xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 52fb57e7)
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 "assigned-dev.h"
31 
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
35 #include <linux/fs.h>
36 #include <linux/vmalloc.h>
37 #include <linux/module.h>
38 #include <linux/mman.h>
39 #include <linux/highmem.h>
40 #include <linux/iommu.h>
41 #include <linux/intel-iommu.h>
42 #include <linux/cpufreq.h>
43 #include <linux/user-return-notifier.h>
44 #include <linux/srcu.h>
45 #include <linux/slab.h>
46 #include <linux/perf_event.h>
47 #include <linux/uaccess.h>
48 #include <linux/hash.h>
49 #include <linux/pci.h>
50 #include <linux/timekeeper_internal.h>
51 #include <linux/pvclock_gtod.h>
52 #include <trace/events/kvm.h>
53 
54 #define CREATE_TRACE_POINTS
55 #include "trace.h"
56 
57 #include <asm/debugreg.h>
58 #include <asm/msr.h>
59 #include <asm/desc.h>
60 #include <asm/mtrr.h>
61 #include <asm/mce.h>
62 #include <asm/i387.h>
63 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/xcr.h>
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 
68 #define MAX_IO_MSRS 256
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 
72 #define emul_to_vcpu(ctxt) \
73 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
74 
75 /* EFER defaults:
76  * - enable syscall per default because its emulated by KVM
77  * - enable LME and LMA per default on 64 bit KVM
78  */
79 #ifdef CONFIG_X86_64
80 static
81 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82 #else
83 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
84 #endif
85 
86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 
89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90 static void process_nmi(struct kvm_vcpu *vcpu);
91 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
92 
93 struct kvm_x86_ops *kvm_x86_ops;
94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
95 
96 static bool ignore_msrs = 0;
97 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
98 
99 unsigned int min_timer_period_us = 500;
100 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
101 
102 bool kvm_has_tsc_control;
103 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
104 u32  kvm_max_guest_tsc_khz;
105 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
106 
107 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
108 static u32 tsc_tolerance_ppm = 250;
109 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
110 
111 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
112 unsigned int lapic_timer_advance_ns = 0;
113 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
114 
115 static bool backwards_tsc_observed = false;
116 
117 #define KVM_NR_SHARED_MSRS 16
118 
119 struct kvm_shared_msrs_global {
120 	int nr;
121 	u32 msrs[KVM_NR_SHARED_MSRS];
122 };
123 
124 struct kvm_shared_msrs {
125 	struct user_return_notifier urn;
126 	bool registered;
127 	struct kvm_shared_msr_values {
128 		u64 host;
129 		u64 curr;
130 	} values[KVM_NR_SHARED_MSRS];
131 };
132 
133 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
134 static struct kvm_shared_msrs __percpu *shared_msrs;
135 
136 struct kvm_stats_debugfs_item debugfs_entries[] = {
137 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
138 	{ "pf_guest", VCPU_STAT(pf_guest) },
139 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
140 	{ "invlpg", VCPU_STAT(invlpg) },
141 	{ "exits", VCPU_STAT(exits) },
142 	{ "io_exits", VCPU_STAT(io_exits) },
143 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
144 	{ "signal_exits", VCPU_STAT(signal_exits) },
145 	{ "irq_window", VCPU_STAT(irq_window_exits) },
146 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
147 	{ "halt_exits", VCPU_STAT(halt_exits) },
148 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
149 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
150 	{ "hypercalls", VCPU_STAT(hypercalls) },
151 	{ "request_irq", VCPU_STAT(request_irq_exits) },
152 	{ "irq_exits", VCPU_STAT(irq_exits) },
153 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
154 	{ "efer_reload", VCPU_STAT(efer_reload) },
155 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
156 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
157 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
158 	{ "irq_injections", VCPU_STAT(irq_injections) },
159 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
160 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
161 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
162 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
163 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
164 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
165 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
166 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
167 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
168 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
169 	{ "largepages", VM_STAT(lpages) },
170 	{ NULL }
171 };
172 
173 u64 __read_mostly host_xcr0;
174 
175 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
176 
177 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
178 {
179 	int i;
180 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
181 		vcpu->arch.apf.gfns[i] = ~0;
182 }
183 
184 static void kvm_on_user_return(struct user_return_notifier *urn)
185 {
186 	unsigned slot;
187 	struct kvm_shared_msrs *locals
188 		= container_of(urn, struct kvm_shared_msrs, urn);
189 	struct kvm_shared_msr_values *values;
190 
191 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
192 		values = &locals->values[slot];
193 		if (values->host != values->curr) {
194 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
195 			values->curr = values->host;
196 		}
197 	}
198 	locals->registered = false;
199 	user_return_notifier_unregister(urn);
200 }
201 
202 static void shared_msr_update(unsigned slot, u32 msr)
203 {
204 	u64 value;
205 	unsigned int cpu = smp_processor_id();
206 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
207 
208 	/* only read, and nobody should modify it at this time,
209 	 * so don't need lock */
210 	if (slot >= shared_msrs_global.nr) {
211 		printk(KERN_ERR "kvm: invalid MSR slot!");
212 		return;
213 	}
214 	rdmsrl_safe(msr, &value);
215 	smsr->values[slot].host = value;
216 	smsr->values[slot].curr = value;
217 }
218 
219 void kvm_define_shared_msr(unsigned slot, u32 msr)
220 {
221 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
222 	if (slot >= shared_msrs_global.nr)
223 		shared_msrs_global.nr = slot + 1;
224 	shared_msrs_global.msrs[slot] = msr;
225 	/* we need ensured the shared_msr_global have been updated */
226 	smp_wmb();
227 }
228 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
229 
230 static void kvm_shared_msr_cpu_online(void)
231 {
232 	unsigned i;
233 
234 	for (i = 0; i < shared_msrs_global.nr; ++i)
235 		shared_msr_update(i, shared_msrs_global.msrs[i]);
236 }
237 
238 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
239 {
240 	unsigned int cpu = smp_processor_id();
241 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
242 	int err;
243 
244 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
245 		return 0;
246 	smsr->values[slot].curr = value;
247 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
248 	if (err)
249 		return 1;
250 
251 	if (!smsr->registered) {
252 		smsr->urn.on_user_return = kvm_on_user_return;
253 		user_return_notifier_register(&smsr->urn);
254 		smsr->registered = true;
255 	}
256 	return 0;
257 }
258 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
259 
260 static void drop_user_return_notifiers(void)
261 {
262 	unsigned int cpu = smp_processor_id();
263 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
264 
265 	if (smsr->registered)
266 		kvm_on_user_return(&smsr->urn);
267 }
268 
269 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
270 {
271 	return vcpu->arch.apic_base;
272 }
273 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
274 
275 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
276 {
277 	u64 old_state = vcpu->arch.apic_base &
278 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
279 	u64 new_state = msr_info->data &
280 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
282 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
283 
284 	if (!msr_info->host_initiated &&
285 	    ((msr_info->data & reserved_bits) != 0 ||
286 	     new_state == X2APIC_ENABLE ||
287 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
288 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
289 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
290 	      old_state == 0)))
291 		return 1;
292 
293 	kvm_lapic_set_base(vcpu, msr_info->data);
294 	return 0;
295 }
296 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
297 
298 asmlinkage __visible void kvm_spurious_fault(void)
299 {
300 	/* Fault while not rebooting.  We want the trace. */
301 	BUG();
302 }
303 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
304 
305 #define EXCPT_BENIGN		0
306 #define EXCPT_CONTRIBUTORY	1
307 #define EXCPT_PF		2
308 
309 static int exception_class(int vector)
310 {
311 	switch (vector) {
312 	case PF_VECTOR:
313 		return EXCPT_PF;
314 	case DE_VECTOR:
315 	case TS_VECTOR:
316 	case NP_VECTOR:
317 	case SS_VECTOR:
318 	case GP_VECTOR:
319 		return EXCPT_CONTRIBUTORY;
320 	default:
321 		break;
322 	}
323 	return EXCPT_BENIGN;
324 }
325 
326 #define EXCPT_FAULT		0
327 #define EXCPT_TRAP		1
328 #define EXCPT_ABORT		2
329 #define EXCPT_INTERRUPT		3
330 
331 static int exception_type(int vector)
332 {
333 	unsigned int mask;
334 
335 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
336 		return EXCPT_INTERRUPT;
337 
338 	mask = 1 << vector;
339 
340 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
341 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
342 		return EXCPT_TRAP;
343 
344 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
345 		return EXCPT_ABORT;
346 
347 	/* Reserved exceptions will result in fault */
348 	return EXCPT_FAULT;
349 }
350 
351 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
352 		unsigned nr, bool has_error, u32 error_code,
353 		bool reinject)
354 {
355 	u32 prev_nr;
356 	int class1, class2;
357 
358 	kvm_make_request(KVM_REQ_EVENT, vcpu);
359 
360 	if (!vcpu->arch.exception.pending) {
361 	queue:
362 		if (has_error && !is_protmode(vcpu))
363 			has_error = false;
364 		vcpu->arch.exception.pending = true;
365 		vcpu->arch.exception.has_error_code = has_error;
366 		vcpu->arch.exception.nr = nr;
367 		vcpu->arch.exception.error_code = error_code;
368 		vcpu->arch.exception.reinject = reinject;
369 		return;
370 	}
371 
372 	/* to check exception */
373 	prev_nr = vcpu->arch.exception.nr;
374 	if (prev_nr == DF_VECTOR) {
375 		/* triple fault -> shutdown */
376 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
377 		return;
378 	}
379 	class1 = exception_class(prev_nr);
380 	class2 = exception_class(nr);
381 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
382 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
383 		/* generate double fault per SDM Table 5-5 */
384 		vcpu->arch.exception.pending = true;
385 		vcpu->arch.exception.has_error_code = true;
386 		vcpu->arch.exception.nr = DF_VECTOR;
387 		vcpu->arch.exception.error_code = 0;
388 	} else
389 		/* replace previous exception with a new one in a hope
390 		   that instruction re-execution will regenerate lost
391 		   exception */
392 		goto queue;
393 }
394 
395 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
396 {
397 	kvm_multiple_exception(vcpu, nr, false, 0, false);
398 }
399 EXPORT_SYMBOL_GPL(kvm_queue_exception);
400 
401 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
402 {
403 	kvm_multiple_exception(vcpu, nr, false, 0, true);
404 }
405 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
406 
407 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
408 {
409 	if (err)
410 		kvm_inject_gp(vcpu, 0);
411 	else
412 		kvm_x86_ops->skip_emulated_instruction(vcpu);
413 }
414 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
415 
416 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
417 {
418 	++vcpu->stat.pf_guest;
419 	vcpu->arch.cr2 = fault->address;
420 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
421 }
422 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
423 
424 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
425 {
426 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
427 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
428 	else
429 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
430 
431 	return fault->nested_page_fault;
432 }
433 
434 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
435 {
436 	atomic_inc(&vcpu->arch.nmi_queued);
437 	kvm_make_request(KVM_REQ_NMI, vcpu);
438 }
439 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
440 
441 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
442 {
443 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
444 }
445 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
446 
447 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
448 {
449 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
450 }
451 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
452 
453 /*
454  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
455  * a #GP and return false.
456  */
457 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
458 {
459 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
460 		return true;
461 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
462 	return false;
463 }
464 EXPORT_SYMBOL_GPL(kvm_require_cpl);
465 
466 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
467 {
468 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
469 		return true;
470 
471 	kvm_queue_exception(vcpu, UD_VECTOR);
472 	return false;
473 }
474 EXPORT_SYMBOL_GPL(kvm_require_dr);
475 
476 /*
477  * This function will be used to read from the physical memory of the currently
478  * running guest. The difference to kvm_read_guest_page is that this function
479  * can read from guest physical or from the guest's guest physical memory.
480  */
481 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
482 			    gfn_t ngfn, void *data, int offset, int len,
483 			    u32 access)
484 {
485 	struct x86_exception exception;
486 	gfn_t real_gfn;
487 	gpa_t ngpa;
488 
489 	ngpa     = gfn_to_gpa(ngfn);
490 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
491 	if (real_gfn == UNMAPPED_GVA)
492 		return -EFAULT;
493 
494 	real_gfn = gpa_to_gfn(real_gfn);
495 
496 	return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
497 }
498 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
499 
500 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
501 			       void *data, int offset, int len, u32 access)
502 {
503 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
504 				       data, offset, len, access);
505 }
506 
507 /*
508  * Load the pae pdptrs.  Return true is they are all valid.
509  */
510 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
511 {
512 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
513 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
514 	int i;
515 	int ret;
516 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
517 
518 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
519 				      offset * sizeof(u64), sizeof(pdpte),
520 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
521 	if (ret < 0) {
522 		ret = 0;
523 		goto out;
524 	}
525 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
526 		if (is_present_gpte(pdpte[i]) &&
527 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
528 			ret = 0;
529 			goto out;
530 		}
531 	}
532 	ret = 1;
533 
534 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
535 	__set_bit(VCPU_EXREG_PDPTR,
536 		  (unsigned long *)&vcpu->arch.regs_avail);
537 	__set_bit(VCPU_EXREG_PDPTR,
538 		  (unsigned long *)&vcpu->arch.regs_dirty);
539 out:
540 
541 	return ret;
542 }
543 EXPORT_SYMBOL_GPL(load_pdptrs);
544 
545 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
546 {
547 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
548 	bool changed = true;
549 	int offset;
550 	gfn_t gfn;
551 	int r;
552 
553 	if (is_long_mode(vcpu) || !is_pae(vcpu))
554 		return false;
555 
556 	if (!test_bit(VCPU_EXREG_PDPTR,
557 		      (unsigned long *)&vcpu->arch.regs_avail))
558 		return true;
559 
560 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
561 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
562 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
563 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
564 	if (r < 0)
565 		goto out;
566 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
567 out:
568 
569 	return changed;
570 }
571 
572 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
573 {
574 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
575 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
576 				    X86_CR0_CD | X86_CR0_NW;
577 
578 	cr0 |= X86_CR0_ET;
579 
580 #ifdef CONFIG_X86_64
581 	if (cr0 & 0xffffffff00000000UL)
582 		return 1;
583 #endif
584 
585 	cr0 &= ~CR0_RESERVED_BITS;
586 
587 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
588 		return 1;
589 
590 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
591 		return 1;
592 
593 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
594 #ifdef CONFIG_X86_64
595 		if ((vcpu->arch.efer & EFER_LME)) {
596 			int cs_db, cs_l;
597 
598 			if (!is_pae(vcpu))
599 				return 1;
600 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
601 			if (cs_l)
602 				return 1;
603 		} else
604 #endif
605 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
606 						 kvm_read_cr3(vcpu)))
607 			return 1;
608 	}
609 
610 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
611 		return 1;
612 
613 	kvm_x86_ops->set_cr0(vcpu, cr0);
614 
615 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
616 		kvm_clear_async_pf_completion_queue(vcpu);
617 		kvm_async_pf_hash_reset(vcpu);
618 	}
619 
620 	if ((cr0 ^ old_cr0) & update_bits)
621 		kvm_mmu_reset_context(vcpu);
622 	return 0;
623 }
624 EXPORT_SYMBOL_GPL(kvm_set_cr0);
625 
626 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
627 {
628 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
629 }
630 EXPORT_SYMBOL_GPL(kvm_lmsw);
631 
632 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
633 {
634 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
635 			!vcpu->guest_xcr0_loaded) {
636 		/* kvm_set_xcr() also depends on this */
637 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
638 		vcpu->guest_xcr0_loaded = 1;
639 	}
640 }
641 
642 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
643 {
644 	if (vcpu->guest_xcr0_loaded) {
645 		if (vcpu->arch.xcr0 != host_xcr0)
646 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
647 		vcpu->guest_xcr0_loaded = 0;
648 	}
649 }
650 
651 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
652 {
653 	u64 xcr0 = xcr;
654 	u64 old_xcr0 = vcpu->arch.xcr0;
655 	u64 valid_bits;
656 
657 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
658 	if (index != XCR_XFEATURE_ENABLED_MASK)
659 		return 1;
660 	if (!(xcr0 & XSTATE_FP))
661 		return 1;
662 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
663 		return 1;
664 
665 	/*
666 	 * Do not allow the guest to set bits that we do not support
667 	 * saving.  However, xcr0 bit 0 is always set, even if the
668 	 * emulated CPU does not support XSAVE (see fx_init).
669 	 */
670 	valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
671 	if (xcr0 & ~valid_bits)
672 		return 1;
673 
674 	if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
675 		return 1;
676 
677 	if (xcr0 & XSTATE_AVX512) {
678 		if (!(xcr0 & XSTATE_YMM))
679 			return 1;
680 		if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
681 			return 1;
682 	}
683 	kvm_put_guest_xcr0(vcpu);
684 	vcpu->arch.xcr0 = xcr0;
685 
686 	if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
687 		kvm_update_cpuid(vcpu);
688 	return 0;
689 }
690 
691 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
692 {
693 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
694 	    __kvm_set_xcr(vcpu, index, xcr)) {
695 		kvm_inject_gp(vcpu, 0);
696 		return 1;
697 	}
698 	return 0;
699 }
700 EXPORT_SYMBOL_GPL(kvm_set_xcr);
701 
702 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
703 {
704 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
705 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
706 				   X86_CR4_SMEP | X86_CR4_SMAP;
707 
708 	if (cr4 & CR4_RESERVED_BITS)
709 		return 1;
710 
711 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
712 		return 1;
713 
714 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
715 		return 1;
716 
717 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
718 		return 1;
719 
720 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
721 		return 1;
722 
723 	if (is_long_mode(vcpu)) {
724 		if (!(cr4 & X86_CR4_PAE))
725 			return 1;
726 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
727 		   && ((cr4 ^ old_cr4) & pdptr_bits)
728 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
729 				   kvm_read_cr3(vcpu)))
730 		return 1;
731 
732 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
733 		if (!guest_cpuid_has_pcid(vcpu))
734 			return 1;
735 
736 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
737 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
738 			return 1;
739 	}
740 
741 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
742 		return 1;
743 
744 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
745 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
746 		kvm_mmu_reset_context(vcpu);
747 
748 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
749 		kvm_update_cpuid(vcpu);
750 
751 	return 0;
752 }
753 EXPORT_SYMBOL_GPL(kvm_set_cr4);
754 
755 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
756 {
757 #ifdef CONFIG_X86_64
758 	cr3 &= ~CR3_PCID_INVD;
759 #endif
760 
761 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
762 		kvm_mmu_sync_roots(vcpu);
763 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
764 		return 0;
765 	}
766 
767 	if (is_long_mode(vcpu)) {
768 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
769 			return 1;
770 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
771 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
772 		return 1;
773 
774 	vcpu->arch.cr3 = cr3;
775 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
776 	kvm_mmu_new_cr3(vcpu);
777 	return 0;
778 }
779 EXPORT_SYMBOL_GPL(kvm_set_cr3);
780 
781 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
782 {
783 	if (cr8 & CR8_RESERVED_BITS)
784 		return 1;
785 	if (irqchip_in_kernel(vcpu->kvm))
786 		kvm_lapic_set_tpr(vcpu, cr8);
787 	else
788 		vcpu->arch.cr8 = cr8;
789 	return 0;
790 }
791 EXPORT_SYMBOL_GPL(kvm_set_cr8);
792 
793 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
794 {
795 	if (irqchip_in_kernel(vcpu->kvm))
796 		return kvm_lapic_get_cr8(vcpu);
797 	else
798 		return vcpu->arch.cr8;
799 }
800 EXPORT_SYMBOL_GPL(kvm_get_cr8);
801 
802 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
803 {
804 	int i;
805 
806 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
807 		for (i = 0; i < KVM_NR_DB_REGS; i++)
808 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
809 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
810 	}
811 }
812 
813 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
814 {
815 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
816 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
817 }
818 
819 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
820 {
821 	unsigned long dr7;
822 
823 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
824 		dr7 = vcpu->arch.guest_debug_dr7;
825 	else
826 		dr7 = vcpu->arch.dr7;
827 	kvm_x86_ops->set_dr7(vcpu, dr7);
828 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
829 	if (dr7 & DR7_BP_EN_MASK)
830 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
831 }
832 
833 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
834 {
835 	u64 fixed = DR6_FIXED_1;
836 
837 	if (!guest_cpuid_has_rtm(vcpu))
838 		fixed |= DR6_RTM;
839 	return fixed;
840 }
841 
842 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
843 {
844 	switch (dr) {
845 	case 0 ... 3:
846 		vcpu->arch.db[dr] = val;
847 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
848 			vcpu->arch.eff_db[dr] = val;
849 		break;
850 	case 4:
851 		/* fall through */
852 	case 6:
853 		if (val & 0xffffffff00000000ULL)
854 			return -1; /* #GP */
855 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
856 		kvm_update_dr6(vcpu);
857 		break;
858 	case 5:
859 		/* fall through */
860 	default: /* 7 */
861 		if (val & 0xffffffff00000000ULL)
862 			return -1; /* #GP */
863 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
864 		kvm_update_dr7(vcpu);
865 		break;
866 	}
867 
868 	return 0;
869 }
870 
871 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
872 {
873 	if (__kvm_set_dr(vcpu, dr, val)) {
874 		kvm_inject_gp(vcpu, 0);
875 		return 1;
876 	}
877 	return 0;
878 }
879 EXPORT_SYMBOL_GPL(kvm_set_dr);
880 
881 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
882 {
883 	switch (dr) {
884 	case 0 ... 3:
885 		*val = vcpu->arch.db[dr];
886 		break;
887 	case 4:
888 		/* fall through */
889 	case 6:
890 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
891 			*val = vcpu->arch.dr6;
892 		else
893 			*val = kvm_x86_ops->get_dr6(vcpu);
894 		break;
895 	case 5:
896 		/* fall through */
897 	default: /* 7 */
898 		*val = vcpu->arch.dr7;
899 		break;
900 	}
901 	return 0;
902 }
903 EXPORT_SYMBOL_GPL(kvm_get_dr);
904 
905 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
906 {
907 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
908 	u64 data;
909 	int err;
910 
911 	err = kvm_pmu_read_pmc(vcpu, ecx, &data);
912 	if (err)
913 		return err;
914 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
915 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
916 	return err;
917 }
918 EXPORT_SYMBOL_GPL(kvm_rdpmc);
919 
920 /*
921  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
922  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
923  *
924  * This list is modified at module load time to reflect the
925  * capabilities of the host cpu. This capabilities test skips MSRs that are
926  * kvm-specific. Those are put in the beginning of the list.
927  */
928 
929 #define KVM_SAVE_MSRS_BEGIN	12
930 static u32 msrs_to_save[] = {
931 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
932 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
933 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
934 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
935 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
936 	MSR_KVM_PV_EOI_EN,
937 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
938 	MSR_STAR,
939 #ifdef CONFIG_X86_64
940 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
941 #endif
942 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
943 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
944 };
945 
946 static unsigned num_msrs_to_save;
947 
948 static const u32 emulated_msrs[] = {
949 	MSR_IA32_TSC_ADJUST,
950 	MSR_IA32_TSCDEADLINE,
951 	MSR_IA32_MISC_ENABLE,
952 	MSR_IA32_MCG_STATUS,
953 	MSR_IA32_MCG_CTL,
954 };
955 
956 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
957 {
958 	if (efer & efer_reserved_bits)
959 		return false;
960 
961 	if (efer & EFER_FFXSR) {
962 		struct kvm_cpuid_entry2 *feat;
963 
964 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
965 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
966 			return false;
967 	}
968 
969 	if (efer & EFER_SVME) {
970 		struct kvm_cpuid_entry2 *feat;
971 
972 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
973 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
974 			return false;
975 	}
976 
977 	return true;
978 }
979 EXPORT_SYMBOL_GPL(kvm_valid_efer);
980 
981 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
982 {
983 	u64 old_efer = vcpu->arch.efer;
984 
985 	if (!kvm_valid_efer(vcpu, efer))
986 		return 1;
987 
988 	if (is_paging(vcpu)
989 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
990 		return 1;
991 
992 	efer &= ~EFER_LMA;
993 	efer |= vcpu->arch.efer & EFER_LMA;
994 
995 	kvm_x86_ops->set_efer(vcpu, efer);
996 
997 	/* Update reserved bits */
998 	if ((efer ^ old_efer) & EFER_NX)
999 		kvm_mmu_reset_context(vcpu);
1000 
1001 	return 0;
1002 }
1003 
1004 void kvm_enable_efer_bits(u64 mask)
1005 {
1006        efer_reserved_bits &= ~mask;
1007 }
1008 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1009 
1010 /*
1011  * Writes msr value into into the appropriate "register".
1012  * Returns 0 on success, non-0 otherwise.
1013  * Assumes vcpu_load() was already called.
1014  */
1015 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1016 {
1017 	switch (msr->index) {
1018 	case MSR_FS_BASE:
1019 	case MSR_GS_BASE:
1020 	case MSR_KERNEL_GS_BASE:
1021 	case MSR_CSTAR:
1022 	case MSR_LSTAR:
1023 		if (is_noncanonical_address(msr->data))
1024 			return 1;
1025 		break;
1026 	case MSR_IA32_SYSENTER_EIP:
1027 	case MSR_IA32_SYSENTER_ESP:
1028 		/*
1029 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1030 		 * non-canonical address is written on Intel but not on
1031 		 * AMD (which ignores the top 32-bits, because it does
1032 		 * not implement 64-bit SYSENTER).
1033 		 *
1034 		 * 64-bit code should hence be able to write a non-canonical
1035 		 * value on AMD.  Making the address canonical ensures that
1036 		 * vmentry does not fail on Intel after writing a non-canonical
1037 		 * value, and that something deterministic happens if the guest
1038 		 * invokes 64-bit SYSENTER.
1039 		 */
1040 		msr->data = get_canonical(msr->data);
1041 	}
1042 	return kvm_x86_ops->set_msr(vcpu, msr);
1043 }
1044 EXPORT_SYMBOL_GPL(kvm_set_msr);
1045 
1046 /*
1047  * Adapt set_msr() to msr_io()'s calling convention
1048  */
1049 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1050 {
1051 	struct msr_data msr;
1052 
1053 	msr.data = *data;
1054 	msr.index = index;
1055 	msr.host_initiated = true;
1056 	return kvm_set_msr(vcpu, &msr);
1057 }
1058 
1059 #ifdef CONFIG_X86_64
1060 struct pvclock_gtod_data {
1061 	seqcount_t	seq;
1062 
1063 	struct { /* extract of a clocksource struct */
1064 		int vclock_mode;
1065 		cycle_t	cycle_last;
1066 		cycle_t	mask;
1067 		u32	mult;
1068 		u32	shift;
1069 	} clock;
1070 
1071 	u64		boot_ns;
1072 	u64		nsec_base;
1073 };
1074 
1075 static struct pvclock_gtod_data pvclock_gtod_data;
1076 
1077 static void update_pvclock_gtod(struct timekeeper *tk)
1078 {
1079 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1080 	u64 boot_ns;
1081 
1082 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1083 
1084 	write_seqcount_begin(&vdata->seq);
1085 
1086 	/* copy pvclock gtod data */
1087 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1088 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1089 	vdata->clock.mask		= tk->tkr_mono.mask;
1090 	vdata->clock.mult		= tk->tkr_mono.mult;
1091 	vdata->clock.shift		= tk->tkr_mono.shift;
1092 
1093 	vdata->boot_ns			= boot_ns;
1094 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1095 
1096 	write_seqcount_end(&vdata->seq);
1097 }
1098 #endif
1099 
1100 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1101 {
1102 	/*
1103 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1104 	 * vcpu_enter_guest.  This function is only called from
1105 	 * the physical CPU that is running vcpu.
1106 	 */
1107 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1108 }
1109 
1110 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1111 {
1112 	int version;
1113 	int r;
1114 	struct pvclock_wall_clock wc;
1115 	struct timespec boot;
1116 
1117 	if (!wall_clock)
1118 		return;
1119 
1120 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1121 	if (r)
1122 		return;
1123 
1124 	if (version & 1)
1125 		++version;  /* first time write, random junk */
1126 
1127 	++version;
1128 
1129 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1130 
1131 	/*
1132 	 * The guest calculates current wall clock time by adding
1133 	 * system time (updated by kvm_guest_time_update below) to the
1134 	 * wall clock specified here.  guest system time equals host
1135 	 * system time for us, thus we must fill in host boot time here.
1136 	 */
1137 	getboottime(&boot);
1138 
1139 	if (kvm->arch.kvmclock_offset) {
1140 		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1141 		boot = timespec_sub(boot, ts);
1142 	}
1143 	wc.sec = boot.tv_sec;
1144 	wc.nsec = boot.tv_nsec;
1145 	wc.version = version;
1146 
1147 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1148 
1149 	version++;
1150 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1151 }
1152 
1153 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1154 {
1155 	uint32_t quotient, remainder;
1156 
1157 	/* Don't try to replace with do_div(), this one calculates
1158 	 * "(dividend << 32) / divisor" */
1159 	__asm__ ( "divl %4"
1160 		  : "=a" (quotient), "=d" (remainder)
1161 		  : "0" (0), "1" (dividend), "r" (divisor) );
1162 	return quotient;
1163 }
1164 
1165 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1166 			       s8 *pshift, u32 *pmultiplier)
1167 {
1168 	uint64_t scaled64;
1169 	int32_t  shift = 0;
1170 	uint64_t tps64;
1171 	uint32_t tps32;
1172 
1173 	tps64 = base_khz * 1000LL;
1174 	scaled64 = scaled_khz * 1000LL;
1175 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1176 		tps64 >>= 1;
1177 		shift--;
1178 	}
1179 
1180 	tps32 = (uint32_t)tps64;
1181 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1182 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1183 			scaled64 >>= 1;
1184 		else
1185 			tps32 <<= 1;
1186 		shift++;
1187 	}
1188 
1189 	*pshift = shift;
1190 	*pmultiplier = div_frac(scaled64, tps32);
1191 
1192 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1193 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1194 }
1195 
1196 static inline u64 get_kernel_ns(void)
1197 {
1198 	return ktime_get_boot_ns();
1199 }
1200 
1201 #ifdef CONFIG_X86_64
1202 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1203 #endif
1204 
1205 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1206 static unsigned long max_tsc_khz;
1207 
1208 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1209 {
1210 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1211 				   vcpu->arch.virtual_tsc_shift);
1212 }
1213 
1214 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1215 {
1216 	u64 v = (u64)khz * (1000000 + ppm);
1217 	do_div(v, 1000000);
1218 	return v;
1219 }
1220 
1221 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1222 {
1223 	u32 thresh_lo, thresh_hi;
1224 	int use_scaling = 0;
1225 
1226 	/* tsc_khz can be zero if TSC calibration fails */
1227 	if (this_tsc_khz == 0)
1228 		return;
1229 
1230 	/* Compute a scale to convert nanoseconds in TSC cycles */
1231 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1232 			   &vcpu->arch.virtual_tsc_shift,
1233 			   &vcpu->arch.virtual_tsc_mult);
1234 	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1235 
1236 	/*
1237 	 * Compute the variation in TSC rate which is acceptable
1238 	 * within the range of tolerance and decide if the
1239 	 * rate being applied is within that bounds of the hardware
1240 	 * rate.  If so, no scaling or compensation need be done.
1241 	 */
1242 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1243 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1244 	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1245 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1246 		use_scaling = 1;
1247 	}
1248 	kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1249 }
1250 
1251 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1252 {
1253 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1254 				      vcpu->arch.virtual_tsc_mult,
1255 				      vcpu->arch.virtual_tsc_shift);
1256 	tsc += vcpu->arch.this_tsc_write;
1257 	return tsc;
1258 }
1259 
1260 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1261 {
1262 #ifdef CONFIG_X86_64
1263 	bool vcpus_matched;
1264 	struct kvm_arch *ka = &vcpu->kvm->arch;
1265 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1266 
1267 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1268 			 atomic_read(&vcpu->kvm->online_vcpus));
1269 
1270 	/*
1271 	 * Once the masterclock is enabled, always perform request in
1272 	 * order to update it.
1273 	 *
1274 	 * In order to enable masterclock, the host clocksource must be TSC
1275 	 * and the vcpus need to have matched TSCs.  When that happens,
1276 	 * perform request to enable masterclock.
1277 	 */
1278 	if (ka->use_master_clock ||
1279 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1280 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1281 
1282 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1283 			    atomic_read(&vcpu->kvm->online_vcpus),
1284 		            ka->use_master_clock, gtod->clock.vclock_mode);
1285 #endif
1286 }
1287 
1288 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1289 {
1290 	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1291 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1292 }
1293 
1294 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1295 {
1296 	struct kvm *kvm = vcpu->kvm;
1297 	u64 offset, ns, elapsed;
1298 	unsigned long flags;
1299 	s64 usdiff;
1300 	bool matched;
1301 	bool already_matched;
1302 	u64 data = msr->data;
1303 
1304 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1305 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1306 	ns = get_kernel_ns();
1307 	elapsed = ns - kvm->arch.last_tsc_nsec;
1308 
1309 	if (vcpu->arch.virtual_tsc_khz) {
1310 		int faulted = 0;
1311 
1312 		/* n.b - signed multiplication and division required */
1313 		usdiff = data - kvm->arch.last_tsc_write;
1314 #ifdef CONFIG_X86_64
1315 		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1316 #else
1317 		/* do_div() only does unsigned */
1318 		asm("1: idivl %[divisor]\n"
1319 		    "2: xor %%edx, %%edx\n"
1320 		    "   movl $0, %[faulted]\n"
1321 		    "3:\n"
1322 		    ".section .fixup,\"ax\"\n"
1323 		    "4: movl $1, %[faulted]\n"
1324 		    "   jmp  3b\n"
1325 		    ".previous\n"
1326 
1327 		_ASM_EXTABLE(1b, 4b)
1328 
1329 		: "=A"(usdiff), [faulted] "=r" (faulted)
1330 		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1331 
1332 #endif
1333 		do_div(elapsed, 1000);
1334 		usdiff -= elapsed;
1335 		if (usdiff < 0)
1336 			usdiff = -usdiff;
1337 
1338 		/* idivl overflow => difference is larger than USEC_PER_SEC */
1339 		if (faulted)
1340 			usdiff = USEC_PER_SEC;
1341 	} else
1342 		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1343 
1344 	/*
1345 	 * Special case: TSC write with a small delta (1 second) of virtual
1346 	 * cycle time against real time is interpreted as an attempt to
1347 	 * synchronize the CPU.
1348          *
1349 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1350 	 * TSC, we add elapsed time in this computation.  We could let the
1351 	 * compensation code attempt to catch up if we fall behind, but
1352 	 * it's better to try to match offsets from the beginning.
1353          */
1354 	if (usdiff < USEC_PER_SEC &&
1355 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1356 		if (!check_tsc_unstable()) {
1357 			offset = kvm->arch.cur_tsc_offset;
1358 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1359 		} else {
1360 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1361 			data += delta;
1362 			offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1363 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1364 		}
1365 		matched = true;
1366 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1367 	} else {
1368 		/*
1369 		 * We split periods of matched TSC writes into generations.
1370 		 * For each generation, we track the original measured
1371 		 * nanosecond time, offset, and write, so if TSCs are in
1372 		 * sync, we can match exact offset, and if not, we can match
1373 		 * exact software computation in compute_guest_tsc()
1374 		 *
1375 		 * These values are tracked in kvm->arch.cur_xxx variables.
1376 		 */
1377 		kvm->arch.cur_tsc_generation++;
1378 		kvm->arch.cur_tsc_nsec = ns;
1379 		kvm->arch.cur_tsc_write = data;
1380 		kvm->arch.cur_tsc_offset = offset;
1381 		matched = false;
1382 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1383 			 kvm->arch.cur_tsc_generation, data);
1384 	}
1385 
1386 	/*
1387 	 * We also track th most recent recorded KHZ, write and time to
1388 	 * allow the matching interval to be extended at each write.
1389 	 */
1390 	kvm->arch.last_tsc_nsec = ns;
1391 	kvm->arch.last_tsc_write = data;
1392 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1393 
1394 	vcpu->arch.last_guest_tsc = data;
1395 
1396 	/* Keep track of which generation this VCPU has synchronized to */
1397 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1398 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1399 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1400 
1401 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1402 		update_ia32_tsc_adjust_msr(vcpu, offset);
1403 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1404 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1405 
1406 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1407 	if (!matched) {
1408 		kvm->arch.nr_vcpus_matched_tsc = 0;
1409 	} else if (!already_matched) {
1410 		kvm->arch.nr_vcpus_matched_tsc++;
1411 	}
1412 
1413 	kvm_track_tsc_matching(vcpu);
1414 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1415 }
1416 
1417 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1418 
1419 #ifdef CONFIG_X86_64
1420 
1421 static cycle_t read_tsc(void)
1422 {
1423 	cycle_t ret;
1424 	u64 last;
1425 
1426 	/*
1427 	 * Empirically, a fence (of type that depends on the CPU)
1428 	 * before rdtsc is enough to ensure that rdtsc is ordered
1429 	 * with respect to loads.  The various CPU manuals are unclear
1430 	 * as to whether rdtsc can be reordered with later loads,
1431 	 * but no one has ever seen it happen.
1432 	 */
1433 	rdtsc_barrier();
1434 	ret = (cycle_t)vget_cycles();
1435 
1436 	last = pvclock_gtod_data.clock.cycle_last;
1437 
1438 	if (likely(ret >= last))
1439 		return ret;
1440 
1441 	/*
1442 	 * GCC likes to generate cmov here, but this branch is extremely
1443 	 * predictable (it's just a funciton of time and the likely is
1444 	 * very likely) and there's a data dependence, so force GCC
1445 	 * to generate a branch instead.  I don't barrier() because
1446 	 * we don't actually need a barrier, and if this function
1447 	 * ever gets inlined it will generate worse code.
1448 	 */
1449 	asm volatile ("");
1450 	return last;
1451 }
1452 
1453 static inline u64 vgettsc(cycle_t *cycle_now)
1454 {
1455 	long v;
1456 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1457 
1458 	*cycle_now = read_tsc();
1459 
1460 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1461 	return v * gtod->clock.mult;
1462 }
1463 
1464 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1465 {
1466 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1467 	unsigned long seq;
1468 	int mode;
1469 	u64 ns;
1470 
1471 	do {
1472 		seq = read_seqcount_begin(&gtod->seq);
1473 		mode = gtod->clock.vclock_mode;
1474 		ns = gtod->nsec_base;
1475 		ns += vgettsc(cycle_now);
1476 		ns >>= gtod->clock.shift;
1477 		ns += gtod->boot_ns;
1478 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1479 	*t = ns;
1480 
1481 	return mode;
1482 }
1483 
1484 /* returns true if host is using tsc clocksource */
1485 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1486 {
1487 	/* checked again under seqlock below */
1488 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1489 		return false;
1490 
1491 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1492 }
1493 #endif
1494 
1495 /*
1496  *
1497  * Assuming a stable TSC across physical CPUS, and a stable TSC
1498  * across virtual CPUs, the following condition is possible.
1499  * Each numbered line represents an event visible to both
1500  * CPUs at the next numbered event.
1501  *
1502  * "timespecX" represents host monotonic time. "tscX" represents
1503  * RDTSC value.
1504  *
1505  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1506  *
1507  * 1.  read timespec0,tsc0
1508  * 2.					| timespec1 = timespec0 + N
1509  * 					| tsc1 = tsc0 + M
1510  * 3. transition to guest		| transition to guest
1511  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1512  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1513  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1514  *
1515  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1516  *
1517  * 	- ret0 < ret1
1518  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1519  *		...
1520  *	- 0 < N - M => M < N
1521  *
1522  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1523  * always the case (the difference between two distinct xtime instances
1524  * might be smaller then the difference between corresponding TSC reads,
1525  * when updating guest vcpus pvclock areas).
1526  *
1527  * To avoid that problem, do not allow visibility of distinct
1528  * system_timestamp/tsc_timestamp values simultaneously: use a master
1529  * copy of host monotonic time values. Update that master copy
1530  * in lockstep.
1531  *
1532  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1533  *
1534  */
1535 
1536 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1537 {
1538 #ifdef CONFIG_X86_64
1539 	struct kvm_arch *ka = &kvm->arch;
1540 	int vclock_mode;
1541 	bool host_tsc_clocksource, vcpus_matched;
1542 
1543 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1544 			atomic_read(&kvm->online_vcpus));
1545 
1546 	/*
1547 	 * If the host uses TSC clock, then passthrough TSC as stable
1548 	 * to the guest.
1549 	 */
1550 	host_tsc_clocksource = kvm_get_time_and_clockread(
1551 					&ka->master_kernel_ns,
1552 					&ka->master_cycle_now);
1553 
1554 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1555 				&& !backwards_tsc_observed
1556 				&& !ka->boot_vcpu_runs_old_kvmclock;
1557 
1558 	if (ka->use_master_clock)
1559 		atomic_set(&kvm_guest_has_master_clock, 1);
1560 
1561 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1562 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1563 					vcpus_matched);
1564 #endif
1565 }
1566 
1567 static void kvm_gen_update_masterclock(struct kvm *kvm)
1568 {
1569 #ifdef CONFIG_X86_64
1570 	int i;
1571 	struct kvm_vcpu *vcpu;
1572 	struct kvm_arch *ka = &kvm->arch;
1573 
1574 	spin_lock(&ka->pvclock_gtod_sync_lock);
1575 	kvm_make_mclock_inprogress_request(kvm);
1576 	/* no guest entries from this point */
1577 	pvclock_update_vm_gtod_copy(kvm);
1578 
1579 	kvm_for_each_vcpu(i, vcpu, kvm)
1580 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1581 
1582 	/* guest entries allowed */
1583 	kvm_for_each_vcpu(i, vcpu, kvm)
1584 		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1585 
1586 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1587 #endif
1588 }
1589 
1590 static int kvm_guest_time_update(struct kvm_vcpu *v)
1591 {
1592 	unsigned long flags, this_tsc_khz;
1593 	struct kvm_vcpu_arch *vcpu = &v->arch;
1594 	struct kvm_arch *ka = &v->kvm->arch;
1595 	s64 kernel_ns;
1596 	u64 tsc_timestamp, host_tsc;
1597 	struct pvclock_vcpu_time_info guest_hv_clock;
1598 	u8 pvclock_flags;
1599 	bool use_master_clock;
1600 
1601 	kernel_ns = 0;
1602 	host_tsc = 0;
1603 
1604 	/*
1605 	 * If the host uses TSC clock, then passthrough TSC as stable
1606 	 * to the guest.
1607 	 */
1608 	spin_lock(&ka->pvclock_gtod_sync_lock);
1609 	use_master_clock = ka->use_master_clock;
1610 	if (use_master_clock) {
1611 		host_tsc = ka->master_cycle_now;
1612 		kernel_ns = ka->master_kernel_ns;
1613 	}
1614 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1615 
1616 	/* Keep irq disabled to prevent changes to the clock */
1617 	local_irq_save(flags);
1618 	this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1619 	if (unlikely(this_tsc_khz == 0)) {
1620 		local_irq_restore(flags);
1621 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1622 		return 1;
1623 	}
1624 	if (!use_master_clock) {
1625 		host_tsc = native_read_tsc();
1626 		kernel_ns = get_kernel_ns();
1627 	}
1628 
1629 	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1630 
1631 	/*
1632 	 * We may have to catch up the TSC to match elapsed wall clock
1633 	 * time for two reasons, even if kvmclock is used.
1634 	 *   1) CPU could have been running below the maximum TSC rate
1635 	 *   2) Broken TSC compensation resets the base at each VCPU
1636 	 *      entry to avoid unknown leaps of TSC even when running
1637 	 *      again on the same CPU.  This may cause apparent elapsed
1638 	 *      time to disappear, and the guest to stand still or run
1639 	 *	very slowly.
1640 	 */
1641 	if (vcpu->tsc_catchup) {
1642 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1643 		if (tsc > tsc_timestamp) {
1644 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1645 			tsc_timestamp = tsc;
1646 		}
1647 	}
1648 
1649 	local_irq_restore(flags);
1650 
1651 	if (!vcpu->pv_time_enabled)
1652 		return 0;
1653 
1654 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1655 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1656 				   &vcpu->hv_clock.tsc_shift,
1657 				   &vcpu->hv_clock.tsc_to_system_mul);
1658 		vcpu->hw_tsc_khz = this_tsc_khz;
1659 	}
1660 
1661 	/* With all the info we got, fill in the values */
1662 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1663 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1664 	vcpu->last_guest_tsc = tsc_timestamp;
1665 
1666 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1667 		&guest_hv_clock, sizeof(guest_hv_clock))))
1668 		return 0;
1669 
1670 	/* This VCPU is paused, but it's legal for a guest to read another
1671 	 * VCPU's kvmclock, so we really have to follow the specification where
1672 	 * it says that version is odd if data is being modified, and even after
1673 	 * it is consistent.
1674 	 *
1675 	 * Version field updates must be kept separate.  This is because
1676 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1677 	 * writes within a string instruction are weakly ordered.  So there
1678 	 * are three writes overall.
1679 	 *
1680 	 * As a small optimization, only write the version field in the first
1681 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1682 	 * version field is the first in the struct.
1683 	 */
1684 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1685 
1686 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1687 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1688 				&vcpu->hv_clock,
1689 				sizeof(vcpu->hv_clock.version));
1690 
1691 	smp_wmb();
1692 
1693 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1694 	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1695 
1696 	if (vcpu->pvclock_set_guest_stopped_request) {
1697 		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1698 		vcpu->pvclock_set_guest_stopped_request = false;
1699 	}
1700 
1701 	/* If the host uses TSC clocksource, then it is stable */
1702 	if (use_master_clock)
1703 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1704 
1705 	vcpu->hv_clock.flags = pvclock_flags;
1706 
1707 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1708 
1709 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1710 				&vcpu->hv_clock,
1711 				sizeof(vcpu->hv_clock));
1712 
1713 	smp_wmb();
1714 
1715 	vcpu->hv_clock.version++;
1716 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1717 				&vcpu->hv_clock,
1718 				sizeof(vcpu->hv_clock.version));
1719 	return 0;
1720 }
1721 
1722 /*
1723  * kvmclock updates which are isolated to a given vcpu, such as
1724  * vcpu->cpu migration, should not allow system_timestamp from
1725  * the rest of the vcpus to remain static. Otherwise ntp frequency
1726  * correction applies to one vcpu's system_timestamp but not
1727  * the others.
1728  *
1729  * So in those cases, request a kvmclock update for all vcpus.
1730  * We need to rate-limit these requests though, as they can
1731  * considerably slow guests that have a large number of vcpus.
1732  * The time for a remote vcpu to update its kvmclock is bound
1733  * by the delay we use to rate-limit the updates.
1734  */
1735 
1736 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1737 
1738 static void kvmclock_update_fn(struct work_struct *work)
1739 {
1740 	int i;
1741 	struct delayed_work *dwork = to_delayed_work(work);
1742 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1743 					   kvmclock_update_work);
1744 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1745 	struct kvm_vcpu *vcpu;
1746 
1747 	kvm_for_each_vcpu(i, vcpu, kvm) {
1748 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1749 		kvm_vcpu_kick(vcpu);
1750 	}
1751 }
1752 
1753 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1754 {
1755 	struct kvm *kvm = v->kvm;
1756 
1757 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1758 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1759 					KVMCLOCK_UPDATE_DELAY);
1760 }
1761 
1762 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1763 
1764 static void kvmclock_sync_fn(struct work_struct *work)
1765 {
1766 	struct delayed_work *dwork = to_delayed_work(work);
1767 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1768 					   kvmclock_sync_work);
1769 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1770 
1771 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1772 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1773 					KVMCLOCK_SYNC_PERIOD);
1774 }
1775 
1776 static bool msr_mtrr_valid(unsigned msr)
1777 {
1778 	switch (msr) {
1779 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1780 	case MSR_MTRRfix64K_00000:
1781 	case MSR_MTRRfix16K_80000:
1782 	case MSR_MTRRfix16K_A0000:
1783 	case MSR_MTRRfix4K_C0000:
1784 	case MSR_MTRRfix4K_C8000:
1785 	case MSR_MTRRfix4K_D0000:
1786 	case MSR_MTRRfix4K_D8000:
1787 	case MSR_MTRRfix4K_E0000:
1788 	case MSR_MTRRfix4K_E8000:
1789 	case MSR_MTRRfix4K_F0000:
1790 	case MSR_MTRRfix4K_F8000:
1791 	case MSR_MTRRdefType:
1792 	case MSR_IA32_CR_PAT:
1793 		return true;
1794 	case 0x2f8:
1795 		return true;
1796 	}
1797 	return false;
1798 }
1799 
1800 static bool valid_pat_type(unsigned t)
1801 {
1802 	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1803 }
1804 
1805 static bool valid_mtrr_type(unsigned t)
1806 {
1807 	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1808 }
1809 
1810 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1811 {
1812 	int i;
1813 	u64 mask;
1814 
1815 	if (!msr_mtrr_valid(msr))
1816 		return false;
1817 
1818 	if (msr == MSR_IA32_CR_PAT) {
1819 		for (i = 0; i < 8; i++)
1820 			if (!valid_pat_type((data >> (i * 8)) & 0xff))
1821 				return false;
1822 		return true;
1823 	} else if (msr == MSR_MTRRdefType) {
1824 		if (data & ~0xcff)
1825 			return false;
1826 		return valid_mtrr_type(data & 0xff);
1827 	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1828 		for (i = 0; i < 8 ; i++)
1829 			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1830 				return false;
1831 		return true;
1832 	}
1833 
1834 	/* variable MTRRs */
1835 	WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1836 
1837 	mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1838 	if ((msr & 1) == 0) {
1839 		/* MTRR base */
1840 		if (!valid_mtrr_type(data & 0xff))
1841 			return false;
1842 		mask |= 0xf00;
1843 	} else
1844 		/* MTRR mask */
1845 		mask |= 0x7ff;
1846 	if (data & mask) {
1847 		kvm_inject_gp(vcpu, 0);
1848 		return false;
1849 	}
1850 
1851 	return true;
1852 }
1853 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1854 
1855 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1856 {
1857 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1858 
1859 	if (!kvm_mtrr_valid(vcpu, msr, data))
1860 		return 1;
1861 
1862 	if (msr == MSR_MTRRdefType) {
1863 		vcpu->arch.mtrr_state.def_type = data;
1864 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1865 	} else if (msr == MSR_MTRRfix64K_00000)
1866 		p[0] = data;
1867 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1868 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
1869 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1870 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1871 	else if (msr == MSR_IA32_CR_PAT)
1872 		vcpu->arch.pat = data;
1873 	else {	/* Variable MTRRs */
1874 		int idx, is_mtrr_mask;
1875 		u64 *pt;
1876 
1877 		idx = (msr - 0x200) / 2;
1878 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1879 		if (!is_mtrr_mask)
1880 			pt =
1881 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1882 		else
1883 			pt =
1884 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1885 		*pt = data;
1886 	}
1887 
1888 	kvm_mmu_reset_context(vcpu);
1889 	return 0;
1890 }
1891 
1892 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1893 {
1894 	u64 mcg_cap = vcpu->arch.mcg_cap;
1895 	unsigned bank_num = mcg_cap & 0xff;
1896 
1897 	switch (msr) {
1898 	case MSR_IA32_MCG_STATUS:
1899 		vcpu->arch.mcg_status = data;
1900 		break;
1901 	case MSR_IA32_MCG_CTL:
1902 		if (!(mcg_cap & MCG_CTL_P))
1903 			return 1;
1904 		if (data != 0 && data != ~(u64)0)
1905 			return -1;
1906 		vcpu->arch.mcg_ctl = data;
1907 		break;
1908 	default:
1909 		if (msr >= MSR_IA32_MC0_CTL &&
1910 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
1911 			u32 offset = msr - MSR_IA32_MC0_CTL;
1912 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1913 			 * some Linux kernels though clear bit 10 in bank 4 to
1914 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1915 			 * this to avoid an uncatched #GP in the guest
1916 			 */
1917 			if ((offset & 0x3) == 0 &&
1918 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1919 				return -1;
1920 			vcpu->arch.mce_banks[offset] = data;
1921 			break;
1922 		}
1923 		return 1;
1924 	}
1925 	return 0;
1926 }
1927 
1928 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1929 {
1930 	struct kvm *kvm = vcpu->kvm;
1931 	int lm = is_long_mode(vcpu);
1932 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1933 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1934 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1935 		: kvm->arch.xen_hvm_config.blob_size_32;
1936 	u32 page_num = data & ~PAGE_MASK;
1937 	u64 page_addr = data & PAGE_MASK;
1938 	u8 *page;
1939 	int r;
1940 
1941 	r = -E2BIG;
1942 	if (page_num >= blob_size)
1943 		goto out;
1944 	r = -ENOMEM;
1945 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1946 	if (IS_ERR(page)) {
1947 		r = PTR_ERR(page);
1948 		goto out;
1949 	}
1950 	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1951 		goto out_free;
1952 	r = 0;
1953 out_free:
1954 	kfree(page);
1955 out:
1956 	return r;
1957 }
1958 
1959 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1960 {
1961 	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1962 }
1963 
1964 static bool kvm_hv_msr_partition_wide(u32 msr)
1965 {
1966 	bool r = false;
1967 	switch (msr) {
1968 	case HV_X64_MSR_GUEST_OS_ID:
1969 	case HV_X64_MSR_HYPERCALL:
1970 	case HV_X64_MSR_REFERENCE_TSC:
1971 	case HV_X64_MSR_TIME_REF_COUNT:
1972 		r = true;
1973 		break;
1974 	}
1975 
1976 	return r;
1977 }
1978 
1979 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1980 {
1981 	struct kvm *kvm = vcpu->kvm;
1982 
1983 	switch (msr) {
1984 	case HV_X64_MSR_GUEST_OS_ID:
1985 		kvm->arch.hv_guest_os_id = data;
1986 		/* setting guest os id to zero disables hypercall page */
1987 		if (!kvm->arch.hv_guest_os_id)
1988 			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1989 		break;
1990 	case HV_X64_MSR_HYPERCALL: {
1991 		u64 gfn;
1992 		unsigned long addr;
1993 		u8 instructions[4];
1994 
1995 		/* if guest os id is not set hypercall should remain disabled */
1996 		if (!kvm->arch.hv_guest_os_id)
1997 			break;
1998 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1999 			kvm->arch.hv_hypercall = data;
2000 			break;
2001 		}
2002 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
2003 		addr = gfn_to_hva(kvm, gfn);
2004 		if (kvm_is_error_hva(addr))
2005 			return 1;
2006 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
2007 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
2008 		if (__copy_to_user((void __user *)addr, instructions, 4))
2009 			return 1;
2010 		kvm->arch.hv_hypercall = data;
2011 		mark_page_dirty(kvm, gfn);
2012 		break;
2013 	}
2014 	case HV_X64_MSR_REFERENCE_TSC: {
2015 		u64 gfn;
2016 		HV_REFERENCE_TSC_PAGE tsc_ref;
2017 		memset(&tsc_ref, 0, sizeof(tsc_ref));
2018 		kvm->arch.hv_tsc_page = data;
2019 		if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
2020 			break;
2021 		gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
2022 		if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
2023 			&tsc_ref, sizeof(tsc_ref)))
2024 			return 1;
2025 		mark_page_dirty(kvm, gfn);
2026 		break;
2027 	}
2028 	default:
2029 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2030 			    "data 0x%llx\n", msr, data);
2031 		return 1;
2032 	}
2033 	return 0;
2034 }
2035 
2036 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2037 {
2038 	switch (msr) {
2039 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
2040 		u64 gfn;
2041 		unsigned long addr;
2042 
2043 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2044 			vcpu->arch.hv_vapic = data;
2045 			if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2046 				return 1;
2047 			break;
2048 		}
2049 		gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2050 		addr = gfn_to_hva(vcpu->kvm, gfn);
2051 		if (kvm_is_error_hva(addr))
2052 			return 1;
2053 		if (__clear_user((void __user *)addr, PAGE_SIZE))
2054 			return 1;
2055 		vcpu->arch.hv_vapic = data;
2056 		mark_page_dirty(vcpu->kvm, gfn);
2057 		if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2058 			return 1;
2059 		break;
2060 	}
2061 	case HV_X64_MSR_EOI:
2062 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2063 	case HV_X64_MSR_ICR:
2064 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2065 	case HV_X64_MSR_TPR:
2066 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2067 	default:
2068 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2069 			    "data 0x%llx\n", msr, data);
2070 		return 1;
2071 	}
2072 
2073 	return 0;
2074 }
2075 
2076 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2077 {
2078 	gpa_t gpa = data & ~0x3f;
2079 
2080 	/* Bits 2:5 are reserved, Should be zero */
2081 	if (data & 0x3c)
2082 		return 1;
2083 
2084 	vcpu->arch.apf.msr_val = data;
2085 
2086 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2087 		kvm_clear_async_pf_completion_queue(vcpu);
2088 		kvm_async_pf_hash_reset(vcpu);
2089 		return 0;
2090 	}
2091 
2092 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2093 					sizeof(u32)))
2094 		return 1;
2095 
2096 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2097 	kvm_async_pf_wakeup_all(vcpu);
2098 	return 0;
2099 }
2100 
2101 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2102 {
2103 	vcpu->arch.pv_time_enabled = false;
2104 }
2105 
2106 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2107 {
2108 	u64 delta;
2109 
2110 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2111 		return;
2112 
2113 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2114 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2115 	vcpu->arch.st.accum_steal = delta;
2116 }
2117 
2118 static void record_steal_time(struct kvm_vcpu *vcpu)
2119 {
2120 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2121 		return;
2122 
2123 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2124 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2125 		return;
2126 
2127 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2128 	vcpu->arch.st.steal.version += 2;
2129 	vcpu->arch.st.accum_steal = 0;
2130 
2131 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2132 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2133 }
2134 
2135 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2136 {
2137 	bool pr = false;
2138 	u32 msr = msr_info->index;
2139 	u64 data = msr_info->data;
2140 
2141 	switch (msr) {
2142 	case MSR_AMD64_NB_CFG:
2143 	case MSR_IA32_UCODE_REV:
2144 	case MSR_IA32_UCODE_WRITE:
2145 	case MSR_VM_HSAVE_PA:
2146 	case MSR_AMD64_PATCH_LOADER:
2147 	case MSR_AMD64_BU_CFG2:
2148 		break;
2149 
2150 	case MSR_EFER:
2151 		return set_efer(vcpu, data);
2152 	case MSR_K7_HWCR:
2153 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2154 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2155 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2156 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2157 		if (data != 0) {
2158 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2159 				    data);
2160 			return 1;
2161 		}
2162 		break;
2163 	case MSR_FAM10H_MMIO_CONF_BASE:
2164 		if (data != 0) {
2165 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2166 				    "0x%llx\n", data);
2167 			return 1;
2168 		}
2169 		break;
2170 	case MSR_IA32_DEBUGCTLMSR:
2171 		if (!data) {
2172 			/* We support the non-activated case already */
2173 			break;
2174 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2175 			/* Values other than LBR and BTF are vendor-specific,
2176 			   thus reserved and should throw a #GP */
2177 			return 1;
2178 		}
2179 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2180 			    __func__, data);
2181 		break;
2182 	case 0x200 ... 0x2ff:
2183 		return set_msr_mtrr(vcpu, msr, data);
2184 	case MSR_IA32_APICBASE:
2185 		return kvm_set_apic_base(vcpu, msr_info);
2186 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2187 		return kvm_x2apic_msr_write(vcpu, msr, data);
2188 	case MSR_IA32_TSCDEADLINE:
2189 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2190 		break;
2191 	case MSR_IA32_TSC_ADJUST:
2192 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2193 			if (!msr_info->host_initiated) {
2194 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2195 				kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2196 			}
2197 			vcpu->arch.ia32_tsc_adjust_msr = data;
2198 		}
2199 		break;
2200 	case MSR_IA32_MISC_ENABLE:
2201 		vcpu->arch.ia32_misc_enable_msr = data;
2202 		break;
2203 	case MSR_KVM_WALL_CLOCK_NEW:
2204 	case MSR_KVM_WALL_CLOCK:
2205 		vcpu->kvm->arch.wall_clock = data;
2206 		kvm_write_wall_clock(vcpu->kvm, data);
2207 		break;
2208 	case MSR_KVM_SYSTEM_TIME_NEW:
2209 	case MSR_KVM_SYSTEM_TIME: {
2210 		u64 gpa_offset;
2211 		struct kvm_arch *ka = &vcpu->kvm->arch;
2212 
2213 		kvmclock_reset(vcpu);
2214 
2215 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2216 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2217 
2218 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2219 				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2220 					&vcpu->requests);
2221 
2222 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2223 		}
2224 
2225 		vcpu->arch.time = data;
2226 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2227 
2228 		/* we verify if the enable bit is set... */
2229 		if (!(data & 1))
2230 			break;
2231 
2232 		gpa_offset = data & ~(PAGE_MASK | 1);
2233 
2234 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2235 		     &vcpu->arch.pv_time, data & ~1ULL,
2236 		     sizeof(struct pvclock_vcpu_time_info)))
2237 			vcpu->arch.pv_time_enabled = false;
2238 		else
2239 			vcpu->arch.pv_time_enabled = true;
2240 
2241 		break;
2242 	}
2243 	case MSR_KVM_ASYNC_PF_EN:
2244 		if (kvm_pv_enable_async_pf(vcpu, data))
2245 			return 1;
2246 		break;
2247 	case MSR_KVM_STEAL_TIME:
2248 
2249 		if (unlikely(!sched_info_on()))
2250 			return 1;
2251 
2252 		if (data & KVM_STEAL_RESERVED_MASK)
2253 			return 1;
2254 
2255 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2256 						data & KVM_STEAL_VALID_BITS,
2257 						sizeof(struct kvm_steal_time)))
2258 			return 1;
2259 
2260 		vcpu->arch.st.msr_val = data;
2261 
2262 		if (!(data & KVM_MSR_ENABLED))
2263 			break;
2264 
2265 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
2266 
2267 		preempt_disable();
2268 		accumulate_steal_time(vcpu);
2269 		preempt_enable();
2270 
2271 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2272 
2273 		break;
2274 	case MSR_KVM_PV_EOI_EN:
2275 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2276 			return 1;
2277 		break;
2278 
2279 	case MSR_IA32_MCG_CTL:
2280 	case MSR_IA32_MCG_STATUS:
2281 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2282 		return set_msr_mce(vcpu, msr, data);
2283 
2284 	/* Performance counters are not protected by a CPUID bit,
2285 	 * so we should check all of them in the generic path for the sake of
2286 	 * cross vendor migration.
2287 	 * Writing a zero into the event select MSRs disables them,
2288 	 * which we perfectly emulate ;-). Any other value should be at least
2289 	 * reported, some guests depend on them.
2290 	 */
2291 	case MSR_K7_EVNTSEL0:
2292 	case MSR_K7_EVNTSEL1:
2293 	case MSR_K7_EVNTSEL2:
2294 	case MSR_K7_EVNTSEL3:
2295 		if (data != 0)
2296 			vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2297 				    "0x%x data 0x%llx\n", msr, data);
2298 		break;
2299 	/* at least RHEL 4 unconditionally writes to the perfctr registers,
2300 	 * so we ignore writes to make it happy.
2301 	 */
2302 	case MSR_K7_PERFCTR0:
2303 	case MSR_K7_PERFCTR1:
2304 	case MSR_K7_PERFCTR2:
2305 	case MSR_K7_PERFCTR3:
2306 		vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2307 			    "0x%x data 0x%llx\n", msr, data);
2308 		break;
2309 	case MSR_P6_PERFCTR0:
2310 	case MSR_P6_PERFCTR1:
2311 		pr = true;
2312 	case MSR_P6_EVNTSEL0:
2313 	case MSR_P6_EVNTSEL1:
2314 		if (kvm_pmu_msr(vcpu, msr))
2315 			return kvm_pmu_set_msr(vcpu, msr_info);
2316 
2317 		if (pr || data != 0)
2318 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2319 				    "0x%x data 0x%llx\n", msr, data);
2320 		break;
2321 	case MSR_K7_CLK_CTL:
2322 		/*
2323 		 * Ignore all writes to this no longer documented MSR.
2324 		 * Writes are only relevant for old K7 processors,
2325 		 * all pre-dating SVM, but a recommended workaround from
2326 		 * AMD for these chips. It is possible to specify the
2327 		 * affected processor models on the command line, hence
2328 		 * the need to ignore the workaround.
2329 		 */
2330 		break;
2331 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2332 		if (kvm_hv_msr_partition_wide(msr)) {
2333 			int r;
2334 			mutex_lock(&vcpu->kvm->lock);
2335 			r = set_msr_hyperv_pw(vcpu, msr, data);
2336 			mutex_unlock(&vcpu->kvm->lock);
2337 			return r;
2338 		} else
2339 			return set_msr_hyperv(vcpu, msr, data);
2340 		break;
2341 	case MSR_IA32_BBL_CR_CTL3:
2342 		/* Drop writes to this legacy MSR -- see rdmsr
2343 		 * counterpart for further detail.
2344 		 */
2345 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2346 		break;
2347 	case MSR_AMD64_OSVW_ID_LENGTH:
2348 		if (!guest_cpuid_has_osvw(vcpu))
2349 			return 1;
2350 		vcpu->arch.osvw.length = data;
2351 		break;
2352 	case MSR_AMD64_OSVW_STATUS:
2353 		if (!guest_cpuid_has_osvw(vcpu))
2354 			return 1;
2355 		vcpu->arch.osvw.status = data;
2356 		break;
2357 	default:
2358 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2359 			return xen_hvm_config(vcpu, data);
2360 		if (kvm_pmu_msr(vcpu, msr))
2361 			return kvm_pmu_set_msr(vcpu, msr_info);
2362 		if (!ignore_msrs) {
2363 			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2364 				    msr, data);
2365 			return 1;
2366 		} else {
2367 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2368 				    msr, data);
2369 			break;
2370 		}
2371 	}
2372 	return 0;
2373 }
2374 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2375 
2376 
2377 /*
2378  * Reads an msr value (of 'msr_index') into 'pdata'.
2379  * Returns 0 on success, non-0 otherwise.
2380  * Assumes vcpu_load() was already called.
2381  */
2382 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2383 {
2384 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2385 }
2386 EXPORT_SYMBOL_GPL(kvm_get_msr);
2387 
2388 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2389 {
2390 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2391 
2392 	if (!msr_mtrr_valid(msr))
2393 		return 1;
2394 
2395 	if (msr == MSR_MTRRdefType)
2396 		*pdata = vcpu->arch.mtrr_state.def_type +
2397 			 (vcpu->arch.mtrr_state.enabled << 10);
2398 	else if (msr == MSR_MTRRfix64K_00000)
2399 		*pdata = p[0];
2400 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2401 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2402 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2403 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2404 	else if (msr == MSR_IA32_CR_PAT)
2405 		*pdata = vcpu->arch.pat;
2406 	else {	/* Variable MTRRs */
2407 		int idx, is_mtrr_mask;
2408 		u64 *pt;
2409 
2410 		idx = (msr - 0x200) / 2;
2411 		is_mtrr_mask = msr - 0x200 - 2 * idx;
2412 		if (!is_mtrr_mask)
2413 			pt =
2414 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2415 		else
2416 			pt =
2417 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2418 		*pdata = *pt;
2419 	}
2420 
2421 	return 0;
2422 }
2423 
2424 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2425 {
2426 	u64 data;
2427 	u64 mcg_cap = vcpu->arch.mcg_cap;
2428 	unsigned bank_num = mcg_cap & 0xff;
2429 
2430 	switch (msr) {
2431 	case MSR_IA32_P5_MC_ADDR:
2432 	case MSR_IA32_P5_MC_TYPE:
2433 		data = 0;
2434 		break;
2435 	case MSR_IA32_MCG_CAP:
2436 		data = vcpu->arch.mcg_cap;
2437 		break;
2438 	case MSR_IA32_MCG_CTL:
2439 		if (!(mcg_cap & MCG_CTL_P))
2440 			return 1;
2441 		data = vcpu->arch.mcg_ctl;
2442 		break;
2443 	case MSR_IA32_MCG_STATUS:
2444 		data = vcpu->arch.mcg_status;
2445 		break;
2446 	default:
2447 		if (msr >= MSR_IA32_MC0_CTL &&
2448 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2449 			u32 offset = msr - MSR_IA32_MC0_CTL;
2450 			data = vcpu->arch.mce_banks[offset];
2451 			break;
2452 		}
2453 		return 1;
2454 	}
2455 	*pdata = data;
2456 	return 0;
2457 }
2458 
2459 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2460 {
2461 	u64 data = 0;
2462 	struct kvm *kvm = vcpu->kvm;
2463 
2464 	switch (msr) {
2465 	case HV_X64_MSR_GUEST_OS_ID:
2466 		data = kvm->arch.hv_guest_os_id;
2467 		break;
2468 	case HV_X64_MSR_HYPERCALL:
2469 		data = kvm->arch.hv_hypercall;
2470 		break;
2471 	case HV_X64_MSR_TIME_REF_COUNT: {
2472 		data =
2473 		     div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2474 		break;
2475 	}
2476 	case HV_X64_MSR_REFERENCE_TSC:
2477 		data = kvm->arch.hv_tsc_page;
2478 		break;
2479 	default:
2480 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2481 		return 1;
2482 	}
2483 
2484 	*pdata = data;
2485 	return 0;
2486 }
2487 
2488 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2489 {
2490 	u64 data = 0;
2491 
2492 	switch (msr) {
2493 	case HV_X64_MSR_VP_INDEX: {
2494 		int r;
2495 		struct kvm_vcpu *v;
2496 		kvm_for_each_vcpu(r, v, vcpu->kvm) {
2497 			if (v == vcpu) {
2498 				data = r;
2499 				break;
2500 			}
2501 		}
2502 		break;
2503 	}
2504 	case HV_X64_MSR_EOI:
2505 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2506 	case HV_X64_MSR_ICR:
2507 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2508 	case HV_X64_MSR_TPR:
2509 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2510 	case HV_X64_MSR_APIC_ASSIST_PAGE:
2511 		data = vcpu->arch.hv_vapic;
2512 		break;
2513 	default:
2514 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2515 		return 1;
2516 	}
2517 	*pdata = data;
2518 	return 0;
2519 }
2520 
2521 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2522 {
2523 	u64 data;
2524 
2525 	switch (msr) {
2526 	case MSR_IA32_PLATFORM_ID:
2527 	case MSR_IA32_EBL_CR_POWERON:
2528 	case MSR_IA32_DEBUGCTLMSR:
2529 	case MSR_IA32_LASTBRANCHFROMIP:
2530 	case MSR_IA32_LASTBRANCHTOIP:
2531 	case MSR_IA32_LASTINTFROMIP:
2532 	case MSR_IA32_LASTINTTOIP:
2533 	case MSR_K8_SYSCFG:
2534 	case MSR_K7_HWCR:
2535 	case MSR_VM_HSAVE_PA:
2536 	case MSR_K7_EVNTSEL0:
2537 	case MSR_K7_EVNTSEL1:
2538 	case MSR_K7_EVNTSEL2:
2539 	case MSR_K7_EVNTSEL3:
2540 	case MSR_K7_PERFCTR0:
2541 	case MSR_K7_PERFCTR1:
2542 	case MSR_K7_PERFCTR2:
2543 	case MSR_K7_PERFCTR3:
2544 	case MSR_K8_INT_PENDING_MSG:
2545 	case MSR_AMD64_NB_CFG:
2546 	case MSR_FAM10H_MMIO_CONF_BASE:
2547 	case MSR_AMD64_BU_CFG2:
2548 		data = 0;
2549 		break;
2550 	case MSR_P6_PERFCTR0:
2551 	case MSR_P6_PERFCTR1:
2552 	case MSR_P6_EVNTSEL0:
2553 	case MSR_P6_EVNTSEL1:
2554 		if (kvm_pmu_msr(vcpu, msr))
2555 			return kvm_pmu_get_msr(vcpu, msr, pdata);
2556 		data = 0;
2557 		break;
2558 	case MSR_IA32_UCODE_REV:
2559 		data = 0x100000000ULL;
2560 		break;
2561 	case MSR_MTRRcap:
2562 		data = 0x500 | KVM_NR_VAR_MTRR;
2563 		break;
2564 	case 0x200 ... 0x2ff:
2565 		return get_msr_mtrr(vcpu, msr, pdata);
2566 	case 0xcd: /* fsb frequency */
2567 		data = 3;
2568 		break;
2569 		/*
2570 		 * MSR_EBC_FREQUENCY_ID
2571 		 * Conservative value valid for even the basic CPU models.
2572 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2573 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2574 		 * and 266MHz for model 3, or 4. Set Core Clock
2575 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2576 		 * 31:24) even though these are only valid for CPU
2577 		 * models > 2, however guests may end up dividing or
2578 		 * multiplying by zero otherwise.
2579 		 */
2580 	case MSR_EBC_FREQUENCY_ID:
2581 		data = 1 << 24;
2582 		break;
2583 	case MSR_IA32_APICBASE:
2584 		data = kvm_get_apic_base(vcpu);
2585 		break;
2586 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2587 		return kvm_x2apic_msr_read(vcpu, msr, pdata);
2588 		break;
2589 	case MSR_IA32_TSCDEADLINE:
2590 		data = kvm_get_lapic_tscdeadline_msr(vcpu);
2591 		break;
2592 	case MSR_IA32_TSC_ADJUST:
2593 		data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2594 		break;
2595 	case MSR_IA32_MISC_ENABLE:
2596 		data = vcpu->arch.ia32_misc_enable_msr;
2597 		break;
2598 	case MSR_IA32_PERF_STATUS:
2599 		/* TSC increment by tick */
2600 		data = 1000ULL;
2601 		/* CPU multiplier */
2602 		data |= (((uint64_t)4ULL) << 40);
2603 		break;
2604 	case MSR_EFER:
2605 		data = vcpu->arch.efer;
2606 		break;
2607 	case MSR_KVM_WALL_CLOCK:
2608 	case MSR_KVM_WALL_CLOCK_NEW:
2609 		data = vcpu->kvm->arch.wall_clock;
2610 		break;
2611 	case MSR_KVM_SYSTEM_TIME:
2612 	case MSR_KVM_SYSTEM_TIME_NEW:
2613 		data = vcpu->arch.time;
2614 		break;
2615 	case MSR_KVM_ASYNC_PF_EN:
2616 		data = vcpu->arch.apf.msr_val;
2617 		break;
2618 	case MSR_KVM_STEAL_TIME:
2619 		data = vcpu->arch.st.msr_val;
2620 		break;
2621 	case MSR_KVM_PV_EOI_EN:
2622 		data = vcpu->arch.pv_eoi.msr_val;
2623 		break;
2624 	case MSR_IA32_P5_MC_ADDR:
2625 	case MSR_IA32_P5_MC_TYPE:
2626 	case MSR_IA32_MCG_CAP:
2627 	case MSR_IA32_MCG_CTL:
2628 	case MSR_IA32_MCG_STATUS:
2629 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2630 		return get_msr_mce(vcpu, msr, pdata);
2631 	case MSR_K7_CLK_CTL:
2632 		/*
2633 		 * Provide expected ramp-up count for K7. All other
2634 		 * are set to zero, indicating minimum divisors for
2635 		 * every field.
2636 		 *
2637 		 * This prevents guest kernels on AMD host with CPU
2638 		 * type 6, model 8 and higher from exploding due to
2639 		 * the rdmsr failing.
2640 		 */
2641 		data = 0x20000000;
2642 		break;
2643 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2644 		if (kvm_hv_msr_partition_wide(msr)) {
2645 			int r;
2646 			mutex_lock(&vcpu->kvm->lock);
2647 			r = get_msr_hyperv_pw(vcpu, msr, pdata);
2648 			mutex_unlock(&vcpu->kvm->lock);
2649 			return r;
2650 		} else
2651 			return get_msr_hyperv(vcpu, msr, pdata);
2652 		break;
2653 	case MSR_IA32_BBL_CR_CTL3:
2654 		/* This legacy MSR exists but isn't fully documented in current
2655 		 * silicon.  It is however accessed by winxp in very narrow
2656 		 * scenarios where it sets bit #19, itself documented as
2657 		 * a "reserved" bit.  Best effort attempt to source coherent
2658 		 * read data here should the balance of the register be
2659 		 * interpreted by the guest:
2660 		 *
2661 		 * L2 cache control register 3: 64GB range, 256KB size,
2662 		 * enabled, latency 0x1, configured
2663 		 */
2664 		data = 0xbe702111;
2665 		break;
2666 	case MSR_AMD64_OSVW_ID_LENGTH:
2667 		if (!guest_cpuid_has_osvw(vcpu))
2668 			return 1;
2669 		data = vcpu->arch.osvw.length;
2670 		break;
2671 	case MSR_AMD64_OSVW_STATUS:
2672 		if (!guest_cpuid_has_osvw(vcpu))
2673 			return 1;
2674 		data = vcpu->arch.osvw.status;
2675 		break;
2676 	default:
2677 		if (kvm_pmu_msr(vcpu, msr))
2678 			return kvm_pmu_get_msr(vcpu, msr, pdata);
2679 		if (!ignore_msrs) {
2680 			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2681 			return 1;
2682 		} else {
2683 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2684 			data = 0;
2685 		}
2686 		break;
2687 	}
2688 	*pdata = data;
2689 	return 0;
2690 }
2691 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2692 
2693 /*
2694  * Read or write a bunch of msrs. All parameters are kernel addresses.
2695  *
2696  * @return number of msrs set successfully.
2697  */
2698 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2699 		    struct kvm_msr_entry *entries,
2700 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2701 				  unsigned index, u64 *data))
2702 {
2703 	int i, idx;
2704 
2705 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2706 	for (i = 0; i < msrs->nmsrs; ++i)
2707 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2708 			break;
2709 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2710 
2711 	return i;
2712 }
2713 
2714 /*
2715  * Read or write a bunch of msrs. Parameters are user addresses.
2716  *
2717  * @return number of msrs set successfully.
2718  */
2719 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2720 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2721 				unsigned index, u64 *data),
2722 		  int writeback)
2723 {
2724 	struct kvm_msrs msrs;
2725 	struct kvm_msr_entry *entries;
2726 	int r, n;
2727 	unsigned size;
2728 
2729 	r = -EFAULT;
2730 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2731 		goto out;
2732 
2733 	r = -E2BIG;
2734 	if (msrs.nmsrs >= MAX_IO_MSRS)
2735 		goto out;
2736 
2737 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2738 	entries = memdup_user(user_msrs->entries, size);
2739 	if (IS_ERR(entries)) {
2740 		r = PTR_ERR(entries);
2741 		goto out;
2742 	}
2743 
2744 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2745 	if (r < 0)
2746 		goto out_free;
2747 
2748 	r = -EFAULT;
2749 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2750 		goto out_free;
2751 
2752 	r = n;
2753 
2754 out_free:
2755 	kfree(entries);
2756 out:
2757 	return r;
2758 }
2759 
2760 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2761 {
2762 	int r;
2763 
2764 	switch (ext) {
2765 	case KVM_CAP_IRQCHIP:
2766 	case KVM_CAP_HLT:
2767 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2768 	case KVM_CAP_SET_TSS_ADDR:
2769 	case KVM_CAP_EXT_CPUID:
2770 	case KVM_CAP_EXT_EMUL_CPUID:
2771 	case KVM_CAP_CLOCKSOURCE:
2772 	case KVM_CAP_PIT:
2773 	case KVM_CAP_NOP_IO_DELAY:
2774 	case KVM_CAP_MP_STATE:
2775 	case KVM_CAP_SYNC_MMU:
2776 	case KVM_CAP_USER_NMI:
2777 	case KVM_CAP_REINJECT_CONTROL:
2778 	case KVM_CAP_IRQ_INJECT_STATUS:
2779 	case KVM_CAP_IOEVENTFD:
2780 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2781 	case KVM_CAP_PIT2:
2782 	case KVM_CAP_PIT_STATE2:
2783 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2784 	case KVM_CAP_XEN_HVM:
2785 	case KVM_CAP_ADJUST_CLOCK:
2786 	case KVM_CAP_VCPU_EVENTS:
2787 	case KVM_CAP_HYPERV:
2788 	case KVM_CAP_HYPERV_VAPIC:
2789 	case KVM_CAP_HYPERV_SPIN:
2790 	case KVM_CAP_PCI_SEGMENT:
2791 	case KVM_CAP_DEBUGREGS:
2792 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2793 	case KVM_CAP_XSAVE:
2794 	case KVM_CAP_ASYNC_PF:
2795 	case KVM_CAP_GET_TSC_KHZ:
2796 	case KVM_CAP_KVMCLOCK_CTRL:
2797 	case KVM_CAP_READONLY_MEM:
2798 	case KVM_CAP_HYPERV_TIME:
2799 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2800 	case KVM_CAP_TSC_DEADLINE_TIMER:
2801 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2802 	case KVM_CAP_ASSIGN_DEV_IRQ:
2803 	case KVM_CAP_PCI_2_3:
2804 #endif
2805 		r = 1;
2806 		break;
2807 	case KVM_CAP_COALESCED_MMIO:
2808 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2809 		break;
2810 	case KVM_CAP_VAPIC:
2811 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2812 		break;
2813 	case KVM_CAP_NR_VCPUS:
2814 		r = KVM_SOFT_MAX_VCPUS;
2815 		break;
2816 	case KVM_CAP_MAX_VCPUS:
2817 		r = KVM_MAX_VCPUS;
2818 		break;
2819 	case KVM_CAP_NR_MEMSLOTS:
2820 		r = KVM_USER_MEM_SLOTS;
2821 		break;
2822 	case KVM_CAP_PV_MMU:	/* obsolete */
2823 		r = 0;
2824 		break;
2825 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2826 	case KVM_CAP_IOMMU:
2827 		r = iommu_present(&pci_bus_type);
2828 		break;
2829 #endif
2830 	case KVM_CAP_MCE:
2831 		r = KVM_MAX_MCE_BANKS;
2832 		break;
2833 	case KVM_CAP_XCRS:
2834 		r = cpu_has_xsave;
2835 		break;
2836 	case KVM_CAP_TSC_CONTROL:
2837 		r = kvm_has_tsc_control;
2838 		break;
2839 	default:
2840 		r = 0;
2841 		break;
2842 	}
2843 	return r;
2844 
2845 }
2846 
2847 long kvm_arch_dev_ioctl(struct file *filp,
2848 			unsigned int ioctl, unsigned long arg)
2849 {
2850 	void __user *argp = (void __user *)arg;
2851 	long r;
2852 
2853 	switch (ioctl) {
2854 	case KVM_GET_MSR_INDEX_LIST: {
2855 		struct kvm_msr_list __user *user_msr_list = argp;
2856 		struct kvm_msr_list msr_list;
2857 		unsigned n;
2858 
2859 		r = -EFAULT;
2860 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2861 			goto out;
2862 		n = msr_list.nmsrs;
2863 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2864 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2865 			goto out;
2866 		r = -E2BIG;
2867 		if (n < msr_list.nmsrs)
2868 			goto out;
2869 		r = -EFAULT;
2870 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2871 				 num_msrs_to_save * sizeof(u32)))
2872 			goto out;
2873 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2874 				 &emulated_msrs,
2875 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2876 			goto out;
2877 		r = 0;
2878 		break;
2879 	}
2880 	case KVM_GET_SUPPORTED_CPUID:
2881 	case KVM_GET_EMULATED_CPUID: {
2882 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2883 		struct kvm_cpuid2 cpuid;
2884 
2885 		r = -EFAULT;
2886 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2887 			goto out;
2888 
2889 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2890 					    ioctl);
2891 		if (r)
2892 			goto out;
2893 
2894 		r = -EFAULT;
2895 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2896 			goto out;
2897 		r = 0;
2898 		break;
2899 	}
2900 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2901 		u64 mce_cap;
2902 
2903 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2904 		r = -EFAULT;
2905 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2906 			goto out;
2907 		r = 0;
2908 		break;
2909 	}
2910 	default:
2911 		r = -EINVAL;
2912 	}
2913 out:
2914 	return r;
2915 }
2916 
2917 static void wbinvd_ipi(void *garbage)
2918 {
2919 	wbinvd();
2920 }
2921 
2922 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2923 {
2924 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2925 }
2926 
2927 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2928 {
2929 	/* Address WBINVD may be executed by guest */
2930 	if (need_emulate_wbinvd(vcpu)) {
2931 		if (kvm_x86_ops->has_wbinvd_exit())
2932 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2933 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2934 			smp_call_function_single(vcpu->cpu,
2935 					wbinvd_ipi, NULL, 1);
2936 	}
2937 
2938 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2939 
2940 	/* Apply any externally detected TSC adjustments (due to suspend) */
2941 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2942 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2943 		vcpu->arch.tsc_offset_adjustment = 0;
2944 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2945 	}
2946 
2947 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2948 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2949 				native_read_tsc() - vcpu->arch.last_host_tsc;
2950 		if (tsc_delta < 0)
2951 			mark_tsc_unstable("KVM discovered backwards TSC");
2952 		if (check_tsc_unstable()) {
2953 			u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2954 						vcpu->arch.last_guest_tsc);
2955 			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2956 			vcpu->arch.tsc_catchup = 1;
2957 		}
2958 		/*
2959 		 * On a host with synchronized TSC, there is no need to update
2960 		 * kvmclock on vcpu->cpu migration
2961 		 */
2962 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2963 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2964 		if (vcpu->cpu != cpu)
2965 			kvm_migrate_timers(vcpu);
2966 		vcpu->cpu = cpu;
2967 	}
2968 
2969 	accumulate_steal_time(vcpu);
2970 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2971 }
2972 
2973 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2974 {
2975 	kvm_x86_ops->vcpu_put(vcpu);
2976 	kvm_put_guest_fpu(vcpu);
2977 	vcpu->arch.last_host_tsc = native_read_tsc();
2978 }
2979 
2980 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2981 				    struct kvm_lapic_state *s)
2982 {
2983 	kvm_x86_ops->sync_pir_to_irr(vcpu);
2984 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2985 
2986 	return 0;
2987 }
2988 
2989 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2990 				    struct kvm_lapic_state *s)
2991 {
2992 	kvm_apic_post_state_restore(vcpu, s);
2993 	update_cr8_intercept(vcpu);
2994 
2995 	return 0;
2996 }
2997 
2998 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2999 				    struct kvm_interrupt *irq)
3000 {
3001 	if (irq->irq >= KVM_NR_INTERRUPTS)
3002 		return -EINVAL;
3003 	if (irqchip_in_kernel(vcpu->kvm))
3004 		return -ENXIO;
3005 
3006 	kvm_queue_interrupt(vcpu, irq->irq, false);
3007 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3008 
3009 	return 0;
3010 }
3011 
3012 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3013 {
3014 	kvm_inject_nmi(vcpu);
3015 
3016 	return 0;
3017 }
3018 
3019 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3020 					   struct kvm_tpr_access_ctl *tac)
3021 {
3022 	if (tac->flags)
3023 		return -EINVAL;
3024 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3025 	return 0;
3026 }
3027 
3028 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3029 					u64 mcg_cap)
3030 {
3031 	int r;
3032 	unsigned bank_num = mcg_cap & 0xff, bank;
3033 
3034 	r = -EINVAL;
3035 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3036 		goto out;
3037 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
3038 		goto out;
3039 	r = 0;
3040 	vcpu->arch.mcg_cap = mcg_cap;
3041 	/* Init IA32_MCG_CTL to all 1s */
3042 	if (mcg_cap & MCG_CTL_P)
3043 		vcpu->arch.mcg_ctl = ~(u64)0;
3044 	/* Init IA32_MCi_CTL to all 1s */
3045 	for (bank = 0; bank < bank_num; bank++)
3046 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3047 out:
3048 	return r;
3049 }
3050 
3051 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3052 				      struct kvm_x86_mce *mce)
3053 {
3054 	u64 mcg_cap = vcpu->arch.mcg_cap;
3055 	unsigned bank_num = mcg_cap & 0xff;
3056 	u64 *banks = vcpu->arch.mce_banks;
3057 
3058 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3059 		return -EINVAL;
3060 	/*
3061 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3062 	 * reporting is disabled
3063 	 */
3064 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3065 	    vcpu->arch.mcg_ctl != ~(u64)0)
3066 		return 0;
3067 	banks += 4 * mce->bank;
3068 	/*
3069 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3070 	 * reporting is disabled for the bank
3071 	 */
3072 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3073 		return 0;
3074 	if (mce->status & MCI_STATUS_UC) {
3075 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3076 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3077 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3078 			return 0;
3079 		}
3080 		if (banks[1] & MCI_STATUS_VAL)
3081 			mce->status |= MCI_STATUS_OVER;
3082 		banks[2] = mce->addr;
3083 		banks[3] = mce->misc;
3084 		vcpu->arch.mcg_status = mce->mcg_status;
3085 		banks[1] = mce->status;
3086 		kvm_queue_exception(vcpu, MC_VECTOR);
3087 	} else if (!(banks[1] & MCI_STATUS_VAL)
3088 		   || !(banks[1] & MCI_STATUS_UC)) {
3089 		if (banks[1] & MCI_STATUS_VAL)
3090 			mce->status |= MCI_STATUS_OVER;
3091 		banks[2] = mce->addr;
3092 		banks[3] = mce->misc;
3093 		banks[1] = mce->status;
3094 	} else
3095 		banks[1] |= MCI_STATUS_OVER;
3096 	return 0;
3097 }
3098 
3099 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3100 					       struct kvm_vcpu_events *events)
3101 {
3102 	process_nmi(vcpu);
3103 	events->exception.injected =
3104 		vcpu->arch.exception.pending &&
3105 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
3106 	events->exception.nr = vcpu->arch.exception.nr;
3107 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3108 	events->exception.pad = 0;
3109 	events->exception.error_code = vcpu->arch.exception.error_code;
3110 
3111 	events->interrupt.injected =
3112 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3113 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3114 	events->interrupt.soft = 0;
3115 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3116 
3117 	events->nmi.injected = vcpu->arch.nmi_injected;
3118 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3119 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3120 	events->nmi.pad = 0;
3121 
3122 	events->sipi_vector = 0; /* never valid when reporting to user space */
3123 
3124 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3125 			 | KVM_VCPUEVENT_VALID_SHADOW);
3126 	memset(&events->reserved, 0, sizeof(events->reserved));
3127 }
3128 
3129 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3130 					      struct kvm_vcpu_events *events)
3131 {
3132 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3133 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3134 			      | KVM_VCPUEVENT_VALID_SHADOW))
3135 		return -EINVAL;
3136 
3137 	process_nmi(vcpu);
3138 	vcpu->arch.exception.pending = events->exception.injected;
3139 	vcpu->arch.exception.nr = events->exception.nr;
3140 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3141 	vcpu->arch.exception.error_code = events->exception.error_code;
3142 
3143 	vcpu->arch.interrupt.pending = events->interrupt.injected;
3144 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3145 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3146 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3147 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3148 						  events->interrupt.shadow);
3149 
3150 	vcpu->arch.nmi_injected = events->nmi.injected;
3151 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3152 		vcpu->arch.nmi_pending = events->nmi.pending;
3153 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3154 
3155 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3156 	    kvm_vcpu_has_lapic(vcpu))
3157 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3158 
3159 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3160 
3161 	return 0;
3162 }
3163 
3164 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3165 					     struct kvm_debugregs *dbgregs)
3166 {
3167 	unsigned long val;
3168 
3169 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3170 	kvm_get_dr(vcpu, 6, &val);
3171 	dbgregs->dr6 = val;
3172 	dbgregs->dr7 = vcpu->arch.dr7;
3173 	dbgregs->flags = 0;
3174 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3175 }
3176 
3177 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3178 					    struct kvm_debugregs *dbgregs)
3179 {
3180 	if (dbgregs->flags)
3181 		return -EINVAL;
3182 
3183 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3184 	kvm_update_dr0123(vcpu);
3185 	vcpu->arch.dr6 = dbgregs->dr6;
3186 	kvm_update_dr6(vcpu);
3187 	vcpu->arch.dr7 = dbgregs->dr7;
3188 	kvm_update_dr7(vcpu);
3189 
3190 	return 0;
3191 }
3192 
3193 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3194 
3195 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3196 {
3197 	struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3198 	u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3199 	u64 valid;
3200 
3201 	/*
3202 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3203 	 * leaves 0 and 1 in the loop below.
3204 	 */
3205 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3206 
3207 	/* Set XSTATE_BV */
3208 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3209 
3210 	/*
3211 	 * Copy each region from the possibly compacted offset to the
3212 	 * non-compacted offset.
3213 	 */
3214 	valid = xstate_bv & ~XSTATE_FPSSE;
3215 	while (valid) {
3216 		u64 feature = valid & -valid;
3217 		int index = fls64(feature) - 1;
3218 		void *src = get_xsave_addr(xsave, feature);
3219 
3220 		if (src) {
3221 			u32 size, offset, ecx, edx;
3222 			cpuid_count(XSTATE_CPUID, index,
3223 				    &size, &offset, &ecx, &edx);
3224 			memcpy(dest + offset, src, size);
3225 		}
3226 
3227 		valid -= feature;
3228 	}
3229 }
3230 
3231 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3232 {
3233 	struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3234 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3235 	u64 valid;
3236 
3237 	/*
3238 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3239 	 * leaves 0 and 1 in the loop below.
3240 	 */
3241 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3242 
3243 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3244 	xsave->xsave_hdr.xstate_bv = xstate_bv;
3245 	if (cpu_has_xsaves)
3246 		xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3247 
3248 	/*
3249 	 * Copy each region from the non-compacted offset to the
3250 	 * possibly compacted offset.
3251 	 */
3252 	valid = xstate_bv & ~XSTATE_FPSSE;
3253 	while (valid) {
3254 		u64 feature = valid & -valid;
3255 		int index = fls64(feature) - 1;
3256 		void *dest = get_xsave_addr(xsave, feature);
3257 
3258 		if (dest) {
3259 			u32 size, offset, ecx, edx;
3260 			cpuid_count(XSTATE_CPUID, index,
3261 				    &size, &offset, &ecx, &edx);
3262 			memcpy(dest, src + offset, size);
3263 		} else
3264 			WARN_ON_ONCE(1);
3265 
3266 		valid -= feature;
3267 	}
3268 }
3269 
3270 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3271 					 struct kvm_xsave *guest_xsave)
3272 {
3273 	if (cpu_has_xsave) {
3274 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3275 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3276 	} else {
3277 		memcpy(guest_xsave->region,
3278 			&vcpu->arch.guest_fpu.state->fxsave,
3279 			sizeof(struct i387_fxsave_struct));
3280 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3281 			XSTATE_FPSSE;
3282 	}
3283 }
3284 
3285 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3286 					struct kvm_xsave *guest_xsave)
3287 {
3288 	u64 xstate_bv =
3289 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3290 
3291 	if (cpu_has_xsave) {
3292 		/*
3293 		 * Here we allow setting states that are not present in
3294 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3295 		 * with old userspace.
3296 		 */
3297 		if (xstate_bv & ~kvm_supported_xcr0())
3298 			return -EINVAL;
3299 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3300 	} else {
3301 		if (xstate_bv & ~XSTATE_FPSSE)
3302 			return -EINVAL;
3303 		memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3304 			guest_xsave->region, sizeof(struct i387_fxsave_struct));
3305 	}
3306 	return 0;
3307 }
3308 
3309 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3310 					struct kvm_xcrs *guest_xcrs)
3311 {
3312 	if (!cpu_has_xsave) {
3313 		guest_xcrs->nr_xcrs = 0;
3314 		return;
3315 	}
3316 
3317 	guest_xcrs->nr_xcrs = 1;
3318 	guest_xcrs->flags = 0;
3319 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3320 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3321 }
3322 
3323 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3324 				       struct kvm_xcrs *guest_xcrs)
3325 {
3326 	int i, r = 0;
3327 
3328 	if (!cpu_has_xsave)
3329 		return -EINVAL;
3330 
3331 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3332 		return -EINVAL;
3333 
3334 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3335 		/* Only support XCR0 currently */
3336 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3337 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3338 				guest_xcrs->xcrs[i].value);
3339 			break;
3340 		}
3341 	if (r)
3342 		r = -EINVAL;
3343 	return r;
3344 }
3345 
3346 /*
3347  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3348  * stopped by the hypervisor.  This function will be called from the host only.
3349  * EINVAL is returned when the host attempts to set the flag for a guest that
3350  * does not support pv clocks.
3351  */
3352 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3353 {
3354 	if (!vcpu->arch.pv_time_enabled)
3355 		return -EINVAL;
3356 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3357 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3358 	return 0;
3359 }
3360 
3361 long kvm_arch_vcpu_ioctl(struct file *filp,
3362 			 unsigned int ioctl, unsigned long arg)
3363 {
3364 	struct kvm_vcpu *vcpu = filp->private_data;
3365 	void __user *argp = (void __user *)arg;
3366 	int r;
3367 	union {
3368 		struct kvm_lapic_state *lapic;
3369 		struct kvm_xsave *xsave;
3370 		struct kvm_xcrs *xcrs;
3371 		void *buffer;
3372 	} u;
3373 
3374 	u.buffer = NULL;
3375 	switch (ioctl) {
3376 	case KVM_GET_LAPIC: {
3377 		r = -EINVAL;
3378 		if (!vcpu->arch.apic)
3379 			goto out;
3380 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3381 
3382 		r = -ENOMEM;
3383 		if (!u.lapic)
3384 			goto out;
3385 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3386 		if (r)
3387 			goto out;
3388 		r = -EFAULT;
3389 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3390 			goto out;
3391 		r = 0;
3392 		break;
3393 	}
3394 	case KVM_SET_LAPIC: {
3395 		r = -EINVAL;
3396 		if (!vcpu->arch.apic)
3397 			goto out;
3398 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3399 		if (IS_ERR(u.lapic))
3400 			return PTR_ERR(u.lapic);
3401 
3402 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3403 		break;
3404 	}
3405 	case KVM_INTERRUPT: {
3406 		struct kvm_interrupt irq;
3407 
3408 		r = -EFAULT;
3409 		if (copy_from_user(&irq, argp, sizeof irq))
3410 			goto out;
3411 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3412 		break;
3413 	}
3414 	case KVM_NMI: {
3415 		r = kvm_vcpu_ioctl_nmi(vcpu);
3416 		break;
3417 	}
3418 	case KVM_SET_CPUID: {
3419 		struct kvm_cpuid __user *cpuid_arg = argp;
3420 		struct kvm_cpuid cpuid;
3421 
3422 		r = -EFAULT;
3423 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3424 			goto out;
3425 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3426 		break;
3427 	}
3428 	case KVM_SET_CPUID2: {
3429 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3430 		struct kvm_cpuid2 cpuid;
3431 
3432 		r = -EFAULT;
3433 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3434 			goto out;
3435 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3436 					      cpuid_arg->entries);
3437 		break;
3438 	}
3439 	case KVM_GET_CPUID2: {
3440 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3441 		struct kvm_cpuid2 cpuid;
3442 
3443 		r = -EFAULT;
3444 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3445 			goto out;
3446 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3447 					      cpuid_arg->entries);
3448 		if (r)
3449 			goto out;
3450 		r = -EFAULT;
3451 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3452 			goto out;
3453 		r = 0;
3454 		break;
3455 	}
3456 	case KVM_GET_MSRS:
3457 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
3458 		break;
3459 	case KVM_SET_MSRS:
3460 		r = msr_io(vcpu, argp, do_set_msr, 0);
3461 		break;
3462 	case KVM_TPR_ACCESS_REPORTING: {
3463 		struct kvm_tpr_access_ctl tac;
3464 
3465 		r = -EFAULT;
3466 		if (copy_from_user(&tac, argp, sizeof tac))
3467 			goto out;
3468 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3469 		if (r)
3470 			goto out;
3471 		r = -EFAULT;
3472 		if (copy_to_user(argp, &tac, sizeof tac))
3473 			goto out;
3474 		r = 0;
3475 		break;
3476 	};
3477 	case KVM_SET_VAPIC_ADDR: {
3478 		struct kvm_vapic_addr va;
3479 
3480 		r = -EINVAL;
3481 		if (!irqchip_in_kernel(vcpu->kvm))
3482 			goto out;
3483 		r = -EFAULT;
3484 		if (copy_from_user(&va, argp, sizeof va))
3485 			goto out;
3486 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3487 		break;
3488 	}
3489 	case KVM_X86_SETUP_MCE: {
3490 		u64 mcg_cap;
3491 
3492 		r = -EFAULT;
3493 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3494 			goto out;
3495 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3496 		break;
3497 	}
3498 	case KVM_X86_SET_MCE: {
3499 		struct kvm_x86_mce mce;
3500 
3501 		r = -EFAULT;
3502 		if (copy_from_user(&mce, argp, sizeof mce))
3503 			goto out;
3504 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3505 		break;
3506 	}
3507 	case KVM_GET_VCPU_EVENTS: {
3508 		struct kvm_vcpu_events events;
3509 
3510 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3511 
3512 		r = -EFAULT;
3513 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3514 			break;
3515 		r = 0;
3516 		break;
3517 	}
3518 	case KVM_SET_VCPU_EVENTS: {
3519 		struct kvm_vcpu_events events;
3520 
3521 		r = -EFAULT;
3522 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3523 			break;
3524 
3525 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3526 		break;
3527 	}
3528 	case KVM_GET_DEBUGREGS: {
3529 		struct kvm_debugregs dbgregs;
3530 
3531 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3532 
3533 		r = -EFAULT;
3534 		if (copy_to_user(argp, &dbgregs,
3535 				 sizeof(struct kvm_debugregs)))
3536 			break;
3537 		r = 0;
3538 		break;
3539 	}
3540 	case KVM_SET_DEBUGREGS: {
3541 		struct kvm_debugregs dbgregs;
3542 
3543 		r = -EFAULT;
3544 		if (copy_from_user(&dbgregs, argp,
3545 				   sizeof(struct kvm_debugregs)))
3546 			break;
3547 
3548 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3549 		break;
3550 	}
3551 	case KVM_GET_XSAVE: {
3552 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3553 		r = -ENOMEM;
3554 		if (!u.xsave)
3555 			break;
3556 
3557 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3558 
3559 		r = -EFAULT;
3560 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3561 			break;
3562 		r = 0;
3563 		break;
3564 	}
3565 	case KVM_SET_XSAVE: {
3566 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3567 		if (IS_ERR(u.xsave))
3568 			return PTR_ERR(u.xsave);
3569 
3570 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3571 		break;
3572 	}
3573 	case KVM_GET_XCRS: {
3574 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3575 		r = -ENOMEM;
3576 		if (!u.xcrs)
3577 			break;
3578 
3579 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3580 
3581 		r = -EFAULT;
3582 		if (copy_to_user(argp, u.xcrs,
3583 				 sizeof(struct kvm_xcrs)))
3584 			break;
3585 		r = 0;
3586 		break;
3587 	}
3588 	case KVM_SET_XCRS: {
3589 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3590 		if (IS_ERR(u.xcrs))
3591 			return PTR_ERR(u.xcrs);
3592 
3593 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3594 		break;
3595 	}
3596 	case KVM_SET_TSC_KHZ: {
3597 		u32 user_tsc_khz;
3598 
3599 		r = -EINVAL;
3600 		user_tsc_khz = (u32)arg;
3601 
3602 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3603 			goto out;
3604 
3605 		if (user_tsc_khz == 0)
3606 			user_tsc_khz = tsc_khz;
3607 
3608 		kvm_set_tsc_khz(vcpu, user_tsc_khz);
3609 
3610 		r = 0;
3611 		goto out;
3612 	}
3613 	case KVM_GET_TSC_KHZ: {
3614 		r = vcpu->arch.virtual_tsc_khz;
3615 		goto out;
3616 	}
3617 	case KVM_KVMCLOCK_CTRL: {
3618 		r = kvm_set_guest_paused(vcpu);
3619 		goto out;
3620 	}
3621 	default:
3622 		r = -EINVAL;
3623 	}
3624 out:
3625 	kfree(u.buffer);
3626 	return r;
3627 }
3628 
3629 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3630 {
3631 	return VM_FAULT_SIGBUS;
3632 }
3633 
3634 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3635 {
3636 	int ret;
3637 
3638 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3639 		return -EINVAL;
3640 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3641 	return ret;
3642 }
3643 
3644 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3645 					      u64 ident_addr)
3646 {
3647 	kvm->arch.ept_identity_map_addr = ident_addr;
3648 	return 0;
3649 }
3650 
3651 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3652 					  u32 kvm_nr_mmu_pages)
3653 {
3654 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3655 		return -EINVAL;
3656 
3657 	mutex_lock(&kvm->slots_lock);
3658 
3659 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3660 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3661 
3662 	mutex_unlock(&kvm->slots_lock);
3663 	return 0;
3664 }
3665 
3666 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3667 {
3668 	return kvm->arch.n_max_mmu_pages;
3669 }
3670 
3671 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3672 {
3673 	int r;
3674 
3675 	r = 0;
3676 	switch (chip->chip_id) {
3677 	case KVM_IRQCHIP_PIC_MASTER:
3678 		memcpy(&chip->chip.pic,
3679 			&pic_irqchip(kvm)->pics[0],
3680 			sizeof(struct kvm_pic_state));
3681 		break;
3682 	case KVM_IRQCHIP_PIC_SLAVE:
3683 		memcpy(&chip->chip.pic,
3684 			&pic_irqchip(kvm)->pics[1],
3685 			sizeof(struct kvm_pic_state));
3686 		break;
3687 	case KVM_IRQCHIP_IOAPIC:
3688 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3689 		break;
3690 	default:
3691 		r = -EINVAL;
3692 		break;
3693 	}
3694 	return r;
3695 }
3696 
3697 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3698 {
3699 	int r;
3700 
3701 	r = 0;
3702 	switch (chip->chip_id) {
3703 	case KVM_IRQCHIP_PIC_MASTER:
3704 		spin_lock(&pic_irqchip(kvm)->lock);
3705 		memcpy(&pic_irqchip(kvm)->pics[0],
3706 			&chip->chip.pic,
3707 			sizeof(struct kvm_pic_state));
3708 		spin_unlock(&pic_irqchip(kvm)->lock);
3709 		break;
3710 	case KVM_IRQCHIP_PIC_SLAVE:
3711 		spin_lock(&pic_irqchip(kvm)->lock);
3712 		memcpy(&pic_irqchip(kvm)->pics[1],
3713 			&chip->chip.pic,
3714 			sizeof(struct kvm_pic_state));
3715 		spin_unlock(&pic_irqchip(kvm)->lock);
3716 		break;
3717 	case KVM_IRQCHIP_IOAPIC:
3718 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3719 		break;
3720 	default:
3721 		r = -EINVAL;
3722 		break;
3723 	}
3724 	kvm_pic_update_irq(pic_irqchip(kvm));
3725 	return r;
3726 }
3727 
3728 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3729 {
3730 	int r = 0;
3731 
3732 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3733 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3734 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3735 	return r;
3736 }
3737 
3738 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3739 {
3740 	int r = 0;
3741 
3742 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3743 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3744 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3745 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3746 	return r;
3747 }
3748 
3749 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3750 {
3751 	int r = 0;
3752 
3753 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3754 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3755 		sizeof(ps->channels));
3756 	ps->flags = kvm->arch.vpit->pit_state.flags;
3757 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3758 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3759 	return r;
3760 }
3761 
3762 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3763 {
3764 	int r = 0, start = 0;
3765 	u32 prev_legacy, cur_legacy;
3766 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3767 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3768 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3769 	if (!prev_legacy && cur_legacy)
3770 		start = 1;
3771 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3772 	       sizeof(kvm->arch.vpit->pit_state.channels));
3773 	kvm->arch.vpit->pit_state.flags = ps->flags;
3774 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3775 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3776 	return r;
3777 }
3778 
3779 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3780 				 struct kvm_reinject_control *control)
3781 {
3782 	if (!kvm->arch.vpit)
3783 		return -ENXIO;
3784 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3785 	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3786 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3787 	return 0;
3788 }
3789 
3790 /**
3791  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3792  * @kvm: kvm instance
3793  * @log: slot id and address to which we copy the log
3794  *
3795  * Steps 1-4 below provide general overview of dirty page logging. See
3796  * kvm_get_dirty_log_protect() function description for additional details.
3797  *
3798  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3799  * always flush the TLB (step 4) even if previous step failed  and the dirty
3800  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3801  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3802  * writes will be marked dirty for next log read.
3803  *
3804  *   1. Take a snapshot of the bit and clear it if needed.
3805  *   2. Write protect the corresponding page.
3806  *   3. Copy the snapshot to the userspace.
3807  *   4. Flush TLB's if needed.
3808  */
3809 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3810 {
3811 	bool is_dirty = false;
3812 	int r;
3813 
3814 	mutex_lock(&kvm->slots_lock);
3815 
3816 	/*
3817 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3818 	 */
3819 	if (kvm_x86_ops->flush_log_dirty)
3820 		kvm_x86_ops->flush_log_dirty(kvm);
3821 
3822 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3823 
3824 	/*
3825 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3826 	 * kvm_mmu_slot_remove_write_access().
3827 	 */
3828 	lockdep_assert_held(&kvm->slots_lock);
3829 	if (is_dirty)
3830 		kvm_flush_remote_tlbs(kvm);
3831 
3832 	mutex_unlock(&kvm->slots_lock);
3833 	return r;
3834 }
3835 
3836 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3837 			bool line_status)
3838 {
3839 	if (!irqchip_in_kernel(kvm))
3840 		return -ENXIO;
3841 
3842 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3843 					irq_event->irq, irq_event->level,
3844 					line_status);
3845 	return 0;
3846 }
3847 
3848 long kvm_arch_vm_ioctl(struct file *filp,
3849 		       unsigned int ioctl, unsigned long arg)
3850 {
3851 	struct kvm *kvm = filp->private_data;
3852 	void __user *argp = (void __user *)arg;
3853 	int r = -ENOTTY;
3854 	/*
3855 	 * This union makes it completely explicit to gcc-3.x
3856 	 * that these two variables' stack usage should be
3857 	 * combined, not added together.
3858 	 */
3859 	union {
3860 		struct kvm_pit_state ps;
3861 		struct kvm_pit_state2 ps2;
3862 		struct kvm_pit_config pit_config;
3863 	} u;
3864 
3865 	switch (ioctl) {
3866 	case KVM_SET_TSS_ADDR:
3867 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3868 		break;
3869 	case KVM_SET_IDENTITY_MAP_ADDR: {
3870 		u64 ident_addr;
3871 
3872 		r = -EFAULT;
3873 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3874 			goto out;
3875 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3876 		break;
3877 	}
3878 	case KVM_SET_NR_MMU_PAGES:
3879 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3880 		break;
3881 	case KVM_GET_NR_MMU_PAGES:
3882 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3883 		break;
3884 	case KVM_CREATE_IRQCHIP: {
3885 		struct kvm_pic *vpic;
3886 
3887 		mutex_lock(&kvm->lock);
3888 		r = -EEXIST;
3889 		if (kvm->arch.vpic)
3890 			goto create_irqchip_unlock;
3891 		r = -EINVAL;
3892 		if (atomic_read(&kvm->online_vcpus))
3893 			goto create_irqchip_unlock;
3894 		r = -ENOMEM;
3895 		vpic = kvm_create_pic(kvm);
3896 		if (vpic) {
3897 			r = kvm_ioapic_init(kvm);
3898 			if (r) {
3899 				mutex_lock(&kvm->slots_lock);
3900 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3901 							  &vpic->dev_master);
3902 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3903 							  &vpic->dev_slave);
3904 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3905 							  &vpic->dev_eclr);
3906 				mutex_unlock(&kvm->slots_lock);
3907 				kfree(vpic);
3908 				goto create_irqchip_unlock;
3909 			}
3910 		} else
3911 			goto create_irqchip_unlock;
3912 		smp_wmb();
3913 		kvm->arch.vpic = vpic;
3914 		smp_wmb();
3915 		r = kvm_setup_default_irq_routing(kvm);
3916 		if (r) {
3917 			mutex_lock(&kvm->slots_lock);
3918 			mutex_lock(&kvm->irq_lock);
3919 			kvm_ioapic_destroy(kvm);
3920 			kvm_destroy_pic(kvm);
3921 			mutex_unlock(&kvm->irq_lock);
3922 			mutex_unlock(&kvm->slots_lock);
3923 		}
3924 	create_irqchip_unlock:
3925 		mutex_unlock(&kvm->lock);
3926 		break;
3927 	}
3928 	case KVM_CREATE_PIT:
3929 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3930 		goto create_pit;
3931 	case KVM_CREATE_PIT2:
3932 		r = -EFAULT;
3933 		if (copy_from_user(&u.pit_config, argp,
3934 				   sizeof(struct kvm_pit_config)))
3935 			goto out;
3936 	create_pit:
3937 		mutex_lock(&kvm->slots_lock);
3938 		r = -EEXIST;
3939 		if (kvm->arch.vpit)
3940 			goto create_pit_unlock;
3941 		r = -ENOMEM;
3942 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3943 		if (kvm->arch.vpit)
3944 			r = 0;
3945 	create_pit_unlock:
3946 		mutex_unlock(&kvm->slots_lock);
3947 		break;
3948 	case KVM_GET_IRQCHIP: {
3949 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3950 		struct kvm_irqchip *chip;
3951 
3952 		chip = memdup_user(argp, sizeof(*chip));
3953 		if (IS_ERR(chip)) {
3954 			r = PTR_ERR(chip);
3955 			goto out;
3956 		}
3957 
3958 		r = -ENXIO;
3959 		if (!irqchip_in_kernel(kvm))
3960 			goto get_irqchip_out;
3961 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3962 		if (r)
3963 			goto get_irqchip_out;
3964 		r = -EFAULT;
3965 		if (copy_to_user(argp, chip, sizeof *chip))
3966 			goto get_irqchip_out;
3967 		r = 0;
3968 	get_irqchip_out:
3969 		kfree(chip);
3970 		break;
3971 	}
3972 	case KVM_SET_IRQCHIP: {
3973 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3974 		struct kvm_irqchip *chip;
3975 
3976 		chip = memdup_user(argp, sizeof(*chip));
3977 		if (IS_ERR(chip)) {
3978 			r = PTR_ERR(chip);
3979 			goto out;
3980 		}
3981 
3982 		r = -ENXIO;
3983 		if (!irqchip_in_kernel(kvm))
3984 			goto set_irqchip_out;
3985 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3986 		if (r)
3987 			goto set_irqchip_out;
3988 		r = 0;
3989 	set_irqchip_out:
3990 		kfree(chip);
3991 		break;
3992 	}
3993 	case KVM_GET_PIT: {
3994 		r = -EFAULT;
3995 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3996 			goto out;
3997 		r = -ENXIO;
3998 		if (!kvm->arch.vpit)
3999 			goto out;
4000 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4001 		if (r)
4002 			goto out;
4003 		r = -EFAULT;
4004 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4005 			goto out;
4006 		r = 0;
4007 		break;
4008 	}
4009 	case KVM_SET_PIT: {
4010 		r = -EFAULT;
4011 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
4012 			goto out;
4013 		r = -ENXIO;
4014 		if (!kvm->arch.vpit)
4015 			goto out;
4016 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4017 		break;
4018 	}
4019 	case KVM_GET_PIT2: {
4020 		r = -ENXIO;
4021 		if (!kvm->arch.vpit)
4022 			goto out;
4023 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4024 		if (r)
4025 			goto out;
4026 		r = -EFAULT;
4027 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4028 			goto out;
4029 		r = 0;
4030 		break;
4031 	}
4032 	case KVM_SET_PIT2: {
4033 		r = -EFAULT;
4034 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4035 			goto out;
4036 		r = -ENXIO;
4037 		if (!kvm->arch.vpit)
4038 			goto out;
4039 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4040 		break;
4041 	}
4042 	case KVM_REINJECT_CONTROL: {
4043 		struct kvm_reinject_control control;
4044 		r =  -EFAULT;
4045 		if (copy_from_user(&control, argp, sizeof(control)))
4046 			goto out;
4047 		r = kvm_vm_ioctl_reinject(kvm, &control);
4048 		break;
4049 	}
4050 	case KVM_XEN_HVM_CONFIG: {
4051 		r = -EFAULT;
4052 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4053 				   sizeof(struct kvm_xen_hvm_config)))
4054 			goto out;
4055 		r = -EINVAL;
4056 		if (kvm->arch.xen_hvm_config.flags)
4057 			goto out;
4058 		r = 0;
4059 		break;
4060 	}
4061 	case KVM_SET_CLOCK: {
4062 		struct kvm_clock_data user_ns;
4063 		u64 now_ns;
4064 		s64 delta;
4065 
4066 		r = -EFAULT;
4067 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4068 			goto out;
4069 
4070 		r = -EINVAL;
4071 		if (user_ns.flags)
4072 			goto out;
4073 
4074 		r = 0;
4075 		local_irq_disable();
4076 		now_ns = get_kernel_ns();
4077 		delta = user_ns.clock - now_ns;
4078 		local_irq_enable();
4079 		kvm->arch.kvmclock_offset = delta;
4080 		kvm_gen_update_masterclock(kvm);
4081 		break;
4082 	}
4083 	case KVM_GET_CLOCK: {
4084 		struct kvm_clock_data user_ns;
4085 		u64 now_ns;
4086 
4087 		local_irq_disable();
4088 		now_ns = get_kernel_ns();
4089 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4090 		local_irq_enable();
4091 		user_ns.flags = 0;
4092 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4093 
4094 		r = -EFAULT;
4095 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4096 			goto out;
4097 		r = 0;
4098 		break;
4099 	}
4100 
4101 	default:
4102 		r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4103 	}
4104 out:
4105 	return r;
4106 }
4107 
4108 static void kvm_init_msr_list(void)
4109 {
4110 	u32 dummy[2];
4111 	unsigned i, j;
4112 
4113 	/* skip the first msrs in the list. KVM-specific */
4114 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
4115 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4116 			continue;
4117 
4118 		/*
4119 		 * Even MSRs that are valid in the host may not be exposed
4120 		 * to the guests in some cases.  We could work around this
4121 		 * in VMX with the generic MSR save/load machinery, but it
4122 		 * is not really worthwhile since it will really only
4123 		 * happen with nested virtualization.
4124 		 */
4125 		switch (msrs_to_save[i]) {
4126 		case MSR_IA32_BNDCFGS:
4127 			if (!kvm_x86_ops->mpx_supported())
4128 				continue;
4129 			break;
4130 		default:
4131 			break;
4132 		}
4133 
4134 		if (j < i)
4135 			msrs_to_save[j] = msrs_to_save[i];
4136 		j++;
4137 	}
4138 	num_msrs_to_save = j;
4139 }
4140 
4141 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4142 			   const void *v)
4143 {
4144 	int handled = 0;
4145 	int n;
4146 
4147 	do {
4148 		n = min(len, 8);
4149 		if (!(vcpu->arch.apic &&
4150 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4151 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4152 			break;
4153 		handled += n;
4154 		addr += n;
4155 		len -= n;
4156 		v += n;
4157 	} while (len);
4158 
4159 	return handled;
4160 }
4161 
4162 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4163 {
4164 	int handled = 0;
4165 	int n;
4166 
4167 	do {
4168 		n = min(len, 8);
4169 		if (!(vcpu->arch.apic &&
4170 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4171 					 addr, n, v))
4172 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4173 			break;
4174 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4175 		handled += n;
4176 		addr += n;
4177 		len -= n;
4178 		v += n;
4179 	} while (len);
4180 
4181 	return handled;
4182 }
4183 
4184 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4185 			struct kvm_segment *var, int seg)
4186 {
4187 	kvm_x86_ops->set_segment(vcpu, var, seg);
4188 }
4189 
4190 void kvm_get_segment(struct kvm_vcpu *vcpu,
4191 		     struct kvm_segment *var, int seg)
4192 {
4193 	kvm_x86_ops->get_segment(vcpu, var, seg);
4194 }
4195 
4196 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4197 			   struct x86_exception *exception)
4198 {
4199 	gpa_t t_gpa;
4200 
4201 	BUG_ON(!mmu_is_nested(vcpu));
4202 
4203 	/* NPT walks are always user-walks */
4204 	access |= PFERR_USER_MASK;
4205 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4206 
4207 	return t_gpa;
4208 }
4209 
4210 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4211 			      struct x86_exception *exception)
4212 {
4213 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4214 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4215 }
4216 
4217  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4218 				struct x86_exception *exception)
4219 {
4220 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4221 	access |= PFERR_FETCH_MASK;
4222 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4223 }
4224 
4225 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4226 			       struct x86_exception *exception)
4227 {
4228 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4229 	access |= PFERR_WRITE_MASK;
4230 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4231 }
4232 
4233 /* uses this to access any guest's mapped memory without checking CPL */
4234 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4235 				struct x86_exception *exception)
4236 {
4237 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4238 }
4239 
4240 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4241 				      struct kvm_vcpu *vcpu, u32 access,
4242 				      struct x86_exception *exception)
4243 {
4244 	void *data = val;
4245 	int r = X86EMUL_CONTINUE;
4246 
4247 	while (bytes) {
4248 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4249 							    exception);
4250 		unsigned offset = addr & (PAGE_SIZE-1);
4251 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4252 		int ret;
4253 
4254 		if (gpa == UNMAPPED_GVA)
4255 			return X86EMUL_PROPAGATE_FAULT;
4256 		ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4257 					  offset, toread);
4258 		if (ret < 0) {
4259 			r = X86EMUL_IO_NEEDED;
4260 			goto out;
4261 		}
4262 
4263 		bytes -= toread;
4264 		data += toread;
4265 		addr += toread;
4266 	}
4267 out:
4268 	return r;
4269 }
4270 
4271 /* used for instruction fetching */
4272 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4273 				gva_t addr, void *val, unsigned int bytes,
4274 				struct x86_exception *exception)
4275 {
4276 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4277 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4278 	unsigned offset;
4279 	int ret;
4280 
4281 	/* Inline kvm_read_guest_virt_helper for speed.  */
4282 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4283 						    exception);
4284 	if (unlikely(gpa == UNMAPPED_GVA))
4285 		return X86EMUL_PROPAGATE_FAULT;
4286 
4287 	offset = addr & (PAGE_SIZE-1);
4288 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4289 		bytes = (unsigned)PAGE_SIZE - offset;
4290 	ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4291 				  offset, bytes);
4292 	if (unlikely(ret < 0))
4293 		return X86EMUL_IO_NEEDED;
4294 
4295 	return X86EMUL_CONTINUE;
4296 }
4297 
4298 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4299 			       gva_t addr, void *val, unsigned int bytes,
4300 			       struct x86_exception *exception)
4301 {
4302 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4303 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4304 
4305 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4306 					  exception);
4307 }
4308 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4309 
4310 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4311 				      gva_t addr, void *val, unsigned int bytes,
4312 				      struct x86_exception *exception)
4313 {
4314 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4315 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4316 }
4317 
4318 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4319 				       gva_t addr, void *val,
4320 				       unsigned int bytes,
4321 				       struct x86_exception *exception)
4322 {
4323 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4324 	void *data = val;
4325 	int r = X86EMUL_CONTINUE;
4326 
4327 	while (bytes) {
4328 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4329 							     PFERR_WRITE_MASK,
4330 							     exception);
4331 		unsigned offset = addr & (PAGE_SIZE-1);
4332 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4333 		int ret;
4334 
4335 		if (gpa == UNMAPPED_GVA)
4336 			return X86EMUL_PROPAGATE_FAULT;
4337 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4338 		if (ret < 0) {
4339 			r = X86EMUL_IO_NEEDED;
4340 			goto out;
4341 		}
4342 
4343 		bytes -= towrite;
4344 		data += towrite;
4345 		addr += towrite;
4346 	}
4347 out:
4348 	return r;
4349 }
4350 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4351 
4352 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4353 				gpa_t *gpa, struct x86_exception *exception,
4354 				bool write)
4355 {
4356 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4357 		| (write ? PFERR_WRITE_MASK : 0);
4358 
4359 	if (vcpu_match_mmio_gva(vcpu, gva)
4360 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4361 				 vcpu->arch.access, access)) {
4362 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4363 					(gva & (PAGE_SIZE - 1));
4364 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4365 		return 1;
4366 	}
4367 
4368 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4369 
4370 	if (*gpa == UNMAPPED_GVA)
4371 		return -1;
4372 
4373 	/* For APIC access vmexit */
4374 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4375 		return 1;
4376 
4377 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4378 		trace_vcpu_match_mmio(gva, *gpa, write, true);
4379 		return 1;
4380 	}
4381 
4382 	return 0;
4383 }
4384 
4385 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4386 			const void *val, int bytes)
4387 {
4388 	int ret;
4389 
4390 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4391 	if (ret < 0)
4392 		return 0;
4393 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4394 	return 1;
4395 }
4396 
4397 struct read_write_emulator_ops {
4398 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4399 				  int bytes);
4400 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4401 				  void *val, int bytes);
4402 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4403 			       int bytes, void *val);
4404 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4405 				    void *val, int bytes);
4406 	bool write;
4407 };
4408 
4409 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4410 {
4411 	if (vcpu->mmio_read_completed) {
4412 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4413 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4414 		vcpu->mmio_read_completed = 0;
4415 		return 1;
4416 	}
4417 
4418 	return 0;
4419 }
4420 
4421 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4422 			void *val, int bytes)
4423 {
4424 	return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4425 }
4426 
4427 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4428 			 void *val, int bytes)
4429 {
4430 	return emulator_write_phys(vcpu, gpa, val, bytes);
4431 }
4432 
4433 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4434 {
4435 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4436 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4437 }
4438 
4439 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4440 			  void *val, int bytes)
4441 {
4442 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4443 	return X86EMUL_IO_NEEDED;
4444 }
4445 
4446 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4447 			   void *val, int bytes)
4448 {
4449 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4450 
4451 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4452 	return X86EMUL_CONTINUE;
4453 }
4454 
4455 static const struct read_write_emulator_ops read_emultor = {
4456 	.read_write_prepare = read_prepare,
4457 	.read_write_emulate = read_emulate,
4458 	.read_write_mmio = vcpu_mmio_read,
4459 	.read_write_exit_mmio = read_exit_mmio,
4460 };
4461 
4462 static const struct read_write_emulator_ops write_emultor = {
4463 	.read_write_emulate = write_emulate,
4464 	.read_write_mmio = write_mmio,
4465 	.read_write_exit_mmio = write_exit_mmio,
4466 	.write = true,
4467 };
4468 
4469 static int emulator_read_write_onepage(unsigned long addr, void *val,
4470 				       unsigned int bytes,
4471 				       struct x86_exception *exception,
4472 				       struct kvm_vcpu *vcpu,
4473 				       const struct read_write_emulator_ops *ops)
4474 {
4475 	gpa_t gpa;
4476 	int handled, ret;
4477 	bool write = ops->write;
4478 	struct kvm_mmio_fragment *frag;
4479 
4480 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4481 
4482 	if (ret < 0)
4483 		return X86EMUL_PROPAGATE_FAULT;
4484 
4485 	/* For APIC access vmexit */
4486 	if (ret)
4487 		goto mmio;
4488 
4489 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4490 		return X86EMUL_CONTINUE;
4491 
4492 mmio:
4493 	/*
4494 	 * Is this MMIO handled locally?
4495 	 */
4496 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4497 	if (handled == bytes)
4498 		return X86EMUL_CONTINUE;
4499 
4500 	gpa += handled;
4501 	bytes -= handled;
4502 	val += handled;
4503 
4504 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4505 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4506 	frag->gpa = gpa;
4507 	frag->data = val;
4508 	frag->len = bytes;
4509 	return X86EMUL_CONTINUE;
4510 }
4511 
4512 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4513 			unsigned long addr,
4514 			void *val, unsigned int bytes,
4515 			struct x86_exception *exception,
4516 			const struct read_write_emulator_ops *ops)
4517 {
4518 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4519 	gpa_t gpa;
4520 	int rc;
4521 
4522 	if (ops->read_write_prepare &&
4523 		  ops->read_write_prepare(vcpu, val, bytes))
4524 		return X86EMUL_CONTINUE;
4525 
4526 	vcpu->mmio_nr_fragments = 0;
4527 
4528 	/* Crossing a page boundary? */
4529 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4530 		int now;
4531 
4532 		now = -addr & ~PAGE_MASK;
4533 		rc = emulator_read_write_onepage(addr, val, now, exception,
4534 						 vcpu, ops);
4535 
4536 		if (rc != X86EMUL_CONTINUE)
4537 			return rc;
4538 		addr += now;
4539 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4540 			addr = (u32)addr;
4541 		val += now;
4542 		bytes -= now;
4543 	}
4544 
4545 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4546 					 vcpu, ops);
4547 	if (rc != X86EMUL_CONTINUE)
4548 		return rc;
4549 
4550 	if (!vcpu->mmio_nr_fragments)
4551 		return rc;
4552 
4553 	gpa = vcpu->mmio_fragments[0].gpa;
4554 
4555 	vcpu->mmio_needed = 1;
4556 	vcpu->mmio_cur_fragment = 0;
4557 
4558 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4559 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4560 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4561 	vcpu->run->mmio.phys_addr = gpa;
4562 
4563 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4564 }
4565 
4566 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4567 				  unsigned long addr,
4568 				  void *val,
4569 				  unsigned int bytes,
4570 				  struct x86_exception *exception)
4571 {
4572 	return emulator_read_write(ctxt, addr, val, bytes,
4573 				   exception, &read_emultor);
4574 }
4575 
4576 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4577 			    unsigned long addr,
4578 			    const void *val,
4579 			    unsigned int bytes,
4580 			    struct x86_exception *exception)
4581 {
4582 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4583 				   exception, &write_emultor);
4584 }
4585 
4586 #define CMPXCHG_TYPE(t, ptr, old, new) \
4587 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4588 
4589 #ifdef CONFIG_X86_64
4590 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4591 #else
4592 #  define CMPXCHG64(ptr, old, new) \
4593 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4594 #endif
4595 
4596 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4597 				     unsigned long addr,
4598 				     const void *old,
4599 				     const void *new,
4600 				     unsigned int bytes,
4601 				     struct x86_exception *exception)
4602 {
4603 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4604 	gpa_t gpa;
4605 	struct page *page;
4606 	char *kaddr;
4607 	bool exchanged;
4608 
4609 	/* guests cmpxchg8b have to be emulated atomically */
4610 	if (bytes > 8 || (bytes & (bytes - 1)))
4611 		goto emul_write;
4612 
4613 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4614 
4615 	if (gpa == UNMAPPED_GVA ||
4616 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4617 		goto emul_write;
4618 
4619 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4620 		goto emul_write;
4621 
4622 	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4623 	if (is_error_page(page))
4624 		goto emul_write;
4625 
4626 	kaddr = kmap_atomic(page);
4627 	kaddr += offset_in_page(gpa);
4628 	switch (bytes) {
4629 	case 1:
4630 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4631 		break;
4632 	case 2:
4633 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4634 		break;
4635 	case 4:
4636 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4637 		break;
4638 	case 8:
4639 		exchanged = CMPXCHG64(kaddr, old, new);
4640 		break;
4641 	default:
4642 		BUG();
4643 	}
4644 	kunmap_atomic(kaddr);
4645 	kvm_release_page_dirty(page);
4646 
4647 	if (!exchanged)
4648 		return X86EMUL_CMPXCHG_FAILED;
4649 
4650 	mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4651 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4652 
4653 	return X86EMUL_CONTINUE;
4654 
4655 emul_write:
4656 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4657 
4658 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4659 }
4660 
4661 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4662 {
4663 	/* TODO: String I/O for in kernel device */
4664 	int r;
4665 
4666 	if (vcpu->arch.pio.in)
4667 		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4668 				    vcpu->arch.pio.size, pd);
4669 	else
4670 		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4671 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4672 				     pd);
4673 	return r;
4674 }
4675 
4676 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4677 			       unsigned short port, void *val,
4678 			       unsigned int count, bool in)
4679 {
4680 	vcpu->arch.pio.port = port;
4681 	vcpu->arch.pio.in = in;
4682 	vcpu->arch.pio.count  = count;
4683 	vcpu->arch.pio.size = size;
4684 
4685 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4686 		vcpu->arch.pio.count = 0;
4687 		return 1;
4688 	}
4689 
4690 	vcpu->run->exit_reason = KVM_EXIT_IO;
4691 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4692 	vcpu->run->io.size = size;
4693 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4694 	vcpu->run->io.count = count;
4695 	vcpu->run->io.port = port;
4696 
4697 	return 0;
4698 }
4699 
4700 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4701 				    int size, unsigned short port, void *val,
4702 				    unsigned int count)
4703 {
4704 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4705 	int ret;
4706 
4707 	if (vcpu->arch.pio.count)
4708 		goto data_avail;
4709 
4710 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4711 	if (ret) {
4712 data_avail:
4713 		memcpy(val, vcpu->arch.pio_data, size * count);
4714 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4715 		vcpu->arch.pio.count = 0;
4716 		return 1;
4717 	}
4718 
4719 	return 0;
4720 }
4721 
4722 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4723 				     int size, unsigned short port,
4724 				     const void *val, unsigned int count)
4725 {
4726 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4727 
4728 	memcpy(vcpu->arch.pio_data, val, size * count);
4729 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4730 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4731 }
4732 
4733 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4734 {
4735 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4736 }
4737 
4738 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4739 {
4740 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4741 }
4742 
4743 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4744 {
4745 	if (!need_emulate_wbinvd(vcpu))
4746 		return X86EMUL_CONTINUE;
4747 
4748 	if (kvm_x86_ops->has_wbinvd_exit()) {
4749 		int cpu = get_cpu();
4750 
4751 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4752 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4753 				wbinvd_ipi, NULL, 1);
4754 		put_cpu();
4755 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4756 	} else
4757 		wbinvd();
4758 	return X86EMUL_CONTINUE;
4759 }
4760 
4761 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4762 {
4763 	kvm_x86_ops->skip_emulated_instruction(vcpu);
4764 	return kvm_emulate_wbinvd_noskip(vcpu);
4765 }
4766 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4767 
4768 
4769 
4770 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4771 {
4772 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4773 }
4774 
4775 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4776 			   unsigned long *dest)
4777 {
4778 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4779 }
4780 
4781 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4782 			   unsigned long value)
4783 {
4784 
4785 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4786 }
4787 
4788 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4789 {
4790 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4791 }
4792 
4793 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4794 {
4795 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4796 	unsigned long value;
4797 
4798 	switch (cr) {
4799 	case 0:
4800 		value = kvm_read_cr0(vcpu);
4801 		break;
4802 	case 2:
4803 		value = vcpu->arch.cr2;
4804 		break;
4805 	case 3:
4806 		value = kvm_read_cr3(vcpu);
4807 		break;
4808 	case 4:
4809 		value = kvm_read_cr4(vcpu);
4810 		break;
4811 	case 8:
4812 		value = kvm_get_cr8(vcpu);
4813 		break;
4814 	default:
4815 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4816 		return 0;
4817 	}
4818 
4819 	return value;
4820 }
4821 
4822 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4823 {
4824 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4825 	int res = 0;
4826 
4827 	switch (cr) {
4828 	case 0:
4829 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4830 		break;
4831 	case 2:
4832 		vcpu->arch.cr2 = val;
4833 		break;
4834 	case 3:
4835 		res = kvm_set_cr3(vcpu, val);
4836 		break;
4837 	case 4:
4838 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4839 		break;
4840 	case 8:
4841 		res = kvm_set_cr8(vcpu, val);
4842 		break;
4843 	default:
4844 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4845 		res = -1;
4846 	}
4847 
4848 	return res;
4849 }
4850 
4851 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4852 {
4853 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4854 }
4855 
4856 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4857 {
4858 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4859 }
4860 
4861 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4862 {
4863 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4864 }
4865 
4866 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4867 {
4868 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4869 }
4870 
4871 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4872 {
4873 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4874 }
4875 
4876 static unsigned long emulator_get_cached_segment_base(
4877 	struct x86_emulate_ctxt *ctxt, int seg)
4878 {
4879 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4880 }
4881 
4882 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4883 				 struct desc_struct *desc, u32 *base3,
4884 				 int seg)
4885 {
4886 	struct kvm_segment var;
4887 
4888 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4889 	*selector = var.selector;
4890 
4891 	if (var.unusable) {
4892 		memset(desc, 0, sizeof(*desc));
4893 		return false;
4894 	}
4895 
4896 	if (var.g)
4897 		var.limit >>= 12;
4898 	set_desc_limit(desc, var.limit);
4899 	set_desc_base(desc, (unsigned long)var.base);
4900 #ifdef CONFIG_X86_64
4901 	if (base3)
4902 		*base3 = var.base >> 32;
4903 #endif
4904 	desc->type = var.type;
4905 	desc->s = var.s;
4906 	desc->dpl = var.dpl;
4907 	desc->p = var.present;
4908 	desc->avl = var.avl;
4909 	desc->l = var.l;
4910 	desc->d = var.db;
4911 	desc->g = var.g;
4912 
4913 	return true;
4914 }
4915 
4916 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4917 				 struct desc_struct *desc, u32 base3,
4918 				 int seg)
4919 {
4920 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4921 	struct kvm_segment var;
4922 
4923 	var.selector = selector;
4924 	var.base = get_desc_base(desc);
4925 #ifdef CONFIG_X86_64
4926 	var.base |= ((u64)base3) << 32;
4927 #endif
4928 	var.limit = get_desc_limit(desc);
4929 	if (desc->g)
4930 		var.limit = (var.limit << 12) | 0xfff;
4931 	var.type = desc->type;
4932 	var.dpl = desc->dpl;
4933 	var.db = desc->d;
4934 	var.s = desc->s;
4935 	var.l = desc->l;
4936 	var.g = desc->g;
4937 	var.avl = desc->avl;
4938 	var.present = desc->p;
4939 	var.unusable = !var.present;
4940 	var.padding = 0;
4941 
4942 	kvm_set_segment(vcpu, &var, seg);
4943 	return;
4944 }
4945 
4946 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4947 			    u32 msr_index, u64 *pdata)
4948 {
4949 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4950 }
4951 
4952 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4953 			    u32 msr_index, u64 data)
4954 {
4955 	struct msr_data msr;
4956 
4957 	msr.data = data;
4958 	msr.index = msr_index;
4959 	msr.host_initiated = false;
4960 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4961 }
4962 
4963 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4964 			      u32 pmc)
4965 {
4966 	return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4967 }
4968 
4969 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4970 			     u32 pmc, u64 *pdata)
4971 {
4972 	return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4973 }
4974 
4975 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4976 {
4977 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4978 }
4979 
4980 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4981 {
4982 	preempt_disable();
4983 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4984 	/*
4985 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4986 	 * so it may be clear at this point.
4987 	 */
4988 	clts();
4989 }
4990 
4991 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4992 {
4993 	preempt_enable();
4994 }
4995 
4996 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4997 			      struct x86_instruction_info *info,
4998 			      enum x86_intercept_stage stage)
4999 {
5000 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5001 }
5002 
5003 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5004 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5005 {
5006 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5007 }
5008 
5009 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5010 {
5011 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5012 }
5013 
5014 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5015 {
5016 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5017 }
5018 
5019 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5020 {
5021 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5022 }
5023 
5024 static const struct x86_emulate_ops emulate_ops = {
5025 	.read_gpr            = emulator_read_gpr,
5026 	.write_gpr           = emulator_write_gpr,
5027 	.read_std            = kvm_read_guest_virt_system,
5028 	.write_std           = kvm_write_guest_virt_system,
5029 	.fetch               = kvm_fetch_guest_virt,
5030 	.read_emulated       = emulator_read_emulated,
5031 	.write_emulated      = emulator_write_emulated,
5032 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5033 	.invlpg              = emulator_invlpg,
5034 	.pio_in_emulated     = emulator_pio_in_emulated,
5035 	.pio_out_emulated    = emulator_pio_out_emulated,
5036 	.get_segment         = emulator_get_segment,
5037 	.set_segment         = emulator_set_segment,
5038 	.get_cached_segment_base = emulator_get_cached_segment_base,
5039 	.get_gdt             = emulator_get_gdt,
5040 	.get_idt	     = emulator_get_idt,
5041 	.set_gdt             = emulator_set_gdt,
5042 	.set_idt	     = emulator_set_idt,
5043 	.get_cr              = emulator_get_cr,
5044 	.set_cr              = emulator_set_cr,
5045 	.cpl                 = emulator_get_cpl,
5046 	.get_dr              = emulator_get_dr,
5047 	.set_dr              = emulator_set_dr,
5048 	.set_msr             = emulator_set_msr,
5049 	.get_msr             = emulator_get_msr,
5050 	.check_pmc	     = emulator_check_pmc,
5051 	.read_pmc            = emulator_read_pmc,
5052 	.halt                = emulator_halt,
5053 	.wbinvd              = emulator_wbinvd,
5054 	.fix_hypercall       = emulator_fix_hypercall,
5055 	.get_fpu             = emulator_get_fpu,
5056 	.put_fpu             = emulator_put_fpu,
5057 	.intercept           = emulator_intercept,
5058 	.get_cpuid           = emulator_get_cpuid,
5059 	.set_nmi_mask        = emulator_set_nmi_mask,
5060 };
5061 
5062 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5063 {
5064 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5065 	/*
5066 	 * an sti; sti; sequence only disable interrupts for the first
5067 	 * instruction. So, if the last instruction, be it emulated or
5068 	 * not, left the system with the INT_STI flag enabled, it
5069 	 * means that the last instruction is an sti. We should not
5070 	 * leave the flag on in this case. The same goes for mov ss
5071 	 */
5072 	if (int_shadow & mask)
5073 		mask = 0;
5074 	if (unlikely(int_shadow || mask)) {
5075 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5076 		if (!mask)
5077 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5078 	}
5079 }
5080 
5081 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5082 {
5083 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5084 	if (ctxt->exception.vector == PF_VECTOR)
5085 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5086 
5087 	if (ctxt->exception.error_code_valid)
5088 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5089 				      ctxt->exception.error_code);
5090 	else
5091 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5092 	return false;
5093 }
5094 
5095 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5096 {
5097 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5098 	int cs_db, cs_l;
5099 
5100 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5101 
5102 	ctxt->eflags = kvm_get_rflags(vcpu);
5103 	ctxt->eip = kvm_rip_read(vcpu);
5104 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5105 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5106 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5107 		     cs_db				? X86EMUL_MODE_PROT32 :
5108 							  X86EMUL_MODE_PROT16;
5109 	ctxt->guest_mode = is_guest_mode(vcpu);
5110 
5111 	init_decode_cache(ctxt);
5112 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5113 }
5114 
5115 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5116 {
5117 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5118 	int ret;
5119 
5120 	init_emulate_ctxt(vcpu);
5121 
5122 	ctxt->op_bytes = 2;
5123 	ctxt->ad_bytes = 2;
5124 	ctxt->_eip = ctxt->eip + inc_eip;
5125 	ret = emulate_int_real(ctxt, irq);
5126 
5127 	if (ret != X86EMUL_CONTINUE)
5128 		return EMULATE_FAIL;
5129 
5130 	ctxt->eip = ctxt->_eip;
5131 	kvm_rip_write(vcpu, ctxt->eip);
5132 	kvm_set_rflags(vcpu, ctxt->eflags);
5133 
5134 	if (irq == NMI_VECTOR)
5135 		vcpu->arch.nmi_pending = 0;
5136 	else
5137 		vcpu->arch.interrupt.pending = false;
5138 
5139 	return EMULATE_DONE;
5140 }
5141 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5142 
5143 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5144 {
5145 	int r = EMULATE_DONE;
5146 
5147 	++vcpu->stat.insn_emulation_fail;
5148 	trace_kvm_emulate_insn_failed(vcpu);
5149 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5150 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5151 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5152 		vcpu->run->internal.ndata = 0;
5153 		r = EMULATE_FAIL;
5154 	}
5155 	kvm_queue_exception(vcpu, UD_VECTOR);
5156 
5157 	return r;
5158 }
5159 
5160 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5161 				  bool write_fault_to_shadow_pgtable,
5162 				  int emulation_type)
5163 {
5164 	gpa_t gpa = cr2;
5165 	pfn_t pfn;
5166 
5167 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5168 		return false;
5169 
5170 	if (!vcpu->arch.mmu.direct_map) {
5171 		/*
5172 		 * Write permission should be allowed since only
5173 		 * write access need to be emulated.
5174 		 */
5175 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5176 
5177 		/*
5178 		 * If the mapping is invalid in guest, let cpu retry
5179 		 * it to generate fault.
5180 		 */
5181 		if (gpa == UNMAPPED_GVA)
5182 			return true;
5183 	}
5184 
5185 	/*
5186 	 * Do not retry the unhandleable instruction if it faults on the
5187 	 * readonly host memory, otherwise it will goto a infinite loop:
5188 	 * retry instruction -> write #PF -> emulation fail -> retry
5189 	 * instruction -> ...
5190 	 */
5191 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5192 
5193 	/*
5194 	 * If the instruction failed on the error pfn, it can not be fixed,
5195 	 * report the error to userspace.
5196 	 */
5197 	if (is_error_noslot_pfn(pfn))
5198 		return false;
5199 
5200 	kvm_release_pfn_clean(pfn);
5201 
5202 	/* The instructions are well-emulated on direct mmu. */
5203 	if (vcpu->arch.mmu.direct_map) {
5204 		unsigned int indirect_shadow_pages;
5205 
5206 		spin_lock(&vcpu->kvm->mmu_lock);
5207 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5208 		spin_unlock(&vcpu->kvm->mmu_lock);
5209 
5210 		if (indirect_shadow_pages)
5211 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5212 
5213 		return true;
5214 	}
5215 
5216 	/*
5217 	 * if emulation was due to access to shadowed page table
5218 	 * and it failed try to unshadow page and re-enter the
5219 	 * guest to let CPU execute the instruction.
5220 	 */
5221 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5222 
5223 	/*
5224 	 * If the access faults on its page table, it can not
5225 	 * be fixed by unprotecting shadow page and it should
5226 	 * be reported to userspace.
5227 	 */
5228 	return !write_fault_to_shadow_pgtable;
5229 }
5230 
5231 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5232 			      unsigned long cr2,  int emulation_type)
5233 {
5234 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5235 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5236 
5237 	last_retry_eip = vcpu->arch.last_retry_eip;
5238 	last_retry_addr = vcpu->arch.last_retry_addr;
5239 
5240 	/*
5241 	 * If the emulation is caused by #PF and it is non-page_table
5242 	 * writing instruction, it means the VM-EXIT is caused by shadow
5243 	 * page protected, we can zap the shadow page and retry this
5244 	 * instruction directly.
5245 	 *
5246 	 * Note: if the guest uses a non-page-table modifying instruction
5247 	 * on the PDE that points to the instruction, then we will unmap
5248 	 * the instruction and go to an infinite loop. So, we cache the
5249 	 * last retried eip and the last fault address, if we meet the eip
5250 	 * and the address again, we can break out of the potential infinite
5251 	 * loop.
5252 	 */
5253 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5254 
5255 	if (!(emulation_type & EMULTYPE_RETRY))
5256 		return false;
5257 
5258 	if (x86_page_table_writing_insn(ctxt))
5259 		return false;
5260 
5261 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5262 		return false;
5263 
5264 	vcpu->arch.last_retry_eip = ctxt->eip;
5265 	vcpu->arch.last_retry_addr = cr2;
5266 
5267 	if (!vcpu->arch.mmu.direct_map)
5268 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5269 
5270 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5271 
5272 	return true;
5273 }
5274 
5275 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5276 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5277 
5278 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5279 				unsigned long *db)
5280 {
5281 	u32 dr6 = 0;
5282 	int i;
5283 	u32 enable, rwlen;
5284 
5285 	enable = dr7;
5286 	rwlen = dr7 >> 16;
5287 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5288 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5289 			dr6 |= (1 << i);
5290 	return dr6;
5291 }
5292 
5293 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5294 {
5295 	struct kvm_run *kvm_run = vcpu->run;
5296 
5297 	/*
5298 	 * rflags is the old, "raw" value of the flags.  The new value has
5299 	 * not been saved yet.
5300 	 *
5301 	 * This is correct even for TF set by the guest, because "the
5302 	 * processor will not generate this exception after the instruction
5303 	 * that sets the TF flag".
5304 	 */
5305 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5306 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5307 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5308 						  DR6_RTM;
5309 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5310 			kvm_run->debug.arch.exception = DB_VECTOR;
5311 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5312 			*r = EMULATE_USER_EXIT;
5313 		} else {
5314 			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5315 			/*
5316 			 * "Certain debug exceptions may clear bit 0-3.  The
5317 			 * remaining contents of the DR6 register are never
5318 			 * cleared by the processor".
5319 			 */
5320 			vcpu->arch.dr6 &= ~15;
5321 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5322 			kvm_queue_exception(vcpu, DB_VECTOR);
5323 		}
5324 	}
5325 }
5326 
5327 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5328 {
5329 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5330 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5331 		struct kvm_run *kvm_run = vcpu->run;
5332 		unsigned long eip = kvm_get_linear_rip(vcpu);
5333 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5334 					   vcpu->arch.guest_debug_dr7,
5335 					   vcpu->arch.eff_db);
5336 
5337 		if (dr6 != 0) {
5338 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5339 			kvm_run->debug.arch.pc = eip;
5340 			kvm_run->debug.arch.exception = DB_VECTOR;
5341 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5342 			*r = EMULATE_USER_EXIT;
5343 			return true;
5344 		}
5345 	}
5346 
5347 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5348 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5349 		unsigned long eip = kvm_get_linear_rip(vcpu);
5350 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5351 					   vcpu->arch.dr7,
5352 					   vcpu->arch.db);
5353 
5354 		if (dr6 != 0) {
5355 			vcpu->arch.dr6 &= ~15;
5356 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5357 			kvm_queue_exception(vcpu, DB_VECTOR);
5358 			*r = EMULATE_DONE;
5359 			return true;
5360 		}
5361 	}
5362 
5363 	return false;
5364 }
5365 
5366 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5367 			    unsigned long cr2,
5368 			    int emulation_type,
5369 			    void *insn,
5370 			    int insn_len)
5371 {
5372 	int r;
5373 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5374 	bool writeback = true;
5375 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5376 
5377 	/*
5378 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5379 	 * never reused.
5380 	 */
5381 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5382 	kvm_clear_exception_queue(vcpu);
5383 
5384 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5385 		init_emulate_ctxt(vcpu);
5386 
5387 		/*
5388 		 * We will reenter on the same instruction since
5389 		 * we do not set complete_userspace_io.  This does not
5390 		 * handle watchpoints yet, those would be handled in
5391 		 * the emulate_ops.
5392 		 */
5393 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5394 			return r;
5395 
5396 		ctxt->interruptibility = 0;
5397 		ctxt->have_exception = false;
5398 		ctxt->exception.vector = -1;
5399 		ctxt->perm_ok = false;
5400 
5401 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5402 
5403 		r = x86_decode_insn(ctxt, insn, insn_len);
5404 
5405 		trace_kvm_emulate_insn_start(vcpu);
5406 		++vcpu->stat.insn_emulation;
5407 		if (r != EMULATION_OK)  {
5408 			if (emulation_type & EMULTYPE_TRAP_UD)
5409 				return EMULATE_FAIL;
5410 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5411 						emulation_type))
5412 				return EMULATE_DONE;
5413 			if (emulation_type & EMULTYPE_SKIP)
5414 				return EMULATE_FAIL;
5415 			return handle_emulation_failure(vcpu);
5416 		}
5417 	}
5418 
5419 	if (emulation_type & EMULTYPE_SKIP) {
5420 		kvm_rip_write(vcpu, ctxt->_eip);
5421 		if (ctxt->eflags & X86_EFLAGS_RF)
5422 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5423 		return EMULATE_DONE;
5424 	}
5425 
5426 	if (retry_instruction(ctxt, cr2, emulation_type))
5427 		return EMULATE_DONE;
5428 
5429 	/* this is needed for vmware backdoor interface to work since it
5430 	   changes registers values  during IO operation */
5431 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5432 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5433 		emulator_invalidate_register_cache(ctxt);
5434 	}
5435 
5436 restart:
5437 	r = x86_emulate_insn(ctxt);
5438 
5439 	if (r == EMULATION_INTERCEPTED)
5440 		return EMULATE_DONE;
5441 
5442 	if (r == EMULATION_FAILED) {
5443 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5444 					emulation_type))
5445 			return EMULATE_DONE;
5446 
5447 		return handle_emulation_failure(vcpu);
5448 	}
5449 
5450 	if (ctxt->have_exception) {
5451 		r = EMULATE_DONE;
5452 		if (inject_emulated_exception(vcpu))
5453 			return r;
5454 	} else if (vcpu->arch.pio.count) {
5455 		if (!vcpu->arch.pio.in) {
5456 			/* FIXME: return into emulator if single-stepping.  */
5457 			vcpu->arch.pio.count = 0;
5458 		} else {
5459 			writeback = false;
5460 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5461 		}
5462 		r = EMULATE_USER_EXIT;
5463 	} else if (vcpu->mmio_needed) {
5464 		if (!vcpu->mmio_is_write)
5465 			writeback = false;
5466 		r = EMULATE_USER_EXIT;
5467 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5468 	} else if (r == EMULATION_RESTART)
5469 		goto restart;
5470 	else
5471 		r = EMULATE_DONE;
5472 
5473 	if (writeback) {
5474 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5475 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5476 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5477 		kvm_rip_write(vcpu, ctxt->eip);
5478 		if (r == EMULATE_DONE)
5479 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5480 		if (!ctxt->have_exception ||
5481 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5482 			__kvm_set_rflags(vcpu, ctxt->eflags);
5483 
5484 		/*
5485 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5486 		 * do nothing, and it will be requested again as soon as
5487 		 * the shadow expires.  But we still need to check here,
5488 		 * because POPF has no interrupt shadow.
5489 		 */
5490 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5491 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5492 	} else
5493 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5494 
5495 	return r;
5496 }
5497 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5498 
5499 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5500 {
5501 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5502 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5503 					    size, port, &val, 1);
5504 	/* do not return to emulator after return from userspace */
5505 	vcpu->arch.pio.count = 0;
5506 	return ret;
5507 }
5508 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5509 
5510 static void tsc_bad(void *info)
5511 {
5512 	__this_cpu_write(cpu_tsc_khz, 0);
5513 }
5514 
5515 static void tsc_khz_changed(void *data)
5516 {
5517 	struct cpufreq_freqs *freq = data;
5518 	unsigned long khz = 0;
5519 
5520 	if (data)
5521 		khz = freq->new;
5522 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5523 		khz = cpufreq_quick_get(raw_smp_processor_id());
5524 	if (!khz)
5525 		khz = tsc_khz;
5526 	__this_cpu_write(cpu_tsc_khz, khz);
5527 }
5528 
5529 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5530 				     void *data)
5531 {
5532 	struct cpufreq_freqs *freq = data;
5533 	struct kvm *kvm;
5534 	struct kvm_vcpu *vcpu;
5535 	int i, send_ipi = 0;
5536 
5537 	/*
5538 	 * We allow guests to temporarily run on slowing clocks,
5539 	 * provided we notify them after, or to run on accelerating
5540 	 * clocks, provided we notify them before.  Thus time never
5541 	 * goes backwards.
5542 	 *
5543 	 * However, we have a problem.  We can't atomically update
5544 	 * the frequency of a given CPU from this function; it is
5545 	 * merely a notifier, which can be called from any CPU.
5546 	 * Changing the TSC frequency at arbitrary points in time
5547 	 * requires a recomputation of local variables related to
5548 	 * the TSC for each VCPU.  We must flag these local variables
5549 	 * to be updated and be sure the update takes place with the
5550 	 * new frequency before any guests proceed.
5551 	 *
5552 	 * Unfortunately, the combination of hotplug CPU and frequency
5553 	 * change creates an intractable locking scenario; the order
5554 	 * of when these callouts happen is undefined with respect to
5555 	 * CPU hotplug, and they can race with each other.  As such,
5556 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5557 	 * undefined; you can actually have a CPU frequency change take
5558 	 * place in between the computation of X and the setting of the
5559 	 * variable.  To protect against this problem, all updates of
5560 	 * the per_cpu tsc_khz variable are done in an interrupt
5561 	 * protected IPI, and all callers wishing to update the value
5562 	 * must wait for a synchronous IPI to complete (which is trivial
5563 	 * if the caller is on the CPU already).  This establishes the
5564 	 * necessary total order on variable updates.
5565 	 *
5566 	 * Note that because a guest time update may take place
5567 	 * anytime after the setting of the VCPU's request bit, the
5568 	 * correct TSC value must be set before the request.  However,
5569 	 * to ensure the update actually makes it to any guest which
5570 	 * starts running in hardware virtualization between the set
5571 	 * and the acquisition of the spinlock, we must also ping the
5572 	 * CPU after setting the request bit.
5573 	 *
5574 	 */
5575 
5576 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5577 		return 0;
5578 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5579 		return 0;
5580 
5581 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5582 
5583 	spin_lock(&kvm_lock);
5584 	list_for_each_entry(kvm, &vm_list, vm_list) {
5585 		kvm_for_each_vcpu(i, vcpu, kvm) {
5586 			if (vcpu->cpu != freq->cpu)
5587 				continue;
5588 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5589 			if (vcpu->cpu != smp_processor_id())
5590 				send_ipi = 1;
5591 		}
5592 	}
5593 	spin_unlock(&kvm_lock);
5594 
5595 	if (freq->old < freq->new && send_ipi) {
5596 		/*
5597 		 * We upscale the frequency.  Must make the guest
5598 		 * doesn't see old kvmclock values while running with
5599 		 * the new frequency, otherwise we risk the guest sees
5600 		 * time go backwards.
5601 		 *
5602 		 * In case we update the frequency for another cpu
5603 		 * (which might be in guest context) send an interrupt
5604 		 * to kick the cpu out of guest context.  Next time
5605 		 * guest context is entered kvmclock will be updated,
5606 		 * so the guest will not see stale values.
5607 		 */
5608 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5609 	}
5610 	return 0;
5611 }
5612 
5613 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5614 	.notifier_call  = kvmclock_cpufreq_notifier
5615 };
5616 
5617 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5618 					unsigned long action, void *hcpu)
5619 {
5620 	unsigned int cpu = (unsigned long)hcpu;
5621 
5622 	switch (action) {
5623 		case CPU_ONLINE:
5624 		case CPU_DOWN_FAILED:
5625 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5626 			break;
5627 		case CPU_DOWN_PREPARE:
5628 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5629 			break;
5630 	}
5631 	return NOTIFY_OK;
5632 }
5633 
5634 static struct notifier_block kvmclock_cpu_notifier_block = {
5635 	.notifier_call  = kvmclock_cpu_notifier,
5636 	.priority = -INT_MAX
5637 };
5638 
5639 static void kvm_timer_init(void)
5640 {
5641 	int cpu;
5642 
5643 	max_tsc_khz = tsc_khz;
5644 
5645 	cpu_notifier_register_begin();
5646 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5647 #ifdef CONFIG_CPU_FREQ
5648 		struct cpufreq_policy policy;
5649 		memset(&policy, 0, sizeof(policy));
5650 		cpu = get_cpu();
5651 		cpufreq_get_policy(&policy, cpu);
5652 		if (policy.cpuinfo.max_freq)
5653 			max_tsc_khz = policy.cpuinfo.max_freq;
5654 		put_cpu();
5655 #endif
5656 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5657 					  CPUFREQ_TRANSITION_NOTIFIER);
5658 	}
5659 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5660 	for_each_online_cpu(cpu)
5661 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5662 
5663 	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5664 	cpu_notifier_register_done();
5665 
5666 }
5667 
5668 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5669 
5670 int kvm_is_in_guest(void)
5671 {
5672 	return __this_cpu_read(current_vcpu) != NULL;
5673 }
5674 
5675 static int kvm_is_user_mode(void)
5676 {
5677 	int user_mode = 3;
5678 
5679 	if (__this_cpu_read(current_vcpu))
5680 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5681 
5682 	return user_mode != 0;
5683 }
5684 
5685 static unsigned long kvm_get_guest_ip(void)
5686 {
5687 	unsigned long ip = 0;
5688 
5689 	if (__this_cpu_read(current_vcpu))
5690 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5691 
5692 	return ip;
5693 }
5694 
5695 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5696 	.is_in_guest		= kvm_is_in_guest,
5697 	.is_user_mode		= kvm_is_user_mode,
5698 	.get_guest_ip		= kvm_get_guest_ip,
5699 };
5700 
5701 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5702 {
5703 	__this_cpu_write(current_vcpu, vcpu);
5704 }
5705 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5706 
5707 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5708 {
5709 	__this_cpu_write(current_vcpu, NULL);
5710 }
5711 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5712 
5713 static void kvm_set_mmio_spte_mask(void)
5714 {
5715 	u64 mask;
5716 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5717 
5718 	/*
5719 	 * Set the reserved bits and the present bit of an paging-structure
5720 	 * entry to generate page fault with PFER.RSV = 1.
5721 	 */
5722 	 /* Mask the reserved physical address bits. */
5723 	mask = rsvd_bits(maxphyaddr, 51);
5724 
5725 	/* Bit 62 is always reserved for 32bit host. */
5726 	mask |= 0x3ull << 62;
5727 
5728 	/* Set the present bit. */
5729 	mask |= 1ull;
5730 
5731 #ifdef CONFIG_X86_64
5732 	/*
5733 	 * If reserved bit is not supported, clear the present bit to disable
5734 	 * mmio page fault.
5735 	 */
5736 	if (maxphyaddr == 52)
5737 		mask &= ~1ull;
5738 #endif
5739 
5740 	kvm_mmu_set_mmio_spte_mask(mask);
5741 }
5742 
5743 #ifdef CONFIG_X86_64
5744 static void pvclock_gtod_update_fn(struct work_struct *work)
5745 {
5746 	struct kvm *kvm;
5747 
5748 	struct kvm_vcpu *vcpu;
5749 	int i;
5750 
5751 	spin_lock(&kvm_lock);
5752 	list_for_each_entry(kvm, &vm_list, vm_list)
5753 		kvm_for_each_vcpu(i, vcpu, kvm)
5754 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5755 	atomic_set(&kvm_guest_has_master_clock, 0);
5756 	spin_unlock(&kvm_lock);
5757 }
5758 
5759 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5760 
5761 /*
5762  * Notification about pvclock gtod data update.
5763  */
5764 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5765 			       void *priv)
5766 {
5767 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5768 	struct timekeeper *tk = priv;
5769 
5770 	update_pvclock_gtod(tk);
5771 
5772 	/* disable master clock if host does not trust, or does not
5773 	 * use, TSC clocksource
5774 	 */
5775 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5776 	    atomic_read(&kvm_guest_has_master_clock) != 0)
5777 		queue_work(system_long_wq, &pvclock_gtod_work);
5778 
5779 	return 0;
5780 }
5781 
5782 static struct notifier_block pvclock_gtod_notifier = {
5783 	.notifier_call = pvclock_gtod_notify,
5784 };
5785 #endif
5786 
5787 int kvm_arch_init(void *opaque)
5788 {
5789 	int r;
5790 	struct kvm_x86_ops *ops = opaque;
5791 
5792 	if (kvm_x86_ops) {
5793 		printk(KERN_ERR "kvm: already loaded the other module\n");
5794 		r = -EEXIST;
5795 		goto out;
5796 	}
5797 
5798 	if (!ops->cpu_has_kvm_support()) {
5799 		printk(KERN_ERR "kvm: no hardware support\n");
5800 		r = -EOPNOTSUPP;
5801 		goto out;
5802 	}
5803 	if (ops->disabled_by_bios()) {
5804 		printk(KERN_ERR "kvm: disabled by bios\n");
5805 		r = -EOPNOTSUPP;
5806 		goto out;
5807 	}
5808 
5809 	r = -ENOMEM;
5810 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5811 	if (!shared_msrs) {
5812 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5813 		goto out;
5814 	}
5815 
5816 	r = kvm_mmu_module_init();
5817 	if (r)
5818 		goto out_free_percpu;
5819 
5820 	kvm_set_mmio_spte_mask();
5821 
5822 	kvm_x86_ops = ops;
5823 
5824 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5825 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5826 
5827 	kvm_timer_init();
5828 
5829 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5830 
5831 	if (cpu_has_xsave)
5832 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5833 
5834 	kvm_lapic_init();
5835 #ifdef CONFIG_X86_64
5836 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5837 #endif
5838 
5839 	return 0;
5840 
5841 out_free_percpu:
5842 	free_percpu(shared_msrs);
5843 out:
5844 	return r;
5845 }
5846 
5847 void kvm_arch_exit(void)
5848 {
5849 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5850 
5851 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5852 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5853 					    CPUFREQ_TRANSITION_NOTIFIER);
5854 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5855 #ifdef CONFIG_X86_64
5856 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5857 #endif
5858 	kvm_x86_ops = NULL;
5859 	kvm_mmu_module_exit();
5860 	free_percpu(shared_msrs);
5861 }
5862 
5863 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5864 {
5865 	++vcpu->stat.halt_exits;
5866 	if (irqchip_in_kernel(vcpu->kvm)) {
5867 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5868 		return 1;
5869 	} else {
5870 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5871 		return 0;
5872 	}
5873 }
5874 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5875 
5876 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5877 {
5878 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5879 	return kvm_vcpu_halt(vcpu);
5880 }
5881 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5882 
5883 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5884 {
5885 	u64 param, ingpa, outgpa, ret;
5886 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5887 	bool fast, longmode;
5888 
5889 	/*
5890 	 * hypercall generates UD from non zero cpl and real mode
5891 	 * per HYPER-V spec
5892 	 */
5893 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5894 		kvm_queue_exception(vcpu, UD_VECTOR);
5895 		return 0;
5896 	}
5897 
5898 	longmode = is_64_bit_mode(vcpu);
5899 
5900 	if (!longmode) {
5901 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5902 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5903 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5904 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5905 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5906 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5907 	}
5908 #ifdef CONFIG_X86_64
5909 	else {
5910 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5911 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5912 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5913 	}
5914 #endif
5915 
5916 	code = param & 0xffff;
5917 	fast = (param >> 16) & 0x1;
5918 	rep_cnt = (param >> 32) & 0xfff;
5919 	rep_idx = (param >> 48) & 0xfff;
5920 
5921 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5922 
5923 	switch (code) {
5924 	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5925 		kvm_vcpu_on_spin(vcpu);
5926 		break;
5927 	default:
5928 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
5929 		break;
5930 	}
5931 
5932 	ret = res | (((u64)rep_done & 0xfff) << 32);
5933 	if (longmode) {
5934 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5935 	} else {
5936 		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5937 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5938 	}
5939 
5940 	return 1;
5941 }
5942 
5943 /*
5944  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5945  *
5946  * @apicid - apicid of vcpu to be kicked.
5947  */
5948 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5949 {
5950 	struct kvm_lapic_irq lapic_irq;
5951 
5952 	lapic_irq.shorthand = 0;
5953 	lapic_irq.dest_mode = 0;
5954 	lapic_irq.dest_id = apicid;
5955 
5956 	lapic_irq.delivery_mode = APIC_DM_REMRD;
5957 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5958 }
5959 
5960 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5961 {
5962 	unsigned long nr, a0, a1, a2, a3, ret;
5963 	int op_64_bit, r = 1;
5964 
5965 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5966 
5967 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5968 		return kvm_hv_hypercall(vcpu);
5969 
5970 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5971 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5972 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5973 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5974 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5975 
5976 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5977 
5978 	op_64_bit = is_64_bit_mode(vcpu);
5979 	if (!op_64_bit) {
5980 		nr &= 0xFFFFFFFF;
5981 		a0 &= 0xFFFFFFFF;
5982 		a1 &= 0xFFFFFFFF;
5983 		a2 &= 0xFFFFFFFF;
5984 		a3 &= 0xFFFFFFFF;
5985 	}
5986 
5987 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5988 		ret = -KVM_EPERM;
5989 		goto out;
5990 	}
5991 
5992 	switch (nr) {
5993 	case KVM_HC_VAPIC_POLL_IRQ:
5994 		ret = 0;
5995 		break;
5996 	case KVM_HC_KICK_CPU:
5997 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5998 		ret = 0;
5999 		break;
6000 	default:
6001 		ret = -KVM_ENOSYS;
6002 		break;
6003 	}
6004 out:
6005 	if (!op_64_bit)
6006 		ret = (u32)ret;
6007 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6008 	++vcpu->stat.hypercalls;
6009 	return r;
6010 }
6011 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6012 
6013 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6014 {
6015 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6016 	char instruction[3];
6017 	unsigned long rip = kvm_rip_read(vcpu);
6018 
6019 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6020 
6021 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6022 }
6023 
6024 /*
6025  * Check if userspace requested an interrupt window, and that the
6026  * interrupt window is open.
6027  *
6028  * No need to exit to userspace if we already have an interrupt queued.
6029  */
6030 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6031 {
6032 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
6033 		vcpu->run->request_interrupt_window &&
6034 		kvm_arch_interrupt_allowed(vcpu));
6035 }
6036 
6037 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6038 {
6039 	struct kvm_run *kvm_run = vcpu->run;
6040 
6041 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6042 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6043 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6044 	if (irqchip_in_kernel(vcpu->kvm))
6045 		kvm_run->ready_for_interrupt_injection = 1;
6046 	else
6047 		kvm_run->ready_for_interrupt_injection =
6048 			kvm_arch_interrupt_allowed(vcpu) &&
6049 			!kvm_cpu_has_interrupt(vcpu) &&
6050 			!kvm_event_needs_reinjection(vcpu);
6051 }
6052 
6053 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6054 {
6055 	int max_irr, tpr;
6056 
6057 	if (!kvm_x86_ops->update_cr8_intercept)
6058 		return;
6059 
6060 	if (!vcpu->arch.apic)
6061 		return;
6062 
6063 	if (!vcpu->arch.apic->vapic_addr)
6064 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6065 	else
6066 		max_irr = -1;
6067 
6068 	if (max_irr != -1)
6069 		max_irr >>= 4;
6070 
6071 	tpr = kvm_lapic_get_cr8(vcpu);
6072 
6073 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6074 }
6075 
6076 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6077 {
6078 	int r;
6079 
6080 	/* try to reinject previous events if any */
6081 	if (vcpu->arch.exception.pending) {
6082 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6083 					vcpu->arch.exception.has_error_code,
6084 					vcpu->arch.exception.error_code);
6085 
6086 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6087 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6088 					     X86_EFLAGS_RF);
6089 
6090 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6091 		    (vcpu->arch.dr7 & DR7_GD)) {
6092 			vcpu->arch.dr7 &= ~DR7_GD;
6093 			kvm_update_dr7(vcpu);
6094 		}
6095 
6096 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6097 					  vcpu->arch.exception.has_error_code,
6098 					  vcpu->arch.exception.error_code,
6099 					  vcpu->arch.exception.reinject);
6100 		return 0;
6101 	}
6102 
6103 	if (vcpu->arch.nmi_injected) {
6104 		kvm_x86_ops->set_nmi(vcpu);
6105 		return 0;
6106 	}
6107 
6108 	if (vcpu->arch.interrupt.pending) {
6109 		kvm_x86_ops->set_irq(vcpu);
6110 		return 0;
6111 	}
6112 
6113 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6114 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6115 		if (r != 0)
6116 			return r;
6117 	}
6118 
6119 	/* try to inject new event if pending */
6120 	if (vcpu->arch.nmi_pending) {
6121 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
6122 			--vcpu->arch.nmi_pending;
6123 			vcpu->arch.nmi_injected = true;
6124 			kvm_x86_ops->set_nmi(vcpu);
6125 		}
6126 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6127 		/*
6128 		 * Because interrupts can be injected asynchronously, we are
6129 		 * calling check_nested_events again here to avoid a race condition.
6130 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6131 		 * proposal and current concerns.  Perhaps we should be setting
6132 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6133 		 */
6134 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6135 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6136 			if (r != 0)
6137 				return r;
6138 		}
6139 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6140 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6141 					    false);
6142 			kvm_x86_ops->set_irq(vcpu);
6143 		}
6144 	}
6145 	return 0;
6146 }
6147 
6148 static void process_nmi(struct kvm_vcpu *vcpu)
6149 {
6150 	unsigned limit = 2;
6151 
6152 	/*
6153 	 * x86 is limited to one NMI running, and one NMI pending after it.
6154 	 * If an NMI is already in progress, limit further NMIs to just one.
6155 	 * Otherwise, allow two (and we'll inject the first one immediately).
6156 	 */
6157 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6158 		limit = 1;
6159 
6160 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6161 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6162 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6163 }
6164 
6165 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6166 {
6167 	u64 eoi_exit_bitmap[4];
6168 	u32 tmr[8];
6169 
6170 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6171 		return;
6172 
6173 	memset(eoi_exit_bitmap, 0, 32);
6174 	memset(tmr, 0, 32);
6175 
6176 	kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6177 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6178 	kvm_apic_update_tmr(vcpu, tmr);
6179 }
6180 
6181 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6182 {
6183 	++vcpu->stat.tlb_flush;
6184 	kvm_x86_ops->tlb_flush(vcpu);
6185 }
6186 
6187 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6188 {
6189 	struct page *page = NULL;
6190 
6191 	if (!irqchip_in_kernel(vcpu->kvm))
6192 		return;
6193 
6194 	if (!kvm_x86_ops->set_apic_access_page_addr)
6195 		return;
6196 
6197 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6198 	if (is_error_page(page))
6199 		return;
6200 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6201 
6202 	/*
6203 	 * Do not pin apic access page in memory, the MMU notifier
6204 	 * will call us again if it is migrated or swapped out.
6205 	 */
6206 	put_page(page);
6207 }
6208 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6209 
6210 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6211 					   unsigned long address)
6212 {
6213 	/*
6214 	 * The physical address of apic access page is stored in the VMCS.
6215 	 * Update it when it becomes invalid.
6216 	 */
6217 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6218 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6219 }
6220 
6221 /*
6222  * Returns 1 to let vcpu_run() continue the guest execution loop without
6223  * exiting to the userspace.  Otherwise, the value will be returned to the
6224  * userspace.
6225  */
6226 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6227 {
6228 	int r;
6229 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6230 		vcpu->run->request_interrupt_window;
6231 	bool req_immediate_exit = false;
6232 
6233 	if (vcpu->requests) {
6234 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6235 			kvm_mmu_unload(vcpu);
6236 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6237 			__kvm_migrate_timers(vcpu);
6238 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6239 			kvm_gen_update_masterclock(vcpu->kvm);
6240 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6241 			kvm_gen_kvmclock_update(vcpu);
6242 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6243 			r = kvm_guest_time_update(vcpu);
6244 			if (unlikely(r))
6245 				goto out;
6246 		}
6247 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6248 			kvm_mmu_sync_roots(vcpu);
6249 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6250 			kvm_vcpu_flush_tlb(vcpu);
6251 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6252 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6253 			r = 0;
6254 			goto out;
6255 		}
6256 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6257 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6258 			r = 0;
6259 			goto out;
6260 		}
6261 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6262 			vcpu->fpu_active = 0;
6263 			kvm_x86_ops->fpu_deactivate(vcpu);
6264 		}
6265 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6266 			/* Page is swapped out. Do synthetic halt */
6267 			vcpu->arch.apf.halted = true;
6268 			r = 1;
6269 			goto out;
6270 		}
6271 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6272 			record_steal_time(vcpu);
6273 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6274 			process_nmi(vcpu);
6275 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6276 			kvm_handle_pmu_event(vcpu);
6277 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6278 			kvm_deliver_pmi(vcpu);
6279 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6280 			vcpu_scan_ioapic(vcpu);
6281 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6282 			kvm_vcpu_reload_apic_access_page(vcpu);
6283 	}
6284 
6285 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6286 		kvm_apic_accept_events(vcpu);
6287 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6288 			r = 1;
6289 			goto out;
6290 		}
6291 
6292 		if (inject_pending_event(vcpu, req_int_win) != 0)
6293 			req_immediate_exit = true;
6294 		/* enable NMI/IRQ window open exits if needed */
6295 		else if (vcpu->arch.nmi_pending)
6296 			kvm_x86_ops->enable_nmi_window(vcpu);
6297 		else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6298 			kvm_x86_ops->enable_irq_window(vcpu);
6299 
6300 		if (kvm_lapic_enabled(vcpu)) {
6301 			/*
6302 			 * Update architecture specific hints for APIC
6303 			 * virtual interrupt delivery.
6304 			 */
6305 			if (kvm_x86_ops->hwapic_irr_update)
6306 				kvm_x86_ops->hwapic_irr_update(vcpu,
6307 					kvm_lapic_find_highest_irr(vcpu));
6308 			update_cr8_intercept(vcpu);
6309 			kvm_lapic_sync_to_vapic(vcpu);
6310 		}
6311 	}
6312 
6313 	r = kvm_mmu_reload(vcpu);
6314 	if (unlikely(r)) {
6315 		goto cancel_injection;
6316 	}
6317 
6318 	preempt_disable();
6319 
6320 	kvm_x86_ops->prepare_guest_switch(vcpu);
6321 	if (vcpu->fpu_active)
6322 		kvm_load_guest_fpu(vcpu);
6323 	kvm_load_guest_xcr0(vcpu);
6324 
6325 	vcpu->mode = IN_GUEST_MODE;
6326 
6327 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6328 
6329 	/* We should set ->mode before check ->requests,
6330 	 * see the comment in make_all_cpus_request.
6331 	 */
6332 	smp_mb__after_srcu_read_unlock();
6333 
6334 	local_irq_disable();
6335 
6336 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6337 	    || need_resched() || signal_pending(current)) {
6338 		vcpu->mode = OUTSIDE_GUEST_MODE;
6339 		smp_wmb();
6340 		local_irq_enable();
6341 		preempt_enable();
6342 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6343 		r = 1;
6344 		goto cancel_injection;
6345 	}
6346 
6347 	if (req_immediate_exit)
6348 		smp_send_reschedule(vcpu->cpu);
6349 
6350 	kvm_guest_enter();
6351 
6352 	if (unlikely(vcpu->arch.switch_db_regs)) {
6353 		set_debugreg(0, 7);
6354 		set_debugreg(vcpu->arch.eff_db[0], 0);
6355 		set_debugreg(vcpu->arch.eff_db[1], 1);
6356 		set_debugreg(vcpu->arch.eff_db[2], 2);
6357 		set_debugreg(vcpu->arch.eff_db[3], 3);
6358 		set_debugreg(vcpu->arch.dr6, 6);
6359 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6360 	}
6361 
6362 	trace_kvm_entry(vcpu->vcpu_id);
6363 	wait_lapic_expire(vcpu);
6364 	kvm_x86_ops->run(vcpu);
6365 
6366 	/*
6367 	 * Do this here before restoring debug registers on the host.  And
6368 	 * since we do this before handling the vmexit, a DR access vmexit
6369 	 * can (a) read the correct value of the debug registers, (b) set
6370 	 * KVM_DEBUGREG_WONT_EXIT again.
6371 	 */
6372 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6373 		int i;
6374 
6375 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6376 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6377 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6378 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6379 	}
6380 
6381 	/*
6382 	 * If the guest has used debug registers, at least dr7
6383 	 * will be disabled while returning to the host.
6384 	 * If we don't have active breakpoints in the host, we don't
6385 	 * care about the messed up debug address registers. But if
6386 	 * we have some of them active, restore the old state.
6387 	 */
6388 	if (hw_breakpoint_active())
6389 		hw_breakpoint_restore();
6390 
6391 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6392 							   native_read_tsc());
6393 
6394 	vcpu->mode = OUTSIDE_GUEST_MODE;
6395 	smp_wmb();
6396 
6397 	/* Interrupt is enabled by handle_external_intr() */
6398 	kvm_x86_ops->handle_external_intr(vcpu);
6399 
6400 	++vcpu->stat.exits;
6401 
6402 	/*
6403 	 * We must have an instruction between local_irq_enable() and
6404 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6405 	 * the interrupt shadow.  The stat.exits increment will do nicely.
6406 	 * But we need to prevent reordering, hence this barrier():
6407 	 */
6408 	barrier();
6409 
6410 	kvm_guest_exit();
6411 
6412 	preempt_enable();
6413 
6414 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6415 
6416 	/*
6417 	 * Profile KVM exit RIPs:
6418 	 */
6419 	if (unlikely(prof_on == KVM_PROFILING)) {
6420 		unsigned long rip = kvm_rip_read(vcpu);
6421 		profile_hit(KVM_PROFILING, (void *)rip);
6422 	}
6423 
6424 	if (unlikely(vcpu->arch.tsc_always_catchup))
6425 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6426 
6427 	if (vcpu->arch.apic_attention)
6428 		kvm_lapic_sync_from_vapic(vcpu);
6429 
6430 	r = kvm_x86_ops->handle_exit(vcpu);
6431 	return r;
6432 
6433 cancel_injection:
6434 	kvm_x86_ops->cancel_injection(vcpu);
6435 	if (unlikely(vcpu->arch.apic_attention))
6436 		kvm_lapic_sync_from_vapic(vcpu);
6437 out:
6438 	return r;
6439 }
6440 
6441 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6442 {
6443 	if (!kvm_arch_vcpu_runnable(vcpu)) {
6444 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6445 		kvm_vcpu_block(vcpu);
6446 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6447 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6448 			return 1;
6449 	}
6450 
6451 	kvm_apic_accept_events(vcpu);
6452 	switch(vcpu->arch.mp_state) {
6453 	case KVM_MP_STATE_HALTED:
6454 		vcpu->arch.pv.pv_unhalted = false;
6455 		vcpu->arch.mp_state =
6456 			KVM_MP_STATE_RUNNABLE;
6457 	case KVM_MP_STATE_RUNNABLE:
6458 		vcpu->arch.apf.halted = false;
6459 		break;
6460 	case KVM_MP_STATE_INIT_RECEIVED:
6461 		break;
6462 	default:
6463 		return -EINTR;
6464 		break;
6465 	}
6466 	return 1;
6467 }
6468 
6469 static int vcpu_run(struct kvm_vcpu *vcpu)
6470 {
6471 	int r;
6472 	struct kvm *kvm = vcpu->kvm;
6473 
6474 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6475 
6476 	for (;;) {
6477 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6478 		    !vcpu->arch.apf.halted)
6479 			r = vcpu_enter_guest(vcpu);
6480 		else
6481 			r = vcpu_block(kvm, vcpu);
6482 		if (r <= 0)
6483 			break;
6484 
6485 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6486 		if (kvm_cpu_has_pending_timer(vcpu))
6487 			kvm_inject_pending_timer_irqs(vcpu);
6488 
6489 		if (dm_request_for_irq_injection(vcpu)) {
6490 			r = -EINTR;
6491 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6492 			++vcpu->stat.request_irq_exits;
6493 			break;
6494 		}
6495 
6496 		kvm_check_async_pf_completion(vcpu);
6497 
6498 		if (signal_pending(current)) {
6499 			r = -EINTR;
6500 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6501 			++vcpu->stat.signal_exits;
6502 			break;
6503 		}
6504 		if (need_resched()) {
6505 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6506 			cond_resched();
6507 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6508 		}
6509 	}
6510 
6511 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6512 
6513 	return r;
6514 }
6515 
6516 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6517 {
6518 	int r;
6519 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6520 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6521 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6522 	if (r != EMULATE_DONE)
6523 		return 0;
6524 	return 1;
6525 }
6526 
6527 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6528 {
6529 	BUG_ON(!vcpu->arch.pio.count);
6530 
6531 	return complete_emulated_io(vcpu);
6532 }
6533 
6534 /*
6535  * Implements the following, as a state machine:
6536  *
6537  * read:
6538  *   for each fragment
6539  *     for each mmio piece in the fragment
6540  *       write gpa, len
6541  *       exit
6542  *       copy data
6543  *   execute insn
6544  *
6545  * write:
6546  *   for each fragment
6547  *     for each mmio piece in the fragment
6548  *       write gpa, len
6549  *       copy data
6550  *       exit
6551  */
6552 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6553 {
6554 	struct kvm_run *run = vcpu->run;
6555 	struct kvm_mmio_fragment *frag;
6556 	unsigned len;
6557 
6558 	BUG_ON(!vcpu->mmio_needed);
6559 
6560 	/* Complete previous fragment */
6561 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6562 	len = min(8u, frag->len);
6563 	if (!vcpu->mmio_is_write)
6564 		memcpy(frag->data, run->mmio.data, len);
6565 
6566 	if (frag->len <= 8) {
6567 		/* Switch to the next fragment. */
6568 		frag++;
6569 		vcpu->mmio_cur_fragment++;
6570 	} else {
6571 		/* Go forward to the next mmio piece. */
6572 		frag->data += len;
6573 		frag->gpa += len;
6574 		frag->len -= len;
6575 	}
6576 
6577 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6578 		vcpu->mmio_needed = 0;
6579 
6580 		/* FIXME: return into emulator if single-stepping.  */
6581 		if (vcpu->mmio_is_write)
6582 			return 1;
6583 		vcpu->mmio_read_completed = 1;
6584 		return complete_emulated_io(vcpu);
6585 	}
6586 
6587 	run->exit_reason = KVM_EXIT_MMIO;
6588 	run->mmio.phys_addr = frag->gpa;
6589 	if (vcpu->mmio_is_write)
6590 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6591 	run->mmio.len = min(8u, frag->len);
6592 	run->mmio.is_write = vcpu->mmio_is_write;
6593 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6594 	return 0;
6595 }
6596 
6597 
6598 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6599 {
6600 	int r;
6601 	sigset_t sigsaved;
6602 
6603 	if (!tsk_used_math(current) && init_fpu(current))
6604 		return -ENOMEM;
6605 
6606 	if (vcpu->sigset_active)
6607 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6608 
6609 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6610 		kvm_vcpu_block(vcpu);
6611 		kvm_apic_accept_events(vcpu);
6612 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6613 		r = -EAGAIN;
6614 		goto out;
6615 	}
6616 
6617 	/* re-sync apic's tpr */
6618 	if (!irqchip_in_kernel(vcpu->kvm)) {
6619 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6620 			r = -EINVAL;
6621 			goto out;
6622 		}
6623 	}
6624 
6625 	if (unlikely(vcpu->arch.complete_userspace_io)) {
6626 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6627 		vcpu->arch.complete_userspace_io = NULL;
6628 		r = cui(vcpu);
6629 		if (r <= 0)
6630 			goto out;
6631 	} else
6632 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6633 
6634 	r = vcpu_run(vcpu);
6635 
6636 out:
6637 	post_kvm_run_save(vcpu);
6638 	if (vcpu->sigset_active)
6639 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6640 
6641 	return r;
6642 }
6643 
6644 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6645 {
6646 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6647 		/*
6648 		 * We are here if userspace calls get_regs() in the middle of
6649 		 * instruction emulation. Registers state needs to be copied
6650 		 * back from emulation context to vcpu. Userspace shouldn't do
6651 		 * that usually, but some bad designed PV devices (vmware
6652 		 * backdoor interface) need this to work
6653 		 */
6654 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6655 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6656 	}
6657 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6658 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6659 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6660 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6661 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6662 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6663 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6664 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6665 #ifdef CONFIG_X86_64
6666 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6667 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6668 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6669 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6670 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6671 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6672 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6673 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6674 #endif
6675 
6676 	regs->rip = kvm_rip_read(vcpu);
6677 	regs->rflags = kvm_get_rflags(vcpu);
6678 
6679 	return 0;
6680 }
6681 
6682 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6683 {
6684 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6685 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6686 
6687 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6688 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6689 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6690 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6691 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6692 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6693 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6694 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6695 #ifdef CONFIG_X86_64
6696 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6697 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6698 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6699 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6700 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6701 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6702 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6703 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6704 #endif
6705 
6706 	kvm_rip_write(vcpu, regs->rip);
6707 	kvm_set_rflags(vcpu, regs->rflags);
6708 
6709 	vcpu->arch.exception.pending = false;
6710 
6711 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6712 
6713 	return 0;
6714 }
6715 
6716 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6717 {
6718 	struct kvm_segment cs;
6719 
6720 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6721 	*db = cs.db;
6722 	*l = cs.l;
6723 }
6724 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6725 
6726 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6727 				  struct kvm_sregs *sregs)
6728 {
6729 	struct desc_ptr dt;
6730 
6731 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6732 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6733 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6734 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6735 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6736 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6737 
6738 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6739 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6740 
6741 	kvm_x86_ops->get_idt(vcpu, &dt);
6742 	sregs->idt.limit = dt.size;
6743 	sregs->idt.base = dt.address;
6744 	kvm_x86_ops->get_gdt(vcpu, &dt);
6745 	sregs->gdt.limit = dt.size;
6746 	sregs->gdt.base = dt.address;
6747 
6748 	sregs->cr0 = kvm_read_cr0(vcpu);
6749 	sregs->cr2 = vcpu->arch.cr2;
6750 	sregs->cr3 = kvm_read_cr3(vcpu);
6751 	sregs->cr4 = kvm_read_cr4(vcpu);
6752 	sregs->cr8 = kvm_get_cr8(vcpu);
6753 	sregs->efer = vcpu->arch.efer;
6754 	sregs->apic_base = kvm_get_apic_base(vcpu);
6755 
6756 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6757 
6758 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6759 		set_bit(vcpu->arch.interrupt.nr,
6760 			(unsigned long *)sregs->interrupt_bitmap);
6761 
6762 	return 0;
6763 }
6764 
6765 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6766 				    struct kvm_mp_state *mp_state)
6767 {
6768 	kvm_apic_accept_events(vcpu);
6769 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6770 					vcpu->arch.pv.pv_unhalted)
6771 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6772 	else
6773 		mp_state->mp_state = vcpu->arch.mp_state;
6774 
6775 	return 0;
6776 }
6777 
6778 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6779 				    struct kvm_mp_state *mp_state)
6780 {
6781 	if (!kvm_vcpu_has_lapic(vcpu) &&
6782 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6783 		return -EINVAL;
6784 
6785 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6786 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6787 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6788 	} else
6789 		vcpu->arch.mp_state = mp_state->mp_state;
6790 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6791 	return 0;
6792 }
6793 
6794 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6795 		    int reason, bool has_error_code, u32 error_code)
6796 {
6797 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6798 	int ret;
6799 
6800 	init_emulate_ctxt(vcpu);
6801 
6802 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6803 				   has_error_code, error_code);
6804 
6805 	if (ret)
6806 		return EMULATE_FAIL;
6807 
6808 	kvm_rip_write(vcpu, ctxt->eip);
6809 	kvm_set_rflags(vcpu, ctxt->eflags);
6810 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6811 	return EMULATE_DONE;
6812 }
6813 EXPORT_SYMBOL_GPL(kvm_task_switch);
6814 
6815 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6816 				  struct kvm_sregs *sregs)
6817 {
6818 	struct msr_data apic_base_msr;
6819 	int mmu_reset_needed = 0;
6820 	int pending_vec, max_bits, idx;
6821 	struct desc_ptr dt;
6822 
6823 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6824 		return -EINVAL;
6825 
6826 	dt.size = sregs->idt.limit;
6827 	dt.address = sregs->idt.base;
6828 	kvm_x86_ops->set_idt(vcpu, &dt);
6829 	dt.size = sregs->gdt.limit;
6830 	dt.address = sregs->gdt.base;
6831 	kvm_x86_ops->set_gdt(vcpu, &dt);
6832 
6833 	vcpu->arch.cr2 = sregs->cr2;
6834 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6835 	vcpu->arch.cr3 = sregs->cr3;
6836 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6837 
6838 	kvm_set_cr8(vcpu, sregs->cr8);
6839 
6840 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6841 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
6842 	apic_base_msr.data = sregs->apic_base;
6843 	apic_base_msr.host_initiated = true;
6844 	kvm_set_apic_base(vcpu, &apic_base_msr);
6845 
6846 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6847 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6848 	vcpu->arch.cr0 = sregs->cr0;
6849 
6850 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6851 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6852 	if (sregs->cr4 & X86_CR4_OSXSAVE)
6853 		kvm_update_cpuid(vcpu);
6854 
6855 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6856 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6857 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6858 		mmu_reset_needed = 1;
6859 	}
6860 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6861 
6862 	if (mmu_reset_needed)
6863 		kvm_mmu_reset_context(vcpu);
6864 
6865 	max_bits = KVM_NR_INTERRUPTS;
6866 	pending_vec = find_first_bit(
6867 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
6868 	if (pending_vec < max_bits) {
6869 		kvm_queue_interrupt(vcpu, pending_vec, false);
6870 		pr_debug("Set back pending irq %d\n", pending_vec);
6871 	}
6872 
6873 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6874 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6875 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6876 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6877 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6878 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6879 
6880 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6881 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6882 
6883 	update_cr8_intercept(vcpu);
6884 
6885 	/* Older userspace won't unhalt the vcpu on reset. */
6886 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6887 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6888 	    !is_protmode(vcpu))
6889 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6890 
6891 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6892 
6893 	return 0;
6894 }
6895 
6896 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6897 					struct kvm_guest_debug *dbg)
6898 {
6899 	unsigned long rflags;
6900 	int i, r;
6901 
6902 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6903 		r = -EBUSY;
6904 		if (vcpu->arch.exception.pending)
6905 			goto out;
6906 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6907 			kvm_queue_exception(vcpu, DB_VECTOR);
6908 		else
6909 			kvm_queue_exception(vcpu, BP_VECTOR);
6910 	}
6911 
6912 	/*
6913 	 * Read rflags as long as potentially injected trace flags are still
6914 	 * filtered out.
6915 	 */
6916 	rflags = kvm_get_rflags(vcpu);
6917 
6918 	vcpu->guest_debug = dbg->control;
6919 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6920 		vcpu->guest_debug = 0;
6921 
6922 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6923 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
6924 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6925 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6926 	} else {
6927 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6928 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6929 	}
6930 	kvm_update_dr7(vcpu);
6931 
6932 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6933 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6934 			get_segment_base(vcpu, VCPU_SREG_CS);
6935 
6936 	/*
6937 	 * Trigger an rflags update that will inject or remove the trace
6938 	 * flags.
6939 	 */
6940 	kvm_set_rflags(vcpu, rflags);
6941 
6942 	kvm_x86_ops->update_db_bp_intercept(vcpu);
6943 
6944 	r = 0;
6945 
6946 out:
6947 
6948 	return r;
6949 }
6950 
6951 /*
6952  * Translate a guest virtual address to a guest physical address.
6953  */
6954 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6955 				    struct kvm_translation *tr)
6956 {
6957 	unsigned long vaddr = tr->linear_address;
6958 	gpa_t gpa;
6959 	int idx;
6960 
6961 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6962 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6963 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6964 	tr->physical_address = gpa;
6965 	tr->valid = gpa != UNMAPPED_GVA;
6966 	tr->writeable = 1;
6967 	tr->usermode = 0;
6968 
6969 	return 0;
6970 }
6971 
6972 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6973 {
6974 	struct i387_fxsave_struct *fxsave =
6975 			&vcpu->arch.guest_fpu.state->fxsave;
6976 
6977 	memcpy(fpu->fpr, fxsave->st_space, 128);
6978 	fpu->fcw = fxsave->cwd;
6979 	fpu->fsw = fxsave->swd;
6980 	fpu->ftwx = fxsave->twd;
6981 	fpu->last_opcode = fxsave->fop;
6982 	fpu->last_ip = fxsave->rip;
6983 	fpu->last_dp = fxsave->rdp;
6984 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6985 
6986 	return 0;
6987 }
6988 
6989 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6990 {
6991 	struct i387_fxsave_struct *fxsave =
6992 			&vcpu->arch.guest_fpu.state->fxsave;
6993 
6994 	memcpy(fxsave->st_space, fpu->fpr, 128);
6995 	fxsave->cwd = fpu->fcw;
6996 	fxsave->swd = fpu->fsw;
6997 	fxsave->twd = fpu->ftwx;
6998 	fxsave->fop = fpu->last_opcode;
6999 	fxsave->rip = fpu->last_ip;
7000 	fxsave->rdp = fpu->last_dp;
7001 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7002 
7003 	return 0;
7004 }
7005 
7006 int fx_init(struct kvm_vcpu *vcpu)
7007 {
7008 	int err;
7009 
7010 	err = fpu_alloc(&vcpu->arch.guest_fpu);
7011 	if (err)
7012 		return err;
7013 
7014 	fpu_finit(&vcpu->arch.guest_fpu);
7015 	if (cpu_has_xsaves)
7016 		vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
7017 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7018 
7019 	/*
7020 	 * Ensure guest xcr0 is valid for loading
7021 	 */
7022 	vcpu->arch.xcr0 = XSTATE_FP;
7023 
7024 	vcpu->arch.cr0 |= X86_CR0_ET;
7025 
7026 	return 0;
7027 }
7028 EXPORT_SYMBOL_GPL(fx_init);
7029 
7030 static void fx_free(struct kvm_vcpu *vcpu)
7031 {
7032 	fpu_free(&vcpu->arch.guest_fpu);
7033 }
7034 
7035 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7036 {
7037 	if (vcpu->guest_fpu_loaded)
7038 		return;
7039 
7040 	/*
7041 	 * Restore all possible states in the guest,
7042 	 * and assume host would use all available bits.
7043 	 * Guest xcr0 would be loaded later.
7044 	 */
7045 	kvm_put_guest_xcr0(vcpu);
7046 	vcpu->guest_fpu_loaded = 1;
7047 	__kernel_fpu_begin();
7048 	fpu_restore_checking(&vcpu->arch.guest_fpu);
7049 	trace_kvm_fpu(1);
7050 }
7051 
7052 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7053 {
7054 	kvm_put_guest_xcr0(vcpu);
7055 
7056 	if (!vcpu->guest_fpu_loaded)
7057 		return;
7058 
7059 	vcpu->guest_fpu_loaded = 0;
7060 	fpu_save_init(&vcpu->arch.guest_fpu);
7061 	__kernel_fpu_end();
7062 	++vcpu->stat.fpu_reload;
7063 	if (!vcpu->arch.eager_fpu)
7064 		kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7065 
7066 	trace_kvm_fpu(0);
7067 }
7068 
7069 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7070 {
7071 	kvmclock_reset(vcpu);
7072 
7073 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7074 	fx_free(vcpu);
7075 	kvm_x86_ops->vcpu_free(vcpu);
7076 }
7077 
7078 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7079 						unsigned int id)
7080 {
7081 	struct kvm_vcpu *vcpu;
7082 
7083 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7084 		printk_once(KERN_WARNING
7085 		"kvm: SMP vm created on host with unstable TSC; "
7086 		"guest TSC will not be reliable\n");
7087 
7088 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7089 
7090 	/*
7091 	 * Activate fpu unconditionally in case the guest needs eager FPU.  It will be
7092 	 * deactivated soon if it doesn't.
7093 	 */
7094 	kvm_x86_ops->fpu_activate(vcpu);
7095 	return vcpu;
7096 }
7097 
7098 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7099 {
7100 	int r;
7101 
7102 	vcpu->arch.mtrr_state.have_fixed = 1;
7103 	r = vcpu_load(vcpu);
7104 	if (r)
7105 		return r;
7106 	kvm_vcpu_reset(vcpu);
7107 	kvm_mmu_setup(vcpu);
7108 	vcpu_put(vcpu);
7109 
7110 	return r;
7111 }
7112 
7113 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7114 {
7115 	struct msr_data msr;
7116 	struct kvm *kvm = vcpu->kvm;
7117 
7118 	if (vcpu_load(vcpu))
7119 		return;
7120 	msr.data = 0x0;
7121 	msr.index = MSR_IA32_TSC;
7122 	msr.host_initiated = true;
7123 	kvm_write_tsc(vcpu, &msr);
7124 	vcpu_put(vcpu);
7125 
7126 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7127 					KVMCLOCK_SYNC_PERIOD);
7128 }
7129 
7130 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7131 {
7132 	int r;
7133 	vcpu->arch.apf.msr_val = 0;
7134 
7135 	r = vcpu_load(vcpu);
7136 	BUG_ON(r);
7137 	kvm_mmu_unload(vcpu);
7138 	vcpu_put(vcpu);
7139 
7140 	fx_free(vcpu);
7141 	kvm_x86_ops->vcpu_free(vcpu);
7142 }
7143 
7144 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
7145 {
7146 	atomic_set(&vcpu->arch.nmi_queued, 0);
7147 	vcpu->arch.nmi_pending = 0;
7148 	vcpu->arch.nmi_injected = false;
7149 	kvm_clear_interrupt_queue(vcpu);
7150 	kvm_clear_exception_queue(vcpu);
7151 
7152 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7153 	kvm_update_dr0123(vcpu);
7154 	vcpu->arch.dr6 = DR6_INIT;
7155 	kvm_update_dr6(vcpu);
7156 	vcpu->arch.dr7 = DR7_FIXED_1;
7157 	kvm_update_dr7(vcpu);
7158 
7159 	vcpu->arch.cr2 = 0;
7160 
7161 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7162 	vcpu->arch.apf.msr_val = 0;
7163 	vcpu->arch.st.msr_val = 0;
7164 
7165 	kvmclock_reset(vcpu);
7166 
7167 	kvm_clear_async_pf_completion_queue(vcpu);
7168 	kvm_async_pf_hash_reset(vcpu);
7169 	vcpu->arch.apf.halted = false;
7170 
7171 	kvm_pmu_reset(vcpu);
7172 
7173 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7174 	vcpu->arch.regs_avail = ~0;
7175 	vcpu->arch.regs_dirty = ~0;
7176 
7177 	kvm_x86_ops->vcpu_reset(vcpu);
7178 }
7179 
7180 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7181 {
7182 	struct kvm_segment cs;
7183 
7184 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7185 	cs.selector = vector << 8;
7186 	cs.base = vector << 12;
7187 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7188 	kvm_rip_write(vcpu, 0);
7189 }
7190 
7191 int kvm_arch_hardware_enable(void)
7192 {
7193 	struct kvm *kvm;
7194 	struct kvm_vcpu *vcpu;
7195 	int i;
7196 	int ret;
7197 	u64 local_tsc;
7198 	u64 max_tsc = 0;
7199 	bool stable, backwards_tsc = false;
7200 
7201 	kvm_shared_msr_cpu_online();
7202 	ret = kvm_x86_ops->hardware_enable();
7203 	if (ret != 0)
7204 		return ret;
7205 
7206 	local_tsc = native_read_tsc();
7207 	stable = !check_tsc_unstable();
7208 	list_for_each_entry(kvm, &vm_list, vm_list) {
7209 		kvm_for_each_vcpu(i, vcpu, kvm) {
7210 			if (!stable && vcpu->cpu == smp_processor_id())
7211 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7212 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7213 				backwards_tsc = true;
7214 				if (vcpu->arch.last_host_tsc > max_tsc)
7215 					max_tsc = vcpu->arch.last_host_tsc;
7216 			}
7217 		}
7218 	}
7219 
7220 	/*
7221 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7222 	 * platforms that reset TSC during suspend or hibernate actions, but
7223 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7224 	 * detect that condition here, which happens early in CPU bringup,
7225 	 * before any KVM threads can be running.  Unfortunately, we can't
7226 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7227 	 * enough into CPU bringup that we know how much real time has actually
7228 	 * elapsed; our helper function, get_kernel_ns() will be using boot
7229 	 * variables that haven't been updated yet.
7230 	 *
7231 	 * So we simply find the maximum observed TSC above, then record the
7232 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7233 	 * the adjustment will be applied.  Note that we accumulate
7234 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7235 	 * gets a chance to run again.  In the event that no KVM threads get a
7236 	 * chance to run, we will miss the entire elapsed period, as we'll have
7237 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7238 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7239 	 * uniform across all VCPUs (not to mention the scenario is extremely
7240 	 * unlikely). It is possible that a second hibernate recovery happens
7241 	 * much faster than a first, causing the observed TSC here to be
7242 	 * smaller; this would require additional padding adjustment, which is
7243 	 * why we set last_host_tsc to the local tsc observed here.
7244 	 *
7245 	 * N.B. - this code below runs only on platforms with reliable TSC,
7246 	 * as that is the only way backwards_tsc is set above.  Also note
7247 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7248 	 * have the same delta_cyc adjustment applied if backwards_tsc
7249 	 * is detected.  Note further, this adjustment is only done once,
7250 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7251 	 * called multiple times (one for each physical CPU bringup).
7252 	 *
7253 	 * Platforms with unreliable TSCs don't have to deal with this, they
7254 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7255 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7256 	 * guarantee that they stay in perfect synchronization.
7257 	 */
7258 	if (backwards_tsc) {
7259 		u64 delta_cyc = max_tsc - local_tsc;
7260 		backwards_tsc_observed = true;
7261 		list_for_each_entry(kvm, &vm_list, vm_list) {
7262 			kvm_for_each_vcpu(i, vcpu, kvm) {
7263 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7264 				vcpu->arch.last_host_tsc = local_tsc;
7265 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7266 			}
7267 
7268 			/*
7269 			 * We have to disable TSC offset matching.. if you were
7270 			 * booting a VM while issuing an S4 host suspend....
7271 			 * you may have some problem.  Solving this issue is
7272 			 * left as an exercise to the reader.
7273 			 */
7274 			kvm->arch.last_tsc_nsec = 0;
7275 			kvm->arch.last_tsc_write = 0;
7276 		}
7277 
7278 	}
7279 	return 0;
7280 }
7281 
7282 void kvm_arch_hardware_disable(void)
7283 {
7284 	kvm_x86_ops->hardware_disable();
7285 	drop_user_return_notifiers();
7286 }
7287 
7288 int kvm_arch_hardware_setup(void)
7289 {
7290 	int r;
7291 
7292 	r = kvm_x86_ops->hardware_setup();
7293 	if (r != 0)
7294 		return r;
7295 
7296 	kvm_init_msr_list();
7297 	return 0;
7298 }
7299 
7300 void kvm_arch_hardware_unsetup(void)
7301 {
7302 	kvm_x86_ops->hardware_unsetup();
7303 }
7304 
7305 void kvm_arch_check_processor_compat(void *rtn)
7306 {
7307 	kvm_x86_ops->check_processor_compatibility(rtn);
7308 }
7309 
7310 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7311 {
7312 	return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7313 }
7314 
7315 struct static_key kvm_no_apic_vcpu __read_mostly;
7316 
7317 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7318 {
7319 	struct page *page;
7320 	struct kvm *kvm;
7321 	int r;
7322 
7323 	BUG_ON(vcpu->kvm == NULL);
7324 	kvm = vcpu->kvm;
7325 
7326 	vcpu->arch.pv.pv_unhalted = false;
7327 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7328 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7329 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7330 	else
7331 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7332 
7333 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7334 	if (!page) {
7335 		r = -ENOMEM;
7336 		goto fail;
7337 	}
7338 	vcpu->arch.pio_data = page_address(page);
7339 
7340 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7341 
7342 	r = kvm_mmu_create(vcpu);
7343 	if (r < 0)
7344 		goto fail_free_pio_data;
7345 
7346 	if (irqchip_in_kernel(kvm)) {
7347 		r = kvm_create_lapic(vcpu);
7348 		if (r < 0)
7349 			goto fail_mmu_destroy;
7350 	} else
7351 		static_key_slow_inc(&kvm_no_apic_vcpu);
7352 
7353 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7354 				       GFP_KERNEL);
7355 	if (!vcpu->arch.mce_banks) {
7356 		r = -ENOMEM;
7357 		goto fail_free_lapic;
7358 	}
7359 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7360 
7361 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7362 		r = -ENOMEM;
7363 		goto fail_free_mce_banks;
7364 	}
7365 
7366 	r = fx_init(vcpu);
7367 	if (r)
7368 		goto fail_free_wbinvd_dirty_mask;
7369 
7370 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7371 	vcpu->arch.pv_time_enabled = false;
7372 
7373 	vcpu->arch.guest_supported_xcr0 = 0;
7374 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7375 
7376 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7377 
7378 	kvm_async_pf_hash_reset(vcpu);
7379 	kvm_pmu_init(vcpu);
7380 
7381 	return 0;
7382 fail_free_wbinvd_dirty_mask:
7383 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7384 fail_free_mce_banks:
7385 	kfree(vcpu->arch.mce_banks);
7386 fail_free_lapic:
7387 	kvm_free_lapic(vcpu);
7388 fail_mmu_destroy:
7389 	kvm_mmu_destroy(vcpu);
7390 fail_free_pio_data:
7391 	free_page((unsigned long)vcpu->arch.pio_data);
7392 fail:
7393 	return r;
7394 }
7395 
7396 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7397 {
7398 	int idx;
7399 
7400 	kvm_pmu_destroy(vcpu);
7401 	kfree(vcpu->arch.mce_banks);
7402 	kvm_free_lapic(vcpu);
7403 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7404 	kvm_mmu_destroy(vcpu);
7405 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7406 	free_page((unsigned long)vcpu->arch.pio_data);
7407 	if (!irqchip_in_kernel(vcpu->kvm))
7408 		static_key_slow_dec(&kvm_no_apic_vcpu);
7409 }
7410 
7411 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7412 {
7413 	kvm_x86_ops->sched_in(vcpu, cpu);
7414 }
7415 
7416 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7417 {
7418 	if (type)
7419 		return -EINVAL;
7420 
7421 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7422 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7423 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7424 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7425 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7426 
7427 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7428 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7429 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7430 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7431 		&kvm->arch.irq_sources_bitmap);
7432 
7433 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7434 	mutex_init(&kvm->arch.apic_map_lock);
7435 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7436 
7437 	pvclock_update_vm_gtod_copy(kvm);
7438 
7439 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7440 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7441 
7442 	return 0;
7443 }
7444 
7445 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7446 {
7447 	int r;
7448 	r = vcpu_load(vcpu);
7449 	BUG_ON(r);
7450 	kvm_mmu_unload(vcpu);
7451 	vcpu_put(vcpu);
7452 }
7453 
7454 static void kvm_free_vcpus(struct kvm *kvm)
7455 {
7456 	unsigned int i;
7457 	struct kvm_vcpu *vcpu;
7458 
7459 	/*
7460 	 * Unpin any mmu pages first.
7461 	 */
7462 	kvm_for_each_vcpu(i, vcpu, kvm) {
7463 		kvm_clear_async_pf_completion_queue(vcpu);
7464 		kvm_unload_vcpu_mmu(vcpu);
7465 	}
7466 	kvm_for_each_vcpu(i, vcpu, kvm)
7467 		kvm_arch_vcpu_free(vcpu);
7468 
7469 	mutex_lock(&kvm->lock);
7470 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7471 		kvm->vcpus[i] = NULL;
7472 
7473 	atomic_set(&kvm->online_vcpus, 0);
7474 	mutex_unlock(&kvm->lock);
7475 }
7476 
7477 void kvm_arch_sync_events(struct kvm *kvm)
7478 {
7479 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7480 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7481 	kvm_free_all_assigned_devices(kvm);
7482 	kvm_free_pit(kvm);
7483 }
7484 
7485 void kvm_arch_destroy_vm(struct kvm *kvm)
7486 {
7487 	if (current->mm == kvm->mm) {
7488 		/*
7489 		 * Free memory regions allocated on behalf of userspace,
7490 		 * unless the the memory map has changed due to process exit
7491 		 * or fd copying.
7492 		 */
7493 		struct kvm_userspace_memory_region mem;
7494 		memset(&mem, 0, sizeof(mem));
7495 		mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7496 		kvm_set_memory_region(kvm, &mem);
7497 
7498 		mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7499 		kvm_set_memory_region(kvm, &mem);
7500 
7501 		mem.slot = TSS_PRIVATE_MEMSLOT;
7502 		kvm_set_memory_region(kvm, &mem);
7503 	}
7504 	kvm_iommu_unmap_guest(kvm);
7505 	kfree(kvm->arch.vpic);
7506 	kfree(kvm->arch.vioapic);
7507 	kvm_free_vcpus(kvm);
7508 	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7509 }
7510 
7511 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7512 			   struct kvm_memory_slot *dont)
7513 {
7514 	int i;
7515 
7516 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7517 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7518 			kvfree(free->arch.rmap[i]);
7519 			free->arch.rmap[i] = NULL;
7520 		}
7521 		if (i == 0)
7522 			continue;
7523 
7524 		if (!dont || free->arch.lpage_info[i - 1] !=
7525 			     dont->arch.lpage_info[i - 1]) {
7526 			kvfree(free->arch.lpage_info[i - 1]);
7527 			free->arch.lpage_info[i - 1] = NULL;
7528 		}
7529 	}
7530 }
7531 
7532 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7533 			    unsigned long npages)
7534 {
7535 	int i;
7536 
7537 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7538 		unsigned long ugfn;
7539 		int lpages;
7540 		int level = i + 1;
7541 
7542 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7543 				      slot->base_gfn, level) + 1;
7544 
7545 		slot->arch.rmap[i] =
7546 			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7547 		if (!slot->arch.rmap[i])
7548 			goto out_free;
7549 		if (i == 0)
7550 			continue;
7551 
7552 		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7553 					sizeof(*slot->arch.lpage_info[i - 1]));
7554 		if (!slot->arch.lpage_info[i - 1])
7555 			goto out_free;
7556 
7557 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7558 			slot->arch.lpage_info[i - 1][0].write_count = 1;
7559 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7560 			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7561 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7562 		/*
7563 		 * If the gfn and userspace address are not aligned wrt each
7564 		 * other, or if explicitly asked to, disable large page
7565 		 * support for this slot
7566 		 */
7567 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7568 		    !kvm_largepages_enabled()) {
7569 			unsigned long j;
7570 
7571 			for (j = 0; j < lpages; ++j)
7572 				slot->arch.lpage_info[i - 1][j].write_count = 1;
7573 		}
7574 	}
7575 
7576 	return 0;
7577 
7578 out_free:
7579 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7580 		kvfree(slot->arch.rmap[i]);
7581 		slot->arch.rmap[i] = NULL;
7582 		if (i == 0)
7583 			continue;
7584 
7585 		kvfree(slot->arch.lpage_info[i - 1]);
7586 		slot->arch.lpage_info[i - 1] = NULL;
7587 	}
7588 	return -ENOMEM;
7589 }
7590 
7591 void kvm_arch_memslots_updated(struct kvm *kvm)
7592 {
7593 	/*
7594 	 * memslots->generation has been incremented.
7595 	 * mmio generation may have reached its maximum value.
7596 	 */
7597 	kvm_mmu_invalidate_mmio_sptes(kvm);
7598 }
7599 
7600 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7601 				struct kvm_memory_slot *memslot,
7602 				struct kvm_userspace_memory_region *mem,
7603 				enum kvm_mr_change change)
7604 {
7605 	/*
7606 	 * Only private memory slots need to be mapped here since
7607 	 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7608 	 */
7609 	if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7610 		unsigned long userspace_addr;
7611 
7612 		/*
7613 		 * MAP_SHARED to prevent internal slot pages from being moved
7614 		 * by fork()/COW.
7615 		 */
7616 		userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7617 					 PROT_READ | PROT_WRITE,
7618 					 MAP_SHARED | MAP_ANONYMOUS, 0);
7619 
7620 		if (IS_ERR((void *)userspace_addr))
7621 			return PTR_ERR((void *)userspace_addr);
7622 
7623 		memslot->userspace_addr = userspace_addr;
7624 	}
7625 
7626 	return 0;
7627 }
7628 
7629 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7630 				     struct kvm_memory_slot *new)
7631 {
7632 	/* Still write protect RO slot */
7633 	if (new->flags & KVM_MEM_READONLY) {
7634 		kvm_mmu_slot_remove_write_access(kvm, new);
7635 		return;
7636 	}
7637 
7638 	/*
7639 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
7640 	 *
7641 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
7642 	 *
7643 	 *  - KVM_MR_CREATE with dirty logging is disabled
7644 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7645 	 *
7646 	 * The reason is, in case of PML, we need to set D-bit for any slots
7647 	 * with dirty logging disabled in order to eliminate unnecessary GPA
7648 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
7649 	 * guarantees leaving PML enabled during guest's lifetime won't have
7650 	 * any additonal overhead from PML when guest is running with dirty
7651 	 * logging disabled for memory slots.
7652 	 *
7653 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7654 	 * to dirty logging mode.
7655 	 *
7656 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7657 	 *
7658 	 * In case of write protect:
7659 	 *
7660 	 * Write protect all pages for dirty logging.
7661 	 *
7662 	 * All the sptes including the large sptes which point to this
7663 	 * slot are set to readonly. We can not create any new large
7664 	 * spte on this slot until the end of the logging.
7665 	 *
7666 	 * See the comments in fast_page_fault().
7667 	 */
7668 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7669 		if (kvm_x86_ops->slot_enable_log_dirty)
7670 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7671 		else
7672 			kvm_mmu_slot_remove_write_access(kvm, new);
7673 	} else {
7674 		if (kvm_x86_ops->slot_disable_log_dirty)
7675 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7676 	}
7677 }
7678 
7679 void kvm_arch_commit_memory_region(struct kvm *kvm,
7680 				struct kvm_userspace_memory_region *mem,
7681 				const struct kvm_memory_slot *old,
7682 				enum kvm_mr_change change)
7683 {
7684 	struct kvm_memory_slot *new;
7685 	int nr_mmu_pages = 0;
7686 
7687 	if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7688 		int ret;
7689 
7690 		ret = vm_munmap(old->userspace_addr,
7691 				old->npages * PAGE_SIZE);
7692 		if (ret < 0)
7693 			printk(KERN_WARNING
7694 			       "kvm_vm_ioctl_set_memory_region: "
7695 			       "failed to munmap memory\n");
7696 	}
7697 
7698 	if (!kvm->arch.n_requested_mmu_pages)
7699 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7700 
7701 	if (nr_mmu_pages)
7702 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7703 
7704 	/* It's OK to get 'new' slot here as it has already been installed */
7705 	new = id_to_memslot(kvm->memslots, mem->slot);
7706 
7707 	/*
7708 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
7709 	 * sptes have to be split.  If live migration is successful, the guest
7710 	 * in the source machine will be destroyed and large sptes will be
7711 	 * created in the destination. However, if the guest continues to run
7712 	 * in the source machine (for example if live migration fails), small
7713 	 * sptes will remain around and cause bad performance.
7714 	 *
7715 	 * Scan sptes if dirty logging has been stopped, dropping those
7716 	 * which can be collapsed into a single large-page spte.  Later
7717 	 * page faults will create the large-page sptes.
7718 	 */
7719 	if ((change != KVM_MR_DELETE) &&
7720 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7721 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7722 		kvm_mmu_zap_collapsible_sptes(kvm, new);
7723 
7724 	/*
7725 	 * Set up write protection and/or dirty logging for the new slot.
7726 	 *
7727 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7728 	 * been zapped so no dirty logging staff is needed for old slot. For
7729 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7730 	 * new and it's also covered when dealing with the new slot.
7731 	 */
7732 	if (change != KVM_MR_DELETE)
7733 		kvm_mmu_slot_apply_flags(kvm, new);
7734 }
7735 
7736 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7737 {
7738 	kvm_mmu_invalidate_zap_all_pages(kvm);
7739 }
7740 
7741 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7742 				   struct kvm_memory_slot *slot)
7743 {
7744 	kvm_mmu_invalidate_zap_all_pages(kvm);
7745 }
7746 
7747 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7748 {
7749 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7750 		kvm_x86_ops->check_nested_events(vcpu, false);
7751 
7752 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7753 		!vcpu->arch.apf.halted)
7754 		|| !list_empty_careful(&vcpu->async_pf.done)
7755 		|| kvm_apic_has_events(vcpu)
7756 		|| vcpu->arch.pv.pv_unhalted
7757 		|| atomic_read(&vcpu->arch.nmi_queued) ||
7758 		(kvm_arch_interrupt_allowed(vcpu) &&
7759 		 kvm_cpu_has_interrupt(vcpu));
7760 }
7761 
7762 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7763 {
7764 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7765 }
7766 
7767 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7768 {
7769 	return kvm_x86_ops->interrupt_allowed(vcpu);
7770 }
7771 
7772 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7773 {
7774 	if (is_64_bit_mode(vcpu))
7775 		return kvm_rip_read(vcpu);
7776 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7777 		     kvm_rip_read(vcpu));
7778 }
7779 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7780 
7781 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7782 {
7783 	return kvm_get_linear_rip(vcpu) == linear_rip;
7784 }
7785 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7786 
7787 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7788 {
7789 	unsigned long rflags;
7790 
7791 	rflags = kvm_x86_ops->get_rflags(vcpu);
7792 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7793 		rflags &= ~X86_EFLAGS_TF;
7794 	return rflags;
7795 }
7796 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7797 
7798 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7799 {
7800 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7801 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7802 		rflags |= X86_EFLAGS_TF;
7803 	kvm_x86_ops->set_rflags(vcpu, rflags);
7804 }
7805 
7806 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7807 {
7808 	__kvm_set_rflags(vcpu, rflags);
7809 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7810 }
7811 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7812 
7813 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7814 {
7815 	int r;
7816 
7817 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7818 	      work->wakeup_all)
7819 		return;
7820 
7821 	r = kvm_mmu_reload(vcpu);
7822 	if (unlikely(r))
7823 		return;
7824 
7825 	if (!vcpu->arch.mmu.direct_map &&
7826 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7827 		return;
7828 
7829 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7830 }
7831 
7832 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7833 {
7834 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7835 }
7836 
7837 static inline u32 kvm_async_pf_next_probe(u32 key)
7838 {
7839 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7840 }
7841 
7842 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7843 {
7844 	u32 key = kvm_async_pf_hash_fn(gfn);
7845 
7846 	while (vcpu->arch.apf.gfns[key] != ~0)
7847 		key = kvm_async_pf_next_probe(key);
7848 
7849 	vcpu->arch.apf.gfns[key] = gfn;
7850 }
7851 
7852 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7853 {
7854 	int i;
7855 	u32 key = kvm_async_pf_hash_fn(gfn);
7856 
7857 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7858 		     (vcpu->arch.apf.gfns[key] != gfn &&
7859 		      vcpu->arch.apf.gfns[key] != ~0); i++)
7860 		key = kvm_async_pf_next_probe(key);
7861 
7862 	return key;
7863 }
7864 
7865 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7866 {
7867 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7868 }
7869 
7870 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7871 {
7872 	u32 i, j, k;
7873 
7874 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7875 	while (true) {
7876 		vcpu->arch.apf.gfns[i] = ~0;
7877 		do {
7878 			j = kvm_async_pf_next_probe(j);
7879 			if (vcpu->arch.apf.gfns[j] == ~0)
7880 				return;
7881 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7882 			/*
7883 			 * k lies cyclically in ]i,j]
7884 			 * |    i.k.j |
7885 			 * |....j i.k.| or  |.k..j i...|
7886 			 */
7887 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7888 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7889 		i = j;
7890 	}
7891 }
7892 
7893 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7894 {
7895 
7896 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7897 				      sizeof(val));
7898 }
7899 
7900 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7901 				     struct kvm_async_pf *work)
7902 {
7903 	struct x86_exception fault;
7904 
7905 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7906 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7907 
7908 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7909 	    (vcpu->arch.apf.send_user_only &&
7910 	     kvm_x86_ops->get_cpl(vcpu) == 0))
7911 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7912 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7913 		fault.vector = PF_VECTOR;
7914 		fault.error_code_valid = true;
7915 		fault.error_code = 0;
7916 		fault.nested_page_fault = false;
7917 		fault.address = work->arch.token;
7918 		kvm_inject_page_fault(vcpu, &fault);
7919 	}
7920 }
7921 
7922 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7923 				 struct kvm_async_pf *work)
7924 {
7925 	struct x86_exception fault;
7926 
7927 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
7928 	if (work->wakeup_all)
7929 		work->arch.token = ~0; /* broadcast wakeup */
7930 	else
7931 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7932 
7933 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7934 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7935 		fault.vector = PF_VECTOR;
7936 		fault.error_code_valid = true;
7937 		fault.error_code = 0;
7938 		fault.nested_page_fault = false;
7939 		fault.address = work->arch.token;
7940 		kvm_inject_page_fault(vcpu, &fault);
7941 	}
7942 	vcpu->arch.apf.halted = false;
7943 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7944 }
7945 
7946 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7947 {
7948 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7949 		return true;
7950 	else
7951 		return !kvm_event_needs_reinjection(vcpu) &&
7952 			kvm_x86_ops->interrupt_allowed(vcpu);
7953 }
7954 
7955 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7956 {
7957 	atomic_inc(&kvm->arch.noncoherent_dma_count);
7958 }
7959 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7960 
7961 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7962 {
7963 	atomic_dec(&kvm->arch.noncoherent_dma_count);
7964 }
7965 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7966 
7967 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7968 {
7969 	return atomic_read(&kvm->arch.noncoherent_dma_count);
7970 }
7971 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7972 
7973 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7974 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7975 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7976 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7977 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7978 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7979 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7980 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7981 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7982 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7983 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7984 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7985 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7986 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
7987 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
7988