf_put_user(vcpu, 0)) {
10452 			vcpu->arch.exception.injected = false;
10453 			vcpu->arch.exception.pending = false;
10454 			vcpu->arch.exception.nr = 0;
10455 			vcpu->arch.exception.has_error_code = false;
10456 			vcpu->arch.exception.error_code = 0;
10457 			vcpu->arch.exception.has_payload = false;
10458 			vcpu->arch.exception.payload = 0;
10459 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10460 			fault.vector = PF_VECTOR;
10461 			fault.error_code_valid = true;
10462 			fault.error_code = 0;
10463 			fault.nested_page_fault = false;
10464 			fault.address = work->arch.token;
10465 			fault.async_page_fault = true;
10466 			kvm_inject_page_fault(vcpu, &fault);
10467 		}
10468 	}
10469 	vcpu->arch.apf.halted = false;
10470 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10471 }
10472 
10473 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10474 {
10475 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10476 		return true;
10477 	else
10478 		return kvm_can_do_async_pf(vcpu);
10479 }
10480 
10481 void kvm_arch_start_assignment(struct kvm *kvm)
10482 {
10483 	atomic_inc(&kvm->arch.assigned_device_count);
10484 }
10485 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10486 
10487 void kvm_arch_end_assignment(struct kvm *kvm)
10488 {
10489 	atomic_dec(&kvm->arch.assigned_device_count);
10490 }
10491 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10492 
10493 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10494 {
10495 	return atomic_read(&kvm->arch.assigned_device_count);
10496 }
10497 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10498 
10499 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10500 {
10501 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10502 }
10503 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10504 
10505 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10506 {
10507 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10508 }
10509 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10510 
10511 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10512 {
10513 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10514 }
10515 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10516 
10517 bool kvm_arch_has_irq_bypass(void)
10518 {
10519 	return true;
10520 }
10521 
10522 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10523 				      struct irq_bypass_producer *prod)
10524 {
10525 	struct kvm_kernel_irqfd *irqfd =
10526 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10527 
10528 	irqfd->producer = prod;
10529 
10530 	return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10531 					   prod->irq, irqfd->gsi, 1);
10532 }
10533 
10534 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10535 				      struct irq_bypass_producer *prod)
10536 {
10537 	int ret;
10538 	struct kvm_kernel_irqfd *irqfd =
10539 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10540 
10541 	WARN_ON(irqfd->producer != prod);
10542 	irqfd->producer = NULL;
10543 
10544 	/*
10545 	 * When producer of consumer is unregistered, we change back to
10546 	 * remapped mode, so we can re-use the current implementation
10547 	 * when the irq is masked/disabled or the consumer side (KVM
10548 	 * int this case doesn't want to receive the interrupts.
10549 	*/
10550 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10551 	if (ret)
10552 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10553 		       " fails: %d\n", irqfd->consumer.token, ret);
10554 }
10555 
10556 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10557 				   uint32_t guest_irq, bool set)
10558 {
10559 	return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10560 }
10561 
10562 bool kvm_vector_hashing_enabled(void)
10563 {
10564 	return vector_hashing;
10565 }
10566 
10567 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10568 {
10569 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10570 }
10571 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10572 
10573 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10574 {
10575 	uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10576 
10577 	/* The STIBP bit doesn't fault even if it's not advertised */
10578 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10579 	    !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10580 		bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10581 	if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10582 	    !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10583 		bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10584 
10585 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10586 	    !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10587 		bits &= ~SPEC_CTRL_SSBD;
10588 	if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10589 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10590 		bits &= ~SPEC_CTRL_SSBD;
10591 
10592 	return bits;
10593 }
10594 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10595 
10596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10599 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10600 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10601 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10602 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10603 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10604 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10605 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10606 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10607 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10608 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10609 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10610 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10611 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10612 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10613 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10614 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10615 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10616 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10617 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
10618