xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 239480ab)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 
58 #include <trace/events/kvm.h>
59 
60 #include <asm/debugreg.h>
61 #include <asm/msr.h>
62 #include <asm/desc.h>
63 #include <asm/mce.h>
64 #include <linux/kernel_stat.h>
65 #include <asm/fpu/internal.h> /* Ugh! */
66 #include <asm/pvclock.h>
67 #include <asm/div64.h>
68 #include <asm/irq_remapping.h>
69 
70 #define CREATE_TRACE_POINTS
71 #include "trace.h"
72 
73 #define MAX_IO_MSRS 256
74 #define KVM_MAX_MCE_BANKS 32
75 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
76 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
77 
78 #define emul_to_vcpu(ctxt) \
79 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
80 
81 /* EFER defaults:
82  * - enable syscall per default because its emulated by KVM
83  * - enable LME and LMA per default on 64 bit KVM
84  */
85 #ifdef CONFIG_X86_64
86 static
87 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
88 #else
89 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
90 #endif
91 
92 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
93 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
94 
95 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
96                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
97 
98 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
99 static void process_nmi(struct kvm_vcpu *vcpu);
100 static void enter_smm(struct kvm_vcpu *vcpu);
101 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
102 
103 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
104 EXPORT_SYMBOL_GPL(kvm_x86_ops);
105 
106 static bool __read_mostly ignore_msrs = 0;
107 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
108 
109 unsigned int min_timer_period_us = 500;
110 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
111 
112 static bool __read_mostly kvmclock_periodic_sync = true;
113 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
114 
115 bool __read_mostly kvm_has_tsc_control;
116 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
117 u32  __read_mostly kvm_max_guest_tsc_khz;
118 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
119 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
120 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
121 u64  __read_mostly kvm_max_tsc_scaling_ratio;
122 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
123 u64 __read_mostly kvm_default_tsc_scaling_ratio;
124 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
125 
126 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
127 static u32 __read_mostly tsc_tolerance_ppm = 250;
128 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
129 
130 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
131 unsigned int __read_mostly lapic_timer_advance_ns = 0;
132 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
133 
134 static bool __read_mostly vector_hashing = true;
135 module_param(vector_hashing, bool, S_IRUGO);
136 
137 static bool __read_mostly backwards_tsc_observed = false;
138 
139 #define KVM_NR_SHARED_MSRS 16
140 
141 struct kvm_shared_msrs_global {
142 	int nr;
143 	u32 msrs[KVM_NR_SHARED_MSRS];
144 };
145 
146 struct kvm_shared_msrs {
147 	struct user_return_notifier urn;
148 	bool registered;
149 	struct kvm_shared_msr_values {
150 		u64 host;
151 		u64 curr;
152 	} values[KVM_NR_SHARED_MSRS];
153 };
154 
155 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
156 static struct kvm_shared_msrs __percpu *shared_msrs;
157 
158 struct kvm_stats_debugfs_item debugfs_entries[] = {
159 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
160 	{ "pf_guest", VCPU_STAT(pf_guest) },
161 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
162 	{ "invlpg", VCPU_STAT(invlpg) },
163 	{ "exits", VCPU_STAT(exits) },
164 	{ "io_exits", VCPU_STAT(io_exits) },
165 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
166 	{ "signal_exits", VCPU_STAT(signal_exits) },
167 	{ "irq_window", VCPU_STAT(irq_window_exits) },
168 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
169 	{ "halt_exits", VCPU_STAT(halt_exits) },
170 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
171 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
172 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
173 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
174 	{ "hypercalls", VCPU_STAT(hypercalls) },
175 	{ "request_irq", VCPU_STAT(request_irq_exits) },
176 	{ "irq_exits", VCPU_STAT(irq_exits) },
177 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
178 	{ "efer_reload", VCPU_STAT(efer_reload) },
179 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
180 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
181 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
182 	{ "irq_injections", VCPU_STAT(irq_injections) },
183 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
184 	{ "req_event", VCPU_STAT(req_event) },
185 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
186 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
187 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
188 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
189 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
190 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
191 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
192 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
193 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
194 	{ "largepages", VM_STAT(lpages) },
195 	{ "max_mmu_page_hash_collisions",
196 		VM_STAT(max_mmu_page_hash_collisions) },
197 	{ NULL }
198 };
199 
200 u64 __read_mostly host_xcr0;
201 
202 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
203 
204 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
205 {
206 	int i;
207 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
208 		vcpu->arch.apf.gfns[i] = ~0;
209 }
210 
211 static void kvm_on_user_return(struct user_return_notifier *urn)
212 {
213 	unsigned slot;
214 	struct kvm_shared_msrs *locals
215 		= container_of(urn, struct kvm_shared_msrs, urn);
216 	struct kvm_shared_msr_values *values;
217 	unsigned long flags;
218 
219 	/*
220 	 * Disabling irqs at this point since the following code could be
221 	 * interrupted and executed through kvm_arch_hardware_disable()
222 	 */
223 	local_irq_save(flags);
224 	if (locals->registered) {
225 		locals->registered = false;
226 		user_return_notifier_unregister(urn);
227 	}
228 	local_irq_restore(flags);
229 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
230 		values = &locals->values[slot];
231 		if (values->host != values->curr) {
232 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
233 			values->curr = values->host;
234 		}
235 	}
236 }
237 
238 static void shared_msr_update(unsigned slot, u32 msr)
239 {
240 	u64 value;
241 	unsigned int cpu = smp_processor_id();
242 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
243 
244 	/* only read, and nobody should modify it at this time,
245 	 * so don't need lock */
246 	if (slot >= shared_msrs_global.nr) {
247 		printk(KERN_ERR "kvm: invalid MSR slot!");
248 		return;
249 	}
250 	rdmsrl_safe(msr, &value);
251 	smsr->values[slot].host = value;
252 	smsr->values[slot].curr = value;
253 }
254 
255 void kvm_define_shared_msr(unsigned slot, u32 msr)
256 {
257 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
258 	shared_msrs_global.msrs[slot] = msr;
259 	if (slot >= shared_msrs_global.nr)
260 		shared_msrs_global.nr = slot + 1;
261 }
262 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
263 
264 static void kvm_shared_msr_cpu_online(void)
265 {
266 	unsigned i;
267 
268 	for (i = 0; i < shared_msrs_global.nr; ++i)
269 		shared_msr_update(i, shared_msrs_global.msrs[i]);
270 }
271 
272 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
273 {
274 	unsigned int cpu = smp_processor_id();
275 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
276 	int err;
277 
278 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
279 		return 0;
280 	smsr->values[slot].curr = value;
281 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
282 	if (err)
283 		return 1;
284 
285 	if (!smsr->registered) {
286 		smsr->urn.on_user_return = kvm_on_user_return;
287 		user_return_notifier_register(&smsr->urn);
288 		smsr->registered = true;
289 	}
290 	return 0;
291 }
292 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
293 
294 static void drop_user_return_notifiers(void)
295 {
296 	unsigned int cpu = smp_processor_id();
297 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
298 
299 	if (smsr->registered)
300 		kvm_on_user_return(&smsr->urn);
301 }
302 
303 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
304 {
305 	return vcpu->arch.apic_base;
306 }
307 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
308 
309 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
310 {
311 	u64 old_state = vcpu->arch.apic_base &
312 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
313 	u64 new_state = msr_info->data &
314 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
316 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
317 
318 	if (!msr_info->host_initiated &&
319 	    ((msr_info->data & reserved_bits) != 0 ||
320 	     new_state == X2APIC_ENABLE ||
321 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
322 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
323 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
324 	      old_state == 0)))
325 		return 1;
326 
327 	kvm_lapic_set_base(vcpu, msr_info->data);
328 	return 0;
329 }
330 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
331 
332 asmlinkage __visible void kvm_spurious_fault(void)
333 {
334 	/* Fault while not rebooting.  We want the trace. */
335 	BUG();
336 }
337 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
338 
339 #define EXCPT_BENIGN		0
340 #define EXCPT_CONTRIBUTORY	1
341 #define EXCPT_PF		2
342 
343 static int exception_class(int vector)
344 {
345 	switch (vector) {
346 	case PF_VECTOR:
347 		return EXCPT_PF;
348 	case DE_VECTOR:
349 	case TS_VECTOR:
350 	case NP_VECTOR:
351 	case SS_VECTOR:
352 	case GP_VECTOR:
353 		return EXCPT_CONTRIBUTORY;
354 	default:
355 		break;
356 	}
357 	return EXCPT_BENIGN;
358 }
359 
360 #define EXCPT_FAULT		0
361 #define EXCPT_TRAP		1
362 #define EXCPT_ABORT		2
363 #define EXCPT_INTERRUPT		3
364 
365 static int exception_type(int vector)
366 {
367 	unsigned int mask;
368 
369 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
370 		return EXCPT_INTERRUPT;
371 
372 	mask = 1 << vector;
373 
374 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
375 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
376 		return EXCPT_TRAP;
377 
378 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
379 		return EXCPT_ABORT;
380 
381 	/* Reserved exceptions will result in fault */
382 	return EXCPT_FAULT;
383 }
384 
385 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
386 		unsigned nr, bool has_error, u32 error_code,
387 		bool reinject)
388 {
389 	u32 prev_nr;
390 	int class1, class2;
391 
392 	kvm_make_request(KVM_REQ_EVENT, vcpu);
393 
394 	if (!vcpu->arch.exception.pending) {
395 	queue:
396 		if (has_error && !is_protmode(vcpu))
397 			has_error = false;
398 		vcpu->arch.exception.pending = true;
399 		vcpu->arch.exception.has_error_code = has_error;
400 		vcpu->arch.exception.nr = nr;
401 		vcpu->arch.exception.error_code = error_code;
402 		vcpu->arch.exception.reinject = reinject;
403 		return;
404 	}
405 
406 	/* to check exception */
407 	prev_nr = vcpu->arch.exception.nr;
408 	if (prev_nr == DF_VECTOR) {
409 		/* triple fault -> shutdown */
410 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
411 		return;
412 	}
413 	class1 = exception_class(prev_nr);
414 	class2 = exception_class(nr);
415 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
416 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
417 		/* generate double fault per SDM Table 5-5 */
418 		vcpu->arch.exception.pending = true;
419 		vcpu->arch.exception.has_error_code = true;
420 		vcpu->arch.exception.nr = DF_VECTOR;
421 		vcpu->arch.exception.error_code = 0;
422 	} else
423 		/* replace previous exception with a new one in a hope
424 		   that instruction re-execution will regenerate lost
425 		   exception */
426 		goto queue;
427 }
428 
429 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
430 {
431 	kvm_multiple_exception(vcpu, nr, false, 0, false);
432 }
433 EXPORT_SYMBOL_GPL(kvm_queue_exception);
434 
435 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
436 {
437 	kvm_multiple_exception(vcpu, nr, false, 0, true);
438 }
439 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
440 
441 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
442 {
443 	if (err)
444 		kvm_inject_gp(vcpu, 0);
445 	else
446 		return kvm_skip_emulated_instruction(vcpu);
447 
448 	return 1;
449 }
450 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
451 
452 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
453 {
454 	++vcpu->stat.pf_guest;
455 	vcpu->arch.cr2 = fault->address;
456 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
457 }
458 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
459 
460 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
461 {
462 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
463 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
464 	else
465 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
466 
467 	return fault->nested_page_fault;
468 }
469 
470 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
471 {
472 	atomic_inc(&vcpu->arch.nmi_queued);
473 	kvm_make_request(KVM_REQ_NMI, vcpu);
474 }
475 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
476 
477 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
478 {
479 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
480 }
481 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
482 
483 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
484 {
485 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
486 }
487 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
488 
489 /*
490  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
491  * a #GP and return false.
492  */
493 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
494 {
495 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
496 		return true;
497 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
498 	return false;
499 }
500 EXPORT_SYMBOL_GPL(kvm_require_cpl);
501 
502 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
503 {
504 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
505 		return true;
506 
507 	kvm_queue_exception(vcpu, UD_VECTOR);
508 	return false;
509 }
510 EXPORT_SYMBOL_GPL(kvm_require_dr);
511 
512 /*
513  * This function will be used to read from the physical memory of the currently
514  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
515  * can read from guest physical or from the guest's guest physical memory.
516  */
517 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
518 			    gfn_t ngfn, void *data, int offset, int len,
519 			    u32 access)
520 {
521 	struct x86_exception exception;
522 	gfn_t real_gfn;
523 	gpa_t ngpa;
524 
525 	ngpa     = gfn_to_gpa(ngfn);
526 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
527 	if (real_gfn == UNMAPPED_GVA)
528 		return -EFAULT;
529 
530 	real_gfn = gpa_to_gfn(real_gfn);
531 
532 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
533 }
534 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
535 
536 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
537 			       void *data, int offset, int len, u32 access)
538 {
539 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
540 				       data, offset, len, access);
541 }
542 
543 /*
544  * Load the pae pdptrs.  Return true is they are all valid.
545  */
546 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
547 {
548 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
549 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
550 	int i;
551 	int ret;
552 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
553 
554 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
555 				      offset * sizeof(u64), sizeof(pdpte),
556 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
557 	if (ret < 0) {
558 		ret = 0;
559 		goto out;
560 	}
561 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
562 		if ((pdpte[i] & PT_PRESENT_MASK) &&
563 		    (pdpte[i] &
564 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
565 			ret = 0;
566 			goto out;
567 		}
568 	}
569 	ret = 1;
570 
571 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
572 	__set_bit(VCPU_EXREG_PDPTR,
573 		  (unsigned long *)&vcpu->arch.regs_avail);
574 	__set_bit(VCPU_EXREG_PDPTR,
575 		  (unsigned long *)&vcpu->arch.regs_dirty);
576 out:
577 
578 	return ret;
579 }
580 EXPORT_SYMBOL_GPL(load_pdptrs);
581 
582 bool pdptrs_changed(struct kvm_vcpu *vcpu)
583 {
584 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
585 	bool changed = true;
586 	int offset;
587 	gfn_t gfn;
588 	int r;
589 
590 	if (is_long_mode(vcpu) || !is_pae(vcpu))
591 		return false;
592 
593 	if (!test_bit(VCPU_EXREG_PDPTR,
594 		      (unsigned long *)&vcpu->arch.regs_avail))
595 		return true;
596 
597 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
598 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
599 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
600 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
601 	if (r < 0)
602 		goto out;
603 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
604 out:
605 
606 	return changed;
607 }
608 EXPORT_SYMBOL_GPL(pdptrs_changed);
609 
610 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
611 {
612 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
613 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
614 
615 	cr0 |= X86_CR0_ET;
616 
617 #ifdef CONFIG_X86_64
618 	if (cr0 & 0xffffffff00000000UL)
619 		return 1;
620 #endif
621 
622 	cr0 &= ~CR0_RESERVED_BITS;
623 
624 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
625 		return 1;
626 
627 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
628 		return 1;
629 
630 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
631 #ifdef CONFIG_X86_64
632 		if ((vcpu->arch.efer & EFER_LME)) {
633 			int cs_db, cs_l;
634 
635 			if (!is_pae(vcpu))
636 				return 1;
637 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
638 			if (cs_l)
639 				return 1;
640 		} else
641 #endif
642 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
643 						 kvm_read_cr3(vcpu)))
644 			return 1;
645 	}
646 
647 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
648 		return 1;
649 
650 	kvm_x86_ops->set_cr0(vcpu, cr0);
651 
652 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
653 		kvm_clear_async_pf_completion_queue(vcpu);
654 		kvm_async_pf_hash_reset(vcpu);
655 	}
656 
657 	if ((cr0 ^ old_cr0) & update_bits)
658 		kvm_mmu_reset_context(vcpu);
659 
660 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
661 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
662 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
663 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
664 
665 	return 0;
666 }
667 EXPORT_SYMBOL_GPL(kvm_set_cr0);
668 
669 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
670 {
671 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
672 }
673 EXPORT_SYMBOL_GPL(kvm_lmsw);
674 
675 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
676 {
677 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
678 			!vcpu->guest_xcr0_loaded) {
679 		/* kvm_set_xcr() also depends on this */
680 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
681 		vcpu->guest_xcr0_loaded = 1;
682 	}
683 }
684 
685 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
686 {
687 	if (vcpu->guest_xcr0_loaded) {
688 		if (vcpu->arch.xcr0 != host_xcr0)
689 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
690 		vcpu->guest_xcr0_loaded = 0;
691 	}
692 }
693 
694 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
695 {
696 	u64 xcr0 = xcr;
697 	u64 old_xcr0 = vcpu->arch.xcr0;
698 	u64 valid_bits;
699 
700 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
701 	if (index != XCR_XFEATURE_ENABLED_MASK)
702 		return 1;
703 	if (!(xcr0 & XFEATURE_MASK_FP))
704 		return 1;
705 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
706 		return 1;
707 
708 	/*
709 	 * Do not allow the guest to set bits that we do not support
710 	 * saving.  However, xcr0 bit 0 is always set, even if the
711 	 * emulated CPU does not support XSAVE (see fx_init).
712 	 */
713 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
714 	if (xcr0 & ~valid_bits)
715 		return 1;
716 
717 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
718 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
719 		return 1;
720 
721 	if (xcr0 & XFEATURE_MASK_AVX512) {
722 		if (!(xcr0 & XFEATURE_MASK_YMM))
723 			return 1;
724 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
725 			return 1;
726 	}
727 	vcpu->arch.xcr0 = xcr0;
728 
729 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
730 		kvm_update_cpuid(vcpu);
731 	return 0;
732 }
733 
734 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
735 {
736 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
737 	    __kvm_set_xcr(vcpu, index, xcr)) {
738 		kvm_inject_gp(vcpu, 0);
739 		return 1;
740 	}
741 	return 0;
742 }
743 EXPORT_SYMBOL_GPL(kvm_set_xcr);
744 
745 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
746 {
747 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
748 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
749 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
750 
751 	if (cr4 & CR4_RESERVED_BITS)
752 		return 1;
753 
754 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
755 		return 1;
756 
757 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
758 		return 1;
759 
760 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
761 		return 1;
762 
763 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
764 		return 1;
765 
766 	if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
767 		return 1;
768 
769 	if (is_long_mode(vcpu)) {
770 		if (!(cr4 & X86_CR4_PAE))
771 			return 1;
772 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
773 		   && ((cr4 ^ old_cr4) & pdptr_bits)
774 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
775 				   kvm_read_cr3(vcpu)))
776 		return 1;
777 
778 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
779 		if (!guest_cpuid_has_pcid(vcpu))
780 			return 1;
781 
782 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
783 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
784 			return 1;
785 	}
786 
787 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
788 		return 1;
789 
790 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
791 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
792 		kvm_mmu_reset_context(vcpu);
793 
794 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
795 		kvm_update_cpuid(vcpu);
796 
797 	return 0;
798 }
799 EXPORT_SYMBOL_GPL(kvm_set_cr4);
800 
801 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
802 {
803 #ifdef CONFIG_X86_64
804 	cr3 &= ~CR3_PCID_INVD;
805 #endif
806 
807 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
808 		kvm_mmu_sync_roots(vcpu);
809 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
810 		return 0;
811 	}
812 
813 	if (is_long_mode(vcpu)) {
814 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
815 			return 1;
816 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
817 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
818 		return 1;
819 
820 	vcpu->arch.cr3 = cr3;
821 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
822 	kvm_mmu_new_cr3(vcpu);
823 	return 0;
824 }
825 EXPORT_SYMBOL_GPL(kvm_set_cr3);
826 
827 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
828 {
829 	if (cr8 & CR8_RESERVED_BITS)
830 		return 1;
831 	if (lapic_in_kernel(vcpu))
832 		kvm_lapic_set_tpr(vcpu, cr8);
833 	else
834 		vcpu->arch.cr8 = cr8;
835 	return 0;
836 }
837 EXPORT_SYMBOL_GPL(kvm_set_cr8);
838 
839 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
840 {
841 	if (lapic_in_kernel(vcpu))
842 		return kvm_lapic_get_cr8(vcpu);
843 	else
844 		return vcpu->arch.cr8;
845 }
846 EXPORT_SYMBOL_GPL(kvm_get_cr8);
847 
848 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
849 {
850 	int i;
851 
852 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
853 		for (i = 0; i < KVM_NR_DB_REGS; i++)
854 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
855 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
856 	}
857 }
858 
859 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
860 {
861 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
862 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
863 }
864 
865 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
866 {
867 	unsigned long dr7;
868 
869 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
870 		dr7 = vcpu->arch.guest_debug_dr7;
871 	else
872 		dr7 = vcpu->arch.dr7;
873 	kvm_x86_ops->set_dr7(vcpu, dr7);
874 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
875 	if (dr7 & DR7_BP_EN_MASK)
876 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
877 }
878 
879 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
880 {
881 	u64 fixed = DR6_FIXED_1;
882 
883 	if (!guest_cpuid_has_rtm(vcpu))
884 		fixed |= DR6_RTM;
885 	return fixed;
886 }
887 
888 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
889 {
890 	switch (dr) {
891 	case 0 ... 3:
892 		vcpu->arch.db[dr] = val;
893 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
894 			vcpu->arch.eff_db[dr] = val;
895 		break;
896 	case 4:
897 		/* fall through */
898 	case 6:
899 		if (val & 0xffffffff00000000ULL)
900 			return -1; /* #GP */
901 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
902 		kvm_update_dr6(vcpu);
903 		break;
904 	case 5:
905 		/* fall through */
906 	default: /* 7 */
907 		if (val & 0xffffffff00000000ULL)
908 			return -1; /* #GP */
909 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
910 		kvm_update_dr7(vcpu);
911 		break;
912 	}
913 
914 	return 0;
915 }
916 
917 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
918 {
919 	if (__kvm_set_dr(vcpu, dr, val)) {
920 		kvm_inject_gp(vcpu, 0);
921 		return 1;
922 	}
923 	return 0;
924 }
925 EXPORT_SYMBOL_GPL(kvm_set_dr);
926 
927 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
928 {
929 	switch (dr) {
930 	case 0 ... 3:
931 		*val = vcpu->arch.db[dr];
932 		break;
933 	case 4:
934 		/* fall through */
935 	case 6:
936 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
937 			*val = vcpu->arch.dr6;
938 		else
939 			*val = kvm_x86_ops->get_dr6(vcpu);
940 		break;
941 	case 5:
942 		/* fall through */
943 	default: /* 7 */
944 		*val = vcpu->arch.dr7;
945 		break;
946 	}
947 	return 0;
948 }
949 EXPORT_SYMBOL_GPL(kvm_get_dr);
950 
951 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
952 {
953 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
954 	u64 data;
955 	int err;
956 
957 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
958 	if (err)
959 		return err;
960 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
961 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
962 	return err;
963 }
964 EXPORT_SYMBOL_GPL(kvm_rdpmc);
965 
966 /*
967  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
968  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
969  *
970  * This list is modified at module load time to reflect the
971  * capabilities of the host cpu. This capabilities test skips MSRs that are
972  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
973  * may depend on host virtualization features rather than host cpu features.
974  */
975 
976 static u32 msrs_to_save[] = {
977 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
978 	MSR_STAR,
979 #ifdef CONFIG_X86_64
980 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
981 #endif
982 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
983 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
984 };
985 
986 static unsigned num_msrs_to_save;
987 
988 static u32 emulated_msrs[] = {
989 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
990 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
991 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
992 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
993 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
994 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
995 	HV_X64_MSR_RESET,
996 	HV_X64_MSR_VP_INDEX,
997 	HV_X64_MSR_VP_RUNTIME,
998 	HV_X64_MSR_SCONTROL,
999 	HV_X64_MSR_STIMER0_CONFIG,
1000 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1001 	MSR_KVM_PV_EOI_EN,
1002 
1003 	MSR_IA32_TSC_ADJUST,
1004 	MSR_IA32_TSCDEADLINE,
1005 	MSR_IA32_MISC_ENABLE,
1006 	MSR_IA32_MCG_STATUS,
1007 	MSR_IA32_MCG_CTL,
1008 	MSR_IA32_MCG_EXT_CTL,
1009 	MSR_IA32_SMBASE,
1010 	MSR_PLATFORM_INFO,
1011 	MSR_MISC_FEATURES_ENABLES,
1012 };
1013 
1014 static unsigned num_emulated_msrs;
1015 
1016 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1017 {
1018 	if (efer & efer_reserved_bits)
1019 		return false;
1020 
1021 	if (efer & EFER_FFXSR) {
1022 		struct kvm_cpuid_entry2 *feat;
1023 
1024 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1025 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1026 			return false;
1027 	}
1028 
1029 	if (efer & EFER_SVME) {
1030 		struct kvm_cpuid_entry2 *feat;
1031 
1032 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1033 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1034 			return false;
1035 	}
1036 
1037 	return true;
1038 }
1039 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1040 
1041 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1042 {
1043 	u64 old_efer = vcpu->arch.efer;
1044 
1045 	if (!kvm_valid_efer(vcpu, efer))
1046 		return 1;
1047 
1048 	if (is_paging(vcpu)
1049 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1050 		return 1;
1051 
1052 	efer &= ~EFER_LMA;
1053 	efer |= vcpu->arch.efer & EFER_LMA;
1054 
1055 	kvm_x86_ops->set_efer(vcpu, efer);
1056 
1057 	/* Update reserved bits */
1058 	if ((efer ^ old_efer) & EFER_NX)
1059 		kvm_mmu_reset_context(vcpu);
1060 
1061 	return 0;
1062 }
1063 
1064 void kvm_enable_efer_bits(u64 mask)
1065 {
1066        efer_reserved_bits &= ~mask;
1067 }
1068 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1069 
1070 /*
1071  * Writes msr value into into the appropriate "register".
1072  * Returns 0 on success, non-0 otherwise.
1073  * Assumes vcpu_load() was already called.
1074  */
1075 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1076 {
1077 	switch (msr->index) {
1078 	case MSR_FS_BASE:
1079 	case MSR_GS_BASE:
1080 	case MSR_KERNEL_GS_BASE:
1081 	case MSR_CSTAR:
1082 	case MSR_LSTAR:
1083 		if (is_noncanonical_address(msr->data))
1084 			return 1;
1085 		break;
1086 	case MSR_IA32_SYSENTER_EIP:
1087 	case MSR_IA32_SYSENTER_ESP:
1088 		/*
1089 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1090 		 * non-canonical address is written on Intel but not on
1091 		 * AMD (which ignores the top 32-bits, because it does
1092 		 * not implement 64-bit SYSENTER).
1093 		 *
1094 		 * 64-bit code should hence be able to write a non-canonical
1095 		 * value on AMD.  Making the address canonical ensures that
1096 		 * vmentry does not fail on Intel after writing a non-canonical
1097 		 * value, and that something deterministic happens if the guest
1098 		 * invokes 64-bit SYSENTER.
1099 		 */
1100 		msr->data = get_canonical(msr->data);
1101 	}
1102 	return kvm_x86_ops->set_msr(vcpu, msr);
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_set_msr);
1105 
1106 /*
1107  * Adapt set_msr() to msr_io()'s calling convention
1108  */
1109 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1110 {
1111 	struct msr_data msr;
1112 	int r;
1113 
1114 	msr.index = index;
1115 	msr.host_initiated = true;
1116 	r = kvm_get_msr(vcpu, &msr);
1117 	if (r)
1118 		return r;
1119 
1120 	*data = msr.data;
1121 	return 0;
1122 }
1123 
1124 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1125 {
1126 	struct msr_data msr;
1127 
1128 	msr.data = *data;
1129 	msr.index = index;
1130 	msr.host_initiated = true;
1131 	return kvm_set_msr(vcpu, &msr);
1132 }
1133 
1134 #ifdef CONFIG_X86_64
1135 struct pvclock_gtod_data {
1136 	seqcount_t	seq;
1137 
1138 	struct { /* extract of a clocksource struct */
1139 		int vclock_mode;
1140 		u64	cycle_last;
1141 		u64	mask;
1142 		u32	mult;
1143 		u32	shift;
1144 	} clock;
1145 
1146 	u64		boot_ns;
1147 	u64		nsec_base;
1148 	u64		wall_time_sec;
1149 };
1150 
1151 static struct pvclock_gtod_data pvclock_gtod_data;
1152 
1153 static void update_pvclock_gtod(struct timekeeper *tk)
1154 {
1155 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1156 	u64 boot_ns;
1157 
1158 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1159 
1160 	write_seqcount_begin(&vdata->seq);
1161 
1162 	/* copy pvclock gtod data */
1163 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1164 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1165 	vdata->clock.mask		= tk->tkr_mono.mask;
1166 	vdata->clock.mult		= tk->tkr_mono.mult;
1167 	vdata->clock.shift		= tk->tkr_mono.shift;
1168 
1169 	vdata->boot_ns			= boot_ns;
1170 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1171 
1172 	vdata->wall_time_sec            = tk->xtime_sec;
1173 
1174 	write_seqcount_end(&vdata->seq);
1175 }
1176 #endif
1177 
1178 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1179 {
1180 	/*
1181 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1182 	 * vcpu_enter_guest.  This function is only called from
1183 	 * the physical CPU that is running vcpu.
1184 	 */
1185 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1186 }
1187 
1188 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1189 {
1190 	int version;
1191 	int r;
1192 	struct pvclock_wall_clock wc;
1193 	struct timespec64 boot;
1194 
1195 	if (!wall_clock)
1196 		return;
1197 
1198 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1199 	if (r)
1200 		return;
1201 
1202 	if (version & 1)
1203 		++version;  /* first time write, random junk */
1204 
1205 	++version;
1206 
1207 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1208 		return;
1209 
1210 	/*
1211 	 * The guest calculates current wall clock time by adding
1212 	 * system time (updated by kvm_guest_time_update below) to the
1213 	 * wall clock specified here.  guest system time equals host
1214 	 * system time for us, thus we must fill in host boot time here.
1215 	 */
1216 	getboottime64(&boot);
1217 
1218 	if (kvm->arch.kvmclock_offset) {
1219 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1220 		boot = timespec64_sub(boot, ts);
1221 	}
1222 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1223 	wc.nsec = boot.tv_nsec;
1224 	wc.version = version;
1225 
1226 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1227 
1228 	version++;
1229 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1230 }
1231 
1232 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1233 {
1234 	do_shl32_div32(dividend, divisor);
1235 	return dividend;
1236 }
1237 
1238 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1239 			       s8 *pshift, u32 *pmultiplier)
1240 {
1241 	uint64_t scaled64;
1242 	int32_t  shift = 0;
1243 	uint64_t tps64;
1244 	uint32_t tps32;
1245 
1246 	tps64 = base_hz;
1247 	scaled64 = scaled_hz;
1248 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1249 		tps64 >>= 1;
1250 		shift--;
1251 	}
1252 
1253 	tps32 = (uint32_t)tps64;
1254 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1255 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1256 			scaled64 >>= 1;
1257 		else
1258 			tps32 <<= 1;
1259 		shift++;
1260 	}
1261 
1262 	*pshift = shift;
1263 	*pmultiplier = div_frac(scaled64, tps32);
1264 
1265 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1266 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1267 }
1268 
1269 #ifdef CONFIG_X86_64
1270 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1271 #endif
1272 
1273 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1274 static unsigned long max_tsc_khz;
1275 
1276 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1277 {
1278 	u64 v = (u64)khz * (1000000 + ppm);
1279 	do_div(v, 1000000);
1280 	return v;
1281 }
1282 
1283 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1284 {
1285 	u64 ratio;
1286 
1287 	/* Guest TSC same frequency as host TSC? */
1288 	if (!scale) {
1289 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1290 		return 0;
1291 	}
1292 
1293 	/* TSC scaling supported? */
1294 	if (!kvm_has_tsc_control) {
1295 		if (user_tsc_khz > tsc_khz) {
1296 			vcpu->arch.tsc_catchup = 1;
1297 			vcpu->arch.tsc_always_catchup = 1;
1298 			return 0;
1299 		} else {
1300 			WARN(1, "user requested TSC rate below hardware speed\n");
1301 			return -1;
1302 		}
1303 	}
1304 
1305 	/* TSC scaling required  - calculate ratio */
1306 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1307 				user_tsc_khz, tsc_khz);
1308 
1309 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1310 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1311 			  user_tsc_khz);
1312 		return -1;
1313 	}
1314 
1315 	vcpu->arch.tsc_scaling_ratio = ratio;
1316 	return 0;
1317 }
1318 
1319 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1320 {
1321 	u32 thresh_lo, thresh_hi;
1322 	int use_scaling = 0;
1323 
1324 	/* tsc_khz can be zero if TSC calibration fails */
1325 	if (user_tsc_khz == 0) {
1326 		/* set tsc_scaling_ratio to a safe value */
1327 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1328 		return -1;
1329 	}
1330 
1331 	/* Compute a scale to convert nanoseconds in TSC cycles */
1332 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1333 			   &vcpu->arch.virtual_tsc_shift,
1334 			   &vcpu->arch.virtual_tsc_mult);
1335 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1336 
1337 	/*
1338 	 * Compute the variation in TSC rate which is acceptable
1339 	 * within the range of tolerance and decide if the
1340 	 * rate being applied is within that bounds of the hardware
1341 	 * rate.  If so, no scaling or compensation need be done.
1342 	 */
1343 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1344 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1345 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1346 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1347 		use_scaling = 1;
1348 	}
1349 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1350 }
1351 
1352 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1353 {
1354 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1355 				      vcpu->arch.virtual_tsc_mult,
1356 				      vcpu->arch.virtual_tsc_shift);
1357 	tsc += vcpu->arch.this_tsc_write;
1358 	return tsc;
1359 }
1360 
1361 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1362 {
1363 #ifdef CONFIG_X86_64
1364 	bool vcpus_matched;
1365 	struct kvm_arch *ka = &vcpu->kvm->arch;
1366 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1367 
1368 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1369 			 atomic_read(&vcpu->kvm->online_vcpus));
1370 
1371 	/*
1372 	 * Once the masterclock is enabled, always perform request in
1373 	 * order to update it.
1374 	 *
1375 	 * In order to enable masterclock, the host clocksource must be TSC
1376 	 * and the vcpus need to have matched TSCs.  When that happens,
1377 	 * perform request to enable masterclock.
1378 	 */
1379 	if (ka->use_master_clock ||
1380 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1381 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1382 
1383 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1384 			    atomic_read(&vcpu->kvm->online_vcpus),
1385 		            ka->use_master_clock, gtod->clock.vclock_mode);
1386 #endif
1387 }
1388 
1389 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1390 {
1391 	u64 curr_offset = vcpu->arch.tsc_offset;
1392 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1393 }
1394 
1395 /*
1396  * Multiply tsc by a fixed point number represented by ratio.
1397  *
1398  * The most significant 64-N bits (mult) of ratio represent the
1399  * integral part of the fixed point number; the remaining N bits
1400  * (frac) represent the fractional part, ie. ratio represents a fixed
1401  * point number (mult + frac * 2^(-N)).
1402  *
1403  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1404  */
1405 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1406 {
1407 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1408 }
1409 
1410 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1411 {
1412 	u64 _tsc = tsc;
1413 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1414 
1415 	if (ratio != kvm_default_tsc_scaling_ratio)
1416 		_tsc = __scale_tsc(ratio, tsc);
1417 
1418 	return _tsc;
1419 }
1420 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1421 
1422 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1423 {
1424 	u64 tsc;
1425 
1426 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1427 
1428 	return target_tsc - tsc;
1429 }
1430 
1431 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1432 {
1433 	return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1436 
1437 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1438 {
1439 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1440 	vcpu->arch.tsc_offset = offset;
1441 }
1442 
1443 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1444 {
1445 	struct kvm *kvm = vcpu->kvm;
1446 	u64 offset, ns, elapsed;
1447 	unsigned long flags;
1448 	bool matched;
1449 	bool already_matched;
1450 	u64 data = msr->data;
1451 	bool synchronizing = false;
1452 
1453 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1454 	offset = kvm_compute_tsc_offset(vcpu, data);
1455 	ns = ktime_get_boot_ns();
1456 	elapsed = ns - kvm->arch.last_tsc_nsec;
1457 
1458 	if (vcpu->arch.virtual_tsc_khz) {
1459 		if (data == 0 && msr->host_initiated) {
1460 			/*
1461 			 * detection of vcpu initialization -- need to sync
1462 			 * with other vCPUs. This particularly helps to keep
1463 			 * kvm_clock stable after CPU hotplug
1464 			 */
1465 			synchronizing = true;
1466 		} else {
1467 			u64 tsc_exp = kvm->arch.last_tsc_write +
1468 						nsec_to_cycles(vcpu, elapsed);
1469 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1470 			/*
1471 			 * Special case: TSC write with a small delta (1 second)
1472 			 * of virtual cycle time against real time is
1473 			 * interpreted as an attempt to synchronize the CPU.
1474 			 */
1475 			synchronizing = data < tsc_exp + tsc_hz &&
1476 					data + tsc_hz > tsc_exp;
1477 		}
1478 	}
1479 
1480 	/*
1481 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1482 	 * TSC, we add elapsed time in this computation.  We could let the
1483 	 * compensation code attempt to catch up if we fall behind, but
1484 	 * it's better to try to match offsets from the beginning.
1485          */
1486 	if (synchronizing &&
1487 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1488 		if (!check_tsc_unstable()) {
1489 			offset = kvm->arch.cur_tsc_offset;
1490 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1491 		} else {
1492 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1493 			data += delta;
1494 			offset = kvm_compute_tsc_offset(vcpu, data);
1495 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1496 		}
1497 		matched = true;
1498 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1499 	} else {
1500 		/*
1501 		 * We split periods of matched TSC writes into generations.
1502 		 * For each generation, we track the original measured
1503 		 * nanosecond time, offset, and write, so if TSCs are in
1504 		 * sync, we can match exact offset, and if not, we can match
1505 		 * exact software computation in compute_guest_tsc()
1506 		 *
1507 		 * These values are tracked in kvm->arch.cur_xxx variables.
1508 		 */
1509 		kvm->arch.cur_tsc_generation++;
1510 		kvm->arch.cur_tsc_nsec = ns;
1511 		kvm->arch.cur_tsc_write = data;
1512 		kvm->arch.cur_tsc_offset = offset;
1513 		matched = false;
1514 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1515 			 kvm->arch.cur_tsc_generation, data);
1516 	}
1517 
1518 	/*
1519 	 * We also track th most recent recorded KHZ, write and time to
1520 	 * allow the matching interval to be extended at each write.
1521 	 */
1522 	kvm->arch.last_tsc_nsec = ns;
1523 	kvm->arch.last_tsc_write = data;
1524 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1525 
1526 	vcpu->arch.last_guest_tsc = data;
1527 
1528 	/* Keep track of which generation this VCPU has synchronized to */
1529 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1530 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1531 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1532 
1533 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1534 		update_ia32_tsc_adjust_msr(vcpu, offset);
1535 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1536 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1537 
1538 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1539 	if (!matched) {
1540 		kvm->arch.nr_vcpus_matched_tsc = 0;
1541 	} else if (!already_matched) {
1542 		kvm->arch.nr_vcpus_matched_tsc++;
1543 	}
1544 
1545 	kvm_track_tsc_matching(vcpu);
1546 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1547 }
1548 
1549 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1550 
1551 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1552 					   s64 adjustment)
1553 {
1554 	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1555 }
1556 
1557 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1558 {
1559 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1560 		WARN_ON(adjustment < 0);
1561 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1562 	adjust_tsc_offset_guest(vcpu, adjustment);
1563 }
1564 
1565 #ifdef CONFIG_X86_64
1566 
1567 static u64 read_tsc(void)
1568 {
1569 	u64 ret = (u64)rdtsc_ordered();
1570 	u64 last = pvclock_gtod_data.clock.cycle_last;
1571 
1572 	if (likely(ret >= last))
1573 		return ret;
1574 
1575 	/*
1576 	 * GCC likes to generate cmov here, but this branch is extremely
1577 	 * predictable (it's just a function of time and the likely is
1578 	 * very likely) and there's a data dependence, so force GCC
1579 	 * to generate a branch instead.  I don't barrier() because
1580 	 * we don't actually need a barrier, and if this function
1581 	 * ever gets inlined it will generate worse code.
1582 	 */
1583 	asm volatile ("");
1584 	return last;
1585 }
1586 
1587 static inline u64 vgettsc(u64 *cycle_now)
1588 {
1589 	long v;
1590 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1591 
1592 	*cycle_now = read_tsc();
1593 
1594 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1595 	return v * gtod->clock.mult;
1596 }
1597 
1598 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1599 {
1600 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1601 	unsigned long seq;
1602 	int mode;
1603 	u64 ns;
1604 
1605 	do {
1606 		seq = read_seqcount_begin(&gtod->seq);
1607 		mode = gtod->clock.vclock_mode;
1608 		ns = gtod->nsec_base;
1609 		ns += vgettsc(cycle_now);
1610 		ns >>= gtod->clock.shift;
1611 		ns += gtod->boot_ns;
1612 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1613 	*t = ns;
1614 
1615 	return mode;
1616 }
1617 
1618 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1619 {
1620 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1621 	unsigned long seq;
1622 	int mode;
1623 	u64 ns;
1624 
1625 	do {
1626 		seq = read_seqcount_begin(&gtod->seq);
1627 		mode = gtod->clock.vclock_mode;
1628 		ts->tv_sec = gtod->wall_time_sec;
1629 		ns = gtod->nsec_base;
1630 		ns += vgettsc(cycle_now);
1631 		ns >>= gtod->clock.shift;
1632 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1633 
1634 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1635 	ts->tv_nsec = ns;
1636 
1637 	return mode;
1638 }
1639 
1640 /* returns true if host is using tsc clocksource */
1641 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1642 {
1643 	/* checked again under seqlock below */
1644 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1645 		return false;
1646 
1647 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1648 }
1649 
1650 /* returns true if host is using tsc clocksource */
1651 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1652 					   u64 *cycle_now)
1653 {
1654 	/* checked again under seqlock below */
1655 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1656 		return false;
1657 
1658 	return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1659 }
1660 #endif
1661 
1662 /*
1663  *
1664  * Assuming a stable TSC across physical CPUS, and a stable TSC
1665  * across virtual CPUs, the following condition is possible.
1666  * Each numbered line represents an event visible to both
1667  * CPUs at the next numbered event.
1668  *
1669  * "timespecX" represents host monotonic time. "tscX" represents
1670  * RDTSC value.
1671  *
1672  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1673  *
1674  * 1.  read timespec0,tsc0
1675  * 2.					| timespec1 = timespec0 + N
1676  * 					| tsc1 = tsc0 + M
1677  * 3. transition to guest		| transition to guest
1678  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1679  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1680  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1681  *
1682  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1683  *
1684  * 	- ret0 < ret1
1685  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1686  *		...
1687  *	- 0 < N - M => M < N
1688  *
1689  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1690  * always the case (the difference between two distinct xtime instances
1691  * might be smaller then the difference between corresponding TSC reads,
1692  * when updating guest vcpus pvclock areas).
1693  *
1694  * To avoid that problem, do not allow visibility of distinct
1695  * system_timestamp/tsc_timestamp values simultaneously: use a master
1696  * copy of host monotonic time values. Update that master copy
1697  * in lockstep.
1698  *
1699  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1700  *
1701  */
1702 
1703 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1704 {
1705 #ifdef CONFIG_X86_64
1706 	struct kvm_arch *ka = &kvm->arch;
1707 	int vclock_mode;
1708 	bool host_tsc_clocksource, vcpus_matched;
1709 
1710 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1711 			atomic_read(&kvm->online_vcpus));
1712 
1713 	/*
1714 	 * If the host uses TSC clock, then passthrough TSC as stable
1715 	 * to the guest.
1716 	 */
1717 	host_tsc_clocksource = kvm_get_time_and_clockread(
1718 					&ka->master_kernel_ns,
1719 					&ka->master_cycle_now);
1720 
1721 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1722 				&& !backwards_tsc_observed
1723 				&& !ka->boot_vcpu_runs_old_kvmclock;
1724 
1725 	if (ka->use_master_clock)
1726 		atomic_set(&kvm_guest_has_master_clock, 1);
1727 
1728 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1729 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1730 					vcpus_matched);
1731 #endif
1732 }
1733 
1734 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1735 {
1736 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1737 }
1738 
1739 static void kvm_gen_update_masterclock(struct kvm *kvm)
1740 {
1741 #ifdef CONFIG_X86_64
1742 	int i;
1743 	struct kvm_vcpu *vcpu;
1744 	struct kvm_arch *ka = &kvm->arch;
1745 
1746 	spin_lock(&ka->pvclock_gtod_sync_lock);
1747 	kvm_make_mclock_inprogress_request(kvm);
1748 	/* no guest entries from this point */
1749 	pvclock_update_vm_gtod_copy(kvm);
1750 
1751 	kvm_for_each_vcpu(i, vcpu, kvm)
1752 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1753 
1754 	/* guest entries allowed */
1755 	kvm_for_each_vcpu(i, vcpu, kvm)
1756 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1757 
1758 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1759 #endif
1760 }
1761 
1762 u64 get_kvmclock_ns(struct kvm *kvm)
1763 {
1764 	struct kvm_arch *ka = &kvm->arch;
1765 	struct pvclock_vcpu_time_info hv_clock;
1766 
1767 	spin_lock(&ka->pvclock_gtod_sync_lock);
1768 	if (!ka->use_master_clock) {
1769 		spin_unlock(&ka->pvclock_gtod_sync_lock);
1770 		return ktime_get_boot_ns() + ka->kvmclock_offset;
1771 	}
1772 
1773 	hv_clock.tsc_timestamp = ka->master_cycle_now;
1774 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1775 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1776 
1777 	kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1778 			   &hv_clock.tsc_shift,
1779 			   &hv_clock.tsc_to_system_mul);
1780 	return __pvclock_read_cycles(&hv_clock, rdtsc());
1781 }
1782 
1783 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1784 {
1785 	struct kvm_vcpu_arch *vcpu = &v->arch;
1786 	struct pvclock_vcpu_time_info guest_hv_clock;
1787 
1788 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1789 		&guest_hv_clock, sizeof(guest_hv_clock))))
1790 		return;
1791 
1792 	/* This VCPU is paused, but it's legal for a guest to read another
1793 	 * VCPU's kvmclock, so we really have to follow the specification where
1794 	 * it says that version is odd if data is being modified, and even after
1795 	 * it is consistent.
1796 	 *
1797 	 * Version field updates must be kept separate.  This is because
1798 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1799 	 * writes within a string instruction are weakly ordered.  So there
1800 	 * are three writes overall.
1801 	 *
1802 	 * As a small optimization, only write the version field in the first
1803 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1804 	 * version field is the first in the struct.
1805 	 */
1806 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1807 
1808 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1809 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1810 				&vcpu->hv_clock,
1811 				sizeof(vcpu->hv_clock.version));
1812 
1813 	smp_wmb();
1814 
1815 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1816 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1817 
1818 	if (vcpu->pvclock_set_guest_stopped_request) {
1819 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1820 		vcpu->pvclock_set_guest_stopped_request = false;
1821 	}
1822 
1823 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1824 
1825 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1826 				&vcpu->hv_clock,
1827 				sizeof(vcpu->hv_clock));
1828 
1829 	smp_wmb();
1830 
1831 	vcpu->hv_clock.version++;
1832 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1833 				&vcpu->hv_clock,
1834 				sizeof(vcpu->hv_clock.version));
1835 }
1836 
1837 static int kvm_guest_time_update(struct kvm_vcpu *v)
1838 {
1839 	unsigned long flags, tgt_tsc_khz;
1840 	struct kvm_vcpu_arch *vcpu = &v->arch;
1841 	struct kvm_arch *ka = &v->kvm->arch;
1842 	s64 kernel_ns;
1843 	u64 tsc_timestamp, host_tsc;
1844 	u8 pvclock_flags;
1845 	bool use_master_clock;
1846 
1847 	kernel_ns = 0;
1848 	host_tsc = 0;
1849 
1850 	/*
1851 	 * If the host uses TSC clock, then passthrough TSC as stable
1852 	 * to the guest.
1853 	 */
1854 	spin_lock(&ka->pvclock_gtod_sync_lock);
1855 	use_master_clock = ka->use_master_clock;
1856 	if (use_master_clock) {
1857 		host_tsc = ka->master_cycle_now;
1858 		kernel_ns = ka->master_kernel_ns;
1859 	}
1860 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1861 
1862 	/* Keep irq disabled to prevent changes to the clock */
1863 	local_irq_save(flags);
1864 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1865 	if (unlikely(tgt_tsc_khz == 0)) {
1866 		local_irq_restore(flags);
1867 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1868 		return 1;
1869 	}
1870 	if (!use_master_clock) {
1871 		host_tsc = rdtsc();
1872 		kernel_ns = ktime_get_boot_ns();
1873 	}
1874 
1875 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1876 
1877 	/*
1878 	 * We may have to catch up the TSC to match elapsed wall clock
1879 	 * time for two reasons, even if kvmclock is used.
1880 	 *   1) CPU could have been running below the maximum TSC rate
1881 	 *   2) Broken TSC compensation resets the base at each VCPU
1882 	 *      entry to avoid unknown leaps of TSC even when running
1883 	 *      again on the same CPU.  This may cause apparent elapsed
1884 	 *      time to disappear, and the guest to stand still or run
1885 	 *	very slowly.
1886 	 */
1887 	if (vcpu->tsc_catchup) {
1888 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1889 		if (tsc > tsc_timestamp) {
1890 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1891 			tsc_timestamp = tsc;
1892 		}
1893 	}
1894 
1895 	local_irq_restore(flags);
1896 
1897 	/* With all the info we got, fill in the values */
1898 
1899 	if (kvm_has_tsc_control)
1900 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1901 
1902 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1903 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1904 				   &vcpu->hv_clock.tsc_shift,
1905 				   &vcpu->hv_clock.tsc_to_system_mul);
1906 		vcpu->hw_tsc_khz = tgt_tsc_khz;
1907 	}
1908 
1909 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1910 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1911 	vcpu->last_guest_tsc = tsc_timestamp;
1912 
1913 	/* If the host uses TSC clocksource, then it is stable */
1914 	pvclock_flags = 0;
1915 	if (use_master_clock)
1916 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1917 
1918 	vcpu->hv_clock.flags = pvclock_flags;
1919 
1920 	if (vcpu->pv_time_enabled)
1921 		kvm_setup_pvclock_page(v);
1922 	if (v == kvm_get_vcpu(v->kvm, 0))
1923 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1924 	return 0;
1925 }
1926 
1927 /*
1928  * kvmclock updates which are isolated to a given vcpu, such as
1929  * vcpu->cpu migration, should not allow system_timestamp from
1930  * the rest of the vcpus to remain static. Otherwise ntp frequency
1931  * correction applies to one vcpu's system_timestamp but not
1932  * the others.
1933  *
1934  * So in those cases, request a kvmclock update for all vcpus.
1935  * We need to rate-limit these requests though, as they can
1936  * considerably slow guests that have a large number of vcpus.
1937  * The time for a remote vcpu to update its kvmclock is bound
1938  * by the delay we use to rate-limit the updates.
1939  */
1940 
1941 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1942 
1943 static void kvmclock_update_fn(struct work_struct *work)
1944 {
1945 	int i;
1946 	struct delayed_work *dwork = to_delayed_work(work);
1947 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1948 					   kvmclock_update_work);
1949 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1950 	struct kvm_vcpu *vcpu;
1951 
1952 	kvm_for_each_vcpu(i, vcpu, kvm) {
1953 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1954 		kvm_vcpu_kick(vcpu);
1955 	}
1956 }
1957 
1958 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1959 {
1960 	struct kvm *kvm = v->kvm;
1961 
1962 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1963 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1964 					KVMCLOCK_UPDATE_DELAY);
1965 }
1966 
1967 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1968 
1969 static void kvmclock_sync_fn(struct work_struct *work)
1970 {
1971 	struct delayed_work *dwork = to_delayed_work(work);
1972 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1973 					   kvmclock_sync_work);
1974 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1975 
1976 	if (!kvmclock_periodic_sync)
1977 		return;
1978 
1979 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1980 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1981 					KVMCLOCK_SYNC_PERIOD);
1982 }
1983 
1984 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1985 {
1986 	u64 mcg_cap = vcpu->arch.mcg_cap;
1987 	unsigned bank_num = mcg_cap & 0xff;
1988 
1989 	switch (msr) {
1990 	case MSR_IA32_MCG_STATUS:
1991 		vcpu->arch.mcg_status = data;
1992 		break;
1993 	case MSR_IA32_MCG_CTL:
1994 		if (!(mcg_cap & MCG_CTL_P))
1995 			return 1;
1996 		if (data != 0 && data != ~(u64)0)
1997 			return -1;
1998 		vcpu->arch.mcg_ctl = data;
1999 		break;
2000 	default:
2001 		if (msr >= MSR_IA32_MC0_CTL &&
2002 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2003 			u32 offset = msr - MSR_IA32_MC0_CTL;
2004 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2005 			 * some Linux kernels though clear bit 10 in bank 4 to
2006 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2007 			 * this to avoid an uncatched #GP in the guest
2008 			 */
2009 			if ((offset & 0x3) == 0 &&
2010 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2011 				return -1;
2012 			vcpu->arch.mce_banks[offset] = data;
2013 			break;
2014 		}
2015 		return 1;
2016 	}
2017 	return 0;
2018 }
2019 
2020 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2021 {
2022 	struct kvm *kvm = vcpu->kvm;
2023 	int lm = is_long_mode(vcpu);
2024 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2025 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2026 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2027 		: kvm->arch.xen_hvm_config.blob_size_32;
2028 	u32 page_num = data & ~PAGE_MASK;
2029 	u64 page_addr = data & PAGE_MASK;
2030 	u8 *page;
2031 	int r;
2032 
2033 	r = -E2BIG;
2034 	if (page_num >= blob_size)
2035 		goto out;
2036 	r = -ENOMEM;
2037 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2038 	if (IS_ERR(page)) {
2039 		r = PTR_ERR(page);
2040 		goto out;
2041 	}
2042 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2043 		goto out_free;
2044 	r = 0;
2045 out_free:
2046 	kfree(page);
2047 out:
2048 	return r;
2049 }
2050 
2051 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2052 {
2053 	gpa_t gpa = data & ~0x3f;
2054 
2055 	/* Bits 2:5 are reserved, Should be zero */
2056 	if (data & 0x3c)
2057 		return 1;
2058 
2059 	vcpu->arch.apf.msr_val = data;
2060 
2061 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2062 		kvm_clear_async_pf_completion_queue(vcpu);
2063 		kvm_async_pf_hash_reset(vcpu);
2064 		return 0;
2065 	}
2066 
2067 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2068 					sizeof(u32)))
2069 		return 1;
2070 
2071 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2072 	kvm_async_pf_wakeup_all(vcpu);
2073 	return 0;
2074 }
2075 
2076 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2077 {
2078 	vcpu->arch.pv_time_enabled = false;
2079 }
2080 
2081 static void record_steal_time(struct kvm_vcpu *vcpu)
2082 {
2083 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2084 		return;
2085 
2086 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2087 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2088 		return;
2089 
2090 	vcpu->arch.st.steal.preempted = 0;
2091 
2092 	if (vcpu->arch.st.steal.version & 1)
2093 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2094 
2095 	vcpu->arch.st.steal.version += 1;
2096 
2097 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2098 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2099 
2100 	smp_wmb();
2101 
2102 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2103 		vcpu->arch.st.last_steal;
2104 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2105 
2106 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2107 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2108 
2109 	smp_wmb();
2110 
2111 	vcpu->arch.st.steal.version += 1;
2112 
2113 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2114 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2115 }
2116 
2117 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2118 {
2119 	bool pr = false;
2120 	u32 msr = msr_info->index;
2121 	u64 data = msr_info->data;
2122 
2123 	switch (msr) {
2124 	case MSR_AMD64_NB_CFG:
2125 	case MSR_IA32_UCODE_REV:
2126 	case MSR_IA32_UCODE_WRITE:
2127 	case MSR_VM_HSAVE_PA:
2128 	case MSR_AMD64_PATCH_LOADER:
2129 	case MSR_AMD64_BU_CFG2:
2130 	case MSR_AMD64_DC_CFG:
2131 		break;
2132 
2133 	case MSR_EFER:
2134 		return set_efer(vcpu, data);
2135 	case MSR_K7_HWCR:
2136 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2137 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2138 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2139 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2140 		if (data != 0) {
2141 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2142 				    data);
2143 			return 1;
2144 		}
2145 		break;
2146 	case MSR_FAM10H_MMIO_CONF_BASE:
2147 		if (data != 0) {
2148 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2149 				    "0x%llx\n", data);
2150 			return 1;
2151 		}
2152 		break;
2153 	case MSR_IA32_DEBUGCTLMSR:
2154 		if (!data) {
2155 			/* We support the non-activated case already */
2156 			break;
2157 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2158 			/* Values other than LBR and BTF are vendor-specific,
2159 			   thus reserved and should throw a #GP */
2160 			return 1;
2161 		}
2162 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2163 			    __func__, data);
2164 		break;
2165 	case 0x200 ... 0x2ff:
2166 		return kvm_mtrr_set_msr(vcpu, msr, data);
2167 	case MSR_IA32_APICBASE:
2168 		return kvm_set_apic_base(vcpu, msr_info);
2169 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2170 		return kvm_x2apic_msr_write(vcpu, msr, data);
2171 	case MSR_IA32_TSCDEADLINE:
2172 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2173 		break;
2174 	case MSR_IA32_TSC_ADJUST:
2175 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2176 			if (!msr_info->host_initiated) {
2177 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2178 				adjust_tsc_offset_guest(vcpu, adj);
2179 			}
2180 			vcpu->arch.ia32_tsc_adjust_msr = data;
2181 		}
2182 		break;
2183 	case MSR_IA32_MISC_ENABLE:
2184 		vcpu->arch.ia32_misc_enable_msr = data;
2185 		break;
2186 	case MSR_IA32_SMBASE:
2187 		if (!msr_info->host_initiated)
2188 			return 1;
2189 		vcpu->arch.smbase = data;
2190 		break;
2191 	case MSR_KVM_WALL_CLOCK_NEW:
2192 	case MSR_KVM_WALL_CLOCK:
2193 		vcpu->kvm->arch.wall_clock = data;
2194 		kvm_write_wall_clock(vcpu->kvm, data);
2195 		break;
2196 	case MSR_KVM_SYSTEM_TIME_NEW:
2197 	case MSR_KVM_SYSTEM_TIME: {
2198 		struct kvm_arch *ka = &vcpu->kvm->arch;
2199 
2200 		kvmclock_reset(vcpu);
2201 
2202 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2203 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2204 
2205 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2206 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2207 
2208 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2209 		}
2210 
2211 		vcpu->arch.time = data;
2212 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2213 
2214 		/* we verify if the enable bit is set... */
2215 		if (!(data & 1))
2216 			break;
2217 
2218 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2219 		     &vcpu->arch.pv_time, data & ~1ULL,
2220 		     sizeof(struct pvclock_vcpu_time_info)))
2221 			vcpu->arch.pv_time_enabled = false;
2222 		else
2223 			vcpu->arch.pv_time_enabled = true;
2224 
2225 		break;
2226 	}
2227 	case MSR_KVM_ASYNC_PF_EN:
2228 		if (kvm_pv_enable_async_pf(vcpu, data))
2229 			return 1;
2230 		break;
2231 	case MSR_KVM_STEAL_TIME:
2232 
2233 		if (unlikely(!sched_info_on()))
2234 			return 1;
2235 
2236 		if (data & KVM_STEAL_RESERVED_MASK)
2237 			return 1;
2238 
2239 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2240 						data & KVM_STEAL_VALID_BITS,
2241 						sizeof(struct kvm_steal_time)))
2242 			return 1;
2243 
2244 		vcpu->arch.st.msr_val = data;
2245 
2246 		if (!(data & KVM_MSR_ENABLED))
2247 			break;
2248 
2249 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2250 
2251 		break;
2252 	case MSR_KVM_PV_EOI_EN:
2253 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2254 			return 1;
2255 		break;
2256 
2257 	case MSR_IA32_MCG_CTL:
2258 	case MSR_IA32_MCG_STATUS:
2259 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2260 		return set_msr_mce(vcpu, msr, data);
2261 
2262 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2263 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2264 		pr = true; /* fall through */
2265 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2266 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2267 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2268 			return kvm_pmu_set_msr(vcpu, msr_info);
2269 
2270 		if (pr || data != 0)
2271 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2272 				    "0x%x data 0x%llx\n", msr, data);
2273 		break;
2274 	case MSR_K7_CLK_CTL:
2275 		/*
2276 		 * Ignore all writes to this no longer documented MSR.
2277 		 * Writes are only relevant for old K7 processors,
2278 		 * all pre-dating SVM, but a recommended workaround from
2279 		 * AMD for these chips. It is possible to specify the
2280 		 * affected processor models on the command line, hence
2281 		 * the need to ignore the workaround.
2282 		 */
2283 		break;
2284 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2285 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2286 	case HV_X64_MSR_CRASH_CTL:
2287 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2288 		return kvm_hv_set_msr_common(vcpu, msr, data,
2289 					     msr_info->host_initiated);
2290 	case MSR_IA32_BBL_CR_CTL3:
2291 		/* Drop writes to this legacy MSR -- see rdmsr
2292 		 * counterpart for further detail.
2293 		 */
2294 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2295 		break;
2296 	case MSR_AMD64_OSVW_ID_LENGTH:
2297 		if (!guest_cpuid_has_osvw(vcpu))
2298 			return 1;
2299 		vcpu->arch.osvw.length = data;
2300 		break;
2301 	case MSR_AMD64_OSVW_STATUS:
2302 		if (!guest_cpuid_has_osvw(vcpu))
2303 			return 1;
2304 		vcpu->arch.osvw.status = data;
2305 		break;
2306 	case MSR_PLATFORM_INFO:
2307 		if (!msr_info->host_initiated ||
2308 		    data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2309 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2310 		     cpuid_fault_enabled(vcpu)))
2311 			return 1;
2312 		vcpu->arch.msr_platform_info = data;
2313 		break;
2314 	case MSR_MISC_FEATURES_ENABLES:
2315 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2316 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2317 		     !supports_cpuid_fault(vcpu)))
2318 			return 1;
2319 		vcpu->arch.msr_misc_features_enables = data;
2320 		break;
2321 	default:
2322 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2323 			return xen_hvm_config(vcpu, data);
2324 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2325 			return kvm_pmu_set_msr(vcpu, msr_info);
2326 		if (!ignore_msrs) {
2327 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2328 				    msr, data);
2329 			return 1;
2330 		} else {
2331 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2332 				    msr, data);
2333 			break;
2334 		}
2335 	}
2336 	return 0;
2337 }
2338 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2339 
2340 
2341 /*
2342  * Reads an msr value (of 'msr_index') into 'pdata'.
2343  * Returns 0 on success, non-0 otherwise.
2344  * Assumes vcpu_load() was already called.
2345  */
2346 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2347 {
2348 	return kvm_x86_ops->get_msr(vcpu, msr);
2349 }
2350 EXPORT_SYMBOL_GPL(kvm_get_msr);
2351 
2352 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2353 {
2354 	u64 data;
2355 	u64 mcg_cap = vcpu->arch.mcg_cap;
2356 	unsigned bank_num = mcg_cap & 0xff;
2357 
2358 	switch (msr) {
2359 	case MSR_IA32_P5_MC_ADDR:
2360 	case MSR_IA32_P5_MC_TYPE:
2361 		data = 0;
2362 		break;
2363 	case MSR_IA32_MCG_CAP:
2364 		data = vcpu->arch.mcg_cap;
2365 		break;
2366 	case MSR_IA32_MCG_CTL:
2367 		if (!(mcg_cap & MCG_CTL_P))
2368 			return 1;
2369 		data = vcpu->arch.mcg_ctl;
2370 		break;
2371 	case MSR_IA32_MCG_STATUS:
2372 		data = vcpu->arch.mcg_status;
2373 		break;
2374 	default:
2375 		if (msr >= MSR_IA32_MC0_CTL &&
2376 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2377 			u32 offset = msr - MSR_IA32_MC0_CTL;
2378 			data = vcpu->arch.mce_banks[offset];
2379 			break;
2380 		}
2381 		return 1;
2382 	}
2383 	*pdata = data;
2384 	return 0;
2385 }
2386 
2387 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2388 {
2389 	switch (msr_info->index) {
2390 	case MSR_IA32_PLATFORM_ID:
2391 	case MSR_IA32_EBL_CR_POWERON:
2392 	case MSR_IA32_DEBUGCTLMSR:
2393 	case MSR_IA32_LASTBRANCHFROMIP:
2394 	case MSR_IA32_LASTBRANCHTOIP:
2395 	case MSR_IA32_LASTINTFROMIP:
2396 	case MSR_IA32_LASTINTTOIP:
2397 	case MSR_K8_SYSCFG:
2398 	case MSR_K8_TSEG_ADDR:
2399 	case MSR_K8_TSEG_MASK:
2400 	case MSR_K7_HWCR:
2401 	case MSR_VM_HSAVE_PA:
2402 	case MSR_K8_INT_PENDING_MSG:
2403 	case MSR_AMD64_NB_CFG:
2404 	case MSR_FAM10H_MMIO_CONF_BASE:
2405 	case MSR_AMD64_BU_CFG2:
2406 	case MSR_IA32_PERF_CTL:
2407 	case MSR_AMD64_DC_CFG:
2408 		msr_info->data = 0;
2409 		break;
2410 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2411 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2412 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2413 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2414 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2415 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2416 		msr_info->data = 0;
2417 		break;
2418 	case MSR_IA32_UCODE_REV:
2419 		msr_info->data = 0x100000000ULL;
2420 		break;
2421 	case MSR_MTRRcap:
2422 	case 0x200 ... 0x2ff:
2423 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2424 	case 0xcd: /* fsb frequency */
2425 		msr_info->data = 3;
2426 		break;
2427 		/*
2428 		 * MSR_EBC_FREQUENCY_ID
2429 		 * Conservative value valid for even the basic CPU models.
2430 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2431 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2432 		 * and 266MHz for model 3, or 4. Set Core Clock
2433 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2434 		 * 31:24) even though these are only valid for CPU
2435 		 * models > 2, however guests may end up dividing or
2436 		 * multiplying by zero otherwise.
2437 		 */
2438 	case MSR_EBC_FREQUENCY_ID:
2439 		msr_info->data = 1 << 24;
2440 		break;
2441 	case MSR_IA32_APICBASE:
2442 		msr_info->data = kvm_get_apic_base(vcpu);
2443 		break;
2444 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2445 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2446 		break;
2447 	case MSR_IA32_TSCDEADLINE:
2448 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2449 		break;
2450 	case MSR_IA32_TSC_ADJUST:
2451 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2452 		break;
2453 	case MSR_IA32_MISC_ENABLE:
2454 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2455 		break;
2456 	case MSR_IA32_SMBASE:
2457 		if (!msr_info->host_initiated)
2458 			return 1;
2459 		msr_info->data = vcpu->arch.smbase;
2460 		break;
2461 	case MSR_IA32_PERF_STATUS:
2462 		/* TSC increment by tick */
2463 		msr_info->data = 1000ULL;
2464 		/* CPU multiplier */
2465 		msr_info->data |= (((uint64_t)4ULL) << 40);
2466 		break;
2467 	case MSR_EFER:
2468 		msr_info->data = vcpu->arch.efer;
2469 		break;
2470 	case MSR_KVM_WALL_CLOCK:
2471 	case MSR_KVM_WALL_CLOCK_NEW:
2472 		msr_info->data = vcpu->kvm->arch.wall_clock;
2473 		break;
2474 	case MSR_KVM_SYSTEM_TIME:
2475 	case MSR_KVM_SYSTEM_TIME_NEW:
2476 		msr_info->data = vcpu->arch.time;
2477 		break;
2478 	case MSR_KVM_ASYNC_PF_EN:
2479 		msr_info->data = vcpu->arch.apf.msr_val;
2480 		break;
2481 	case MSR_KVM_STEAL_TIME:
2482 		msr_info->data = vcpu->arch.st.msr_val;
2483 		break;
2484 	case MSR_KVM_PV_EOI_EN:
2485 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2486 		break;
2487 	case MSR_IA32_P5_MC_ADDR:
2488 	case MSR_IA32_P5_MC_TYPE:
2489 	case MSR_IA32_MCG_CAP:
2490 	case MSR_IA32_MCG_CTL:
2491 	case MSR_IA32_MCG_STATUS:
2492 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2493 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2494 	case MSR_K7_CLK_CTL:
2495 		/*
2496 		 * Provide expected ramp-up count for K7. All other
2497 		 * are set to zero, indicating minimum divisors for
2498 		 * every field.
2499 		 *
2500 		 * This prevents guest kernels on AMD host with CPU
2501 		 * type 6, model 8 and higher from exploding due to
2502 		 * the rdmsr failing.
2503 		 */
2504 		msr_info->data = 0x20000000;
2505 		break;
2506 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2507 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2508 	case HV_X64_MSR_CRASH_CTL:
2509 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2510 		return kvm_hv_get_msr_common(vcpu,
2511 					     msr_info->index, &msr_info->data);
2512 		break;
2513 	case MSR_IA32_BBL_CR_CTL3:
2514 		/* This legacy MSR exists but isn't fully documented in current
2515 		 * silicon.  It is however accessed by winxp in very narrow
2516 		 * scenarios where it sets bit #19, itself documented as
2517 		 * a "reserved" bit.  Best effort attempt to source coherent
2518 		 * read data here should the balance of the register be
2519 		 * interpreted by the guest:
2520 		 *
2521 		 * L2 cache control register 3: 64GB range, 256KB size,
2522 		 * enabled, latency 0x1, configured
2523 		 */
2524 		msr_info->data = 0xbe702111;
2525 		break;
2526 	case MSR_AMD64_OSVW_ID_LENGTH:
2527 		if (!guest_cpuid_has_osvw(vcpu))
2528 			return 1;
2529 		msr_info->data = vcpu->arch.osvw.length;
2530 		break;
2531 	case MSR_AMD64_OSVW_STATUS:
2532 		if (!guest_cpuid_has_osvw(vcpu))
2533 			return 1;
2534 		msr_info->data = vcpu->arch.osvw.status;
2535 		break;
2536 	case MSR_PLATFORM_INFO:
2537 		msr_info->data = vcpu->arch.msr_platform_info;
2538 		break;
2539 	case MSR_MISC_FEATURES_ENABLES:
2540 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2541 		break;
2542 	default:
2543 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2544 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2545 		if (!ignore_msrs) {
2546 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2547 					       msr_info->index);
2548 			return 1;
2549 		} else {
2550 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2551 			msr_info->data = 0;
2552 		}
2553 		break;
2554 	}
2555 	return 0;
2556 }
2557 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2558 
2559 /*
2560  * Read or write a bunch of msrs. All parameters are kernel addresses.
2561  *
2562  * @return number of msrs set successfully.
2563  */
2564 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2565 		    struct kvm_msr_entry *entries,
2566 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2567 				  unsigned index, u64 *data))
2568 {
2569 	int i, idx;
2570 
2571 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2572 	for (i = 0; i < msrs->nmsrs; ++i)
2573 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2574 			break;
2575 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2576 
2577 	return i;
2578 }
2579 
2580 /*
2581  * Read or write a bunch of msrs. Parameters are user addresses.
2582  *
2583  * @return number of msrs set successfully.
2584  */
2585 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2586 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2587 				unsigned index, u64 *data),
2588 		  int writeback)
2589 {
2590 	struct kvm_msrs msrs;
2591 	struct kvm_msr_entry *entries;
2592 	int r, n;
2593 	unsigned size;
2594 
2595 	r = -EFAULT;
2596 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2597 		goto out;
2598 
2599 	r = -E2BIG;
2600 	if (msrs.nmsrs >= MAX_IO_MSRS)
2601 		goto out;
2602 
2603 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2604 	entries = memdup_user(user_msrs->entries, size);
2605 	if (IS_ERR(entries)) {
2606 		r = PTR_ERR(entries);
2607 		goto out;
2608 	}
2609 
2610 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2611 	if (r < 0)
2612 		goto out_free;
2613 
2614 	r = -EFAULT;
2615 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2616 		goto out_free;
2617 
2618 	r = n;
2619 
2620 out_free:
2621 	kfree(entries);
2622 out:
2623 	return r;
2624 }
2625 
2626 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2627 {
2628 	int r;
2629 
2630 	switch (ext) {
2631 	case KVM_CAP_IRQCHIP:
2632 	case KVM_CAP_HLT:
2633 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2634 	case KVM_CAP_SET_TSS_ADDR:
2635 	case KVM_CAP_EXT_CPUID:
2636 	case KVM_CAP_EXT_EMUL_CPUID:
2637 	case KVM_CAP_CLOCKSOURCE:
2638 	case KVM_CAP_PIT:
2639 	case KVM_CAP_NOP_IO_DELAY:
2640 	case KVM_CAP_MP_STATE:
2641 	case KVM_CAP_SYNC_MMU:
2642 	case KVM_CAP_USER_NMI:
2643 	case KVM_CAP_REINJECT_CONTROL:
2644 	case KVM_CAP_IRQ_INJECT_STATUS:
2645 	case KVM_CAP_IOEVENTFD:
2646 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2647 	case KVM_CAP_PIT2:
2648 	case KVM_CAP_PIT_STATE2:
2649 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2650 	case KVM_CAP_XEN_HVM:
2651 	case KVM_CAP_VCPU_EVENTS:
2652 	case KVM_CAP_HYPERV:
2653 	case KVM_CAP_HYPERV_VAPIC:
2654 	case KVM_CAP_HYPERV_SPIN:
2655 	case KVM_CAP_HYPERV_SYNIC:
2656 	case KVM_CAP_PCI_SEGMENT:
2657 	case KVM_CAP_DEBUGREGS:
2658 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2659 	case KVM_CAP_XSAVE:
2660 	case KVM_CAP_ASYNC_PF:
2661 	case KVM_CAP_GET_TSC_KHZ:
2662 	case KVM_CAP_KVMCLOCK_CTRL:
2663 	case KVM_CAP_READONLY_MEM:
2664 	case KVM_CAP_HYPERV_TIME:
2665 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2666 	case KVM_CAP_TSC_DEADLINE_TIMER:
2667 	case KVM_CAP_ENABLE_CAP_VM:
2668 	case KVM_CAP_DISABLE_QUIRKS:
2669 	case KVM_CAP_SET_BOOT_CPU_ID:
2670  	case KVM_CAP_SPLIT_IRQCHIP:
2671 	case KVM_CAP_IMMEDIATE_EXIT:
2672 		r = 1;
2673 		break;
2674 	case KVM_CAP_ADJUST_CLOCK:
2675 		r = KVM_CLOCK_TSC_STABLE;
2676 		break;
2677 	case KVM_CAP_X86_GUEST_MWAIT:
2678 		r = kvm_mwait_in_guest();
2679 		break;
2680 	case KVM_CAP_X86_SMM:
2681 		/* SMBASE is usually relocated above 1M on modern chipsets,
2682 		 * and SMM handlers might indeed rely on 4G segment limits,
2683 		 * so do not report SMM to be available if real mode is
2684 		 * emulated via vm86 mode.  Still, do not go to great lengths
2685 		 * to avoid userspace's usage of the feature, because it is a
2686 		 * fringe case that is not enabled except via specific settings
2687 		 * of the module parameters.
2688 		 */
2689 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2690 		break;
2691 	case KVM_CAP_VAPIC:
2692 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2693 		break;
2694 	case KVM_CAP_NR_VCPUS:
2695 		r = KVM_SOFT_MAX_VCPUS;
2696 		break;
2697 	case KVM_CAP_MAX_VCPUS:
2698 		r = KVM_MAX_VCPUS;
2699 		break;
2700 	case KVM_CAP_NR_MEMSLOTS:
2701 		r = KVM_USER_MEM_SLOTS;
2702 		break;
2703 	case KVM_CAP_PV_MMU:	/* obsolete */
2704 		r = 0;
2705 		break;
2706 	case KVM_CAP_MCE:
2707 		r = KVM_MAX_MCE_BANKS;
2708 		break;
2709 	case KVM_CAP_XCRS:
2710 		r = boot_cpu_has(X86_FEATURE_XSAVE);
2711 		break;
2712 	case KVM_CAP_TSC_CONTROL:
2713 		r = kvm_has_tsc_control;
2714 		break;
2715 	case KVM_CAP_X2APIC_API:
2716 		r = KVM_X2APIC_API_VALID_FLAGS;
2717 		break;
2718 	default:
2719 		r = 0;
2720 		break;
2721 	}
2722 	return r;
2723 
2724 }
2725 
2726 long kvm_arch_dev_ioctl(struct file *filp,
2727 			unsigned int ioctl, unsigned long arg)
2728 {
2729 	void __user *argp = (void __user *)arg;
2730 	long r;
2731 
2732 	switch (ioctl) {
2733 	case KVM_GET_MSR_INDEX_LIST: {
2734 		struct kvm_msr_list __user *user_msr_list = argp;
2735 		struct kvm_msr_list msr_list;
2736 		unsigned n;
2737 
2738 		r = -EFAULT;
2739 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2740 			goto out;
2741 		n = msr_list.nmsrs;
2742 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2743 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2744 			goto out;
2745 		r = -E2BIG;
2746 		if (n < msr_list.nmsrs)
2747 			goto out;
2748 		r = -EFAULT;
2749 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2750 				 num_msrs_to_save * sizeof(u32)))
2751 			goto out;
2752 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2753 				 &emulated_msrs,
2754 				 num_emulated_msrs * sizeof(u32)))
2755 			goto out;
2756 		r = 0;
2757 		break;
2758 	}
2759 	case KVM_GET_SUPPORTED_CPUID:
2760 	case KVM_GET_EMULATED_CPUID: {
2761 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2762 		struct kvm_cpuid2 cpuid;
2763 
2764 		r = -EFAULT;
2765 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2766 			goto out;
2767 
2768 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2769 					    ioctl);
2770 		if (r)
2771 			goto out;
2772 
2773 		r = -EFAULT;
2774 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2775 			goto out;
2776 		r = 0;
2777 		break;
2778 	}
2779 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2780 		r = -EFAULT;
2781 		if (copy_to_user(argp, &kvm_mce_cap_supported,
2782 				 sizeof(kvm_mce_cap_supported)))
2783 			goto out;
2784 		r = 0;
2785 		break;
2786 	}
2787 	default:
2788 		r = -EINVAL;
2789 	}
2790 out:
2791 	return r;
2792 }
2793 
2794 static void wbinvd_ipi(void *garbage)
2795 {
2796 	wbinvd();
2797 }
2798 
2799 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2800 {
2801 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2802 }
2803 
2804 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2805 {
2806 	/* Address WBINVD may be executed by guest */
2807 	if (need_emulate_wbinvd(vcpu)) {
2808 		if (kvm_x86_ops->has_wbinvd_exit())
2809 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2810 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2811 			smp_call_function_single(vcpu->cpu,
2812 					wbinvd_ipi, NULL, 1);
2813 	}
2814 
2815 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2816 
2817 	/* Apply any externally detected TSC adjustments (due to suspend) */
2818 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2819 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2820 		vcpu->arch.tsc_offset_adjustment = 0;
2821 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2822 	}
2823 
2824 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2825 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2826 				rdtsc() - vcpu->arch.last_host_tsc;
2827 		if (tsc_delta < 0)
2828 			mark_tsc_unstable("KVM discovered backwards TSC");
2829 
2830 		if (check_tsc_unstable()) {
2831 			u64 offset = kvm_compute_tsc_offset(vcpu,
2832 						vcpu->arch.last_guest_tsc);
2833 			kvm_vcpu_write_tsc_offset(vcpu, offset);
2834 			vcpu->arch.tsc_catchup = 1;
2835 		}
2836 		if (kvm_lapic_hv_timer_in_use(vcpu) &&
2837 				kvm_x86_ops->set_hv_timer(vcpu,
2838 					kvm_get_lapic_target_expiration_tsc(vcpu)))
2839 			kvm_lapic_switch_to_sw_timer(vcpu);
2840 		/*
2841 		 * On a host with synchronized TSC, there is no need to update
2842 		 * kvmclock on vcpu->cpu migration
2843 		 */
2844 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2845 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2846 		if (vcpu->cpu != cpu)
2847 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2848 		vcpu->cpu = cpu;
2849 	}
2850 
2851 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2852 }
2853 
2854 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2855 {
2856 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2857 		return;
2858 
2859 	vcpu->arch.st.steal.preempted = 1;
2860 
2861 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2862 			&vcpu->arch.st.steal.preempted,
2863 			offsetof(struct kvm_steal_time, preempted),
2864 			sizeof(vcpu->arch.st.steal.preempted));
2865 }
2866 
2867 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2868 {
2869 	int idx;
2870 	/*
2871 	 * Disable page faults because we're in atomic context here.
2872 	 * kvm_write_guest_offset_cached() would call might_fault()
2873 	 * that relies on pagefault_disable() to tell if there's a
2874 	 * bug. NOTE: the write to guest memory may not go through if
2875 	 * during postcopy live migration or if there's heavy guest
2876 	 * paging.
2877 	 */
2878 	pagefault_disable();
2879 	/*
2880 	 * kvm_memslots() will be called by
2881 	 * kvm_write_guest_offset_cached() so take the srcu lock.
2882 	 */
2883 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2884 	kvm_steal_time_set_preempted(vcpu);
2885 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2886 	pagefault_enable();
2887 	kvm_x86_ops->vcpu_put(vcpu);
2888 	kvm_put_guest_fpu(vcpu);
2889 	vcpu->arch.last_host_tsc = rdtsc();
2890 }
2891 
2892 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2893 				    struct kvm_lapic_state *s)
2894 {
2895 	if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2896 		kvm_x86_ops->sync_pir_to_irr(vcpu);
2897 
2898 	return kvm_apic_get_state(vcpu, s);
2899 }
2900 
2901 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2902 				    struct kvm_lapic_state *s)
2903 {
2904 	int r;
2905 
2906 	r = kvm_apic_set_state(vcpu, s);
2907 	if (r)
2908 		return r;
2909 	update_cr8_intercept(vcpu);
2910 
2911 	return 0;
2912 }
2913 
2914 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2915 {
2916 	return (!lapic_in_kernel(vcpu) ||
2917 		kvm_apic_accept_pic_intr(vcpu));
2918 }
2919 
2920 /*
2921  * if userspace requested an interrupt window, check that the
2922  * interrupt window is open.
2923  *
2924  * No need to exit to userspace if we already have an interrupt queued.
2925  */
2926 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2927 {
2928 	return kvm_arch_interrupt_allowed(vcpu) &&
2929 		!kvm_cpu_has_interrupt(vcpu) &&
2930 		!kvm_event_needs_reinjection(vcpu) &&
2931 		kvm_cpu_accept_dm_intr(vcpu);
2932 }
2933 
2934 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2935 				    struct kvm_interrupt *irq)
2936 {
2937 	if (irq->irq >= KVM_NR_INTERRUPTS)
2938 		return -EINVAL;
2939 
2940 	if (!irqchip_in_kernel(vcpu->kvm)) {
2941 		kvm_queue_interrupt(vcpu, irq->irq, false);
2942 		kvm_make_request(KVM_REQ_EVENT, vcpu);
2943 		return 0;
2944 	}
2945 
2946 	/*
2947 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2948 	 * fail for in-kernel 8259.
2949 	 */
2950 	if (pic_in_kernel(vcpu->kvm))
2951 		return -ENXIO;
2952 
2953 	if (vcpu->arch.pending_external_vector != -1)
2954 		return -EEXIST;
2955 
2956 	vcpu->arch.pending_external_vector = irq->irq;
2957 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2958 	return 0;
2959 }
2960 
2961 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2962 {
2963 	kvm_inject_nmi(vcpu);
2964 
2965 	return 0;
2966 }
2967 
2968 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2969 {
2970 	kvm_make_request(KVM_REQ_SMI, vcpu);
2971 
2972 	return 0;
2973 }
2974 
2975 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2976 					   struct kvm_tpr_access_ctl *tac)
2977 {
2978 	if (tac->flags)
2979 		return -EINVAL;
2980 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2981 	return 0;
2982 }
2983 
2984 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2985 					u64 mcg_cap)
2986 {
2987 	int r;
2988 	unsigned bank_num = mcg_cap & 0xff, bank;
2989 
2990 	r = -EINVAL;
2991 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2992 		goto out;
2993 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
2994 		goto out;
2995 	r = 0;
2996 	vcpu->arch.mcg_cap = mcg_cap;
2997 	/* Init IA32_MCG_CTL to all 1s */
2998 	if (mcg_cap & MCG_CTL_P)
2999 		vcpu->arch.mcg_ctl = ~(u64)0;
3000 	/* Init IA32_MCi_CTL to all 1s */
3001 	for (bank = 0; bank < bank_num; bank++)
3002 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3003 
3004 	if (kvm_x86_ops->setup_mce)
3005 		kvm_x86_ops->setup_mce(vcpu);
3006 out:
3007 	return r;
3008 }
3009 
3010 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3011 				      struct kvm_x86_mce *mce)
3012 {
3013 	u64 mcg_cap = vcpu->arch.mcg_cap;
3014 	unsigned bank_num = mcg_cap & 0xff;
3015 	u64 *banks = vcpu->arch.mce_banks;
3016 
3017 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3018 		return -EINVAL;
3019 	/*
3020 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3021 	 * reporting is disabled
3022 	 */
3023 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3024 	    vcpu->arch.mcg_ctl != ~(u64)0)
3025 		return 0;
3026 	banks += 4 * mce->bank;
3027 	/*
3028 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3029 	 * reporting is disabled for the bank
3030 	 */
3031 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3032 		return 0;
3033 	if (mce->status & MCI_STATUS_UC) {
3034 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3035 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3036 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3037 			return 0;
3038 		}
3039 		if (banks[1] & MCI_STATUS_VAL)
3040 			mce->status |= MCI_STATUS_OVER;
3041 		banks[2] = mce->addr;
3042 		banks[3] = mce->misc;
3043 		vcpu->arch.mcg_status = mce->mcg_status;
3044 		banks[1] = mce->status;
3045 		kvm_queue_exception(vcpu, MC_VECTOR);
3046 	} else if (!(banks[1] & MCI_STATUS_VAL)
3047 		   || !(banks[1] & MCI_STATUS_UC)) {
3048 		if (banks[1] & MCI_STATUS_VAL)
3049 			mce->status |= MCI_STATUS_OVER;
3050 		banks[2] = mce->addr;
3051 		banks[3] = mce->misc;
3052 		banks[1] = mce->status;
3053 	} else
3054 		banks[1] |= MCI_STATUS_OVER;
3055 	return 0;
3056 }
3057 
3058 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3059 					       struct kvm_vcpu_events *events)
3060 {
3061 	process_nmi(vcpu);
3062 	events->exception.injected =
3063 		vcpu->arch.exception.pending &&
3064 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3065 	events->exception.nr = vcpu->arch.exception.nr;
3066 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3067 	events->exception.pad = 0;
3068 	events->exception.error_code = vcpu->arch.exception.error_code;
3069 
3070 	events->interrupt.injected =
3071 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3072 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3073 	events->interrupt.soft = 0;
3074 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3075 
3076 	events->nmi.injected = vcpu->arch.nmi_injected;
3077 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3078 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3079 	events->nmi.pad = 0;
3080 
3081 	events->sipi_vector = 0; /* never valid when reporting to user space */
3082 
3083 	events->smi.smm = is_smm(vcpu);
3084 	events->smi.pending = vcpu->arch.smi_pending;
3085 	events->smi.smm_inside_nmi =
3086 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3087 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3088 
3089 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3090 			 | KVM_VCPUEVENT_VALID_SHADOW
3091 			 | KVM_VCPUEVENT_VALID_SMM);
3092 	memset(&events->reserved, 0, sizeof(events->reserved));
3093 }
3094 
3095 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3096 
3097 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3098 					      struct kvm_vcpu_events *events)
3099 {
3100 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3101 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3102 			      | KVM_VCPUEVENT_VALID_SHADOW
3103 			      | KVM_VCPUEVENT_VALID_SMM))
3104 		return -EINVAL;
3105 
3106 	if (events->exception.injected &&
3107 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3108 	     is_guest_mode(vcpu)))
3109 		return -EINVAL;
3110 
3111 	/* INITs are latched while in SMM */
3112 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3113 	    (events->smi.smm || events->smi.pending) &&
3114 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3115 		return -EINVAL;
3116 
3117 	process_nmi(vcpu);
3118 	vcpu->arch.exception.pending = events->exception.injected;
3119 	vcpu->arch.exception.nr = events->exception.nr;
3120 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3121 	vcpu->arch.exception.error_code = events->exception.error_code;
3122 
3123 	vcpu->arch.interrupt.pending = events->interrupt.injected;
3124 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3125 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3126 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3127 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3128 						  events->interrupt.shadow);
3129 
3130 	vcpu->arch.nmi_injected = events->nmi.injected;
3131 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3132 		vcpu->arch.nmi_pending = events->nmi.pending;
3133 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3134 
3135 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3136 	    lapic_in_kernel(vcpu))
3137 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3138 
3139 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3140 		u32 hflags = vcpu->arch.hflags;
3141 		if (events->smi.smm)
3142 			hflags |= HF_SMM_MASK;
3143 		else
3144 			hflags &= ~HF_SMM_MASK;
3145 		kvm_set_hflags(vcpu, hflags);
3146 
3147 		vcpu->arch.smi_pending = events->smi.pending;
3148 		if (events->smi.smm_inside_nmi)
3149 			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3150 		else
3151 			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3152 		if (lapic_in_kernel(vcpu)) {
3153 			if (events->smi.latched_init)
3154 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3155 			else
3156 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3157 		}
3158 	}
3159 
3160 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3161 
3162 	return 0;
3163 }
3164 
3165 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3166 					     struct kvm_debugregs *dbgregs)
3167 {
3168 	unsigned long val;
3169 
3170 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3171 	kvm_get_dr(vcpu, 6, &val);
3172 	dbgregs->dr6 = val;
3173 	dbgregs->dr7 = vcpu->arch.dr7;
3174 	dbgregs->flags = 0;
3175 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3176 }
3177 
3178 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3179 					    struct kvm_debugregs *dbgregs)
3180 {
3181 	if (dbgregs->flags)
3182 		return -EINVAL;
3183 
3184 	if (dbgregs->dr6 & ~0xffffffffull)
3185 		return -EINVAL;
3186 	if (dbgregs->dr7 & ~0xffffffffull)
3187 		return -EINVAL;
3188 
3189 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3190 	kvm_update_dr0123(vcpu);
3191 	vcpu->arch.dr6 = dbgregs->dr6;
3192 	kvm_update_dr6(vcpu);
3193 	vcpu->arch.dr7 = dbgregs->dr7;
3194 	kvm_update_dr7(vcpu);
3195 
3196 	return 0;
3197 }
3198 
3199 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3200 
3201 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3202 {
3203 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3204 	u64 xstate_bv = xsave->header.xfeatures;
3205 	u64 valid;
3206 
3207 	/*
3208 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3209 	 * leaves 0 and 1 in the loop below.
3210 	 */
3211 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3212 
3213 	/* Set XSTATE_BV */
3214 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3215 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3216 
3217 	/*
3218 	 * Copy each region from the possibly compacted offset to the
3219 	 * non-compacted offset.
3220 	 */
3221 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3222 	while (valid) {
3223 		u64 feature = valid & -valid;
3224 		int index = fls64(feature) - 1;
3225 		void *src = get_xsave_addr(xsave, feature);
3226 
3227 		if (src) {
3228 			u32 size, offset, ecx, edx;
3229 			cpuid_count(XSTATE_CPUID, index,
3230 				    &size, &offset, &ecx, &edx);
3231 			memcpy(dest + offset, src, size);
3232 		}
3233 
3234 		valid -= feature;
3235 	}
3236 }
3237 
3238 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3239 {
3240 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3241 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3242 	u64 valid;
3243 
3244 	/*
3245 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3246 	 * leaves 0 and 1 in the loop below.
3247 	 */
3248 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3249 
3250 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3251 	xsave->header.xfeatures = xstate_bv;
3252 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3253 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3254 
3255 	/*
3256 	 * Copy each region from the non-compacted offset to the
3257 	 * possibly compacted offset.
3258 	 */
3259 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3260 	while (valid) {
3261 		u64 feature = valid & -valid;
3262 		int index = fls64(feature) - 1;
3263 		void *dest = get_xsave_addr(xsave, feature);
3264 
3265 		if (dest) {
3266 			u32 size, offset, ecx, edx;
3267 			cpuid_count(XSTATE_CPUID, index,
3268 				    &size, &offset, &ecx, &edx);
3269 			memcpy(dest, src + offset, size);
3270 		}
3271 
3272 		valid -= feature;
3273 	}
3274 }
3275 
3276 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3277 					 struct kvm_xsave *guest_xsave)
3278 {
3279 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3280 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3281 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3282 	} else {
3283 		memcpy(guest_xsave->region,
3284 			&vcpu->arch.guest_fpu.state.fxsave,
3285 			sizeof(struct fxregs_state));
3286 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3287 			XFEATURE_MASK_FPSSE;
3288 	}
3289 }
3290 
3291 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3292 					struct kvm_xsave *guest_xsave)
3293 {
3294 	u64 xstate_bv =
3295 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3296 
3297 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3298 		/*
3299 		 * Here we allow setting states that are not present in
3300 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3301 		 * with old userspace.
3302 		 */
3303 		if (xstate_bv & ~kvm_supported_xcr0())
3304 			return -EINVAL;
3305 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3306 	} else {
3307 		if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3308 			return -EINVAL;
3309 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3310 			guest_xsave->region, sizeof(struct fxregs_state));
3311 	}
3312 	return 0;
3313 }
3314 
3315 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3316 					struct kvm_xcrs *guest_xcrs)
3317 {
3318 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3319 		guest_xcrs->nr_xcrs = 0;
3320 		return;
3321 	}
3322 
3323 	guest_xcrs->nr_xcrs = 1;
3324 	guest_xcrs->flags = 0;
3325 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3326 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3327 }
3328 
3329 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3330 				       struct kvm_xcrs *guest_xcrs)
3331 {
3332 	int i, r = 0;
3333 
3334 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3335 		return -EINVAL;
3336 
3337 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3338 		return -EINVAL;
3339 
3340 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3341 		/* Only support XCR0 currently */
3342 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3343 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3344 				guest_xcrs->xcrs[i].value);
3345 			break;
3346 		}
3347 	if (r)
3348 		r = -EINVAL;
3349 	return r;
3350 }
3351 
3352 /*
3353  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3354  * stopped by the hypervisor.  This function will be called from the host only.
3355  * EINVAL is returned when the host attempts to set the flag for a guest that
3356  * does not support pv clocks.
3357  */
3358 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3359 {
3360 	if (!vcpu->arch.pv_time_enabled)
3361 		return -EINVAL;
3362 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3363 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3364 	return 0;
3365 }
3366 
3367 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3368 				     struct kvm_enable_cap *cap)
3369 {
3370 	if (cap->flags)
3371 		return -EINVAL;
3372 
3373 	switch (cap->cap) {
3374 	case KVM_CAP_HYPERV_SYNIC:
3375 		if (!irqchip_in_kernel(vcpu->kvm))
3376 			return -EINVAL;
3377 		return kvm_hv_activate_synic(vcpu);
3378 	default:
3379 		return -EINVAL;
3380 	}
3381 }
3382 
3383 long kvm_arch_vcpu_ioctl(struct file *filp,
3384 			 unsigned int ioctl, unsigned long arg)
3385 {
3386 	struct kvm_vcpu *vcpu = filp->private_data;
3387 	void __user *argp = (void __user *)arg;
3388 	int r;
3389 	union {
3390 		struct kvm_lapic_state *lapic;
3391 		struct kvm_xsave *xsave;
3392 		struct kvm_xcrs *xcrs;
3393 		void *buffer;
3394 	} u;
3395 
3396 	u.buffer = NULL;
3397 	switch (ioctl) {
3398 	case KVM_GET_LAPIC: {
3399 		r = -EINVAL;
3400 		if (!lapic_in_kernel(vcpu))
3401 			goto out;
3402 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3403 
3404 		r = -ENOMEM;
3405 		if (!u.lapic)
3406 			goto out;
3407 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3408 		if (r)
3409 			goto out;
3410 		r = -EFAULT;
3411 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3412 			goto out;
3413 		r = 0;
3414 		break;
3415 	}
3416 	case KVM_SET_LAPIC: {
3417 		r = -EINVAL;
3418 		if (!lapic_in_kernel(vcpu))
3419 			goto out;
3420 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3421 		if (IS_ERR(u.lapic))
3422 			return PTR_ERR(u.lapic);
3423 
3424 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3425 		break;
3426 	}
3427 	case KVM_INTERRUPT: {
3428 		struct kvm_interrupt irq;
3429 
3430 		r = -EFAULT;
3431 		if (copy_from_user(&irq, argp, sizeof irq))
3432 			goto out;
3433 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3434 		break;
3435 	}
3436 	case KVM_NMI: {
3437 		r = kvm_vcpu_ioctl_nmi(vcpu);
3438 		break;
3439 	}
3440 	case KVM_SMI: {
3441 		r = kvm_vcpu_ioctl_smi(vcpu);
3442 		break;
3443 	}
3444 	case KVM_SET_CPUID: {
3445 		struct kvm_cpuid __user *cpuid_arg = argp;
3446 		struct kvm_cpuid cpuid;
3447 
3448 		r = -EFAULT;
3449 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3450 			goto out;
3451 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3452 		break;
3453 	}
3454 	case KVM_SET_CPUID2: {
3455 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3456 		struct kvm_cpuid2 cpuid;
3457 
3458 		r = -EFAULT;
3459 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3460 			goto out;
3461 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3462 					      cpuid_arg->entries);
3463 		break;
3464 	}
3465 	case KVM_GET_CPUID2: {
3466 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3467 		struct kvm_cpuid2 cpuid;
3468 
3469 		r = -EFAULT;
3470 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3471 			goto out;
3472 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3473 					      cpuid_arg->entries);
3474 		if (r)
3475 			goto out;
3476 		r = -EFAULT;
3477 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3478 			goto out;
3479 		r = 0;
3480 		break;
3481 	}
3482 	case KVM_GET_MSRS:
3483 		r = msr_io(vcpu, argp, do_get_msr, 1);
3484 		break;
3485 	case KVM_SET_MSRS:
3486 		r = msr_io(vcpu, argp, do_set_msr, 0);
3487 		break;
3488 	case KVM_TPR_ACCESS_REPORTING: {
3489 		struct kvm_tpr_access_ctl tac;
3490 
3491 		r = -EFAULT;
3492 		if (copy_from_user(&tac, argp, sizeof tac))
3493 			goto out;
3494 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3495 		if (r)
3496 			goto out;
3497 		r = -EFAULT;
3498 		if (copy_to_user(argp, &tac, sizeof tac))
3499 			goto out;
3500 		r = 0;
3501 		break;
3502 	};
3503 	case KVM_SET_VAPIC_ADDR: {
3504 		struct kvm_vapic_addr va;
3505 		int idx;
3506 
3507 		r = -EINVAL;
3508 		if (!lapic_in_kernel(vcpu))
3509 			goto out;
3510 		r = -EFAULT;
3511 		if (copy_from_user(&va, argp, sizeof va))
3512 			goto out;
3513 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3514 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3515 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3516 		break;
3517 	}
3518 	case KVM_X86_SETUP_MCE: {
3519 		u64 mcg_cap;
3520 
3521 		r = -EFAULT;
3522 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3523 			goto out;
3524 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3525 		break;
3526 	}
3527 	case KVM_X86_SET_MCE: {
3528 		struct kvm_x86_mce mce;
3529 
3530 		r = -EFAULT;
3531 		if (copy_from_user(&mce, argp, sizeof mce))
3532 			goto out;
3533 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3534 		break;
3535 	}
3536 	case KVM_GET_VCPU_EVENTS: {
3537 		struct kvm_vcpu_events events;
3538 
3539 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3540 
3541 		r = -EFAULT;
3542 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3543 			break;
3544 		r = 0;
3545 		break;
3546 	}
3547 	case KVM_SET_VCPU_EVENTS: {
3548 		struct kvm_vcpu_events events;
3549 
3550 		r = -EFAULT;
3551 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3552 			break;
3553 
3554 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3555 		break;
3556 	}
3557 	case KVM_GET_DEBUGREGS: {
3558 		struct kvm_debugregs dbgregs;
3559 
3560 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3561 
3562 		r = -EFAULT;
3563 		if (copy_to_user(argp, &dbgregs,
3564 				 sizeof(struct kvm_debugregs)))
3565 			break;
3566 		r = 0;
3567 		break;
3568 	}
3569 	case KVM_SET_DEBUGREGS: {
3570 		struct kvm_debugregs dbgregs;
3571 
3572 		r = -EFAULT;
3573 		if (copy_from_user(&dbgregs, argp,
3574 				   sizeof(struct kvm_debugregs)))
3575 			break;
3576 
3577 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3578 		break;
3579 	}
3580 	case KVM_GET_XSAVE: {
3581 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3582 		r = -ENOMEM;
3583 		if (!u.xsave)
3584 			break;
3585 
3586 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3587 
3588 		r = -EFAULT;
3589 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3590 			break;
3591 		r = 0;
3592 		break;
3593 	}
3594 	case KVM_SET_XSAVE: {
3595 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3596 		if (IS_ERR(u.xsave))
3597 			return PTR_ERR(u.xsave);
3598 
3599 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3600 		break;
3601 	}
3602 	case KVM_GET_XCRS: {
3603 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3604 		r = -ENOMEM;
3605 		if (!u.xcrs)
3606 			break;
3607 
3608 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3609 
3610 		r = -EFAULT;
3611 		if (copy_to_user(argp, u.xcrs,
3612 				 sizeof(struct kvm_xcrs)))
3613 			break;
3614 		r = 0;
3615 		break;
3616 	}
3617 	case KVM_SET_XCRS: {
3618 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3619 		if (IS_ERR(u.xcrs))
3620 			return PTR_ERR(u.xcrs);
3621 
3622 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3623 		break;
3624 	}
3625 	case KVM_SET_TSC_KHZ: {
3626 		u32 user_tsc_khz;
3627 
3628 		r = -EINVAL;
3629 		user_tsc_khz = (u32)arg;
3630 
3631 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3632 			goto out;
3633 
3634 		if (user_tsc_khz == 0)
3635 			user_tsc_khz = tsc_khz;
3636 
3637 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3638 			r = 0;
3639 
3640 		goto out;
3641 	}
3642 	case KVM_GET_TSC_KHZ: {
3643 		r = vcpu->arch.virtual_tsc_khz;
3644 		goto out;
3645 	}
3646 	case KVM_KVMCLOCK_CTRL: {
3647 		r = kvm_set_guest_paused(vcpu);
3648 		goto out;
3649 	}
3650 	case KVM_ENABLE_CAP: {
3651 		struct kvm_enable_cap cap;
3652 
3653 		r = -EFAULT;
3654 		if (copy_from_user(&cap, argp, sizeof(cap)))
3655 			goto out;
3656 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3657 		break;
3658 	}
3659 	default:
3660 		r = -EINVAL;
3661 	}
3662 out:
3663 	kfree(u.buffer);
3664 	return r;
3665 }
3666 
3667 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3668 {
3669 	return VM_FAULT_SIGBUS;
3670 }
3671 
3672 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3673 {
3674 	int ret;
3675 
3676 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3677 		return -EINVAL;
3678 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3679 	return ret;
3680 }
3681 
3682 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3683 					      u64 ident_addr)
3684 {
3685 	kvm->arch.ept_identity_map_addr = ident_addr;
3686 	return 0;
3687 }
3688 
3689 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3690 					  u32 kvm_nr_mmu_pages)
3691 {
3692 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3693 		return -EINVAL;
3694 
3695 	mutex_lock(&kvm->slots_lock);
3696 
3697 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3698 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3699 
3700 	mutex_unlock(&kvm->slots_lock);
3701 	return 0;
3702 }
3703 
3704 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3705 {
3706 	return kvm->arch.n_max_mmu_pages;
3707 }
3708 
3709 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3710 {
3711 	struct kvm_pic *pic = kvm->arch.vpic;
3712 	int r;
3713 
3714 	r = 0;
3715 	switch (chip->chip_id) {
3716 	case KVM_IRQCHIP_PIC_MASTER:
3717 		memcpy(&chip->chip.pic, &pic->pics[0],
3718 			sizeof(struct kvm_pic_state));
3719 		break;
3720 	case KVM_IRQCHIP_PIC_SLAVE:
3721 		memcpy(&chip->chip.pic, &pic->pics[1],
3722 			sizeof(struct kvm_pic_state));
3723 		break;
3724 	case KVM_IRQCHIP_IOAPIC:
3725 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
3726 		break;
3727 	default:
3728 		r = -EINVAL;
3729 		break;
3730 	}
3731 	return r;
3732 }
3733 
3734 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3735 {
3736 	struct kvm_pic *pic = kvm->arch.vpic;
3737 	int r;
3738 
3739 	r = 0;
3740 	switch (chip->chip_id) {
3741 	case KVM_IRQCHIP_PIC_MASTER:
3742 		spin_lock(&pic->lock);
3743 		memcpy(&pic->pics[0], &chip->chip.pic,
3744 			sizeof(struct kvm_pic_state));
3745 		spin_unlock(&pic->lock);
3746 		break;
3747 	case KVM_IRQCHIP_PIC_SLAVE:
3748 		spin_lock(&pic->lock);
3749 		memcpy(&pic->pics[1], &chip->chip.pic,
3750 			sizeof(struct kvm_pic_state));
3751 		spin_unlock(&pic->lock);
3752 		break;
3753 	case KVM_IRQCHIP_IOAPIC:
3754 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
3755 		break;
3756 	default:
3757 		r = -EINVAL;
3758 		break;
3759 	}
3760 	kvm_pic_update_irq(pic);
3761 	return r;
3762 }
3763 
3764 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3765 {
3766 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3767 
3768 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3769 
3770 	mutex_lock(&kps->lock);
3771 	memcpy(ps, &kps->channels, sizeof(*ps));
3772 	mutex_unlock(&kps->lock);
3773 	return 0;
3774 }
3775 
3776 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3777 {
3778 	int i;
3779 	struct kvm_pit *pit = kvm->arch.vpit;
3780 
3781 	mutex_lock(&pit->pit_state.lock);
3782 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3783 	for (i = 0; i < 3; i++)
3784 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3785 	mutex_unlock(&pit->pit_state.lock);
3786 	return 0;
3787 }
3788 
3789 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3790 {
3791 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3792 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3793 		sizeof(ps->channels));
3794 	ps->flags = kvm->arch.vpit->pit_state.flags;
3795 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3796 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3797 	return 0;
3798 }
3799 
3800 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3801 {
3802 	int start = 0;
3803 	int i;
3804 	u32 prev_legacy, cur_legacy;
3805 	struct kvm_pit *pit = kvm->arch.vpit;
3806 
3807 	mutex_lock(&pit->pit_state.lock);
3808 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3809 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3810 	if (!prev_legacy && cur_legacy)
3811 		start = 1;
3812 	memcpy(&pit->pit_state.channels, &ps->channels,
3813 	       sizeof(pit->pit_state.channels));
3814 	pit->pit_state.flags = ps->flags;
3815 	for (i = 0; i < 3; i++)
3816 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3817 				   start && i == 0);
3818 	mutex_unlock(&pit->pit_state.lock);
3819 	return 0;
3820 }
3821 
3822 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3823 				 struct kvm_reinject_control *control)
3824 {
3825 	struct kvm_pit *pit = kvm->arch.vpit;
3826 
3827 	if (!pit)
3828 		return -ENXIO;
3829 
3830 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
3831 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3832 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
3833 	 */
3834 	mutex_lock(&pit->pit_state.lock);
3835 	kvm_pit_set_reinject(pit, control->pit_reinject);
3836 	mutex_unlock(&pit->pit_state.lock);
3837 
3838 	return 0;
3839 }
3840 
3841 /**
3842  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3843  * @kvm: kvm instance
3844  * @log: slot id and address to which we copy the log
3845  *
3846  * Steps 1-4 below provide general overview of dirty page logging. See
3847  * kvm_get_dirty_log_protect() function description for additional details.
3848  *
3849  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3850  * always flush the TLB (step 4) even if previous step failed  and the dirty
3851  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3852  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3853  * writes will be marked dirty for next log read.
3854  *
3855  *   1. Take a snapshot of the bit and clear it if needed.
3856  *   2. Write protect the corresponding page.
3857  *   3. Copy the snapshot to the userspace.
3858  *   4. Flush TLB's if needed.
3859  */
3860 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3861 {
3862 	bool is_dirty = false;
3863 	int r;
3864 
3865 	mutex_lock(&kvm->slots_lock);
3866 
3867 	/*
3868 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3869 	 */
3870 	if (kvm_x86_ops->flush_log_dirty)
3871 		kvm_x86_ops->flush_log_dirty(kvm);
3872 
3873 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3874 
3875 	/*
3876 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3877 	 * kvm_mmu_slot_remove_write_access().
3878 	 */
3879 	lockdep_assert_held(&kvm->slots_lock);
3880 	if (is_dirty)
3881 		kvm_flush_remote_tlbs(kvm);
3882 
3883 	mutex_unlock(&kvm->slots_lock);
3884 	return r;
3885 }
3886 
3887 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3888 			bool line_status)
3889 {
3890 	if (!irqchip_in_kernel(kvm))
3891 		return -ENXIO;
3892 
3893 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3894 					irq_event->irq, irq_event->level,
3895 					line_status);
3896 	return 0;
3897 }
3898 
3899 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3900 				   struct kvm_enable_cap *cap)
3901 {
3902 	int r;
3903 
3904 	if (cap->flags)
3905 		return -EINVAL;
3906 
3907 	switch (cap->cap) {
3908 	case KVM_CAP_DISABLE_QUIRKS:
3909 		kvm->arch.disabled_quirks = cap->args[0];
3910 		r = 0;
3911 		break;
3912 	case KVM_CAP_SPLIT_IRQCHIP: {
3913 		mutex_lock(&kvm->lock);
3914 		r = -EINVAL;
3915 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3916 			goto split_irqchip_unlock;
3917 		r = -EEXIST;
3918 		if (irqchip_in_kernel(kvm))
3919 			goto split_irqchip_unlock;
3920 		if (kvm->created_vcpus)
3921 			goto split_irqchip_unlock;
3922 		r = kvm_setup_empty_irq_routing(kvm);
3923 		if (r)
3924 			goto split_irqchip_unlock;
3925 		/* Pairs with irqchip_in_kernel. */
3926 		smp_wmb();
3927 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3928 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3929 		r = 0;
3930 split_irqchip_unlock:
3931 		mutex_unlock(&kvm->lock);
3932 		break;
3933 	}
3934 	case KVM_CAP_X2APIC_API:
3935 		r = -EINVAL;
3936 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3937 			break;
3938 
3939 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3940 			kvm->arch.x2apic_format = true;
3941 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3942 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
3943 
3944 		r = 0;
3945 		break;
3946 	default:
3947 		r = -EINVAL;
3948 		break;
3949 	}
3950 	return r;
3951 }
3952 
3953 long kvm_arch_vm_ioctl(struct file *filp,
3954 		       unsigned int ioctl, unsigned long arg)
3955 {
3956 	struct kvm *kvm = filp->private_data;
3957 	void __user *argp = (void __user *)arg;
3958 	int r = -ENOTTY;
3959 	/*
3960 	 * This union makes it completely explicit to gcc-3.x
3961 	 * that these two variables' stack usage should be
3962 	 * combined, not added together.
3963 	 */
3964 	union {
3965 		struct kvm_pit_state ps;
3966 		struct kvm_pit_state2 ps2;
3967 		struct kvm_pit_config pit_config;
3968 	} u;
3969 
3970 	switch (ioctl) {
3971 	case KVM_SET_TSS_ADDR:
3972 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3973 		break;
3974 	case KVM_SET_IDENTITY_MAP_ADDR: {
3975 		u64 ident_addr;
3976 
3977 		r = -EFAULT;
3978 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3979 			goto out;
3980 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3981 		break;
3982 	}
3983 	case KVM_SET_NR_MMU_PAGES:
3984 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3985 		break;
3986 	case KVM_GET_NR_MMU_PAGES:
3987 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3988 		break;
3989 	case KVM_CREATE_IRQCHIP: {
3990 		mutex_lock(&kvm->lock);
3991 
3992 		r = -EEXIST;
3993 		if (irqchip_in_kernel(kvm))
3994 			goto create_irqchip_unlock;
3995 
3996 		r = -EINVAL;
3997 		if (kvm->created_vcpus)
3998 			goto create_irqchip_unlock;
3999 
4000 		r = kvm_pic_init(kvm);
4001 		if (r)
4002 			goto create_irqchip_unlock;
4003 
4004 		r = kvm_ioapic_init(kvm);
4005 		if (r) {
4006 			kvm_pic_destroy(kvm);
4007 			goto create_irqchip_unlock;
4008 		}
4009 
4010 		r = kvm_setup_default_irq_routing(kvm);
4011 		if (r) {
4012 			kvm_ioapic_destroy(kvm);
4013 			kvm_pic_destroy(kvm);
4014 			goto create_irqchip_unlock;
4015 		}
4016 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4017 		smp_wmb();
4018 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4019 	create_irqchip_unlock:
4020 		mutex_unlock(&kvm->lock);
4021 		break;
4022 	}
4023 	case KVM_CREATE_PIT:
4024 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4025 		goto create_pit;
4026 	case KVM_CREATE_PIT2:
4027 		r = -EFAULT;
4028 		if (copy_from_user(&u.pit_config, argp,
4029 				   sizeof(struct kvm_pit_config)))
4030 			goto out;
4031 	create_pit:
4032 		mutex_lock(&kvm->lock);
4033 		r = -EEXIST;
4034 		if (kvm->arch.vpit)
4035 			goto create_pit_unlock;
4036 		r = -ENOMEM;
4037 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4038 		if (kvm->arch.vpit)
4039 			r = 0;
4040 	create_pit_unlock:
4041 		mutex_unlock(&kvm->lock);
4042 		break;
4043 	case KVM_GET_IRQCHIP: {
4044 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4045 		struct kvm_irqchip *chip;
4046 
4047 		chip = memdup_user(argp, sizeof(*chip));
4048 		if (IS_ERR(chip)) {
4049 			r = PTR_ERR(chip);
4050 			goto out;
4051 		}
4052 
4053 		r = -ENXIO;
4054 		if (!irqchip_kernel(kvm))
4055 			goto get_irqchip_out;
4056 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4057 		if (r)
4058 			goto get_irqchip_out;
4059 		r = -EFAULT;
4060 		if (copy_to_user(argp, chip, sizeof *chip))
4061 			goto get_irqchip_out;
4062 		r = 0;
4063 	get_irqchip_out:
4064 		kfree(chip);
4065 		break;
4066 	}
4067 	case KVM_SET_IRQCHIP: {
4068 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4069 		struct kvm_irqchip *chip;
4070 
4071 		chip = memdup_user(argp, sizeof(*chip));
4072 		if (IS_ERR(chip)) {
4073 			r = PTR_ERR(chip);
4074 			goto out;
4075 		}
4076 
4077 		r = -ENXIO;
4078 		if (!irqchip_kernel(kvm))
4079 			goto set_irqchip_out;
4080 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4081 		if (r)
4082 			goto set_irqchip_out;
4083 		r = 0;
4084 	set_irqchip_out:
4085 		kfree(chip);
4086 		break;
4087 	}
4088 	case KVM_GET_PIT: {
4089 		r = -EFAULT;
4090 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4091 			goto out;
4092 		r = -ENXIO;
4093 		if (!kvm->arch.vpit)
4094 			goto out;
4095 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4096 		if (r)
4097 			goto out;
4098 		r = -EFAULT;
4099 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4100 			goto out;
4101 		r = 0;
4102 		break;
4103 	}
4104 	case KVM_SET_PIT: {
4105 		r = -EFAULT;
4106 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4107 			goto out;
4108 		r = -ENXIO;
4109 		if (!kvm->arch.vpit)
4110 			goto out;
4111 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4112 		break;
4113 	}
4114 	case KVM_GET_PIT2: {
4115 		r = -ENXIO;
4116 		if (!kvm->arch.vpit)
4117 			goto out;
4118 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4119 		if (r)
4120 			goto out;
4121 		r = -EFAULT;
4122 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4123 			goto out;
4124 		r = 0;
4125 		break;
4126 	}
4127 	case KVM_SET_PIT2: {
4128 		r = -EFAULT;
4129 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4130 			goto out;
4131 		r = -ENXIO;
4132 		if (!kvm->arch.vpit)
4133 			goto out;
4134 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4135 		break;
4136 	}
4137 	case KVM_REINJECT_CONTROL: {
4138 		struct kvm_reinject_control control;
4139 		r =  -EFAULT;
4140 		if (copy_from_user(&control, argp, sizeof(control)))
4141 			goto out;
4142 		r = kvm_vm_ioctl_reinject(kvm, &control);
4143 		break;
4144 	}
4145 	case KVM_SET_BOOT_CPU_ID:
4146 		r = 0;
4147 		mutex_lock(&kvm->lock);
4148 		if (kvm->created_vcpus)
4149 			r = -EBUSY;
4150 		else
4151 			kvm->arch.bsp_vcpu_id = arg;
4152 		mutex_unlock(&kvm->lock);
4153 		break;
4154 	case KVM_XEN_HVM_CONFIG: {
4155 		r = -EFAULT;
4156 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4157 				   sizeof(struct kvm_xen_hvm_config)))
4158 			goto out;
4159 		r = -EINVAL;
4160 		if (kvm->arch.xen_hvm_config.flags)
4161 			goto out;
4162 		r = 0;
4163 		break;
4164 	}
4165 	case KVM_SET_CLOCK: {
4166 		struct kvm_clock_data user_ns;
4167 		u64 now_ns;
4168 
4169 		r = -EFAULT;
4170 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4171 			goto out;
4172 
4173 		r = -EINVAL;
4174 		if (user_ns.flags)
4175 			goto out;
4176 
4177 		r = 0;
4178 		now_ns = get_kvmclock_ns(kvm);
4179 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4180 		kvm_gen_update_masterclock(kvm);
4181 		break;
4182 	}
4183 	case KVM_GET_CLOCK: {
4184 		struct kvm_clock_data user_ns;
4185 		u64 now_ns;
4186 
4187 		now_ns = get_kvmclock_ns(kvm);
4188 		user_ns.clock = now_ns;
4189 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4190 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4191 
4192 		r = -EFAULT;
4193 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4194 			goto out;
4195 		r = 0;
4196 		break;
4197 	}
4198 	case KVM_ENABLE_CAP: {
4199 		struct kvm_enable_cap cap;
4200 
4201 		r = -EFAULT;
4202 		if (copy_from_user(&cap, argp, sizeof(cap)))
4203 			goto out;
4204 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4205 		break;
4206 	}
4207 	default:
4208 		r = -ENOTTY;
4209 	}
4210 out:
4211 	return r;
4212 }
4213 
4214 static void kvm_init_msr_list(void)
4215 {
4216 	u32 dummy[2];
4217 	unsigned i, j;
4218 
4219 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4220 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4221 			continue;
4222 
4223 		/*
4224 		 * Even MSRs that are valid in the host may not be exposed
4225 		 * to the guests in some cases.
4226 		 */
4227 		switch (msrs_to_save[i]) {
4228 		case MSR_IA32_BNDCFGS:
4229 			if (!kvm_x86_ops->mpx_supported())
4230 				continue;
4231 			break;
4232 		case MSR_TSC_AUX:
4233 			if (!kvm_x86_ops->rdtscp_supported())
4234 				continue;
4235 			break;
4236 		default:
4237 			break;
4238 		}
4239 
4240 		if (j < i)
4241 			msrs_to_save[j] = msrs_to_save[i];
4242 		j++;
4243 	}
4244 	num_msrs_to_save = j;
4245 
4246 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4247 		switch (emulated_msrs[i]) {
4248 		case MSR_IA32_SMBASE:
4249 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4250 				continue;
4251 			break;
4252 		default:
4253 			break;
4254 		}
4255 
4256 		if (j < i)
4257 			emulated_msrs[j] = emulated_msrs[i];
4258 		j++;
4259 	}
4260 	num_emulated_msrs = j;
4261 }
4262 
4263 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4264 			   const void *v)
4265 {
4266 	int handled = 0;
4267 	int n;
4268 
4269 	do {
4270 		n = min(len, 8);
4271 		if (!(lapic_in_kernel(vcpu) &&
4272 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4273 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4274 			break;
4275 		handled += n;
4276 		addr += n;
4277 		len -= n;
4278 		v += n;
4279 	} while (len);
4280 
4281 	return handled;
4282 }
4283 
4284 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4285 {
4286 	int handled = 0;
4287 	int n;
4288 
4289 	do {
4290 		n = min(len, 8);
4291 		if (!(lapic_in_kernel(vcpu) &&
4292 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4293 					 addr, n, v))
4294 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4295 			break;
4296 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4297 		handled += n;
4298 		addr += n;
4299 		len -= n;
4300 		v += n;
4301 	} while (len);
4302 
4303 	return handled;
4304 }
4305 
4306 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4307 			struct kvm_segment *var, int seg)
4308 {
4309 	kvm_x86_ops->set_segment(vcpu, var, seg);
4310 }
4311 
4312 void kvm_get_segment(struct kvm_vcpu *vcpu,
4313 		     struct kvm_segment *var, int seg)
4314 {
4315 	kvm_x86_ops->get_segment(vcpu, var, seg);
4316 }
4317 
4318 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4319 			   struct x86_exception *exception)
4320 {
4321 	gpa_t t_gpa;
4322 
4323 	BUG_ON(!mmu_is_nested(vcpu));
4324 
4325 	/* NPT walks are always user-walks */
4326 	access |= PFERR_USER_MASK;
4327 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4328 
4329 	return t_gpa;
4330 }
4331 
4332 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4333 			      struct x86_exception *exception)
4334 {
4335 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4336 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4337 }
4338 
4339  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4340 				struct x86_exception *exception)
4341 {
4342 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4343 	access |= PFERR_FETCH_MASK;
4344 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4345 }
4346 
4347 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4348 			       struct x86_exception *exception)
4349 {
4350 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4351 	access |= PFERR_WRITE_MASK;
4352 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4353 }
4354 
4355 /* uses this to access any guest's mapped memory without checking CPL */
4356 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4357 				struct x86_exception *exception)
4358 {
4359 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4360 }
4361 
4362 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4363 				      struct kvm_vcpu *vcpu, u32 access,
4364 				      struct x86_exception *exception)
4365 {
4366 	void *data = val;
4367 	int r = X86EMUL_CONTINUE;
4368 
4369 	while (bytes) {
4370 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4371 							    exception);
4372 		unsigned offset = addr & (PAGE_SIZE-1);
4373 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4374 		int ret;
4375 
4376 		if (gpa == UNMAPPED_GVA)
4377 			return X86EMUL_PROPAGATE_FAULT;
4378 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4379 					       offset, toread);
4380 		if (ret < 0) {
4381 			r = X86EMUL_IO_NEEDED;
4382 			goto out;
4383 		}
4384 
4385 		bytes -= toread;
4386 		data += toread;
4387 		addr += toread;
4388 	}
4389 out:
4390 	return r;
4391 }
4392 
4393 /* used for instruction fetching */
4394 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4395 				gva_t addr, void *val, unsigned int bytes,
4396 				struct x86_exception *exception)
4397 {
4398 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4399 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4400 	unsigned offset;
4401 	int ret;
4402 
4403 	/* Inline kvm_read_guest_virt_helper for speed.  */
4404 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4405 						    exception);
4406 	if (unlikely(gpa == UNMAPPED_GVA))
4407 		return X86EMUL_PROPAGATE_FAULT;
4408 
4409 	offset = addr & (PAGE_SIZE-1);
4410 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4411 		bytes = (unsigned)PAGE_SIZE - offset;
4412 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4413 				       offset, bytes);
4414 	if (unlikely(ret < 0))
4415 		return X86EMUL_IO_NEEDED;
4416 
4417 	return X86EMUL_CONTINUE;
4418 }
4419 
4420 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4421 			       gva_t addr, void *val, unsigned int bytes,
4422 			       struct x86_exception *exception)
4423 {
4424 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4425 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4426 
4427 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4428 					  exception);
4429 }
4430 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4431 
4432 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4433 				      gva_t addr, void *val, unsigned int bytes,
4434 				      struct x86_exception *exception)
4435 {
4436 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4437 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4438 }
4439 
4440 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4441 		unsigned long addr, void *val, unsigned int bytes)
4442 {
4443 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4444 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4445 
4446 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4447 }
4448 
4449 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4450 				       gva_t addr, void *val,
4451 				       unsigned int bytes,
4452 				       struct x86_exception *exception)
4453 {
4454 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4455 	void *data = val;
4456 	int r = X86EMUL_CONTINUE;
4457 
4458 	while (bytes) {
4459 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4460 							     PFERR_WRITE_MASK,
4461 							     exception);
4462 		unsigned offset = addr & (PAGE_SIZE-1);
4463 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4464 		int ret;
4465 
4466 		if (gpa == UNMAPPED_GVA)
4467 			return X86EMUL_PROPAGATE_FAULT;
4468 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4469 		if (ret < 0) {
4470 			r = X86EMUL_IO_NEEDED;
4471 			goto out;
4472 		}
4473 
4474 		bytes -= towrite;
4475 		data += towrite;
4476 		addr += towrite;
4477 	}
4478 out:
4479 	return r;
4480 }
4481 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4482 
4483 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4484 			    gpa_t gpa, bool write)
4485 {
4486 	/* For APIC access vmexit */
4487 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4488 		return 1;
4489 
4490 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4491 		trace_vcpu_match_mmio(gva, gpa, write, true);
4492 		return 1;
4493 	}
4494 
4495 	return 0;
4496 }
4497 
4498 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4499 				gpa_t *gpa, struct x86_exception *exception,
4500 				bool write)
4501 {
4502 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4503 		| (write ? PFERR_WRITE_MASK : 0);
4504 
4505 	/*
4506 	 * currently PKRU is only applied to ept enabled guest so
4507 	 * there is no pkey in EPT page table for L1 guest or EPT
4508 	 * shadow page table for L2 guest.
4509 	 */
4510 	if (vcpu_match_mmio_gva(vcpu, gva)
4511 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4512 				 vcpu->arch.access, 0, access)) {
4513 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4514 					(gva & (PAGE_SIZE - 1));
4515 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4516 		return 1;
4517 	}
4518 
4519 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4520 
4521 	if (*gpa == UNMAPPED_GVA)
4522 		return -1;
4523 
4524 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4525 }
4526 
4527 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4528 			const void *val, int bytes)
4529 {
4530 	int ret;
4531 
4532 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4533 	if (ret < 0)
4534 		return 0;
4535 	kvm_page_track_write(vcpu, gpa, val, bytes);
4536 	return 1;
4537 }
4538 
4539 struct read_write_emulator_ops {
4540 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4541 				  int bytes);
4542 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4543 				  void *val, int bytes);
4544 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4545 			       int bytes, void *val);
4546 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4547 				    void *val, int bytes);
4548 	bool write;
4549 };
4550 
4551 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4552 {
4553 	if (vcpu->mmio_read_completed) {
4554 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4555 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4556 		vcpu->mmio_read_completed = 0;
4557 		return 1;
4558 	}
4559 
4560 	return 0;
4561 }
4562 
4563 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4564 			void *val, int bytes)
4565 {
4566 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4567 }
4568 
4569 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4570 			 void *val, int bytes)
4571 {
4572 	return emulator_write_phys(vcpu, gpa, val, bytes);
4573 }
4574 
4575 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4576 {
4577 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4578 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4579 }
4580 
4581 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4582 			  void *val, int bytes)
4583 {
4584 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4585 	return X86EMUL_IO_NEEDED;
4586 }
4587 
4588 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4589 			   void *val, int bytes)
4590 {
4591 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4592 
4593 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4594 	return X86EMUL_CONTINUE;
4595 }
4596 
4597 static const struct read_write_emulator_ops read_emultor = {
4598 	.read_write_prepare = read_prepare,
4599 	.read_write_emulate = read_emulate,
4600 	.read_write_mmio = vcpu_mmio_read,
4601 	.read_write_exit_mmio = read_exit_mmio,
4602 };
4603 
4604 static const struct read_write_emulator_ops write_emultor = {
4605 	.read_write_emulate = write_emulate,
4606 	.read_write_mmio = write_mmio,
4607 	.read_write_exit_mmio = write_exit_mmio,
4608 	.write = true,
4609 };
4610 
4611 static int emulator_read_write_onepage(unsigned long addr, void *val,
4612 				       unsigned int bytes,
4613 				       struct x86_exception *exception,
4614 				       struct kvm_vcpu *vcpu,
4615 				       const struct read_write_emulator_ops *ops)
4616 {
4617 	gpa_t gpa;
4618 	int handled, ret;
4619 	bool write = ops->write;
4620 	struct kvm_mmio_fragment *frag;
4621 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4622 
4623 	/*
4624 	 * If the exit was due to a NPF we may already have a GPA.
4625 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4626 	 * Note, this cannot be used on string operations since string
4627 	 * operation using rep will only have the initial GPA from the NPF
4628 	 * occurred.
4629 	 */
4630 	if (vcpu->arch.gpa_available &&
4631 	    emulator_can_use_gpa(ctxt) &&
4632 	    vcpu_is_mmio_gpa(vcpu, addr, exception->address, write) &&
4633 	    (addr & ~PAGE_MASK) == (exception->address & ~PAGE_MASK)) {
4634 		gpa = exception->address;
4635 		goto mmio;
4636 	}
4637 
4638 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4639 
4640 	if (ret < 0)
4641 		return X86EMUL_PROPAGATE_FAULT;
4642 
4643 	/* For APIC access vmexit */
4644 	if (ret)
4645 		goto mmio;
4646 
4647 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4648 		return X86EMUL_CONTINUE;
4649 
4650 mmio:
4651 	/*
4652 	 * Is this MMIO handled locally?
4653 	 */
4654 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4655 	if (handled == bytes)
4656 		return X86EMUL_CONTINUE;
4657 
4658 	gpa += handled;
4659 	bytes -= handled;
4660 	val += handled;
4661 
4662 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4663 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4664 	frag->gpa = gpa;
4665 	frag->data = val;
4666 	frag->len = bytes;
4667 	return X86EMUL_CONTINUE;
4668 }
4669 
4670 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4671 			unsigned long addr,
4672 			void *val, unsigned int bytes,
4673 			struct x86_exception *exception,
4674 			const struct read_write_emulator_ops *ops)
4675 {
4676 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4677 	gpa_t gpa;
4678 	int rc;
4679 
4680 	if (ops->read_write_prepare &&
4681 		  ops->read_write_prepare(vcpu, val, bytes))
4682 		return X86EMUL_CONTINUE;
4683 
4684 	vcpu->mmio_nr_fragments = 0;
4685 
4686 	/* Crossing a page boundary? */
4687 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4688 		int now;
4689 
4690 		now = -addr & ~PAGE_MASK;
4691 		rc = emulator_read_write_onepage(addr, val, now, exception,
4692 						 vcpu, ops);
4693 
4694 		if (rc != X86EMUL_CONTINUE)
4695 			return rc;
4696 		addr += now;
4697 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4698 			addr = (u32)addr;
4699 		val += now;
4700 		bytes -= now;
4701 	}
4702 
4703 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4704 					 vcpu, ops);
4705 	if (rc != X86EMUL_CONTINUE)
4706 		return rc;
4707 
4708 	if (!vcpu->mmio_nr_fragments)
4709 		return rc;
4710 
4711 	gpa = vcpu->mmio_fragments[0].gpa;
4712 
4713 	vcpu->mmio_needed = 1;
4714 	vcpu->mmio_cur_fragment = 0;
4715 
4716 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4717 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4718 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4719 	vcpu->run->mmio.phys_addr = gpa;
4720 
4721 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4722 }
4723 
4724 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4725 				  unsigned long addr,
4726 				  void *val,
4727 				  unsigned int bytes,
4728 				  struct x86_exception *exception)
4729 {
4730 	return emulator_read_write(ctxt, addr, val, bytes,
4731 				   exception, &read_emultor);
4732 }
4733 
4734 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4735 			    unsigned long addr,
4736 			    const void *val,
4737 			    unsigned int bytes,
4738 			    struct x86_exception *exception)
4739 {
4740 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4741 				   exception, &write_emultor);
4742 }
4743 
4744 #define CMPXCHG_TYPE(t, ptr, old, new) \
4745 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4746 
4747 #ifdef CONFIG_X86_64
4748 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4749 #else
4750 #  define CMPXCHG64(ptr, old, new) \
4751 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4752 #endif
4753 
4754 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4755 				     unsigned long addr,
4756 				     const void *old,
4757 				     const void *new,
4758 				     unsigned int bytes,
4759 				     struct x86_exception *exception)
4760 {
4761 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4762 	gpa_t gpa;
4763 	struct page *page;
4764 	char *kaddr;
4765 	bool exchanged;
4766 
4767 	/* guests cmpxchg8b have to be emulated atomically */
4768 	if (bytes > 8 || (bytes & (bytes - 1)))
4769 		goto emul_write;
4770 
4771 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4772 
4773 	if (gpa == UNMAPPED_GVA ||
4774 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4775 		goto emul_write;
4776 
4777 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4778 		goto emul_write;
4779 
4780 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4781 	if (is_error_page(page))
4782 		goto emul_write;
4783 
4784 	kaddr = kmap_atomic(page);
4785 	kaddr += offset_in_page(gpa);
4786 	switch (bytes) {
4787 	case 1:
4788 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4789 		break;
4790 	case 2:
4791 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4792 		break;
4793 	case 4:
4794 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4795 		break;
4796 	case 8:
4797 		exchanged = CMPXCHG64(kaddr, old, new);
4798 		break;
4799 	default:
4800 		BUG();
4801 	}
4802 	kunmap_atomic(kaddr);
4803 	kvm_release_page_dirty(page);
4804 
4805 	if (!exchanged)
4806 		return X86EMUL_CMPXCHG_FAILED;
4807 
4808 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4809 	kvm_page_track_write(vcpu, gpa, new, bytes);
4810 
4811 	return X86EMUL_CONTINUE;
4812 
4813 emul_write:
4814 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4815 
4816 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4817 }
4818 
4819 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4820 {
4821 	/* TODO: String I/O for in kernel device */
4822 	int r;
4823 
4824 	if (vcpu->arch.pio.in)
4825 		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4826 				    vcpu->arch.pio.size, pd);
4827 	else
4828 		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4829 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4830 				     pd);
4831 	return r;
4832 }
4833 
4834 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4835 			       unsigned short port, void *val,
4836 			       unsigned int count, bool in)
4837 {
4838 	vcpu->arch.pio.port = port;
4839 	vcpu->arch.pio.in = in;
4840 	vcpu->arch.pio.count  = count;
4841 	vcpu->arch.pio.size = size;
4842 
4843 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4844 		vcpu->arch.pio.count = 0;
4845 		return 1;
4846 	}
4847 
4848 	vcpu->run->exit_reason = KVM_EXIT_IO;
4849 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4850 	vcpu->run->io.size = size;
4851 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4852 	vcpu->run->io.count = count;
4853 	vcpu->run->io.port = port;
4854 
4855 	return 0;
4856 }
4857 
4858 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4859 				    int size, unsigned short port, void *val,
4860 				    unsigned int count)
4861 {
4862 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4863 	int ret;
4864 
4865 	if (vcpu->arch.pio.count)
4866 		goto data_avail;
4867 
4868 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4869 	if (ret) {
4870 data_avail:
4871 		memcpy(val, vcpu->arch.pio_data, size * count);
4872 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4873 		vcpu->arch.pio.count = 0;
4874 		return 1;
4875 	}
4876 
4877 	return 0;
4878 }
4879 
4880 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4881 				     int size, unsigned short port,
4882 				     const void *val, unsigned int count)
4883 {
4884 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4885 
4886 	memcpy(vcpu->arch.pio_data, val, size * count);
4887 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4888 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4889 }
4890 
4891 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4892 {
4893 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4894 }
4895 
4896 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4897 {
4898 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4899 }
4900 
4901 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4902 {
4903 	if (!need_emulate_wbinvd(vcpu))
4904 		return X86EMUL_CONTINUE;
4905 
4906 	if (kvm_x86_ops->has_wbinvd_exit()) {
4907 		int cpu = get_cpu();
4908 
4909 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4910 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4911 				wbinvd_ipi, NULL, 1);
4912 		put_cpu();
4913 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4914 	} else
4915 		wbinvd();
4916 	return X86EMUL_CONTINUE;
4917 }
4918 
4919 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4920 {
4921 	kvm_emulate_wbinvd_noskip(vcpu);
4922 	return kvm_skip_emulated_instruction(vcpu);
4923 }
4924 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4925 
4926 
4927 
4928 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4929 {
4930 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4931 }
4932 
4933 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4934 			   unsigned long *dest)
4935 {
4936 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4937 }
4938 
4939 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4940 			   unsigned long value)
4941 {
4942 
4943 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4944 }
4945 
4946 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4947 {
4948 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4949 }
4950 
4951 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4952 {
4953 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4954 	unsigned long value;
4955 
4956 	switch (cr) {
4957 	case 0:
4958 		value = kvm_read_cr0(vcpu);
4959 		break;
4960 	case 2:
4961 		value = vcpu->arch.cr2;
4962 		break;
4963 	case 3:
4964 		value = kvm_read_cr3(vcpu);
4965 		break;
4966 	case 4:
4967 		value = kvm_read_cr4(vcpu);
4968 		break;
4969 	case 8:
4970 		value = kvm_get_cr8(vcpu);
4971 		break;
4972 	default:
4973 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4974 		return 0;
4975 	}
4976 
4977 	return value;
4978 }
4979 
4980 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4981 {
4982 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4983 	int res = 0;
4984 
4985 	switch (cr) {
4986 	case 0:
4987 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4988 		break;
4989 	case 2:
4990 		vcpu->arch.cr2 = val;
4991 		break;
4992 	case 3:
4993 		res = kvm_set_cr3(vcpu, val);
4994 		break;
4995 	case 4:
4996 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4997 		break;
4998 	case 8:
4999 		res = kvm_set_cr8(vcpu, val);
5000 		break;
5001 	default:
5002 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5003 		res = -1;
5004 	}
5005 
5006 	return res;
5007 }
5008 
5009 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5010 {
5011 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5012 }
5013 
5014 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5015 {
5016 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5017 }
5018 
5019 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5020 {
5021 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5022 }
5023 
5024 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5025 {
5026 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5027 }
5028 
5029 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5030 {
5031 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5032 }
5033 
5034 static unsigned long emulator_get_cached_segment_base(
5035 	struct x86_emulate_ctxt *ctxt, int seg)
5036 {
5037 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5038 }
5039 
5040 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5041 				 struct desc_struct *desc, u32 *base3,
5042 				 int seg)
5043 {
5044 	struct kvm_segment var;
5045 
5046 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5047 	*selector = var.selector;
5048 
5049 	if (var.unusable) {
5050 		memset(desc, 0, sizeof(*desc));
5051 		return false;
5052 	}
5053 
5054 	if (var.g)
5055 		var.limit >>= 12;
5056 	set_desc_limit(desc, var.limit);
5057 	set_desc_base(desc, (unsigned long)var.base);
5058 #ifdef CONFIG_X86_64
5059 	if (base3)
5060 		*base3 = var.base >> 32;
5061 #endif
5062 	desc->type = var.type;
5063 	desc->s = var.s;
5064 	desc->dpl = var.dpl;
5065 	desc->p = var.present;
5066 	desc->avl = var.avl;
5067 	desc->l = var.l;
5068 	desc->d = var.db;
5069 	desc->g = var.g;
5070 
5071 	return true;
5072 }
5073 
5074 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5075 				 struct desc_struct *desc, u32 base3,
5076 				 int seg)
5077 {
5078 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5079 	struct kvm_segment var;
5080 
5081 	var.selector = selector;
5082 	var.base = get_desc_base(desc);
5083 #ifdef CONFIG_X86_64
5084 	var.base |= ((u64)base3) << 32;
5085 #endif
5086 	var.limit = get_desc_limit(desc);
5087 	if (desc->g)
5088 		var.limit = (var.limit << 12) | 0xfff;
5089 	var.type = desc->type;
5090 	var.dpl = desc->dpl;
5091 	var.db = desc->d;
5092 	var.s = desc->s;
5093 	var.l = desc->l;
5094 	var.g = desc->g;
5095 	var.avl = desc->avl;
5096 	var.present = desc->p;
5097 	var.unusable = !var.present;
5098 	var.padding = 0;
5099 
5100 	kvm_set_segment(vcpu, &var, seg);
5101 	return;
5102 }
5103 
5104 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5105 			    u32 msr_index, u64 *pdata)
5106 {
5107 	struct msr_data msr;
5108 	int r;
5109 
5110 	msr.index = msr_index;
5111 	msr.host_initiated = false;
5112 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5113 	if (r)
5114 		return r;
5115 
5116 	*pdata = msr.data;
5117 	return 0;
5118 }
5119 
5120 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5121 			    u32 msr_index, u64 data)
5122 {
5123 	struct msr_data msr;
5124 
5125 	msr.data = data;
5126 	msr.index = msr_index;
5127 	msr.host_initiated = false;
5128 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5129 }
5130 
5131 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5132 {
5133 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5134 
5135 	return vcpu->arch.smbase;
5136 }
5137 
5138 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5139 {
5140 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5141 
5142 	vcpu->arch.smbase = smbase;
5143 }
5144 
5145 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5146 			      u32 pmc)
5147 {
5148 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5149 }
5150 
5151 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5152 			     u32 pmc, u64 *pdata)
5153 {
5154 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5155 }
5156 
5157 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5158 {
5159 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5160 }
5161 
5162 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5163 {
5164 	preempt_disable();
5165 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5166 }
5167 
5168 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5169 {
5170 	preempt_enable();
5171 }
5172 
5173 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5174 			      struct x86_instruction_info *info,
5175 			      enum x86_intercept_stage stage)
5176 {
5177 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5178 }
5179 
5180 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5181 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5182 {
5183 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5184 }
5185 
5186 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5187 {
5188 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5189 }
5190 
5191 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5192 {
5193 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5194 }
5195 
5196 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5197 {
5198 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5199 }
5200 
5201 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5202 {
5203 	return emul_to_vcpu(ctxt)->arch.hflags;
5204 }
5205 
5206 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5207 {
5208 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5209 }
5210 
5211 static const struct x86_emulate_ops emulate_ops = {
5212 	.read_gpr            = emulator_read_gpr,
5213 	.write_gpr           = emulator_write_gpr,
5214 	.read_std            = kvm_read_guest_virt_system,
5215 	.write_std           = kvm_write_guest_virt_system,
5216 	.read_phys           = kvm_read_guest_phys_system,
5217 	.fetch               = kvm_fetch_guest_virt,
5218 	.read_emulated       = emulator_read_emulated,
5219 	.write_emulated      = emulator_write_emulated,
5220 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5221 	.invlpg              = emulator_invlpg,
5222 	.pio_in_emulated     = emulator_pio_in_emulated,
5223 	.pio_out_emulated    = emulator_pio_out_emulated,
5224 	.get_segment         = emulator_get_segment,
5225 	.set_segment         = emulator_set_segment,
5226 	.get_cached_segment_base = emulator_get_cached_segment_base,
5227 	.get_gdt             = emulator_get_gdt,
5228 	.get_idt	     = emulator_get_idt,
5229 	.set_gdt             = emulator_set_gdt,
5230 	.set_idt	     = emulator_set_idt,
5231 	.get_cr              = emulator_get_cr,
5232 	.set_cr              = emulator_set_cr,
5233 	.cpl                 = emulator_get_cpl,
5234 	.get_dr              = emulator_get_dr,
5235 	.set_dr              = emulator_set_dr,
5236 	.get_smbase          = emulator_get_smbase,
5237 	.set_smbase          = emulator_set_smbase,
5238 	.set_msr             = emulator_set_msr,
5239 	.get_msr             = emulator_get_msr,
5240 	.check_pmc	     = emulator_check_pmc,
5241 	.read_pmc            = emulator_read_pmc,
5242 	.halt                = emulator_halt,
5243 	.wbinvd              = emulator_wbinvd,
5244 	.fix_hypercall       = emulator_fix_hypercall,
5245 	.get_fpu             = emulator_get_fpu,
5246 	.put_fpu             = emulator_put_fpu,
5247 	.intercept           = emulator_intercept,
5248 	.get_cpuid           = emulator_get_cpuid,
5249 	.set_nmi_mask        = emulator_set_nmi_mask,
5250 	.get_hflags          = emulator_get_hflags,
5251 	.set_hflags          = emulator_set_hflags,
5252 };
5253 
5254 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5255 {
5256 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5257 	/*
5258 	 * an sti; sti; sequence only disable interrupts for the first
5259 	 * instruction. So, if the last instruction, be it emulated or
5260 	 * not, left the system with the INT_STI flag enabled, it
5261 	 * means that the last instruction is an sti. We should not
5262 	 * leave the flag on in this case. The same goes for mov ss
5263 	 */
5264 	if (int_shadow & mask)
5265 		mask = 0;
5266 	if (unlikely(int_shadow || mask)) {
5267 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5268 		if (!mask)
5269 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5270 	}
5271 }
5272 
5273 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5274 {
5275 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5276 	if (ctxt->exception.vector == PF_VECTOR)
5277 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5278 
5279 	if (ctxt->exception.error_code_valid)
5280 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5281 				      ctxt->exception.error_code);
5282 	else
5283 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5284 	return false;
5285 }
5286 
5287 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5288 {
5289 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5290 	int cs_db, cs_l;
5291 
5292 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5293 
5294 	ctxt->eflags = kvm_get_rflags(vcpu);
5295 	ctxt->eip = kvm_rip_read(vcpu);
5296 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5297 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5298 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5299 		     cs_db				? X86EMUL_MODE_PROT32 :
5300 							  X86EMUL_MODE_PROT16;
5301 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5302 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5303 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5304 
5305 	init_decode_cache(ctxt);
5306 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5307 }
5308 
5309 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5310 {
5311 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5312 	int ret;
5313 
5314 	init_emulate_ctxt(vcpu);
5315 
5316 	ctxt->op_bytes = 2;
5317 	ctxt->ad_bytes = 2;
5318 	ctxt->_eip = ctxt->eip + inc_eip;
5319 	ret = emulate_int_real(ctxt, irq);
5320 
5321 	if (ret != X86EMUL_CONTINUE)
5322 		return EMULATE_FAIL;
5323 
5324 	ctxt->eip = ctxt->_eip;
5325 	kvm_rip_write(vcpu, ctxt->eip);
5326 	kvm_set_rflags(vcpu, ctxt->eflags);
5327 
5328 	if (irq == NMI_VECTOR)
5329 		vcpu->arch.nmi_pending = 0;
5330 	else
5331 		vcpu->arch.interrupt.pending = false;
5332 
5333 	return EMULATE_DONE;
5334 }
5335 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5336 
5337 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5338 {
5339 	int r = EMULATE_DONE;
5340 
5341 	++vcpu->stat.insn_emulation_fail;
5342 	trace_kvm_emulate_insn_failed(vcpu);
5343 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5344 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5345 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5346 		vcpu->run->internal.ndata = 0;
5347 		r = EMULATE_FAIL;
5348 	}
5349 	kvm_queue_exception(vcpu, UD_VECTOR);
5350 
5351 	return r;
5352 }
5353 
5354 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5355 				  bool write_fault_to_shadow_pgtable,
5356 				  int emulation_type)
5357 {
5358 	gpa_t gpa = cr2;
5359 	kvm_pfn_t pfn;
5360 
5361 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5362 		return false;
5363 
5364 	if (!vcpu->arch.mmu.direct_map) {
5365 		/*
5366 		 * Write permission should be allowed since only
5367 		 * write access need to be emulated.
5368 		 */
5369 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5370 
5371 		/*
5372 		 * If the mapping is invalid in guest, let cpu retry
5373 		 * it to generate fault.
5374 		 */
5375 		if (gpa == UNMAPPED_GVA)
5376 			return true;
5377 	}
5378 
5379 	/*
5380 	 * Do not retry the unhandleable instruction if it faults on the
5381 	 * readonly host memory, otherwise it will goto a infinite loop:
5382 	 * retry instruction -> write #PF -> emulation fail -> retry
5383 	 * instruction -> ...
5384 	 */
5385 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5386 
5387 	/*
5388 	 * If the instruction failed on the error pfn, it can not be fixed,
5389 	 * report the error to userspace.
5390 	 */
5391 	if (is_error_noslot_pfn(pfn))
5392 		return false;
5393 
5394 	kvm_release_pfn_clean(pfn);
5395 
5396 	/* The instructions are well-emulated on direct mmu. */
5397 	if (vcpu->arch.mmu.direct_map) {
5398 		unsigned int indirect_shadow_pages;
5399 
5400 		spin_lock(&vcpu->kvm->mmu_lock);
5401 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5402 		spin_unlock(&vcpu->kvm->mmu_lock);
5403 
5404 		if (indirect_shadow_pages)
5405 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5406 
5407 		return true;
5408 	}
5409 
5410 	/*
5411 	 * if emulation was due to access to shadowed page table
5412 	 * and it failed try to unshadow page and re-enter the
5413 	 * guest to let CPU execute the instruction.
5414 	 */
5415 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5416 
5417 	/*
5418 	 * If the access faults on its page table, it can not
5419 	 * be fixed by unprotecting shadow page and it should
5420 	 * be reported to userspace.
5421 	 */
5422 	return !write_fault_to_shadow_pgtable;
5423 }
5424 
5425 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5426 			      unsigned long cr2,  int emulation_type)
5427 {
5428 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5429 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5430 
5431 	last_retry_eip = vcpu->arch.last_retry_eip;
5432 	last_retry_addr = vcpu->arch.last_retry_addr;
5433 
5434 	/*
5435 	 * If the emulation is caused by #PF and it is non-page_table
5436 	 * writing instruction, it means the VM-EXIT is caused by shadow
5437 	 * page protected, we can zap the shadow page and retry this
5438 	 * instruction directly.
5439 	 *
5440 	 * Note: if the guest uses a non-page-table modifying instruction
5441 	 * on the PDE that points to the instruction, then we will unmap
5442 	 * the instruction and go to an infinite loop. So, we cache the
5443 	 * last retried eip and the last fault address, if we meet the eip
5444 	 * and the address again, we can break out of the potential infinite
5445 	 * loop.
5446 	 */
5447 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5448 
5449 	if (!(emulation_type & EMULTYPE_RETRY))
5450 		return false;
5451 
5452 	if (x86_page_table_writing_insn(ctxt))
5453 		return false;
5454 
5455 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5456 		return false;
5457 
5458 	vcpu->arch.last_retry_eip = ctxt->eip;
5459 	vcpu->arch.last_retry_addr = cr2;
5460 
5461 	if (!vcpu->arch.mmu.direct_map)
5462 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5463 
5464 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5465 
5466 	return true;
5467 }
5468 
5469 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5470 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5471 
5472 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5473 {
5474 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5475 		/* This is a good place to trace that we are exiting SMM.  */
5476 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5477 
5478 		/* Process a latched INIT or SMI, if any.  */
5479 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5480 	}
5481 
5482 	kvm_mmu_reset_context(vcpu);
5483 }
5484 
5485 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5486 {
5487 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5488 
5489 	vcpu->arch.hflags = emul_flags;
5490 
5491 	if (changed & HF_SMM_MASK)
5492 		kvm_smm_changed(vcpu);
5493 }
5494 
5495 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5496 				unsigned long *db)
5497 {
5498 	u32 dr6 = 0;
5499 	int i;
5500 	u32 enable, rwlen;
5501 
5502 	enable = dr7;
5503 	rwlen = dr7 >> 16;
5504 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5505 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5506 			dr6 |= (1 << i);
5507 	return dr6;
5508 }
5509 
5510 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5511 {
5512 	struct kvm_run *kvm_run = vcpu->run;
5513 
5514 	/*
5515 	 * rflags is the old, "raw" value of the flags.  The new value has
5516 	 * not been saved yet.
5517 	 *
5518 	 * This is correct even for TF set by the guest, because "the
5519 	 * processor will not generate this exception after the instruction
5520 	 * that sets the TF flag".
5521 	 */
5522 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5523 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5524 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5525 						  DR6_RTM;
5526 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5527 			kvm_run->debug.arch.exception = DB_VECTOR;
5528 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5529 			*r = EMULATE_USER_EXIT;
5530 		} else {
5531 			/*
5532 			 * "Certain debug exceptions may clear bit 0-3.  The
5533 			 * remaining contents of the DR6 register are never
5534 			 * cleared by the processor".
5535 			 */
5536 			vcpu->arch.dr6 &= ~15;
5537 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5538 			kvm_queue_exception(vcpu, DB_VECTOR);
5539 		}
5540 	}
5541 }
5542 
5543 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5544 {
5545 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5546 	int r = EMULATE_DONE;
5547 
5548 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5549 	kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5550 	return r == EMULATE_DONE;
5551 }
5552 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5553 
5554 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5555 {
5556 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5557 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5558 		struct kvm_run *kvm_run = vcpu->run;
5559 		unsigned long eip = kvm_get_linear_rip(vcpu);
5560 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5561 					   vcpu->arch.guest_debug_dr7,
5562 					   vcpu->arch.eff_db);
5563 
5564 		if (dr6 != 0) {
5565 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5566 			kvm_run->debug.arch.pc = eip;
5567 			kvm_run->debug.arch.exception = DB_VECTOR;
5568 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5569 			*r = EMULATE_USER_EXIT;
5570 			return true;
5571 		}
5572 	}
5573 
5574 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5575 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5576 		unsigned long eip = kvm_get_linear_rip(vcpu);
5577 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5578 					   vcpu->arch.dr7,
5579 					   vcpu->arch.db);
5580 
5581 		if (dr6 != 0) {
5582 			vcpu->arch.dr6 &= ~15;
5583 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5584 			kvm_queue_exception(vcpu, DB_VECTOR);
5585 			*r = EMULATE_DONE;
5586 			return true;
5587 		}
5588 	}
5589 
5590 	return false;
5591 }
5592 
5593 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5594 			    unsigned long cr2,
5595 			    int emulation_type,
5596 			    void *insn,
5597 			    int insn_len)
5598 {
5599 	int r;
5600 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5601 	bool writeback = true;
5602 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5603 
5604 	/*
5605 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5606 	 * never reused.
5607 	 */
5608 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5609 	kvm_clear_exception_queue(vcpu);
5610 
5611 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5612 		init_emulate_ctxt(vcpu);
5613 
5614 		/*
5615 		 * We will reenter on the same instruction since
5616 		 * we do not set complete_userspace_io.  This does not
5617 		 * handle watchpoints yet, those would be handled in
5618 		 * the emulate_ops.
5619 		 */
5620 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5621 			return r;
5622 
5623 		ctxt->interruptibility = 0;
5624 		ctxt->have_exception = false;
5625 		ctxt->exception.vector = -1;
5626 		ctxt->perm_ok = false;
5627 
5628 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5629 
5630 		r = x86_decode_insn(ctxt, insn, insn_len);
5631 
5632 		trace_kvm_emulate_insn_start(vcpu);
5633 		++vcpu->stat.insn_emulation;
5634 		if (r != EMULATION_OK)  {
5635 			if (emulation_type & EMULTYPE_TRAP_UD)
5636 				return EMULATE_FAIL;
5637 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5638 						emulation_type))
5639 				return EMULATE_DONE;
5640 			if (emulation_type & EMULTYPE_SKIP)
5641 				return EMULATE_FAIL;
5642 			return handle_emulation_failure(vcpu);
5643 		}
5644 	}
5645 
5646 	if (emulation_type & EMULTYPE_SKIP) {
5647 		kvm_rip_write(vcpu, ctxt->_eip);
5648 		if (ctxt->eflags & X86_EFLAGS_RF)
5649 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5650 		return EMULATE_DONE;
5651 	}
5652 
5653 	if (retry_instruction(ctxt, cr2, emulation_type))
5654 		return EMULATE_DONE;
5655 
5656 	/* this is needed for vmware backdoor interface to work since it
5657 	   changes registers values  during IO operation */
5658 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5659 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5660 		emulator_invalidate_register_cache(ctxt);
5661 	}
5662 
5663 restart:
5664 	/* Save the faulting GPA (cr2) in the address field */
5665 	ctxt->exception.address = cr2;
5666 
5667 	r = x86_emulate_insn(ctxt);
5668 
5669 	if (r == EMULATION_INTERCEPTED)
5670 		return EMULATE_DONE;
5671 
5672 	if (r == EMULATION_FAILED) {
5673 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5674 					emulation_type))
5675 			return EMULATE_DONE;
5676 
5677 		return handle_emulation_failure(vcpu);
5678 	}
5679 
5680 	if (ctxt->have_exception) {
5681 		r = EMULATE_DONE;
5682 		if (inject_emulated_exception(vcpu))
5683 			return r;
5684 	} else if (vcpu->arch.pio.count) {
5685 		if (!vcpu->arch.pio.in) {
5686 			/* FIXME: return into emulator if single-stepping.  */
5687 			vcpu->arch.pio.count = 0;
5688 		} else {
5689 			writeback = false;
5690 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5691 		}
5692 		r = EMULATE_USER_EXIT;
5693 	} else if (vcpu->mmio_needed) {
5694 		if (!vcpu->mmio_is_write)
5695 			writeback = false;
5696 		r = EMULATE_USER_EXIT;
5697 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5698 	} else if (r == EMULATION_RESTART)
5699 		goto restart;
5700 	else
5701 		r = EMULATE_DONE;
5702 
5703 	if (writeback) {
5704 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5705 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5706 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5707 		kvm_rip_write(vcpu, ctxt->eip);
5708 		if (r == EMULATE_DONE)
5709 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5710 		if (!ctxt->have_exception ||
5711 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5712 			__kvm_set_rflags(vcpu, ctxt->eflags);
5713 
5714 		/*
5715 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5716 		 * do nothing, and it will be requested again as soon as
5717 		 * the shadow expires.  But we still need to check here,
5718 		 * because POPF has no interrupt shadow.
5719 		 */
5720 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5721 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5722 	} else
5723 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5724 
5725 	return r;
5726 }
5727 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5728 
5729 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5730 {
5731 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5732 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5733 					    size, port, &val, 1);
5734 	/* do not return to emulator after return from userspace */
5735 	vcpu->arch.pio.count = 0;
5736 	return ret;
5737 }
5738 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5739 
5740 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5741 {
5742 	unsigned long val;
5743 
5744 	/* We should only ever be called with arch.pio.count equal to 1 */
5745 	BUG_ON(vcpu->arch.pio.count != 1);
5746 
5747 	/* For size less than 4 we merge, else we zero extend */
5748 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5749 					: 0;
5750 
5751 	/*
5752 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5753 	 * the copy and tracing
5754 	 */
5755 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5756 				 vcpu->arch.pio.port, &val, 1);
5757 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5758 
5759 	return 1;
5760 }
5761 
5762 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5763 {
5764 	unsigned long val;
5765 	int ret;
5766 
5767 	/* For size less than 4 we merge, else we zero extend */
5768 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5769 
5770 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5771 				       &val, 1);
5772 	if (ret) {
5773 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5774 		return ret;
5775 	}
5776 
5777 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5778 
5779 	return 0;
5780 }
5781 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5782 
5783 static int kvmclock_cpu_down_prep(unsigned int cpu)
5784 {
5785 	__this_cpu_write(cpu_tsc_khz, 0);
5786 	return 0;
5787 }
5788 
5789 static void tsc_khz_changed(void *data)
5790 {
5791 	struct cpufreq_freqs *freq = data;
5792 	unsigned long khz = 0;
5793 
5794 	if (data)
5795 		khz = freq->new;
5796 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5797 		khz = cpufreq_quick_get(raw_smp_processor_id());
5798 	if (!khz)
5799 		khz = tsc_khz;
5800 	__this_cpu_write(cpu_tsc_khz, khz);
5801 }
5802 
5803 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5804 				     void *data)
5805 {
5806 	struct cpufreq_freqs *freq = data;
5807 	struct kvm *kvm;
5808 	struct kvm_vcpu *vcpu;
5809 	int i, send_ipi = 0;
5810 
5811 	/*
5812 	 * We allow guests to temporarily run on slowing clocks,
5813 	 * provided we notify them after, or to run on accelerating
5814 	 * clocks, provided we notify them before.  Thus time never
5815 	 * goes backwards.
5816 	 *
5817 	 * However, we have a problem.  We can't atomically update
5818 	 * the frequency of a given CPU from this function; it is
5819 	 * merely a notifier, which can be called from any CPU.
5820 	 * Changing the TSC frequency at arbitrary points in time
5821 	 * requires a recomputation of local variables related to
5822 	 * the TSC for each VCPU.  We must flag these local variables
5823 	 * to be updated and be sure the update takes place with the
5824 	 * new frequency before any guests proceed.
5825 	 *
5826 	 * Unfortunately, the combination of hotplug CPU and frequency
5827 	 * change creates an intractable locking scenario; the order
5828 	 * of when these callouts happen is undefined with respect to
5829 	 * CPU hotplug, and they can race with each other.  As such,
5830 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5831 	 * undefined; you can actually have a CPU frequency change take
5832 	 * place in between the computation of X and the setting of the
5833 	 * variable.  To protect against this problem, all updates of
5834 	 * the per_cpu tsc_khz variable are done in an interrupt
5835 	 * protected IPI, and all callers wishing to update the value
5836 	 * must wait for a synchronous IPI to complete (which is trivial
5837 	 * if the caller is on the CPU already).  This establishes the
5838 	 * necessary total order on variable updates.
5839 	 *
5840 	 * Note that because a guest time update may take place
5841 	 * anytime after the setting of the VCPU's request bit, the
5842 	 * correct TSC value must be set before the request.  However,
5843 	 * to ensure the update actually makes it to any guest which
5844 	 * starts running in hardware virtualization between the set
5845 	 * and the acquisition of the spinlock, we must also ping the
5846 	 * CPU after setting the request bit.
5847 	 *
5848 	 */
5849 
5850 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5851 		return 0;
5852 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5853 		return 0;
5854 
5855 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5856 
5857 	spin_lock(&kvm_lock);
5858 	list_for_each_entry(kvm, &vm_list, vm_list) {
5859 		kvm_for_each_vcpu(i, vcpu, kvm) {
5860 			if (vcpu->cpu != freq->cpu)
5861 				continue;
5862 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5863 			if (vcpu->cpu != smp_processor_id())
5864 				send_ipi = 1;
5865 		}
5866 	}
5867 	spin_unlock(&kvm_lock);
5868 
5869 	if (freq->old < freq->new && send_ipi) {
5870 		/*
5871 		 * We upscale the frequency.  Must make the guest
5872 		 * doesn't see old kvmclock values while running with
5873 		 * the new frequency, otherwise we risk the guest sees
5874 		 * time go backwards.
5875 		 *
5876 		 * In case we update the frequency for another cpu
5877 		 * (which might be in guest context) send an interrupt
5878 		 * to kick the cpu out of guest context.  Next time
5879 		 * guest context is entered kvmclock will be updated,
5880 		 * so the guest will not see stale values.
5881 		 */
5882 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5883 	}
5884 	return 0;
5885 }
5886 
5887 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5888 	.notifier_call  = kvmclock_cpufreq_notifier
5889 };
5890 
5891 static int kvmclock_cpu_online(unsigned int cpu)
5892 {
5893 	tsc_khz_changed(NULL);
5894 	return 0;
5895 }
5896 
5897 static void kvm_timer_init(void)
5898 {
5899 	max_tsc_khz = tsc_khz;
5900 
5901 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5902 #ifdef CONFIG_CPU_FREQ
5903 		struct cpufreq_policy policy;
5904 		int cpu;
5905 
5906 		memset(&policy, 0, sizeof(policy));
5907 		cpu = get_cpu();
5908 		cpufreq_get_policy(&policy, cpu);
5909 		if (policy.cpuinfo.max_freq)
5910 			max_tsc_khz = policy.cpuinfo.max_freq;
5911 		put_cpu();
5912 #endif
5913 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5914 					  CPUFREQ_TRANSITION_NOTIFIER);
5915 	}
5916 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5917 
5918 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5919 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
5920 }
5921 
5922 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5923 
5924 int kvm_is_in_guest(void)
5925 {
5926 	return __this_cpu_read(current_vcpu) != NULL;
5927 }
5928 
5929 static int kvm_is_user_mode(void)
5930 {
5931 	int user_mode = 3;
5932 
5933 	if (__this_cpu_read(current_vcpu))
5934 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5935 
5936 	return user_mode != 0;
5937 }
5938 
5939 static unsigned long kvm_get_guest_ip(void)
5940 {
5941 	unsigned long ip = 0;
5942 
5943 	if (__this_cpu_read(current_vcpu))
5944 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5945 
5946 	return ip;
5947 }
5948 
5949 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5950 	.is_in_guest		= kvm_is_in_guest,
5951 	.is_user_mode		= kvm_is_user_mode,
5952 	.get_guest_ip		= kvm_get_guest_ip,
5953 };
5954 
5955 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5956 {
5957 	__this_cpu_write(current_vcpu, vcpu);
5958 }
5959 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5960 
5961 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5962 {
5963 	__this_cpu_write(current_vcpu, NULL);
5964 }
5965 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5966 
5967 static void kvm_set_mmio_spte_mask(void)
5968 {
5969 	u64 mask;
5970 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5971 
5972 	/*
5973 	 * Set the reserved bits and the present bit of an paging-structure
5974 	 * entry to generate page fault with PFER.RSV = 1.
5975 	 */
5976 	 /* Mask the reserved physical address bits. */
5977 	mask = rsvd_bits(maxphyaddr, 51);
5978 
5979 	/* Set the present bit. */
5980 	mask |= 1ull;
5981 
5982 #ifdef CONFIG_X86_64
5983 	/*
5984 	 * If reserved bit is not supported, clear the present bit to disable
5985 	 * mmio page fault.
5986 	 */
5987 	if (maxphyaddr == 52)
5988 		mask &= ~1ull;
5989 #endif
5990 
5991 	kvm_mmu_set_mmio_spte_mask(mask);
5992 }
5993 
5994 #ifdef CONFIG_X86_64
5995 static void pvclock_gtod_update_fn(struct work_struct *work)
5996 {
5997 	struct kvm *kvm;
5998 
5999 	struct kvm_vcpu *vcpu;
6000 	int i;
6001 
6002 	spin_lock(&kvm_lock);
6003 	list_for_each_entry(kvm, &vm_list, vm_list)
6004 		kvm_for_each_vcpu(i, vcpu, kvm)
6005 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6006 	atomic_set(&kvm_guest_has_master_clock, 0);
6007 	spin_unlock(&kvm_lock);
6008 }
6009 
6010 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6011 
6012 /*
6013  * Notification about pvclock gtod data update.
6014  */
6015 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6016 			       void *priv)
6017 {
6018 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6019 	struct timekeeper *tk = priv;
6020 
6021 	update_pvclock_gtod(tk);
6022 
6023 	/* disable master clock if host does not trust, or does not
6024 	 * use, TSC clocksource
6025 	 */
6026 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6027 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6028 		queue_work(system_long_wq, &pvclock_gtod_work);
6029 
6030 	return 0;
6031 }
6032 
6033 static struct notifier_block pvclock_gtod_notifier = {
6034 	.notifier_call = pvclock_gtod_notify,
6035 };
6036 #endif
6037 
6038 int kvm_arch_init(void *opaque)
6039 {
6040 	int r;
6041 	struct kvm_x86_ops *ops = opaque;
6042 
6043 	if (kvm_x86_ops) {
6044 		printk(KERN_ERR "kvm: already loaded the other module\n");
6045 		r = -EEXIST;
6046 		goto out;
6047 	}
6048 
6049 	if (!ops->cpu_has_kvm_support()) {
6050 		printk(KERN_ERR "kvm: no hardware support\n");
6051 		r = -EOPNOTSUPP;
6052 		goto out;
6053 	}
6054 	if (ops->disabled_by_bios()) {
6055 		printk(KERN_ERR "kvm: disabled by bios\n");
6056 		r = -EOPNOTSUPP;
6057 		goto out;
6058 	}
6059 
6060 	r = -ENOMEM;
6061 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6062 	if (!shared_msrs) {
6063 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6064 		goto out;
6065 	}
6066 
6067 	r = kvm_mmu_module_init();
6068 	if (r)
6069 		goto out_free_percpu;
6070 
6071 	kvm_set_mmio_spte_mask();
6072 
6073 	kvm_x86_ops = ops;
6074 
6075 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6076 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6077 			PT_PRESENT_MASK, 0);
6078 	kvm_timer_init();
6079 
6080 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6081 
6082 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6083 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6084 
6085 	kvm_lapic_init();
6086 #ifdef CONFIG_X86_64
6087 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6088 #endif
6089 
6090 	return 0;
6091 
6092 out_free_percpu:
6093 	free_percpu(shared_msrs);
6094 out:
6095 	return r;
6096 }
6097 
6098 void kvm_arch_exit(void)
6099 {
6100 	kvm_lapic_exit();
6101 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6102 
6103 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6104 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6105 					    CPUFREQ_TRANSITION_NOTIFIER);
6106 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6107 #ifdef CONFIG_X86_64
6108 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6109 #endif
6110 	kvm_x86_ops = NULL;
6111 	kvm_mmu_module_exit();
6112 	free_percpu(shared_msrs);
6113 }
6114 
6115 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6116 {
6117 	++vcpu->stat.halt_exits;
6118 	if (lapic_in_kernel(vcpu)) {
6119 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6120 		return 1;
6121 	} else {
6122 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6123 		return 0;
6124 	}
6125 }
6126 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6127 
6128 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6129 {
6130 	int ret = kvm_skip_emulated_instruction(vcpu);
6131 	/*
6132 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6133 	 * KVM_EXIT_DEBUG here.
6134 	 */
6135 	return kvm_vcpu_halt(vcpu) && ret;
6136 }
6137 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6138 
6139 #ifdef CONFIG_X86_64
6140 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6141 			        unsigned long clock_type)
6142 {
6143 	struct kvm_clock_pairing clock_pairing;
6144 	struct timespec ts;
6145 	u64 cycle;
6146 	int ret;
6147 
6148 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6149 		return -KVM_EOPNOTSUPP;
6150 
6151 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6152 		return -KVM_EOPNOTSUPP;
6153 
6154 	clock_pairing.sec = ts.tv_sec;
6155 	clock_pairing.nsec = ts.tv_nsec;
6156 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6157 	clock_pairing.flags = 0;
6158 
6159 	ret = 0;
6160 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6161 			    sizeof(struct kvm_clock_pairing)))
6162 		ret = -KVM_EFAULT;
6163 
6164 	return ret;
6165 }
6166 #endif
6167 
6168 /*
6169  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6170  *
6171  * @apicid - apicid of vcpu to be kicked.
6172  */
6173 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6174 {
6175 	struct kvm_lapic_irq lapic_irq;
6176 
6177 	lapic_irq.shorthand = 0;
6178 	lapic_irq.dest_mode = 0;
6179 	lapic_irq.dest_id = apicid;
6180 	lapic_irq.msi_redir_hint = false;
6181 
6182 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6183 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6184 }
6185 
6186 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6187 {
6188 	vcpu->arch.apicv_active = false;
6189 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6190 }
6191 
6192 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6193 {
6194 	unsigned long nr, a0, a1, a2, a3, ret;
6195 	int op_64_bit, r;
6196 
6197 	r = kvm_skip_emulated_instruction(vcpu);
6198 
6199 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6200 		return kvm_hv_hypercall(vcpu);
6201 
6202 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6203 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6204 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6205 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6206 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6207 
6208 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6209 
6210 	op_64_bit = is_64_bit_mode(vcpu);
6211 	if (!op_64_bit) {
6212 		nr &= 0xFFFFFFFF;
6213 		a0 &= 0xFFFFFFFF;
6214 		a1 &= 0xFFFFFFFF;
6215 		a2 &= 0xFFFFFFFF;
6216 		a3 &= 0xFFFFFFFF;
6217 	}
6218 
6219 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6220 		ret = -KVM_EPERM;
6221 		goto out;
6222 	}
6223 
6224 	switch (nr) {
6225 	case KVM_HC_VAPIC_POLL_IRQ:
6226 		ret = 0;
6227 		break;
6228 	case KVM_HC_KICK_CPU:
6229 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6230 		ret = 0;
6231 		break;
6232 #ifdef CONFIG_X86_64
6233 	case KVM_HC_CLOCK_PAIRING:
6234 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6235 		break;
6236 #endif
6237 	default:
6238 		ret = -KVM_ENOSYS;
6239 		break;
6240 	}
6241 out:
6242 	if (!op_64_bit)
6243 		ret = (u32)ret;
6244 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6245 	++vcpu->stat.hypercalls;
6246 	return r;
6247 }
6248 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6249 
6250 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6251 {
6252 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6253 	char instruction[3];
6254 	unsigned long rip = kvm_rip_read(vcpu);
6255 
6256 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6257 
6258 	return emulator_write_emulated(ctxt, rip, instruction, 3,
6259 		&ctxt->exception);
6260 }
6261 
6262 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6263 {
6264 	return vcpu->run->request_interrupt_window &&
6265 		likely(!pic_in_kernel(vcpu->kvm));
6266 }
6267 
6268 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6269 {
6270 	struct kvm_run *kvm_run = vcpu->run;
6271 
6272 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6273 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6274 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6275 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6276 	kvm_run->ready_for_interrupt_injection =
6277 		pic_in_kernel(vcpu->kvm) ||
6278 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6279 }
6280 
6281 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6282 {
6283 	int max_irr, tpr;
6284 
6285 	if (!kvm_x86_ops->update_cr8_intercept)
6286 		return;
6287 
6288 	if (!lapic_in_kernel(vcpu))
6289 		return;
6290 
6291 	if (vcpu->arch.apicv_active)
6292 		return;
6293 
6294 	if (!vcpu->arch.apic->vapic_addr)
6295 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6296 	else
6297 		max_irr = -1;
6298 
6299 	if (max_irr != -1)
6300 		max_irr >>= 4;
6301 
6302 	tpr = kvm_lapic_get_cr8(vcpu);
6303 
6304 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6305 }
6306 
6307 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6308 {
6309 	int r;
6310 
6311 	/* try to reinject previous events if any */
6312 	if (vcpu->arch.exception.pending) {
6313 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6314 					vcpu->arch.exception.has_error_code,
6315 					vcpu->arch.exception.error_code);
6316 
6317 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6318 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6319 					     X86_EFLAGS_RF);
6320 
6321 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6322 		    (vcpu->arch.dr7 & DR7_GD)) {
6323 			vcpu->arch.dr7 &= ~DR7_GD;
6324 			kvm_update_dr7(vcpu);
6325 		}
6326 
6327 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6328 					  vcpu->arch.exception.has_error_code,
6329 					  vcpu->arch.exception.error_code,
6330 					  vcpu->arch.exception.reinject);
6331 		return 0;
6332 	}
6333 
6334 	if (vcpu->arch.nmi_injected) {
6335 		kvm_x86_ops->set_nmi(vcpu);
6336 		return 0;
6337 	}
6338 
6339 	if (vcpu->arch.interrupt.pending) {
6340 		kvm_x86_ops->set_irq(vcpu);
6341 		return 0;
6342 	}
6343 
6344 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6345 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6346 		if (r != 0)
6347 			return r;
6348 	}
6349 
6350 	/* try to inject new event if pending */
6351 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6352 		vcpu->arch.smi_pending = false;
6353 		enter_smm(vcpu);
6354 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6355 		--vcpu->arch.nmi_pending;
6356 		vcpu->arch.nmi_injected = true;
6357 		kvm_x86_ops->set_nmi(vcpu);
6358 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6359 		/*
6360 		 * Because interrupts can be injected asynchronously, we are
6361 		 * calling check_nested_events again here to avoid a race condition.
6362 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6363 		 * proposal and current concerns.  Perhaps we should be setting
6364 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6365 		 */
6366 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6367 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6368 			if (r != 0)
6369 				return r;
6370 		}
6371 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6372 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6373 					    false);
6374 			kvm_x86_ops->set_irq(vcpu);
6375 		}
6376 	}
6377 
6378 	return 0;
6379 }
6380 
6381 static void process_nmi(struct kvm_vcpu *vcpu)
6382 {
6383 	unsigned limit = 2;
6384 
6385 	/*
6386 	 * x86 is limited to one NMI running, and one NMI pending after it.
6387 	 * If an NMI is already in progress, limit further NMIs to just one.
6388 	 * Otherwise, allow two (and we'll inject the first one immediately).
6389 	 */
6390 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6391 		limit = 1;
6392 
6393 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6394 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6395 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6396 }
6397 
6398 #define put_smstate(type, buf, offset, val)			  \
6399 	*(type *)((buf) + (offset) - 0x7e00) = val
6400 
6401 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6402 {
6403 	u32 flags = 0;
6404 	flags |= seg->g       << 23;
6405 	flags |= seg->db      << 22;
6406 	flags |= seg->l       << 21;
6407 	flags |= seg->avl     << 20;
6408 	flags |= seg->present << 15;
6409 	flags |= seg->dpl     << 13;
6410 	flags |= seg->s       << 12;
6411 	flags |= seg->type    << 8;
6412 	return flags;
6413 }
6414 
6415 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6416 {
6417 	struct kvm_segment seg;
6418 	int offset;
6419 
6420 	kvm_get_segment(vcpu, &seg, n);
6421 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6422 
6423 	if (n < 3)
6424 		offset = 0x7f84 + n * 12;
6425 	else
6426 		offset = 0x7f2c + (n - 3) * 12;
6427 
6428 	put_smstate(u32, buf, offset + 8, seg.base);
6429 	put_smstate(u32, buf, offset + 4, seg.limit);
6430 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6431 }
6432 
6433 #ifdef CONFIG_X86_64
6434 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6435 {
6436 	struct kvm_segment seg;
6437 	int offset;
6438 	u16 flags;
6439 
6440 	kvm_get_segment(vcpu, &seg, n);
6441 	offset = 0x7e00 + n * 16;
6442 
6443 	flags = enter_smm_get_segment_flags(&seg) >> 8;
6444 	put_smstate(u16, buf, offset, seg.selector);
6445 	put_smstate(u16, buf, offset + 2, flags);
6446 	put_smstate(u32, buf, offset + 4, seg.limit);
6447 	put_smstate(u64, buf, offset + 8, seg.base);
6448 }
6449 #endif
6450 
6451 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6452 {
6453 	struct desc_ptr dt;
6454 	struct kvm_segment seg;
6455 	unsigned long val;
6456 	int i;
6457 
6458 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6459 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6460 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6461 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6462 
6463 	for (i = 0; i < 8; i++)
6464 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6465 
6466 	kvm_get_dr(vcpu, 6, &val);
6467 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6468 	kvm_get_dr(vcpu, 7, &val);
6469 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6470 
6471 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6472 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6473 	put_smstate(u32, buf, 0x7f64, seg.base);
6474 	put_smstate(u32, buf, 0x7f60, seg.limit);
6475 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6476 
6477 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6478 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6479 	put_smstate(u32, buf, 0x7f80, seg.base);
6480 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6481 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6482 
6483 	kvm_x86_ops->get_gdt(vcpu, &dt);
6484 	put_smstate(u32, buf, 0x7f74, dt.address);
6485 	put_smstate(u32, buf, 0x7f70, dt.size);
6486 
6487 	kvm_x86_ops->get_idt(vcpu, &dt);
6488 	put_smstate(u32, buf, 0x7f58, dt.address);
6489 	put_smstate(u32, buf, 0x7f54, dt.size);
6490 
6491 	for (i = 0; i < 6; i++)
6492 		enter_smm_save_seg_32(vcpu, buf, i);
6493 
6494 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6495 
6496 	/* revision id */
6497 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6498 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6499 }
6500 
6501 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6502 {
6503 #ifdef CONFIG_X86_64
6504 	struct desc_ptr dt;
6505 	struct kvm_segment seg;
6506 	unsigned long val;
6507 	int i;
6508 
6509 	for (i = 0; i < 16; i++)
6510 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6511 
6512 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6513 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6514 
6515 	kvm_get_dr(vcpu, 6, &val);
6516 	put_smstate(u64, buf, 0x7f68, val);
6517 	kvm_get_dr(vcpu, 7, &val);
6518 	put_smstate(u64, buf, 0x7f60, val);
6519 
6520 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6521 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6522 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6523 
6524 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6525 
6526 	/* revision id */
6527 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6528 
6529 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6530 
6531 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6532 	put_smstate(u16, buf, 0x7e90, seg.selector);
6533 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6534 	put_smstate(u32, buf, 0x7e94, seg.limit);
6535 	put_smstate(u64, buf, 0x7e98, seg.base);
6536 
6537 	kvm_x86_ops->get_idt(vcpu, &dt);
6538 	put_smstate(u32, buf, 0x7e84, dt.size);
6539 	put_smstate(u64, buf, 0x7e88, dt.address);
6540 
6541 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6542 	put_smstate(u16, buf, 0x7e70, seg.selector);
6543 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6544 	put_smstate(u32, buf, 0x7e74, seg.limit);
6545 	put_smstate(u64, buf, 0x7e78, seg.base);
6546 
6547 	kvm_x86_ops->get_gdt(vcpu, &dt);
6548 	put_smstate(u32, buf, 0x7e64, dt.size);
6549 	put_smstate(u64, buf, 0x7e68, dt.address);
6550 
6551 	for (i = 0; i < 6; i++)
6552 		enter_smm_save_seg_64(vcpu, buf, i);
6553 #else
6554 	WARN_ON_ONCE(1);
6555 #endif
6556 }
6557 
6558 static void enter_smm(struct kvm_vcpu *vcpu)
6559 {
6560 	struct kvm_segment cs, ds;
6561 	struct desc_ptr dt;
6562 	char buf[512];
6563 	u32 cr0;
6564 
6565 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6566 	vcpu->arch.hflags |= HF_SMM_MASK;
6567 	memset(buf, 0, 512);
6568 	if (guest_cpuid_has_longmode(vcpu))
6569 		enter_smm_save_state_64(vcpu, buf);
6570 	else
6571 		enter_smm_save_state_32(vcpu, buf);
6572 
6573 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6574 
6575 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6576 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6577 	else
6578 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6579 
6580 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6581 	kvm_rip_write(vcpu, 0x8000);
6582 
6583 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6584 	kvm_x86_ops->set_cr0(vcpu, cr0);
6585 	vcpu->arch.cr0 = cr0;
6586 
6587 	kvm_x86_ops->set_cr4(vcpu, 0);
6588 
6589 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6590 	dt.address = dt.size = 0;
6591 	kvm_x86_ops->set_idt(vcpu, &dt);
6592 
6593 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6594 
6595 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6596 	cs.base = vcpu->arch.smbase;
6597 
6598 	ds.selector = 0;
6599 	ds.base = 0;
6600 
6601 	cs.limit    = ds.limit = 0xffffffff;
6602 	cs.type     = ds.type = 0x3;
6603 	cs.dpl      = ds.dpl = 0;
6604 	cs.db       = ds.db = 0;
6605 	cs.s        = ds.s = 1;
6606 	cs.l        = ds.l = 0;
6607 	cs.g        = ds.g = 1;
6608 	cs.avl      = ds.avl = 0;
6609 	cs.present  = ds.present = 1;
6610 	cs.unusable = ds.unusable = 0;
6611 	cs.padding  = ds.padding = 0;
6612 
6613 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6614 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6615 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6616 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6617 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6618 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6619 
6620 	if (guest_cpuid_has_longmode(vcpu))
6621 		kvm_x86_ops->set_efer(vcpu, 0);
6622 
6623 	kvm_update_cpuid(vcpu);
6624 	kvm_mmu_reset_context(vcpu);
6625 }
6626 
6627 static void process_smi(struct kvm_vcpu *vcpu)
6628 {
6629 	vcpu->arch.smi_pending = true;
6630 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6631 }
6632 
6633 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6634 {
6635 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6636 }
6637 
6638 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6639 {
6640 	u64 eoi_exit_bitmap[4];
6641 
6642 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6643 		return;
6644 
6645 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6646 
6647 	if (irqchip_split(vcpu->kvm))
6648 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6649 	else {
6650 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6651 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6652 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6653 	}
6654 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6655 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
6656 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6657 }
6658 
6659 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6660 {
6661 	++vcpu->stat.tlb_flush;
6662 	kvm_x86_ops->tlb_flush(vcpu);
6663 }
6664 
6665 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6666 {
6667 	struct page *page = NULL;
6668 
6669 	if (!lapic_in_kernel(vcpu))
6670 		return;
6671 
6672 	if (!kvm_x86_ops->set_apic_access_page_addr)
6673 		return;
6674 
6675 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6676 	if (is_error_page(page))
6677 		return;
6678 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6679 
6680 	/*
6681 	 * Do not pin apic access page in memory, the MMU notifier
6682 	 * will call us again if it is migrated or swapped out.
6683 	 */
6684 	put_page(page);
6685 }
6686 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6687 
6688 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6689 					   unsigned long address)
6690 {
6691 	/*
6692 	 * The physical address of apic access page is stored in the VMCS.
6693 	 * Update it when it becomes invalid.
6694 	 */
6695 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6696 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6697 }
6698 
6699 /*
6700  * Returns 1 to let vcpu_run() continue the guest execution loop without
6701  * exiting to the userspace.  Otherwise, the value will be returned to the
6702  * userspace.
6703  */
6704 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6705 {
6706 	int r;
6707 	bool req_int_win =
6708 		dm_request_for_irq_injection(vcpu) &&
6709 		kvm_cpu_accept_dm_intr(vcpu);
6710 
6711 	bool req_immediate_exit = false;
6712 
6713 	if (vcpu->requests) {
6714 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6715 			kvm_mmu_unload(vcpu);
6716 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6717 			__kvm_migrate_timers(vcpu);
6718 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6719 			kvm_gen_update_masterclock(vcpu->kvm);
6720 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6721 			kvm_gen_kvmclock_update(vcpu);
6722 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6723 			r = kvm_guest_time_update(vcpu);
6724 			if (unlikely(r))
6725 				goto out;
6726 		}
6727 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6728 			kvm_mmu_sync_roots(vcpu);
6729 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6730 			kvm_vcpu_flush_tlb(vcpu);
6731 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6732 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6733 			r = 0;
6734 			goto out;
6735 		}
6736 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6737 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6738 			r = 0;
6739 			goto out;
6740 		}
6741 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6742 			/* Page is swapped out. Do synthetic halt */
6743 			vcpu->arch.apf.halted = true;
6744 			r = 1;
6745 			goto out;
6746 		}
6747 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6748 			record_steal_time(vcpu);
6749 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6750 			process_smi(vcpu);
6751 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6752 			process_nmi(vcpu);
6753 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6754 			kvm_pmu_handle_event(vcpu);
6755 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6756 			kvm_pmu_deliver_pmi(vcpu);
6757 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6758 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6759 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
6760 				     vcpu->arch.ioapic_handled_vectors)) {
6761 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6762 				vcpu->run->eoi.vector =
6763 						vcpu->arch.pending_ioapic_eoi;
6764 				r = 0;
6765 				goto out;
6766 			}
6767 		}
6768 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6769 			vcpu_scan_ioapic(vcpu);
6770 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6771 			kvm_vcpu_reload_apic_access_page(vcpu);
6772 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6773 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6774 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6775 			r = 0;
6776 			goto out;
6777 		}
6778 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6779 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6780 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6781 			r = 0;
6782 			goto out;
6783 		}
6784 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6785 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6786 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6787 			r = 0;
6788 			goto out;
6789 		}
6790 
6791 		/*
6792 		 * KVM_REQ_HV_STIMER has to be processed after
6793 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6794 		 * depend on the guest clock being up-to-date
6795 		 */
6796 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6797 			kvm_hv_process_stimers(vcpu);
6798 	}
6799 
6800 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6801 		++vcpu->stat.req_event;
6802 		kvm_apic_accept_events(vcpu);
6803 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6804 			r = 1;
6805 			goto out;
6806 		}
6807 
6808 		if (inject_pending_event(vcpu, req_int_win) != 0)
6809 			req_immediate_exit = true;
6810 		else {
6811 			/* Enable NMI/IRQ window open exits if needed.
6812 			 *
6813 			 * SMIs have two cases: 1) they can be nested, and
6814 			 * then there is nothing to do here because RSM will
6815 			 * cause a vmexit anyway; 2) or the SMI can be pending
6816 			 * because inject_pending_event has completed the
6817 			 * injection of an IRQ or NMI from the previous vmexit,
6818 			 * and then we request an immediate exit to inject the SMI.
6819 			 */
6820 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
6821 				req_immediate_exit = true;
6822 			if (vcpu->arch.nmi_pending)
6823 				kvm_x86_ops->enable_nmi_window(vcpu);
6824 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6825 				kvm_x86_ops->enable_irq_window(vcpu);
6826 		}
6827 
6828 		if (kvm_lapic_enabled(vcpu)) {
6829 			update_cr8_intercept(vcpu);
6830 			kvm_lapic_sync_to_vapic(vcpu);
6831 		}
6832 	}
6833 
6834 	r = kvm_mmu_reload(vcpu);
6835 	if (unlikely(r)) {
6836 		goto cancel_injection;
6837 	}
6838 
6839 	preempt_disable();
6840 
6841 	kvm_x86_ops->prepare_guest_switch(vcpu);
6842 	kvm_load_guest_fpu(vcpu);
6843 
6844 	/*
6845 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
6846 	 * IPI are then delayed after guest entry, which ensures that they
6847 	 * result in virtual interrupt delivery.
6848 	 */
6849 	local_irq_disable();
6850 	vcpu->mode = IN_GUEST_MODE;
6851 
6852 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6853 
6854 	/*
6855 	 * 1) We should set ->mode before checking ->requests.  Please see
6856 	 * the comment in kvm_vcpu_exiting_guest_mode().
6857 	 *
6858 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
6859 	 * pairs with the memory barrier implicit in pi_test_and_set_on
6860 	 * (see vmx_deliver_posted_interrupt).
6861 	 *
6862 	 * 3) This also orders the write to mode from any reads to the page
6863 	 * tables done while the VCPU is running.  Please see the comment
6864 	 * in kvm_flush_remote_tlbs.
6865 	 */
6866 	smp_mb__after_srcu_read_unlock();
6867 
6868 	/*
6869 	 * This handles the case where a posted interrupt was
6870 	 * notified with kvm_vcpu_kick.
6871 	 */
6872 	if (kvm_lapic_enabled(vcpu)) {
6873 		if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6874 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6875 	}
6876 
6877 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6878 	    || need_resched() || signal_pending(current)) {
6879 		vcpu->mode = OUTSIDE_GUEST_MODE;
6880 		smp_wmb();
6881 		local_irq_enable();
6882 		preempt_enable();
6883 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6884 		r = 1;
6885 		goto cancel_injection;
6886 	}
6887 
6888 	kvm_load_guest_xcr0(vcpu);
6889 
6890 	if (req_immediate_exit) {
6891 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6892 		smp_send_reschedule(vcpu->cpu);
6893 	}
6894 
6895 	trace_kvm_entry(vcpu->vcpu_id);
6896 	wait_lapic_expire(vcpu);
6897 	guest_enter_irqoff();
6898 
6899 	if (unlikely(vcpu->arch.switch_db_regs)) {
6900 		set_debugreg(0, 7);
6901 		set_debugreg(vcpu->arch.eff_db[0], 0);
6902 		set_debugreg(vcpu->arch.eff_db[1], 1);
6903 		set_debugreg(vcpu->arch.eff_db[2], 2);
6904 		set_debugreg(vcpu->arch.eff_db[3], 3);
6905 		set_debugreg(vcpu->arch.dr6, 6);
6906 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6907 	}
6908 
6909 	kvm_x86_ops->run(vcpu);
6910 
6911 	/*
6912 	 * Do this here before restoring debug registers on the host.  And
6913 	 * since we do this before handling the vmexit, a DR access vmexit
6914 	 * can (a) read the correct value of the debug registers, (b) set
6915 	 * KVM_DEBUGREG_WONT_EXIT again.
6916 	 */
6917 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6918 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6919 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6920 		kvm_update_dr0123(vcpu);
6921 		kvm_update_dr6(vcpu);
6922 		kvm_update_dr7(vcpu);
6923 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6924 	}
6925 
6926 	/*
6927 	 * If the guest has used debug registers, at least dr7
6928 	 * will be disabled while returning to the host.
6929 	 * If we don't have active breakpoints in the host, we don't
6930 	 * care about the messed up debug address registers. But if
6931 	 * we have some of them active, restore the old state.
6932 	 */
6933 	if (hw_breakpoint_active())
6934 		hw_breakpoint_restore();
6935 
6936 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6937 
6938 	vcpu->mode = OUTSIDE_GUEST_MODE;
6939 	smp_wmb();
6940 
6941 	kvm_put_guest_xcr0(vcpu);
6942 
6943 	kvm_x86_ops->handle_external_intr(vcpu);
6944 
6945 	++vcpu->stat.exits;
6946 
6947 	guest_exit_irqoff();
6948 
6949 	local_irq_enable();
6950 	preempt_enable();
6951 
6952 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6953 
6954 	/*
6955 	 * Profile KVM exit RIPs:
6956 	 */
6957 	if (unlikely(prof_on == KVM_PROFILING)) {
6958 		unsigned long rip = kvm_rip_read(vcpu);
6959 		profile_hit(KVM_PROFILING, (void *)rip);
6960 	}
6961 
6962 	if (unlikely(vcpu->arch.tsc_always_catchup))
6963 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6964 
6965 	if (vcpu->arch.apic_attention)
6966 		kvm_lapic_sync_from_vapic(vcpu);
6967 
6968 	r = kvm_x86_ops->handle_exit(vcpu);
6969 	return r;
6970 
6971 cancel_injection:
6972 	kvm_x86_ops->cancel_injection(vcpu);
6973 	if (unlikely(vcpu->arch.apic_attention))
6974 		kvm_lapic_sync_from_vapic(vcpu);
6975 out:
6976 	return r;
6977 }
6978 
6979 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6980 {
6981 	if (!kvm_arch_vcpu_runnable(vcpu) &&
6982 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6983 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6984 		kvm_vcpu_block(vcpu);
6985 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6986 
6987 		if (kvm_x86_ops->post_block)
6988 			kvm_x86_ops->post_block(vcpu);
6989 
6990 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6991 			return 1;
6992 	}
6993 
6994 	kvm_apic_accept_events(vcpu);
6995 	switch(vcpu->arch.mp_state) {
6996 	case KVM_MP_STATE_HALTED:
6997 		vcpu->arch.pv.pv_unhalted = false;
6998 		vcpu->arch.mp_state =
6999 			KVM_MP_STATE_RUNNABLE;
7000 	case KVM_MP_STATE_RUNNABLE:
7001 		vcpu->arch.apf.halted = false;
7002 		break;
7003 	case KVM_MP_STATE_INIT_RECEIVED:
7004 		break;
7005 	default:
7006 		return -EINTR;
7007 		break;
7008 	}
7009 	return 1;
7010 }
7011 
7012 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7013 {
7014 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7015 		kvm_x86_ops->check_nested_events(vcpu, false);
7016 
7017 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7018 		!vcpu->arch.apf.halted);
7019 }
7020 
7021 static int vcpu_run(struct kvm_vcpu *vcpu)
7022 {
7023 	int r;
7024 	struct kvm *kvm = vcpu->kvm;
7025 
7026 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7027 
7028 	for (;;) {
7029 		if (kvm_vcpu_running(vcpu)) {
7030 			r = vcpu_enter_guest(vcpu);
7031 		} else {
7032 			r = vcpu_block(kvm, vcpu);
7033 		}
7034 
7035 		if (r <= 0)
7036 			break;
7037 
7038 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7039 		if (kvm_cpu_has_pending_timer(vcpu))
7040 			kvm_inject_pending_timer_irqs(vcpu);
7041 
7042 		if (dm_request_for_irq_injection(vcpu) &&
7043 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7044 			r = 0;
7045 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7046 			++vcpu->stat.request_irq_exits;
7047 			break;
7048 		}
7049 
7050 		kvm_check_async_pf_completion(vcpu);
7051 
7052 		if (signal_pending(current)) {
7053 			r = -EINTR;
7054 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7055 			++vcpu->stat.signal_exits;
7056 			break;
7057 		}
7058 		if (need_resched()) {
7059 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7060 			cond_resched();
7061 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7062 		}
7063 	}
7064 
7065 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7066 
7067 	return r;
7068 }
7069 
7070 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7071 {
7072 	int r;
7073 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7074 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7075 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7076 	if (r != EMULATE_DONE)
7077 		return 0;
7078 	return 1;
7079 }
7080 
7081 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7082 {
7083 	BUG_ON(!vcpu->arch.pio.count);
7084 
7085 	return complete_emulated_io(vcpu);
7086 }
7087 
7088 /*
7089  * Implements the following, as a state machine:
7090  *
7091  * read:
7092  *   for each fragment
7093  *     for each mmio piece in the fragment
7094  *       write gpa, len
7095  *       exit
7096  *       copy data
7097  *   execute insn
7098  *
7099  * write:
7100  *   for each fragment
7101  *     for each mmio piece in the fragment
7102  *       write gpa, len
7103  *       copy data
7104  *       exit
7105  */
7106 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7107 {
7108 	struct kvm_run *run = vcpu->run;
7109 	struct kvm_mmio_fragment *frag;
7110 	unsigned len;
7111 
7112 	BUG_ON(!vcpu->mmio_needed);
7113 
7114 	/* Complete previous fragment */
7115 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7116 	len = min(8u, frag->len);
7117 	if (!vcpu->mmio_is_write)
7118 		memcpy(frag->data, run->mmio.data, len);
7119 
7120 	if (frag->len <= 8) {
7121 		/* Switch to the next fragment. */
7122 		frag++;
7123 		vcpu->mmio_cur_fragment++;
7124 	} else {
7125 		/* Go forward to the next mmio piece. */
7126 		frag->data += len;
7127 		frag->gpa += len;
7128 		frag->len -= len;
7129 	}
7130 
7131 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7132 		vcpu->mmio_needed = 0;
7133 
7134 		/* FIXME: return into emulator if single-stepping.  */
7135 		if (vcpu->mmio_is_write)
7136 			return 1;
7137 		vcpu->mmio_read_completed = 1;
7138 		return complete_emulated_io(vcpu);
7139 	}
7140 
7141 	run->exit_reason = KVM_EXIT_MMIO;
7142 	run->mmio.phys_addr = frag->gpa;
7143 	if (vcpu->mmio_is_write)
7144 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7145 	run->mmio.len = min(8u, frag->len);
7146 	run->mmio.is_write = vcpu->mmio_is_write;
7147 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7148 	return 0;
7149 }
7150 
7151 
7152 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7153 {
7154 	struct fpu *fpu = &current->thread.fpu;
7155 	int r;
7156 	sigset_t sigsaved;
7157 
7158 	fpu__activate_curr(fpu);
7159 
7160 	if (vcpu->sigset_active)
7161 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7162 
7163 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7164 		kvm_vcpu_block(vcpu);
7165 		kvm_apic_accept_events(vcpu);
7166 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7167 		r = -EAGAIN;
7168 		goto out;
7169 	}
7170 
7171 	/* re-sync apic's tpr */
7172 	if (!lapic_in_kernel(vcpu)) {
7173 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7174 			r = -EINVAL;
7175 			goto out;
7176 		}
7177 	}
7178 
7179 	if (unlikely(vcpu->arch.complete_userspace_io)) {
7180 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7181 		vcpu->arch.complete_userspace_io = NULL;
7182 		r = cui(vcpu);
7183 		if (r <= 0)
7184 			goto out;
7185 	} else
7186 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7187 
7188 	if (kvm_run->immediate_exit)
7189 		r = -EINTR;
7190 	else
7191 		r = vcpu_run(vcpu);
7192 
7193 out:
7194 	post_kvm_run_save(vcpu);
7195 	if (vcpu->sigset_active)
7196 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7197 
7198 	return r;
7199 }
7200 
7201 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7202 {
7203 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7204 		/*
7205 		 * We are here if userspace calls get_regs() in the middle of
7206 		 * instruction emulation. Registers state needs to be copied
7207 		 * back from emulation context to vcpu. Userspace shouldn't do
7208 		 * that usually, but some bad designed PV devices (vmware
7209 		 * backdoor interface) need this to work
7210 		 */
7211 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7212 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7213 	}
7214 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7215 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7216 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7217 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7218 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7219 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7220 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7221 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7222 #ifdef CONFIG_X86_64
7223 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7224 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7225 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7226 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7227 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7228 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7229 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7230 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7231 #endif
7232 
7233 	regs->rip = kvm_rip_read(vcpu);
7234 	regs->rflags = kvm_get_rflags(vcpu);
7235 
7236 	return 0;
7237 }
7238 
7239 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7240 {
7241 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7242 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7243 
7244 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7245 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7246 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7247 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7248 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7249 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7250 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7251 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7252 #ifdef CONFIG_X86_64
7253 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7254 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7255 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7256 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7257 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7258 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7259 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7260 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7261 #endif
7262 
7263 	kvm_rip_write(vcpu, regs->rip);
7264 	kvm_set_rflags(vcpu, regs->rflags);
7265 
7266 	vcpu->arch.exception.pending = false;
7267 
7268 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7269 
7270 	return 0;
7271 }
7272 
7273 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7274 {
7275 	struct kvm_segment cs;
7276 
7277 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7278 	*db = cs.db;
7279 	*l = cs.l;
7280 }
7281 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7282 
7283 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7284 				  struct kvm_sregs *sregs)
7285 {
7286 	struct desc_ptr dt;
7287 
7288 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7289 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7290 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7291 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7292 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7293 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7294 
7295 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7296 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7297 
7298 	kvm_x86_ops->get_idt(vcpu, &dt);
7299 	sregs->idt.limit = dt.size;
7300 	sregs->idt.base = dt.address;
7301 	kvm_x86_ops->get_gdt(vcpu, &dt);
7302 	sregs->gdt.limit = dt.size;
7303 	sregs->gdt.base = dt.address;
7304 
7305 	sregs->cr0 = kvm_read_cr0(vcpu);
7306 	sregs->cr2 = vcpu->arch.cr2;
7307 	sregs->cr3 = kvm_read_cr3(vcpu);
7308 	sregs->cr4 = kvm_read_cr4(vcpu);
7309 	sregs->cr8 = kvm_get_cr8(vcpu);
7310 	sregs->efer = vcpu->arch.efer;
7311 	sregs->apic_base = kvm_get_apic_base(vcpu);
7312 
7313 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7314 
7315 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7316 		set_bit(vcpu->arch.interrupt.nr,
7317 			(unsigned long *)sregs->interrupt_bitmap);
7318 
7319 	return 0;
7320 }
7321 
7322 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7323 				    struct kvm_mp_state *mp_state)
7324 {
7325 	kvm_apic_accept_events(vcpu);
7326 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7327 					vcpu->arch.pv.pv_unhalted)
7328 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7329 	else
7330 		mp_state->mp_state = vcpu->arch.mp_state;
7331 
7332 	return 0;
7333 }
7334 
7335 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7336 				    struct kvm_mp_state *mp_state)
7337 {
7338 	if (!lapic_in_kernel(vcpu) &&
7339 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7340 		return -EINVAL;
7341 
7342 	/* INITs are latched while in SMM */
7343 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7344 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7345 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7346 		return -EINVAL;
7347 
7348 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7349 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7350 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7351 	} else
7352 		vcpu->arch.mp_state = mp_state->mp_state;
7353 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7354 	return 0;
7355 }
7356 
7357 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7358 		    int reason, bool has_error_code, u32 error_code)
7359 {
7360 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7361 	int ret;
7362 
7363 	init_emulate_ctxt(vcpu);
7364 
7365 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7366 				   has_error_code, error_code);
7367 
7368 	if (ret)
7369 		return EMULATE_FAIL;
7370 
7371 	kvm_rip_write(vcpu, ctxt->eip);
7372 	kvm_set_rflags(vcpu, ctxt->eflags);
7373 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7374 	return EMULATE_DONE;
7375 }
7376 EXPORT_SYMBOL_GPL(kvm_task_switch);
7377 
7378 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7379 				  struct kvm_sregs *sregs)
7380 {
7381 	struct msr_data apic_base_msr;
7382 	int mmu_reset_needed = 0;
7383 	int pending_vec, max_bits, idx;
7384 	struct desc_ptr dt;
7385 
7386 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7387 		return -EINVAL;
7388 
7389 	dt.size = sregs->idt.limit;
7390 	dt.address = sregs->idt.base;
7391 	kvm_x86_ops->set_idt(vcpu, &dt);
7392 	dt.size = sregs->gdt.limit;
7393 	dt.address = sregs->gdt.base;
7394 	kvm_x86_ops->set_gdt(vcpu, &dt);
7395 
7396 	vcpu->arch.cr2 = sregs->cr2;
7397 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7398 	vcpu->arch.cr3 = sregs->cr3;
7399 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7400 
7401 	kvm_set_cr8(vcpu, sregs->cr8);
7402 
7403 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7404 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7405 	apic_base_msr.data = sregs->apic_base;
7406 	apic_base_msr.host_initiated = true;
7407 	kvm_set_apic_base(vcpu, &apic_base_msr);
7408 
7409 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7410 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7411 	vcpu->arch.cr0 = sregs->cr0;
7412 
7413 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7414 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7415 	if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7416 		kvm_update_cpuid(vcpu);
7417 
7418 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7419 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7420 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7421 		mmu_reset_needed = 1;
7422 	}
7423 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7424 
7425 	if (mmu_reset_needed)
7426 		kvm_mmu_reset_context(vcpu);
7427 
7428 	max_bits = KVM_NR_INTERRUPTS;
7429 	pending_vec = find_first_bit(
7430 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7431 	if (pending_vec < max_bits) {
7432 		kvm_queue_interrupt(vcpu, pending_vec, false);
7433 		pr_debug("Set back pending irq %d\n", pending_vec);
7434 	}
7435 
7436 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7437 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7438 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7439 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7440 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7441 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7442 
7443 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7444 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7445 
7446 	update_cr8_intercept(vcpu);
7447 
7448 	/* Older userspace won't unhalt the vcpu on reset. */
7449 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7450 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7451 	    !is_protmode(vcpu))
7452 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7453 
7454 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7455 
7456 	return 0;
7457 }
7458 
7459 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7460 					struct kvm_guest_debug *dbg)
7461 {
7462 	unsigned long rflags;
7463 	int i, r;
7464 
7465 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7466 		r = -EBUSY;
7467 		if (vcpu->arch.exception.pending)
7468 			goto out;
7469 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7470 			kvm_queue_exception(vcpu, DB_VECTOR);
7471 		else
7472 			kvm_queue_exception(vcpu, BP_VECTOR);
7473 	}
7474 
7475 	/*
7476 	 * Read rflags as long as potentially injected trace flags are still
7477 	 * filtered out.
7478 	 */
7479 	rflags = kvm_get_rflags(vcpu);
7480 
7481 	vcpu->guest_debug = dbg->control;
7482 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7483 		vcpu->guest_debug = 0;
7484 
7485 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7486 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7487 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7488 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7489 	} else {
7490 		for (i = 0; i < KVM_NR_DB_REGS; i++)
7491 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7492 	}
7493 	kvm_update_dr7(vcpu);
7494 
7495 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7496 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7497 			get_segment_base(vcpu, VCPU_SREG_CS);
7498 
7499 	/*
7500 	 * Trigger an rflags update that will inject or remove the trace
7501 	 * flags.
7502 	 */
7503 	kvm_set_rflags(vcpu, rflags);
7504 
7505 	kvm_x86_ops->update_bp_intercept(vcpu);
7506 
7507 	r = 0;
7508 
7509 out:
7510 
7511 	return r;
7512 }
7513 
7514 /*
7515  * Translate a guest virtual address to a guest physical address.
7516  */
7517 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7518 				    struct kvm_translation *tr)
7519 {
7520 	unsigned long vaddr = tr->linear_address;
7521 	gpa_t gpa;
7522 	int idx;
7523 
7524 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7525 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7526 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7527 	tr->physical_address = gpa;
7528 	tr->valid = gpa != UNMAPPED_GVA;
7529 	tr->writeable = 1;
7530 	tr->usermode = 0;
7531 
7532 	return 0;
7533 }
7534 
7535 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7536 {
7537 	struct fxregs_state *fxsave =
7538 			&vcpu->arch.guest_fpu.state.fxsave;
7539 
7540 	memcpy(fpu->fpr, fxsave->st_space, 128);
7541 	fpu->fcw = fxsave->cwd;
7542 	fpu->fsw = fxsave->swd;
7543 	fpu->ftwx = fxsave->twd;
7544 	fpu->last_opcode = fxsave->fop;
7545 	fpu->last_ip = fxsave->rip;
7546 	fpu->last_dp = fxsave->rdp;
7547 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7548 
7549 	return 0;
7550 }
7551 
7552 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7553 {
7554 	struct fxregs_state *fxsave =
7555 			&vcpu->arch.guest_fpu.state.fxsave;
7556 
7557 	memcpy(fxsave->st_space, fpu->fpr, 128);
7558 	fxsave->cwd = fpu->fcw;
7559 	fxsave->swd = fpu->fsw;
7560 	fxsave->twd = fpu->ftwx;
7561 	fxsave->fop = fpu->last_opcode;
7562 	fxsave->rip = fpu->last_ip;
7563 	fxsave->rdp = fpu->last_dp;
7564 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7565 
7566 	return 0;
7567 }
7568 
7569 static void fx_init(struct kvm_vcpu *vcpu)
7570 {
7571 	fpstate_init(&vcpu->arch.guest_fpu.state);
7572 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7573 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7574 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7575 
7576 	/*
7577 	 * Ensure guest xcr0 is valid for loading
7578 	 */
7579 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7580 
7581 	vcpu->arch.cr0 |= X86_CR0_ET;
7582 }
7583 
7584 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7585 {
7586 	if (vcpu->guest_fpu_loaded)
7587 		return;
7588 
7589 	/*
7590 	 * Restore all possible states in the guest,
7591 	 * and assume host would use all available bits.
7592 	 * Guest xcr0 would be loaded later.
7593 	 */
7594 	vcpu->guest_fpu_loaded = 1;
7595 	__kernel_fpu_begin();
7596 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7597 	trace_kvm_fpu(1);
7598 }
7599 
7600 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7601 {
7602 	if (!vcpu->guest_fpu_loaded)
7603 		return;
7604 
7605 	vcpu->guest_fpu_loaded = 0;
7606 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7607 	__kernel_fpu_end();
7608 	++vcpu->stat.fpu_reload;
7609 	trace_kvm_fpu(0);
7610 }
7611 
7612 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7613 {
7614 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7615 
7616 	kvmclock_reset(vcpu);
7617 
7618 	kvm_x86_ops->vcpu_free(vcpu);
7619 	free_cpumask_var(wbinvd_dirty_mask);
7620 }
7621 
7622 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7623 						unsigned int id)
7624 {
7625 	struct kvm_vcpu *vcpu;
7626 
7627 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7628 		printk_once(KERN_WARNING
7629 		"kvm: SMP vm created on host with unstable TSC; "
7630 		"guest TSC will not be reliable\n");
7631 
7632 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7633 
7634 	return vcpu;
7635 }
7636 
7637 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7638 {
7639 	int r;
7640 
7641 	kvm_vcpu_mtrr_init(vcpu);
7642 	r = vcpu_load(vcpu);
7643 	if (r)
7644 		return r;
7645 	kvm_vcpu_reset(vcpu, false);
7646 	kvm_mmu_setup(vcpu);
7647 	vcpu_put(vcpu);
7648 	return r;
7649 }
7650 
7651 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7652 {
7653 	struct msr_data msr;
7654 	struct kvm *kvm = vcpu->kvm;
7655 
7656 	if (vcpu_load(vcpu))
7657 		return;
7658 	msr.data = 0x0;
7659 	msr.index = MSR_IA32_TSC;
7660 	msr.host_initiated = true;
7661 	kvm_write_tsc(vcpu, &msr);
7662 	vcpu_put(vcpu);
7663 
7664 	if (!kvmclock_periodic_sync)
7665 		return;
7666 
7667 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7668 					KVMCLOCK_SYNC_PERIOD);
7669 }
7670 
7671 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7672 {
7673 	int r;
7674 	vcpu->arch.apf.msr_val = 0;
7675 
7676 	r = vcpu_load(vcpu);
7677 	BUG_ON(r);
7678 	kvm_mmu_unload(vcpu);
7679 	vcpu_put(vcpu);
7680 
7681 	kvm_x86_ops->vcpu_free(vcpu);
7682 }
7683 
7684 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7685 {
7686 	vcpu->arch.hflags = 0;
7687 
7688 	vcpu->arch.smi_pending = 0;
7689 	atomic_set(&vcpu->arch.nmi_queued, 0);
7690 	vcpu->arch.nmi_pending = 0;
7691 	vcpu->arch.nmi_injected = false;
7692 	kvm_clear_interrupt_queue(vcpu);
7693 	kvm_clear_exception_queue(vcpu);
7694 
7695 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7696 	kvm_update_dr0123(vcpu);
7697 	vcpu->arch.dr6 = DR6_INIT;
7698 	kvm_update_dr6(vcpu);
7699 	vcpu->arch.dr7 = DR7_FIXED_1;
7700 	kvm_update_dr7(vcpu);
7701 
7702 	vcpu->arch.cr2 = 0;
7703 
7704 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7705 	vcpu->arch.apf.msr_val = 0;
7706 	vcpu->arch.st.msr_val = 0;
7707 
7708 	kvmclock_reset(vcpu);
7709 
7710 	kvm_clear_async_pf_completion_queue(vcpu);
7711 	kvm_async_pf_hash_reset(vcpu);
7712 	vcpu->arch.apf.halted = false;
7713 
7714 	if (!init_event) {
7715 		kvm_pmu_reset(vcpu);
7716 		vcpu->arch.smbase = 0x30000;
7717 
7718 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7719 		vcpu->arch.msr_misc_features_enables = 0;
7720 	}
7721 
7722 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7723 	vcpu->arch.regs_avail = ~0;
7724 	vcpu->arch.regs_dirty = ~0;
7725 
7726 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7727 }
7728 
7729 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7730 {
7731 	struct kvm_segment cs;
7732 
7733 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7734 	cs.selector = vector << 8;
7735 	cs.base = vector << 12;
7736 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7737 	kvm_rip_write(vcpu, 0);
7738 }
7739 
7740 int kvm_arch_hardware_enable(void)
7741 {
7742 	struct kvm *kvm;
7743 	struct kvm_vcpu *vcpu;
7744 	int i;
7745 	int ret;
7746 	u64 local_tsc;
7747 	u64 max_tsc = 0;
7748 	bool stable, backwards_tsc = false;
7749 
7750 	kvm_shared_msr_cpu_online();
7751 	ret = kvm_x86_ops->hardware_enable();
7752 	if (ret != 0)
7753 		return ret;
7754 
7755 	local_tsc = rdtsc();
7756 	stable = !check_tsc_unstable();
7757 	list_for_each_entry(kvm, &vm_list, vm_list) {
7758 		kvm_for_each_vcpu(i, vcpu, kvm) {
7759 			if (!stable && vcpu->cpu == smp_processor_id())
7760 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7761 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7762 				backwards_tsc = true;
7763 				if (vcpu->arch.last_host_tsc > max_tsc)
7764 					max_tsc = vcpu->arch.last_host_tsc;
7765 			}
7766 		}
7767 	}
7768 
7769 	/*
7770 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7771 	 * platforms that reset TSC during suspend or hibernate actions, but
7772 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7773 	 * detect that condition here, which happens early in CPU bringup,
7774 	 * before any KVM threads can be running.  Unfortunately, we can't
7775 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7776 	 * enough into CPU bringup that we know how much real time has actually
7777 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7778 	 * variables that haven't been updated yet.
7779 	 *
7780 	 * So we simply find the maximum observed TSC above, then record the
7781 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7782 	 * the adjustment will be applied.  Note that we accumulate
7783 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7784 	 * gets a chance to run again.  In the event that no KVM threads get a
7785 	 * chance to run, we will miss the entire elapsed period, as we'll have
7786 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7787 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7788 	 * uniform across all VCPUs (not to mention the scenario is extremely
7789 	 * unlikely). It is possible that a second hibernate recovery happens
7790 	 * much faster than a first, causing the observed TSC here to be
7791 	 * smaller; this would require additional padding adjustment, which is
7792 	 * why we set last_host_tsc to the local tsc observed here.
7793 	 *
7794 	 * N.B. - this code below runs only on platforms with reliable TSC,
7795 	 * as that is the only way backwards_tsc is set above.  Also note
7796 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7797 	 * have the same delta_cyc adjustment applied if backwards_tsc
7798 	 * is detected.  Note further, this adjustment is only done once,
7799 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7800 	 * called multiple times (one for each physical CPU bringup).
7801 	 *
7802 	 * Platforms with unreliable TSCs don't have to deal with this, they
7803 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7804 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7805 	 * guarantee that they stay in perfect synchronization.
7806 	 */
7807 	if (backwards_tsc) {
7808 		u64 delta_cyc = max_tsc - local_tsc;
7809 		backwards_tsc_observed = true;
7810 		list_for_each_entry(kvm, &vm_list, vm_list) {
7811 			kvm_for_each_vcpu(i, vcpu, kvm) {
7812 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7813 				vcpu->arch.last_host_tsc = local_tsc;
7814 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7815 			}
7816 
7817 			/*
7818 			 * We have to disable TSC offset matching.. if you were
7819 			 * booting a VM while issuing an S4 host suspend....
7820 			 * you may have some problem.  Solving this issue is
7821 			 * left as an exercise to the reader.
7822 			 */
7823 			kvm->arch.last_tsc_nsec = 0;
7824 			kvm->arch.last_tsc_write = 0;
7825 		}
7826 
7827 	}
7828 	return 0;
7829 }
7830 
7831 void kvm_arch_hardware_disable(void)
7832 {
7833 	kvm_x86_ops->hardware_disable();
7834 	drop_user_return_notifiers();
7835 }
7836 
7837 int kvm_arch_hardware_setup(void)
7838 {
7839 	int r;
7840 
7841 	r = kvm_x86_ops->hardware_setup();
7842 	if (r != 0)
7843 		return r;
7844 
7845 	if (kvm_has_tsc_control) {
7846 		/*
7847 		 * Make sure the user can only configure tsc_khz values that
7848 		 * fit into a signed integer.
7849 		 * A min value is not calculated needed because it will always
7850 		 * be 1 on all machines.
7851 		 */
7852 		u64 max = min(0x7fffffffULL,
7853 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7854 		kvm_max_guest_tsc_khz = max;
7855 
7856 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7857 	}
7858 
7859 	kvm_init_msr_list();
7860 	return 0;
7861 }
7862 
7863 void kvm_arch_hardware_unsetup(void)
7864 {
7865 	kvm_x86_ops->hardware_unsetup();
7866 }
7867 
7868 void kvm_arch_check_processor_compat(void *rtn)
7869 {
7870 	kvm_x86_ops->check_processor_compatibility(rtn);
7871 }
7872 
7873 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7874 {
7875 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7876 }
7877 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7878 
7879 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7880 {
7881 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7882 }
7883 
7884 struct static_key kvm_no_apic_vcpu __read_mostly;
7885 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7886 
7887 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7888 {
7889 	struct page *page;
7890 	struct kvm *kvm;
7891 	int r;
7892 
7893 	BUG_ON(vcpu->kvm == NULL);
7894 	kvm = vcpu->kvm;
7895 
7896 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7897 	vcpu->arch.pv.pv_unhalted = false;
7898 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7899 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7900 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7901 	else
7902 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7903 
7904 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7905 	if (!page) {
7906 		r = -ENOMEM;
7907 		goto fail;
7908 	}
7909 	vcpu->arch.pio_data = page_address(page);
7910 
7911 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7912 
7913 	r = kvm_mmu_create(vcpu);
7914 	if (r < 0)
7915 		goto fail_free_pio_data;
7916 
7917 	if (irqchip_in_kernel(kvm)) {
7918 		r = kvm_create_lapic(vcpu);
7919 		if (r < 0)
7920 			goto fail_mmu_destroy;
7921 	} else
7922 		static_key_slow_inc(&kvm_no_apic_vcpu);
7923 
7924 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7925 				       GFP_KERNEL);
7926 	if (!vcpu->arch.mce_banks) {
7927 		r = -ENOMEM;
7928 		goto fail_free_lapic;
7929 	}
7930 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7931 
7932 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7933 		r = -ENOMEM;
7934 		goto fail_free_mce_banks;
7935 	}
7936 
7937 	fx_init(vcpu);
7938 
7939 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7940 	vcpu->arch.pv_time_enabled = false;
7941 
7942 	vcpu->arch.guest_supported_xcr0 = 0;
7943 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7944 
7945 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7946 
7947 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7948 
7949 	kvm_async_pf_hash_reset(vcpu);
7950 	kvm_pmu_init(vcpu);
7951 
7952 	vcpu->arch.pending_external_vector = -1;
7953 
7954 	kvm_hv_vcpu_init(vcpu);
7955 
7956 	return 0;
7957 
7958 fail_free_mce_banks:
7959 	kfree(vcpu->arch.mce_banks);
7960 fail_free_lapic:
7961 	kvm_free_lapic(vcpu);
7962 fail_mmu_destroy:
7963 	kvm_mmu_destroy(vcpu);
7964 fail_free_pio_data:
7965 	free_page((unsigned long)vcpu->arch.pio_data);
7966 fail:
7967 	return r;
7968 }
7969 
7970 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7971 {
7972 	int idx;
7973 
7974 	kvm_hv_vcpu_uninit(vcpu);
7975 	kvm_pmu_destroy(vcpu);
7976 	kfree(vcpu->arch.mce_banks);
7977 	kvm_free_lapic(vcpu);
7978 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7979 	kvm_mmu_destroy(vcpu);
7980 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7981 	free_page((unsigned long)vcpu->arch.pio_data);
7982 	if (!lapic_in_kernel(vcpu))
7983 		static_key_slow_dec(&kvm_no_apic_vcpu);
7984 }
7985 
7986 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7987 {
7988 	kvm_x86_ops->sched_in(vcpu, cpu);
7989 }
7990 
7991 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7992 {
7993 	if (type)
7994 		return -EINVAL;
7995 
7996 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7997 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7998 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7999 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8000 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8001 
8002 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8003 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8004 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8005 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8006 		&kvm->arch.irq_sources_bitmap);
8007 
8008 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8009 	mutex_init(&kvm->arch.apic_map_lock);
8010 	mutex_init(&kvm->arch.hyperv.hv_lock);
8011 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8012 
8013 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8014 	pvclock_update_vm_gtod_copy(kvm);
8015 
8016 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8017 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8018 
8019 	kvm_page_track_init(kvm);
8020 	kvm_mmu_init_vm(kvm);
8021 
8022 	if (kvm_x86_ops->vm_init)
8023 		return kvm_x86_ops->vm_init(kvm);
8024 
8025 	return 0;
8026 }
8027 
8028 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8029 {
8030 	int r;
8031 	r = vcpu_load(vcpu);
8032 	BUG_ON(r);
8033 	kvm_mmu_unload(vcpu);
8034 	vcpu_put(vcpu);
8035 }
8036 
8037 static void kvm_free_vcpus(struct kvm *kvm)
8038 {
8039 	unsigned int i;
8040 	struct kvm_vcpu *vcpu;
8041 
8042 	/*
8043 	 * Unpin any mmu pages first.
8044 	 */
8045 	kvm_for_each_vcpu(i, vcpu, kvm) {
8046 		kvm_clear_async_pf_completion_queue(vcpu);
8047 		kvm_unload_vcpu_mmu(vcpu);
8048 	}
8049 	kvm_for_each_vcpu(i, vcpu, kvm)
8050 		kvm_arch_vcpu_free(vcpu);
8051 
8052 	mutex_lock(&kvm->lock);
8053 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8054 		kvm->vcpus[i] = NULL;
8055 
8056 	atomic_set(&kvm->online_vcpus, 0);
8057 	mutex_unlock(&kvm->lock);
8058 }
8059 
8060 void kvm_arch_sync_events(struct kvm *kvm)
8061 {
8062 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8063 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8064 	kvm_free_pit(kvm);
8065 }
8066 
8067 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8068 {
8069 	int i, r;
8070 	unsigned long hva;
8071 	struct kvm_memslots *slots = kvm_memslots(kvm);
8072 	struct kvm_memory_slot *slot, old;
8073 
8074 	/* Called with kvm->slots_lock held.  */
8075 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8076 		return -EINVAL;
8077 
8078 	slot = id_to_memslot(slots, id);
8079 	if (size) {
8080 		if (slot->npages)
8081 			return -EEXIST;
8082 
8083 		/*
8084 		 * MAP_SHARED to prevent internal slot pages from being moved
8085 		 * by fork()/COW.
8086 		 */
8087 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8088 			      MAP_SHARED | MAP_ANONYMOUS, 0);
8089 		if (IS_ERR((void *)hva))
8090 			return PTR_ERR((void *)hva);
8091 	} else {
8092 		if (!slot->npages)
8093 			return 0;
8094 
8095 		hva = 0;
8096 	}
8097 
8098 	old = *slot;
8099 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8100 		struct kvm_userspace_memory_region m;
8101 
8102 		m.slot = id | (i << 16);
8103 		m.flags = 0;
8104 		m.guest_phys_addr = gpa;
8105 		m.userspace_addr = hva;
8106 		m.memory_size = size;
8107 		r = __kvm_set_memory_region(kvm, &m);
8108 		if (r < 0)
8109 			return r;
8110 	}
8111 
8112 	if (!size) {
8113 		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8114 		WARN_ON(r < 0);
8115 	}
8116 
8117 	return 0;
8118 }
8119 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8120 
8121 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8122 {
8123 	int r;
8124 
8125 	mutex_lock(&kvm->slots_lock);
8126 	r = __x86_set_memory_region(kvm, id, gpa, size);
8127 	mutex_unlock(&kvm->slots_lock);
8128 
8129 	return r;
8130 }
8131 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8132 
8133 void kvm_arch_destroy_vm(struct kvm *kvm)
8134 {
8135 	if (current->mm == kvm->mm) {
8136 		/*
8137 		 * Free memory regions allocated on behalf of userspace,
8138 		 * unless the the memory map has changed due to process exit
8139 		 * or fd copying.
8140 		 */
8141 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8142 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8143 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8144 	}
8145 	if (kvm_x86_ops->vm_destroy)
8146 		kvm_x86_ops->vm_destroy(kvm);
8147 	kvm_pic_destroy(kvm);
8148 	kvm_ioapic_destroy(kvm);
8149 	kvm_free_vcpus(kvm);
8150 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8151 	kvm_mmu_uninit_vm(kvm);
8152 	kvm_page_track_cleanup(kvm);
8153 }
8154 
8155 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8156 			   struct kvm_memory_slot *dont)
8157 {
8158 	int i;
8159 
8160 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8161 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8162 			kvfree(free->arch.rmap[i]);
8163 			free->arch.rmap[i] = NULL;
8164 		}
8165 		if (i == 0)
8166 			continue;
8167 
8168 		if (!dont || free->arch.lpage_info[i - 1] !=
8169 			     dont->arch.lpage_info[i - 1]) {
8170 			kvfree(free->arch.lpage_info[i - 1]);
8171 			free->arch.lpage_info[i - 1] = NULL;
8172 		}
8173 	}
8174 
8175 	kvm_page_track_free_memslot(free, dont);
8176 }
8177 
8178 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8179 			    unsigned long npages)
8180 {
8181 	int i;
8182 
8183 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8184 		struct kvm_lpage_info *linfo;
8185 		unsigned long ugfn;
8186 		int lpages;
8187 		int level = i + 1;
8188 
8189 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
8190 				      slot->base_gfn, level) + 1;
8191 
8192 		slot->arch.rmap[i] =
8193 			kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8194 		if (!slot->arch.rmap[i])
8195 			goto out_free;
8196 		if (i == 0)
8197 			continue;
8198 
8199 		linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8200 		if (!linfo)
8201 			goto out_free;
8202 
8203 		slot->arch.lpage_info[i - 1] = linfo;
8204 
8205 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8206 			linfo[0].disallow_lpage = 1;
8207 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8208 			linfo[lpages - 1].disallow_lpage = 1;
8209 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
8210 		/*
8211 		 * If the gfn and userspace address are not aligned wrt each
8212 		 * other, or if explicitly asked to, disable large page
8213 		 * support for this slot
8214 		 */
8215 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8216 		    !kvm_largepages_enabled()) {
8217 			unsigned long j;
8218 
8219 			for (j = 0; j < lpages; ++j)
8220 				linfo[j].disallow_lpage = 1;
8221 		}
8222 	}
8223 
8224 	if (kvm_page_track_create_memslot(slot, npages))
8225 		goto out_free;
8226 
8227 	return 0;
8228 
8229 out_free:
8230 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8231 		kvfree(slot->arch.rmap[i]);
8232 		slot->arch.rmap[i] = NULL;
8233 		if (i == 0)
8234 			continue;
8235 
8236 		kvfree(slot->arch.lpage_info[i - 1]);
8237 		slot->arch.lpage_info[i - 1] = NULL;
8238 	}
8239 	return -ENOMEM;
8240 }
8241 
8242 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8243 {
8244 	/*
8245 	 * memslots->generation has been incremented.
8246 	 * mmio generation may have reached its maximum value.
8247 	 */
8248 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8249 }
8250 
8251 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8252 				struct kvm_memory_slot *memslot,
8253 				const struct kvm_userspace_memory_region *mem,
8254 				enum kvm_mr_change change)
8255 {
8256 	return 0;
8257 }
8258 
8259 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8260 				     struct kvm_memory_slot *new)
8261 {
8262 	/* Still write protect RO slot */
8263 	if (new->flags & KVM_MEM_READONLY) {
8264 		kvm_mmu_slot_remove_write_access(kvm, new);
8265 		return;
8266 	}
8267 
8268 	/*
8269 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
8270 	 *
8271 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
8272 	 *
8273 	 *  - KVM_MR_CREATE with dirty logging is disabled
8274 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8275 	 *
8276 	 * The reason is, in case of PML, we need to set D-bit for any slots
8277 	 * with dirty logging disabled in order to eliminate unnecessary GPA
8278 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
8279 	 * guarantees leaving PML enabled during guest's lifetime won't have
8280 	 * any additonal overhead from PML when guest is running with dirty
8281 	 * logging disabled for memory slots.
8282 	 *
8283 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8284 	 * to dirty logging mode.
8285 	 *
8286 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8287 	 *
8288 	 * In case of write protect:
8289 	 *
8290 	 * Write protect all pages for dirty logging.
8291 	 *
8292 	 * All the sptes including the large sptes which point to this
8293 	 * slot are set to readonly. We can not create any new large
8294 	 * spte on this slot until the end of the logging.
8295 	 *
8296 	 * See the comments in fast_page_fault().
8297 	 */
8298 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8299 		if (kvm_x86_ops->slot_enable_log_dirty)
8300 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8301 		else
8302 			kvm_mmu_slot_remove_write_access(kvm, new);
8303 	} else {
8304 		if (kvm_x86_ops->slot_disable_log_dirty)
8305 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8306 	}
8307 }
8308 
8309 void kvm_arch_commit_memory_region(struct kvm *kvm,
8310 				const struct kvm_userspace_memory_region *mem,
8311 				const struct kvm_memory_slot *old,
8312 				const struct kvm_memory_slot *new,
8313 				enum kvm_mr_change change)
8314 {
8315 	int nr_mmu_pages = 0;
8316 
8317 	if (!kvm->arch.n_requested_mmu_pages)
8318 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8319 
8320 	if (nr_mmu_pages)
8321 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8322 
8323 	/*
8324 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
8325 	 * sptes have to be split.  If live migration is successful, the guest
8326 	 * in the source machine will be destroyed and large sptes will be
8327 	 * created in the destination. However, if the guest continues to run
8328 	 * in the source machine (for example if live migration fails), small
8329 	 * sptes will remain around and cause bad performance.
8330 	 *
8331 	 * Scan sptes if dirty logging has been stopped, dropping those
8332 	 * which can be collapsed into a single large-page spte.  Later
8333 	 * page faults will create the large-page sptes.
8334 	 */
8335 	if ((change != KVM_MR_DELETE) &&
8336 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8337 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8338 		kvm_mmu_zap_collapsible_sptes(kvm, new);
8339 
8340 	/*
8341 	 * Set up write protection and/or dirty logging for the new slot.
8342 	 *
8343 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8344 	 * been zapped so no dirty logging staff is needed for old slot. For
8345 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8346 	 * new and it's also covered when dealing with the new slot.
8347 	 *
8348 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
8349 	 */
8350 	if (change != KVM_MR_DELETE)
8351 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8352 }
8353 
8354 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8355 {
8356 	kvm_mmu_invalidate_zap_all_pages(kvm);
8357 }
8358 
8359 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8360 				   struct kvm_memory_slot *slot)
8361 {
8362 	kvm_page_track_flush_slot(kvm, slot);
8363 }
8364 
8365 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8366 {
8367 	if (!list_empty_careful(&vcpu->async_pf.done))
8368 		return true;
8369 
8370 	if (kvm_apic_has_events(vcpu))
8371 		return true;
8372 
8373 	if (vcpu->arch.pv.pv_unhalted)
8374 		return true;
8375 
8376 	if (atomic_read(&vcpu->arch.nmi_queued))
8377 		return true;
8378 
8379 	if (kvm_test_request(KVM_REQ_SMI, vcpu))
8380 		return true;
8381 
8382 	if (kvm_arch_interrupt_allowed(vcpu) &&
8383 	    kvm_cpu_has_interrupt(vcpu))
8384 		return true;
8385 
8386 	if (kvm_hv_has_stimer_pending(vcpu))
8387 		return true;
8388 
8389 	return false;
8390 }
8391 
8392 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8393 {
8394 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8395 }
8396 
8397 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8398 {
8399 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8400 }
8401 
8402 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8403 {
8404 	return kvm_x86_ops->interrupt_allowed(vcpu);
8405 }
8406 
8407 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8408 {
8409 	if (is_64_bit_mode(vcpu))
8410 		return kvm_rip_read(vcpu);
8411 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8412 		     kvm_rip_read(vcpu));
8413 }
8414 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8415 
8416 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8417 {
8418 	return kvm_get_linear_rip(vcpu) == linear_rip;
8419 }
8420 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8421 
8422 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8423 {
8424 	unsigned long rflags;
8425 
8426 	rflags = kvm_x86_ops->get_rflags(vcpu);
8427 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8428 		rflags &= ~X86_EFLAGS_TF;
8429 	return rflags;
8430 }
8431 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8432 
8433 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8434 {
8435 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8436 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8437 		rflags |= X86_EFLAGS_TF;
8438 	kvm_x86_ops->set_rflags(vcpu, rflags);
8439 }
8440 
8441 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8442 {
8443 	__kvm_set_rflags(vcpu, rflags);
8444 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8445 }
8446 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8447 
8448 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8449 {
8450 	int r;
8451 
8452 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8453 	      work->wakeup_all)
8454 		return;
8455 
8456 	r = kvm_mmu_reload(vcpu);
8457 	if (unlikely(r))
8458 		return;
8459 
8460 	if (!vcpu->arch.mmu.direct_map &&
8461 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8462 		return;
8463 
8464 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8465 }
8466 
8467 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8468 {
8469 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8470 }
8471 
8472 static inline u32 kvm_async_pf_next_probe(u32 key)
8473 {
8474 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8475 }
8476 
8477 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8478 {
8479 	u32 key = kvm_async_pf_hash_fn(gfn);
8480 
8481 	while (vcpu->arch.apf.gfns[key] != ~0)
8482 		key = kvm_async_pf_next_probe(key);
8483 
8484 	vcpu->arch.apf.gfns[key] = gfn;
8485 }
8486 
8487 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8488 {
8489 	int i;
8490 	u32 key = kvm_async_pf_hash_fn(gfn);
8491 
8492 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8493 		     (vcpu->arch.apf.gfns[key] != gfn &&
8494 		      vcpu->arch.apf.gfns[key] != ~0); i++)
8495 		key = kvm_async_pf_next_probe(key);
8496 
8497 	return key;
8498 }
8499 
8500 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8501 {
8502 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8503 }
8504 
8505 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8506 {
8507 	u32 i, j, k;
8508 
8509 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8510 	while (true) {
8511 		vcpu->arch.apf.gfns[i] = ~0;
8512 		do {
8513 			j = kvm_async_pf_next_probe(j);
8514 			if (vcpu->arch.apf.gfns[j] == ~0)
8515 				return;
8516 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8517 			/*
8518 			 * k lies cyclically in ]i,j]
8519 			 * |    i.k.j |
8520 			 * |....j i.k.| or  |.k..j i...|
8521 			 */
8522 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8523 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8524 		i = j;
8525 	}
8526 }
8527 
8528 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8529 {
8530 
8531 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8532 				      sizeof(val));
8533 }
8534 
8535 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8536 				     struct kvm_async_pf *work)
8537 {
8538 	struct x86_exception fault;
8539 
8540 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8541 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8542 
8543 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8544 	    (vcpu->arch.apf.send_user_only &&
8545 	     kvm_x86_ops->get_cpl(vcpu) == 0))
8546 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8547 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8548 		fault.vector = PF_VECTOR;
8549 		fault.error_code_valid = true;
8550 		fault.error_code = 0;
8551 		fault.nested_page_fault = false;
8552 		fault.address = work->arch.token;
8553 		kvm_inject_page_fault(vcpu, &fault);
8554 	}
8555 }
8556 
8557 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8558 				 struct kvm_async_pf *work)
8559 {
8560 	struct x86_exception fault;
8561 
8562 	if (work->wakeup_all)
8563 		work->arch.token = ~0; /* broadcast wakeup */
8564 	else
8565 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8566 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8567 
8568 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8569 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8570 		fault.vector = PF_VECTOR;
8571 		fault.error_code_valid = true;
8572 		fault.error_code = 0;
8573 		fault.nested_page_fault = false;
8574 		fault.address = work->arch.token;
8575 		kvm_inject_page_fault(vcpu, &fault);
8576 	}
8577 	vcpu->arch.apf.halted = false;
8578 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8579 }
8580 
8581 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8582 {
8583 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8584 		return true;
8585 	else
8586 		return !kvm_event_needs_reinjection(vcpu) &&
8587 			kvm_x86_ops->interrupt_allowed(vcpu);
8588 }
8589 
8590 void kvm_arch_start_assignment(struct kvm *kvm)
8591 {
8592 	atomic_inc(&kvm->arch.assigned_device_count);
8593 }
8594 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8595 
8596 void kvm_arch_end_assignment(struct kvm *kvm)
8597 {
8598 	atomic_dec(&kvm->arch.assigned_device_count);
8599 }
8600 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8601 
8602 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8603 {
8604 	return atomic_read(&kvm->arch.assigned_device_count);
8605 }
8606 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8607 
8608 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8609 {
8610 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8611 }
8612 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8613 
8614 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8615 {
8616 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8617 }
8618 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8619 
8620 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8621 {
8622 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8623 }
8624 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8625 
8626 bool kvm_arch_has_irq_bypass(void)
8627 {
8628 	return kvm_x86_ops->update_pi_irte != NULL;
8629 }
8630 
8631 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8632 				      struct irq_bypass_producer *prod)
8633 {
8634 	struct kvm_kernel_irqfd *irqfd =
8635 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8636 
8637 	irqfd->producer = prod;
8638 
8639 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8640 					   prod->irq, irqfd->gsi, 1);
8641 }
8642 
8643 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8644 				      struct irq_bypass_producer *prod)
8645 {
8646 	int ret;
8647 	struct kvm_kernel_irqfd *irqfd =
8648 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8649 
8650 	WARN_ON(irqfd->producer != prod);
8651 	irqfd->producer = NULL;
8652 
8653 	/*
8654 	 * When producer of consumer is unregistered, we change back to
8655 	 * remapped mode, so we can re-use the current implementation
8656 	 * when the irq is masked/disabled or the consumer side (KVM
8657 	 * int this case doesn't want to receive the interrupts.
8658 	*/
8659 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8660 	if (ret)
8661 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8662 		       " fails: %d\n", irqfd->consumer.token, ret);
8663 }
8664 
8665 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8666 				   uint32_t guest_irq, bool set)
8667 {
8668 	if (!kvm_x86_ops->update_pi_irte)
8669 		return -EINVAL;
8670 
8671 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8672 }
8673 
8674 bool kvm_vector_hashing_enabled(void)
8675 {
8676 	return vector_hashing;
8677 }
8678 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8679 
8680 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8681 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8682 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8683 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8684 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8685 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8686 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8687 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8688 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8689 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8690 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8691 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8692 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8693 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8694 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8695 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8696 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8697 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8698 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
8699