xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 9123e3a7)
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 "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
60 
61 #include <trace/events/kvm.h>
62 
63 #include <asm/debugreg.h>
64 #include <asm/msr.h>
65 #include <asm/desc.h>
66 #include <asm/mce.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/intel_pt.h>
75 #include <asm/emulate_prefix.h>
76 #include <clocksource/hyperv_timer.h>
77 
78 #define CREATE_TRACE_POINTS
79 #include "trace.h"
80 
81 #define MAX_IO_MSRS 256
82 #define KVM_MAX_MCE_BANKS 32
83 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
84 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
85 
86 #define emul_to_vcpu(ctxt) \
87 	((struct kvm_vcpu *)(ctxt)->vcpu)
88 
89 /* EFER defaults:
90  * - enable syscall per default because its emulated by KVM
91  * - enable LME and LMA per default on 64 bit KVM
92  */
93 #ifdef CONFIG_X86_64
94 static
95 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
96 #else
97 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
98 #endif
99 
100 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
101 
102 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
103                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
104 
105 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
106 static void process_nmi(struct kvm_vcpu *vcpu);
107 static void enter_smm(struct kvm_vcpu *vcpu);
108 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
109 static void store_regs(struct kvm_vcpu *vcpu);
110 static int sync_regs(struct kvm_vcpu *vcpu);
111 
112 struct kvm_x86_ops kvm_x86_ops __read_mostly;
113 EXPORT_SYMBOL_GPL(kvm_x86_ops);
114 
115 static bool __read_mostly ignore_msrs = 0;
116 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
117 
118 static bool __read_mostly report_ignored_msrs = true;
119 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
120 
121 unsigned int min_timer_period_us = 200;
122 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
123 
124 static bool __read_mostly kvmclock_periodic_sync = true;
125 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
126 
127 bool __read_mostly kvm_has_tsc_control;
128 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
129 u32  __read_mostly kvm_max_guest_tsc_khz;
130 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
131 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
132 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
133 u64  __read_mostly kvm_max_tsc_scaling_ratio;
134 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
135 u64 __read_mostly kvm_default_tsc_scaling_ratio;
136 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
137 
138 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
139 static u32 __read_mostly tsc_tolerance_ppm = 250;
140 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
141 
142 /*
143  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
144  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
145  * advancement entirely.  Any other value is used as-is and disables adaptive
146  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
147  */
148 static int __read_mostly lapic_timer_advance_ns = -1;
149 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
150 
151 static bool __read_mostly vector_hashing = true;
152 module_param(vector_hashing, bool, S_IRUGO);
153 
154 bool __read_mostly enable_vmware_backdoor = false;
155 module_param(enable_vmware_backdoor, bool, S_IRUGO);
156 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
157 
158 static bool __read_mostly force_emulation_prefix = false;
159 module_param(force_emulation_prefix, bool, S_IRUGO);
160 
161 int __read_mostly pi_inject_timer = -1;
162 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
163 
164 #define KVM_NR_SHARED_MSRS 16
165 
166 struct kvm_shared_msrs_global {
167 	int nr;
168 	u32 msrs[KVM_NR_SHARED_MSRS];
169 };
170 
171 struct kvm_shared_msrs {
172 	struct user_return_notifier urn;
173 	bool registered;
174 	struct kvm_shared_msr_values {
175 		u64 host;
176 		u64 curr;
177 	} values[KVM_NR_SHARED_MSRS];
178 };
179 
180 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
181 static struct kvm_shared_msrs __percpu *shared_msrs;
182 
183 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
184 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
185 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
186 				| XFEATURE_MASK_PKRU)
187 
188 u64 __read_mostly host_efer;
189 EXPORT_SYMBOL_GPL(host_efer);
190 
191 bool __read_mostly allow_smaller_maxphyaddr;
192 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
193 
194 static u64 __read_mostly host_xss;
195 u64 __read_mostly supported_xss;
196 EXPORT_SYMBOL_GPL(supported_xss);
197 
198 struct kvm_stats_debugfs_item debugfs_entries[] = {
199 	VCPU_STAT("pf_fixed", pf_fixed),
200 	VCPU_STAT("pf_guest", pf_guest),
201 	VCPU_STAT("tlb_flush", tlb_flush),
202 	VCPU_STAT("invlpg", invlpg),
203 	VCPU_STAT("exits", exits),
204 	VCPU_STAT("io_exits", io_exits),
205 	VCPU_STAT("mmio_exits", mmio_exits),
206 	VCPU_STAT("signal_exits", signal_exits),
207 	VCPU_STAT("irq_window", irq_window_exits),
208 	VCPU_STAT("nmi_window", nmi_window_exits),
209 	VCPU_STAT("halt_exits", halt_exits),
210 	VCPU_STAT("halt_successful_poll", halt_successful_poll),
211 	VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
212 	VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
213 	VCPU_STAT("halt_wakeup", halt_wakeup),
214 	VCPU_STAT("hypercalls", hypercalls),
215 	VCPU_STAT("request_irq", request_irq_exits),
216 	VCPU_STAT("irq_exits", irq_exits),
217 	VCPU_STAT("host_state_reload", host_state_reload),
218 	VCPU_STAT("fpu_reload", fpu_reload),
219 	VCPU_STAT("insn_emulation", insn_emulation),
220 	VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
221 	VCPU_STAT("irq_injections", irq_injections),
222 	VCPU_STAT("nmi_injections", nmi_injections),
223 	VCPU_STAT("req_event", req_event),
224 	VCPU_STAT("l1d_flush", l1d_flush),
225 	VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
226 	VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
227 	VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
228 	VM_STAT("mmu_pte_write", mmu_pte_write),
229 	VM_STAT("mmu_pte_updated", mmu_pte_updated),
230 	VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
231 	VM_STAT("mmu_flooded", mmu_flooded),
232 	VM_STAT("mmu_recycled", mmu_recycled),
233 	VM_STAT("mmu_cache_miss", mmu_cache_miss),
234 	VM_STAT("mmu_unsync", mmu_unsync),
235 	VM_STAT("remote_tlb_flush", remote_tlb_flush),
236 	VM_STAT("largepages", lpages, .mode = 0444),
237 	VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
238 	VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
239 	{ NULL }
240 };
241 
242 u64 __read_mostly host_xcr0;
243 u64 __read_mostly supported_xcr0;
244 EXPORT_SYMBOL_GPL(supported_xcr0);
245 
246 static struct kmem_cache *x86_fpu_cache;
247 
248 static struct kmem_cache *x86_emulator_cache;
249 
250 /*
251  * When called, it means the previous get/set msr reached an invalid msr.
252  * Return 0 if we want to ignore/silent this failed msr access, or 1 if we want
253  * to fail the caller.
254  */
255 static int kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
256 				 u64 data, bool write)
257 {
258 	const char *op = write ? "wrmsr" : "rdmsr";
259 
260 	if (ignore_msrs) {
261 		if (report_ignored_msrs)
262 			vcpu_unimpl(vcpu, "ignored %s: 0x%x data 0x%llx\n",
263 				    op, msr, data);
264 		/* Mask the error */
265 		return 0;
266 	} else {
267 		vcpu_debug_ratelimited(vcpu, "unhandled %s: 0x%x data 0x%llx\n",
268 				       op, msr, data);
269 		return 1;
270 	}
271 }
272 
273 static struct kmem_cache *kvm_alloc_emulator_cache(void)
274 {
275 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
276 	unsigned int size = sizeof(struct x86_emulate_ctxt);
277 
278 	return kmem_cache_create_usercopy("x86_emulator", size,
279 					  __alignof__(struct x86_emulate_ctxt),
280 					  SLAB_ACCOUNT, useroffset,
281 					  size - useroffset, NULL);
282 }
283 
284 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
285 
286 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
287 {
288 	int i;
289 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
290 		vcpu->arch.apf.gfns[i] = ~0;
291 }
292 
293 static void kvm_on_user_return(struct user_return_notifier *urn)
294 {
295 	unsigned slot;
296 	struct kvm_shared_msrs *locals
297 		= container_of(urn, struct kvm_shared_msrs, urn);
298 	struct kvm_shared_msr_values *values;
299 	unsigned long flags;
300 
301 	/*
302 	 * Disabling irqs at this point since the following code could be
303 	 * interrupted and executed through kvm_arch_hardware_disable()
304 	 */
305 	local_irq_save(flags);
306 	if (locals->registered) {
307 		locals->registered = false;
308 		user_return_notifier_unregister(urn);
309 	}
310 	local_irq_restore(flags);
311 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
312 		values = &locals->values[slot];
313 		if (values->host != values->curr) {
314 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
315 			values->curr = values->host;
316 		}
317 	}
318 }
319 
320 void kvm_define_shared_msr(unsigned slot, u32 msr)
321 {
322 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
323 	shared_msrs_global.msrs[slot] = msr;
324 	if (slot >= shared_msrs_global.nr)
325 		shared_msrs_global.nr = slot + 1;
326 }
327 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
328 
329 static void kvm_shared_msr_cpu_online(void)
330 {
331 	unsigned int cpu = smp_processor_id();
332 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
333 	u64 value;
334 	int i;
335 
336 	for (i = 0; i < shared_msrs_global.nr; ++i) {
337 		rdmsrl_safe(shared_msrs_global.msrs[i], &value);
338 		smsr->values[i].host = value;
339 		smsr->values[i].curr = value;
340 	}
341 }
342 
343 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
344 {
345 	unsigned int cpu = smp_processor_id();
346 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
347 	int err;
348 
349 	value = (value & mask) | (smsr->values[slot].host & ~mask);
350 	if (value == smsr->values[slot].curr)
351 		return 0;
352 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
353 	if (err)
354 		return 1;
355 
356 	smsr->values[slot].curr = value;
357 	if (!smsr->registered) {
358 		smsr->urn.on_user_return = kvm_on_user_return;
359 		user_return_notifier_register(&smsr->urn);
360 		smsr->registered = true;
361 	}
362 	return 0;
363 }
364 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
365 
366 static void drop_user_return_notifiers(void)
367 {
368 	unsigned int cpu = smp_processor_id();
369 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
370 
371 	if (smsr->registered)
372 		kvm_on_user_return(&smsr->urn);
373 }
374 
375 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
376 {
377 	return vcpu->arch.apic_base;
378 }
379 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
380 
381 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
382 {
383 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
384 }
385 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
386 
387 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
388 {
389 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
390 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
391 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
392 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
393 
394 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
395 		return 1;
396 	if (!msr_info->host_initiated) {
397 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
398 			return 1;
399 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
400 			return 1;
401 	}
402 
403 	kvm_lapic_set_base(vcpu, msr_info->data);
404 	kvm_recalculate_apic_map(vcpu->kvm);
405 	return 0;
406 }
407 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
408 
409 asmlinkage __visible noinstr void kvm_spurious_fault(void)
410 {
411 	/* Fault while not rebooting.  We want the trace. */
412 	BUG_ON(!kvm_rebooting);
413 }
414 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
415 
416 #define EXCPT_BENIGN		0
417 #define EXCPT_CONTRIBUTORY	1
418 #define EXCPT_PF		2
419 
420 static int exception_class(int vector)
421 {
422 	switch (vector) {
423 	case PF_VECTOR:
424 		return EXCPT_PF;
425 	case DE_VECTOR:
426 	case TS_VECTOR:
427 	case NP_VECTOR:
428 	case SS_VECTOR:
429 	case GP_VECTOR:
430 		return EXCPT_CONTRIBUTORY;
431 	default:
432 		break;
433 	}
434 	return EXCPT_BENIGN;
435 }
436 
437 #define EXCPT_FAULT		0
438 #define EXCPT_TRAP		1
439 #define EXCPT_ABORT		2
440 #define EXCPT_INTERRUPT		3
441 
442 static int exception_type(int vector)
443 {
444 	unsigned int mask;
445 
446 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
447 		return EXCPT_INTERRUPT;
448 
449 	mask = 1 << vector;
450 
451 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
452 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
453 		return EXCPT_TRAP;
454 
455 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
456 		return EXCPT_ABORT;
457 
458 	/* Reserved exceptions will result in fault */
459 	return EXCPT_FAULT;
460 }
461 
462 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
463 {
464 	unsigned nr = vcpu->arch.exception.nr;
465 	bool has_payload = vcpu->arch.exception.has_payload;
466 	unsigned long payload = vcpu->arch.exception.payload;
467 
468 	if (!has_payload)
469 		return;
470 
471 	switch (nr) {
472 	case DB_VECTOR:
473 		/*
474 		 * "Certain debug exceptions may clear bit 0-3.  The
475 		 * remaining contents of the DR6 register are never
476 		 * cleared by the processor".
477 		 */
478 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
479 		/*
480 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
481 		 */
482 		vcpu->arch.dr6 |= DR6_RTM;
483 		vcpu->arch.dr6 |= payload;
484 		/*
485 		 * Bit 16 should be set in the payload whenever the #DB
486 		 * exception should clear DR6.RTM. This makes the payload
487 		 * compatible with the pending debug exceptions under VMX.
488 		 * Though not currently documented in the SDM, this also
489 		 * makes the payload compatible with the exit qualification
490 		 * for #DB exceptions under VMX.
491 		 */
492 		vcpu->arch.dr6 ^= payload & DR6_RTM;
493 
494 		/*
495 		 * The #DB payload is defined as compatible with the 'pending
496 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
497 		 * defined in the 'pending debug exceptions' field (enabled
498 		 * breakpoint), it is reserved and must be zero in DR6.
499 		 */
500 		vcpu->arch.dr6 &= ~BIT(12);
501 		break;
502 	case PF_VECTOR:
503 		vcpu->arch.cr2 = payload;
504 		break;
505 	}
506 
507 	vcpu->arch.exception.has_payload = false;
508 	vcpu->arch.exception.payload = 0;
509 }
510 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
511 
512 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
513 		unsigned nr, bool has_error, u32 error_code,
514 	        bool has_payload, unsigned long payload, bool reinject)
515 {
516 	u32 prev_nr;
517 	int class1, class2;
518 
519 	kvm_make_request(KVM_REQ_EVENT, vcpu);
520 
521 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
522 	queue:
523 		if (has_error && !is_protmode(vcpu))
524 			has_error = false;
525 		if (reinject) {
526 			/*
527 			 * On vmentry, vcpu->arch.exception.pending is only
528 			 * true if an event injection was blocked by
529 			 * nested_run_pending.  In that case, however,
530 			 * vcpu_enter_guest requests an immediate exit,
531 			 * and the guest shouldn't proceed far enough to
532 			 * need reinjection.
533 			 */
534 			WARN_ON_ONCE(vcpu->arch.exception.pending);
535 			vcpu->arch.exception.injected = true;
536 			if (WARN_ON_ONCE(has_payload)) {
537 				/*
538 				 * A reinjected event has already
539 				 * delivered its payload.
540 				 */
541 				has_payload = false;
542 				payload = 0;
543 			}
544 		} else {
545 			vcpu->arch.exception.pending = true;
546 			vcpu->arch.exception.injected = false;
547 		}
548 		vcpu->arch.exception.has_error_code = has_error;
549 		vcpu->arch.exception.nr = nr;
550 		vcpu->arch.exception.error_code = error_code;
551 		vcpu->arch.exception.has_payload = has_payload;
552 		vcpu->arch.exception.payload = payload;
553 		if (!is_guest_mode(vcpu))
554 			kvm_deliver_exception_payload(vcpu);
555 		return;
556 	}
557 
558 	/* to check exception */
559 	prev_nr = vcpu->arch.exception.nr;
560 	if (prev_nr == DF_VECTOR) {
561 		/* triple fault -> shutdown */
562 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
563 		return;
564 	}
565 	class1 = exception_class(prev_nr);
566 	class2 = exception_class(nr);
567 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
568 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
569 		/*
570 		 * Generate double fault per SDM Table 5-5.  Set
571 		 * exception.pending = true so that the double fault
572 		 * can trigger a nested vmexit.
573 		 */
574 		vcpu->arch.exception.pending = true;
575 		vcpu->arch.exception.injected = false;
576 		vcpu->arch.exception.has_error_code = true;
577 		vcpu->arch.exception.nr = DF_VECTOR;
578 		vcpu->arch.exception.error_code = 0;
579 		vcpu->arch.exception.has_payload = false;
580 		vcpu->arch.exception.payload = 0;
581 	} else
582 		/* replace previous exception with a new one in a hope
583 		   that instruction re-execution will regenerate lost
584 		   exception */
585 		goto queue;
586 }
587 
588 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
589 {
590 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
591 }
592 EXPORT_SYMBOL_GPL(kvm_queue_exception);
593 
594 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
595 {
596 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
597 }
598 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
599 
600 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
601 			   unsigned long payload)
602 {
603 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
604 }
605 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
606 
607 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
608 				    u32 error_code, unsigned long payload)
609 {
610 	kvm_multiple_exception(vcpu, nr, true, error_code,
611 			       true, payload, false);
612 }
613 
614 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
615 {
616 	if (err)
617 		kvm_inject_gp(vcpu, 0);
618 	else
619 		return kvm_skip_emulated_instruction(vcpu);
620 
621 	return 1;
622 }
623 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
624 
625 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
626 {
627 	++vcpu->stat.pf_guest;
628 	vcpu->arch.exception.nested_apf =
629 		is_guest_mode(vcpu) && fault->async_page_fault;
630 	if (vcpu->arch.exception.nested_apf) {
631 		vcpu->arch.apf.nested_apf_token = fault->address;
632 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
633 	} else {
634 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
635 					fault->address);
636 	}
637 }
638 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
639 
640 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
641 				    struct x86_exception *fault)
642 {
643 	struct kvm_mmu *fault_mmu;
644 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
645 
646 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
647 					       vcpu->arch.walk_mmu;
648 
649 	/*
650 	 * Invalidate the TLB entry for the faulting address, if it exists,
651 	 * else the access will fault indefinitely (and to emulate hardware).
652 	 */
653 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
654 	    !(fault->error_code & PFERR_RSVD_MASK))
655 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
656 				       fault_mmu->root_hpa);
657 
658 	fault_mmu->inject_page_fault(vcpu, fault);
659 	return fault->nested_page_fault;
660 }
661 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
662 
663 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
664 {
665 	atomic_inc(&vcpu->arch.nmi_queued);
666 	kvm_make_request(KVM_REQ_NMI, vcpu);
667 }
668 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
669 
670 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
671 {
672 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
673 }
674 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
675 
676 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
677 {
678 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
679 }
680 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
681 
682 /*
683  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
684  * a #GP and return false.
685  */
686 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
687 {
688 	if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
689 		return true;
690 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
691 	return false;
692 }
693 EXPORT_SYMBOL_GPL(kvm_require_cpl);
694 
695 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
696 {
697 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
698 		return true;
699 
700 	kvm_queue_exception(vcpu, UD_VECTOR);
701 	return false;
702 }
703 EXPORT_SYMBOL_GPL(kvm_require_dr);
704 
705 /*
706  * This function will be used to read from the physical memory of the currently
707  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
708  * can read from guest physical or from the guest's guest physical memory.
709  */
710 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
711 			    gfn_t ngfn, void *data, int offset, int len,
712 			    u32 access)
713 {
714 	struct x86_exception exception;
715 	gfn_t real_gfn;
716 	gpa_t ngpa;
717 
718 	ngpa     = gfn_to_gpa(ngfn);
719 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
720 	if (real_gfn == UNMAPPED_GVA)
721 		return -EFAULT;
722 
723 	real_gfn = gpa_to_gfn(real_gfn);
724 
725 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
726 }
727 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
728 
729 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
730 			       void *data, int offset, int len, u32 access)
731 {
732 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
733 				       data, offset, len, access);
734 }
735 
736 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
737 {
738 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
739 	       rsvd_bits(1, 2);
740 }
741 
742 /*
743  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
744  */
745 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
746 {
747 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
748 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
749 	int i;
750 	int ret;
751 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
752 
753 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
754 				      offset * sizeof(u64), sizeof(pdpte),
755 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
756 	if (ret < 0) {
757 		ret = 0;
758 		goto out;
759 	}
760 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
761 		if ((pdpte[i] & PT_PRESENT_MASK) &&
762 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
763 			ret = 0;
764 			goto out;
765 		}
766 	}
767 	ret = 1;
768 
769 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
770 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
771 
772 out:
773 
774 	return ret;
775 }
776 EXPORT_SYMBOL_GPL(load_pdptrs);
777 
778 bool pdptrs_changed(struct kvm_vcpu *vcpu)
779 {
780 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
781 	int offset;
782 	gfn_t gfn;
783 	int r;
784 
785 	if (!is_pae_paging(vcpu))
786 		return false;
787 
788 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
789 		return true;
790 
791 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
792 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
793 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
794 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
795 	if (r < 0)
796 		return true;
797 
798 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
799 }
800 EXPORT_SYMBOL_GPL(pdptrs_changed);
801 
802 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
803 {
804 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
805 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
806 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
807 
808 	cr0 |= X86_CR0_ET;
809 
810 #ifdef CONFIG_X86_64
811 	if (cr0 & 0xffffffff00000000UL)
812 		return 1;
813 #endif
814 
815 	cr0 &= ~CR0_RESERVED_BITS;
816 
817 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
818 		return 1;
819 
820 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
821 		return 1;
822 
823 #ifdef CONFIG_X86_64
824 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
825 	    (cr0 & X86_CR0_PG)) {
826 		int cs_db, cs_l;
827 
828 		if (!is_pae(vcpu))
829 			return 1;
830 		kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
831 		if (cs_l)
832 			return 1;
833 	}
834 #endif
835 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
836 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
837 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
838 		return 1;
839 
840 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
841 		return 1;
842 
843 	kvm_x86_ops.set_cr0(vcpu, cr0);
844 
845 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
846 		kvm_clear_async_pf_completion_queue(vcpu);
847 		kvm_async_pf_hash_reset(vcpu);
848 	}
849 
850 	if ((cr0 ^ old_cr0) & update_bits)
851 		kvm_mmu_reset_context(vcpu);
852 
853 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
854 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
855 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
856 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
857 
858 	return 0;
859 }
860 EXPORT_SYMBOL_GPL(kvm_set_cr0);
861 
862 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
863 {
864 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
865 }
866 EXPORT_SYMBOL_GPL(kvm_lmsw);
867 
868 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
869 {
870 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
871 
872 		if (vcpu->arch.xcr0 != host_xcr0)
873 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
874 
875 		if (vcpu->arch.xsaves_enabled &&
876 		    vcpu->arch.ia32_xss != host_xss)
877 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
878 	}
879 
880 	if (static_cpu_has(X86_FEATURE_PKU) &&
881 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
882 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
883 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
884 		__write_pkru(vcpu->arch.pkru);
885 }
886 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
887 
888 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
889 {
890 	if (static_cpu_has(X86_FEATURE_PKU) &&
891 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
892 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
893 		vcpu->arch.pkru = rdpkru();
894 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
895 			__write_pkru(vcpu->arch.host_pkru);
896 	}
897 
898 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
899 
900 		if (vcpu->arch.xcr0 != host_xcr0)
901 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
902 
903 		if (vcpu->arch.xsaves_enabled &&
904 		    vcpu->arch.ia32_xss != host_xss)
905 			wrmsrl(MSR_IA32_XSS, host_xss);
906 	}
907 
908 }
909 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
910 
911 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
912 {
913 	u64 xcr0 = xcr;
914 	u64 old_xcr0 = vcpu->arch.xcr0;
915 	u64 valid_bits;
916 
917 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
918 	if (index != XCR_XFEATURE_ENABLED_MASK)
919 		return 1;
920 	if (!(xcr0 & XFEATURE_MASK_FP))
921 		return 1;
922 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
923 		return 1;
924 
925 	/*
926 	 * Do not allow the guest to set bits that we do not support
927 	 * saving.  However, xcr0 bit 0 is always set, even if the
928 	 * emulated CPU does not support XSAVE (see fx_init).
929 	 */
930 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
931 	if (xcr0 & ~valid_bits)
932 		return 1;
933 
934 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
935 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
936 		return 1;
937 
938 	if (xcr0 & XFEATURE_MASK_AVX512) {
939 		if (!(xcr0 & XFEATURE_MASK_YMM))
940 			return 1;
941 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
942 			return 1;
943 	}
944 	vcpu->arch.xcr0 = xcr0;
945 
946 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
947 		kvm_update_cpuid_runtime(vcpu);
948 	return 0;
949 }
950 
951 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
952 {
953 	if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
954 	    __kvm_set_xcr(vcpu, index, xcr)) {
955 		kvm_inject_gp(vcpu, 0);
956 		return 1;
957 	}
958 	return 0;
959 }
960 EXPORT_SYMBOL_GPL(kvm_set_xcr);
961 
962 int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
963 {
964 	if (cr4 & cr4_reserved_bits)
965 		return -EINVAL;
966 
967 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
968 		return -EINVAL;
969 
970 	return 0;
971 }
972 EXPORT_SYMBOL_GPL(kvm_valid_cr4);
973 
974 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
975 {
976 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
977 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
978 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
979 
980 	if (kvm_valid_cr4(vcpu, cr4))
981 		return 1;
982 
983 	if (is_long_mode(vcpu)) {
984 		if (!(cr4 & X86_CR4_PAE))
985 			return 1;
986 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
987 			return 1;
988 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
989 		   && ((cr4 ^ old_cr4) & pdptr_bits)
990 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
991 				   kvm_read_cr3(vcpu)))
992 		return 1;
993 
994 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
995 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
996 			return 1;
997 
998 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
999 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1000 			return 1;
1001 	}
1002 
1003 	if (kvm_x86_ops.set_cr4(vcpu, cr4))
1004 		return 1;
1005 
1006 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
1007 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1008 		kvm_mmu_reset_context(vcpu);
1009 
1010 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1011 		kvm_update_cpuid_runtime(vcpu);
1012 
1013 	return 0;
1014 }
1015 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1016 
1017 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1018 {
1019 	bool skip_tlb_flush = false;
1020 #ifdef CONFIG_X86_64
1021 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1022 
1023 	if (pcid_enabled) {
1024 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1025 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1026 	}
1027 #endif
1028 
1029 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1030 		if (!skip_tlb_flush) {
1031 			kvm_mmu_sync_roots(vcpu);
1032 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1033 		}
1034 		return 0;
1035 	}
1036 
1037 	if (is_long_mode(vcpu) &&
1038 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1039 		return 1;
1040 	else if (is_pae_paging(vcpu) &&
1041 		 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1042 		return 1;
1043 
1044 	kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1045 	vcpu->arch.cr3 = cr3;
1046 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1047 
1048 	return 0;
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1051 
1052 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1053 {
1054 	if (cr8 & CR8_RESERVED_BITS)
1055 		return 1;
1056 	if (lapic_in_kernel(vcpu))
1057 		kvm_lapic_set_tpr(vcpu, cr8);
1058 	else
1059 		vcpu->arch.cr8 = cr8;
1060 	return 0;
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1063 
1064 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1065 {
1066 	if (lapic_in_kernel(vcpu))
1067 		return kvm_lapic_get_cr8(vcpu);
1068 	else
1069 		return vcpu->arch.cr8;
1070 }
1071 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1072 
1073 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1074 {
1075 	int i;
1076 
1077 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1078 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1079 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1080 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1081 	}
1082 }
1083 
1084 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1085 {
1086 	unsigned long dr7;
1087 
1088 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1089 		dr7 = vcpu->arch.guest_debug_dr7;
1090 	else
1091 		dr7 = vcpu->arch.dr7;
1092 	kvm_x86_ops.set_dr7(vcpu, dr7);
1093 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1094 	if (dr7 & DR7_BP_EN_MASK)
1095 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1096 }
1097 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1098 
1099 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1100 {
1101 	u64 fixed = DR6_FIXED_1;
1102 
1103 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1104 		fixed |= DR6_RTM;
1105 	return fixed;
1106 }
1107 
1108 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1109 {
1110 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1111 
1112 	switch (dr) {
1113 	case 0 ... 3:
1114 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1115 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1116 			vcpu->arch.eff_db[dr] = val;
1117 		break;
1118 	case 4:
1119 		/* fall through */
1120 	case 6:
1121 		if (!kvm_dr6_valid(val))
1122 			return -1; /* #GP */
1123 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1124 		break;
1125 	case 5:
1126 		/* fall through */
1127 	default: /* 7 */
1128 		if (!kvm_dr7_valid(val))
1129 			return -1; /* #GP */
1130 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1131 		kvm_update_dr7(vcpu);
1132 		break;
1133 	}
1134 
1135 	return 0;
1136 }
1137 
1138 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1139 {
1140 	if (__kvm_set_dr(vcpu, dr, val)) {
1141 		kvm_inject_gp(vcpu, 0);
1142 		return 1;
1143 	}
1144 	return 0;
1145 }
1146 EXPORT_SYMBOL_GPL(kvm_set_dr);
1147 
1148 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1149 {
1150 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1151 
1152 	switch (dr) {
1153 	case 0 ... 3:
1154 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1155 		break;
1156 	case 4:
1157 		/* fall through */
1158 	case 6:
1159 		*val = vcpu->arch.dr6;
1160 		break;
1161 	case 5:
1162 		/* fall through */
1163 	default: /* 7 */
1164 		*val = vcpu->arch.dr7;
1165 		break;
1166 	}
1167 	return 0;
1168 }
1169 EXPORT_SYMBOL_GPL(kvm_get_dr);
1170 
1171 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1172 {
1173 	u32 ecx = kvm_rcx_read(vcpu);
1174 	u64 data;
1175 	int err;
1176 
1177 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1178 	if (err)
1179 		return err;
1180 	kvm_rax_write(vcpu, (u32)data);
1181 	kvm_rdx_write(vcpu, data >> 32);
1182 	return err;
1183 }
1184 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1185 
1186 /*
1187  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1188  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1189  *
1190  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1191  * extract the supported MSRs from the related const lists.
1192  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1193  * capabilities of the host cpu. This capabilities test skips MSRs that are
1194  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1195  * may depend on host virtualization features rather than host cpu features.
1196  */
1197 
1198 static const u32 msrs_to_save_all[] = {
1199 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1200 	MSR_STAR,
1201 #ifdef CONFIG_X86_64
1202 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1203 #endif
1204 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1205 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1206 	MSR_IA32_SPEC_CTRL,
1207 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1208 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1209 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1210 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1211 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1212 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1213 	MSR_IA32_UMWAIT_CONTROL,
1214 
1215 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1216 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1217 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1218 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1219 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1220 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1221 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1222 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1223 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1224 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1225 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1226 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1227 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1228 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1229 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1230 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1231 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1232 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1233 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1234 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1235 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1236 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1237 };
1238 
1239 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1240 static unsigned num_msrs_to_save;
1241 
1242 static const u32 emulated_msrs_all[] = {
1243 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1244 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1245 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1246 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1247 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1248 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1249 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1250 	HV_X64_MSR_RESET,
1251 	HV_X64_MSR_VP_INDEX,
1252 	HV_X64_MSR_VP_RUNTIME,
1253 	HV_X64_MSR_SCONTROL,
1254 	HV_X64_MSR_STIMER0_CONFIG,
1255 	HV_X64_MSR_VP_ASSIST_PAGE,
1256 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1257 	HV_X64_MSR_TSC_EMULATION_STATUS,
1258 	HV_X64_MSR_SYNDBG_OPTIONS,
1259 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1260 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1261 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1262 
1263 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1264 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1265 
1266 	MSR_IA32_TSC_ADJUST,
1267 	MSR_IA32_TSCDEADLINE,
1268 	MSR_IA32_ARCH_CAPABILITIES,
1269 	MSR_IA32_PERF_CAPABILITIES,
1270 	MSR_IA32_MISC_ENABLE,
1271 	MSR_IA32_MCG_STATUS,
1272 	MSR_IA32_MCG_CTL,
1273 	MSR_IA32_MCG_EXT_CTL,
1274 	MSR_IA32_SMBASE,
1275 	MSR_SMI_COUNT,
1276 	MSR_PLATFORM_INFO,
1277 	MSR_MISC_FEATURES_ENABLES,
1278 	MSR_AMD64_VIRT_SPEC_CTRL,
1279 	MSR_IA32_POWER_CTL,
1280 	MSR_IA32_UCODE_REV,
1281 
1282 	/*
1283 	 * The following list leaves out MSRs whose values are determined
1284 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1285 	 * We always support the "true" VMX control MSRs, even if the host
1286 	 * processor does not, so I am putting these registers here rather
1287 	 * than in msrs_to_save_all.
1288 	 */
1289 	MSR_IA32_VMX_BASIC,
1290 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1291 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1292 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1293 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1294 	MSR_IA32_VMX_MISC,
1295 	MSR_IA32_VMX_CR0_FIXED0,
1296 	MSR_IA32_VMX_CR4_FIXED0,
1297 	MSR_IA32_VMX_VMCS_ENUM,
1298 	MSR_IA32_VMX_PROCBASED_CTLS2,
1299 	MSR_IA32_VMX_EPT_VPID_CAP,
1300 	MSR_IA32_VMX_VMFUNC,
1301 
1302 	MSR_K7_HWCR,
1303 	MSR_KVM_POLL_CONTROL,
1304 };
1305 
1306 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1307 static unsigned num_emulated_msrs;
1308 
1309 /*
1310  * List of msr numbers which are used to expose MSR-based features that
1311  * can be used by a hypervisor to validate requested CPU features.
1312  */
1313 static const u32 msr_based_features_all[] = {
1314 	MSR_IA32_VMX_BASIC,
1315 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1316 	MSR_IA32_VMX_PINBASED_CTLS,
1317 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1318 	MSR_IA32_VMX_PROCBASED_CTLS,
1319 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1320 	MSR_IA32_VMX_EXIT_CTLS,
1321 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1322 	MSR_IA32_VMX_ENTRY_CTLS,
1323 	MSR_IA32_VMX_MISC,
1324 	MSR_IA32_VMX_CR0_FIXED0,
1325 	MSR_IA32_VMX_CR0_FIXED1,
1326 	MSR_IA32_VMX_CR4_FIXED0,
1327 	MSR_IA32_VMX_CR4_FIXED1,
1328 	MSR_IA32_VMX_VMCS_ENUM,
1329 	MSR_IA32_VMX_PROCBASED_CTLS2,
1330 	MSR_IA32_VMX_EPT_VPID_CAP,
1331 	MSR_IA32_VMX_VMFUNC,
1332 
1333 	MSR_F10H_DECFG,
1334 	MSR_IA32_UCODE_REV,
1335 	MSR_IA32_ARCH_CAPABILITIES,
1336 	MSR_IA32_PERF_CAPABILITIES,
1337 };
1338 
1339 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1340 static unsigned int num_msr_based_features;
1341 
1342 static u64 kvm_get_arch_capabilities(void)
1343 {
1344 	u64 data = 0;
1345 
1346 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1347 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1348 
1349 	/*
1350 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1351 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1352 	 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1353 	 * L1 guests, so it need not worry about its own (L2) guests.
1354 	 */
1355 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1356 
1357 	/*
1358 	 * If we're doing cache flushes (either "always" or "cond")
1359 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1360 	 * If an outer hypervisor is doing the cache flush for us
1361 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1362 	 * capability to the guest too, and if EPT is disabled we're not
1363 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1364 	 * require a nested hypervisor to do a flush of its own.
1365 	 */
1366 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1367 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1368 
1369 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1370 		data |= ARCH_CAP_RDCL_NO;
1371 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1372 		data |= ARCH_CAP_SSB_NO;
1373 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1374 		data |= ARCH_CAP_MDS_NO;
1375 
1376 	/*
1377 	 * On TAA affected systems:
1378 	 *      - nothing to do if TSX is disabled on the host.
1379 	 *      - we emulate TSX_CTRL if present on the host.
1380 	 *	  This lets the guest use VERW to clear CPU buffers.
1381 	 */
1382 	if (!boot_cpu_has(X86_FEATURE_RTM))
1383 		data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1384 	else if (!boot_cpu_has_bug(X86_BUG_TAA))
1385 		data |= ARCH_CAP_TAA_NO;
1386 
1387 	return data;
1388 }
1389 
1390 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1391 {
1392 	switch (msr->index) {
1393 	case MSR_IA32_ARCH_CAPABILITIES:
1394 		msr->data = kvm_get_arch_capabilities();
1395 		break;
1396 	case MSR_IA32_UCODE_REV:
1397 		rdmsrl_safe(msr->index, &msr->data);
1398 		break;
1399 	default:
1400 		return kvm_x86_ops.get_msr_feature(msr);
1401 	}
1402 	return 0;
1403 }
1404 
1405 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1406 {
1407 	struct kvm_msr_entry msr;
1408 	int r;
1409 
1410 	msr.index = index;
1411 	r = kvm_get_msr_feature(&msr);
1412 
1413 	if (r == KVM_MSR_RET_INVALID) {
1414 		/* Unconditionally clear the output for simplicity */
1415 		*data = 0;
1416 		r = kvm_msr_ignored_check(vcpu, index, 0, false);
1417 	}
1418 
1419 	if (r)
1420 		return r;
1421 
1422 	*data = msr.data;
1423 
1424 	return 0;
1425 }
1426 
1427 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1428 {
1429 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1430 		return false;
1431 
1432 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1433 		return false;
1434 
1435 	if (efer & (EFER_LME | EFER_LMA) &&
1436 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1437 		return false;
1438 
1439 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1440 		return false;
1441 
1442 	return true;
1443 
1444 }
1445 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1446 {
1447 	if (efer & efer_reserved_bits)
1448 		return false;
1449 
1450 	return __kvm_valid_efer(vcpu, efer);
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1453 
1454 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1455 {
1456 	u64 old_efer = vcpu->arch.efer;
1457 	u64 efer = msr_info->data;
1458 
1459 	if (efer & efer_reserved_bits)
1460 		return 1;
1461 
1462 	if (!msr_info->host_initiated) {
1463 		if (!__kvm_valid_efer(vcpu, efer))
1464 			return 1;
1465 
1466 		if (is_paging(vcpu) &&
1467 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1468 			return 1;
1469 	}
1470 
1471 	efer &= ~EFER_LMA;
1472 	efer |= vcpu->arch.efer & EFER_LMA;
1473 
1474 	kvm_x86_ops.set_efer(vcpu, efer);
1475 
1476 	/* Update reserved bits */
1477 	if ((efer ^ old_efer) & EFER_NX)
1478 		kvm_mmu_reset_context(vcpu);
1479 
1480 	return 0;
1481 }
1482 
1483 void kvm_enable_efer_bits(u64 mask)
1484 {
1485        efer_reserved_bits &= ~mask;
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1488 
1489 /*
1490  * Write @data into the MSR specified by @index.  Select MSR specific fault
1491  * checks are bypassed if @host_initiated is %true.
1492  * Returns 0 on success, non-0 otherwise.
1493  * Assumes vcpu_load() was already called.
1494  */
1495 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1496 			 bool host_initiated)
1497 {
1498 	struct msr_data msr;
1499 
1500 	switch (index) {
1501 	case MSR_FS_BASE:
1502 	case MSR_GS_BASE:
1503 	case MSR_KERNEL_GS_BASE:
1504 	case MSR_CSTAR:
1505 	case MSR_LSTAR:
1506 		if (is_noncanonical_address(data, vcpu))
1507 			return 1;
1508 		break;
1509 	case MSR_IA32_SYSENTER_EIP:
1510 	case MSR_IA32_SYSENTER_ESP:
1511 		/*
1512 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1513 		 * non-canonical address is written on Intel but not on
1514 		 * AMD (which ignores the top 32-bits, because it does
1515 		 * not implement 64-bit SYSENTER).
1516 		 *
1517 		 * 64-bit code should hence be able to write a non-canonical
1518 		 * value on AMD.  Making the address canonical ensures that
1519 		 * vmentry does not fail on Intel after writing a non-canonical
1520 		 * value, and that something deterministic happens if the guest
1521 		 * invokes 64-bit SYSENTER.
1522 		 */
1523 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1524 	}
1525 
1526 	msr.data = data;
1527 	msr.index = index;
1528 	msr.host_initiated = host_initiated;
1529 
1530 	return kvm_x86_ops.set_msr(vcpu, &msr);
1531 }
1532 
1533 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1534 				     u32 index, u64 data, bool host_initiated)
1535 {
1536 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1537 
1538 	if (ret == KVM_MSR_RET_INVALID)
1539 		ret = kvm_msr_ignored_check(vcpu, index, data, true);
1540 
1541 	return ret;
1542 }
1543 
1544 /*
1545  * Read the MSR specified by @index into @data.  Select MSR specific fault
1546  * checks are bypassed if @host_initiated is %true.
1547  * Returns 0 on success, non-0 otherwise.
1548  * Assumes vcpu_load() was already called.
1549  */
1550 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1551 		  bool host_initiated)
1552 {
1553 	struct msr_data msr;
1554 	int ret;
1555 
1556 	msr.index = index;
1557 	msr.host_initiated = host_initiated;
1558 
1559 	ret = kvm_x86_ops.get_msr(vcpu, &msr);
1560 	if (!ret)
1561 		*data = msr.data;
1562 	return ret;
1563 }
1564 
1565 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1566 				     u32 index, u64 *data, bool host_initiated)
1567 {
1568 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1569 
1570 	if (ret == KVM_MSR_RET_INVALID) {
1571 		/* Unconditionally clear *data for simplicity */
1572 		*data = 0;
1573 		ret = kvm_msr_ignored_check(vcpu, index, 0, false);
1574 	}
1575 
1576 	return ret;
1577 }
1578 
1579 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1580 {
1581 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1582 }
1583 EXPORT_SYMBOL_GPL(kvm_get_msr);
1584 
1585 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1586 {
1587 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1588 }
1589 EXPORT_SYMBOL_GPL(kvm_set_msr);
1590 
1591 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1592 {
1593 	u32 ecx = kvm_rcx_read(vcpu);
1594 	u64 data;
1595 
1596 	if (kvm_get_msr(vcpu, ecx, &data)) {
1597 		trace_kvm_msr_read_ex(ecx);
1598 		kvm_inject_gp(vcpu, 0);
1599 		return 1;
1600 	}
1601 
1602 	trace_kvm_msr_read(ecx, data);
1603 
1604 	kvm_rax_write(vcpu, data & -1u);
1605 	kvm_rdx_write(vcpu, (data >> 32) & -1u);
1606 	return kvm_skip_emulated_instruction(vcpu);
1607 }
1608 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1609 
1610 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1611 {
1612 	u32 ecx = kvm_rcx_read(vcpu);
1613 	u64 data = kvm_read_edx_eax(vcpu);
1614 
1615 	if (kvm_set_msr(vcpu, ecx, data)) {
1616 		trace_kvm_msr_write_ex(ecx, data);
1617 		kvm_inject_gp(vcpu, 0);
1618 		return 1;
1619 	}
1620 
1621 	trace_kvm_msr_write(ecx, data);
1622 	return kvm_skip_emulated_instruction(vcpu);
1623 }
1624 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1625 
1626 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1627 {
1628 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1629 		xfer_to_guest_mode_work_pending();
1630 }
1631 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1632 
1633 /*
1634  * The fast path for frequent and performance sensitive wrmsr emulation,
1635  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1636  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1637  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1638  * other cases which must be called after interrupts are enabled on the host.
1639  */
1640 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1641 {
1642 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1643 		return 1;
1644 
1645 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1646 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1647 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1648 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1649 
1650 		data &= ~(1 << 12);
1651 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1652 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1653 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1654 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1655 		return 0;
1656 	}
1657 
1658 	return 1;
1659 }
1660 
1661 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1662 {
1663 	if (!kvm_can_use_hv_timer(vcpu))
1664 		return 1;
1665 
1666 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1667 	return 0;
1668 }
1669 
1670 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1671 {
1672 	u32 msr = kvm_rcx_read(vcpu);
1673 	u64 data;
1674 	fastpath_t ret = EXIT_FASTPATH_NONE;
1675 
1676 	switch (msr) {
1677 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1678 		data = kvm_read_edx_eax(vcpu);
1679 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1680 			kvm_skip_emulated_instruction(vcpu);
1681 			ret = EXIT_FASTPATH_EXIT_HANDLED;
1682 		}
1683 		break;
1684 	case MSR_IA32_TSCDEADLINE:
1685 		data = kvm_read_edx_eax(vcpu);
1686 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1687 			kvm_skip_emulated_instruction(vcpu);
1688 			ret = EXIT_FASTPATH_REENTER_GUEST;
1689 		}
1690 		break;
1691 	default:
1692 		break;
1693 	}
1694 
1695 	if (ret != EXIT_FASTPATH_NONE)
1696 		trace_kvm_msr_write(msr, data);
1697 
1698 	return ret;
1699 }
1700 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1701 
1702 /*
1703  * Adapt set_msr() to msr_io()'s calling convention
1704  */
1705 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1706 {
1707 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
1708 }
1709 
1710 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1711 {
1712 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1713 }
1714 
1715 #ifdef CONFIG_X86_64
1716 struct pvclock_clock {
1717 	int vclock_mode;
1718 	u64 cycle_last;
1719 	u64 mask;
1720 	u32 mult;
1721 	u32 shift;
1722 	u64 base_cycles;
1723 	u64 offset;
1724 };
1725 
1726 struct pvclock_gtod_data {
1727 	seqcount_t	seq;
1728 
1729 	struct pvclock_clock clock; /* extract of a clocksource struct */
1730 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1731 
1732 	ktime_t		offs_boot;
1733 	u64		wall_time_sec;
1734 };
1735 
1736 static struct pvclock_gtod_data pvclock_gtod_data;
1737 
1738 static void update_pvclock_gtod(struct timekeeper *tk)
1739 {
1740 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1741 
1742 	write_seqcount_begin(&vdata->seq);
1743 
1744 	/* copy pvclock gtod data */
1745 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
1746 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1747 	vdata->clock.mask		= tk->tkr_mono.mask;
1748 	vdata->clock.mult		= tk->tkr_mono.mult;
1749 	vdata->clock.shift		= tk->tkr_mono.shift;
1750 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
1751 	vdata->clock.offset		= tk->tkr_mono.base;
1752 
1753 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
1754 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
1755 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
1756 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
1757 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
1758 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
1759 	vdata->raw_clock.offset		= tk->tkr_raw.base;
1760 
1761 	vdata->wall_time_sec            = tk->xtime_sec;
1762 
1763 	vdata->offs_boot		= tk->offs_boot;
1764 
1765 	write_seqcount_end(&vdata->seq);
1766 }
1767 
1768 static s64 get_kvmclock_base_ns(void)
1769 {
1770 	/* Count up from boot time, but with the frequency of the raw clock.  */
1771 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1772 }
1773 #else
1774 static s64 get_kvmclock_base_ns(void)
1775 {
1776 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
1777 	return ktime_get_boottime_ns();
1778 }
1779 #endif
1780 
1781 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1782 {
1783 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1784 	kvm_vcpu_kick(vcpu);
1785 }
1786 
1787 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1788 {
1789 	int version;
1790 	int r;
1791 	struct pvclock_wall_clock wc;
1792 	u64 wall_nsec;
1793 
1794 	if (!wall_clock)
1795 		return;
1796 
1797 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1798 	if (r)
1799 		return;
1800 
1801 	if (version & 1)
1802 		++version;  /* first time write, random junk */
1803 
1804 	++version;
1805 
1806 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1807 		return;
1808 
1809 	/*
1810 	 * The guest calculates current wall clock time by adding
1811 	 * system time (updated by kvm_guest_time_update below) to the
1812 	 * wall clock specified here.  We do the reverse here.
1813 	 */
1814 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1815 
1816 	wc.nsec = do_div(wall_nsec, 1000000000);
1817 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1818 	wc.version = version;
1819 
1820 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1821 
1822 	version++;
1823 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1824 }
1825 
1826 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1827 {
1828 	do_shl32_div32(dividend, divisor);
1829 	return dividend;
1830 }
1831 
1832 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1833 			       s8 *pshift, u32 *pmultiplier)
1834 {
1835 	uint64_t scaled64;
1836 	int32_t  shift = 0;
1837 	uint64_t tps64;
1838 	uint32_t tps32;
1839 
1840 	tps64 = base_hz;
1841 	scaled64 = scaled_hz;
1842 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1843 		tps64 >>= 1;
1844 		shift--;
1845 	}
1846 
1847 	tps32 = (uint32_t)tps64;
1848 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1849 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1850 			scaled64 >>= 1;
1851 		else
1852 			tps32 <<= 1;
1853 		shift++;
1854 	}
1855 
1856 	*pshift = shift;
1857 	*pmultiplier = div_frac(scaled64, tps32);
1858 }
1859 
1860 #ifdef CONFIG_X86_64
1861 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1862 #endif
1863 
1864 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1865 static unsigned long max_tsc_khz;
1866 
1867 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1868 {
1869 	u64 v = (u64)khz * (1000000 + ppm);
1870 	do_div(v, 1000000);
1871 	return v;
1872 }
1873 
1874 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1875 {
1876 	u64 ratio;
1877 
1878 	/* Guest TSC same frequency as host TSC? */
1879 	if (!scale) {
1880 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1881 		return 0;
1882 	}
1883 
1884 	/* TSC scaling supported? */
1885 	if (!kvm_has_tsc_control) {
1886 		if (user_tsc_khz > tsc_khz) {
1887 			vcpu->arch.tsc_catchup = 1;
1888 			vcpu->arch.tsc_always_catchup = 1;
1889 			return 0;
1890 		} else {
1891 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1892 			return -1;
1893 		}
1894 	}
1895 
1896 	/* TSC scaling required  - calculate ratio */
1897 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1898 				user_tsc_khz, tsc_khz);
1899 
1900 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1901 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1902 			            user_tsc_khz);
1903 		return -1;
1904 	}
1905 
1906 	vcpu->arch.tsc_scaling_ratio = ratio;
1907 	return 0;
1908 }
1909 
1910 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1911 {
1912 	u32 thresh_lo, thresh_hi;
1913 	int use_scaling = 0;
1914 
1915 	/* tsc_khz can be zero if TSC calibration fails */
1916 	if (user_tsc_khz == 0) {
1917 		/* set tsc_scaling_ratio to a safe value */
1918 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1919 		return -1;
1920 	}
1921 
1922 	/* Compute a scale to convert nanoseconds in TSC cycles */
1923 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1924 			   &vcpu->arch.virtual_tsc_shift,
1925 			   &vcpu->arch.virtual_tsc_mult);
1926 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1927 
1928 	/*
1929 	 * Compute the variation in TSC rate which is acceptable
1930 	 * within the range of tolerance and decide if the
1931 	 * rate being applied is within that bounds of the hardware
1932 	 * rate.  If so, no scaling or compensation need be done.
1933 	 */
1934 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1935 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1936 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1937 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1938 		use_scaling = 1;
1939 	}
1940 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1941 }
1942 
1943 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1944 {
1945 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1946 				      vcpu->arch.virtual_tsc_mult,
1947 				      vcpu->arch.virtual_tsc_shift);
1948 	tsc += vcpu->arch.this_tsc_write;
1949 	return tsc;
1950 }
1951 
1952 static inline int gtod_is_based_on_tsc(int mode)
1953 {
1954 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1955 }
1956 
1957 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1958 {
1959 #ifdef CONFIG_X86_64
1960 	bool vcpus_matched;
1961 	struct kvm_arch *ka = &vcpu->kvm->arch;
1962 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1963 
1964 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1965 			 atomic_read(&vcpu->kvm->online_vcpus));
1966 
1967 	/*
1968 	 * Once the masterclock is enabled, always perform request in
1969 	 * order to update it.
1970 	 *
1971 	 * In order to enable masterclock, the host clocksource must be TSC
1972 	 * and the vcpus need to have matched TSCs.  When that happens,
1973 	 * perform request to enable masterclock.
1974 	 */
1975 	if (ka->use_master_clock ||
1976 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1977 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1978 
1979 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1980 			    atomic_read(&vcpu->kvm->online_vcpus),
1981 		            ka->use_master_clock, gtod->clock.vclock_mode);
1982 #endif
1983 }
1984 
1985 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1986 {
1987 	u64 curr_offset = vcpu->arch.l1_tsc_offset;
1988 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1989 }
1990 
1991 /*
1992  * Multiply tsc by a fixed point number represented by ratio.
1993  *
1994  * The most significant 64-N bits (mult) of ratio represent the
1995  * integral part of the fixed point number; the remaining N bits
1996  * (frac) represent the fractional part, ie. ratio represents a fixed
1997  * point number (mult + frac * 2^(-N)).
1998  *
1999  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2000  */
2001 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2002 {
2003 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2004 }
2005 
2006 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2007 {
2008 	u64 _tsc = tsc;
2009 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
2010 
2011 	if (ratio != kvm_default_tsc_scaling_ratio)
2012 		_tsc = __scale_tsc(ratio, tsc);
2013 
2014 	return _tsc;
2015 }
2016 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2017 
2018 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2019 {
2020 	u64 tsc;
2021 
2022 	tsc = kvm_scale_tsc(vcpu, rdtsc());
2023 
2024 	return target_tsc - tsc;
2025 }
2026 
2027 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2028 {
2029 	return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2030 }
2031 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2032 
2033 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2034 {
2035 	vcpu->arch.l1_tsc_offset = offset;
2036 	vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2037 }
2038 
2039 static inline bool kvm_check_tsc_unstable(void)
2040 {
2041 #ifdef CONFIG_X86_64
2042 	/*
2043 	 * TSC is marked unstable when we're running on Hyper-V,
2044 	 * 'TSC page' clocksource is good.
2045 	 */
2046 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2047 		return false;
2048 #endif
2049 	return check_tsc_unstable();
2050 }
2051 
2052 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
2053 {
2054 	struct kvm *kvm = vcpu->kvm;
2055 	u64 offset, ns, elapsed;
2056 	unsigned long flags;
2057 	bool matched;
2058 	bool already_matched;
2059 	u64 data = msr->data;
2060 	bool synchronizing = false;
2061 
2062 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2063 	offset = kvm_compute_tsc_offset(vcpu, data);
2064 	ns = get_kvmclock_base_ns();
2065 	elapsed = ns - kvm->arch.last_tsc_nsec;
2066 
2067 	if (vcpu->arch.virtual_tsc_khz) {
2068 		if (data == 0 && msr->host_initiated) {
2069 			/*
2070 			 * detection of vcpu initialization -- need to sync
2071 			 * with other vCPUs. This particularly helps to keep
2072 			 * kvm_clock stable after CPU hotplug
2073 			 */
2074 			synchronizing = true;
2075 		} else {
2076 			u64 tsc_exp = kvm->arch.last_tsc_write +
2077 						nsec_to_cycles(vcpu, elapsed);
2078 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2079 			/*
2080 			 * Special case: TSC write with a small delta (1 second)
2081 			 * of virtual cycle time against real time is
2082 			 * interpreted as an attempt to synchronize the CPU.
2083 			 */
2084 			synchronizing = data < tsc_exp + tsc_hz &&
2085 					data + tsc_hz > tsc_exp;
2086 		}
2087 	}
2088 
2089 	/*
2090 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2091 	 * TSC, we add elapsed time in this computation.  We could let the
2092 	 * compensation code attempt to catch up if we fall behind, but
2093 	 * it's better to try to match offsets from the beginning.
2094          */
2095 	if (synchronizing &&
2096 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2097 		if (!kvm_check_tsc_unstable()) {
2098 			offset = kvm->arch.cur_tsc_offset;
2099 		} else {
2100 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2101 			data += delta;
2102 			offset = kvm_compute_tsc_offset(vcpu, data);
2103 		}
2104 		matched = true;
2105 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2106 	} else {
2107 		/*
2108 		 * We split periods of matched TSC writes into generations.
2109 		 * For each generation, we track the original measured
2110 		 * nanosecond time, offset, and write, so if TSCs are in
2111 		 * sync, we can match exact offset, and if not, we can match
2112 		 * exact software computation in compute_guest_tsc()
2113 		 *
2114 		 * These values are tracked in kvm->arch.cur_xxx variables.
2115 		 */
2116 		kvm->arch.cur_tsc_generation++;
2117 		kvm->arch.cur_tsc_nsec = ns;
2118 		kvm->arch.cur_tsc_write = data;
2119 		kvm->arch.cur_tsc_offset = offset;
2120 		matched = false;
2121 	}
2122 
2123 	/*
2124 	 * We also track th most recent recorded KHZ, write and time to
2125 	 * allow the matching interval to be extended at each write.
2126 	 */
2127 	kvm->arch.last_tsc_nsec = ns;
2128 	kvm->arch.last_tsc_write = data;
2129 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2130 
2131 	vcpu->arch.last_guest_tsc = data;
2132 
2133 	/* Keep track of which generation this VCPU has synchronized to */
2134 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2135 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2136 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2137 
2138 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2139 		update_ia32_tsc_adjust_msr(vcpu, offset);
2140 
2141 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2142 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2143 
2144 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2145 	if (!matched) {
2146 		kvm->arch.nr_vcpus_matched_tsc = 0;
2147 	} else if (!already_matched) {
2148 		kvm->arch.nr_vcpus_matched_tsc++;
2149 	}
2150 
2151 	kvm_track_tsc_matching(vcpu);
2152 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2153 }
2154 
2155 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2156 
2157 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2158 					   s64 adjustment)
2159 {
2160 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2161 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2162 }
2163 
2164 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2165 {
2166 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2167 		WARN_ON(adjustment < 0);
2168 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2169 	adjust_tsc_offset_guest(vcpu, adjustment);
2170 }
2171 
2172 #ifdef CONFIG_X86_64
2173 
2174 static u64 read_tsc(void)
2175 {
2176 	u64 ret = (u64)rdtsc_ordered();
2177 	u64 last = pvclock_gtod_data.clock.cycle_last;
2178 
2179 	if (likely(ret >= last))
2180 		return ret;
2181 
2182 	/*
2183 	 * GCC likes to generate cmov here, but this branch is extremely
2184 	 * predictable (it's just a function of time and the likely is
2185 	 * very likely) and there's a data dependence, so force GCC
2186 	 * to generate a branch instead.  I don't barrier() because
2187 	 * we don't actually need a barrier, and if this function
2188 	 * ever gets inlined it will generate worse code.
2189 	 */
2190 	asm volatile ("");
2191 	return last;
2192 }
2193 
2194 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2195 			  int *mode)
2196 {
2197 	long v;
2198 	u64 tsc_pg_val;
2199 
2200 	switch (clock->vclock_mode) {
2201 	case VDSO_CLOCKMODE_HVCLOCK:
2202 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2203 						  tsc_timestamp);
2204 		if (tsc_pg_val != U64_MAX) {
2205 			/* TSC page valid */
2206 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2207 			v = (tsc_pg_val - clock->cycle_last) &
2208 				clock->mask;
2209 		} else {
2210 			/* TSC page invalid */
2211 			*mode = VDSO_CLOCKMODE_NONE;
2212 		}
2213 		break;
2214 	case VDSO_CLOCKMODE_TSC:
2215 		*mode = VDSO_CLOCKMODE_TSC;
2216 		*tsc_timestamp = read_tsc();
2217 		v = (*tsc_timestamp - clock->cycle_last) &
2218 			clock->mask;
2219 		break;
2220 	default:
2221 		*mode = VDSO_CLOCKMODE_NONE;
2222 	}
2223 
2224 	if (*mode == VDSO_CLOCKMODE_NONE)
2225 		*tsc_timestamp = v = 0;
2226 
2227 	return v * clock->mult;
2228 }
2229 
2230 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2231 {
2232 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2233 	unsigned long seq;
2234 	int mode;
2235 	u64 ns;
2236 
2237 	do {
2238 		seq = read_seqcount_begin(&gtod->seq);
2239 		ns = gtod->raw_clock.base_cycles;
2240 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2241 		ns >>= gtod->raw_clock.shift;
2242 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2243 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2244 	*t = ns;
2245 
2246 	return mode;
2247 }
2248 
2249 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2250 {
2251 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2252 	unsigned long seq;
2253 	int mode;
2254 	u64 ns;
2255 
2256 	do {
2257 		seq = read_seqcount_begin(&gtod->seq);
2258 		ts->tv_sec = gtod->wall_time_sec;
2259 		ns = gtod->clock.base_cycles;
2260 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2261 		ns >>= gtod->clock.shift;
2262 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2263 
2264 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2265 	ts->tv_nsec = ns;
2266 
2267 	return mode;
2268 }
2269 
2270 /* returns true if host is using TSC based clocksource */
2271 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2272 {
2273 	/* checked again under seqlock below */
2274 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2275 		return false;
2276 
2277 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2278 						      tsc_timestamp));
2279 }
2280 
2281 /* returns true if host is using TSC based clocksource */
2282 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2283 					   u64 *tsc_timestamp)
2284 {
2285 	/* checked again under seqlock below */
2286 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2287 		return false;
2288 
2289 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2290 }
2291 #endif
2292 
2293 /*
2294  *
2295  * Assuming a stable TSC across physical CPUS, and a stable TSC
2296  * across virtual CPUs, the following condition is possible.
2297  * Each numbered line represents an event visible to both
2298  * CPUs at the next numbered event.
2299  *
2300  * "timespecX" represents host monotonic time. "tscX" represents
2301  * RDTSC value.
2302  *
2303  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2304  *
2305  * 1.  read timespec0,tsc0
2306  * 2.					| timespec1 = timespec0 + N
2307  * 					| tsc1 = tsc0 + M
2308  * 3. transition to guest		| transition to guest
2309  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2310  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2311  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2312  *
2313  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2314  *
2315  * 	- ret0 < ret1
2316  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2317  *		...
2318  *	- 0 < N - M => M < N
2319  *
2320  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2321  * always the case (the difference between two distinct xtime instances
2322  * might be smaller then the difference between corresponding TSC reads,
2323  * when updating guest vcpus pvclock areas).
2324  *
2325  * To avoid that problem, do not allow visibility of distinct
2326  * system_timestamp/tsc_timestamp values simultaneously: use a master
2327  * copy of host monotonic time values. Update that master copy
2328  * in lockstep.
2329  *
2330  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2331  *
2332  */
2333 
2334 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2335 {
2336 #ifdef CONFIG_X86_64
2337 	struct kvm_arch *ka = &kvm->arch;
2338 	int vclock_mode;
2339 	bool host_tsc_clocksource, vcpus_matched;
2340 
2341 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2342 			atomic_read(&kvm->online_vcpus));
2343 
2344 	/*
2345 	 * If the host uses TSC clock, then passthrough TSC as stable
2346 	 * to the guest.
2347 	 */
2348 	host_tsc_clocksource = kvm_get_time_and_clockread(
2349 					&ka->master_kernel_ns,
2350 					&ka->master_cycle_now);
2351 
2352 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2353 				&& !ka->backwards_tsc_observed
2354 				&& !ka->boot_vcpu_runs_old_kvmclock;
2355 
2356 	if (ka->use_master_clock)
2357 		atomic_set(&kvm_guest_has_master_clock, 1);
2358 
2359 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2360 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2361 					vcpus_matched);
2362 #endif
2363 }
2364 
2365 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2366 {
2367 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2368 }
2369 
2370 static void kvm_gen_update_masterclock(struct kvm *kvm)
2371 {
2372 #ifdef CONFIG_X86_64
2373 	int i;
2374 	struct kvm_vcpu *vcpu;
2375 	struct kvm_arch *ka = &kvm->arch;
2376 
2377 	spin_lock(&ka->pvclock_gtod_sync_lock);
2378 	kvm_make_mclock_inprogress_request(kvm);
2379 	/* no guest entries from this point */
2380 	pvclock_update_vm_gtod_copy(kvm);
2381 
2382 	kvm_for_each_vcpu(i, vcpu, kvm)
2383 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2384 
2385 	/* guest entries allowed */
2386 	kvm_for_each_vcpu(i, vcpu, kvm)
2387 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2388 
2389 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2390 #endif
2391 }
2392 
2393 u64 get_kvmclock_ns(struct kvm *kvm)
2394 {
2395 	struct kvm_arch *ka = &kvm->arch;
2396 	struct pvclock_vcpu_time_info hv_clock;
2397 	u64 ret;
2398 
2399 	spin_lock(&ka->pvclock_gtod_sync_lock);
2400 	if (!ka->use_master_clock) {
2401 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2402 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2403 	}
2404 
2405 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2406 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2407 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2408 
2409 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2410 	get_cpu();
2411 
2412 	if (__this_cpu_read(cpu_tsc_khz)) {
2413 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2414 				   &hv_clock.tsc_shift,
2415 				   &hv_clock.tsc_to_system_mul);
2416 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2417 	} else
2418 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2419 
2420 	put_cpu();
2421 
2422 	return ret;
2423 }
2424 
2425 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2426 {
2427 	struct kvm_vcpu_arch *vcpu = &v->arch;
2428 	struct pvclock_vcpu_time_info guest_hv_clock;
2429 
2430 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2431 		&guest_hv_clock, sizeof(guest_hv_clock))))
2432 		return;
2433 
2434 	/* This VCPU is paused, but it's legal for a guest to read another
2435 	 * VCPU's kvmclock, so we really have to follow the specification where
2436 	 * it says that version is odd if data is being modified, and even after
2437 	 * it is consistent.
2438 	 *
2439 	 * Version field updates must be kept separate.  This is because
2440 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2441 	 * writes within a string instruction are weakly ordered.  So there
2442 	 * are three writes overall.
2443 	 *
2444 	 * As a small optimization, only write the version field in the first
2445 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2446 	 * version field is the first in the struct.
2447 	 */
2448 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2449 
2450 	if (guest_hv_clock.version & 1)
2451 		++guest_hv_clock.version;  /* first time write, random junk */
2452 
2453 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2454 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2455 				&vcpu->hv_clock,
2456 				sizeof(vcpu->hv_clock.version));
2457 
2458 	smp_wmb();
2459 
2460 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2461 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2462 
2463 	if (vcpu->pvclock_set_guest_stopped_request) {
2464 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2465 		vcpu->pvclock_set_guest_stopped_request = false;
2466 	}
2467 
2468 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2469 
2470 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2471 				&vcpu->hv_clock,
2472 				sizeof(vcpu->hv_clock));
2473 
2474 	smp_wmb();
2475 
2476 	vcpu->hv_clock.version++;
2477 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2478 				&vcpu->hv_clock,
2479 				sizeof(vcpu->hv_clock.version));
2480 }
2481 
2482 static int kvm_guest_time_update(struct kvm_vcpu *v)
2483 {
2484 	unsigned long flags, tgt_tsc_khz;
2485 	struct kvm_vcpu_arch *vcpu = &v->arch;
2486 	struct kvm_arch *ka = &v->kvm->arch;
2487 	s64 kernel_ns;
2488 	u64 tsc_timestamp, host_tsc;
2489 	u8 pvclock_flags;
2490 	bool use_master_clock;
2491 
2492 	kernel_ns = 0;
2493 	host_tsc = 0;
2494 
2495 	/*
2496 	 * If the host uses TSC clock, then passthrough TSC as stable
2497 	 * to the guest.
2498 	 */
2499 	spin_lock(&ka->pvclock_gtod_sync_lock);
2500 	use_master_clock = ka->use_master_clock;
2501 	if (use_master_clock) {
2502 		host_tsc = ka->master_cycle_now;
2503 		kernel_ns = ka->master_kernel_ns;
2504 	}
2505 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2506 
2507 	/* Keep irq disabled to prevent changes to the clock */
2508 	local_irq_save(flags);
2509 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2510 	if (unlikely(tgt_tsc_khz == 0)) {
2511 		local_irq_restore(flags);
2512 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2513 		return 1;
2514 	}
2515 	if (!use_master_clock) {
2516 		host_tsc = rdtsc();
2517 		kernel_ns = get_kvmclock_base_ns();
2518 	}
2519 
2520 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2521 
2522 	/*
2523 	 * We may have to catch up the TSC to match elapsed wall clock
2524 	 * time for two reasons, even if kvmclock is used.
2525 	 *   1) CPU could have been running below the maximum TSC rate
2526 	 *   2) Broken TSC compensation resets the base at each VCPU
2527 	 *      entry to avoid unknown leaps of TSC even when running
2528 	 *      again on the same CPU.  This may cause apparent elapsed
2529 	 *      time to disappear, and the guest to stand still or run
2530 	 *	very slowly.
2531 	 */
2532 	if (vcpu->tsc_catchup) {
2533 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2534 		if (tsc > tsc_timestamp) {
2535 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2536 			tsc_timestamp = tsc;
2537 		}
2538 	}
2539 
2540 	local_irq_restore(flags);
2541 
2542 	/* With all the info we got, fill in the values */
2543 
2544 	if (kvm_has_tsc_control)
2545 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2546 
2547 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2548 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2549 				   &vcpu->hv_clock.tsc_shift,
2550 				   &vcpu->hv_clock.tsc_to_system_mul);
2551 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2552 	}
2553 
2554 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2555 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2556 	vcpu->last_guest_tsc = tsc_timestamp;
2557 
2558 	/* If the host uses TSC clocksource, then it is stable */
2559 	pvclock_flags = 0;
2560 	if (use_master_clock)
2561 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2562 
2563 	vcpu->hv_clock.flags = pvclock_flags;
2564 
2565 	if (vcpu->pv_time_enabled)
2566 		kvm_setup_pvclock_page(v);
2567 	if (v == kvm_get_vcpu(v->kvm, 0))
2568 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2569 	return 0;
2570 }
2571 
2572 /*
2573  * kvmclock updates which are isolated to a given vcpu, such as
2574  * vcpu->cpu migration, should not allow system_timestamp from
2575  * the rest of the vcpus to remain static. Otherwise ntp frequency
2576  * correction applies to one vcpu's system_timestamp but not
2577  * the others.
2578  *
2579  * So in those cases, request a kvmclock update for all vcpus.
2580  * We need to rate-limit these requests though, as they can
2581  * considerably slow guests that have a large number of vcpus.
2582  * The time for a remote vcpu to update its kvmclock is bound
2583  * by the delay we use to rate-limit the updates.
2584  */
2585 
2586 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2587 
2588 static void kvmclock_update_fn(struct work_struct *work)
2589 {
2590 	int i;
2591 	struct delayed_work *dwork = to_delayed_work(work);
2592 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2593 					   kvmclock_update_work);
2594 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2595 	struct kvm_vcpu *vcpu;
2596 
2597 	kvm_for_each_vcpu(i, vcpu, kvm) {
2598 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2599 		kvm_vcpu_kick(vcpu);
2600 	}
2601 }
2602 
2603 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2604 {
2605 	struct kvm *kvm = v->kvm;
2606 
2607 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2608 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2609 					KVMCLOCK_UPDATE_DELAY);
2610 }
2611 
2612 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2613 
2614 static void kvmclock_sync_fn(struct work_struct *work)
2615 {
2616 	struct delayed_work *dwork = to_delayed_work(work);
2617 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2618 					   kvmclock_sync_work);
2619 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2620 
2621 	if (!kvmclock_periodic_sync)
2622 		return;
2623 
2624 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2625 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2626 					KVMCLOCK_SYNC_PERIOD);
2627 }
2628 
2629 /*
2630  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2631  */
2632 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2633 {
2634 	/* McStatusWrEn enabled? */
2635 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2636 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2637 
2638 	return false;
2639 }
2640 
2641 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2642 {
2643 	u64 mcg_cap = vcpu->arch.mcg_cap;
2644 	unsigned bank_num = mcg_cap & 0xff;
2645 	u32 msr = msr_info->index;
2646 	u64 data = msr_info->data;
2647 
2648 	switch (msr) {
2649 	case MSR_IA32_MCG_STATUS:
2650 		vcpu->arch.mcg_status = data;
2651 		break;
2652 	case MSR_IA32_MCG_CTL:
2653 		if (!(mcg_cap & MCG_CTL_P) &&
2654 		    (data || !msr_info->host_initiated))
2655 			return 1;
2656 		if (data != 0 && data != ~(u64)0)
2657 			return 1;
2658 		vcpu->arch.mcg_ctl = data;
2659 		break;
2660 	default:
2661 		if (msr >= MSR_IA32_MC0_CTL &&
2662 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2663 			u32 offset = array_index_nospec(
2664 				msr - MSR_IA32_MC0_CTL,
2665 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2666 
2667 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2668 			 * some Linux kernels though clear bit 10 in bank 4 to
2669 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2670 			 * this to avoid an uncatched #GP in the guest
2671 			 */
2672 			if ((offset & 0x3) == 0 &&
2673 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2674 				return -1;
2675 
2676 			/* MCi_STATUS */
2677 			if (!msr_info->host_initiated &&
2678 			    (offset & 0x3) == 1 && data != 0) {
2679 				if (!can_set_mci_status(vcpu))
2680 					return -1;
2681 			}
2682 
2683 			vcpu->arch.mce_banks[offset] = data;
2684 			break;
2685 		}
2686 		return 1;
2687 	}
2688 	return 0;
2689 }
2690 
2691 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2692 {
2693 	struct kvm *kvm = vcpu->kvm;
2694 	int lm = is_long_mode(vcpu);
2695 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2696 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2697 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2698 		: kvm->arch.xen_hvm_config.blob_size_32;
2699 	u32 page_num = data & ~PAGE_MASK;
2700 	u64 page_addr = data & PAGE_MASK;
2701 	u8 *page;
2702 	int r;
2703 
2704 	r = -E2BIG;
2705 	if (page_num >= blob_size)
2706 		goto out;
2707 	r = -ENOMEM;
2708 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2709 	if (IS_ERR(page)) {
2710 		r = PTR_ERR(page);
2711 		goto out;
2712 	}
2713 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2714 		goto out_free;
2715 	r = 0;
2716 out_free:
2717 	kfree(page);
2718 out:
2719 	return r;
2720 }
2721 
2722 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2723 {
2724 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2725 
2726 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
2727 }
2728 
2729 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2730 {
2731 	gpa_t gpa = data & ~0x3f;
2732 
2733 	/* Bits 4:5 are reserved, Should be zero */
2734 	if (data & 0x30)
2735 		return 1;
2736 
2737 	if (!lapic_in_kernel(vcpu))
2738 		return 1;
2739 
2740 	vcpu->arch.apf.msr_en_val = data;
2741 
2742 	if (!kvm_pv_async_pf_enabled(vcpu)) {
2743 		kvm_clear_async_pf_completion_queue(vcpu);
2744 		kvm_async_pf_hash_reset(vcpu);
2745 		return 0;
2746 	}
2747 
2748 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2749 					sizeof(u64)))
2750 		return 1;
2751 
2752 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2753 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2754 
2755 	kvm_async_pf_wakeup_all(vcpu);
2756 
2757 	return 0;
2758 }
2759 
2760 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2761 {
2762 	/* Bits 8-63 are reserved */
2763 	if (data >> 8)
2764 		return 1;
2765 
2766 	if (!lapic_in_kernel(vcpu))
2767 		return 1;
2768 
2769 	vcpu->arch.apf.msr_int_val = data;
2770 
2771 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2772 
2773 	return 0;
2774 }
2775 
2776 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2777 {
2778 	vcpu->arch.pv_time_enabled = false;
2779 	vcpu->arch.time = 0;
2780 }
2781 
2782 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2783 {
2784 	++vcpu->stat.tlb_flush;
2785 	kvm_x86_ops.tlb_flush_all(vcpu);
2786 }
2787 
2788 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2789 {
2790 	++vcpu->stat.tlb_flush;
2791 	kvm_x86_ops.tlb_flush_guest(vcpu);
2792 }
2793 
2794 static void record_steal_time(struct kvm_vcpu *vcpu)
2795 {
2796 	struct kvm_host_map map;
2797 	struct kvm_steal_time *st;
2798 
2799 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2800 		return;
2801 
2802 	/* -EAGAIN is returned in atomic context so we can just return. */
2803 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2804 			&map, &vcpu->arch.st.cache, false))
2805 		return;
2806 
2807 	st = map.hva +
2808 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2809 
2810 	/*
2811 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2812 	 * expensive IPIs.
2813 	 */
2814 	trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2815 		st->preempted & KVM_VCPU_FLUSH_TLB);
2816 	if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2817 		kvm_vcpu_flush_tlb_guest(vcpu);
2818 
2819 	vcpu->arch.st.preempted = 0;
2820 
2821 	if (st->version & 1)
2822 		st->version += 1;  /* first time write, random junk */
2823 
2824 	st->version += 1;
2825 
2826 	smp_wmb();
2827 
2828 	st->steal += current->sched_info.run_delay -
2829 		vcpu->arch.st.last_steal;
2830 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2831 
2832 	smp_wmb();
2833 
2834 	st->version += 1;
2835 
2836 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2837 }
2838 
2839 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2840 {
2841 	bool pr = false;
2842 	u32 msr = msr_info->index;
2843 	u64 data = msr_info->data;
2844 
2845 	switch (msr) {
2846 	case MSR_AMD64_NB_CFG:
2847 	case MSR_IA32_UCODE_WRITE:
2848 	case MSR_VM_HSAVE_PA:
2849 	case MSR_AMD64_PATCH_LOADER:
2850 	case MSR_AMD64_BU_CFG2:
2851 	case MSR_AMD64_DC_CFG:
2852 	case MSR_F15H_EX_CFG:
2853 		break;
2854 
2855 	case MSR_IA32_UCODE_REV:
2856 		if (msr_info->host_initiated)
2857 			vcpu->arch.microcode_version = data;
2858 		break;
2859 	case MSR_IA32_ARCH_CAPABILITIES:
2860 		if (!msr_info->host_initiated)
2861 			return 1;
2862 		vcpu->arch.arch_capabilities = data;
2863 		break;
2864 	case MSR_IA32_PERF_CAPABILITIES: {
2865 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
2866 
2867 		if (!msr_info->host_initiated)
2868 			return 1;
2869 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
2870 			return 1;
2871 		if (data & ~msr_ent.data)
2872 			return 1;
2873 
2874 		vcpu->arch.perf_capabilities = data;
2875 
2876 		return 0;
2877 		}
2878 	case MSR_EFER:
2879 		return set_efer(vcpu, msr_info);
2880 	case MSR_K7_HWCR:
2881 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2882 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2883 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2884 
2885 		/* Handle McStatusWrEn */
2886 		if (data == BIT_ULL(18)) {
2887 			vcpu->arch.msr_hwcr = data;
2888 		} else if (data != 0) {
2889 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2890 				    data);
2891 			return 1;
2892 		}
2893 		break;
2894 	case MSR_FAM10H_MMIO_CONF_BASE:
2895 		if (data != 0) {
2896 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2897 				    "0x%llx\n", data);
2898 			return 1;
2899 		}
2900 		break;
2901 	case MSR_IA32_DEBUGCTLMSR:
2902 		if (!data) {
2903 			/* We support the non-activated case already */
2904 			break;
2905 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2906 			/* Values other than LBR and BTF are vendor-specific,
2907 			   thus reserved and should throw a #GP */
2908 			return 1;
2909 		}
2910 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2911 			    __func__, data);
2912 		break;
2913 	case 0x200 ... 0x2ff:
2914 		return kvm_mtrr_set_msr(vcpu, msr, data);
2915 	case MSR_IA32_APICBASE:
2916 		return kvm_set_apic_base(vcpu, msr_info);
2917 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
2918 		return kvm_x2apic_msr_write(vcpu, msr, data);
2919 	case MSR_IA32_TSCDEADLINE:
2920 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2921 		break;
2922 	case MSR_IA32_TSC_ADJUST:
2923 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2924 			if (!msr_info->host_initiated) {
2925 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2926 				adjust_tsc_offset_guest(vcpu, adj);
2927 			}
2928 			vcpu->arch.ia32_tsc_adjust_msr = data;
2929 		}
2930 		break;
2931 	case MSR_IA32_MISC_ENABLE:
2932 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2933 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2934 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2935 				return 1;
2936 			vcpu->arch.ia32_misc_enable_msr = data;
2937 			kvm_update_cpuid_runtime(vcpu);
2938 		} else {
2939 			vcpu->arch.ia32_misc_enable_msr = data;
2940 		}
2941 		break;
2942 	case MSR_IA32_SMBASE:
2943 		if (!msr_info->host_initiated)
2944 			return 1;
2945 		vcpu->arch.smbase = data;
2946 		break;
2947 	case MSR_IA32_POWER_CTL:
2948 		vcpu->arch.msr_ia32_power_ctl = data;
2949 		break;
2950 	case MSR_IA32_TSC:
2951 		kvm_write_tsc(vcpu, msr_info);
2952 		break;
2953 	case MSR_IA32_XSS:
2954 		if (!msr_info->host_initiated &&
2955 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2956 			return 1;
2957 		/*
2958 		 * KVM supports exposing PT to the guest, but does not support
2959 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2960 		 * XSAVES/XRSTORS to save/restore PT MSRs.
2961 		 */
2962 		if (data & ~supported_xss)
2963 			return 1;
2964 		vcpu->arch.ia32_xss = data;
2965 		break;
2966 	case MSR_SMI_COUNT:
2967 		if (!msr_info->host_initiated)
2968 			return 1;
2969 		vcpu->arch.smi_count = data;
2970 		break;
2971 	case MSR_KVM_WALL_CLOCK_NEW:
2972 	case MSR_KVM_WALL_CLOCK:
2973 		vcpu->kvm->arch.wall_clock = data;
2974 		kvm_write_wall_clock(vcpu->kvm, data);
2975 		break;
2976 	case MSR_KVM_SYSTEM_TIME_NEW:
2977 	case MSR_KVM_SYSTEM_TIME: {
2978 		struct kvm_arch *ka = &vcpu->kvm->arch;
2979 
2980 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2981 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2982 
2983 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2984 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2985 
2986 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2987 		}
2988 
2989 		vcpu->arch.time = data;
2990 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2991 
2992 		/* we verify if the enable bit is set... */
2993 		vcpu->arch.pv_time_enabled = false;
2994 		if (!(data & 1))
2995 			break;
2996 
2997 		if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2998 		     &vcpu->arch.pv_time, data & ~1ULL,
2999 		     sizeof(struct pvclock_vcpu_time_info)))
3000 			vcpu->arch.pv_time_enabled = true;
3001 
3002 		break;
3003 	}
3004 	case MSR_KVM_ASYNC_PF_EN:
3005 		if (kvm_pv_enable_async_pf(vcpu, data))
3006 			return 1;
3007 		break;
3008 	case MSR_KVM_ASYNC_PF_INT:
3009 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3010 			return 1;
3011 		break;
3012 	case MSR_KVM_ASYNC_PF_ACK:
3013 		if (data & 0x1) {
3014 			vcpu->arch.apf.pageready_pending = false;
3015 			kvm_check_async_pf_completion(vcpu);
3016 		}
3017 		break;
3018 	case MSR_KVM_STEAL_TIME:
3019 
3020 		if (unlikely(!sched_info_on()))
3021 			return 1;
3022 
3023 		if (data & KVM_STEAL_RESERVED_MASK)
3024 			return 1;
3025 
3026 		vcpu->arch.st.msr_val = data;
3027 
3028 		if (!(data & KVM_MSR_ENABLED))
3029 			break;
3030 
3031 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3032 
3033 		break;
3034 	case MSR_KVM_PV_EOI_EN:
3035 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3036 			return 1;
3037 		break;
3038 
3039 	case MSR_KVM_POLL_CONTROL:
3040 		/* only enable bit supported */
3041 		if (data & (-1ULL << 1))
3042 			return 1;
3043 
3044 		vcpu->arch.msr_kvm_poll_control = data;
3045 		break;
3046 
3047 	case MSR_IA32_MCG_CTL:
3048 	case MSR_IA32_MCG_STATUS:
3049 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3050 		return set_msr_mce(vcpu, msr_info);
3051 
3052 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3053 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3054 		pr = true; /* fall through */
3055 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3056 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3057 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3058 			return kvm_pmu_set_msr(vcpu, msr_info);
3059 
3060 		if (pr || data != 0)
3061 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3062 				    "0x%x data 0x%llx\n", msr, data);
3063 		break;
3064 	case MSR_K7_CLK_CTL:
3065 		/*
3066 		 * Ignore all writes to this no longer documented MSR.
3067 		 * Writes are only relevant for old K7 processors,
3068 		 * all pre-dating SVM, but a recommended workaround from
3069 		 * AMD for these chips. It is possible to specify the
3070 		 * affected processor models on the command line, hence
3071 		 * the need to ignore the workaround.
3072 		 */
3073 		break;
3074 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3075 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3076 	case HV_X64_MSR_SYNDBG_OPTIONS:
3077 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3078 	case HV_X64_MSR_CRASH_CTL:
3079 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3080 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3081 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3082 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3083 		return kvm_hv_set_msr_common(vcpu, msr, data,
3084 					     msr_info->host_initiated);
3085 	case MSR_IA32_BBL_CR_CTL3:
3086 		/* Drop writes to this legacy MSR -- see rdmsr
3087 		 * counterpart for further detail.
3088 		 */
3089 		if (report_ignored_msrs)
3090 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3091 				msr, data);
3092 		break;
3093 	case MSR_AMD64_OSVW_ID_LENGTH:
3094 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3095 			return 1;
3096 		vcpu->arch.osvw.length = data;
3097 		break;
3098 	case MSR_AMD64_OSVW_STATUS:
3099 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3100 			return 1;
3101 		vcpu->arch.osvw.status = data;
3102 		break;
3103 	case MSR_PLATFORM_INFO:
3104 		if (!msr_info->host_initiated ||
3105 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3106 		     cpuid_fault_enabled(vcpu)))
3107 			return 1;
3108 		vcpu->arch.msr_platform_info = data;
3109 		break;
3110 	case MSR_MISC_FEATURES_ENABLES:
3111 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3112 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3113 		     !supports_cpuid_fault(vcpu)))
3114 			return 1;
3115 		vcpu->arch.msr_misc_features_enables = data;
3116 		break;
3117 	default:
3118 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3119 			return xen_hvm_config(vcpu, data);
3120 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3121 			return kvm_pmu_set_msr(vcpu, msr_info);
3122 		return KVM_MSR_RET_INVALID;
3123 	}
3124 	return 0;
3125 }
3126 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3127 
3128 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3129 {
3130 	u64 data;
3131 	u64 mcg_cap = vcpu->arch.mcg_cap;
3132 	unsigned bank_num = mcg_cap & 0xff;
3133 
3134 	switch (msr) {
3135 	case MSR_IA32_P5_MC_ADDR:
3136 	case MSR_IA32_P5_MC_TYPE:
3137 		data = 0;
3138 		break;
3139 	case MSR_IA32_MCG_CAP:
3140 		data = vcpu->arch.mcg_cap;
3141 		break;
3142 	case MSR_IA32_MCG_CTL:
3143 		if (!(mcg_cap & MCG_CTL_P) && !host)
3144 			return 1;
3145 		data = vcpu->arch.mcg_ctl;
3146 		break;
3147 	case MSR_IA32_MCG_STATUS:
3148 		data = vcpu->arch.mcg_status;
3149 		break;
3150 	default:
3151 		if (msr >= MSR_IA32_MC0_CTL &&
3152 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3153 			u32 offset = array_index_nospec(
3154 				msr - MSR_IA32_MC0_CTL,
3155 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3156 
3157 			data = vcpu->arch.mce_banks[offset];
3158 			break;
3159 		}
3160 		return 1;
3161 	}
3162 	*pdata = data;
3163 	return 0;
3164 }
3165 
3166 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3167 {
3168 	switch (msr_info->index) {
3169 	case MSR_IA32_PLATFORM_ID:
3170 	case MSR_IA32_EBL_CR_POWERON:
3171 	case MSR_IA32_DEBUGCTLMSR:
3172 	case MSR_IA32_LASTBRANCHFROMIP:
3173 	case MSR_IA32_LASTBRANCHTOIP:
3174 	case MSR_IA32_LASTINTFROMIP:
3175 	case MSR_IA32_LASTINTTOIP:
3176 	case MSR_K8_SYSCFG:
3177 	case MSR_K8_TSEG_ADDR:
3178 	case MSR_K8_TSEG_MASK:
3179 	case MSR_VM_HSAVE_PA:
3180 	case MSR_K8_INT_PENDING_MSG:
3181 	case MSR_AMD64_NB_CFG:
3182 	case MSR_FAM10H_MMIO_CONF_BASE:
3183 	case MSR_AMD64_BU_CFG2:
3184 	case MSR_IA32_PERF_CTL:
3185 	case MSR_AMD64_DC_CFG:
3186 	case MSR_F15H_EX_CFG:
3187 	/*
3188 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3189 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3190 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3191 	 * so for existing CPU-specific MSRs.
3192 	 */
3193 	case MSR_RAPL_POWER_UNIT:
3194 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3195 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3196 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3197 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3198 		msr_info->data = 0;
3199 		break;
3200 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3201 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3202 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3203 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3204 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3205 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3206 			return kvm_pmu_get_msr(vcpu, msr_info);
3207 		msr_info->data = 0;
3208 		break;
3209 	case MSR_IA32_UCODE_REV:
3210 		msr_info->data = vcpu->arch.microcode_version;
3211 		break;
3212 	case MSR_IA32_ARCH_CAPABILITIES:
3213 		if (!msr_info->host_initiated &&
3214 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3215 			return 1;
3216 		msr_info->data = vcpu->arch.arch_capabilities;
3217 		break;
3218 	case MSR_IA32_PERF_CAPABILITIES:
3219 		if (!msr_info->host_initiated &&
3220 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3221 			return 1;
3222 		msr_info->data = vcpu->arch.perf_capabilities;
3223 		break;
3224 	case MSR_IA32_POWER_CTL:
3225 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3226 		break;
3227 	case MSR_IA32_TSC:
3228 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3229 		break;
3230 	case MSR_MTRRcap:
3231 	case 0x200 ... 0x2ff:
3232 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3233 	case 0xcd: /* fsb frequency */
3234 		msr_info->data = 3;
3235 		break;
3236 		/*
3237 		 * MSR_EBC_FREQUENCY_ID
3238 		 * Conservative value valid for even the basic CPU models.
3239 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3240 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3241 		 * and 266MHz for model 3, or 4. Set Core Clock
3242 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3243 		 * 31:24) even though these are only valid for CPU
3244 		 * models > 2, however guests may end up dividing or
3245 		 * multiplying by zero otherwise.
3246 		 */
3247 	case MSR_EBC_FREQUENCY_ID:
3248 		msr_info->data = 1 << 24;
3249 		break;
3250 	case MSR_IA32_APICBASE:
3251 		msr_info->data = kvm_get_apic_base(vcpu);
3252 		break;
3253 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3254 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3255 	case MSR_IA32_TSCDEADLINE:
3256 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3257 		break;
3258 	case MSR_IA32_TSC_ADJUST:
3259 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3260 		break;
3261 	case MSR_IA32_MISC_ENABLE:
3262 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3263 		break;
3264 	case MSR_IA32_SMBASE:
3265 		if (!msr_info->host_initiated)
3266 			return 1;
3267 		msr_info->data = vcpu->arch.smbase;
3268 		break;
3269 	case MSR_SMI_COUNT:
3270 		msr_info->data = vcpu->arch.smi_count;
3271 		break;
3272 	case MSR_IA32_PERF_STATUS:
3273 		/* TSC increment by tick */
3274 		msr_info->data = 1000ULL;
3275 		/* CPU multiplier */
3276 		msr_info->data |= (((uint64_t)4ULL) << 40);
3277 		break;
3278 	case MSR_EFER:
3279 		msr_info->data = vcpu->arch.efer;
3280 		break;
3281 	case MSR_KVM_WALL_CLOCK:
3282 	case MSR_KVM_WALL_CLOCK_NEW:
3283 		msr_info->data = vcpu->kvm->arch.wall_clock;
3284 		break;
3285 	case MSR_KVM_SYSTEM_TIME:
3286 	case MSR_KVM_SYSTEM_TIME_NEW:
3287 		msr_info->data = vcpu->arch.time;
3288 		break;
3289 	case MSR_KVM_ASYNC_PF_EN:
3290 		msr_info->data = vcpu->arch.apf.msr_en_val;
3291 		break;
3292 	case MSR_KVM_ASYNC_PF_INT:
3293 		msr_info->data = vcpu->arch.apf.msr_int_val;
3294 		break;
3295 	case MSR_KVM_ASYNC_PF_ACK:
3296 		msr_info->data = 0;
3297 		break;
3298 	case MSR_KVM_STEAL_TIME:
3299 		msr_info->data = vcpu->arch.st.msr_val;
3300 		break;
3301 	case MSR_KVM_PV_EOI_EN:
3302 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3303 		break;
3304 	case MSR_KVM_POLL_CONTROL:
3305 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3306 		break;
3307 	case MSR_IA32_P5_MC_ADDR:
3308 	case MSR_IA32_P5_MC_TYPE:
3309 	case MSR_IA32_MCG_CAP:
3310 	case MSR_IA32_MCG_CTL:
3311 	case MSR_IA32_MCG_STATUS:
3312 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3313 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3314 				   msr_info->host_initiated);
3315 	case MSR_IA32_XSS:
3316 		if (!msr_info->host_initiated &&
3317 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3318 			return 1;
3319 		msr_info->data = vcpu->arch.ia32_xss;
3320 		break;
3321 	case MSR_K7_CLK_CTL:
3322 		/*
3323 		 * Provide expected ramp-up count for K7. All other
3324 		 * are set to zero, indicating minimum divisors for
3325 		 * every field.
3326 		 *
3327 		 * This prevents guest kernels on AMD host with CPU
3328 		 * type 6, model 8 and higher from exploding due to
3329 		 * the rdmsr failing.
3330 		 */
3331 		msr_info->data = 0x20000000;
3332 		break;
3333 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3334 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3335 	case HV_X64_MSR_SYNDBG_OPTIONS:
3336 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3337 	case HV_X64_MSR_CRASH_CTL:
3338 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3339 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3340 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3341 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3342 		return kvm_hv_get_msr_common(vcpu,
3343 					     msr_info->index, &msr_info->data,
3344 					     msr_info->host_initiated);
3345 	case MSR_IA32_BBL_CR_CTL3:
3346 		/* This legacy MSR exists but isn't fully documented in current
3347 		 * silicon.  It is however accessed by winxp in very narrow
3348 		 * scenarios where it sets bit #19, itself documented as
3349 		 * a "reserved" bit.  Best effort attempt to source coherent
3350 		 * read data here should the balance of the register be
3351 		 * interpreted by the guest:
3352 		 *
3353 		 * L2 cache control register 3: 64GB range, 256KB size,
3354 		 * enabled, latency 0x1, configured
3355 		 */
3356 		msr_info->data = 0xbe702111;
3357 		break;
3358 	case MSR_AMD64_OSVW_ID_LENGTH:
3359 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3360 			return 1;
3361 		msr_info->data = vcpu->arch.osvw.length;
3362 		break;
3363 	case MSR_AMD64_OSVW_STATUS:
3364 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3365 			return 1;
3366 		msr_info->data = vcpu->arch.osvw.status;
3367 		break;
3368 	case MSR_PLATFORM_INFO:
3369 		if (!msr_info->host_initiated &&
3370 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3371 			return 1;
3372 		msr_info->data = vcpu->arch.msr_platform_info;
3373 		break;
3374 	case MSR_MISC_FEATURES_ENABLES:
3375 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3376 		break;
3377 	case MSR_K7_HWCR:
3378 		msr_info->data = vcpu->arch.msr_hwcr;
3379 		break;
3380 	default:
3381 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3382 			return kvm_pmu_get_msr(vcpu, msr_info);
3383 		return KVM_MSR_RET_INVALID;
3384 	}
3385 	return 0;
3386 }
3387 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3388 
3389 /*
3390  * Read or write a bunch of msrs. All parameters are kernel addresses.
3391  *
3392  * @return number of msrs set successfully.
3393  */
3394 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3395 		    struct kvm_msr_entry *entries,
3396 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3397 				  unsigned index, u64 *data))
3398 {
3399 	int i;
3400 
3401 	for (i = 0; i < msrs->nmsrs; ++i)
3402 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3403 			break;
3404 
3405 	return i;
3406 }
3407 
3408 /*
3409  * Read or write a bunch of msrs. Parameters are user addresses.
3410  *
3411  * @return number of msrs set successfully.
3412  */
3413 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3414 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3415 				unsigned index, u64 *data),
3416 		  int writeback)
3417 {
3418 	struct kvm_msrs msrs;
3419 	struct kvm_msr_entry *entries;
3420 	int r, n;
3421 	unsigned size;
3422 
3423 	r = -EFAULT;
3424 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3425 		goto out;
3426 
3427 	r = -E2BIG;
3428 	if (msrs.nmsrs >= MAX_IO_MSRS)
3429 		goto out;
3430 
3431 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3432 	entries = memdup_user(user_msrs->entries, size);
3433 	if (IS_ERR(entries)) {
3434 		r = PTR_ERR(entries);
3435 		goto out;
3436 	}
3437 
3438 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3439 	if (r < 0)
3440 		goto out_free;
3441 
3442 	r = -EFAULT;
3443 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3444 		goto out_free;
3445 
3446 	r = n;
3447 
3448 out_free:
3449 	kfree(entries);
3450 out:
3451 	return r;
3452 }
3453 
3454 static inline bool kvm_can_mwait_in_guest(void)
3455 {
3456 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3457 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3458 		boot_cpu_has(X86_FEATURE_ARAT);
3459 }
3460 
3461 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3462 {
3463 	int r = 0;
3464 
3465 	switch (ext) {
3466 	case KVM_CAP_IRQCHIP:
3467 	case KVM_CAP_HLT:
3468 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3469 	case KVM_CAP_SET_TSS_ADDR:
3470 	case KVM_CAP_EXT_CPUID:
3471 	case KVM_CAP_EXT_EMUL_CPUID:
3472 	case KVM_CAP_CLOCKSOURCE:
3473 	case KVM_CAP_PIT:
3474 	case KVM_CAP_NOP_IO_DELAY:
3475 	case KVM_CAP_MP_STATE:
3476 	case KVM_CAP_SYNC_MMU:
3477 	case KVM_CAP_USER_NMI:
3478 	case KVM_CAP_REINJECT_CONTROL:
3479 	case KVM_CAP_IRQ_INJECT_STATUS:
3480 	case KVM_CAP_IOEVENTFD:
3481 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3482 	case KVM_CAP_PIT2:
3483 	case KVM_CAP_PIT_STATE2:
3484 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3485 	case KVM_CAP_XEN_HVM:
3486 	case KVM_CAP_VCPU_EVENTS:
3487 	case KVM_CAP_HYPERV:
3488 	case KVM_CAP_HYPERV_VAPIC:
3489 	case KVM_CAP_HYPERV_SPIN:
3490 	case KVM_CAP_HYPERV_SYNIC:
3491 	case KVM_CAP_HYPERV_SYNIC2:
3492 	case KVM_CAP_HYPERV_VP_INDEX:
3493 	case KVM_CAP_HYPERV_EVENTFD:
3494 	case KVM_CAP_HYPERV_TLBFLUSH:
3495 	case KVM_CAP_HYPERV_SEND_IPI:
3496 	case KVM_CAP_HYPERV_CPUID:
3497 	case KVM_CAP_PCI_SEGMENT:
3498 	case KVM_CAP_DEBUGREGS:
3499 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3500 	case KVM_CAP_XSAVE:
3501 	case KVM_CAP_ASYNC_PF:
3502 	case KVM_CAP_ASYNC_PF_INT:
3503 	case KVM_CAP_GET_TSC_KHZ:
3504 	case KVM_CAP_KVMCLOCK_CTRL:
3505 	case KVM_CAP_READONLY_MEM:
3506 	case KVM_CAP_HYPERV_TIME:
3507 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3508 	case KVM_CAP_TSC_DEADLINE_TIMER:
3509 	case KVM_CAP_DISABLE_QUIRKS:
3510 	case KVM_CAP_SET_BOOT_CPU_ID:
3511  	case KVM_CAP_SPLIT_IRQCHIP:
3512 	case KVM_CAP_IMMEDIATE_EXIT:
3513 	case KVM_CAP_PMU_EVENT_FILTER:
3514 	case KVM_CAP_GET_MSR_FEATURES:
3515 	case KVM_CAP_MSR_PLATFORM_INFO:
3516 	case KVM_CAP_EXCEPTION_PAYLOAD:
3517 	case KVM_CAP_SET_GUEST_DEBUG:
3518 	case KVM_CAP_LAST_CPU:
3519 		r = 1;
3520 		break;
3521 	case KVM_CAP_SYNC_REGS:
3522 		r = KVM_SYNC_X86_VALID_FIELDS;
3523 		break;
3524 	case KVM_CAP_ADJUST_CLOCK:
3525 		r = KVM_CLOCK_TSC_STABLE;
3526 		break;
3527 	case KVM_CAP_X86_DISABLE_EXITS:
3528 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3529 		      KVM_X86_DISABLE_EXITS_CSTATE;
3530 		if(kvm_can_mwait_in_guest())
3531 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3532 		break;
3533 	case KVM_CAP_X86_SMM:
3534 		/* SMBASE is usually relocated above 1M on modern chipsets,
3535 		 * and SMM handlers might indeed rely on 4G segment limits,
3536 		 * so do not report SMM to be available if real mode is
3537 		 * emulated via vm86 mode.  Still, do not go to great lengths
3538 		 * to avoid userspace's usage of the feature, because it is a
3539 		 * fringe case that is not enabled except via specific settings
3540 		 * of the module parameters.
3541 		 */
3542 		r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3543 		break;
3544 	case KVM_CAP_VAPIC:
3545 		r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3546 		break;
3547 	case KVM_CAP_NR_VCPUS:
3548 		r = KVM_SOFT_MAX_VCPUS;
3549 		break;
3550 	case KVM_CAP_MAX_VCPUS:
3551 		r = KVM_MAX_VCPUS;
3552 		break;
3553 	case KVM_CAP_MAX_VCPU_ID:
3554 		r = KVM_MAX_VCPU_ID;
3555 		break;
3556 	case KVM_CAP_PV_MMU:	/* obsolete */
3557 		r = 0;
3558 		break;
3559 	case KVM_CAP_MCE:
3560 		r = KVM_MAX_MCE_BANKS;
3561 		break;
3562 	case KVM_CAP_XCRS:
3563 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3564 		break;
3565 	case KVM_CAP_TSC_CONTROL:
3566 		r = kvm_has_tsc_control;
3567 		break;
3568 	case KVM_CAP_X2APIC_API:
3569 		r = KVM_X2APIC_API_VALID_FLAGS;
3570 		break;
3571 	case KVM_CAP_NESTED_STATE:
3572 		r = kvm_x86_ops.nested_ops->get_state ?
3573 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3574 		break;
3575 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3576 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3577 		break;
3578 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3579 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3580 		break;
3581 	case KVM_CAP_SMALLER_MAXPHYADDR:
3582 		r = (int) allow_smaller_maxphyaddr;
3583 		break;
3584 	default:
3585 		break;
3586 	}
3587 	return r;
3588 
3589 }
3590 
3591 long kvm_arch_dev_ioctl(struct file *filp,
3592 			unsigned int ioctl, unsigned long arg)
3593 {
3594 	void __user *argp = (void __user *)arg;
3595 	long r;
3596 
3597 	switch (ioctl) {
3598 	case KVM_GET_MSR_INDEX_LIST: {
3599 		struct kvm_msr_list __user *user_msr_list = argp;
3600 		struct kvm_msr_list msr_list;
3601 		unsigned n;
3602 
3603 		r = -EFAULT;
3604 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3605 			goto out;
3606 		n = msr_list.nmsrs;
3607 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3608 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3609 			goto out;
3610 		r = -E2BIG;
3611 		if (n < msr_list.nmsrs)
3612 			goto out;
3613 		r = -EFAULT;
3614 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3615 				 num_msrs_to_save * sizeof(u32)))
3616 			goto out;
3617 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3618 				 &emulated_msrs,
3619 				 num_emulated_msrs * sizeof(u32)))
3620 			goto out;
3621 		r = 0;
3622 		break;
3623 	}
3624 	case KVM_GET_SUPPORTED_CPUID:
3625 	case KVM_GET_EMULATED_CPUID: {
3626 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3627 		struct kvm_cpuid2 cpuid;
3628 
3629 		r = -EFAULT;
3630 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3631 			goto out;
3632 
3633 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3634 					    ioctl);
3635 		if (r)
3636 			goto out;
3637 
3638 		r = -EFAULT;
3639 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3640 			goto out;
3641 		r = 0;
3642 		break;
3643 	}
3644 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
3645 		r = -EFAULT;
3646 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3647 				 sizeof(kvm_mce_cap_supported)))
3648 			goto out;
3649 		r = 0;
3650 		break;
3651 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3652 		struct kvm_msr_list __user *user_msr_list = argp;
3653 		struct kvm_msr_list msr_list;
3654 		unsigned int n;
3655 
3656 		r = -EFAULT;
3657 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3658 			goto out;
3659 		n = msr_list.nmsrs;
3660 		msr_list.nmsrs = num_msr_based_features;
3661 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3662 			goto out;
3663 		r = -E2BIG;
3664 		if (n < msr_list.nmsrs)
3665 			goto out;
3666 		r = -EFAULT;
3667 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3668 				 num_msr_based_features * sizeof(u32)))
3669 			goto out;
3670 		r = 0;
3671 		break;
3672 	}
3673 	case KVM_GET_MSRS:
3674 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3675 		break;
3676 	default:
3677 		r = -EINVAL;
3678 		break;
3679 	}
3680 out:
3681 	return r;
3682 }
3683 
3684 static void wbinvd_ipi(void *garbage)
3685 {
3686 	wbinvd();
3687 }
3688 
3689 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3690 {
3691 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3692 }
3693 
3694 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3695 {
3696 	/* Address WBINVD may be executed by guest */
3697 	if (need_emulate_wbinvd(vcpu)) {
3698 		if (kvm_x86_ops.has_wbinvd_exit())
3699 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3700 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3701 			smp_call_function_single(vcpu->cpu,
3702 					wbinvd_ipi, NULL, 1);
3703 	}
3704 
3705 	kvm_x86_ops.vcpu_load(vcpu, cpu);
3706 
3707 	/* Save host pkru register if supported */
3708 	vcpu->arch.host_pkru = read_pkru();
3709 
3710 	/* Apply any externally detected TSC adjustments (due to suspend) */
3711 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3712 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3713 		vcpu->arch.tsc_offset_adjustment = 0;
3714 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3715 	}
3716 
3717 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3718 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3719 				rdtsc() - vcpu->arch.last_host_tsc;
3720 		if (tsc_delta < 0)
3721 			mark_tsc_unstable("KVM discovered backwards TSC");
3722 
3723 		if (kvm_check_tsc_unstable()) {
3724 			u64 offset = kvm_compute_tsc_offset(vcpu,
3725 						vcpu->arch.last_guest_tsc);
3726 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3727 			vcpu->arch.tsc_catchup = 1;
3728 		}
3729 
3730 		if (kvm_lapic_hv_timer_in_use(vcpu))
3731 			kvm_lapic_restart_hv_timer(vcpu);
3732 
3733 		/*
3734 		 * On a host with synchronized TSC, there is no need to update
3735 		 * kvmclock on vcpu->cpu migration
3736 		 */
3737 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3738 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3739 		if (vcpu->cpu != cpu)
3740 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3741 		vcpu->cpu = cpu;
3742 	}
3743 
3744 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3745 }
3746 
3747 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3748 {
3749 	struct kvm_host_map map;
3750 	struct kvm_steal_time *st;
3751 
3752 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3753 		return;
3754 
3755 	if (vcpu->arch.st.preempted)
3756 		return;
3757 
3758 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3759 			&vcpu->arch.st.cache, true))
3760 		return;
3761 
3762 	st = map.hva +
3763 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3764 
3765 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3766 
3767 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3768 }
3769 
3770 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3771 {
3772 	int idx;
3773 
3774 	if (vcpu->preempted)
3775 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3776 
3777 	/*
3778 	 * Disable page faults because we're in atomic context here.
3779 	 * kvm_write_guest_offset_cached() would call might_fault()
3780 	 * that relies on pagefault_disable() to tell if there's a
3781 	 * bug. NOTE: the write to guest memory may not go through if
3782 	 * during postcopy live migration or if there's heavy guest
3783 	 * paging.
3784 	 */
3785 	pagefault_disable();
3786 	/*
3787 	 * kvm_memslots() will be called by
3788 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3789 	 */
3790 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3791 	kvm_steal_time_set_preempted(vcpu);
3792 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3793 	pagefault_enable();
3794 	kvm_x86_ops.vcpu_put(vcpu);
3795 	vcpu->arch.last_host_tsc = rdtsc();
3796 	/*
3797 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3798 	 * on every vmexit, but if not, we might have a stale dr6 from the
3799 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3800 	 */
3801 	set_debugreg(0, 6);
3802 }
3803 
3804 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3805 				    struct kvm_lapic_state *s)
3806 {
3807 	if (vcpu->arch.apicv_active)
3808 		kvm_x86_ops.sync_pir_to_irr(vcpu);
3809 
3810 	return kvm_apic_get_state(vcpu, s);
3811 }
3812 
3813 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3814 				    struct kvm_lapic_state *s)
3815 {
3816 	int r;
3817 
3818 	r = kvm_apic_set_state(vcpu, s);
3819 	if (r)
3820 		return r;
3821 	update_cr8_intercept(vcpu);
3822 
3823 	return 0;
3824 }
3825 
3826 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3827 {
3828 	return (!lapic_in_kernel(vcpu) ||
3829 		kvm_apic_accept_pic_intr(vcpu));
3830 }
3831 
3832 /*
3833  * if userspace requested an interrupt window, check that the
3834  * interrupt window is open.
3835  *
3836  * No need to exit to userspace if we already have an interrupt queued.
3837  */
3838 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3839 {
3840 	return kvm_arch_interrupt_allowed(vcpu) &&
3841 		!kvm_cpu_has_interrupt(vcpu) &&
3842 		!kvm_event_needs_reinjection(vcpu) &&
3843 		kvm_cpu_accept_dm_intr(vcpu);
3844 }
3845 
3846 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3847 				    struct kvm_interrupt *irq)
3848 {
3849 	if (irq->irq >= KVM_NR_INTERRUPTS)
3850 		return -EINVAL;
3851 
3852 	if (!irqchip_in_kernel(vcpu->kvm)) {
3853 		kvm_queue_interrupt(vcpu, irq->irq, false);
3854 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3855 		return 0;
3856 	}
3857 
3858 	/*
3859 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3860 	 * fail for in-kernel 8259.
3861 	 */
3862 	if (pic_in_kernel(vcpu->kvm))
3863 		return -ENXIO;
3864 
3865 	if (vcpu->arch.pending_external_vector != -1)
3866 		return -EEXIST;
3867 
3868 	vcpu->arch.pending_external_vector = irq->irq;
3869 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3870 	return 0;
3871 }
3872 
3873 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3874 {
3875 	kvm_inject_nmi(vcpu);
3876 
3877 	return 0;
3878 }
3879 
3880 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3881 {
3882 	kvm_make_request(KVM_REQ_SMI, vcpu);
3883 
3884 	return 0;
3885 }
3886 
3887 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3888 					   struct kvm_tpr_access_ctl *tac)
3889 {
3890 	if (tac->flags)
3891 		return -EINVAL;
3892 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3893 	return 0;
3894 }
3895 
3896 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3897 					u64 mcg_cap)
3898 {
3899 	int r;
3900 	unsigned bank_num = mcg_cap & 0xff, bank;
3901 
3902 	r = -EINVAL;
3903 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
3904 		goto out;
3905 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3906 		goto out;
3907 	r = 0;
3908 	vcpu->arch.mcg_cap = mcg_cap;
3909 	/* Init IA32_MCG_CTL to all 1s */
3910 	if (mcg_cap & MCG_CTL_P)
3911 		vcpu->arch.mcg_ctl = ~(u64)0;
3912 	/* Init IA32_MCi_CTL to all 1s */
3913 	for (bank = 0; bank < bank_num; bank++)
3914 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3915 
3916 	kvm_x86_ops.setup_mce(vcpu);
3917 out:
3918 	return r;
3919 }
3920 
3921 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3922 				      struct kvm_x86_mce *mce)
3923 {
3924 	u64 mcg_cap = vcpu->arch.mcg_cap;
3925 	unsigned bank_num = mcg_cap & 0xff;
3926 	u64 *banks = vcpu->arch.mce_banks;
3927 
3928 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3929 		return -EINVAL;
3930 	/*
3931 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3932 	 * reporting is disabled
3933 	 */
3934 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3935 	    vcpu->arch.mcg_ctl != ~(u64)0)
3936 		return 0;
3937 	banks += 4 * mce->bank;
3938 	/*
3939 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3940 	 * reporting is disabled for the bank
3941 	 */
3942 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3943 		return 0;
3944 	if (mce->status & MCI_STATUS_UC) {
3945 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3946 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3947 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3948 			return 0;
3949 		}
3950 		if (banks[1] & MCI_STATUS_VAL)
3951 			mce->status |= MCI_STATUS_OVER;
3952 		banks[2] = mce->addr;
3953 		banks[3] = mce->misc;
3954 		vcpu->arch.mcg_status = mce->mcg_status;
3955 		banks[1] = mce->status;
3956 		kvm_queue_exception(vcpu, MC_VECTOR);
3957 	} else if (!(banks[1] & MCI_STATUS_VAL)
3958 		   || !(banks[1] & MCI_STATUS_UC)) {
3959 		if (banks[1] & MCI_STATUS_VAL)
3960 			mce->status |= MCI_STATUS_OVER;
3961 		banks[2] = mce->addr;
3962 		banks[3] = mce->misc;
3963 		banks[1] = mce->status;
3964 	} else
3965 		banks[1] |= MCI_STATUS_OVER;
3966 	return 0;
3967 }
3968 
3969 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3970 					       struct kvm_vcpu_events *events)
3971 {
3972 	process_nmi(vcpu);
3973 
3974 	/*
3975 	 * In guest mode, payload delivery should be deferred,
3976 	 * so that the L1 hypervisor can intercept #PF before
3977 	 * CR2 is modified (or intercept #DB before DR6 is
3978 	 * modified under nVMX). Unless the per-VM capability,
3979 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3980 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3981 	 * opportunistically defer the exception payload, deliver it if the
3982 	 * capability hasn't been requested before processing a
3983 	 * KVM_GET_VCPU_EVENTS.
3984 	 */
3985 	if (!vcpu->kvm->arch.exception_payload_enabled &&
3986 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3987 		kvm_deliver_exception_payload(vcpu);
3988 
3989 	/*
3990 	 * The API doesn't provide the instruction length for software
3991 	 * exceptions, so don't report them. As long as the guest RIP
3992 	 * isn't advanced, we should expect to encounter the exception
3993 	 * again.
3994 	 */
3995 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3996 		events->exception.injected = 0;
3997 		events->exception.pending = 0;
3998 	} else {
3999 		events->exception.injected = vcpu->arch.exception.injected;
4000 		events->exception.pending = vcpu->arch.exception.pending;
4001 		/*
4002 		 * For ABI compatibility, deliberately conflate
4003 		 * pending and injected exceptions when
4004 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4005 		 */
4006 		if (!vcpu->kvm->arch.exception_payload_enabled)
4007 			events->exception.injected |=
4008 				vcpu->arch.exception.pending;
4009 	}
4010 	events->exception.nr = vcpu->arch.exception.nr;
4011 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4012 	events->exception.error_code = vcpu->arch.exception.error_code;
4013 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4014 	events->exception_payload = vcpu->arch.exception.payload;
4015 
4016 	events->interrupt.injected =
4017 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4018 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4019 	events->interrupt.soft = 0;
4020 	events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4021 
4022 	events->nmi.injected = vcpu->arch.nmi_injected;
4023 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4024 	events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4025 	events->nmi.pad = 0;
4026 
4027 	events->sipi_vector = 0; /* never valid when reporting to user space */
4028 
4029 	events->smi.smm = is_smm(vcpu);
4030 	events->smi.pending = vcpu->arch.smi_pending;
4031 	events->smi.smm_inside_nmi =
4032 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4033 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4034 
4035 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4036 			 | KVM_VCPUEVENT_VALID_SHADOW
4037 			 | KVM_VCPUEVENT_VALID_SMM);
4038 	if (vcpu->kvm->arch.exception_payload_enabled)
4039 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4040 
4041 	memset(&events->reserved, 0, sizeof(events->reserved));
4042 }
4043 
4044 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4045 
4046 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4047 					      struct kvm_vcpu_events *events)
4048 {
4049 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4050 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4051 			      | KVM_VCPUEVENT_VALID_SHADOW
4052 			      | KVM_VCPUEVENT_VALID_SMM
4053 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4054 		return -EINVAL;
4055 
4056 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4057 		if (!vcpu->kvm->arch.exception_payload_enabled)
4058 			return -EINVAL;
4059 		if (events->exception.pending)
4060 			events->exception.injected = 0;
4061 		else
4062 			events->exception_has_payload = 0;
4063 	} else {
4064 		events->exception.pending = 0;
4065 		events->exception_has_payload = 0;
4066 	}
4067 
4068 	if ((events->exception.injected || events->exception.pending) &&
4069 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4070 		return -EINVAL;
4071 
4072 	/* INITs are latched while in SMM */
4073 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4074 	    (events->smi.smm || events->smi.pending) &&
4075 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4076 		return -EINVAL;
4077 
4078 	process_nmi(vcpu);
4079 	vcpu->arch.exception.injected = events->exception.injected;
4080 	vcpu->arch.exception.pending = events->exception.pending;
4081 	vcpu->arch.exception.nr = events->exception.nr;
4082 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4083 	vcpu->arch.exception.error_code = events->exception.error_code;
4084 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4085 	vcpu->arch.exception.payload = events->exception_payload;
4086 
4087 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4088 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4089 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4090 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4091 		kvm_x86_ops.set_interrupt_shadow(vcpu,
4092 						  events->interrupt.shadow);
4093 
4094 	vcpu->arch.nmi_injected = events->nmi.injected;
4095 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4096 		vcpu->arch.nmi_pending = events->nmi.pending;
4097 	kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4098 
4099 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4100 	    lapic_in_kernel(vcpu))
4101 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4102 
4103 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4104 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4105 			if (events->smi.smm)
4106 				vcpu->arch.hflags |= HF_SMM_MASK;
4107 			else
4108 				vcpu->arch.hflags &= ~HF_SMM_MASK;
4109 			kvm_smm_changed(vcpu);
4110 		}
4111 
4112 		vcpu->arch.smi_pending = events->smi.pending;
4113 
4114 		if (events->smi.smm) {
4115 			if (events->smi.smm_inside_nmi)
4116 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4117 			else
4118 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4119 		}
4120 
4121 		if (lapic_in_kernel(vcpu)) {
4122 			if (events->smi.latched_init)
4123 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4124 			else
4125 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4126 		}
4127 	}
4128 
4129 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4130 
4131 	return 0;
4132 }
4133 
4134 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4135 					     struct kvm_debugregs *dbgregs)
4136 {
4137 	unsigned long val;
4138 
4139 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4140 	kvm_get_dr(vcpu, 6, &val);
4141 	dbgregs->dr6 = val;
4142 	dbgregs->dr7 = vcpu->arch.dr7;
4143 	dbgregs->flags = 0;
4144 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4145 }
4146 
4147 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4148 					    struct kvm_debugregs *dbgregs)
4149 {
4150 	if (dbgregs->flags)
4151 		return -EINVAL;
4152 
4153 	if (dbgregs->dr6 & ~0xffffffffull)
4154 		return -EINVAL;
4155 	if (dbgregs->dr7 & ~0xffffffffull)
4156 		return -EINVAL;
4157 
4158 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4159 	kvm_update_dr0123(vcpu);
4160 	vcpu->arch.dr6 = dbgregs->dr6;
4161 	vcpu->arch.dr7 = dbgregs->dr7;
4162 	kvm_update_dr7(vcpu);
4163 
4164 	return 0;
4165 }
4166 
4167 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4168 
4169 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4170 {
4171 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4172 	u64 xstate_bv = xsave->header.xfeatures;
4173 	u64 valid;
4174 
4175 	/*
4176 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4177 	 * leaves 0 and 1 in the loop below.
4178 	 */
4179 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4180 
4181 	/* Set XSTATE_BV */
4182 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4183 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4184 
4185 	/*
4186 	 * Copy each region from the possibly compacted offset to the
4187 	 * non-compacted offset.
4188 	 */
4189 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4190 	while (valid) {
4191 		u64 xfeature_mask = valid & -valid;
4192 		int xfeature_nr = fls64(xfeature_mask) - 1;
4193 		void *src = get_xsave_addr(xsave, xfeature_nr);
4194 
4195 		if (src) {
4196 			u32 size, offset, ecx, edx;
4197 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4198 				    &size, &offset, &ecx, &edx);
4199 			if (xfeature_nr == XFEATURE_PKRU)
4200 				memcpy(dest + offset, &vcpu->arch.pkru,
4201 				       sizeof(vcpu->arch.pkru));
4202 			else
4203 				memcpy(dest + offset, src, size);
4204 
4205 		}
4206 
4207 		valid -= xfeature_mask;
4208 	}
4209 }
4210 
4211 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4212 {
4213 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4214 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4215 	u64 valid;
4216 
4217 	/*
4218 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4219 	 * leaves 0 and 1 in the loop below.
4220 	 */
4221 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4222 
4223 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4224 	xsave->header.xfeatures = xstate_bv;
4225 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4226 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4227 
4228 	/*
4229 	 * Copy each region from the non-compacted offset to the
4230 	 * possibly compacted offset.
4231 	 */
4232 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4233 	while (valid) {
4234 		u64 xfeature_mask = valid & -valid;
4235 		int xfeature_nr = fls64(xfeature_mask) - 1;
4236 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4237 
4238 		if (dest) {
4239 			u32 size, offset, ecx, edx;
4240 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4241 				    &size, &offset, &ecx, &edx);
4242 			if (xfeature_nr == XFEATURE_PKRU)
4243 				memcpy(&vcpu->arch.pkru, src + offset,
4244 				       sizeof(vcpu->arch.pkru));
4245 			else
4246 				memcpy(dest, src + offset, size);
4247 		}
4248 
4249 		valid -= xfeature_mask;
4250 	}
4251 }
4252 
4253 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4254 					 struct kvm_xsave *guest_xsave)
4255 {
4256 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4257 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4258 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4259 	} else {
4260 		memcpy(guest_xsave->region,
4261 			&vcpu->arch.guest_fpu->state.fxsave,
4262 			sizeof(struct fxregs_state));
4263 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4264 			XFEATURE_MASK_FPSSE;
4265 	}
4266 }
4267 
4268 #define XSAVE_MXCSR_OFFSET 24
4269 
4270 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4271 					struct kvm_xsave *guest_xsave)
4272 {
4273 	u64 xstate_bv =
4274 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4275 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4276 
4277 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4278 		/*
4279 		 * Here we allow setting states that are not present in
4280 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4281 		 * with old userspace.
4282 		 */
4283 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4284 			return -EINVAL;
4285 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4286 	} else {
4287 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4288 			mxcsr & ~mxcsr_feature_mask)
4289 			return -EINVAL;
4290 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4291 			guest_xsave->region, sizeof(struct fxregs_state));
4292 	}
4293 	return 0;
4294 }
4295 
4296 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4297 					struct kvm_xcrs *guest_xcrs)
4298 {
4299 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4300 		guest_xcrs->nr_xcrs = 0;
4301 		return;
4302 	}
4303 
4304 	guest_xcrs->nr_xcrs = 1;
4305 	guest_xcrs->flags = 0;
4306 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4307 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4308 }
4309 
4310 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4311 				       struct kvm_xcrs *guest_xcrs)
4312 {
4313 	int i, r = 0;
4314 
4315 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4316 		return -EINVAL;
4317 
4318 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4319 		return -EINVAL;
4320 
4321 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4322 		/* Only support XCR0 currently */
4323 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4324 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4325 				guest_xcrs->xcrs[i].value);
4326 			break;
4327 		}
4328 	if (r)
4329 		r = -EINVAL;
4330 	return r;
4331 }
4332 
4333 /*
4334  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4335  * stopped by the hypervisor.  This function will be called from the host only.
4336  * EINVAL is returned when the host attempts to set the flag for a guest that
4337  * does not support pv clocks.
4338  */
4339 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4340 {
4341 	if (!vcpu->arch.pv_time_enabled)
4342 		return -EINVAL;
4343 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4344 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4345 	return 0;
4346 }
4347 
4348 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4349 				     struct kvm_enable_cap *cap)
4350 {
4351 	int r;
4352 	uint16_t vmcs_version;
4353 	void __user *user_ptr;
4354 
4355 	if (cap->flags)
4356 		return -EINVAL;
4357 
4358 	switch (cap->cap) {
4359 	case KVM_CAP_HYPERV_SYNIC2:
4360 		if (cap->args[0])
4361 			return -EINVAL;
4362 		/* fall through */
4363 
4364 	case KVM_CAP_HYPERV_SYNIC:
4365 		if (!irqchip_in_kernel(vcpu->kvm))
4366 			return -EINVAL;
4367 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4368 					     KVM_CAP_HYPERV_SYNIC2);
4369 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4370 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4371 			return -ENOTTY;
4372 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4373 		if (!r) {
4374 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4375 			if (copy_to_user(user_ptr, &vmcs_version,
4376 					 sizeof(vmcs_version)))
4377 				r = -EFAULT;
4378 		}
4379 		return r;
4380 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4381 		if (!kvm_x86_ops.enable_direct_tlbflush)
4382 			return -ENOTTY;
4383 
4384 		return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4385 
4386 	default:
4387 		return -EINVAL;
4388 	}
4389 }
4390 
4391 long kvm_arch_vcpu_ioctl(struct file *filp,
4392 			 unsigned int ioctl, unsigned long arg)
4393 {
4394 	struct kvm_vcpu *vcpu = filp->private_data;
4395 	void __user *argp = (void __user *)arg;
4396 	int r;
4397 	union {
4398 		struct kvm_lapic_state *lapic;
4399 		struct kvm_xsave *xsave;
4400 		struct kvm_xcrs *xcrs;
4401 		void *buffer;
4402 	} u;
4403 
4404 	vcpu_load(vcpu);
4405 
4406 	u.buffer = NULL;
4407 	switch (ioctl) {
4408 	case KVM_GET_LAPIC: {
4409 		r = -EINVAL;
4410 		if (!lapic_in_kernel(vcpu))
4411 			goto out;
4412 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4413 				GFP_KERNEL_ACCOUNT);
4414 
4415 		r = -ENOMEM;
4416 		if (!u.lapic)
4417 			goto out;
4418 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4419 		if (r)
4420 			goto out;
4421 		r = -EFAULT;
4422 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4423 			goto out;
4424 		r = 0;
4425 		break;
4426 	}
4427 	case KVM_SET_LAPIC: {
4428 		r = -EINVAL;
4429 		if (!lapic_in_kernel(vcpu))
4430 			goto out;
4431 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4432 		if (IS_ERR(u.lapic)) {
4433 			r = PTR_ERR(u.lapic);
4434 			goto out_nofree;
4435 		}
4436 
4437 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4438 		break;
4439 	}
4440 	case KVM_INTERRUPT: {
4441 		struct kvm_interrupt irq;
4442 
4443 		r = -EFAULT;
4444 		if (copy_from_user(&irq, argp, sizeof(irq)))
4445 			goto out;
4446 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4447 		break;
4448 	}
4449 	case KVM_NMI: {
4450 		r = kvm_vcpu_ioctl_nmi(vcpu);
4451 		break;
4452 	}
4453 	case KVM_SMI: {
4454 		r = kvm_vcpu_ioctl_smi(vcpu);
4455 		break;
4456 	}
4457 	case KVM_SET_CPUID: {
4458 		struct kvm_cpuid __user *cpuid_arg = argp;
4459 		struct kvm_cpuid cpuid;
4460 
4461 		r = -EFAULT;
4462 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4463 			goto out;
4464 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4465 		break;
4466 	}
4467 	case KVM_SET_CPUID2: {
4468 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4469 		struct kvm_cpuid2 cpuid;
4470 
4471 		r = -EFAULT;
4472 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4473 			goto out;
4474 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4475 					      cpuid_arg->entries);
4476 		break;
4477 	}
4478 	case KVM_GET_CPUID2: {
4479 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4480 		struct kvm_cpuid2 cpuid;
4481 
4482 		r = -EFAULT;
4483 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4484 			goto out;
4485 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4486 					      cpuid_arg->entries);
4487 		if (r)
4488 			goto out;
4489 		r = -EFAULT;
4490 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4491 			goto out;
4492 		r = 0;
4493 		break;
4494 	}
4495 	case KVM_GET_MSRS: {
4496 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4497 		r = msr_io(vcpu, argp, do_get_msr, 1);
4498 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4499 		break;
4500 	}
4501 	case KVM_SET_MSRS: {
4502 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4503 		r = msr_io(vcpu, argp, do_set_msr, 0);
4504 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4505 		break;
4506 	}
4507 	case KVM_TPR_ACCESS_REPORTING: {
4508 		struct kvm_tpr_access_ctl tac;
4509 
4510 		r = -EFAULT;
4511 		if (copy_from_user(&tac, argp, sizeof(tac)))
4512 			goto out;
4513 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4514 		if (r)
4515 			goto out;
4516 		r = -EFAULT;
4517 		if (copy_to_user(argp, &tac, sizeof(tac)))
4518 			goto out;
4519 		r = 0;
4520 		break;
4521 	};
4522 	case KVM_SET_VAPIC_ADDR: {
4523 		struct kvm_vapic_addr va;
4524 		int idx;
4525 
4526 		r = -EINVAL;
4527 		if (!lapic_in_kernel(vcpu))
4528 			goto out;
4529 		r = -EFAULT;
4530 		if (copy_from_user(&va, argp, sizeof(va)))
4531 			goto out;
4532 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4533 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4534 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4535 		break;
4536 	}
4537 	case KVM_X86_SETUP_MCE: {
4538 		u64 mcg_cap;
4539 
4540 		r = -EFAULT;
4541 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4542 			goto out;
4543 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4544 		break;
4545 	}
4546 	case KVM_X86_SET_MCE: {
4547 		struct kvm_x86_mce mce;
4548 
4549 		r = -EFAULT;
4550 		if (copy_from_user(&mce, argp, sizeof(mce)))
4551 			goto out;
4552 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4553 		break;
4554 	}
4555 	case KVM_GET_VCPU_EVENTS: {
4556 		struct kvm_vcpu_events events;
4557 
4558 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4559 
4560 		r = -EFAULT;
4561 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4562 			break;
4563 		r = 0;
4564 		break;
4565 	}
4566 	case KVM_SET_VCPU_EVENTS: {
4567 		struct kvm_vcpu_events events;
4568 
4569 		r = -EFAULT;
4570 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4571 			break;
4572 
4573 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4574 		break;
4575 	}
4576 	case KVM_GET_DEBUGREGS: {
4577 		struct kvm_debugregs dbgregs;
4578 
4579 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4580 
4581 		r = -EFAULT;
4582 		if (copy_to_user(argp, &dbgregs,
4583 				 sizeof(struct kvm_debugregs)))
4584 			break;
4585 		r = 0;
4586 		break;
4587 	}
4588 	case KVM_SET_DEBUGREGS: {
4589 		struct kvm_debugregs dbgregs;
4590 
4591 		r = -EFAULT;
4592 		if (copy_from_user(&dbgregs, argp,
4593 				   sizeof(struct kvm_debugregs)))
4594 			break;
4595 
4596 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4597 		break;
4598 	}
4599 	case KVM_GET_XSAVE: {
4600 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4601 		r = -ENOMEM;
4602 		if (!u.xsave)
4603 			break;
4604 
4605 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4606 
4607 		r = -EFAULT;
4608 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4609 			break;
4610 		r = 0;
4611 		break;
4612 	}
4613 	case KVM_SET_XSAVE: {
4614 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4615 		if (IS_ERR(u.xsave)) {
4616 			r = PTR_ERR(u.xsave);
4617 			goto out_nofree;
4618 		}
4619 
4620 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4621 		break;
4622 	}
4623 	case KVM_GET_XCRS: {
4624 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4625 		r = -ENOMEM;
4626 		if (!u.xcrs)
4627 			break;
4628 
4629 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4630 
4631 		r = -EFAULT;
4632 		if (copy_to_user(argp, u.xcrs,
4633 				 sizeof(struct kvm_xcrs)))
4634 			break;
4635 		r = 0;
4636 		break;
4637 	}
4638 	case KVM_SET_XCRS: {
4639 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4640 		if (IS_ERR(u.xcrs)) {
4641 			r = PTR_ERR(u.xcrs);
4642 			goto out_nofree;
4643 		}
4644 
4645 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4646 		break;
4647 	}
4648 	case KVM_SET_TSC_KHZ: {
4649 		u32 user_tsc_khz;
4650 
4651 		r = -EINVAL;
4652 		user_tsc_khz = (u32)arg;
4653 
4654 		if (kvm_has_tsc_control &&
4655 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
4656 			goto out;
4657 
4658 		if (user_tsc_khz == 0)
4659 			user_tsc_khz = tsc_khz;
4660 
4661 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4662 			r = 0;
4663 
4664 		goto out;
4665 	}
4666 	case KVM_GET_TSC_KHZ: {
4667 		r = vcpu->arch.virtual_tsc_khz;
4668 		goto out;
4669 	}
4670 	case KVM_KVMCLOCK_CTRL: {
4671 		r = kvm_set_guest_paused(vcpu);
4672 		goto out;
4673 	}
4674 	case KVM_ENABLE_CAP: {
4675 		struct kvm_enable_cap cap;
4676 
4677 		r = -EFAULT;
4678 		if (copy_from_user(&cap, argp, sizeof(cap)))
4679 			goto out;
4680 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4681 		break;
4682 	}
4683 	case KVM_GET_NESTED_STATE: {
4684 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4685 		u32 user_data_size;
4686 
4687 		r = -EINVAL;
4688 		if (!kvm_x86_ops.nested_ops->get_state)
4689 			break;
4690 
4691 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4692 		r = -EFAULT;
4693 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4694 			break;
4695 
4696 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4697 						     user_data_size);
4698 		if (r < 0)
4699 			break;
4700 
4701 		if (r > user_data_size) {
4702 			if (put_user(r, &user_kvm_nested_state->size))
4703 				r = -EFAULT;
4704 			else
4705 				r = -E2BIG;
4706 			break;
4707 		}
4708 
4709 		r = 0;
4710 		break;
4711 	}
4712 	case KVM_SET_NESTED_STATE: {
4713 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4714 		struct kvm_nested_state kvm_state;
4715 		int idx;
4716 
4717 		r = -EINVAL;
4718 		if (!kvm_x86_ops.nested_ops->set_state)
4719 			break;
4720 
4721 		r = -EFAULT;
4722 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4723 			break;
4724 
4725 		r = -EINVAL;
4726 		if (kvm_state.size < sizeof(kvm_state))
4727 			break;
4728 
4729 		if (kvm_state.flags &
4730 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4731 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
4732 		      | KVM_STATE_NESTED_GIF_SET))
4733 			break;
4734 
4735 		/* nested_run_pending implies guest_mode.  */
4736 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4737 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4738 			break;
4739 
4740 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4741 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
4742 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4743 		break;
4744 	}
4745 	case KVM_GET_SUPPORTED_HV_CPUID: {
4746 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4747 		struct kvm_cpuid2 cpuid;
4748 
4749 		r = -EFAULT;
4750 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4751 			goto out;
4752 
4753 		r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4754 						cpuid_arg->entries);
4755 		if (r)
4756 			goto out;
4757 
4758 		r = -EFAULT;
4759 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4760 			goto out;
4761 		r = 0;
4762 		break;
4763 	}
4764 	default:
4765 		r = -EINVAL;
4766 	}
4767 out:
4768 	kfree(u.buffer);
4769 out_nofree:
4770 	vcpu_put(vcpu);
4771 	return r;
4772 }
4773 
4774 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4775 {
4776 	return VM_FAULT_SIGBUS;
4777 }
4778 
4779 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4780 {
4781 	int ret;
4782 
4783 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4784 		return -EINVAL;
4785 	ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4786 	return ret;
4787 }
4788 
4789 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4790 					      u64 ident_addr)
4791 {
4792 	return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4793 }
4794 
4795 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4796 					 unsigned long kvm_nr_mmu_pages)
4797 {
4798 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4799 		return -EINVAL;
4800 
4801 	mutex_lock(&kvm->slots_lock);
4802 
4803 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4804 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4805 
4806 	mutex_unlock(&kvm->slots_lock);
4807 	return 0;
4808 }
4809 
4810 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4811 {
4812 	return kvm->arch.n_max_mmu_pages;
4813 }
4814 
4815 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4816 {
4817 	struct kvm_pic *pic = kvm->arch.vpic;
4818 	int r;
4819 
4820 	r = 0;
4821 	switch (chip->chip_id) {
4822 	case KVM_IRQCHIP_PIC_MASTER:
4823 		memcpy(&chip->chip.pic, &pic->pics[0],
4824 			sizeof(struct kvm_pic_state));
4825 		break;
4826 	case KVM_IRQCHIP_PIC_SLAVE:
4827 		memcpy(&chip->chip.pic, &pic->pics[1],
4828 			sizeof(struct kvm_pic_state));
4829 		break;
4830 	case KVM_IRQCHIP_IOAPIC:
4831 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4832 		break;
4833 	default:
4834 		r = -EINVAL;
4835 		break;
4836 	}
4837 	return r;
4838 }
4839 
4840 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4841 {
4842 	struct kvm_pic *pic = kvm->arch.vpic;
4843 	int r;
4844 
4845 	r = 0;
4846 	switch (chip->chip_id) {
4847 	case KVM_IRQCHIP_PIC_MASTER:
4848 		spin_lock(&pic->lock);
4849 		memcpy(&pic->pics[0], &chip->chip.pic,
4850 			sizeof(struct kvm_pic_state));
4851 		spin_unlock(&pic->lock);
4852 		break;
4853 	case KVM_IRQCHIP_PIC_SLAVE:
4854 		spin_lock(&pic->lock);
4855 		memcpy(&pic->pics[1], &chip->chip.pic,
4856 			sizeof(struct kvm_pic_state));
4857 		spin_unlock(&pic->lock);
4858 		break;
4859 	case KVM_IRQCHIP_IOAPIC:
4860 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4861 		break;
4862 	default:
4863 		r = -EINVAL;
4864 		break;
4865 	}
4866 	kvm_pic_update_irq(pic);
4867 	return r;
4868 }
4869 
4870 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4871 {
4872 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4873 
4874 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4875 
4876 	mutex_lock(&kps->lock);
4877 	memcpy(ps, &kps->channels, sizeof(*ps));
4878 	mutex_unlock(&kps->lock);
4879 	return 0;
4880 }
4881 
4882 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4883 {
4884 	int i;
4885 	struct kvm_pit *pit = kvm->arch.vpit;
4886 
4887 	mutex_lock(&pit->pit_state.lock);
4888 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4889 	for (i = 0; i < 3; i++)
4890 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4891 	mutex_unlock(&pit->pit_state.lock);
4892 	return 0;
4893 }
4894 
4895 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4896 {
4897 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4898 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4899 		sizeof(ps->channels));
4900 	ps->flags = kvm->arch.vpit->pit_state.flags;
4901 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4902 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4903 	return 0;
4904 }
4905 
4906 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4907 {
4908 	int start = 0;
4909 	int i;
4910 	u32 prev_legacy, cur_legacy;
4911 	struct kvm_pit *pit = kvm->arch.vpit;
4912 
4913 	mutex_lock(&pit->pit_state.lock);
4914 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4915 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4916 	if (!prev_legacy && cur_legacy)
4917 		start = 1;
4918 	memcpy(&pit->pit_state.channels, &ps->channels,
4919 	       sizeof(pit->pit_state.channels));
4920 	pit->pit_state.flags = ps->flags;
4921 	for (i = 0; i < 3; i++)
4922 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4923 				   start && i == 0);
4924 	mutex_unlock(&pit->pit_state.lock);
4925 	return 0;
4926 }
4927 
4928 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4929 				 struct kvm_reinject_control *control)
4930 {
4931 	struct kvm_pit *pit = kvm->arch.vpit;
4932 
4933 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4934 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4935 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4936 	 */
4937 	mutex_lock(&pit->pit_state.lock);
4938 	kvm_pit_set_reinject(pit, control->pit_reinject);
4939 	mutex_unlock(&pit->pit_state.lock);
4940 
4941 	return 0;
4942 }
4943 
4944 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4945 {
4946 	/*
4947 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4948 	 */
4949 	if (kvm_x86_ops.flush_log_dirty)
4950 		kvm_x86_ops.flush_log_dirty(kvm);
4951 }
4952 
4953 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4954 			bool line_status)
4955 {
4956 	if (!irqchip_in_kernel(kvm))
4957 		return -ENXIO;
4958 
4959 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4960 					irq_event->irq, irq_event->level,
4961 					line_status);
4962 	return 0;
4963 }
4964 
4965 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4966 			    struct kvm_enable_cap *cap)
4967 {
4968 	int r;
4969 
4970 	if (cap->flags)
4971 		return -EINVAL;
4972 
4973 	switch (cap->cap) {
4974 	case KVM_CAP_DISABLE_QUIRKS:
4975 		kvm->arch.disabled_quirks = cap->args[0];
4976 		r = 0;
4977 		break;
4978 	case KVM_CAP_SPLIT_IRQCHIP: {
4979 		mutex_lock(&kvm->lock);
4980 		r = -EINVAL;
4981 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4982 			goto split_irqchip_unlock;
4983 		r = -EEXIST;
4984 		if (irqchip_in_kernel(kvm))
4985 			goto split_irqchip_unlock;
4986 		if (kvm->created_vcpus)
4987 			goto split_irqchip_unlock;
4988 		r = kvm_setup_empty_irq_routing(kvm);
4989 		if (r)
4990 			goto split_irqchip_unlock;
4991 		/* Pairs with irqchip_in_kernel. */
4992 		smp_wmb();
4993 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4994 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4995 		r = 0;
4996 split_irqchip_unlock:
4997 		mutex_unlock(&kvm->lock);
4998 		break;
4999 	}
5000 	case KVM_CAP_X2APIC_API:
5001 		r = -EINVAL;
5002 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5003 			break;
5004 
5005 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5006 			kvm->arch.x2apic_format = true;
5007 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5008 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5009 
5010 		r = 0;
5011 		break;
5012 	case KVM_CAP_X86_DISABLE_EXITS:
5013 		r = -EINVAL;
5014 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5015 			break;
5016 
5017 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5018 			kvm_can_mwait_in_guest())
5019 			kvm->arch.mwait_in_guest = true;
5020 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5021 			kvm->arch.hlt_in_guest = true;
5022 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5023 			kvm->arch.pause_in_guest = true;
5024 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5025 			kvm->arch.cstate_in_guest = true;
5026 		r = 0;
5027 		break;
5028 	case KVM_CAP_MSR_PLATFORM_INFO:
5029 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5030 		r = 0;
5031 		break;
5032 	case KVM_CAP_EXCEPTION_PAYLOAD:
5033 		kvm->arch.exception_payload_enabled = cap->args[0];
5034 		r = 0;
5035 		break;
5036 	default:
5037 		r = -EINVAL;
5038 		break;
5039 	}
5040 	return r;
5041 }
5042 
5043 long kvm_arch_vm_ioctl(struct file *filp,
5044 		       unsigned int ioctl, unsigned long arg)
5045 {
5046 	struct kvm *kvm = filp->private_data;
5047 	void __user *argp = (void __user *)arg;
5048 	int r = -ENOTTY;
5049 	/*
5050 	 * This union makes it completely explicit to gcc-3.x
5051 	 * that these two variables' stack usage should be
5052 	 * combined, not added together.
5053 	 */
5054 	union {
5055 		struct kvm_pit_state ps;
5056 		struct kvm_pit_state2 ps2;
5057 		struct kvm_pit_config pit_config;
5058 	} u;
5059 
5060 	switch (ioctl) {
5061 	case KVM_SET_TSS_ADDR:
5062 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5063 		break;
5064 	case KVM_SET_IDENTITY_MAP_ADDR: {
5065 		u64 ident_addr;
5066 
5067 		mutex_lock(&kvm->lock);
5068 		r = -EINVAL;
5069 		if (kvm->created_vcpus)
5070 			goto set_identity_unlock;
5071 		r = -EFAULT;
5072 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5073 			goto set_identity_unlock;
5074 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5075 set_identity_unlock:
5076 		mutex_unlock(&kvm->lock);
5077 		break;
5078 	}
5079 	case KVM_SET_NR_MMU_PAGES:
5080 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5081 		break;
5082 	case KVM_GET_NR_MMU_PAGES:
5083 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5084 		break;
5085 	case KVM_CREATE_IRQCHIP: {
5086 		mutex_lock(&kvm->lock);
5087 
5088 		r = -EEXIST;
5089 		if (irqchip_in_kernel(kvm))
5090 			goto create_irqchip_unlock;
5091 
5092 		r = -EINVAL;
5093 		if (kvm->created_vcpus)
5094 			goto create_irqchip_unlock;
5095 
5096 		r = kvm_pic_init(kvm);
5097 		if (r)
5098 			goto create_irqchip_unlock;
5099 
5100 		r = kvm_ioapic_init(kvm);
5101 		if (r) {
5102 			kvm_pic_destroy(kvm);
5103 			goto create_irqchip_unlock;
5104 		}
5105 
5106 		r = kvm_setup_default_irq_routing(kvm);
5107 		if (r) {
5108 			kvm_ioapic_destroy(kvm);
5109 			kvm_pic_destroy(kvm);
5110 			goto create_irqchip_unlock;
5111 		}
5112 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5113 		smp_wmb();
5114 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5115 	create_irqchip_unlock:
5116 		mutex_unlock(&kvm->lock);
5117 		break;
5118 	}
5119 	case KVM_CREATE_PIT:
5120 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5121 		goto create_pit;
5122 	case KVM_CREATE_PIT2:
5123 		r = -EFAULT;
5124 		if (copy_from_user(&u.pit_config, argp,
5125 				   sizeof(struct kvm_pit_config)))
5126 			goto out;
5127 	create_pit:
5128 		mutex_lock(&kvm->lock);
5129 		r = -EEXIST;
5130 		if (kvm->arch.vpit)
5131 			goto create_pit_unlock;
5132 		r = -ENOMEM;
5133 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5134 		if (kvm->arch.vpit)
5135 			r = 0;
5136 	create_pit_unlock:
5137 		mutex_unlock(&kvm->lock);
5138 		break;
5139 	case KVM_GET_IRQCHIP: {
5140 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5141 		struct kvm_irqchip *chip;
5142 
5143 		chip = memdup_user(argp, sizeof(*chip));
5144 		if (IS_ERR(chip)) {
5145 			r = PTR_ERR(chip);
5146 			goto out;
5147 		}
5148 
5149 		r = -ENXIO;
5150 		if (!irqchip_kernel(kvm))
5151 			goto get_irqchip_out;
5152 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5153 		if (r)
5154 			goto get_irqchip_out;
5155 		r = -EFAULT;
5156 		if (copy_to_user(argp, chip, sizeof(*chip)))
5157 			goto get_irqchip_out;
5158 		r = 0;
5159 	get_irqchip_out:
5160 		kfree(chip);
5161 		break;
5162 	}
5163 	case KVM_SET_IRQCHIP: {
5164 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5165 		struct kvm_irqchip *chip;
5166 
5167 		chip = memdup_user(argp, sizeof(*chip));
5168 		if (IS_ERR(chip)) {
5169 			r = PTR_ERR(chip);
5170 			goto out;
5171 		}
5172 
5173 		r = -ENXIO;
5174 		if (!irqchip_kernel(kvm))
5175 			goto set_irqchip_out;
5176 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5177 	set_irqchip_out:
5178 		kfree(chip);
5179 		break;
5180 	}
5181 	case KVM_GET_PIT: {
5182 		r = -EFAULT;
5183 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5184 			goto out;
5185 		r = -ENXIO;
5186 		if (!kvm->arch.vpit)
5187 			goto out;
5188 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5189 		if (r)
5190 			goto out;
5191 		r = -EFAULT;
5192 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5193 			goto out;
5194 		r = 0;
5195 		break;
5196 	}
5197 	case KVM_SET_PIT: {
5198 		r = -EFAULT;
5199 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5200 			goto out;
5201 		mutex_lock(&kvm->lock);
5202 		r = -ENXIO;
5203 		if (!kvm->arch.vpit)
5204 			goto set_pit_out;
5205 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5206 set_pit_out:
5207 		mutex_unlock(&kvm->lock);
5208 		break;
5209 	}
5210 	case KVM_GET_PIT2: {
5211 		r = -ENXIO;
5212 		if (!kvm->arch.vpit)
5213 			goto out;
5214 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5215 		if (r)
5216 			goto out;
5217 		r = -EFAULT;
5218 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5219 			goto out;
5220 		r = 0;
5221 		break;
5222 	}
5223 	case KVM_SET_PIT2: {
5224 		r = -EFAULT;
5225 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5226 			goto out;
5227 		mutex_lock(&kvm->lock);
5228 		r = -ENXIO;
5229 		if (!kvm->arch.vpit)
5230 			goto set_pit2_out;
5231 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5232 set_pit2_out:
5233 		mutex_unlock(&kvm->lock);
5234 		break;
5235 	}
5236 	case KVM_REINJECT_CONTROL: {
5237 		struct kvm_reinject_control control;
5238 		r =  -EFAULT;
5239 		if (copy_from_user(&control, argp, sizeof(control)))
5240 			goto out;
5241 		r = -ENXIO;
5242 		if (!kvm->arch.vpit)
5243 			goto out;
5244 		r = kvm_vm_ioctl_reinject(kvm, &control);
5245 		break;
5246 	}
5247 	case KVM_SET_BOOT_CPU_ID:
5248 		r = 0;
5249 		mutex_lock(&kvm->lock);
5250 		if (kvm->created_vcpus)
5251 			r = -EBUSY;
5252 		else
5253 			kvm->arch.bsp_vcpu_id = arg;
5254 		mutex_unlock(&kvm->lock);
5255 		break;
5256 	case KVM_XEN_HVM_CONFIG: {
5257 		struct kvm_xen_hvm_config xhc;
5258 		r = -EFAULT;
5259 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5260 			goto out;
5261 		r = -EINVAL;
5262 		if (xhc.flags)
5263 			goto out;
5264 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5265 		r = 0;
5266 		break;
5267 	}
5268 	case KVM_SET_CLOCK: {
5269 		struct kvm_clock_data user_ns;
5270 		u64 now_ns;
5271 
5272 		r = -EFAULT;
5273 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5274 			goto out;
5275 
5276 		r = -EINVAL;
5277 		if (user_ns.flags)
5278 			goto out;
5279 
5280 		r = 0;
5281 		/*
5282 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5283 		 * kvm_gen_update_masterclock() can be cut down to locked
5284 		 * pvclock_update_vm_gtod_copy().
5285 		 */
5286 		kvm_gen_update_masterclock(kvm);
5287 		now_ns = get_kvmclock_ns(kvm);
5288 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5289 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5290 		break;
5291 	}
5292 	case KVM_GET_CLOCK: {
5293 		struct kvm_clock_data user_ns;
5294 		u64 now_ns;
5295 
5296 		now_ns = get_kvmclock_ns(kvm);
5297 		user_ns.clock = now_ns;
5298 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5299 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5300 
5301 		r = -EFAULT;
5302 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5303 			goto out;
5304 		r = 0;
5305 		break;
5306 	}
5307 	case KVM_MEMORY_ENCRYPT_OP: {
5308 		r = -ENOTTY;
5309 		if (kvm_x86_ops.mem_enc_op)
5310 			r = kvm_x86_ops.mem_enc_op(kvm, argp);
5311 		break;
5312 	}
5313 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5314 		struct kvm_enc_region region;
5315 
5316 		r = -EFAULT;
5317 		if (copy_from_user(&region, argp, sizeof(region)))
5318 			goto out;
5319 
5320 		r = -ENOTTY;
5321 		if (kvm_x86_ops.mem_enc_reg_region)
5322 			r = kvm_x86_ops.mem_enc_reg_region(kvm, &region);
5323 		break;
5324 	}
5325 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5326 		struct kvm_enc_region region;
5327 
5328 		r = -EFAULT;
5329 		if (copy_from_user(&region, argp, sizeof(region)))
5330 			goto out;
5331 
5332 		r = -ENOTTY;
5333 		if (kvm_x86_ops.mem_enc_unreg_region)
5334 			r = kvm_x86_ops.mem_enc_unreg_region(kvm, &region);
5335 		break;
5336 	}
5337 	case KVM_HYPERV_EVENTFD: {
5338 		struct kvm_hyperv_eventfd hvevfd;
5339 
5340 		r = -EFAULT;
5341 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5342 			goto out;
5343 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5344 		break;
5345 	}
5346 	case KVM_SET_PMU_EVENT_FILTER:
5347 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5348 		break;
5349 	default:
5350 		r = -ENOTTY;
5351 	}
5352 out:
5353 	return r;
5354 }
5355 
5356 static void kvm_init_msr_list(void)
5357 {
5358 	struct x86_pmu_capability x86_pmu;
5359 	u32 dummy[2];
5360 	unsigned i;
5361 
5362 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5363 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5364 
5365 	perf_get_x86_pmu_capability(&x86_pmu);
5366 
5367 	num_msrs_to_save = 0;
5368 	num_emulated_msrs = 0;
5369 	num_msr_based_features = 0;
5370 
5371 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5372 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5373 			continue;
5374 
5375 		/*
5376 		 * Even MSRs that are valid in the host may not be exposed
5377 		 * to the guests in some cases.
5378 		 */
5379 		switch (msrs_to_save_all[i]) {
5380 		case MSR_IA32_BNDCFGS:
5381 			if (!kvm_mpx_supported())
5382 				continue;
5383 			break;
5384 		case MSR_TSC_AUX:
5385 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5386 				continue;
5387 			break;
5388 		case MSR_IA32_UMWAIT_CONTROL:
5389 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5390 				continue;
5391 			break;
5392 		case MSR_IA32_RTIT_CTL:
5393 		case MSR_IA32_RTIT_STATUS:
5394 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5395 				continue;
5396 			break;
5397 		case MSR_IA32_RTIT_CR3_MATCH:
5398 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5399 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5400 				continue;
5401 			break;
5402 		case MSR_IA32_RTIT_OUTPUT_BASE:
5403 		case MSR_IA32_RTIT_OUTPUT_MASK:
5404 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5405 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5406 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5407 				continue;
5408 			break;
5409 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5410 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5411 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5412 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5413 				continue;
5414 			break;
5415 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5416 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5417 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5418 				continue;
5419 			break;
5420 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5421 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5422 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5423 				continue;
5424 			break;
5425 		default:
5426 			break;
5427 		}
5428 
5429 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5430 	}
5431 
5432 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5433 		if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5434 			continue;
5435 
5436 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5437 	}
5438 
5439 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5440 		struct kvm_msr_entry msr;
5441 
5442 		msr.index = msr_based_features_all[i];
5443 		if (kvm_get_msr_feature(&msr))
5444 			continue;
5445 
5446 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5447 	}
5448 }
5449 
5450 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5451 			   const void *v)
5452 {
5453 	int handled = 0;
5454 	int n;
5455 
5456 	do {
5457 		n = min(len, 8);
5458 		if (!(lapic_in_kernel(vcpu) &&
5459 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5460 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5461 			break;
5462 		handled += n;
5463 		addr += n;
5464 		len -= n;
5465 		v += n;
5466 	} while (len);
5467 
5468 	return handled;
5469 }
5470 
5471 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5472 {
5473 	int handled = 0;
5474 	int n;
5475 
5476 	do {
5477 		n = min(len, 8);
5478 		if (!(lapic_in_kernel(vcpu) &&
5479 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5480 					 addr, n, v))
5481 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5482 			break;
5483 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5484 		handled += n;
5485 		addr += n;
5486 		len -= n;
5487 		v += n;
5488 	} while (len);
5489 
5490 	return handled;
5491 }
5492 
5493 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5494 			struct kvm_segment *var, int seg)
5495 {
5496 	kvm_x86_ops.set_segment(vcpu, var, seg);
5497 }
5498 
5499 void kvm_get_segment(struct kvm_vcpu *vcpu,
5500 		     struct kvm_segment *var, int seg)
5501 {
5502 	kvm_x86_ops.get_segment(vcpu, var, seg);
5503 }
5504 
5505 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5506 			   struct x86_exception *exception)
5507 {
5508 	gpa_t t_gpa;
5509 
5510 	BUG_ON(!mmu_is_nested(vcpu));
5511 
5512 	/* NPT walks are always user-walks */
5513 	access |= PFERR_USER_MASK;
5514 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5515 
5516 	return t_gpa;
5517 }
5518 
5519 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5520 			      struct x86_exception *exception)
5521 {
5522 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5523 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5524 }
5525 
5526  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5527 				struct x86_exception *exception)
5528 {
5529 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5530 	access |= PFERR_FETCH_MASK;
5531 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5532 }
5533 
5534 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5535 			       struct x86_exception *exception)
5536 {
5537 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5538 	access |= PFERR_WRITE_MASK;
5539 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5540 }
5541 
5542 /* uses this to access any guest's mapped memory without checking CPL */
5543 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5544 				struct x86_exception *exception)
5545 {
5546 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5547 }
5548 
5549 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5550 				      struct kvm_vcpu *vcpu, u32 access,
5551 				      struct x86_exception *exception)
5552 {
5553 	void *data = val;
5554 	int r = X86EMUL_CONTINUE;
5555 
5556 	while (bytes) {
5557 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5558 							    exception);
5559 		unsigned offset = addr & (PAGE_SIZE-1);
5560 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5561 		int ret;
5562 
5563 		if (gpa == UNMAPPED_GVA)
5564 			return X86EMUL_PROPAGATE_FAULT;
5565 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5566 					       offset, toread);
5567 		if (ret < 0) {
5568 			r = X86EMUL_IO_NEEDED;
5569 			goto out;
5570 		}
5571 
5572 		bytes -= toread;
5573 		data += toread;
5574 		addr += toread;
5575 	}
5576 out:
5577 	return r;
5578 }
5579 
5580 /* used for instruction fetching */
5581 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5582 				gva_t addr, void *val, unsigned int bytes,
5583 				struct x86_exception *exception)
5584 {
5585 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5586 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5587 	unsigned offset;
5588 	int ret;
5589 
5590 	/* Inline kvm_read_guest_virt_helper for speed.  */
5591 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5592 						    exception);
5593 	if (unlikely(gpa == UNMAPPED_GVA))
5594 		return X86EMUL_PROPAGATE_FAULT;
5595 
5596 	offset = addr & (PAGE_SIZE-1);
5597 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5598 		bytes = (unsigned)PAGE_SIZE - offset;
5599 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5600 				       offset, bytes);
5601 	if (unlikely(ret < 0))
5602 		return X86EMUL_IO_NEEDED;
5603 
5604 	return X86EMUL_CONTINUE;
5605 }
5606 
5607 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5608 			       gva_t addr, void *val, unsigned int bytes,
5609 			       struct x86_exception *exception)
5610 {
5611 	u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5612 
5613 	/*
5614 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5615 	 * is returned, but our callers are not ready for that and they blindly
5616 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
5617 	 * uninitialized kernel stack memory into cr2 and error code.
5618 	 */
5619 	memset(exception, 0, sizeof(*exception));
5620 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5621 					  exception);
5622 }
5623 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5624 
5625 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5626 			     gva_t addr, void *val, unsigned int bytes,
5627 			     struct x86_exception *exception, bool system)
5628 {
5629 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5630 	u32 access = 0;
5631 
5632 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5633 		access |= PFERR_USER_MASK;
5634 
5635 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5636 }
5637 
5638 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5639 		unsigned long addr, void *val, unsigned int bytes)
5640 {
5641 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5642 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5643 
5644 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5645 }
5646 
5647 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5648 				      struct kvm_vcpu *vcpu, u32 access,
5649 				      struct x86_exception *exception)
5650 {
5651 	void *data = val;
5652 	int r = X86EMUL_CONTINUE;
5653 
5654 	while (bytes) {
5655 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5656 							     access,
5657 							     exception);
5658 		unsigned offset = addr & (PAGE_SIZE-1);
5659 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5660 		int ret;
5661 
5662 		if (gpa == UNMAPPED_GVA)
5663 			return X86EMUL_PROPAGATE_FAULT;
5664 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5665 		if (ret < 0) {
5666 			r = X86EMUL_IO_NEEDED;
5667 			goto out;
5668 		}
5669 
5670 		bytes -= towrite;
5671 		data += towrite;
5672 		addr += towrite;
5673 	}
5674 out:
5675 	return r;
5676 }
5677 
5678 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5679 			      unsigned int bytes, struct x86_exception *exception,
5680 			      bool system)
5681 {
5682 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5683 	u32 access = PFERR_WRITE_MASK;
5684 
5685 	if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5686 		access |= PFERR_USER_MASK;
5687 
5688 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5689 					   access, exception);
5690 }
5691 
5692 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5693 				unsigned int bytes, struct x86_exception *exception)
5694 {
5695 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5696 	vcpu->arch.l1tf_flush_l1d = true;
5697 
5698 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5699 					   PFERR_WRITE_MASK, exception);
5700 }
5701 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5702 
5703 int handle_ud(struct kvm_vcpu *vcpu)
5704 {
5705 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5706 	int emul_type = EMULTYPE_TRAP_UD;
5707 	char sig[5]; /* ud2; .ascii "kvm" */
5708 	struct x86_exception e;
5709 
5710 	if (force_emulation_prefix &&
5711 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5712 				sig, sizeof(sig), &e) == 0 &&
5713 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5714 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5715 		emul_type = EMULTYPE_TRAP_UD_FORCED;
5716 	}
5717 
5718 	return kvm_emulate_instruction(vcpu, emul_type);
5719 }
5720 EXPORT_SYMBOL_GPL(handle_ud);
5721 
5722 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5723 			    gpa_t gpa, bool write)
5724 {
5725 	/* For APIC access vmexit */
5726 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5727 		return 1;
5728 
5729 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5730 		trace_vcpu_match_mmio(gva, gpa, write, true);
5731 		return 1;
5732 	}
5733 
5734 	return 0;
5735 }
5736 
5737 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5738 				gpa_t *gpa, struct x86_exception *exception,
5739 				bool write)
5740 {
5741 	u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5742 		| (write ? PFERR_WRITE_MASK : 0);
5743 
5744 	/*
5745 	 * currently PKRU is only applied to ept enabled guest so
5746 	 * there is no pkey in EPT page table for L1 guest or EPT
5747 	 * shadow page table for L2 guest.
5748 	 */
5749 	if (vcpu_match_mmio_gva(vcpu, gva)
5750 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5751 				 vcpu->arch.mmio_access, 0, access)) {
5752 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5753 					(gva & (PAGE_SIZE - 1));
5754 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5755 		return 1;
5756 	}
5757 
5758 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5759 
5760 	if (*gpa == UNMAPPED_GVA)
5761 		return -1;
5762 
5763 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5764 }
5765 
5766 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5767 			const void *val, int bytes)
5768 {
5769 	int ret;
5770 
5771 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5772 	if (ret < 0)
5773 		return 0;
5774 	kvm_page_track_write(vcpu, gpa, val, bytes);
5775 	return 1;
5776 }
5777 
5778 struct read_write_emulator_ops {
5779 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5780 				  int bytes);
5781 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5782 				  void *val, int bytes);
5783 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5784 			       int bytes, void *val);
5785 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5786 				    void *val, int bytes);
5787 	bool write;
5788 };
5789 
5790 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5791 {
5792 	if (vcpu->mmio_read_completed) {
5793 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5794 			       vcpu->mmio_fragments[0].gpa, val);
5795 		vcpu->mmio_read_completed = 0;
5796 		return 1;
5797 	}
5798 
5799 	return 0;
5800 }
5801 
5802 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5803 			void *val, int bytes)
5804 {
5805 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5806 }
5807 
5808 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5809 			 void *val, int bytes)
5810 {
5811 	return emulator_write_phys(vcpu, gpa, val, bytes);
5812 }
5813 
5814 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5815 {
5816 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5817 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5818 }
5819 
5820 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5821 			  void *val, int bytes)
5822 {
5823 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5824 	return X86EMUL_IO_NEEDED;
5825 }
5826 
5827 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5828 			   void *val, int bytes)
5829 {
5830 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5831 
5832 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5833 	return X86EMUL_CONTINUE;
5834 }
5835 
5836 static const struct read_write_emulator_ops read_emultor = {
5837 	.read_write_prepare = read_prepare,
5838 	.read_write_emulate = read_emulate,
5839 	.read_write_mmio = vcpu_mmio_read,
5840 	.read_write_exit_mmio = read_exit_mmio,
5841 };
5842 
5843 static const struct read_write_emulator_ops write_emultor = {
5844 	.read_write_emulate = write_emulate,
5845 	.read_write_mmio = write_mmio,
5846 	.read_write_exit_mmio = write_exit_mmio,
5847 	.write = true,
5848 };
5849 
5850 static int emulator_read_write_onepage(unsigned long addr, void *val,
5851 				       unsigned int bytes,
5852 				       struct x86_exception *exception,
5853 				       struct kvm_vcpu *vcpu,
5854 				       const struct read_write_emulator_ops *ops)
5855 {
5856 	gpa_t gpa;
5857 	int handled, ret;
5858 	bool write = ops->write;
5859 	struct kvm_mmio_fragment *frag;
5860 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5861 
5862 	/*
5863 	 * If the exit was due to a NPF we may already have a GPA.
5864 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5865 	 * Note, this cannot be used on string operations since string
5866 	 * operation using rep will only have the initial GPA from the NPF
5867 	 * occurred.
5868 	 */
5869 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5870 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5871 		gpa = ctxt->gpa_val;
5872 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5873 	} else {
5874 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5875 		if (ret < 0)
5876 			return X86EMUL_PROPAGATE_FAULT;
5877 	}
5878 
5879 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5880 		return X86EMUL_CONTINUE;
5881 
5882 	/*
5883 	 * Is this MMIO handled locally?
5884 	 */
5885 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5886 	if (handled == bytes)
5887 		return X86EMUL_CONTINUE;
5888 
5889 	gpa += handled;
5890 	bytes -= handled;
5891 	val += handled;
5892 
5893 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5894 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5895 	frag->gpa = gpa;
5896 	frag->data = val;
5897 	frag->len = bytes;
5898 	return X86EMUL_CONTINUE;
5899 }
5900 
5901 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5902 			unsigned long addr,
5903 			void *val, unsigned int bytes,
5904 			struct x86_exception *exception,
5905 			const struct read_write_emulator_ops *ops)
5906 {
5907 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5908 	gpa_t gpa;
5909 	int rc;
5910 
5911 	if (ops->read_write_prepare &&
5912 		  ops->read_write_prepare(vcpu, val, bytes))
5913 		return X86EMUL_CONTINUE;
5914 
5915 	vcpu->mmio_nr_fragments = 0;
5916 
5917 	/* Crossing a page boundary? */
5918 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5919 		int now;
5920 
5921 		now = -addr & ~PAGE_MASK;
5922 		rc = emulator_read_write_onepage(addr, val, now, exception,
5923 						 vcpu, ops);
5924 
5925 		if (rc != X86EMUL_CONTINUE)
5926 			return rc;
5927 		addr += now;
5928 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5929 			addr = (u32)addr;
5930 		val += now;
5931 		bytes -= now;
5932 	}
5933 
5934 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5935 					 vcpu, ops);
5936 	if (rc != X86EMUL_CONTINUE)
5937 		return rc;
5938 
5939 	if (!vcpu->mmio_nr_fragments)
5940 		return rc;
5941 
5942 	gpa = vcpu->mmio_fragments[0].gpa;
5943 
5944 	vcpu->mmio_needed = 1;
5945 	vcpu->mmio_cur_fragment = 0;
5946 
5947 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5948 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5949 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5950 	vcpu->run->mmio.phys_addr = gpa;
5951 
5952 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5953 }
5954 
5955 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5956 				  unsigned long addr,
5957 				  void *val,
5958 				  unsigned int bytes,
5959 				  struct x86_exception *exception)
5960 {
5961 	return emulator_read_write(ctxt, addr, val, bytes,
5962 				   exception, &read_emultor);
5963 }
5964 
5965 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5966 			    unsigned long addr,
5967 			    const void *val,
5968 			    unsigned int bytes,
5969 			    struct x86_exception *exception)
5970 {
5971 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5972 				   exception, &write_emultor);
5973 }
5974 
5975 #define CMPXCHG_TYPE(t, ptr, old, new) \
5976 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5977 
5978 #ifdef CONFIG_X86_64
5979 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5980 #else
5981 #  define CMPXCHG64(ptr, old, new) \
5982 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5983 #endif
5984 
5985 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5986 				     unsigned long addr,
5987 				     const void *old,
5988 				     const void *new,
5989 				     unsigned int bytes,
5990 				     struct x86_exception *exception)
5991 {
5992 	struct kvm_host_map map;
5993 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5994 	u64 page_line_mask;
5995 	gpa_t gpa;
5996 	char *kaddr;
5997 	bool exchanged;
5998 
5999 	/* guests cmpxchg8b have to be emulated atomically */
6000 	if (bytes > 8 || (bytes & (bytes - 1)))
6001 		goto emul_write;
6002 
6003 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6004 
6005 	if (gpa == UNMAPPED_GVA ||
6006 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6007 		goto emul_write;
6008 
6009 	/*
6010 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6011 	 * enabled in the host and the access splits a cache line.
6012 	 */
6013 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6014 		page_line_mask = ~(cache_line_size() - 1);
6015 	else
6016 		page_line_mask = PAGE_MASK;
6017 
6018 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6019 		goto emul_write;
6020 
6021 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6022 		goto emul_write;
6023 
6024 	kaddr = map.hva + offset_in_page(gpa);
6025 
6026 	switch (bytes) {
6027 	case 1:
6028 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6029 		break;
6030 	case 2:
6031 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6032 		break;
6033 	case 4:
6034 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6035 		break;
6036 	case 8:
6037 		exchanged = CMPXCHG64(kaddr, old, new);
6038 		break;
6039 	default:
6040 		BUG();
6041 	}
6042 
6043 	kvm_vcpu_unmap(vcpu, &map, true);
6044 
6045 	if (!exchanged)
6046 		return X86EMUL_CMPXCHG_FAILED;
6047 
6048 	kvm_page_track_write(vcpu, gpa, new, bytes);
6049 
6050 	return X86EMUL_CONTINUE;
6051 
6052 emul_write:
6053 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6054 
6055 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6056 }
6057 
6058 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6059 {
6060 	int r = 0, i;
6061 
6062 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6063 		if (vcpu->arch.pio.in)
6064 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6065 					    vcpu->arch.pio.size, pd);
6066 		else
6067 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6068 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6069 					     pd);
6070 		if (r)
6071 			break;
6072 		pd += vcpu->arch.pio.size;
6073 	}
6074 	return r;
6075 }
6076 
6077 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6078 			       unsigned short port, void *val,
6079 			       unsigned int count, bool in)
6080 {
6081 	vcpu->arch.pio.port = port;
6082 	vcpu->arch.pio.in = in;
6083 	vcpu->arch.pio.count  = count;
6084 	vcpu->arch.pio.size = size;
6085 
6086 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6087 		vcpu->arch.pio.count = 0;
6088 		return 1;
6089 	}
6090 
6091 	vcpu->run->exit_reason = KVM_EXIT_IO;
6092 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6093 	vcpu->run->io.size = size;
6094 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6095 	vcpu->run->io.count = count;
6096 	vcpu->run->io.port = port;
6097 
6098 	return 0;
6099 }
6100 
6101 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6102 			   unsigned short port, void *val, unsigned int count)
6103 {
6104 	int ret;
6105 
6106 	if (vcpu->arch.pio.count)
6107 		goto data_avail;
6108 
6109 	memset(vcpu->arch.pio_data, 0, size * count);
6110 
6111 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6112 	if (ret) {
6113 data_avail:
6114 		memcpy(val, vcpu->arch.pio_data, size * count);
6115 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6116 		vcpu->arch.pio.count = 0;
6117 		return 1;
6118 	}
6119 
6120 	return 0;
6121 }
6122 
6123 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6124 				    int size, unsigned short port, void *val,
6125 				    unsigned int count)
6126 {
6127 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6128 
6129 }
6130 
6131 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6132 			    unsigned short port, const void *val,
6133 			    unsigned int count)
6134 {
6135 	memcpy(vcpu->arch.pio_data, val, size * count);
6136 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6137 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6138 }
6139 
6140 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6141 				     int size, unsigned short port,
6142 				     const void *val, unsigned int count)
6143 {
6144 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6145 }
6146 
6147 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6148 {
6149 	return kvm_x86_ops.get_segment_base(vcpu, seg);
6150 }
6151 
6152 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6153 {
6154 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6155 }
6156 
6157 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6158 {
6159 	if (!need_emulate_wbinvd(vcpu))
6160 		return X86EMUL_CONTINUE;
6161 
6162 	if (kvm_x86_ops.has_wbinvd_exit()) {
6163 		int cpu = get_cpu();
6164 
6165 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6166 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6167 				wbinvd_ipi, NULL, 1);
6168 		put_cpu();
6169 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6170 	} else
6171 		wbinvd();
6172 	return X86EMUL_CONTINUE;
6173 }
6174 
6175 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6176 {
6177 	kvm_emulate_wbinvd_noskip(vcpu);
6178 	return kvm_skip_emulated_instruction(vcpu);
6179 }
6180 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6181 
6182 
6183 
6184 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6185 {
6186 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6187 }
6188 
6189 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6190 			   unsigned long *dest)
6191 {
6192 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6193 }
6194 
6195 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6196 			   unsigned long value)
6197 {
6198 
6199 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6200 }
6201 
6202 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6203 {
6204 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6205 }
6206 
6207 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6208 {
6209 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6210 	unsigned long value;
6211 
6212 	switch (cr) {
6213 	case 0:
6214 		value = kvm_read_cr0(vcpu);
6215 		break;
6216 	case 2:
6217 		value = vcpu->arch.cr2;
6218 		break;
6219 	case 3:
6220 		value = kvm_read_cr3(vcpu);
6221 		break;
6222 	case 4:
6223 		value = kvm_read_cr4(vcpu);
6224 		break;
6225 	case 8:
6226 		value = kvm_get_cr8(vcpu);
6227 		break;
6228 	default:
6229 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6230 		return 0;
6231 	}
6232 
6233 	return value;
6234 }
6235 
6236 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6237 {
6238 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6239 	int res = 0;
6240 
6241 	switch (cr) {
6242 	case 0:
6243 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6244 		break;
6245 	case 2:
6246 		vcpu->arch.cr2 = val;
6247 		break;
6248 	case 3:
6249 		res = kvm_set_cr3(vcpu, val);
6250 		break;
6251 	case 4:
6252 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6253 		break;
6254 	case 8:
6255 		res = kvm_set_cr8(vcpu, val);
6256 		break;
6257 	default:
6258 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6259 		res = -1;
6260 	}
6261 
6262 	return res;
6263 }
6264 
6265 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6266 {
6267 	return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6268 }
6269 
6270 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6271 {
6272 	kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6273 }
6274 
6275 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6276 {
6277 	kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6278 }
6279 
6280 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6281 {
6282 	kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6283 }
6284 
6285 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6286 {
6287 	kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6288 }
6289 
6290 static unsigned long emulator_get_cached_segment_base(
6291 	struct x86_emulate_ctxt *ctxt, int seg)
6292 {
6293 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6294 }
6295 
6296 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6297 				 struct desc_struct *desc, u32 *base3,
6298 				 int seg)
6299 {
6300 	struct kvm_segment var;
6301 
6302 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6303 	*selector = var.selector;
6304 
6305 	if (var.unusable) {
6306 		memset(desc, 0, sizeof(*desc));
6307 		if (base3)
6308 			*base3 = 0;
6309 		return false;
6310 	}
6311 
6312 	if (var.g)
6313 		var.limit >>= 12;
6314 	set_desc_limit(desc, var.limit);
6315 	set_desc_base(desc, (unsigned long)var.base);
6316 #ifdef CONFIG_X86_64
6317 	if (base3)
6318 		*base3 = var.base >> 32;
6319 #endif
6320 	desc->type = var.type;
6321 	desc->s = var.s;
6322 	desc->dpl = var.dpl;
6323 	desc->p = var.present;
6324 	desc->avl = var.avl;
6325 	desc->l = var.l;
6326 	desc->d = var.db;
6327 	desc->g = var.g;
6328 
6329 	return true;
6330 }
6331 
6332 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6333 				 struct desc_struct *desc, u32 base3,
6334 				 int seg)
6335 {
6336 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6337 	struct kvm_segment var;
6338 
6339 	var.selector = selector;
6340 	var.base = get_desc_base(desc);
6341 #ifdef CONFIG_X86_64
6342 	var.base |= ((u64)base3) << 32;
6343 #endif
6344 	var.limit = get_desc_limit(desc);
6345 	if (desc->g)
6346 		var.limit = (var.limit << 12) | 0xfff;
6347 	var.type = desc->type;
6348 	var.dpl = desc->dpl;
6349 	var.db = desc->d;
6350 	var.s = desc->s;
6351 	var.l = desc->l;
6352 	var.g = desc->g;
6353 	var.avl = desc->avl;
6354 	var.present = desc->p;
6355 	var.unusable = !var.present;
6356 	var.padding = 0;
6357 
6358 	kvm_set_segment(vcpu, &var, seg);
6359 	return;
6360 }
6361 
6362 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6363 			    u32 msr_index, u64 *pdata)
6364 {
6365 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6366 }
6367 
6368 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6369 			    u32 msr_index, u64 data)
6370 {
6371 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6372 }
6373 
6374 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6375 {
6376 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6377 
6378 	return vcpu->arch.smbase;
6379 }
6380 
6381 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6382 {
6383 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6384 
6385 	vcpu->arch.smbase = smbase;
6386 }
6387 
6388 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6389 			      u32 pmc)
6390 {
6391 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6392 }
6393 
6394 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6395 			     u32 pmc, u64 *pdata)
6396 {
6397 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6398 }
6399 
6400 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6401 {
6402 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
6403 }
6404 
6405 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6406 			      struct x86_instruction_info *info,
6407 			      enum x86_intercept_stage stage)
6408 {
6409 	return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6410 					    &ctxt->exception);
6411 }
6412 
6413 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6414 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6415 			      bool exact_only)
6416 {
6417 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6418 }
6419 
6420 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6421 {
6422 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6423 }
6424 
6425 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6426 {
6427 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6428 }
6429 
6430 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6431 {
6432 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6433 }
6434 
6435 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6436 {
6437 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
6438 }
6439 
6440 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6441 {
6442 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6443 }
6444 
6445 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6446 {
6447 	kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6448 }
6449 
6450 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6451 {
6452 	return emul_to_vcpu(ctxt)->arch.hflags;
6453 }
6454 
6455 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6456 {
6457 	emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6458 }
6459 
6460 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6461 				  const char *smstate)
6462 {
6463 	return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6464 }
6465 
6466 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6467 {
6468 	kvm_smm_changed(emul_to_vcpu(ctxt));
6469 }
6470 
6471 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6472 {
6473 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6474 }
6475 
6476 static const struct x86_emulate_ops emulate_ops = {
6477 	.read_gpr            = emulator_read_gpr,
6478 	.write_gpr           = emulator_write_gpr,
6479 	.read_std            = emulator_read_std,
6480 	.write_std           = emulator_write_std,
6481 	.read_phys           = kvm_read_guest_phys_system,
6482 	.fetch               = kvm_fetch_guest_virt,
6483 	.read_emulated       = emulator_read_emulated,
6484 	.write_emulated      = emulator_write_emulated,
6485 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
6486 	.invlpg              = emulator_invlpg,
6487 	.pio_in_emulated     = emulator_pio_in_emulated,
6488 	.pio_out_emulated    = emulator_pio_out_emulated,
6489 	.get_segment         = emulator_get_segment,
6490 	.set_segment         = emulator_set_segment,
6491 	.get_cached_segment_base = emulator_get_cached_segment_base,
6492 	.get_gdt             = emulator_get_gdt,
6493 	.get_idt	     = emulator_get_idt,
6494 	.set_gdt             = emulator_set_gdt,
6495 	.set_idt	     = emulator_set_idt,
6496 	.get_cr              = emulator_get_cr,
6497 	.set_cr              = emulator_set_cr,
6498 	.cpl                 = emulator_get_cpl,
6499 	.get_dr              = emulator_get_dr,
6500 	.set_dr              = emulator_set_dr,
6501 	.get_smbase          = emulator_get_smbase,
6502 	.set_smbase          = emulator_set_smbase,
6503 	.set_msr             = emulator_set_msr,
6504 	.get_msr             = emulator_get_msr,
6505 	.check_pmc	     = emulator_check_pmc,
6506 	.read_pmc            = emulator_read_pmc,
6507 	.halt                = emulator_halt,
6508 	.wbinvd              = emulator_wbinvd,
6509 	.fix_hypercall       = emulator_fix_hypercall,
6510 	.intercept           = emulator_intercept,
6511 	.get_cpuid           = emulator_get_cpuid,
6512 	.guest_has_long_mode = emulator_guest_has_long_mode,
6513 	.guest_has_movbe     = emulator_guest_has_movbe,
6514 	.guest_has_fxsr      = emulator_guest_has_fxsr,
6515 	.set_nmi_mask        = emulator_set_nmi_mask,
6516 	.get_hflags          = emulator_get_hflags,
6517 	.set_hflags          = emulator_set_hflags,
6518 	.pre_leave_smm       = emulator_pre_leave_smm,
6519 	.post_leave_smm      = emulator_post_leave_smm,
6520 	.set_xcr             = emulator_set_xcr,
6521 };
6522 
6523 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6524 {
6525 	u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6526 	/*
6527 	 * an sti; sti; sequence only disable interrupts for the first
6528 	 * instruction. So, if the last instruction, be it emulated or
6529 	 * not, left the system with the INT_STI flag enabled, it
6530 	 * means that the last instruction is an sti. We should not
6531 	 * leave the flag on in this case. The same goes for mov ss
6532 	 */
6533 	if (int_shadow & mask)
6534 		mask = 0;
6535 	if (unlikely(int_shadow || mask)) {
6536 		kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6537 		if (!mask)
6538 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6539 	}
6540 }
6541 
6542 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6543 {
6544 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6545 	if (ctxt->exception.vector == PF_VECTOR)
6546 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6547 
6548 	if (ctxt->exception.error_code_valid)
6549 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6550 				      ctxt->exception.error_code);
6551 	else
6552 		kvm_queue_exception(vcpu, ctxt->exception.vector);
6553 	return false;
6554 }
6555 
6556 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6557 {
6558 	struct x86_emulate_ctxt *ctxt;
6559 
6560 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6561 	if (!ctxt) {
6562 		pr_err("kvm: failed to allocate vcpu's emulator\n");
6563 		return NULL;
6564 	}
6565 
6566 	ctxt->vcpu = vcpu;
6567 	ctxt->ops = &emulate_ops;
6568 	vcpu->arch.emulate_ctxt = ctxt;
6569 
6570 	return ctxt;
6571 }
6572 
6573 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6574 {
6575 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6576 	int cs_db, cs_l;
6577 
6578 	kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6579 
6580 	ctxt->gpa_available = false;
6581 	ctxt->eflags = kvm_get_rflags(vcpu);
6582 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6583 
6584 	ctxt->eip = kvm_rip_read(vcpu);
6585 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
6586 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
6587 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
6588 		     cs_db				? X86EMUL_MODE_PROT32 :
6589 							  X86EMUL_MODE_PROT16;
6590 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6591 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6592 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6593 
6594 	init_decode_cache(ctxt);
6595 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6596 }
6597 
6598 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6599 {
6600 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6601 	int ret;
6602 
6603 	init_emulate_ctxt(vcpu);
6604 
6605 	ctxt->op_bytes = 2;
6606 	ctxt->ad_bytes = 2;
6607 	ctxt->_eip = ctxt->eip + inc_eip;
6608 	ret = emulate_int_real(ctxt, irq);
6609 
6610 	if (ret != X86EMUL_CONTINUE) {
6611 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6612 	} else {
6613 		ctxt->eip = ctxt->_eip;
6614 		kvm_rip_write(vcpu, ctxt->eip);
6615 		kvm_set_rflags(vcpu, ctxt->eflags);
6616 	}
6617 }
6618 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6619 
6620 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6621 {
6622 	++vcpu->stat.insn_emulation_fail;
6623 	trace_kvm_emulate_insn_failed(vcpu);
6624 
6625 	if (emulation_type & EMULTYPE_VMWARE_GP) {
6626 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6627 		return 1;
6628 	}
6629 
6630 	if (emulation_type & EMULTYPE_SKIP) {
6631 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6632 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6633 		vcpu->run->internal.ndata = 0;
6634 		return 0;
6635 	}
6636 
6637 	kvm_queue_exception(vcpu, UD_VECTOR);
6638 
6639 	if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6640 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6641 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6642 		vcpu->run->internal.ndata = 0;
6643 		return 0;
6644 	}
6645 
6646 	return 1;
6647 }
6648 
6649 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6650 				  bool write_fault_to_shadow_pgtable,
6651 				  int emulation_type)
6652 {
6653 	gpa_t gpa = cr2_or_gpa;
6654 	kvm_pfn_t pfn;
6655 
6656 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6657 		return false;
6658 
6659 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6660 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6661 		return false;
6662 
6663 	if (!vcpu->arch.mmu->direct_map) {
6664 		/*
6665 		 * Write permission should be allowed since only
6666 		 * write access need to be emulated.
6667 		 */
6668 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6669 
6670 		/*
6671 		 * If the mapping is invalid in guest, let cpu retry
6672 		 * it to generate fault.
6673 		 */
6674 		if (gpa == UNMAPPED_GVA)
6675 			return true;
6676 	}
6677 
6678 	/*
6679 	 * Do not retry the unhandleable instruction if it faults on the
6680 	 * readonly host memory, otherwise it will goto a infinite loop:
6681 	 * retry instruction -> write #PF -> emulation fail -> retry
6682 	 * instruction -> ...
6683 	 */
6684 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6685 
6686 	/*
6687 	 * If the instruction failed on the error pfn, it can not be fixed,
6688 	 * report the error to userspace.
6689 	 */
6690 	if (is_error_noslot_pfn(pfn))
6691 		return false;
6692 
6693 	kvm_release_pfn_clean(pfn);
6694 
6695 	/* The instructions are well-emulated on direct mmu. */
6696 	if (vcpu->arch.mmu->direct_map) {
6697 		unsigned int indirect_shadow_pages;
6698 
6699 		spin_lock(&vcpu->kvm->mmu_lock);
6700 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6701 		spin_unlock(&vcpu->kvm->mmu_lock);
6702 
6703 		if (indirect_shadow_pages)
6704 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6705 
6706 		return true;
6707 	}
6708 
6709 	/*
6710 	 * if emulation was due to access to shadowed page table
6711 	 * and it failed try to unshadow page and re-enter the
6712 	 * guest to let CPU execute the instruction.
6713 	 */
6714 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6715 
6716 	/*
6717 	 * If the access faults on its page table, it can not
6718 	 * be fixed by unprotecting shadow page and it should
6719 	 * be reported to userspace.
6720 	 */
6721 	return !write_fault_to_shadow_pgtable;
6722 }
6723 
6724 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6725 			      gpa_t cr2_or_gpa,  int emulation_type)
6726 {
6727 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6728 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6729 
6730 	last_retry_eip = vcpu->arch.last_retry_eip;
6731 	last_retry_addr = vcpu->arch.last_retry_addr;
6732 
6733 	/*
6734 	 * If the emulation is caused by #PF and it is non-page_table
6735 	 * writing instruction, it means the VM-EXIT is caused by shadow
6736 	 * page protected, we can zap the shadow page and retry this
6737 	 * instruction directly.
6738 	 *
6739 	 * Note: if the guest uses a non-page-table modifying instruction
6740 	 * on the PDE that points to the instruction, then we will unmap
6741 	 * the instruction and go to an infinite loop. So, we cache the
6742 	 * last retried eip and the last fault address, if we meet the eip
6743 	 * and the address again, we can break out of the potential infinite
6744 	 * loop.
6745 	 */
6746 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6747 
6748 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6749 		return false;
6750 
6751 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6752 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6753 		return false;
6754 
6755 	if (x86_page_table_writing_insn(ctxt))
6756 		return false;
6757 
6758 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6759 		return false;
6760 
6761 	vcpu->arch.last_retry_eip = ctxt->eip;
6762 	vcpu->arch.last_retry_addr = cr2_or_gpa;
6763 
6764 	if (!vcpu->arch.mmu->direct_map)
6765 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6766 
6767 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6768 
6769 	return true;
6770 }
6771 
6772 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6773 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6774 
6775 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6776 {
6777 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6778 		/* This is a good place to trace that we are exiting SMM.  */
6779 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6780 
6781 		/* Process a latched INIT or SMI, if any.  */
6782 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6783 	}
6784 
6785 	kvm_mmu_reset_context(vcpu);
6786 }
6787 
6788 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6789 				unsigned long *db)
6790 {
6791 	u32 dr6 = 0;
6792 	int i;
6793 	u32 enable, rwlen;
6794 
6795 	enable = dr7;
6796 	rwlen = dr7 >> 16;
6797 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6798 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6799 			dr6 |= (1 << i);
6800 	return dr6;
6801 }
6802 
6803 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6804 {
6805 	struct kvm_run *kvm_run = vcpu->run;
6806 
6807 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6808 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6809 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6810 		kvm_run->debug.arch.exception = DB_VECTOR;
6811 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6812 		return 0;
6813 	}
6814 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6815 	return 1;
6816 }
6817 
6818 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6819 {
6820 	unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6821 	int r;
6822 
6823 	r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6824 	if (unlikely(!r))
6825 		return 0;
6826 
6827 	/*
6828 	 * rflags is the old, "raw" value of the flags.  The new value has
6829 	 * not been saved yet.
6830 	 *
6831 	 * This is correct even for TF set by the guest, because "the
6832 	 * processor will not generate this exception after the instruction
6833 	 * that sets the TF flag".
6834 	 */
6835 	if (unlikely(rflags & X86_EFLAGS_TF))
6836 		r = kvm_vcpu_do_singlestep(vcpu);
6837 	return r;
6838 }
6839 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6840 
6841 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6842 {
6843 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6844 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6845 		struct kvm_run *kvm_run = vcpu->run;
6846 		unsigned long eip = kvm_get_linear_rip(vcpu);
6847 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6848 					   vcpu->arch.guest_debug_dr7,
6849 					   vcpu->arch.eff_db);
6850 
6851 		if (dr6 != 0) {
6852 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6853 			kvm_run->debug.arch.pc = eip;
6854 			kvm_run->debug.arch.exception = DB_VECTOR;
6855 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6856 			*r = 0;
6857 			return true;
6858 		}
6859 	}
6860 
6861 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6862 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6863 		unsigned long eip = kvm_get_linear_rip(vcpu);
6864 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6865 					   vcpu->arch.dr7,
6866 					   vcpu->arch.db);
6867 
6868 		if (dr6 != 0) {
6869 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
6870 			*r = 1;
6871 			return true;
6872 		}
6873 	}
6874 
6875 	return false;
6876 }
6877 
6878 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6879 {
6880 	switch (ctxt->opcode_len) {
6881 	case 1:
6882 		switch (ctxt->b) {
6883 		case 0xe4:	/* IN */
6884 		case 0xe5:
6885 		case 0xec:
6886 		case 0xed:
6887 		case 0xe6:	/* OUT */
6888 		case 0xe7:
6889 		case 0xee:
6890 		case 0xef:
6891 		case 0x6c:	/* INS */
6892 		case 0x6d:
6893 		case 0x6e:	/* OUTS */
6894 		case 0x6f:
6895 			return true;
6896 		}
6897 		break;
6898 	case 2:
6899 		switch (ctxt->b) {
6900 		case 0x33:	/* RDPMC */
6901 			return true;
6902 		}
6903 		break;
6904 	}
6905 
6906 	return false;
6907 }
6908 
6909 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6910 			    int emulation_type, void *insn, int insn_len)
6911 {
6912 	int r;
6913 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6914 	bool writeback = true;
6915 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6916 
6917 	vcpu->arch.l1tf_flush_l1d = true;
6918 
6919 	/*
6920 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6921 	 * never reused.
6922 	 */
6923 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6924 	kvm_clear_exception_queue(vcpu);
6925 
6926 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6927 		init_emulate_ctxt(vcpu);
6928 
6929 		/*
6930 		 * We will reenter on the same instruction since
6931 		 * we do not set complete_userspace_io.  This does not
6932 		 * handle watchpoints yet, those would be handled in
6933 		 * the emulate_ops.
6934 		 */
6935 		if (!(emulation_type & EMULTYPE_SKIP) &&
6936 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6937 			return r;
6938 
6939 		ctxt->interruptibility = 0;
6940 		ctxt->have_exception = false;
6941 		ctxt->exception.vector = -1;
6942 		ctxt->perm_ok = false;
6943 
6944 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6945 
6946 		r = x86_decode_insn(ctxt, insn, insn_len);
6947 
6948 		trace_kvm_emulate_insn_start(vcpu);
6949 		++vcpu->stat.insn_emulation;
6950 		if (r != EMULATION_OK)  {
6951 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
6952 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6953 				kvm_queue_exception(vcpu, UD_VECTOR);
6954 				return 1;
6955 			}
6956 			if (reexecute_instruction(vcpu, cr2_or_gpa,
6957 						  write_fault_to_spt,
6958 						  emulation_type))
6959 				return 1;
6960 			if (ctxt->have_exception) {
6961 				/*
6962 				 * #UD should result in just EMULATION_FAILED, and trap-like
6963 				 * exception should not be encountered during decode.
6964 				 */
6965 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6966 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6967 				inject_emulated_exception(vcpu);
6968 				return 1;
6969 			}
6970 			return handle_emulation_failure(vcpu, emulation_type);
6971 		}
6972 	}
6973 
6974 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6975 	    !is_vmware_backdoor_opcode(ctxt)) {
6976 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6977 		return 1;
6978 	}
6979 
6980 	/*
6981 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6982 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
6983 	 * updating interruptibility state and injecting single-step #DBs.
6984 	 */
6985 	if (emulation_type & EMULTYPE_SKIP) {
6986 		kvm_rip_write(vcpu, ctxt->_eip);
6987 		if (ctxt->eflags & X86_EFLAGS_RF)
6988 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6989 		return 1;
6990 	}
6991 
6992 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6993 		return 1;
6994 
6995 	/* this is needed for vmware backdoor interface to work since it
6996 	   changes registers values  during IO operation */
6997 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6998 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6999 		emulator_invalidate_register_cache(ctxt);
7000 	}
7001 
7002 restart:
7003 	if (emulation_type & EMULTYPE_PF) {
7004 		/* Save the faulting GPA (cr2) in the address field */
7005 		ctxt->exception.address = cr2_or_gpa;
7006 
7007 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7008 		if (vcpu->arch.mmu->direct_map) {
7009 			ctxt->gpa_available = true;
7010 			ctxt->gpa_val = cr2_or_gpa;
7011 		}
7012 	} else {
7013 		/* Sanitize the address out of an abundance of paranoia. */
7014 		ctxt->exception.address = 0;
7015 	}
7016 
7017 	r = x86_emulate_insn(ctxt);
7018 
7019 	if (r == EMULATION_INTERCEPTED)
7020 		return 1;
7021 
7022 	if (r == EMULATION_FAILED) {
7023 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7024 					emulation_type))
7025 			return 1;
7026 
7027 		return handle_emulation_failure(vcpu, emulation_type);
7028 	}
7029 
7030 	if (ctxt->have_exception) {
7031 		r = 1;
7032 		if (inject_emulated_exception(vcpu))
7033 			return r;
7034 	} else if (vcpu->arch.pio.count) {
7035 		if (!vcpu->arch.pio.in) {
7036 			/* FIXME: return into emulator if single-stepping.  */
7037 			vcpu->arch.pio.count = 0;
7038 		} else {
7039 			writeback = false;
7040 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7041 		}
7042 		r = 0;
7043 	} else if (vcpu->mmio_needed) {
7044 		++vcpu->stat.mmio_exits;
7045 
7046 		if (!vcpu->mmio_is_write)
7047 			writeback = false;
7048 		r = 0;
7049 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7050 	} else if (r == EMULATION_RESTART)
7051 		goto restart;
7052 	else
7053 		r = 1;
7054 
7055 	if (writeback) {
7056 		unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7057 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7058 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7059 		if (!ctxt->have_exception ||
7060 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7061 			kvm_rip_write(vcpu, ctxt->eip);
7062 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7063 				r = kvm_vcpu_do_singlestep(vcpu);
7064 			if (kvm_x86_ops.update_emulated_instruction)
7065 				kvm_x86_ops.update_emulated_instruction(vcpu);
7066 			__kvm_set_rflags(vcpu, ctxt->eflags);
7067 		}
7068 
7069 		/*
7070 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7071 		 * do nothing, and it will be requested again as soon as
7072 		 * the shadow expires.  But we still need to check here,
7073 		 * because POPF has no interrupt shadow.
7074 		 */
7075 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7076 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7077 	} else
7078 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7079 
7080 	return r;
7081 }
7082 
7083 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7084 {
7085 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7086 }
7087 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7088 
7089 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7090 					void *insn, int insn_len)
7091 {
7092 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7093 }
7094 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7095 
7096 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7097 {
7098 	vcpu->arch.pio.count = 0;
7099 	return 1;
7100 }
7101 
7102 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7103 {
7104 	vcpu->arch.pio.count = 0;
7105 
7106 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7107 		return 1;
7108 
7109 	return kvm_skip_emulated_instruction(vcpu);
7110 }
7111 
7112 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7113 			    unsigned short port)
7114 {
7115 	unsigned long val = kvm_rax_read(vcpu);
7116 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7117 
7118 	if (ret)
7119 		return ret;
7120 
7121 	/*
7122 	 * Workaround userspace that relies on old KVM behavior of %rip being
7123 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7124 	 */
7125 	if (port == 0x7e &&
7126 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7127 		vcpu->arch.complete_userspace_io =
7128 			complete_fast_pio_out_port_0x7e;
7129 		kvm_skip_emulated_instruction(vcpu);
7130 	} else {
7131 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7132 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7133 	}
7134 	return 0;
7135 }
7136 
7137 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7138 {
7139 	unsigned long val;
7140 
7141 	/* We should only ever be called with arch.pio.count equal to 1 */
7142 	BUG_ON(vcpu->arch.pio.count != 1);
7143 
7144 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7145 		vcpu->arch.pio.count = 0;
7146 		return 1;
7147 	}
7148 
7149 	/* For size less than 4 we merge, else we zero extend */
7150 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7151 
7152 	/*
7153 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7154 	 * the copy and tracing
7155 	 */
7156 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7157 	kvm_rax_write(vcpu, val);
7158 
7159 	return kvm_skip_emulated_instruction(vcpu);
7160 }
7161 
7162 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7163 			   unsigned short port)
7164 {
7165 	unsigned long val;
7166 	int ret;
7167 
7168 	/* For size less than 4 we merge, else we zero extend */
7169 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7170 
7171 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7172 	if (ret) {
7173 		kvm_rax_write(vcpu, val);
7174 		return ret;
7175 	}
7176 
7177 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7178 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7179 
7180 	return 0;
7181 }
7182 
7183 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7184 {
7185 	int ret;
7186 
7187 	if (in)
7188 		ret = kvm_fast_pio_in(vcpu, size, port);
7189 	else
7190 		ret = kvm_fast_pio_out(vcpu, size, port);
7191 	return ret && kvm_skip_emulated_instruction(vcpu);
7192 }
7193 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7194 
7195 static int kvmclock_cpu_down_prep(unsigned int cpu)
7196 {
7197 	__this_cpu_write(cpu_tsc_khz, 0);
7198 	return 0;
7199 }
7200 
7201 static void tsc_khz_changed(void *data)
7202 {
7203 	struct cpufreq_freqs *freq = data;
7204 	unsigned long khz = 0;
7205 
7206 	if (data)
7207 		khz = freq->new;
7208 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7209 		khz = cpufreq_quick_get(raw_smp_processor_id());
7210 	if (!khz)
7211 		khz = tsc_khz;
7212 	__this_cpu_write(cpu_tsc_khz, khz);
7213 }
7214 
7215 #ifdef CONFIG_X86_64
7216 static void kvm_hyperv_tsc_notifier(void)
7217 {
7218 	struct kvm *kvm;
7219 	struct kvm_vcpu *vcpu;
7220 	int cpu;
7221 
7222 	mutex_lock(&kvm_lock);
7223 	list_for_each_entry(kvm, &vm_list, vm_list)
7224 		kvm_make_mclock_inprogress_request(kvm);
7225 
7226 	hyperv_stop_tsc_emulation();
7227 
7228 	/* TSC frequency always matches when on Hyper-V */
7229 	for_each_present_cpu(cpu)
7230 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7231 	kvm_max_guest_tsc_khz = tsc_khz;
7232 
7233 	list_for_each_entry(kvm, &vm_list, vm_list) {
7234 		struct kvm_arch *ka = &kvm->arch;
7235 
7236 		spin_lock(&ka->pvclock_gtod_sync_lock);
7237 
7238 		pvclock_update_vm_gtod_copy(kvm);
7239 
7240 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7241 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7242 
7243 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7244 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7245 
7246 		spin_unlock(&ka->pvclock_gtod_sync_lock);
7247 	}
7248 	mutex_unlock(&kvm_lock);
7249 }
7250 #endif
7251 
7252 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7253 {
7254 	struct kvm *kvm;
7255 	struct kvm_vcpu *vcpu;
7256 	int i, send_ipi = 0;
7257 
7258 	/*
7259 	 * We allow guests to temporarily run on slowing clocks,
7260 	 * provided we notify them after, or to run on accelerating
7261 	 * clocks, provided we notify them before.  Thus time never
7262 	 * goes backwards.
7263 	 *
7264 	 * However, we have a problem.  We can't atomically update
7265 	 * the frequency of a given CPU from this function; it is
7266 	 * merely a notifier, which can be called from any CPU.
7267 	 * Changing the TSC frequency at arbitrary points in time
7268 	 * requires a recomputation of local variables related to
7269 	 * the TSC for each VCPU.  We must flag these local variables
7270 	 * to be updated and be sure the update takes place with the
7271 	 * new frequency before any guests proceed.
7272 	 *
7273 	 * Unfortunately, the combination of hotplug CPU and frequency
7274 	 * change creates an intractable locking scenario; the order
7275 	 * of when these callouts happen is undefined with respect to
7276 	 * CPU hotplug, and they can race with each other.  As such,
7277 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7278 	 * undefined; you can actually have a CPU frequency change take
7279 	 * place in between the computation of X and the setting of the
7280 	 * variable.  To protect against this problem, all updates of
7281 	 * the per_cpu tsc_khz variable are done in an interrupt
7282 	 * protected IPI, and all callers wishing to update the value
7283 	 * must wait for a synchronous IPI to complete (which is trivial
7284 	 * if the caller is on the CPU already).  This establishes the
7285 	 * necessary total order on variable updates.
7286 	 *
7287 	 * Note that because a guest time update may take place
7288 	 * anytime after the setting of the VCPU's request bit, the
7289 	 * correct TSC value must be set before the request.  However,
7290 	 * to ensure the update actually makes it to any guest which
7291 	 * starts running in hardware virtualization between the set
7292 	 * and the acquisition of the spinlock, we must also ping the
7293 	 * CPU after setting the request bit.
7294 	 *
7295 	 */
7296 
7297 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7298 
7299 	mutex_lock(&kvm_lock);
7300 	list_for_each_entry(kvm, &vm_list, vm_list) {
7301 		kvm_for_each_vcpu(i, vcpu, kvm) {
7302 			if (vcpu->cpu != cpu)
7303 				continue;
7304 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7305 			if (vcpu->cpu != raw_smp_processor_id())
7306 				send_ipi = 1;
7307 		}
7308 	}
7309 	mutex_unlock(&kvm_lock);
7310 
7311 	if (freq->old < freq->new && send_ipi) {
7312 		/*
7313 		 * We upscale the frequency.  Must make the guest
7314 		 * doesn't see old kvmclock values while running with
7315 		 * the new frequency, otherwise we risk the guest sees
7316 		 * time go backwards.
7317 		 *
7318 		 * In case we update the frequency for another cpu
7319 		 * (which might be in guest context) send an interrupt
7320 		 * to kick the cpu out of guest context.  Next time
7321 		 * guest context is entered kvmclock will be updated,
7322 		 * so the guest will not see stale values.
7323 		 */
7324 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7325 	}
7326 }
7327 
7328 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7329 				     void *data)
7330 {
7331 	struct cpufreq_freqs *freq = data;
7332 	int cpu;
7333 
7334 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7335 		return 0;
7336 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7337 		return 0;
7338 
7339 	for_each_cpu(cpu, freq->policy->cpus)
7340 		__kvmclock_cpufreq_notifier(freq, cpu);
7341 
7342 	return 0;
7343 }
7344 
7345 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7346 	.notifier_call  = kvmclock_cpufreq_notifier
7347 };
7348 
7349 static int kvmclock_cpu_online(unsigned int cpu)
7350 {
7351 	tsc_khz_changed(NULL);
7352 	return 0;
7353 }
7354 
7355 static void kvm_timer_init(void)
7356 {
7357 	max_tsc_khz = tsc_khz;
7358 
7359 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7360 #ifdef CONFIG_CPU_FREQ
7361 		struct cpufreq_policy *policy;
7362 		int cpu;
7363 
7364 		cpu = get_cpu();
7365 		policy = cpufreq_cpu_get(cpu);
7366 		if (policy) {
7367 			if (policy->cpuinfo.max_freq)
7368 				max_tsc_khz = policy->cpuinfo.max_freq;
7369 			cpufreq_cpu_put(policy);
7370 		}
7371 		put_cpu();
7372 #endif
7373 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7374 					  CPUFREQ_TRANSITION_NOTIFIER);
7375 	}
7376 
7377 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7378 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
7379 }
7380 
7381 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7382 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7383 
7384 int kvm_is_in_guest(void)
7385 {
7386 	return __this_cpu_read(current_vcpu) != NULL;
7387 }
7388 
7389 static int kvm_is_user_mode(void)
7390 {
7391 	int user_mode = 3;
7392 
7393 	if (__this_cpu_read(current_vcpu))
7394 		user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7395 
7396 	return user_mode != 0;
7397 }
7398 
7399 static unsigned long kvm_get_guest_ip(void)
7400 {
7401 	unsigned long ip = 0;
7402 
7403 	if (__this_cpu_read(current_vcpu))
7404 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7405 
7406 	return ip;
7407 }
7408 
7409 static void kvm_handle_intel_pt_intr(void)
7410 {
7411 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7412 
7413 	kvm_make_request(KVM_REQ_PMI, vcpu);
7414 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7415 			(unsigned long *)&vcpu->arch.pmu.global_status);
7416 }
7417 
7418 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7419 	.is_in_guest		= kvm_is_in_guest,
7420 	.is_user_mode		= kvm_is_user_mode,
7421 	.get_guest_ip		= kvm_get_guest_ip,
7422 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
7423 };
7424 
7425 #ifdef CONFIG_X86_64
7426 static void pvclock_gtod_update_fn(struct work_struct *work)
7427 {
7428 	struct kvm *kvm;
7429 
7430 	struct kvm_vcpu *vcpu;
7431 	int i;
7432 
7433 	mutex_lock(&kvm_lock);
7434 	list_for_each_entry(kvm, &vm_list, vm_list)
7435 		kvm_for_each_vcpu(i, vcpu, kvm)
7436 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7437 	atomic_set(&kvm_guest_has_master_clock, 0);
7438 	mutex_unlock(&kvm_lock);
7439 }
7440 
7441 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7442 
7443 /*
7444  * Notification about pvclock gtod data update.
7445  */
7446 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7447 			       void *priv)
7448 {
7449 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7450 	struct timekeeper *tk = priv;
7451 
7452 	update_pvclock_gtod(tk);
7453 
7454 	/* disable master clock if host does not trust, or does not
7455 	 * use, TSC based clocksource.
7456 	 */
7457 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7458 	    atomic_read(&kvm_guest_has_master_clock) != 0)
7459 		queue_work(system_long_wq, &pvclock_gtod_work);
7460 
7461 	return 0;
7462 }
7463 
7464 static struct notifier_block pvclock_gtod_notifier = {
7465 	.notifier_call = pvclock_gtod_notify,
7466 };
7467 #endif
7468 
7469 int kvm_arch_init(void *opaque)
7470 {
7471 	struct kvm_x86_init_ops *ops = opaque;
7472 	int r;
7473 
7474 	if (kvm_x86_ops.hardware_enable) {
7475 		printk(KERN_ERR "kvm: already loaded the other module\n");
7476 		r = -EEXIST;
7477 		goto out;
7478 	}
7479 
7480 	if (!ops->cpu_has_kvm_support()) {
7481 		pr_err_ratelimited("kvm: no hardware support\n");
7482 		r = -EOPNOTSUPP;
7483 		goto out;
7484 	}
7485 	if (ops->disabled_by_bios()) {
7486 		pr_err_ratelimited("kvm: disabled by bios\n");
7487 		r = -EOPNOTSUPP;
7488 		goto out;
7489 	}
7490 
7491 	/*
7492 	 * KVM explicitly assumes that the guest has an FPU and
7493 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7494 	 * vCPU's FPU state as a fxregs_state struct.
7495 	 */
7496 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7497 		printk(KERN_ERR "kvm: inadequate fpu\n");
7498 		r = -EOPNOTSUPP;
7499 		goto out;
7500 	}
7501 
7502 	r = -ENOMEM;
7503 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7504 					  __alignof__(struct fpu), SLAB_ACCOUNT,
7505 					  NULL);
7506 	if (!x86_fpu_cache) {
7507 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7508 		goto out;
7509 	}
7510 
7511 	x86_emulator_cache = kvm_alloc_emulator_cache();
7512 	if (!x86_emulator_cache) {
7513 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
7514 		goto out_free_x86_fpu_cache;
7515 	}
7516 
7517 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7518 	if (!shared_msrs) {
7519 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7520 		goto out_free_x86_emulator_cache;
7521 	}
7522 
7523 	r = kvm_mmu_module_init();
7524 	if (r)
7525 		goto out_free_percpu;
7526 
7527 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7528 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
7529 			PT_PRESENT_MASK, 0, sme_me_mask);
7530 	kvm_timer_init();
7531 
7532 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
7533 
7534 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7535 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7536 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7537 	}
7538 
7539 	kvm_lapic_init();
7540 	if (pi_inject_timer == -1)
7541 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7542 #ifdef CONFIG_X86_64
7543 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7544 
7545 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7546 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7547 #endif
7548 
7549 	return 0;
7550 
7551 out_free_percpu:
7552 	free_percpu(shared_msrs);
7553 out_free_x86_emulator_cache:
7554 	kmem_cache_destroy(x86_emulator_cache);
7555 out_free_x86_fpu_cache:
7556 	kmem_cache_destroy(x86_fpu_cache);
7557 out:
7558 	return r;
7559 }
7560 
7561 void kvm_arch_exit(void)
7562 {
7563 #ifdef CONFIG_X86_64
7564 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7565 		clear_hv_tscchange_cb();
7566 #endif
7567 	kvm_lapic_exit();
7568 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7569 
7570 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7571 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7572 					    CPUFREQ_TRANSITION_NOTIFIER);
7573 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7574 #ifdef CONFIG_X86_64
7575 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7576 #endif
7577 	kvm_x86_ops.hardware_enable = NULL;
7578 	kvm_mmu_module_exit();
7579 	free_percpu(shared_msrs);
7580 	kmem_cache_destroy(x86_fpu_cache);
7581 }
7582 
7583 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7584 {
7585 	++vcpu->stat.halt_exits;
7586 	if (lapic_in_kernel(vcpu)) {
7587 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7588 		return 1;
7589 	} else {
7590 		vcpu->run->exit_reason = KVM_EXIT_HLT;
7591 		return 0;
7592 	}
7593 }
7594 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7595 
7596 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7597 {
7598 	int ret = kvm_skip_emulated_instruction(vcpu);
7599 	/*
7600 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7601 	 * KVM_EXIT_DEBUG here.
7602 	 */
7603 	return kvm_vcpu_halt(vcpu) && ret;
7604 }
7605 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7606 
7607 #ifdef CONFIG_X86_64
7608 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7609 			        unsigned long clock_type)
7610 {
7611 	struct kvm_clock_pairing clock_pairing;
7612 	struct timespec64 ts;
7613 	u64 cycle;
7614 	int ret;
7615 
7616 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7617 		return -KVM_EOPNOTSUPP;
7618 
7619 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7620 		return -KVM_EOPNOTSUPP;
7621 
7622 	clock_pairing.sec = ts.tv_sec;
7623 	clock_pairing.nsec = ts.tv_nsec;
7624 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7625 	clock_pairing.flags = 0;
7626 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7627 
7628 	ret = 0;
7629 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7630 			    sizeof(struct kvm_clock_pairing)))
7631 		ret = -KVM_EFAULT;
7632 
7633 	return ret;
7634 }
7635 #endif
7636 
7637 /*
7638  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7639  *
7640  * @apicid - apicid of vcpu to be kicked.
7641  */
7642 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7643 {
7644 	struct kvm_lapic_irq lapic_irq;
7645 
7646 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
7647 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7648 	lapic_irq.level = 0;
7649 	lapic_irq.dest_id = apicid;
7650 	lapic_irq.msi_redir_hint = false;
7651 
7652 	lapic_irq.delivery_mode = APIC_DM_REMRD;
7653 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7654 }
7655 
7656 bool kvm_apicv_activated(struct kvm *kvm)
7657 {
7658 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7659 }
7660 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7661 
7662 void kvm_apicv_init(struct kvm *kvm, bool enable)
7663 {
7664 	if (enable)
7665 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
7666 			  &kvm->arch.apicv_inhibit_reasons);
7667 	else
7668 		set_bit(APICV_INHIBIT_REASON_DISABLE,
7669 			&kvm->arch.apicv_inhibit_reasons);
7670 }
7671 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7672 
7673 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7674 {
7675 	struct kvm_vcpu *target = NULL;
7676 	struct kvm_apic_map *map;
7677 
7678 	rcu_read_lock();
7679 	map = rcu_dereference(kvm->arch.apic_map);
7680 
7681 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7682 		target = map->phys_map[dest_id]->vcpu;
7683 
7684 	rcu_read_unlock();
7685 
7686 	if (target && READ_ONCE(target->ready))
7687 		kvm_vcpu_yield_to(target);
7688 }
7689 
7690 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7691 {
7692 	unsigned long nr, a0, a1, a2, a3, ret;
7693 	int op_64_bit;
7694 
7695 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
7696 		return kvm_hv_hypercall(vcpu);
7697 
7698 	nr = kvm_rax_read(vcpu);
7699 	a0 = kvm_rbx_read(vcpu);
7700 	a1 = kvm_rcx_read(vcpu);
7701 	a2 = kvm_rdx_read(vcpu);
7702 	a3 = kvm_rsi_read(vcpu);
7703 
7704 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
7705 
7706 	op_64_bit = is_64_bit_mode(vcpu);
7707 	if (!op_64_bit) {
7708 		nr &= 0xFFFFFFFF;
7709 		a0 &= 0xFFFFFFFF;
7710 		a1 &= 0xFFFFFFFF;
7711 		a2 &= 0xFFFFFFFF;
7712 		a3 &= 0xFFFFFFFF;
7713 	}
7714 
7715 	if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7716 		ret = -KVM_EPERM;
7717 		goto out;
7718 	}
7719 
7720 	switch (nr) {
7721 	case KVM_HC_VAPIC_POLL_IRQ:
7722 		ret = 0;
7723 		break;
7724 	case KVM_HC_KICK_CPU:
7725 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7726 		kvm_sched_yield(vcpu->kvm, a1);
7727 		ret = 0;
7728 		break;
7729 #ifdef CONFIG_X86_64
7730 	case KVM_HC_CLOCK_PAIRING:
7731 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7732 		break;
7733 #endif
7734 	case KVM_HC_SEND_IPI:
7735 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7736 		break;
7737 	case KVM_HC_SCHED_YIELD:
7738 		kvm_sched_yield(vcpu->kvm, a0);
7739 		ret = 0;
7740 		break;
7741 	default:
7742 		ret = -KVM_ENOSYS;
7743 		break;
7744 	}
7745 out:
7746 	if (!op_64_bit)
7747 		ret = (u32)ret;
7748 	kvm_rax_write(vcpu, ret);
7749 
7750 	++vcpu->stat.hypercalls;
7751 	return kvm_skip_emulated_instruction(vcpu);
7752 }
7753 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7754 
7755 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7756 {
7757 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7758 	char instruction[3];
7759 	unsigned long rip = kvm_rip_read(vcpu);
7760 
7761 	kvm_x86_ops.patch_hypercall(vcpu, instruction);
7762 
7763 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7764 		&ctxt->exception);
7765 }
7766 
7767 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7768 {
7769 	return vcpu->run->request_interrupt_window &&
7770 		likely(!pic_in_kernel(vcpu->kvm));
7771 }
7772 
7773 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7774 {
7775 	struct kvm_run *kvm_run = vcpu->run;
7776 
7777 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7778 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7779 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7780 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7781 	kvm_run->ready_for_interrupt_injection =
7782 		pic_in_kernel(vcpu->kvm) ||
7783 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7784 }
7785 
7786 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7787 {
7788 	int max_irr, tpr;
7789 
7790 	if (!kvm_x86_ops.update_cr8_intercept)
7791 		return;
7792 
7793 	if (!lapic_in_kernel(vcpu))
7794 		return;
7795 
7796 	if (vcpu->arch.apicv_active)
7797 		return;
7798 
7799 	if (!vcpu->arch.apic->vapic_addr)
7800 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7801 	else
7802 		max_irr = -1;
7803 
7804 	if (max_irr != -1)
7805 		max_irr >>= 4;
7806 
7807 	tpr = kvm_lapic_get_cr8(vcpu);
7808 
7809 	kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7810 }
7811 
7812 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
7813 {
7814 	int r;
7815 	bool can_inject = true;
7816 
7817 	/* try to reinject previous events if any */
7818 
7819 	if (vcpu->arch.exception.injected) {
7820 		kvm_x86_ops.queue_exception(vcpu);
7821 		can_inject = false;
7822 	}
7823 	/*
7824 	 * Do not inject an NMI or interrupt if there is a pending
7825 	 * exception.  Exceptions and interrupts are recognized at
7826 	 * instruction boundaries, i.e. the start of an instruction.
7827 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7828 	 * NMIs and interrupts, i.e. traps are recognized before an
7829 	 * NMI/interrupt that's pending on the same instruction.
7830 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7831 	 * priority, but are only generated (pended) during instruction
7832 	 * execution, i.e. a pending fault-like exception means the
7833 	 * fault occurred on the *previous* instruction and must be
7834 	 * serviced prior to recognizing any new events in order to
7835 	 * fully complete the previous instruction.
7836 	 */
7837 	else if (!vcpu->arch.exception.pending) {
7838 		if (vcpu->arch.nmi_injected) {
7839 			kvm_x86_ops.set_nmi(vcpu);
7840 			can_inject = false;
7841 		} else if (vcpu->arch.interrupt.injected) {
7842 			kvm_x86_ops.set_irq(vcpu);
7843 			can_inject = false;
7844 		}
7845 	}
7846 
7847 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
7848 		     vcpu->arch.exception.pending);
7849 
7850 	/*
7851 	 * Call check_nested_events() even if we reinjected a previous event
7852 	 * in order for caller to determine if it should require immediate-exit
7853 	 * from L2 to L1 due to pending L1 events which require exit
7854 	 * from L2 to L1.
7855 	 */
7856 	if (is_guest_mode(vcpu)) {
7857 		r = kvm_x86_ops.nested_ops->check_events(vcpu);
7858 		if (r < 0)
7859 			goto busy;
7860 	}
7861 
7862 	/* try to inject new event if pending */
7863 	if (vcpu->arch.exception.pending) {
7864 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7865 					vcpu->arch.exception.has_error_code,
7866 					vcpu->arch.exception.error_code);
7867 
7868 		vcpu->arch.exception.pending = false;
7869 		vcpu->arch.exception.injected = true;
7870 
7871 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7872 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7873 					     X86_EFLAGS_RF);
7874 
7875 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7876 			kvm_deliver_exception_payload(vcpu);
7877 			if (vcpu->arch.dr7 & DR7_GD) {
7878 				vcpu->arch.dr7 &= ~DR7_GD;
7879 				kvm_update_dr7(vcpu);
7880 			}
7881 		}
7882 
7883 		kvm_x86_ops.queue_exception(vcpu);
7884 		can_inject = false;
7885 	}
7886 
7887 	/*
7888 	 * Finally, inject interrupt events.  If an event cannot be injected
7889 	 * due to architectural conditions (e.g. IF=0) a window-open exit
7890 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
7891 	 * and can architecturally be injected, but we cannot do it right now:
7892 	 * an interrupt could have arrived just now and we have to inject it
7893 	 * as a vmexit, or there could already an event in the queue, which is
7894 	 * indicated by can_inject.  In that case we request an immediate exit
7895 	 * in order to make progress and get back here for another iteration.
7896 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
7897 	 */
7898 	if (vcpu->arch.smi_pending) {
7899 		r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
7900 		if (r < 0)
7901 			goto busy;
7902 		if (r) {
7903 			vcpu->arch.smi_pending = false;
7904 			++vcpu->arch.smi_count;
7905 			enter_smm(vcpu);
7906 			can_inject = false;
7907 		} else
7908 			kvm_x86_ops.enable_smi_window(vcpu);
7909 	}
7910 
7911 	if (vcpu->arch.nmi_pending) {
7912 		r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
7913 		if (r < 0)
7914 			goto busy;
7915 		if (r) {
7916 			--vcpu->arch.nmi_pending;
7917 			vcpu->arch.nmi_injected = true;
7918 			kvm_x86_ops.set_nmi(vcpu);
7919 			can_inject = false;
7920 			WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
7921 		}
7922 		if (vcpu->arch.nmi_pending)
7923 			kvm_x86_ops.enable_nmi_window(vcpu);
7924 	}
7925 
7926 	if (kvm_cpu_has_injectable_intr(vcpu)) {
7927 		r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
7928 		if (r < 0)
7929 			goto busy;
7930 		if (r) {
7931 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
7932 			kvm_x86_ops.set_irq(vcpu);
7933 			WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
7934 		}
7935 		if (kvm_cpu_has_injectable_intr(vcpu))
7936 			kvm_x86_ops.enable_irq_window(vcpu);
7937 	}
7938 
7939 	if (is_guest_mode(vcpu) &&
7940 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
7941 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
7942 		*req_immediate_exit = true;
7943 
7944 	WARN_ON(vcpu->arch.exception.pending);
7945 	return;
7946 
7947 busy:
7948 	*req_immediate_exit = true;
7949 	return;
7950 }
7951 
7952 static void process_nmi(struct kvm_vcpu *vcpu)
7953 {
7954 	unsigned limit = 2;
7955 
7956 	/*
7957 	 * x86 is limited to one NMI running, and one NMI pending after it.
7958 	 * If an NMI is already in progress, limit further NMIs to just one.
7959 	 * Otherwise, allow two (and we'll inject the first one immediately).
7960 	 */
7961 	if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7962 		limit = 1;
7963 
7964 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7965 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7966 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7967 }
7968 
7969 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7970 {
7971 	u32 flags = 0;
7972 	flags |= seg->g       << 23;
7973 	flags |= seg->db      << 22;
7974 	flags |= seg->l       << 21;
7975 	flags |= seg->avl     << 20;
7976 	flags |= seg->present << 15;
7977 	flags |= seg->dpl     << 13;
7978 	flags |= seg->s       << 12;
7979 	flags |= seg->type    << 8;
7980 	return flags;
7981 }
7982 
7983 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7984 {
7985 	struct kvm_segment seg;
7986 	int offset;
7987 
7988 	kvm_get_segment(vcpu, &seg, n);
7989 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7990 
7991 	if (n < 3)
7992 		offset = 0x7f84 + n * 12;
7993 	else
7994 		offset = 0x7f2c + (n - 3) * 12;
7995 
7996 	put_smstate(u32, buf, offset + 8, seg.base);
7997 	put_smstate(u32, buf, offset + 4, seg.limit);
7998 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7999 }
8000 
8001 #ifdef CONFIG_X86_64
8002 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8003 {
8004 	struct kvm_segment seg;
8005 	int offset;
8006 	u16 flags;
8007 
8008 	kvm_get_segment(vcpu, &seg, n);
8009 	offset = 0x7e00 + n * 16;
8010 
8011 	flags = enter_smm_get_segment_flags(&seg) >> 8;
8012 	put_smstate(u16, buf, offset, seg.selector);
8013 	put_smstate(u16, buf, offset + 2, flags);
8014 	put_smstate(u32, buf, offset + 4, seg.limit);
8015 	put_smstate(u64, buf, offset + 8, seg.base);
8016 }
8017 #endif
8018 
8019 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8020 {
8021 	struct desc_ptr dt;
8022 	struct kvm_segment seg;
8023 	unsigned long val;
8024 	int i;
8025 
8026 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8027 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8028 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8029 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8030 
8031 	for (i = 0; i < 8; i++)
8032 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8033 
8034 	kvm_get_dr(vcpu, 6, &val);
8035 	put_smstate(u32, buf, 0x7fcc, (u32)val);
8036 	kvm_get_dr(vcpu, 7, &val);
8037 	put_smstate(u32, buf, 0x7fc8, (u32)val);
8038 
8039 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8040 	put_smstate(u32, buf, 0x7fc4, seg.selector);
8041 	put_smstate(u32, buf, 0x7f64, seg.base);
8042 	put_smstate(u32, buf, 0x7f60, seg.limit);
8043 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8044 
8045 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8046 	put_smstate(u32, buf, 0x7fc0, seg.selector);
8047 	put_smstate(u32, buf, 0x7f80, seg.base);
8048 	put_smstate(u32, buf, 0x7f7c, seg.limit);
8049 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8050 
8051 	kvm_x86_ops.get_gdt(vcpu, &dt);
8052 	put_smstate(u32, buf, 0x7f74, dt.address);
8053 	put_smstate(u32, buf, 0x7f70, dt.size);
8054 
8055 	kvm_x86_ops.get_idt(vcpu, &dt);
8056 	put_smstate(u32, buf, 0x7f58, dt.address);
8057 	put_smstate(u32, buf, 0x7f54, dt.size);
8058 
8059 	for (i = 0; i < 6; i++)
8060 		enter_smm_save_seg_32(vcpu, buf, i);
8061 
8062 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8063 
8064 	/* revision id */
8065 	put_smstate(u32, buf, 0x7efc, 0x00020000);
8066 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8067 }
8068 
8069 #ifdef CONFIG_X86_64
8070 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8071 {
8072 	struct desc_ptr dt;
8073 	struct kvm_segment seg;
8074 	unsigned long val;
8075 	int i;
8076 
8077 	for (i = 0; i < 16; i++)
8078 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8079 
8080 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8081 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8082 
8083 	kvm_get_dr(vcpu, 6, &val);
8084 	put_smstate(u64, buf, 0x7f68, val);
8085 	kvm_get_dr(vcpu, 7, &val);
8086 	put_smstate(u64, buf, 0x7f60, val);
8087 
8088 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8089 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8090 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8091 
8092 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8093 
8094 	/* revision id */
8095 	put_smstate(u32, buf, 0x7efc, 0x00020064);
8096 
8097 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8098 
8099 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8100 	put_smstate(u16, buf, 0x7e90, seg.selector);
8101 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8102 	put_smstate(u32, buf, 0x7e94, seg.limit);
8103 	put_smstate(u64, buf, 0x7e98, seg.base);
8104 
8105 	kvm_x86_ops.get_idt(vcpu, &dt);
8106 	put_smstate(u32, buf, 0x7e84, dt.size);
8107 	put_smstate(u64, buf, 0x7e88, dt.address);
8108 
8109 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8110 	put_smstate(u16, buf, 0x7e70, seg.selector);
8111 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8112 	put_smstate(u32, buf, 0x7e74, seg.limit);
8113 	put_smstate(u64, buf, 0x7e78, seg.base);
8114 
8115 	kvm_x86_ops.get_gdt(vcpu, &dt);
8116 	put_smstate(u32, buf, 0x7e64, dt.size);
8117 	put_smstate(u64, buf, 0x7e68, dt.address);
8118 
8119 	for (i = 0; i < 6; i++)
8120 		enter_smm_save_seg_64(vcpu, buf, i);
8121 }
8122 #endif
8123 
8124 static void enter_smm(struct kvm_vcpu *vcpu)
8125 {
8126 	struct kvm_segment cs, ds;
8127 	struct desc_ptr dt;
8128 	char buf[512];
8129 	u32 cr0;
8130 
8131 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8132 	memset(buf, 0, 512);
8133 #ifdef CONFIG_X86_64
8134 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8135 		enter_smm_save_state_64(vcpu, buf);
8136 	else
8137 #endif
8138 		enter_smm_save_state_32(vcpu, buf);
8139 
8140 	/*
8141 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8142 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
8143 	 * the SMM state-save area.
8144 	 */
8145 	kvm_x86_ops.pre_enter_smm(vcpu, buf);
8146 
8147 	vcpu->arch.hflags |= HF_SMM_MASK;
8148 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8149 
8150 	if (kvm_x86_ops.get_nmi_mask(vcpu))
8151 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8152 	else
8153 		kvm_x86_ops.set_nmi_mask(vcpu, true);
8154 
8155 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8156 	kvm_rip_write(vcpu, 0x8000);
8157 
8158 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8159 	kvm_x86_ops.set_cr0(vcpu, cr0);
8160 	vcpu->arch.cr0 = cr0;
8161 
8162 	kvm_x86_ops.set_cr4(vcpu, 0);
8163 
8164 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
8165 	dt.address = dt.size = 0;
8166 	kvm_x86_ops.set_idt(vcpu, &dt);
8167 
8168 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8169 
8170 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8171 	cs.base = vcpu->arch.smbase;
8172 
8173 	ds.selector = 0;
8174 	ds.base = 0;
8175 
8176 	cs.limit    = ds.limit = 0xffffffff;
8177 	cs.type     = ds.type = 0x3;
8178 	cs.dpl      = ds.dpl = 0;
8179 	cs.db       = ds.db = 0;
8180 	cs.s        = ds.s = 1;
8181 	cs.l        = ds.l = 0;
8182 	cs.g        = ds.g = 1;
8183 	cs.avl      = ds.avl = 0;
8184 	cs.present  = ds.present = 1;
8185 	cs.unusable = ds.unusable = 0;
8186 	cs.padding  = ds.padding = 0;
8187 
8188 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8189 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8190 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8191 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8192 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8193 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8194 
8195 #ifdef CONFIG_X86_64
8196 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8197 		kvm_x86_ops.set_efer(vcpu, 0);
8198 #endif
8199 
8200 	kvm_update_cpuid_runtime(vcpu);
8201 	kvm_mmu_reset_context(vcpu);
8202 }
8203 
8204 static void process_smi(struct kvm_vcpu *vcpu)
8205 {
8206 	vcpu->arch.smi_pending = true;
8207 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8208 }
8209 
8210 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8211 				       unsigned long *vcpu_bitmap)
8212 {
8213 	cpumask_var_t cpus;
8214 
8215 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8216 
8217 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8218 				    NULL, vcpu_bitmap, cpus);
8219 
8220 	free_cpumask_var(cpus);
8221 }
8222 
8223 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8224 {
8225 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8226 }
8227 
8228 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8229 {
8230 	if (!lapic_in_kernel(vcpu))
8231 		return;
8232 
8233 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8234 	kvm_apic_update_apicv(vcpu);
8235 	kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8236 }
8237 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8238 
8239 /*
8240  * NOTE: Do not hold any lock prior to calling this.
8241  *
8242  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8243  * locked, because it calls __x86_set_memory_region() which does
8244  * synchronize_srcu(&kvm->srcu).
8245  */
8246 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8247 {
8248 	struct kvm_vcpu *except;
8249 	unsigned long old, new, expected;
8250 
8251 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8252 	    !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8253 		return;
8254 
8255 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8256 	do {
8257 		expected = new = old;
8258 		if (activate)
8259 			__clear_bit(bit, &new);
8260 		else
8261 			__set_bit(bit, &new);
8262 		if (new == old)
8263 			break;
8264 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8265 	} while (old != expected);
8266 
8267 	if (!!old == !!new)
8268 		return;
8269 
8270 	trace_kvm_apicv_update_request(activate, bit);
8271 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8272 		kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8273 
8274 	/*
8275 	 * Sending request to update APICV for all other vcpus,
8276 	 * while update the calling vcpu immediately instead of
8277 	 * waiting for another #VMEXIT to handle the request.
8278 	 */
8279 	except = kvm_get_running_vcpu();
8280 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8281 					 except);
8282 	if (except)
8283 		kvm_vcpu_update_apicv(except);
8284 }
8285 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8286 
8287 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8288 {
8289 	if (!kvm_apic_present(vcpu))
8290 		return;
8291 
8292 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8293 
8294 	if (irqchip_split(vcpu->kvm))
8295 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8296 	else {
8297 		if (vcpu->arch.apicv_active)
8298 			kvm_x86_ops.sync_pir_to_irr(vcpu);
8299 		if (ioapic_in_kernel(vcpu->kvm))
8300 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8301 	}
8302 
8303 	if (is_guest_mode(vcpu))
8304 		vcpu->arch.load_eoi_exitmap_pending = true;
8305 	else
8306 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8307 }
8308 
8309 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8310 {
8311 	u64 eoi_exit_bitmap[4];
8312 
8313 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8314 		return;
8315 
8316 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8317 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
8318 	kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8319 }
8320 
8321 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8322 					    unsigned long start, unsigned long end)
8323 {
8324 	unsigned long apic_address;
8325 
8326 	/*
8327 	 * The physical address of apic access page is stored in the VMCS.
8328 	 * Update it when it becomes invalid.
8329 	 */
8330 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8331 	if (start <= apic_address && apic_address < end)
8332 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8333 }
8334 
8335 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8336 {
8337 	if (!lapic_in_kernel(vcpu))
8338 		return;
8339 
8340 	if (!kvm_x86_ops.set_apic_access_page_addr)
8341 		return;
8342 
8343 	kvm_x86_ops.set_apic_access_page_addr(vcpu);
8344 }
8345 
8346 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8347 {
8348 	smp_send_reschedule(vcpu->cpu);
8349 }
8350 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8351 
8352 /*
8353  * Returns 1 to let vcpu_run() continue the guest execution loop without
8354  * exiting to the userspace.  Otherwise, the value will be returned to the
8355  * userspace.
8356  */
8357 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8358 {
8359 	int r;
8360 	bool req_int_win =
8361 		dm_request_for_irq_injection(vcpu) &&
8362 		kvm_cpu_accept_dm_intr(vcpu);
8363 	fastpath_t exit_fastpath;
8364 
8365 	bool req_immediate_exit = false;
8366 
8367 	if (kvm_request_pending(vcpu)) {
8368 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8369 			if (unlikely(!kvm_x86_ops.nested_ops->get_vmcs12_pages(vcpu))) {
8370 				r = 0;
8371 				goto out;
8372 			}
8373 		}
8374 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8375 			kvm_mmu_unload(vcpu);
8376 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8377 			__kvm_migrate_timers(vcpu);
8378 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8379 			kvm_gen_update_masterclock(vcpu->kvm);
8380 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8381 			kvm_gen_kvmclock_update(vcpu);
8382 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8383 			r = kvm_guest_time_update(vcpu);
8384 			if (unlikely(r))
8385 				goto out;
8386 		}
8387 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8388 			kvm_mmu_sync_roots(vcpu);
8389 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8390 			kvm_mmu_load_pgd(vcpu);
8391 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8392 			kvm_vcpu_flush_tlb_all(vcpu);
8393 
8394 			/* Flushing all ASIDs flushes the current ASID... */
8395 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8396 		}
8397 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8398 			kvm_vcpu_flush_tlb_current(vcpu);
8399 		if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8400 			kvm_vcpu_flush_tlb_guest(vcpu);
8401 
8402 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8403 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8404 			r = 0;
8405 			goto out;
8406 		}
8407 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8408 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8409 			vcpu->mmio_needed = 0;
8410 			r = 0;
8411 			goto out;
8412 		}
8413 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8414 			/* Page is swapped out. Do synthetic halt */
8415 			vcpu->arch.apf.halted = true;
8416 			r = 1;
8417 			goto out;
8418 		}
8419 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8420 			record_steal_time(vcpu);
8421 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
8422 			process_smi(vcpu);
8423 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
8424 			process_nmi(vcpu);
8425 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
8426 			kvm_pmu_handle_event(vcpu);
8427 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
8428 			kvm_pmu_deliver_pmi(vcpu);
8429 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8430 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8431 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
8432 				     vcpu->arch.ioapic_handled_vectors)) {
8433 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8434 				vcpu->run->eoi.vector =
8435 						vcpu->arch.pending_ioapic_eoi;
8436 				r = 0;
8437 				goto out;
8438 			}
8439 		}
8440 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8441 			vcpu_scan_ioapic(vcpu);
8442 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8443 			vcpu_load_eoi_exitmap(vcpu);
8444 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8445 			kvm_vcpu_reload_apic_access_page(vcpu);
8446 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8447 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8448 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8449 			r = 0;
8450 			goto out;
8451 		}
8452 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8453 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8454 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8455 			r = 0;
8456 			goto out;
8457 		}
8458 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8459 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8460 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8461 			r = 0;
8462 			goto out;
8463 		}
8464 
8465 		/*
8466 		 * KVM_REQ_HV_STIMER has to be processed after
8467 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8468 		 * depend on the guest clock being up-to-date
8469 		 */
8470 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8471 			kvm_hv_process_stimers(vcpu);
8472 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8473 			kvm_vcpu_update_apicv(vcpu);
8474 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8475 			kvm_check_async_pf_completion(vcpu);
8476 	}
8477 
8478 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8479 		++vcpu->stat.req_event;
8480 		kvm_apic_accept_events(vcpu);
8481 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8482 			r = 1;
8483 			goto out;
8484 		}
8485 
8486 		inject_pending_event(vcpu, &req_immediate_exit);
8487 		if (req_int_win)
8488 			kvm_x86_ops.enable_irq_window(vcpu);
8489 
8490 		if (kvm_lapic_enabled(vcpu)) {
8491 			update_cr8_intercept(vcpu);
8492 			kvm_lapic_sync_to_vapic(vcpu);
8493 		}
8494 	}
8495 
8496 	r = kvm_mmu_reload(vcpu);
8497 	if (unlikely(r)) {
8498 		goto cancel_injection;
8499 	}
8500 
8501 	preempt_disable();
8502 
8503 	kvm_x86_ops.prepare_guest_switch(vcpu);
8504 
8505 	/*
8506 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8507 	 * IPI are then delayed after guest entry, which ensures that they
8508 	 * result in virtual interrupt delivery.
8509 	 */
8510 	local_irq_disable();
8511 	vcpu->mode = IN_GUEST_MODE;
8512 
8513 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8514 
8515 	/*
8516 	 * 1) We should set ->mode before checking ->requests.  Please see
8517 	 * the comment in kvm_vcpu_exiting_guest_mode().
8518 	 *
8519 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
8520 	 * pairs with the memory barrier implicit in pi_test_and_set_on
8521 	 * (see vmx_deliver_posted_interrupt).
8522 	 *
8523 	 * 3) This also orders the write to mode from any reads to the page
8524 	 * tables done while the VCPU is running.  Please see the comment
8525 	 * in kvm_flush_remote_tlbs.
8526 	 */
8527 	smp_mb__after_srcu_read_unlock();
8528 
8529 	/*
8530 	 * This handles the case where a posted interrupt was
8531 	 * notified with kvm_vcpu_kick.
8532 	 */
8533 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8534 		kvm_x86_ops.sync_pir_to_irr(vcpu);
8535 
8536 	if (kvm_vcpu_exit_request(vcpu)) {
8537 		vcpu->mode = OUTSIDE_GUEST_MODE;
8538 		smp_wmb();
8539 		local_irq_enable();
8540 		preempt_enable();
8541 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8542 		r = 1;
8543 		goto cancel_injection;
8544 	}
8545 
8546 	if (req_immediate_exit) {
8547 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8548 		kvm_x86_ops.request_immediate_exit(vcpu);
8549 	}
8550 
8551 	trace_kvm_entry(vcpu->vcpu_id);
8552 
8553 	fpregs_assert_state_consistent();
8554 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8555 		switch_fpu_return();
8556 
8557 	if (unlikely(vcpu->arch.switch_db_regs)) {
8558 		set_debugreg(0, 7);
8559 		set_debugreg(vcpu->arch.eff_db[0], 0);
8560 		set_debugreg(vcpu->arch.eff_db[1], 1);
8561 		set_debugreg(vcpu->arch.eff_db[2], 2);
8562 		set_debugreg(vcpu->arch.eff_db[3], 3);
8563 		set_debugreg(vcpu->arch.dr6, 6);
8564 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8565 	}
8566 
8567 	exit_fastpath = kvm_x86_ops.run(vcpu);
8568 
8569 	/*
8570 	 * Do this here before restoring debug registers on the host.  And
8571 	 * since we do this before handling the vmexit, a DR access vmexit
8572 	 * can (a) read the correct value of the debug registers, (b) set
8573 	 * KVM_DEBUGREG_WONT_EXIT again.
8574 	 */
8575 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8576 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8577 		kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8578 		kvm_update_dr0123(vcpu);
8579 		kvm_update_dr7(vcpu);
8580 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8581 	}
8582 
8583 	/*
8584 	 * If the guest has used debug registers, at least dr7
8585 	 * will be disabled while returning to the host.
8586 	 * If we don't have active breakpoints in the host, we don't
8587 	 * care about the messed up debug address registers. But if
8588 	 * we have some of them active, restore the old state.
8589 	 */
8590 	if (hw_breakpoint_active())
8591 		hw_breakpoint_restore();
8592 
8593 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
8594 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8595 
8596 	vcpu->mode = OUTSIDE_GUEST_MODE;
8597 	smp_wmb();
8598 
8599 	kvm_x86_ops.handle_exit_irqoff(vcpu);
8600 
8601 	/*
8602 	 * Consume any pending interrupts, including the possible source of
8603 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8604 	 * An instruction is required after local_irq_enable() to fully unblock
8605 	 * interrupts on processors that implement an interrupt shadow, the
8606 	 * stat.exits increment will do nicely.
8607 	 */
8608 	kvm_before_interrupt(vcpu);
8609 	local_irq_enable();
8610 	++vcpu->stat.exits;
8611 	local_irq_disable();
8612 	kvm_after_interrupt(vcpu);
8613 
8614 	if (lapic_in_kernel(vcpu)) {
8615 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8616 		if (delta != S64_MIN) {
8617 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8618 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8619 		}
8620 	}
8621 
8622 	local_irq_enable();
8623 	preempt_enable();
8624 
8625 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8626 
8627 	/*
8628 	 * Profile KVM exit RIPs:
8629 	 */
8630 	if (unlikely(prof_on == KVM_PROFILING)) {
8631 		unsigned long rip = kvm_rip_read(vcpu);
8632 		profile_hit(KVM_PROFILING, (void *)rip);
8633 	}
8634 
8635 	if (unlikely(vcpu->arch.tsc_always_catchup))
8636 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8637 
8638 	if (vcpu->arch.apic_attention)
8639 		kvm_lapic_sync_from_vapic(vcpu);
8640 
8641 	r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8642 	return r;
8643 
8644 cancel_injection:
8645 	if (req_immediate_exit)
8646 		kvm_make_request(KVM_REQ_EVENT, vcpu);
8647 	kvm_x86_ops.cancel_injection(vcpu);
8648 	if (unlikely(vcpu->arch.apic_attention))
8649 		kvm_lapic_sync_from_vapic(vcpu);
8650 out:
8651 	return r;
8652 }
8653 
8654 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8655 {
8656 	if (!kvm_arch_vcpu_runnable(vcpu) &&
8657 	    (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8658 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8659 		kvm_vcpu_block(vcpu);
8660 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8661 
8662 		if (kvm_x86_ops.post_block)
8663 			kvm_x86_ops.post_block(vcpu);
8664 
8665 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8666 			return 1;
8667 	}
8668 
8669 	kvm_apic_accept_events(vcpu);
8670 	switch(vcpu->arch.mp_state) {
8671 	case KVM_MP_STATE_HALTED:
8672 		vcpu->arch.pv.pv_unhalted = false;
8673 		vcpu->arch.mp_state =
8674 			KVM_MP_STATE_RUNNABLE;
8675 		/* fall through */
8676 	case KVM_MP_STATE_RUNNABLE:
8677 		vcpu->arch.apf.halted = false;
8678 		break;
8679 	case KVM_MP_STATE_INIT_RECEIVED:
8680 		break;
8681 	default:
8682 		return -EINTR;
8683 	}
8684 	return 1;
8685 }
8686 
8687 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8688 {
8689 	if (is_guest_mode(vcpu))
8690 		kvm_x86_ops.nested_ops->check_events(vcpu);
8691 
8692 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8693 		!vcpu->arch.apf.halted);
8694 }
8695 
8696 static int vcpu_run(struct kvm_vcpu *vcpu)
8697 {
8698 	int r;
8699 	struct kvm *kvm = vcpu->kvm;
8700 
8701 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8702 	vcpu->arch.l1tf_flush_l1d = true;
8703 
8704 	for (;;) {
8705 		if (kvm_vcpu_running(vcpu)) {
8706 			r = vcpu_enter_guest(vcpu);
8707 		} else {
8708 			r = vcpu_block(kvm, vcpu);
8709 		}
8710 
8711 		if (r <= 0)
8712 			break;
8713 
8714 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8715 		if (kvm_cpu_has_pending_timer(vcpu))
8716 			kvm_inject_pending_timer_irqs(vcpu);
8717 
8718 		if (dm_request_for_irq_injection(vcpu) &&
8719 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8720 			r = 0;
8721 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8722 			++vcpu->stat.request_irq_exits;
8723 			break;
8724 		}
8725 
8726 		if (__xfer_to_guest_mode_work_pending()) {
8727 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8728 			r = xfer_to_guest_mode_handle_work(vcpu);
8729 			if (r)
8730 				return r;
8731 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8732 		}
8733 	}
8734 
8735 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8736 
8737 	return r;
8738 }
8739 
8740 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8741 {
8742 	int r;
8743 
8744 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8745 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8746 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8747 	return r;
8748 }
8749 
8750 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8751 {
8752 	BUG_ON(!vcpu->arch.pio.count);
8753 
8754 	return complete_emulated_io(vcpu);
8755 }
8756 
8757 /*
8758  * Implements the following, as a state machine:
8759  *
8760  * read:
8761  *   for each fragment
8762  *     for each mmio piece in the fragment
8763  *       write gpa, len
8764  *       exit
8765  *       copy data
8766  *   execute insn
8767  *
8768  * write:
8769  *   for each fragment
8770  *     for each mmio piece in the fragment
8771  *       write gpa, len
8772  *       copy data
8773  *       exit
8774  */
8775 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8776 {
8777 	struct kvm_run *run = vcpu->run;
8778 	struct kvm_mmio_fragment *frag;
8779 	unsigned len;
8780 
8781 	BUG_ON(!vcpu->mmio_needed);
8782 
8783 	/* Complete previous fragment */
8784 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8785 	len = min(8u, frag->len);
8786 	if (!vcpu->mmio_is_write)
8787 		memcpy(frag->data, run->mmio.data, len);
8788 
8789 	if (frag->len <= 8) {
8790 		/* Switch to the next fragment. */
8791 		frag++;
8792 		vcpu->mmio_cur_fragment++;
8793 	} else {
8794 		/* Go forward to the next mmio piece. */
8795 		frag->data += len;
8796 		frag->gpa += len;
8797 		frag->len -= len;
8798 	}
8799 
8800 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8801 		vcpu->mmio_needed = 0;
8802 
8803 		/* FIXME: return into emulator if single-stepping.  */
8804 		if (vcpu->mmio_is_write)
8805 			return 1;
8806 		vcpu->mmio_read_completed = 1;
8807 		return complete_emulated_io(vcpu);
8808 	}
8809 
8810 	run->exit_reason = KVM_EXIT_MMIO;
8811 	run->mmio.phys_addr = frag->gpa;
8812 	if (vcpu->mmio_is_write)
8813 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8814 	run->mmio.len = min(8u, frag->len);
8815 	run->mmio.is_write = vcpu->mmio_is_write;
8816 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8817 	return 0;
8818 }
8819 
8820 static void kvm_save_current_fpu(struct fpu *fpu)
8821 {
8822 	/*
8823 	 * If the target FPU state is not resident in the CPU registers, just
8824 	 * memcpy() from current, else save CPU state directly to the target.
8825 	 */
8826 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
8827 		memcpy(&fpu->state, &current->thread.fpu.state,
8828 		       fpu_kernel_xstate_size);
8829 	else
8830 		copy_fpregs_to_fpstate(fpu);
8831 }
8832 
8833 /* Swap (qemu) user FPU context for the guest FPU context. */
8834 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8835 {
8836 	fpregs_lock();
8837 
8838 	kvm_save_current_fpu(vcpu->arch.user_fpu);
8839 
8840 	/* PKRU is separately restored in kvm_x86_ops.run.  */
8841 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8842 				~XFEATURE_MASK_PKRU);
8843 
8844 	fpregs_mark_activate();
8845 	fpregs_unlock();
8846 
8847 	trace_kvm_fpu(1);
8848 }
8849 
8850 /* When vcpu_run ends, restore user space FPU context. */
8851 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8852 {
8853 	fpregs_lock();
8854 
8855 	kvm_save_current_fpu(vcpu->arch.guest_fpu);
8856 
8857 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8858 
8859 	fpregs_mark_activate();
8860 	fpregs_unlock();
8861 
8862 	++vcpu->stat.fpu_reload;
8863 	trace_kvm_fpu(0);
8864 }
8865 
8866 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
8867 {
8868 	struct kvm_run *kvm_run = vcpu->run;
8869 	int r;
8870 
8871 	vcpu_load(vcpu);
8872 	kvm_sigset_activate(vcpu);
8873 	kvm_load_guest_fpu(vcpu);
8874 
8875 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8876 		if (kvm_run->immediate_exit) {
8877 			r = -EINTR;
8878 			goto out;
8879 		}
8880 		kvm_vcpu_block(vcpu);
8881 		kvm_apic_accept_events(vcpu);
8882 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8883 		r = -EAGAIN;
8884 		if (signal_pending(current)) {
8885 			r = -EINTR;
8886 			kvm_run->exit_reason = KVM_EXIT_INTR;
8887 			++vcpu->stat.signal_exits;
8888 		}
8889 		goto out;
8890 	}
8891 
8892 	if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8893 		r = -EINVAL;
8894 		goto out;
8895 	}
8896 
8897 	if (kvm_run->kvm_dirty_regs) {
8898 		r = sync_regs(vcpu);
8899 		if (r != 0)
8900 			goto out;
8901 	}
8902 
8903 	/* re-sync apic's tpr */
8904 	if (!lapic_in_kernel(vcpu)) {
8905 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8906 			r = -EINVAL;
8907 			goto out;
8908 		}
8909 	}
8910 
8911 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8912 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8913 		vcpu->arch.complete_userspace_io = NULL;
8914 		r = cui(vcpu);
8915 		if (r <= 0)
8916 			goto out;
8917 	} else
8918 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8919 
8920 	if (kvm_run->immediate_exit)
8921 		r = -EINTR;
8922 	else
8923 		r = vcpu_run(vcpu);
8924 
8925 out:
8926 	kvm_put_guest_fpu(vcpu);
8927 	if (kvm_run->kvm_valid_regs)
8928 		store_regs(vcpu);
8929 	post_kvm_run_save(vcpu);
8930 	kvm_sigset_deactivate(vcpu);
8931 
8932 	vcpu_put(vcpu);
8933 	return r;
8934 }
8935 
8936 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8937 {
8938 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8939 		/*
8940 		 * We are here if userspace calls get_regs() in the middle of
8941 		 * instruction emulation. Registers state needs to be copied
8942 		 * back from emulation context to vcpu. Userspace shouldn't do
8943 		 * that usually, but some bad designed PV devices (vmware
8944 		 * backdoor interface) need this to work
8945 		 */
8946 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8947 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8948 	}
8949 	regs->rax = kvm_rax_read(vcpu);
8950 	regs->rbx = kvm_rbx_read(vcpu);
8951 	regs->rcx = kvm_rcx_read(vcpu);
8952 	regs->rdx = kvm_rdx_read(vcpu);
8953 	regs->rsi = kvm_rsi_read(vcpu);
8954 	regs->rdi = kvm_rdi_read(vcpu);
8955 	regs->rsp = kvm_rsp_read(vcpu);
8956 	regs->rbp = kvm_rbp_read(vcpu);
8957 #ifdef CONFIG_X86_64
8958 	regs->r8 = kvm_r8_read(vcpu);
8959 	regs->r9 = kvm_r9_read(vcpu);
8960 	regs->r10 = kvm_r10_read(vcpu);
8961 	regs->r11 = kvm_r11_read(vcpu);
8962 	regs->r12 = kvm_r12_read(vcpu);
8963 	regs->r13 = kvm_r13_read(vcpu);
8964 	regs->r14 = kvm_r14_read(vcpu);
8965 	regs->r15 = kvm_r15_read(vcpu);
8966 #endif
8967 
8968 	regs->rip = kvm_rip_read(vcpu);
8969 	regs->rflags = kvm_get_rflags(vcpu);
8970 }
8971 
8972 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8973 {
8974 	vcpu_load(vcpu);
8975 	__get_regs(vcpu, regs);
8976 	vcpu_put(vcpu);
8977 	return 0;
8978 }
8979 
8980 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8981 {
8982 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8983 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8984 
8985 	kvm_rax_write(vcpu, regs->rax);
8986 	kvm_rbx_write(vcpu, regs->rbx);
8987 	kvm_rcx_write(vcpu, regs->rcx);
8988 	kvm_rdx_write(vcpu, regs->rdx);
8989 	kvm_rsi_write(vcpu, regs->rsi);
8990 	kvm_rdi_write(vcpu, regs->rdi);
8991 	kvm_rsp_write(vcpu, regs->rsp);
8992 	kvm_rbp_write(vcpu, regs->rbp);
8993 #ifdef CONFIG_X86_64
8994 	kvm_r8_write(vcpu, regs->r8);
8995 	kvm_r9_write(vcpu, regs->r9);
8996 	kvm_r10_write(vcpu, regs->r10);
8997 	kvm_r11_write(vcpu, regs->r11);
8998 	kvm_r12_write(vcpu, regs->r12);
8999 	kvm_r13_write(vcpu, regs->r13);
9000 	kvm_r14_write(vcpu, regs->r14);
9001 	kvm_r15_write(vcpu, regs->r15);
9002 #endif
9003 
9004 	kvm_rip_write(vcpu, regs->rip);
9005 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9006 
9007 	vcpu->arch.exception.pending = false;
9008 
9009 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9010 }
9011 
9012 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9013 {
9014 	vcpu_load(vcpu);
9015 	__set_regs(vcpu, regs);
9016 	vcpu_put(vcpu);
9017 	return 0;
9018 }
9019 
9020 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9021 {
9022 	struct kvm_segment cs;
9023 
9024 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9025 	*db = cs.db;
9026 	*l = cs.l;
9027 }
9028 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9029 
9030 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9031 {
9032 	struct desc_ptr dt;
9033 
9034 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9035 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9036 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9037 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9038 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9039 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9040 
9041 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9042 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9043 
9044 	kvm_x86_ops.get_idt(vcpu, &dt);
9045 	sregs->idt.limit = dt.size;
9046 	sregs->idt.base = dt.address;
9047 	kvm_x86_ops.get_gdt(vcpu, &dt);
9048 	sregs->gdt.limit = dt.size;
9049 	sregs->gdt.base = dt.address;
9050 
9051 	sregs->cr0 = kvm_read_cr0(vcpu);
9052 	sregs->cr2 = vcpu->arch.cr2;
9053 	sregs->cr3 = kvm_read_cr3(vcpu);
9054 	sregs->cr4 = kvm_read_cr4(vcpu);
9055 	sregs->cr8 = kvm_get_cr8(vcpu);
9056 	sregs->efer = vcpu->arch.efer;
9057 	sregs->apic_base = kvm_get_apic_base(vcpu);
9058 
9059 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9060 
9061 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9062 		set_bit(vcpu->arch.interrupt.nr,
9063 			(unsigned long *)sregs->interrupt_bitmap);
9064 }
9065 
9066 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9067 				  struct kvm_sregs *sregs)
9068 {
9069 	vcpu_load(vcpu);
9070 	__get_sregs(vcpu, sregs);
9071 	vcpu_put(vcpu);
9072 	return 0;
9073 }
9074 
9075 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9076 				    struct kvm_mp_state *mp_state)
9077 {
9078 	vcpu_load(vcpu);
9079 	if (kvm_mpx_supported())
9080 		kvm_load_guest_fpu(vcpu);
9081 
9082 	kvm_apic_accept_events(vcpu);
9083 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
9084 					vcpu->arch.pv.pv_unhalted)
9085 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9086 	else
9087 		mp_state->mp_state = vcpu->arch.mp_state;
9088 
9089 	if (kvm_mpx_supported())
9090 		kvm_put_guest_fpu(vcpu);
9091 	vcpu_put(vcpu);
9092 	return 0;
9093 }
9094 
9095 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9096 				    struct kvm_mp_state *mp_state)
9097 {
9098 	int ret = -EINVAL;
9099 
9100 	vcpu_load(vcpu);
9101 
9102 	if (!lapic_in_kernel(vcpu) &&
9103 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9104 		goto out;
9105 
9106 	/*
9107 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9108 	 * INIT state; latched init should be reported using
9109 	 * KVM_SET_VCPU_EVENTS, so reject it here.
9110 	 */
9111 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9112 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9113 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9114 		goto out;
9115 
9116 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9117 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9118 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9119 	} else
9120 		vcpu->arch.mp_state = mp_state->mp_state;
9121 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9122 
9123 	ret = 0;
9124 out:
9125 	vcpu_put(vcpu);
9126 	return ret;
9127 }
9128 
9129 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9130 		    int reason, bool has_error_code, u32 error_code)
9131 {
9132 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9133 	int ret;
9134 
9135 	init_emulate_ctxt(vcpu);
9136 
9137 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9138 				   has_error_code, error_code);
9139 	if (ret) {
9140 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9141 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9142 		vcpu->run->internal.ndata = 0;
9143 		return 0;
9144 	}
9145 
9146 	kvm_rip_write(vcpu, ctxt->eip);
9147 	kvm_set_rflags(vcpu, ctxt->eflags);
9148 	return 1;
9149 }
9150 EXPORT_SYMBOL_GPL(kvm_task_switch);
9151 
9152 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9153 {
9154 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9155 		/*
9156 		 * When EFER.LME and CR0.PG are set, the processor is in
9157 		 * 64-bit mode (though maybe in a 32-bit code segment).
9158 		 * CR4.PAE and EFER.LMA must be set.
9159 		 */
9160 		if (!(sregs->cr4 & X86_CR4_PAE)
9161 		    || !(sregs->efer & EFER_LMA))
9162 			return -EINVAL;
9163 	} else {
9164 		/*
9165 		 * Not in 64-bit mode: EFER.LMA is clear and the code
9166 		 * segment cannot be 64-bit.
9167 		 */
9168 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9169 			return -EINVAL;
9170 	}
9171 
9172 	return kvm_valid_cr4(vcpu, sregs->cr4);
9173 }
9174 
9175 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9176 {
9177 	struct msr_data apic_base_msr;
9178 	int mmu_reset_needed = 0;
9179 	int cpuid_update_needed = 0;
9180 	int pending_vec, max_bits, idx;
9181 	struct desc_ptr dt;
9182 	int ret = -EINVAL;
9183 
9184 	if (kvm_valid_sregs(vcpu, sregs))
9185 		goto out;
9186 
9187 	apic_base_msr.data = sregs->apic_base;
9188 	apic_base_msr.host_initiated = true;
9189 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
9190 		goto out;
9191 
9192 	dt.size = sregs->idt.limit;
9193 	dt.address = sregs->idt.base;
9194 	kvm_x86_ops.set_idt(vcpu, &dt);
9195 	dt.size = sregs->gdt.limit;
9196 	dt.address = sregs->gdt.base;
9197 	kvm_x86_ops.set_gdt(vcpu, &dt);
9198 
9199 	vcpu->arch.cr2 = sregs->cr2;
9200 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9201 	vcpu->arch.cr3 = sregs->cr3;
9202 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9203 
9204 	kvm_set_cr8(vcpu, sregs->cr8);
9205 
9206 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9207 	kvm_x86_ops.set_efer(vcpu, sregs->efer);
9208 
9209 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9210 	kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9211 	vcpu->arch.cr0 = sregs->cr0;
9212 
9213 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9214 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9215 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
9216 	kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9217 	if (cpuid_update_needed)
9218 		kvm_update_cpuid_runtime(vcpu);
9219 
9220 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9221 	if (is_pae_paging(vcpu)) {
9222 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9223 		mmu_reset_needed = 1;
9224 	}
9225 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9226 
9227 	if (mmu_reset_needed)
9228 		kvm_mmu_reset_context(vcpu);
9229 
9230 	max_bits = KVM_NR_INTERRUPTS;
9231 	pending_vec = find_first_bit(
9232 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
9233 	if (pending_vec < max_bits) {
9234 		kvm_queue_interrupt(vcpu, pending_vec, false);
9235 		pr_debug("Set back pending irq %d\n", pending_vec);
9236 	}
9237 
9238 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9239 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9240 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9241 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9242 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9243 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9244 
9245 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9246 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9247 
9248 	update_cr8_intercept(vcpu);
9249 
9250 	/* Older userspace won't unhalt the vcpu on reset. */
9251 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9252 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9253 	    !is_protmode(vcpu))
9254 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9255 
9256 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9257 
9258 	ret = 0;
9259 out:
9260 	return ret;
9261 }
9262 
9263 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9264 				  struct kvm_sregs *sregs)
9265 {
9266 	int ret;
9267 
9268 	vcpu_load(vcpu);
9269 	ret = __set_sregs(vcpu, sregs);
9270 	vcpu_put(vcpu);
9271 	return ret;
9272 }
9273 
9274 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9275 					struct kvm_guest_debug *dbg)
9276 {
9277 	unsigned long rflags;
9278 	int i, r;
9279 
9280 	vcpu_load(vcpu);
9281 
9282 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9283 		r = -EBUSY;
9284 		if (vcpu->arch.exception.pending)
9285 			goto out;
9286 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9287 			kvm_queue_exception(vcpu, DB_VECTOR);
9288 		else
9289 			kvm_queue_exception(vcpu, BP_VECTOR);
9290 	}
9291 
9292 	/*
9293 	 * Read rflags as long as potentially injected trace flags are still
9294 	 * filtered out.
9295 	 */
9296 	rflags = kvm_get_rflags(vcpu);
9297 
9298 	vcpu->guest_debug = dbg->control;
9299 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9300 		vcpu->guest_debug = 0;
9301 
9302 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9303 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
9304 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9305 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9306 	} else {
9307 		for (i = 0; i < KVM_NR_DB_REGS; i++)
9308 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9309 	}
9310 	kvm_update_dr7(vcpu);
9311 
9312 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9313 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9314 			get_segment_base(vcpu, VCPU_SREG_CS);
9315 
9316 	/*
9317 	 * Trigger an rflags update that will inject or remove the trace
9318 	 * flags.
9319 	 */
9320 	kvm_set_rflags(vcpu, rflags);
9321 
9322 	kvm_x86_ops.update_exception_bitmap(vcpu);
9323 
9324 	r = 0;
9325 
9326 out:
9327 	vcpu_put(vcpu);
9328 	return r;
9329 }
9330 
9331 /*
9332  * Translate a guest virtual address to a guest physical address.
9333  */
9334 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9335 				    struct kvm_translation *tr)
9336 {
9337 	unsigned long vaddr = tr->linear_address;
9338 	gpa_t gpa;
9339 	int idx;
9340 
9341 	vcpu_load(vcpu);
9342 
9343 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9344 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9345 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9346 	tr->physical_address = gpa;
9347 	tr->valid = gpa != UNMAPPED_GVA;
9348 	tr->writeable = 1;
9349 	tr->usermode = 0;
9350 
9351 	vcpu_put(vcpu);
9352 	return 0;
9353 }
9354 
9355 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9356 {
9357 	struct fxregs_state *fxsave;
9358 
9359 	vcpu_load(vcpu);
9360 
9361 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9362 	memcpy(fpu->fpr, fxsave->st_space, 128);
9363 	fpu->fcw = fxsave->cwd;
9364 	fpu->fsw = fxsave->swd;
9365 	fpu->ftwx = fxsave->twd;
9366 	fpu->last_opcode = fxsave->fop;
9367 	fpu->last_ip = fxsave->rip;
9368 	fpu->last_dp = fxsave->rdp;
9369 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9370 
9371 	vcpu_put(vcpu);
9372 	return 0;
9373 }
9374 
9375 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9376 {
9377 	struct fxregs_state *fxsave;
9378 
9379 	vcpu_load(vcpu);
9380 
9381 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9382 
9383 	memcpy(fxsave->st_space, fpu->fpr, 128);
9384 	fxsave->cwd = fpu->fcw;
9385 	fxsave->swd = fpu->fsw;
9386 	fxsave->twd = fpu->ftwx;
9387 	fxsave->fop = fpu->last_opcode;
9388 	fxsave->rip = fpu->last_ip;
9389 	fxsave->rdp = fpu->last_dp;
9390 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9391 
9392 	vcpu_put(vcpu);
9393 	return 0;
9394 }
9395 
9396 static void store_regs(struct kvm_vcpu *vcpu)
9397 {
9398 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9399 
9400 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9401 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
9402 
9403 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9404 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9405 
9406 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9407 		kvm_vcpu_ioctl_x86_get_vcpu_events(
9408 				vcpu, &vcpu->run->s.regs.events);
9409 }
9410 
9411 static int sync_regs(struct kvm_vcpu *vcpu)
9412 {
9413 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9414 		return -EINVAL;
9415 
9416 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9417 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
9418 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9419 	}
9420 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9421 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9422 			return -EINVAL;
9423 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9424 	}
9425 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9426 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9427 				vcpu, &vcpu->run->s.regs.events))
9428 			return -EINVAL;
9429 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9430 	}
9431 
9432 	return 0;
9433 }
9434 
9435 static void fx_init(struct kvm_vcpu *vcpu)
9436 {
9437 	fpstate_init(&vcpu->arch.guest_fpu->state);
9438 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9439 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9440 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
9441 
9442 	/*
9443 	 * Ensure guest xcr0 is valid for loading
9444 	 */
9445 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9446 
9447 	vcpu->arch.cr0 |= X86_CR0_ET;
9448 }
9449 
9450 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9451 {
9452 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9453 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9454 			     "guest TSC will not be reliable\n");
9455 
9456 	return 0;
9457 }
9458 
9459 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9460 {
9461 	struct page *page;
9462 	int r;
9463 
9464 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9465 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9466 	else
9467 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9468 
9469 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
9470 
9471 	r = kvm_mmu_create(vcpu);
9472 	if (r < 0)
9473 		return r;
9474 
9475 	if (irqchip_in_kernel(vcpu->kvm)) {
9476 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9477 		if (r < 0)
9478 			goto fail_mmu_destroy;
9479 		if (kvm_apicv_activated(vcpu->kvm))
9480 			vcpu->arch.apicv_active = true;
9481 	} else
9482 		static_key_slow_inc(&kvm_no_apic_vcpu);
9483 
9484 	r = -ENOMEM;
9485 
9486 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9487 	if (!page)
9488 		goto fail_free_lapic;
9489 	vcpu->arch.pio_data = page_address(page);
9490 
9491 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9492 				       GFP_KERNEL_ACCOUNT);
9493 	if (!vcpu->arch.mce_banks)
9494 		goto fail_free_pio_data;
9495 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9496 
9497 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9498 				GFP_KERNEL_ACCOUNT))
9499 		goto fail_free_mce_banks;
9500 
9501 	if (!alloc_emulate_ctxt(vcpu))
9502 		goto free_wbinvd_dirty_mask;
9503 
9504 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9505 						GFP_KERNEL_ACCOUNT);
9506 	if (!vcpu->arch.user_fpu) {
9507 		pr_err("kvm: failed to allocate userspace's fpu\n");
9508 		goto free_emulate_ctxt;
9509 	}
9510 
9511 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9512 						 GFP_KERNEL_ACCOUNT);
9513 	if (!vcpu->arch.guest_fpu) {
9514 		pr_err("kvm: failed to allocate vcpu's fpu\n");
9515 		goto free_user_fpu;
9516 	}
9517 	fx_init(vcpu);
9518 
9519 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9520 
9521 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9522 
9523 	kvm_async_pf_hash_reset(vcpu);
9524 	kvm_pmu_init(vcpu);
9525 
9526 	vcpu->arch.pending_external_vector = -1;
9527 	vcpu->arch.preempted_in_kernel = false;
9528 
9529 	kvm_hv_vcpu_init(vcpu);
9530 
9531 	r = kvm_x86_ops.vcpu_create(vcpu);
9532 	if (r)
9533 		goto free_guest_fpu;
9534 
9535 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9536 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9537 	kvm_vcpu_mtrr_init(vcpu);
9538 	vcpu_load(vcpu);
9539 	kvm_vcpu_reset(vcpu, false);
9540 	kvm_init_mmu(vcpu, false);
9541 	vcpu_put(vcpu);
9542 	return 0;
9543 
9544 free_guest_fpu:
9545 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9546 free_user_fpu:
9547 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9548 free_emulate_ctxt:
9549 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9550 free_wbinvd_dirty_mask:
9551 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9552 fail_free_mce_banks:
9553 	kfree(vcpu->arch.mce_banks);
9554 fail_free_pio_data:
9555 	free_page((unsigned long)vcpu->arch.pio_data);
9556 fail_free_lapic:
9557 	kvm_free_lapic(vcpu);
9558 fail_mmu_destroy:
9559 	kvm_mmu_destroy(vcpu);
9560 	return r;
9561 }
9562 
9563 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9564 {
9565 	struct msr_data msr;
9566 	struct kvm *kvm = vcpu->kvm;
9567 
9568 	kvm_hv_vcpu_postcreate(vcpu);
9569 
9570 	if (mutex_lock_killable(&vcpu->mutex))
9571 		return;
9572 	vcpu_load(vcpu);
9573 	msr.data = 0x0;
9574 	msr.index = MSR_IA32_TSC;
9575 	msr.host_initiated = true;
9576 	kvm_write_tsc(vcpu, &msr);
9577 	vcpu_put(vcpu);
9578 
9579 	/* poll control enabled by default */
9580 	vcpu->arch.msr_kvm_poll_control = 1;
9581 
9582 	mutex_unlock(&vcpu->mutex);
9583 
9584 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9585 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9586 						KVMCLOCK_SYNC_PERIOD);
9587 }
9588 
9589 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9590 {
9591 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9592 	int idx;
9593 
9594 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
9595 
9596 	kvmclock_reset(vcpu);
9597 
9598 	kvm_x86_ops.vcpu_free(vcpu);
9599 
9600 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9601 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9602 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9603 	kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9604 
9605 	kvm_hv_vcpu_uninit(vcpu);
9606 	kvm_pmu_destroy(vcpu);
9607 	kfree(vcpu->arch.mce_banks);
9608 	kvm_free_lapic(vcpu);
9609 	idx = srcu_read_lock(&vcpu->kvm->srcu);
9610 	kvm_mmu_destroy(vcpu);
9611 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
9612 	free_page((unsigned long)vcpu->arch.pio_data);
9613 	if (!lapic_in_kernel(vcpu))
9614 		static_key_slow_dec(&kvm_no_apic_vcpu);
9615 }
9616 
9617 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9618 {
9619 	kvm_lapic_reset(vcpu, init_event);
9620 
9621 	vcpu->arch.hflags = 0;
9622 
9623 	vcpu->arch.smi_pending = 0;
9624 	vcpu->arch.smi_count = 0;
9625 	atomic_set(&vcpu->arch.nmi_queued, 0);
9626 	vcpu->arch.nmi_pending = 0;
9627 	vcpu->arch.nmi_injected = false;
9628 	kvm_clear_interrupt_queue(vcpu);
9629 	kvm_clear_exception_queue(vcpu);
9630 
9631 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9632 	kvm_update_dr0123(vcpu);
9633 	vcpu->arch.dr6 = DR6_INIT;
9634 	vcpu->arch.dr7 = DR7_FIXED_1;
9635 	kvm_update_dr7(vcpu);
9636 
9637 	vcpu->arch.cr2 = 0;
9638 
9639 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9640 	vcpu->arch.apf.msr_en_val = 0;
9641 	vcpu->arch.apf.msr_int_val = 0;
9642 	vcpu->arch.st.msr_val = 0;
9643 
9644 	kvmclock_reset(vcpu);
9645 
9646 	kvm_clear_async_pf_completion_queue(vcpu);
9647 	kvm_async_pf_hash_reset(vcpu);
9648 	vcpu->arch.apf.halted = false;
9649 
9650 	if (kvm_mpx_supported()) {
9651 		void *mpx_state_buffer;
9652 
9653 		/*
9654 		 * To avoid have the INIT path from kvm_apic_has_events() that be
9655 		 * called with loaded FPU and does not let userspace fix the state.
9656 		 */
9657 		if (init_event)
9658 			kvm_put_guest_fpu(vcpu);
9659 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9660 					XFEATURE_BNDREGS);
9661 		if (mpx_state_buffer)
9662 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9663 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9664 					XFEATURE_BNDCSR);
9665 		if (mpx_state_buffer)
9666 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9667 		if (init_event)
9668 			kvm_load_guest_fpu(vcpu);
9669 	}
9670 
9671 	if (!init_event) {
9672 		kvm_pmu_reset(vcpu);
9673 		vcpu->arch.smbase = 0x30000;
9674 
9675 		vcpu->arch.msr_misc_features_enables = 0;
9676 
9677 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9678 	}
9679 
9680 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9681 	vcpu->arch.regs_avail = ~0;
9682 	vcpu->arch.regs_dirty = ~0;
9683 
9684 	vcpu->arch.ia32_xss = 0;
9685 
9686 	kvm_x86_ops.vcpu_reset(vcpu, init_event);
9687 }
9688 
9689 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9690 {
9691 	struct kvm_segment cs;
9692 
9693 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9694 	cs.selector = vector << 8;
9695 	cs.base = vector << 12;
9696 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9697 	kvm_rip_write(vcpu, 0);
9698 }
9699 
9700 int kvm_arch_hardware_enable(void)
9701 {
9702 	struct kvm *kvm;
9703 	struct kvm_vcpu *vcpu;
9704 	int i;
9705 	int ret;
9706 	u64 local_tsc;
9707 	u64 max_tsc = 0;
9708 	bool stable, backwards_tsc = false;
9709 
9710 	kvm_shared_msr_cpu_online();
9711 	ret = kvm_x86_ops.hardware_enable();
9712 	if (ret != 0)
9713 		return ret;
9714 
9715 	local_tsc = rdtsc();
9716 	stable = !kvm_check_tsc_unstable();
9717 	list_for_each_entry(kvm, &vm_list, vm_list) {
9718 		kvm_for_each_vcpu(i, vcpu, kvm) {
9719 			if (!stable && vcpu->cpu == smp_processor_id())
9720 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9721 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9722 				backwards_tsc = true;
9723 				if (vcpu->arch.last_host_tsc > max_tsc)
9724 					max_tsc = vcpu->arch.last_host_tsc;
9725 			}
9726 		}
9727 	}
9728 
9729 	/*
9730 	 * Sometimes, even reliable TSCs go backwards.  This happens on
9731 	 * platforms that reset TSC during suspend or hibernate actions, but
9732 	 * maintain synchronization.  We must compensate.  Fortunately, we can
9733 	 * detect that condition here, which happens early in CPU bringup,
9734 	 * before any KVM threads can be running.  Unfortunately, we can't
9735 	 * bring the TSCs fully up to date with real time, as we aren't yet far
9736 	 * enough into CPU bringup that we know how much real time has actually
9737 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9738 	 * variables that haven't been updated yet.
9739 	 *
9740 	 * So we simply find the maximum observed TSC above, then record the
9741 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9742 	 * the adjustment will be applied.  Note that we accumulate
9743 	 * adjustments, in case multiple suspend cycles happen before some VCPU
9744 	 * gets a chance to run again.  In the event that no KVM threads get a
9745 	 * chance to run, we will miss the entire elapsed period, as we'll have
9746 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9747 	 * loose cycle time.  This isn't too big a deal, since the loss will be
9748 	 * uniform across all VCPUs (not to mention the scenario is extremely
9749 	 * unlikely). It is possible that a second hibernate recovery happens
9750 	 * much faster than a first, causing the observed TSC here to be
9751 	 * smaller; this would require additional padding adjustment, which is
9752 	 * why we set last_host_tsc to the local tsc observed here.
9753 	 *
9754 	 * N.B. - this code below runs only on platforms with reliable TSC,
9755 	 * as that is the only way backwards_tsc is set above.  Also note
9756 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9757 	 * have the same delta_cyc adjustment applied if backwards_tsc
9758 	 * is detected.  Note further, this adjustment is only done once,
9759 	 * as we reset last_host_tsc on all VCPUs to stop this from being
9760 	 * called multiple times (one for each physical CPU bringup).
9761 	 *
9762 	 * Platforms with unreliable TSCs don't have to deal with this, they
9763 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
9764 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
9765 	 * guarantee that they stay in perfect synchronization.
9766 	 */
9767 	if (backwards_tsc) {
9768 		u64 delta_cyc = max_tsc - local_tsc;
9769 		list_for_each_entry(kvm, &vm_list, vm_list) {
9770 			kvm->arch.backwards_tsc_observed = true;
9771 			kvm_for_each_vcpu(i, vcpu, kvm) {
9772 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
9773 				vcpu->arch.last_host_tsc = local_tsc;
9774 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9775 			}
9776 
9777 			/*
9778 			 * We have to disable TSC offset matching.. if you were
9779 			 * booting a VM while issuing an S4 host suspend....
9780 			 * you may have some problem.  Solving this issue is
9781 			 * left as an exercise to the reader.
9782 			 */
9783 			kvm->arch.last_tsc_nsec = 0;
9784 			kvm->arch.last_tsc_write = 0;
9785 		}
9786 
9787 	}
9788 	return 0;
9789 }
9790 
9791 void kvm_arch_hardware_disable(void)
9792 {
9793 	kvm_x86_ops.hardware_disable();
9794 	drop_user_return_notifiers();
9795 }
9796 
9797 int kvm_arch_hardware_setup(void *opaque)
9798 {
9799 	struct kvm_x86_init_ops *ops = opaque;
9800 	int r;
9801 
9802 	rdmsrl_safe(MSR_EFER, &host_efer);
9803 
9804 	if (boot_cpu_has(X86_FEATURE_XSAVES))
9805 		rdmsrl(MSR_IA32_XSS, host_xss);
9806 
9807 	r = ops->hardware_setup();
9808 	if (r != 0)
9809 		return r;
9810 
9811 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9812 
9813 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9814 		supported_xss = 0;
9815 
9816 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9817 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9818 #undef __kvm_cpu_cap_has
9819 
9820 	if (kvm_has_tsc_control) {
9821 		/*
9822 		 * Make sure the user can only configure tsc_khz values that
9823 		 * fit into a signed integer.
9824 		 * A min value is not calculated because it will always
9825 		 * be 1 on all machines.
9826 		 */
9827 		u64 max = min(0x7fffffffULL,
9828 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9829 		kvm_max_guest_tsc_khz = max;
9830 
9831 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9832 	}
9833 
9834 	kvm_init_msr_list();
9835 	return 0;
9836 }
9837 
9838 void kvm_arch_hardware_unsetup(void)
9839 {
9840 	kvm_x86_ops.hardware_unsetup();
9841 }
9842 
9843 int kvm_arch_check_processor_compat(void *opaque)
9844 {
9845 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9846 	struct kvm_x86_init_ops *ops = opaque;
9847 
9848 	WARN_ON(!irqs_disabled());
9849 
9850 	if (__cr4_reserved_bits(cpu_has, c) !=
9851 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9852 		return -EIO;
9853 
9854 	return ops->check_processor_compatibility();
9855 }
9856 
9857 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9858 {
9859 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9860 }
9861 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9862 
9863 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9864 {
9865 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9866 }
9867 
9868 struct static_key kvm_no_apic_vcpu __read_mostly;
9869 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9870 
9871 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9872 {
9873 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9874 
9875 	vcpu->arch.l1tf_flush_l1d = true;
9876 	if (pmu->version && unlikely(pmu->event_count)) {
9877 		pmu->need_cleanup = true;
9878 		kvm_make_request(KVM_REQ_PMU, vcpu);
9879 	}
9880 	kvm_x86_ops.sched_in(vcpu, cpu);
9881 }
9882 
9883 void kvm_arch_free_vm(struct kvm *kvm)
9884 {
9885 	kfree(kvm->arch.hyperv.hv_pa_pg);
9886 	vfree(kvm);
9887 }
9888 
9889 
9890 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9891 {
9892 	if (type)
9893 		return -EINVAL;
9894 
9895 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9896 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9897 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9898 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9899 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9900 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9901 
9902 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9903 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9904 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9905 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9906 		&kvm->arch.irq_sources_bitmap);
9907 
9908 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9909 	mutex_init(&kvm->arch.apic_map_lock);
9910 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9911 
9912 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9913 	pvclock_update_vm_gtod_copy(kvm);
9914 
9915 	kvm->arch.guest_can_read_msr_platform_info = true;
9916 
9917 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9918 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9919 
9920 	kvm_hv_init_vm(kvm);
9921 	kvm_page_track_init(kvm);
9922 	kvm_mmu_init_vm(kvm);
9923 
9924 	return kvm_x86_ops.vm_init(kvm);
9925 }
9926 
9927 int kvm_arch_post_init_vm(struct kvm *kvm)
9928 {
9929 	return kvm_mmu_post_init_vm(kvm);
9930 }
9931 
9932 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9933 {
9934 	vcpu_load(vcpu);
9935 	kvm_mmu_unload(vcpu);
9936 	vcpu_put(vcpu);
9937 }
9938 
9939 static void kvm_free_vcpus(struct kvm *kvm)
9940 {
9941 	unsigned int i;
9942 	struct kvm_vcpu *vcpu;
9943 
9944 	/*
9945 	 * Unpin any mmu pages first.
9946 	 */
9947 	kvm_for_each_vcpu(i, vcpu, kvm) {
9948 		kvm_clear_async_pf_completion_queue(vcpu);
9949 		kvm_unload_vcpu_mmu(vcpu);
9950 	}
9951 	kvm_for_each_vcpu(i, vcpu, kvm)
9952 		kvm_vcpu_destroy(vcpu);
9953 
9954 	mutex_lock(&kvm->lock);
9955 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9956 		kvm->vcpus[i] = NULL;
9957 
9958 	atomic_set(&kvm->online_vcpus, 0);
9959 	mutex_unlock(&kvm->lock);
9960 }
9961 
9962 void kvm_arch_sync_events(struct kvm *kvm)
9963 {
9964 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9965 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9966 	kvm_free_pit(kvm);
9967 }
9968 
9969 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9970 {
9971 	int i, r;
9972 	unsigned long hva, old_npages;
9973 	struct kvm_memslots *slots = kvm_memslots(kvm);
9974 	struct kvm_memory_slot *slot;
9975 
9976 	/* Called with kvm->slots_lock held.  */
9977 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9978 		return -EINVAL;
9979 
9980 	slot = id_to_memslot(slots, id);
9981 	if (size) {
9982 		if (slot && slot->npages)
9983 			return -EEXIST;
9984 
9985 		/*
9986 		 * MAP_SHARED to prevent internal slot pages from being moved
9987 		 * by fork()/COW.
9988 		 */
9989 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9990 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9991 		if (IS_ERR((void *)hva))
9992 			return PTR_ERR((void *)hva);
9993 	} else {
9994 		if (!slot || !slot->npages)
9995 			return 0;
9996 
9997 		old_npages = slot->npages;
9998 		hva = 0;
9999 	}
10000 
10001 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10002 		struct kvm_userspace_memory_region m;
10003 
10004 		m.slot = id | (i << 16);
10005 		m.flags = 0;
10006 		m.guest_phys_addr = gpa;
10007 		m.userspace_addr = hva;
10008 		m.memory_size = size;
10009 		r = __kvm_set_memory_region(kvm, &m);
10010 		if (r < 0)
10011 			return r;
10012 	}
10013 
10014 	if (!size)
10015 		vm_munmap(hva, old_npages * PAGE_SIZE);
10016 
10017 	return 0;
10018 }
10019 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10020 
10021 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10022 {
10023 	kvm_mmu_pre_destroy_vm(kvm);
10024 }
10025 
10026 void kvm_arch_destroy_vm(struct kvm *kvm)
10027 {
10028 	if (current->mm == kvm->mm) {
10029 		/*
10030 		 * Free memory regions allocated on behalf of userspace,
10031 		 * unless the the memory map has changed due to process exit
10032 		 * or fd copying.
10033 		 */
10034 		mutex_lock(&kvm->slots_lock);
10035 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10036 					0, 0);
10037 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10038 					0, 0);
10039 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10040 		mutex_unlock(&kvm->slots_lock);
10041 	}
10042 	if (kvm_x86_ops.vm_destroy)
10043 		kvm_x86_ops.vm_destroy(kvm);
10044 	kvm_pic_destroy(kvm);
10045 	kvm_ioapic_destroy(kvm);
10046 	kvm_free_vcpus(kvm);
10047 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10048 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10049 	kvm_mmu_uninit_vm(kvm);
10050 	kvm_page_track_cleanup(kvm);
10051 	kvm_hv_destroy_vm(kvm);
10052 }
10053 
10054 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10055 {
10056 	int i;
10057 
10058 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10059 		kvfree(slot->arch.rmap[i]);
10060 		slot->arch.rmap[i] = NULL;
10061 
10062 		if (i == 0)
10063 			continue;
10064 
10065 		kvfree(slot->arch.lpage_info[i - 1]);
10066 		slot->arch.lpage_info[i - 1] = NULL;
10067 	}
10068 
10069 	kvm_page_track_free_memslot(slot);
10070 }
10071 
10072 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10073 				      unsigned long npages)
10074 {
10075 	int i;
10076 
10077 	/*
10078 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
10079 	 * old arrays will be freed by __kvm_set_memory_region() if installing
10080 	 * the new memslot is successful.
10081 	 */
10082 	memset(&slot->arch, 0, sizeof(slot->arch));
10083 
10084 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10085 		struct kvm_lpage_info *linfo;
10086 		unsigned long ugfn;
10087 		int lpages;
10088 		int level = i + 1;
10089 
10090 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
10091 				      slot->base_gfn, level) + 1;
10092 
10093 		slot->arch.rmap[i] =
10094 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10095 				 GFP_KERNEL_ACCOUNT);
10096 		if (!slot->arch.rmap[i])
10097 			goto out_free;
10098 		if (i == 0)
10099 			continue;
10100 
10101 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10102 		if (!linfo)
10103 			goto out_free;
10104 
10105 		slot->arch.lpage_info[i - 1] = linfo;
10106 
10107 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10108 			linfo[0].disallow_lpage = 1;
10109 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10110 			linfo[lpages - 1].disallow_lpage = 1;
10111 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
10112 		/*
10113 		 * If the gfn and userspace address are not aligned wrt each
10114 		 * other, disable large page support for this slot.
10115 		 */
10116 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10117 			unsigned long j;
10118 
10119 			for (j = 0; j < lpages; ++j)
10120 				linfo[j].disallow_lpage = 1;
10121 		}
10122 	}
10123 
10124 	if (kvm_page_track_create_memslot(slot, npages))
10125 		goto out_free;
10126 
10127 	return 0;
10128 
10129 out_free:
10130 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10131 		kvfree(slot->arch.rmap[i]);
10132 		slot->arch.rmap[i] = NULL;
10133 		if (i == 0)
10134 			continue;
10135 
10136 		kvfree(slot->arch.lpage_info[i - 1]);
10137 		slot->arch.lpage_info[i - 1] = NULL;
10138 	}
10139 	return -ENOMEM;
10140 }
10141 
10142 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10143 {
10144 	struct kvm_vcpu *vcpu;
10145 	int i;
10146 
10147 	/*
10148 	 * memslots->generation has been incremented.
10149 	 * mmio generation may have reached its maximum value.
10150 	 */
10151 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10152 
10153 	/* Force re-initialization of steal_time cache */
10154 	kvm_for_each_vcpu(i, vcpu, kvm)
10155 		kvm_vcpu_kick(vcpu);
10156 }
10157 
10158 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10159 				struct kvm_memory_slot *memslot,
10160 				const struct kvm_userspace_memory_region *mem,
10161 				enum kvm_mr_change change)
10162 {
10163 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10164 		return kvm_alloc_memslot_metadata(memslot,
10165 						  mem->memory_size >> PAGE_SHIFT);
10166 	return 0;
10167 }
10168 
10169 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10170 				     struct kvm_memory_slot *old,
10171 				     struct kvm_memory_slot *new,
10172 				     enum kvm_mr_change change)
10173 {
10174 	/*
10175 	 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10176 	 * See comments below.
10177 	 */
10178 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10179 		return;
10180 
10181 	/*
10182 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
10183 	 * sptes have to be split.  If live migration is successful, the guest
10184 	 * in the source machine will be destroyed and large sptes will be
10185 	 * created in the destination. However, if the guest continues to run
10186 	 * in the source machine (for example if live migration fails), small
10187 	 * sptes will remain around and cause bad performance.
10188 	 *
10189 	 * Scan sptes if dirty logging has been stopped, dropping those
10190 	 * which can be collapsed into a single large-page spte.  Later
10191 	 * page faults will create the large-page sptes.
10192 	 *
10193 	 * There is no need to do this in any of the following cases:
10194 	 * CREATE:      No dirty mappings will already exist.
10195 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
10196 	 *		kvm_arch_flush_shadow_memslot()
10197 	 */
10198 	if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10199 	    !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10200 		kvm_mmu_zap_collapsible_sptes(kvm, new);
10201 
10202 	/*
10203 	 * Enable or disable dirty logging for the slot.
10204 	 *
10205 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10206 	 * slot have been zapped so no dirty logging updates are needed for
10207 	 * the old slot.
10208 	 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10209 	 * any mappings that might be created in it will consume the
10210 	 * properties of the new slot and do not need to be updated here.
10211 	 *
10212 	 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10213 	 * called to enable/disable dirty logging.
10214 	 *
10215 	 * When disabling dirty logging with PML enabled, the D-bit is set
10216 	 * for sptes in the slot in order to prevent unnecessary GPA
10217 	 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10218 	 * This guarantees leaving PML enabled for the guest's lifetime
10219 	 * won't have any additional overhead from PML when the guest is
10220 	 * running with dirty logging disabled.
10221 	 *
10222 	 * When enabling dirty logging, large sptes are write-protected
10223 	 * so they can be split on first write.  New large sptes cannot
10224 	 * be created for this slot until the end of the logging.
10225 	 * See the comments in fast_page_fault().
10226 	 * For small sptes, nothing is done if the dirty log is in the
10227 	 * initial-all-set state.  Otherwise, depending on whether pml
10228 	 * is enabled the D-bit or the W-bit will be cleared.
10229 	 */
10230 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10231 		if (kvm_x86_ops.slot_enable_log_dirty) {
10232 			kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10233 		} else {
10234 			int level =
10235 				kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10236 				PG_LEVEL_2M : PG_LEVEL_4K;
10237 
10238 			/*
10239 			 * If we're with initial-all-set, we don't need
10240 			 * to write protect any small page because
10241 			 * they're reported as dirty already.  However
10242 			 * we still need to write-protect huge pages
10243 			 * so that the page split can happen lazily on
10244 			 * the first write to the huge page.
10245 			 */
10246 			kvm_mmu_slot_remove_write_access(kvm, new, level);
10247 		}
10248 	} else {
10249 		if (kvm_x86_ops.slot_disable_log_dirty)
10250 			kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10251 	}
10252 }
10253 
10254 void kvm_arch_commit_memory_region(struct kvm *kvm,
10255 				const struct kvm_userspace_memory_region *mem,
10256 				struct kvm_memory_slot *old,
10257 				const struct kvm_memory_slot *new,
10258 				enum kvm_mr_change change)
10259 {
10260 	if (!kvm->arch.n_requested_mmu_pages)
10261 		kvm_mmu_change_mmu_pages(kvm,
10262 				kvm_mmu_calculate_default_mmu_pages(kvm));
10263 
10264 	/*
10265 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
10266 	 */
10267 	kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10268 
10269 	/* Free the arrays associated with the old memslot. */
10270 	if (change == KVM_MR_MOVE)
10271 		kvm_arch_free_memslot(kvm, old);
10272 }
10273 
10274 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10275 {
10276 	kvm_mmu_zap_all(kvm);
10277 }
10278 
10279 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10280 				   struct kvm_memory_slot *slot)
10281 {
10282 	kvm_page_track_flush_slot(kvm, slot);
10283 }
10284 
10285 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10286 {
10287 	return (is_guest_mode(vcpu) &&
10288 			kvm_x86_ops.guest_apic_has_interrupt &&
10289 			kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10290 }
10291 
10292 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10293 {
10294 	if (!list_empty_careful(&vcpu->async_pf.done))
10295 		return true;
10296 
10297 	if (kvm_apic_has_events(vcpu))
10298 		return true;
10299 
10300 	if (vcpu->arch.pv.pv_unhalted)
10301 		return true;
10302 
10303 	if (vcpu->arch.exception.pending)
10304 		return true;
10305 
10306 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10307 	    (vcpu->arch.nmi_pending &&
10308 	     kvm_x86_ops.nmi_allowed(vcpu, false)))
10309 		return true;
10310 
10311 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10312 	    (vcpu->arch.smi_pending &&
10313 	     kvm_x86_ops.smi_allowed(vcpu, false)))
10314 		return true;
10315 
10316 	if (kvm_arch_interrupt_allowed(vcpu) &&
10317 	    (kvm_cpu_has_interrupt(vcpu) ||
10318 	    kvm_guest_apic_has_interrupt(vcpu)))
10319 		return true;
10320 
10321 	if (kvm_hv_has_stimer_pending(vcpu))
10322 		return true;
10323 
10324 	if (is_guest_mode(vcpu) &&
10325 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
10326 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10327 		return true;
10328 
10329 	return false;
10330 }
10331 
10332 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10333 {
10334 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10335 }
10336 
10337 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10338 {
10339 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10340 		return true;
10341 
10342 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10343 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
10344 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
10345 		return true;
10346 
10347 	if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10348 		return true;
10349 
10350 	return false;
10351 }
10352 
10353 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10354 {
10355 	return vcpu->arch.preempted_in_kernel;
10356 }
10357 
10358 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10359 {
10360 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10361 }
10362 
10363 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10364 {
10365 	return kvm_x86_ops.interrupt_allowed(vcpu, false);
10366 }
10367 
10368 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10369 {
10370 	if (is_64_bit_mode(vcpu))
10371 		return kvm_rip_read(vcpu);
10372 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10373 		     kvm_rip_read(vcpu));
10374 }
10375 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10376 
10377 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10378 {
10379 	return kvm_get_linear_rip(vcpu) == linear_rip;
10380 }
10381 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10382 
10383 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10384 {
10385 	unsigned long rflags;
10386 
10387 	rflags = kvm_x86_ops.get_rflags(vcpu);
10388 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10389 		rflags &= ~X86_EFLAGS_TF;
10390 	return rflags;
10391 }
10392 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10393 
10394 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10395 {
10396 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10397 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10398 		rflags |= X86_EFLAGS_TF;
10399 	kvm_x86_ops.set_rflags(vcpu, rflags);
10400 }
10401 
10402 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10403 {
10404 	__kvm_set_rflags(vcpu, rflags);
10405 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10406 }
10407 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10408 
10409 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10410 {
10411 	int r;
10412 
10413 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10414 	      work->wakeup_all)
10415 		return;
10416 
10417 	r = kvm_mmu_reload(vcpu);
10418 	if (unlikely(r))
10419 		return;
10420 
10421 	if (!vcpu->arch.mmu->direct_map &&
10422 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10423 		return;
10424 
10425 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10426 }
10427 
10428 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10429 {
10430 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10431 
10432 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10433 }
10434 
10435 static inline u32 kvm_async_pf_next_probe(u32 key)
10436 {
10437 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10438 }
10439 
10440 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10441 {
10442 	u32 key = kvm_async_pf_hash_fn(gfn);
10443 
10444 	while (vcpu->arch.apf.gfns[key] != ~0)
10445 		key = kvm_async_pf_next_probe(key);
10446 
10447 	vcpu->arch.apf.gfns[key] = gfn;
10448 }
10449 
10450 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10451 {
10452 	int i;
10453 	u32 key = kvm_async_pf_hash_fn(gfn);
10454 
10455 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
10456 		     (vcpu->arch.apf.gfns[key] != gfn &&
10457 		      vcpu->arch.apf.gfns[key] != ~0); i++)
10458 		key = kvm_async_pf_next_probe(key);
10459 
10460 	return key;
10461 }
10462 
10463 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10464 {
10465 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10466 }
10467 
10468 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10469 {
10470 	u32 i, j, k;
10471 
10472 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10473 
10474 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10475 		return;
10476 
10477 	while (true) {
10478 		vcpu->arch.apf.gfns[i] = ~0;
10479 		do {
10480 			j = kvm_async_pf_next_probe(j);
10481 			if (vcpu->arch.apf.gfns[j] == ~0)
10482 				return;
10483 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10484 			/*
10485 			 * k lies cyclically in ]i,j]
10486 			 * |    i.k.j |
10487 			 * |....j i.k.| or  |.k..j i...|
10488 			 */
10489 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10490 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10491 		i = j;
10492 	}
10493 }
10494 
10495 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
10496 {
10497 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
10498 
10499 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
10500 				      sizeof(reason));
10501 }
10502 
10503 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
10504 {
10505 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10506 
10507 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10508 					     &token, offset, sizeof(token));
10509 }
10510 
10511 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
10512 {
10513 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10514 	u32 val;
10515 
10516 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10517 					 &val, offset, sizeof(val)))
10518 		return false;
10519 
10520 	return !val;
10521 }
10522 
10523 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10524 {
10525 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10526 		return false;
10527 
10528 	if (!kvm_pv_async_pf_enabled(vcpu) ||
10529 	    (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
10530 		return false;
10531 
10532 	return true;
10533 }
10534 
10535 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10536 {
10537 	if (unlikely(!lapic_in_kernel(vcpu) ||
10538 		     kvm_event_needs_reinjection(vcpu) ||
10539 		     vcpu->arch.exception.pending))
10540 		return false;
10541 
10542 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10543 		return false;
10544 
10545 	/*
10546 	 * If interrupts are off we cannot even use an artificial
10547 	 * halt state.
10548 	 */
10549 	return kvm_arch_interrupt_allowed(vcpu);
10550 }
10551 
10552 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10553 				     struct kvm_async_pf *work)
10554 {
10555 	struct x86_exception fault;
10556 
10557 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10558 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10559 
10560 	if (kvm_can_deliver_async_pf(vcpu) &&
10561 	    !apf_put_user_notpresent(vcpu)) {
10562 		fault.vector = PF_VECTOR;
10563 		fault.error_code_valid = true;
10564 		fault.error_code = 0;
10565 		fault.nested_page_fault = false;
10566 		fault.address = work->arch.token;
10567 		fault.async_page_fault = true;
10568 		kvm_inject_page_fault(vcpu, &fault);
10569 		return true;
10570 	} else {
10571 		/*
10572 		 * It is not possible to deliver a paravirtualized asynchronous
10573 		 * page fault, but putting the guest in an artificial halt state
10574 		 * can be beneficial nevertheless: if an interrupt arrives, we
10575 		 * can deliver it timely and perhaps the guest will schedule
10576 		 * another process.  When the instruction that triggered a page
10577 		 * fault is retried, hopefully the page will be ready in the host.
10578 		 */
10579 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10580 		return false;
10581 	}
10582 }
10583 
10584 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10585 				 struct kvm_async_pf *work)
10586 {
10587 	struct kvm_lapic_irq irq = {
10588 		.delivery_mode = APIC_DM_FIXED,
10589 		.vector = vcpu->arch.apf.vec
10590 	};
10591 
10592 	if (work->wakeup_all)
10593 		work->arch.token = ~0; /* broadcast wakeup */
10594 	else
10595 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10596 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10597 
10598 	if ((work->wakeup_all || work->notpresent_injected) &&
10599 	    kvm_pv_async_pf_enabled(vcpu) &&
10600 	    !apf_put_user_ready(vcpu, work->arch.token)) {
10601 		vcpu->arch.apf.pageready_pending = true;
10602 		kvm_apic_set_irq(vcpu, &irq, NULL);
10603 	}
10604 
10605 	vcpu->arch.apf.halted = false;
10606 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10607 }
10608 
10609 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
10610 {
10611 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
10612 	if (!vcpu->arch.apf.pageready_pending)
10613 		kvm_vcpu_kick(vcpu);
10614 }
10615 
10616 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
10617 {
10618 	if (!kvm_pv_async_pf_enabled(vcpu))
10619 		return true;
10620 	else
10621 		return apf_pageready_slot_free(vcpu);
10622 }
10623 
10624 void kvm_arch_start_assignment(struct kvm *kvm)
10625 {
10626 	atomic_inc(&kvm->arch.assigned_device_count);
10627 }
10628 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10629 
10630 void kvm_arch_end_assignment(struct kvm *kvm)
10631 {
10632 	atomic_dec(&kvm->arch.assigned_device_count);
10633 }
10634 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10635 
10636 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10637 {
10638 	return atomic_read(&kvm->arch.assigned_device_count);
10639 }
10640 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10641 
10642 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10643 {
10644 	atomic_inc(&kvm->arch.noncoherent_dma_count);
10645 }
10646 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10647 
10648 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10649 {
10650 	atomic_dec(&kvm->arch.noncoherent_dma_count);
10651 }
10652 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10653 
10654 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10655 {
10656 	return atomic_read(&kvm->arch.noncoherent_dma_count);
10657 }
10658 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10659 
10660 bool kvm_arch_has_irq_bypass(void)
10661 {
10662 	return true;
10663 }
10664 
10665 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10666 				      struct irq_bypass_producer *prod)
10667 {
10668 	struct kvm_kernel_irqfd *irqfd =
10669 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10670 	int ret;
10671 
10672 	irqfd->producer = prod;
10673 	kvm_arch_start_assignment(irqfd->kvm);
10674 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
10675 					 prod->irq, irqfd->gsi, 1);
10676 
10677 	if (ret)
10678 		kvm_arch_end_assignment(irqfd->kvm);
10679 
10680 	return ret;
10681 }
10682 
10683 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10684 				      struct irq_bypass_producer *prod)
10685 {
10686 	int ret;
10687 	struct kvm_kernel_irqfd *irqfd =
10688 		container_of(cons, struct kvm_kernel_irqfd, consumer);
10689 
10690 	WARN_ON(irqfd->producer != prod);
10691 	irqfd->producer = NULL;
10692 
10693 	/*
10694 	 * When producer of consumer is unregistered, we change back to
10695 	 * remapped mode, so we can re-use the current implementation
10696 	 * when the irq is masked/disabled or the consumer side (KVM
10697 	 * int this case doesn't want to receive the interrupts.
10698 	*/
10699 	ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10700 	if (ret)
10701 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10702 		       " fails: %d\n", irqfd->consumer.token, ret);
10703 
10704 	kvm_arch_end_assignment(irqfd->kvm);
10705 }
10706 
10707 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10708 				   uint32_t guest_irq, bool set)
10709 {
10710 	return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10711 }
10712 
10713 bool kvm_vector_hashing_enabled(void)
10714 {
10715 	return vector_hashing;
10716 }
10717 
10718 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10719 {
10720 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10721 }
10722 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10723 
10724 
10725 int kvm_spec_ctrl_test_value(u64 value)
10726 {
10727 	/*
10728 	 * test that setting IA32_SPEC_CTRL to given value
10729 	 * is allowed by the host processor
10730 	 */
10731 
10732 	u64 saved_value;
10733 	unsigned long flags;
10734 	int ret = 0;
10735 
10736 	local_irq_save(flags);
10737 
10738 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
10739 		ret = 1;
10740 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
10741 		ret = 1;
10742 	else
10743 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
10744 
10745 	local_irq_restore(flags);
10746 
10747 	return ret;
10748 }
10749 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
10750 
10751 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
10752 {
10753 	struct x86_exception fault;
10754 
10755 	if (!(error_code & PFERR_PRESENT_MASK) ||
10756 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, error_code, &fault) != UNMAPPED_GVA) {
10757 		/*
10758 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
10759 		 * tables probably do not match the TLB.  Just proceed
10760 		 * with the error code that the processor gave.
10761 		 */
10762 		fault.vector = PF_VECTOR;
10763 		fault.error_code_valid = true;
10764 		fault.error_code = error_code;
10765 		fault.nested_page_fault = false;
10766 		fault.address = gva;
10767 	}
10768 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
10769 }
10770 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
10771 
10772 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10773 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10774 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10775 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10776 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10777 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10778 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10779 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10780 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10781 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10782 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10783 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10784 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10785 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10786 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10787 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10788 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10789 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10790 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10791 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10792 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10793 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
10794