xref: /linux/arch/x86/kvm/x86.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
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 #include "pmu.h"
32 #include "hyperv.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <trace/events/kvm.h>
55 
56 #define CREATE_TRACE_POINTS
57 #include "trace.h"
58 
59 #include <asm/debugreg.h>
60 #include <asm/msr.h>
61 #include <asm/desc.h>
62 #include <asm/mce.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
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 static bool __read_mostly kvmclock_periodic_sync = true;
103 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
104 
105 bool kvm_has_tsc_control;
106 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
107 u32  kvm_max_guest_tsc_khz;
108 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
109 
110 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
111 static u32 tsc_tolerance_ppm = 250;
112 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
113 
114 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
115 unsigned int lapic_timer_advance_ns = 0;
116 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
117 
118 static bool backwards_tsc_observed = false;
119 
120 #define KVM_NR_SHARED_MSRS 16
121 
122 struct kvm_shared_msrs_global {
123 	int nr;
124 	u32 msrs[KVM_NR_SHARED_MSRS];
125 };
126 
127 struct kvm_shared_msrs {
128 	struct user_return_notifier urn;
129 	bool registered;
130 	struct kvm_shared_msr_values {
131 		u64 host;
132 		u64 curr;
133 	} values[KVM_NR_SHARED_MSRS];
134 };
135 
136 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
137 static struct kvm_shared_msrs __percpu *shared_msrs;
138 
139 struct kvm_stats_debugfs_item debugfs_entries[] = {
140 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
141 	{ "pf_guest", VCPU_STAT(pf_guest) },
142 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
143 	{ "invlpg", VCPU_STAT(invlpg) },
144 	{ "exits", VCPU_STAT(exits) },
145 	{ "io_exits", VCPU_STAT(io_exits) },
146 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
147 	{ "signal_exits", VCPU_STAT(signal_exits) },
148 	{ "irq_window", VCPU_STAT(irq_window_exits) },
149 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
150 	{ "halt_exits", VCPU_STAT(halt_exits) },
151 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
152 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
153 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
154 	{ "hypercalls", VCPU_STAT(hypercalls) },
155 	{ "request_irq", VCPU_STAT(request_irq_exits) },
156 	{ "irq_exits", VCPU_STAT(irq_exits) },
157 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
158 	{ "efer_reload", VCPU_STAT(efer_reload) },
159 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
160 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
161 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
162 	{ "irq_injections", VCPU_STAT(irq_injections) },
163 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
164 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
165 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
166 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
167 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
168 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
169 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
170 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
171 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
172 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
173 	{ "largepages", VM_STAT(lpages) },
174 	{ NULL }
175 };
176 
177 u64 __read_mostly host_xcr0;
178 
179 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
180 
181 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
182 {
183 	int i;
184 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
185 		vcpu->arch.apf.gfns[i] = ~0;
186 }
187 
188 static void kvm_on_user_return(struct user_return_notifier *urn)
189 {
190 	unsigned slot;
191 	struct kvm_shared_msrs *locals
192 		= container_of(urn, struct kvm_shared_msrs, urn);
193 	struct kvm_shared_msr_values *values;
194 
195 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
196 		values = &locals->values[slot];
197 		if (values->host != values->curr) {
198 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
199 			values->curr = values->host;
200 		}
201 	}
202 	locals->registered = false;
203 	user_return_notifier_unregister(urn);
204 }
205 
206 static void shared_msr_update(unsigned slot, u32 msr)
207 {
208 	u64 value;
209 	unsigned int cpu = smp_processor_id();
210 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
211 
212 	/* only read, and nobody should modify it at this time,
213 	 * so don't need lock */
214 	if (slot >= shared_msrs_global.nr) {
215 		printk(KERN_ERR "kvm: invalid MSR slot!");
216 		return;
217 	}
218 	rdmsrl_safe(msr, &value);
219 	smsr->values[slot].host = value;
220 	smsr->values[slot].curr = value;
221 }
222 
223 void kvm_define_shared_msr(unsigned slot, u32 msr)
224 {
225 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
226 	shared_msrs_global.msrs[slot] = msr;
227 	if (slot >= shared_msrs_global.nr)
228 		shared_msrs_global.nr = slot + 1;
229 }
230 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
231 
232 static void kvm_shared_msr_cpu_online(void)
233 {
234 	unsigned i;
235 
236 	for (i = 0; i < shared_msrs_global.nr; ++i)
237 		shared_msr_update(i, shared_msrs_global.msrs[i]);
238 }
239 
240 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
241 {
242 	unsigned int cpu = smp_processor_id();
243 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
244 	int err;
245 
246 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
247 		return 0;
248 	smsr->values[slot].curr = value;
249 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
250 	if (err)
251 		return 1;
252 
253 	if (!smsr->registered) {
254 		smsr->urn.on_user_return = kvm_on_user_return;
255 		user_return_notifier_register(&smsr->urn);
256 		smsr->registered = true;
257 	}
258 	return 0;
259 }
260 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
261 
262 static void drop_user_return_notifiers(void)
263 {
264 	unsigned int cpu = smp_processor_id();
265 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
266 
267 	if (smsr->registered)
268 		kvm_on_user_return(&smsr->urn);
269 }
270 
271 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
272 {
273 	return vcpu->arch.apic_base;
274 }
275 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
276 
277 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
278 {
279 	u64 old_state = vcpu->arch.apic_base &
280 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281 	u64 new_state = msr_info->data &
282 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
283 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
284 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
285 
286 	if (!msr_info->host_initiated &&
287 	    ((msr_info->data & reserved_bits) != 0 ||
288 	     new_state == X2APIC_ENABLE ||
289 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
290 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
291 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
292 	      old_state == 0)))
293 		return 1;
294 
295 	kvm_lapic_set_base(vcpu, msr_info->data);
296 	return 0;
297 }
298 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
299 
300 asmlinkage __visible void kvm_spurious_fault(void)
301 {
302 	/* Fault while not rebooting.  We want the trace. */
303 	BUG();
304 }
305 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
306 
307 #define EXCPT_BENIGN		0
308 #define EXCPT_CONTRIBUTORY	1
309 #define EXCPT_PF		2
310 
311 static int exception_class(int vector)
312 {
313 	switch (vector) {
314 	case PF_VECTOR:
315 		return EXCPT_PF;
316 	case DE_VECTOR:
317 	case TS_VECTOR:
318 	case NP_VECTOR:
319 	case SS_VECTOR:
320 	case GP_VECTOR:
321 		return EXCPT_CONTRIBUTORY;
322 	default:
323 		break;
324 	}
325 	return EXCPT_BENIGN;
326 }
327 
328 #define EXCPT_FAULT		0
329 #define EXCPT_TRAP		1
330 #define EXCPT_ABORT		2
331 #define EXCPT_INTERRUPT		3
332 
333 static int exception_type(int vector)
334 {
335 	unsigned int mask;
336 
337 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
338 		return EXCPT_INTERRUPT;
339 
340 	mask = 1 << vector;
341 
342 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
343 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
344 		return EXCPT_TRAP;
345 
346 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
347 		return EXCPT_ABORT;
348 
349 	/* Reserved exceptions will result in fault */
350 	return EXCPT_FAULT;
351 }
352 
353 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
354 		unsigned nr, bool has_error, u32 error_code,
355 		bool reinject)
356 {
357 	u32 prev_nr;
358 	int class1, class2;
359 
360 	kvm_make_request(KVM_REQ_EVENT, vcpu);
361 
362 	if (!vcpu->arch.exception.pending) {
363 	queue:
364 		if (has_error && !is_protmode(vcpu))
365 			has_error = false;
366 		vcpu->arch.exception.pending = true;
367 		vcpu->arch.exception.has_error_code = has_error;
368 		vcpu->arch.exception.nr = nr;
369 		vcpu->arch.exception.error_code = error_code;
370 		vcpu->arch.exception.reinject = reinject;
371 		return;
372 	}
373 
374 	/* to check exception */
375 	prev_nr = vcpu->arch.exception.nr;
376 	if (prev_nr == DF_VECTOR) {
377 		/* triple fault -> shutdown */
378 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
379 		return;
380 	}
381 	class1 = exception_class(prev_nr);
382 	class2 = exception_class(nr);
383 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
384 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
385 		/* generate double fault per SDM Table 5-5 */
386 		vcpu->arch.exception.pending = true;
387 		vcpu->arch.exception.has_error_code = true;
388 		vcpu->arch.exception.nr = DF_VECTOR;
389 		vcpu->arch.exception.error_code = 0;
390 	} else
391 		/* replace previous exception with a new one in a hope
392 		   that instruction re-execution will regenerate lost
393 		   exception */
394 		goto queue;
395 }
396 
397 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
398 {
399 	kvm_multiple_exception(vcpu, nr, false, 0, false);
400 }
401 EXPORT_SYMBOL_GPL(kvm_queue_exception);
402 
403 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
404 {
405 	kvm_multiple_exception(vcpu, nr, false, 0, true);
406 }
407 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
408 
409 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
410 {
411 	if (err)
412 		kvm_inject_gp(vcpu, 0);
413 	else
414 		kvm_x86_ops->skip_emulated_instruction(vcpu);
415 }
416 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
417 
418 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
419 {
420 	++vcpu->stat.pf_guest;
421 	vcpu->arch.cr2 = fault->address;
422 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
423 }
424 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
425 
426 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
427 {
428 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
429 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
430 	else
431 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
432 
433 	return fault->nested_page_fault;
434 }
435 
436 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
437 {
438 	atomic_inc(&vcpu->arch.nmi_queued);
439 	kvm_make_request(KVM_REQ_NMI, vcpu);
440 }
441 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
442 
443 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
444 {
445 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
446 }
447 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
448 
449 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
450 {
451 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
452 }
453 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
454 
455 /*
456  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
457  * a #GP and return false.
458  */
459 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
460 {
461 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
462 		return true;
463 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
464 	return false;
465 }
466 EXPORT_SYMBOL_GPL(kvm_require_cpl);
467 
468 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
469 {
470 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
471 		return true;
472 
473 	kvm_queue_exception(vcpu, UD_VECTOR);
474 	return false;
475 }
476 EXPORT_SYMBOL_GPL(kvm_require_dr);
477 
478 /*
479  * This function will be used to read from the physical memory of the currently
480  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
481  * can read from guest physical or from the guest's guest physical memory.
482  */
483 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
484 			    gfn_t ngfn, void *data, int offset, int len,
485 			    u32 access)
486 {
487 	struct x86_exception exception;
488 	gfn_t real_gfn;
489 	gpa_t ngpa;
490 
491 	ngpa     = gfn_to_gpa(ngfn);
492 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
493 	if (real_gfn == UNMAPPED_GVA)
494 		return -EFAULT;
495 
496 	real_gfn = gpa_to_gfn(real_gfn);
497 
498 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
499 }
500 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
501 
502 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
503 			       void *data, int offset, int len, u32 access)
504 {
505 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
506 				       data, offset, len, access);
507 }
508 
509 /*
510  * Load the pae pdptrs.  Return true is they are all valid.
511  */
512 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
513 {
514 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
515 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
516 	int i;
517 	int ret;
518 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
519 
520 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
521 				      offset * sizeof(u64), sizeof(pdpte),
522 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
523 	if (ret < 0) {
524 		ret = 0;
525 		goto out;
526 	}
527 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
528 		if (is_present_gpte(pdpte[i]) &&
529 		    (pdpte[i] &
530 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
531 			ret = 0;
532 			goto out;
533 		}
534 	}
535 	ret = 1;
536 
537 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
538 	__set_bit(VCPU_EXREG_PDPTR,
539 		  (unsigned long *)&vcpu->arch.regs_avail);
540 	__set_bit(VCPU_EXREG_PDPTR,
541 		  (unsigned long *)&vcpu->arch.regs_dirty);
542 out:
543 
544 	return ret;
545 }
546 EXPORT_SYMBOL_GPL(load_pdptrs);
547 
548 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
549 {
550 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
551 	bool changed = true;
552 	int offset;
553 	gfn_t gfn;
554 	int r;
555 
556 	if (is_long_mode(vcpu) || !is_pae(vcpu))
557 		return false;
558 
559 	if (!test_bit(VCPU_EXREG_PDPTR,
560 		      (unsigned long *)&vcpu->arch.regs_avail))
561 		return true;
562 
563 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
564 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
565 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
566 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
567 	if (r < 0)
568 		goto out;
569 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
570 out:
571 
572 	return changed;
573 }
574 
575 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
576 {
577 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
578 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
579 
580 	cr0 |= X86_CR0_ET;
581 
582 #ifdef CONFIG_X86_64
583 	if (cr0 & 0xffffffff00000000UL)
584 		return 1;
585 #endif
586 
587 	cr0 &= ~CR0_RESERVED_BITS;
588 
589 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
590 		return 1;
591 
592 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
593 		return 1;
594 
595 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
596 #ifdef CONFIG_X86_64
597 		if ((vcpu->arch.efer & EFER_LME)) {
598 			int cs_db, cs_l;
599 
600 			if (!is_pae(vcpu))
601 				return 1;
602 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
603 			if (cs_l)
604 				return 1;
605 		} else
606 #endif
607 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
608 						 kvm_read_cr3(vcpu)))
609 			return 1;
610 	}
611 
612 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
613 		return 1;
614 
615 	kvm_x86_ops->set_cr0(vcpu, cr0);
616 
617 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
618 		kvm_clear_async_pf_completion_queue(vcpu);
619 		kvm_async_pf_hash_reset(vcpu);
620 	}
621 
622 	if ((cr0 ^ old_cr0) & update_bits)
623 		kvm_mmu_reset_context(vcpu);
624 
625 	if ((cr0 ^ old_cr0) & X86_CR0_CD)
626 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
627 
628 	return 0;
629 }
630 EXPORT_SYMBOL_GPL(kvm_set_cr0);
631 
632 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
633 {
634 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
635 }
636 EXPORT_SYMBOL_GPL(kvm_lmsw);
637 
638 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
639 {
640 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
641 			!vcpu->guest_xcr0_loaded) {
642 		/* kvm_set_xcr() also depends on this */
643 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
644 		vcpu->guest_xcr0_loaded = 1;
645 	}
646 }
647 
648 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
649 {
650 	if (vcpu->guest_xcr0_loaded) {
651 		if (vcpu->arch.xcr0 != host_xcr0)
652 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
653 		vcpu->guest_xcr0_loaded = 0;
654 	}
655 }
656 
657 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
658 {
659 	u64 xcr0 = xcr;
660 	u64 old_xcr0 = vcpu->arch.xcr0;
661 	u64 valid_bits;
662 
663 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
664 	if (index != XCR_XFEATURE_ENABLED_MASK)
665 		return 1;
666 	if (!(xcr0 & XSTATE_FP))
667 		return 1;
668 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
669 		return 1;
670 
671 	/*
672 	 * Do not allow the guest to set bits that we do not support
673 	 * saving.  However, xcr0 bit 0 is always set, even if the
674 	 * emulated CPU does not support XSAVE (see fx_init).
675 	 */
676 	valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
677 	if (xcr0 & ~valid_bits)
678 		return 1;
679 
680 	if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
681 		return 1;
682 
683 	if (xcr0 & XSTATE_AVX512) {
684 		if (!(xcr0 & XSTATE_YMM))
685 			return 1;
686 		if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
687 			return 1;
688 	}
689 	kvm_put_guest_xcr0(vcpu);
690 	vcpu->arch.xcr0 = xcr0;
691 
692 	if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
693 		kvm_update_cpuid(vcpu);
694 	return 0;
695 }
696 
697 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
698 {
699 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
700 	    __kvm_set_xcr(vcpu, index, xcr)) {
701 		kvm_inject_gp(vcpu, 0);
702 		return 1;
703 	}
704 	return 0;
705 }
706 EXPORT_SYMBOL_GPL(kvm_set_xcr);
707 
708 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
709 {
710 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
711 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
712 				   X86_CR4_SMEP | X86_CR4_SMAP;
713 
714 	if (cr4 & CR4_RESERVED_BITS)
715 		return 1;
716 
717 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
718 		return 1;
719 
720 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
721 		return 1;
722 
723 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
724 		return 1;
725 
726 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
727 		return 1;
728 
729 	if (is_long_mode(vcpu)) {
730 		if (!(cr4 & X86_CR4_PAE))
731 			return 1;
732 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
733 		   && ((cr4 ^ old_cr4) & pdptr_bits)
734 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
735 				   kvm_read_cr3(vcpu)))
736 		return 1;
737 
738 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
739 		if (!guest_cpuid_has_pcid(vcpu))
740 			return 1;
741 
742 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
743 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
744 			return 1;
745 	}
746 
747 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
748 		return 1;
749 
750 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
751 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
752 		kvm_mmu_reset_context(vcpu);
753 
754 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
755 		kvm_update_cpuid(vcpu);
756 
757 	return 0;
758 }
759 EXPORT_SYMBOL_GPL(kvm_set_cr4);
760 
761 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
762 {
763 #ifdef CONFIG_X86_64
764 	cr3 &= ~CR3_PCID_INVD;
765 #endif
766 
767 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
768 		kvm_mmu_sync_roots(vcpu);
769 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
770 		return 0;
771 	}
772 
773 	if (is_long_mode(vcpu)) {
774 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
775 			return 1;
776 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
777 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
778 		return 1;
779 
780 	vcpu->arch.cr3 = cr3;
781 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
782 	kvm_mmu_new_cr3(vcpu);
783 	return 0;
784 }
785 EXPORT_SYMBOL_GPL(kvm_set_cr3);
786 
787 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
788 {
789 	if (cr8 & CR8_RESERVED_BITS)
790 		return 1;
791 	if (irqchip_in_kernel(vcpu->kvm))
792 		kvm_lapic_set_tpr(vcpu, cr8);
793 	else
794 		vcpu->arch.cr8 = cr8;
795 	return 0;
796 }
797 EXPORT_SYMBOL_GPL(kvm_set_cr8);
798 
799 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
800 {
801 	if (irqchip_in_kernel(vcpu->kvm))
802 		return kvm_lapic_get_cr8(vcpu);
803 	else
804 		return vcpu->arch.cr8;
805 }
806 EXPORT_SYMBOL_GPL(kvm_get_cr8);
807 
808 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
809 {
810 	int i;
811 
812 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
813 		for (i = 0; i < KVM_NR_DB_REGS; i++)
814 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
815 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
816 	}
817 }
818 
819 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
820 {
821 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
822 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
823 }
824 
825 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
826 {
827 	unsigned long dr7;
828 
829 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
830 		dr7 = vcpu->arch.guest_debug_dr7;
831 	else
832 		dr7 = vcpu->arch.dr7;
833 	kvm_x86_ops->set_dr7(vcpu, dr7);
834 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
835 	if (dr7 & DR7_BP_EN_MASK)
836 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
837 }
838 
839 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
840 {
841 	u64 fixed = DR6_FIXED_1;
842 
843 	if (!guest_cpuid_has_rtm(vcpu))
844 		fixed |= DR6_RTM;
845 	return fixed;
846 }
847 
848 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
849 {
850 	switch (dr) {
851 	case 0 ... 3:
852 		vcpu->arch.db[dr] = val;
853 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
854 			vcpu->arch.eff_db[dr] = val;
855 		break;
856 	case 4:
857 		/* fall through */
858 	case 6:
859 		if (val & 0xffffffff00000000ULL)
860 			return -1; /* #GP */
861 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
862 		kvm_update_dr6(vcpu);
863 		break;
864 	case 5:
865 		/* fall through */
866 	default: /* 7 */
867 		if (val & 0xffffffff00000000ULL)
868 			return -1; /* #GP */
869 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
870 		kvm_update_dr7(vcpu);
871 		break;
872 	}
873 
874 	return 0;
875 }
876 
877 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
878 {
879 	if (__kvm_set_dr(vcpu, dr, val)) {
880 		kvm_inject_gp(vcpu, 0);
881 		return 1;
882 	}
883 	return 0;
884 }
885 EXPORT_SYMBOL_GPL(kvm_set_dr);
886 
887 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
888 {
889 	switch (dr) {
890 	case 0 ... 3:
891 		*val = vcpu->arch.db[dr];
892 		break;
893 	case 4:
894 		/* fall through */
895 	case 6:
896 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
897 			*val = vcpu->arch.dr6;
898 		else
899 			*val = kvm_x86_ops->get_dr6(vcpu);
900 		break;
901 	case 5:
902 		/* fall through */
903 	default: /* 7 */
904 		*val = vcpu->arch.dr7;
905 		break;
906 	}
907 	return 0;
908 }
909 EXPORT_SYMBOL_GPL(kvm_get_dr);
910 
911 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
912 {
913 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
914 	u64 data;
915 	int err;
916 
917 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
918 	if (err)
919 		return err;
920 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
921 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
922 	return err;
923 }
924 EXPORT_SYMBOL_GPL(kvm_rdpmc);
925 
926 /*
927  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
928  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
929  *
930  * This list is modified at module load time to reflect the
931  * capabilities of the host cpu. This capabilities test skips MSRs that are
932  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
933  * may depend on host virtualization features rather than host cpu features.
934  */
935 
936 static u32 msrs_to_save[] = {
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 u32 emulated_msrs[] = {
949 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
950 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
951 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
952 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
953 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
954 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
955 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
956 	MSR_KVM_PV_EOI_EN,
957 
958 	MSR_IA32_TSC_ADJUST,
959 	MSR_IA32_TSCDEADLINE,
960 	MSR_IA32_MISC_ENABLE,
961 	MSR_IA32_MCG_STATUS,
962 	MSR_IA32_MCG_CTL,
963 	MSR_IA32_SMBASE,
964 };
965 
966 static unsigned num_emulated_msrs;
967 
968 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
969 {
970 	if (efer & efer_reserved_bits)
971 		return false;
972 
973 	if (efer & EFER_FFXSR) {
974 		struct kvm_cpuid_entry2 *feat;
975 
976 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
977 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
978 			return false;
979 	}
980 
981 	if (efer & EFER_SVME) {
982 		struct kvm_cpuid_entry2 *feat;
983 
984 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
985 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
986 			return false;
987 	}
988 
989 	return true;
990 }
991 EXPORT_SYMBOL_GPL(kvm_valid_efer);
992 
993 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
994 {
995 	u64 old_efer = vcpu->arch.efer;
996 
997 	if (!kvm_valid_efer(vcpu, efer))
998 		return 1;
999 
1000 	if (is_paging(vcpu)
1001 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1002 		return 1;
1003 
1004 	efer &= ~EFER_LMA;
1005 	efer |= vcpu->arch.efer & EFER_LMA;
1006 
1007 	kvm_x86_ops->set_efer(vcpu, efer);
1008 
1009 	/* Update reserved bits */
1010 	if ((efer ^ old_efer) & EFER_NX)
1011 		kvm_mmu_reset_context(vcpu);
1012 
1013 	return 0;
1014 }
1015 
1016 void kvm_enable_efer_bits(u64 mask)
1017 {
1018        efer_reserved_bits &= ~mask;
1019 }
1020 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1021 
1022 /*
1023  * Writes msr value into into the appropriate "register".
1024  * Returns 0 on success, non-0 otherwise.
1025  * Assumes vcpu_load() was already called.
1026  */
1027 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1028 {
1029 	switch (msr->index) {
1030 	case MSR_FS_BASE:
1031 	case MSR_GS_BASE:
1032 	case MSR_KERNEL_GS_BASE:
1033 	case MSR_CSTAR:
1034 	case MSR_LSTAR:
1035 		if (is_noncanonical_address(msr->data))
1036 			return 1;
1037 		break;
1038 	case MSR_IA32_SYSENTER_EIP:
1039 	case MSR_IA32_SYSENTER_ESP:
1040 		/*
1041 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1042 		 * non-canonical address is written on Intel but not on
1043 		 * AMD (which ignores the top 32-bits, because it does
1044 		 * not implement 64-bit SYSENTER).
1045 		 *
1046 		 * 64-bit code should hence be able to write a non-canonical
1047 		 * value on AMD.  Making the address canonical ensures that
1048 		 * vmentry does not fail on Intel after writing a non-canonical
1049 		 * value, and that something deterministic happens if the guest
1050 		 * invokes 64-bit SYSENTER.
1051 		 */
1052 		msr->data = get_canonical(msr->data);
1053 	}
1054 	return kvm_x86_ops->set_msr(vcpu, msr);
1055 }
1056 EXPORT_SYMBOL_GPL(kvm_set_msr);
1057 
1058 /*
1059  * Adapt set_msr() to msr_io()'s calling convention
1060  */
1061 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1062 {
1063 	struct msr_data msr;
1064 	int r;
1065 
1066 	msr.index = index;
1067 	msr.host_initiated = true;
1068 	r = kvm_get_msr(vcpu, &msr);
1069 	if (r)
1070 		return r;
1071 
1072 	*data = msr.data;
1073 	return 0;
1074 }
1075 
1076 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1077 {
1078 	struct msr_data msr;
1079 
1080 	msr.data = *data;
1081 	msr.index = index;
1082 	msr.host_initiated = true;
1083 	return kvm_set_msr(vcpu, &msr);
1084 }
1085 
1086 #ifdef CONFIG_X86_64
1087 struct pvclock_gtod_data {
1088 	seqcount_t	seq;
1089 
1090 	struct { /* extract of a clocksource struct */
1091 		int vclock_mode;
1092 		cycle_t	cycle_last;
1093 		cycle_t	mask;
1094 		u32	mult;
1095 		u32	shift;
1096 	} clock;
1097 
1098 	u64		boot_ns;
1099 	u64		nsec_base;
1100 };
1101 
1102 static struct pvclock_gtod_data pvclock_gtod_data;
1103 
1104 static void update_pvclock_gtod(struct timekeeper *tk)
1105 {
1106 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1107 	u64 boot_ns;
1108 
1109 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1110 
1111 	write_seqcount_begin(&vdata->seq);
1112 
1113 	/* copy pvclock gtod data */
1114 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1115 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1116 	vdata->clock.mask		= tk->tkr_mono.mask;
1117 	vdata->clock.mult		= tk->tkr_mono.mult;
1118 	vdata->clock.shift		= tk->tkr_mono.shift;
1119 
1120 	vdata->boot_ns			= boot_ns;
1121 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1122 
1123 	write_seqcount_end(&vdata->seq);
1124 }
1125 #endif
1126 
1127 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1128 {
1129 	/*
1130 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1131 	 * vcpu_enter_guest.  This function is only called from
1132 	 * the physical CPU that is running vcpu.
1133 	 */
1134 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1135 }
1136 
1137 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1138 {
1139 	int version;
1140 	int r;
1141 	struct pvclock_wall_clock wc;
1142 	struct timespec boot;
1143 
1144 	if (!wall_clock)
1145 		return;
1146 
1147 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1148 	if (r)
1149 		return;
1150 
1151 	if (version & 1)
1152 		++version;  /* first time write, random junk */
1153 
1154 	++version;
1155 
1156 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1157 
1158 	/*
1159 	 * The guest calculates current wall clock time by adding
1160 	 * system time (updated by kvm_guest_time_update below) to the
1161 	 * wall clock specified here.  guest system time equals host
1162 	 * system time for us, thus we must fill in host boot time here.
1163 	 */
1164 	getboottime(&boot);
1165 
1166 	if (kvm->arch.kvmclock_offset) {
1167 		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1168 		boot = timespec_sub(boot, ts);
1169 	}
1170 	wc.sec = boot.tv_sec;
1171 	wc.nsec = boot.tv_nsec;
1172 	wc.version = version;
1173 
1174 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1175 
1176 	version++;
1177 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1178 }
1179 
1180 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1181 {
1182 	uint32_t quotient, remainder;
1183 
1184 	/* Don't try to replace with do_div(), this one calculates
1185 	 * "(dividend << 32) / divisor" */
1186 	__asm__ ( "divl %4"
1187 		  : "=a" (quotient), "=d" (remainder)
1188 		  : "0" (0), "1" (dividend), "r" (divisor) );
1189 	return quotient;
1190 }
1191 
1192 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1193 			       s8 *pshift, u32 *pmultiplier)
1194 {
1195 	uint64_t scaled64;
1196 	int32_t  shift = 0;
1197 	uint64_t tps64;
1198 	uint32_t tps32;
1199 
1200 	tps64 = base_khz * 1000LL;
1201 	scaled64 = scaled_khz * 1000LL;
1202 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1203 		tps64 >>= 1;
1204 		shift--;
1205 	}
1206 
1207 	tps32 = (uint32_t)tps64;
1208 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1209 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1210 			scaled64 >>= 1;
1211 		else
1212 			tps32 <<= 1;
1213 		shift++;
1214 	}
1215 
1216 	*pshift = shift;
1217 	*pmultiplier = div_frac(scaled64, tps32);
1218 
1219 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1220 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1221 }
1222 
1223 #ifdef CONFIG_X86_64
1224 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1225 #endif
1226 
1227 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1228 static unsigned long max_tsc_khz;
1229 
1230 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1231 {
1232 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1233 				   vcpu->arch.virtual_tsc_shift);
1234 }
1235 
1236 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1237 {
1238 	u64 v = (u64)khz * (1000000 + ppm);
1239 	do_div(v, 1000000);
1240 	return v;
1241 }
1242 
1243 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1244 {
1245 	u32 thresh_lo, thresh_hi;
1246 	int use_scaling = 0;
1247 
1248 	/* tsc_khz can be zero if TSC calibration fails */
1249 	if (this_tsc_khz == 0)
1250 		return;
1251 
1252 	/* Compute a scale to convert nanoseconds in TSC cycles */
1253 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1254 			   &vcpu->arch.virtual_tsc_shift,
1255 			   &vcpu->arch.virtual_tsc_mult);
1256 	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1257 
1258 	/*
1259 	 * Compute the variation in TSC rate which is acceptable
1260 	 * within the range of tolerance and decide if the
1261 	 * rate being applied is within that bounds of the hardware
1262 	 * rate.  If so, no scaling or compensation need be done.
1263 	 */
1264 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1265 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1266 	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1267 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1268 		use_scaling = 1;
1269 	}
1270 	kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1271 }
1272 
1273 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1274 {
1275 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1276 				      vcpu->arch.virtual_tsc_mult,
1277 				      vcpu->arch.virtual_tsc_shift);
1278 	tsc += vcpu->arch.this_tsc_write;
1279 	return tsc;
1280 }
1281 
1282 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1283 {
1284 #ifdef CONFIG_X86_64
1285 	bool vcpus_matched;
1286 	struct kvm_arch *ka = &vcpu->kvm->arch;
1287 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1288 
1289 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1290 			 atomic_read(&vcpu->kvm->online_vcpus));
1291 
1292 	/*
1293 	 * Once the masterclock is enabled, always perform request in
1294 	 * order to update it.
1295 	 *
1296 	 * In order to enable masterclock, the host clocksource must be TSC
1297 	 * and the vcpus need to have matched TSCs.  When that happens,
1298 	 * perform request to enable masterclock.
1299 	 */
1300 	if (ka->use_master_clock ||
1301 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1302 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1303 
1304 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1305 			    atomic_read(&vcpu->kvm->online_vcpus),
1306 		            ka->use_master_clock, gtod->clock.vclock_mode);
1307 #endif
1308 }
1309 
1310 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1311 {
1312 	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1313 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1314 }
1315 
1316 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1317 {
1318 	struct kvm *kvm = vcpu->kvm;
1319 	u64 offset, ns, elapsed;
1320 	unsigned long flags;
1321 	s64 usdiff;
1322 	bool matched;
1323 	bool already_matched;
1324 	u64 data = msr->data;
1325 
1326 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1327 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1328 	ns = get_kernel_ns();
1329 	elapsed = ns - kvm->arch.last_tsc_nsec;
1330 
1331 	if (vcpu->arch.virtual_tsc_khz) {
1332 		int faulted = 0;
1333 
1334 		/* n.b - signed multiplication and division required */
1335 		usdiff = data - kvm->arch.last_tsc_write;
1336 #ifdef CONFIG_X86_64
1337 		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1338 #else
1339 		/* do_div() only does unsigned */
1340 		asm("1: idivl %[divisor]\n"
1341 		    "2: xor %%edx, %%edx\n"
1342 		    "   movl $0, %[faulted]\n"
1343 		    "3:\n"
1344 		    ".section .fixup,\"ax\"\n"
1345 		    "4: movl $1, %[faulted]\n"
1346 		    "   jmp  3b\n"
1347 		    ".previous\n"
1348 
1349 		_ASM_EXTABLE(1b, 4b)
1350 
1351 		: "=A"(usdiff), [faulted] "=r" (faulted)
1352 		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1353 
1354 #endif
1355 		do_div(elapsed, 1000);
1356 		usdiff -= elapsed;
1357 		if (usdiff < 0)
1358 			usdiff = -usdiff;
1359 
1360 		/* idivl overflow => difference is larger than USEC_PER_SEC */
1361 		if (faulted)
1362 			usdiff = USEC_PER_SEC;
1363 	} else
1364 		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1365 
1366 	/*
1367 	 * Special case: TSC write with a small delta (1 second) of virtual
1368 	 * cycle time against real time is interpreted as an attempt to
1369 	 * synchronize the CPU.
1370          *
1371 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1372 	 * TSC, we add elapsed time in this computation.  We could let the
1373 	 * compensation code attempt to catch up if we fall behind, but
1374 	 * it's better to try to match offsets from the beginning.
1375          */
1376 	if (usdiff < USEC_PER_SEC &&
1377 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1378 		if (!check_tsc_unstable()) {
1379 			offset = kvm->arch.cur_tsc_offset;
1380 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1381 		} else {
1382 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1383 			data += delta;
1384 			offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1385 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1386 		}
1387 		matched = true;
1388 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1389 	} else {
1390 		/*
1391 		 * We split periods of matched TSC writes into generations.
1392 		 * For each generation, we track the original measured
1393 		 * nanosecond time, offset, and write, so if TSCs are in
1394 		 * sync, we can match exact offset, and if not, we can match
1395 		 * exact software computation in compute_guest_tsc()
1396 		 *
1397 		 * These values are tracked in kvm->arch.cur_xxx variables.
1398 		 */
1399 		kvm->arch.cur_tsc_generation++;
1400 		kvm->arch.cur_tsc_nsec = ns;
1401 		kvm->arch.cur_tsc_write = data;
1402 		kvm->arch.cur_tsc_offset = offset;
1403 		matched = false;
1404 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1405 			 kvm->arch.cur_tsc_generation, data);
1406 	}
1407 
1408 	/*
1409 	 * We also track th most recent recorded KHZ, write and time to
1410 	 * allow the matching interval to be extended at each write.
1411 	 */
1412 	kvm->arch.last_tsc_nsec = ns;
1413 	kvm->arch.last_tsc_write = data;
1414 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1415 
1416 	vcpu->arch.last_guest_tsc = data;
1417 
1418 	/* Keep track of which generation this VCPU has synchronized to */
1419 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1420 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1421 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1422 
1423 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1424 		update_ia32_tsc_adjust_msr(vcpu, offset);
1425 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1426 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1427 
1428 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1429 	if (!matched) {
1430 		kvm->arch.nr_vcpus_matched_tsc = 0;
1431 	} else if (!already_matched) {
1432 		kvm->arch.nr_vcpus_matched_tsc++;
1433 	}
1434 
1435 	kvm_track_tsc_matching(vcpu);
1436 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1437 }
1438 
1439 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1440 
1441 #ifdef CONFIG_X86_64
1442 
1443 static cycle_t read_tsc(void)
1444 {
1445 	cycle_t ret = (cycle_t)rdtsc_ordered();
1446 	u64 last = pvclock_gtod_data.clock.cycle_last;
1447 
1448 	if (likely(ret >= last))
1449 		return ret;
1450 
1451 	/*
1452 	 * GCC likes to generate cmov here, but this branch is extremely
1453 	 * predictable (it's just a funciton of time and the likely is
1454 	 * very likely) and there's a data dependence, so force GCC
1455 	 * to generate a branch instead.  I don't barrier() because
1456 	 * we don't actually need a barrier, and if this function
1457 	 * ever gets inlined it will generate worse code.
1458 	 */
1459 	asm volatile ("");
1460 	return last;
1461 }
1462 
1463 static inline u64 vgettsc(cycle_t *cycle_now)
1464 {
1465 	long v;
1466 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1467 
1468 	*cycle_now = read_tsc();
1469 
1470 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1471 	return v * gtod->clock.mult;
1472 }
1473 
1474 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1475 {
1476 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1477 	unsigned long seq;
1478 	int mode;
1479 	u64 ns;
1480 
1481 	do {
1482 		seq = read_seqcount_begin(&gtod->seq);
1483 		mode = gtod->clock.vclock_mode;
1484 		ns = gtod->nsec_base;
1485 		ns += vgettsc(cycle_now);
1486 		ns >>= gtod->clock.shift;
1487 		ns += gtod->boot_ns;
1488 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1489 	*t = ns;
1490 
1491 	return mode;
1492 }
1493 
1494 /* returns true if host is using tsc clocksource */
1495 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1496 {
1497 	/* checked again under seqlock below */
1498 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1499 		return false;
1500 
1501 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1502 }
1503 #endif
1504 
1505 /*
1506  *
1507  * Assuming a stable TSC across physical CPUS, and a stable TSC
1508  * across virtual CPUs, the following condition is possible.
1509  * Each numbered line represents an event visible to both
1510  * CPUs at the next numbered event.
1511  *
1512  * "timespecX" represents host monotonic time. "tscX" represents
1513  * RDTSC value.
1514  *
1515  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1516  *
1517  * 1.  read timespec0,tsc0
1518  * 2.					| timespec1 = timespec0 + N
1519  * 					| tsc1 = tsc0 + M
1520  * 3. transition to guest		| transition to guest
1521  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1522  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1523  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1524  *
1525  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1526  *
1527  * 	- ret0 < ret1
1528  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1529  *		...
1530  *	- 0 < N - M => M < N
1531  *
1532  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1533  * always the case (the difference between two distinct xtime instances
1534  * might be smaller then the difference between corresponding TSC reads,
1535  * when updating guest vcpus pvclock areas).
1536  *
1537  * To avoid that problem, do not allow visibility of distinct
1538  * system_timestamp/tsc_timestamp values simultaneously: use a master
1539  * copy of host monotonic time values. Update that master copy
1540  * in lockstep.
1541  *
1542  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1543  *
1544  */
1545 
1546 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1547 {
1548 #ifdef CONFIG_X86_64
1549 	struct kvm_arch *ka = &kvm->arch;
1550 	int vclock_mode;
1551 	bool host_tsc_clocksource, vcpus_matched;
1552 
1553 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1554 			atomic_read(&kvm->online_vcpus));
1555 
1556 	/*
1557 	 * If the host uses TSC clock, then passthrough TSC as stable
1558 	 * to the guest.
1559 	 */
1560 	host_tsc_clocksource = kvm_get_time_and_clockread(
1561 					&ka->master_kernel_ns,
1562 					&ka->master_cycle_now);
1563 
1564 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1565 				&& !backwards_tsc_observed
1566 				&& !ka->boot_vcpu_runs_old_kvmclock;
1567 
1568 	if (ka->use_master_clock)
1569 		atomic_set(&kvm_guest_has_master_clock, 1);
1570 
1571 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1572 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1573 					vcpus_matched);
1574 #endif
1575 }
1576 
1577 static void kvm_gen_update_masterclock(struct kvm *kvm)
1578 {
1579 #ifdef CONFIG_X86_64
1580 	int i;
1581 	struct kvm_vcpu *vcpu;
1582 	struct kvm_arch *ka = &kvm->arch;
1583 
1584 	spin_lock(&ka->pvclock_gtod_sync_lock);
1585 	kvm_make_mclock_inprogress_request(kvm);
1586 	/* no guest entries from this point */
1587 	pvclock_update_vm_gtod_copy(kvm);
1588 
1589 	kvm_for_each_vcpu(i, vcpu, kvm)
1590 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1591 
1592 	/* guest entries allowed */
1593 	kvm_for_each_vcpu(i, vcpu, kvm)
1594 		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1595 
1596 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1597 #endif
1598 }
1599 
1600 static int kvm_guest_time_update(struct kvm_vcpu *v)
1601 {
1602 	unsigned long flags, this_tsc_khz;
1603 	struct kvm_vcpu_arch *vcpu = &v->arch;
1604 	struct kvm_arch *ka = &v->kvm->arch;
1605 	s64 kernel_ns;
1606 	u64 tsc_timestamp, host_tsc;
1607 	struct pvclock_vcpu_time_info guest_hv_clock;
1608 	u8 pvclock_flags;
1609 	bool use_master_clock;
1610 
1611 	kernel_ns = 0;
1612 	host_tsc = 0;
1613 
1614 	/*
1615 	 * If the host uses TSC clock, then passthrough TSC as stable
1616 	 * to the guest.
1617 	 */
1618 	spin_lock(&ka->pvclock_gtod_sync_lock);
1619 	use_master_clock = ka->use_master_clock;
1620 	if (use_master_clock) {
1621 		host_tsc = ka->master_cycle_now;
1622 		kernel_ns = ka->master_kernel_ns;
1623 	}
1624 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1625 
1626 	/* Keep irq disabled to prevent changes to the clock */
1627 	local_irq_save(flags);
1628 	this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1629 	if (unlikely(this_tsc_khz == 0)) {
1630 		local_irq_restore(flags);
1631 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1632 		return 1;
1633 	}
1634 	if (!use_master_clock) {
1635 		host_tsc = rdtsc();
1636 		kernel_ns = get_kernel_ns();
1637 	}
1638 
1639 	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1640 
1641 	/*
1642 	 * We may have to catch up the TSC to match elapsed wall clock
1643 	 * time for two reasons, even if kvmclock is used.
1644 	 *   1) CPU could have been running below the maximum TSC rate
1645 	 *   2) Broken TSC compensation resets the base at each VCPU
1646 	 *      entry to avoid unknown leaps of TSC even when running
1647 	 *      again on the same CPU.  This may cause apparent elapsed
1648 	 *      time to disappear, and the guest to stand still or run
1649 	 *	very slowly.
1650 	 */
1651 	if (vcpu->tsc_catchup) {
1652 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1653 		if (tsc > tsc_timestamp) {
1654 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1655 			tsc_timestamp = tsc;
1656 		}
1657 	}
1658 
1659 	local_irq_restore(flags);
1660 
1661 	if (!vcpu->pv_time_enabled)
1662 		return 0;
1663 
1664 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1665 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1666 				   &vcpu->hv_clock.tsc_shift,
1667 				   &vcpu->hv_clock.tsc_to_system_mul);
1668 		vcpu->hw_tsc_khz = this_tsc_khz;
1669 	}
1670 
1671 	/* With all the info we got, fill in the values */
1672 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1673 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1674 	vcpu->last_guest_tsc = tsc_timestamp;
1675 
1676 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1677 		&guest_hv_clock, sizeof(guest_hv_clock))))
1678 		return 0;
1679 
1680 	/* This VCPU is paused, but it's legal for a guest to read another
1681 	 * VCPU's kvmclock, so we really have to follow the specification where
1682 	 * it says that version is odd if data is being modified, and even after
1683 	 * it is consistent.
1684 	 *
1685 	 * Version field updates must be kept separate.  This is because
1686 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1687 	 * writes within a string instruction are weakly ordered.  So there
1688 	 * are three writes overall.
1689 	 *
1690 	 * As a small optimization, only write the version field in the first
1691 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1692 	 * version field is the first in the struct.
1693 	 */
1694 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1695 
1696 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1697 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1698 				&vcpu->hv_clock,
1699 				sizeof(vcpu->hv_clock.version));
1700 
1701 	smp_wmb();
1702 
1703 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1704 	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1705 
1706 	if (vcpu->pvclock_set_guest_stopped_request) {
1707 		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1708 		vcpu->pvclock_set_guest_stopped_request = false;
1709 	}
1710 
1711 	/* If the host uses TSC clocksource, then it is stable */
1712 	if (use_master_clock)
1713 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1714 
1715 	vcpu->hv_clock.flags = pvclock_flags;
1716 
1717 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1718 
1719 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1720 				&vcpu->hv_clock,
1721 				sizeof(vcpu->hv_clock));
1722 
1723 	smp_wmb();
1724 
1725 	vcpu->hv_clock.version++;
1726 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1727 				&vcpu->hv_clock,
1728 				sizeof(vcpu->hv_clock.version));
1729 	return 0;
1730 }
1731 
1732 /*
1733  * kvmclock updates which are isolated to a given vcpu, such as
1734  * vcpu->cpu migration, should not allow system_timestamp from
1735  * the rest of the vcpus to remain static. Otherwise ntp frequency
1736  * correction applies to one vcpu's system_timestamp but not
1737  * the others.
1738  *
1739  * So in those cases, request a kvmclock update for all vcpus.
1740  * We need to rate-limit these requests though, as they can
1741  * considerably slow guests that have a large number of vcpus.
1742  * The time for a remote vcpu to update its kvmclock is bound
1743  * by the delay we use to rate-limit the updates.
1744  */
1745 
1746 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1747 
1748 static void kvmclock_update_fn(struct work_struct *work)
1749 {
1750 	int i;
1751 	struct delayed_work *dwork = to_delayed_work(work);
1752 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1753 					   kvmclock_update_work);
1754 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1755 	struct kvm_vcpu *vcpu;
1756 
1757 	kvm_for_each_vcpu(i, vcpu, kvm) {
1758 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1759 		kvm_vcpu_kick(vcpu);
1760 	}
1761 }
1762 
1763 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1764 {
1765 	struct kvm *kvm = v->kvm;
1766 
1767 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1768 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1769 					KVMCLOCK_UPDATE_DELAY);
1770 }
1771 
1772 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1773 
1774 static void kvmclock_sync_fn(struct work_struct *work)
1775 {
1776 	struct delayed_work *dwork = to_delayed_work(work);
1777 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1778 					   kvmclock_sync_work);
1779 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1780 
1781 	if (!kvmclock_periodic_sync)
1782 		return;
1783 
1784 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1785 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1786 					KVMCLOCK_SYNC_PERIOD);
1787 }
1788 
1789 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1790 {
1791 	u64 mcg_cap = vcpu->arch.mcg_cap;
1792 	unsigned bank_num = mcg_cap & 0xff;
1793 
1794 	switch (msr) {
1795 	case MSR_IA32_MCG_STATUS:
1796 		vcpu->arch.mcg_status = data;
1797 		break;
1798 	case MSR_IA32_MCG_CTL:
1799 		if (!(mcg_cap & MCG_CTL_P))
1800 			return 1;
1801 		if (data != 0 && data != ~(u64)0)
1802 			return -1;
1803 		vcpu->arch.mcg_ctl = data;
1804 		break;
1805 	default:
1806 		if (msr >= MSR_IA32_MC0_CTL &&
1807 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
1808 			u32 offset = msr - MSR_IA32_MC0_CTL;
1809 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1810 			 * some Linux kernels though clear bit 10 in bank 4 to
1811 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1812 			 * this to avoid an uncatched #GP in the guest
1813 			 */
1814 			if ((offset & 0x3) == 0 &&
1815 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1816 				return -1;
1817 			vcpu->arch.mce_banks[offset] = data;
1818 			break;
1819 		}
1820 		return 1;
1821 	}
1822 	return 0;
1823 }
1824 
1825 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1826 {
1827 	struct kvm *kvm = vcpu->kvm;
1828 	int lm = is_long_mode(vcpu);
1829 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1830 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1831 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1832 		: kvm->arch.xen_hvm_config.blob_size_32;
1833 	u32 page_num = data & ~PAGE_MASK;
1834 	u64 page_addr = data & PAGE_MASK;
1835 	u8 *page;
1836 	int r;
1837 
1838 	r = -E2BIG;
1839 	if (page_num >= blob_size)
1840 		goto out;
1841 	r = -ENOMEM;
1842 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1843 	if (IS_ERR(page)) {
1844 		r = PTR_ERR(page);
1845 		goto out;
1846 	}
1847 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1848 		goto out_free;
1849 	r = 0;
1850 out_free:
1851 	kfree(page);
1852 out:
1853 	return r;
1854 }
1855 
1856 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1857 {
1858 	gpa_t gpa = data & ~0x3f;
1859 
1860 	/* Bits 2:5 are reserved, Should be zero */
1861 	if (data & 0x3c)
1862 		return 1;
1863 
1864 	vcpu->arch.apf.msr_val = data;
1865 
1866 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1867 		kvm_clear_async_pf_completion_queue(vcpu);
1868 		kvm_async_pf_hash_reset(vcpu);
1869 		return 0;
1870 	}
1871 
1872 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1873 					sizeof(u32)))
1874 		return 1;
1875 
1876 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1877 	kvm_async_pf_wakeup_all(vcpu);
1878 	return 0;
1879 }
1880 
1881 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1882 {
1883 	vcpu->arch.pv_time_enabled = false;
1884 }
1885 
1886 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1887 {
1888 	u64 delta;
1889 
1890 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1891 		return;
1892 
1893 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1894 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
1895 	vcpu->arch.st.accum_steal = delta;
1896 }
1897 
1898 static void record_steal_time(struct kvm_vcpu *vcpu)
1899 {
1900 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1901 		return;
1902 
1903 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1904 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1905 		return;
1906 
1907 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1908 	vcpu->arch.st.steal.version += 2;
1909 	vcpu->arch.st.accum_steal = 0;
1910 
1911 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1912 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1913 }
1914 
1915 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1916 {
1917 	bool pr = false;
1918 	u32 msr = msr_info->index;
1919 	u64 data = msr_info->data;
1920 
1921 	switch (msr) {
1922 	case MSR_AMD64_NB_CFG:
1923 	case MSR_IA32_UCODE_REV:
1924 	case MSR_IA32_UCODE_WRITE:
1925 	case MSR_VM_HSAVE_PA:
1926 	case MSR_AMD64_PATCH_LOADER:
1927 	case MSR_AMD64_BU_CFG2:
1928 		break;
1929 
1930 	case MSR_EFER:
1931 		return set_efer(vcpu, data);
1932 	case MSR_K7_HWCR:
1933 		data &= ~(u64)0x40;	/* ignore flush filter disable */
1934 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
1935 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
1936 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
1937 		if (data != 0) {
1938 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1939 				    data);
1940 			return 1;
1941 		}
1942 		break;
1943 	case MSR_FAM10H_MMIO_CONF_BASE:
1944 		if (data != 0) {
1945 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1946 				    "0x%llx\n", data);
1947 			return 1;
1948 		}
1949 		break;
1950 	case MSR_IA32_DEBUGCTLMSR:
1951 		if (!data) {
1952 			/* We support the non-activated case already */
1953 			break;
1954 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1955 			/* Values other than LBR and BTF are vendor-specific,
1956 			   thus reserved and should throw a #GP */
1957 			return 1;
1958 		}
1959 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1960 			    __func__, data);
1961 		break;
1962 	case 0x200 ... 0x2ff:
1963 		return kvm_mtrr_set_msr(vcpu, msr, data);
1964 	case MSR_IA32_APICBASE:
1965 		return kvm_set_apic_base(vcpu, msr_info);
1966 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1967 		return kvm_x2apic_msr_write(vcpu, msr, data);
1968 	case MSR_IA32_TSCDEADLINE:
1969 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
1970 		break;
1971 	case MSR_IA32_TSC_ADJUST:
1972 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
1973 			if (!msr_info->host_initiated) {
1974 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1975 				adjust_tsc_offset_guest(vcpu, adj);
1976 			}
1977 			vcpu->arch.ia32_tsc_adjust_msr = data;
1978 		}
1979 		break;
1980 	case MSR_IA32_MISC_ENABLE:
1981 		vcpu->arch.ia32_misc_enable_msr = data;
1982 		break;
1983 	case MSR_IA32_SMBASE:
1984 		if (!msr_info->host_initiated)
1985 			return 1;
1986 		vcpu->arch.smbase = data;
1987 		break;
1988 	case MSR_KVM_WALL_CLOCK_NEW:
1989 	case MSR_KVM_WALL_CLOCK:
1990 		vcpu->kvm->arch.wall_clock = data;
1991 		kvm_write_wall_clock(vcpu->kvm, data);
1992 		break;
1993 	case MSR_KVM_SYSTEM_TIME_NEW:
1994 	case MSR_KVM_SYSTEM_TIME: {
1995 		u64 gpa_offset;
1996 		struct kvm_arch *ka = &vcpu->kvm->arch;
1997 
1998 		kvmclock_reset(vcpu);
1999 
2000 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2001 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2002 
2003 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2004 				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2005 					&vcpu->requests);
2006 
2007 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2008 		}
2009 
2010 		vcpu->arch.time = data;
2011 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2012 
2013 		/* we verify if the enable bit is set... */
2014 		if (!(data & 1))
2015 			break;
2016 
2017 		gpa_offset = data & ~(PAGE_MASK | 1);
2018 
2019 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2020 		     &vcpu->arch.pv_time, data & ~1ULL,
2021 		     sizeof(struct pvclock_vcpu_time_info)))
2022 			vcpu->arch.pv_time_enabled = false;
2023 		else
2024 			vcpu->arch.pv_time_enabled = true;
2025 
2026 		break;
2027 	}
2028 	case MSR_KVM_ASYNC_PF_EN:
2029 		if (kvm_pv_enable_async_pf(vcpu, data))
2030 			return 1;
2031 		break;
2032 	case MSR_KVM_STEAL_TIME:
2033 
2034 		if (unlikely(!sched_info_on()))
2035 			return 1;
2036 
2037 		if (data & KVM_STEAL_RESERVED_MASK)
2038 			return 1;
2039 
2040 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2041 						data & KVM_STEAL_VALID_BITS,
2042 						sizeof(struct kvm_steal_time)))
2043 			return 1;
2044 
2045 		vcpu->arch.st.msr_val = data;
2046 
2047 		if (!(data & KVM_MSR_ENABLED))
2048 			break;
2049 
2050 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
2051 
2052 		preempt_disable();
2053 		accumulate_steal_time(vcpu);
2054 		preempt_enable();
2055 
2056 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2057 
2058 		break;
2059 	case MSR_KVM_PV_EOI_EN:
2060 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2061 			return 1;
2062 		break;
2063 
2064 	case MSR_IA32_MCG_CTL:
2065 	case MSR_IA32_MCG_STATUS:
2066 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2067 		return set_msr_mce(vcpu, msr, data);
2068 
2069 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2070 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2071 		pr = true; /* fall through */
2072 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2073 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2074 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2075 			return kvm_pmu_set_msr(vcpu, msr_info);
2076 
2077 		if (pr || data != 0)
2078 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2079 				    "0x%x data 0x%llx\n", msr, data);
2080 		break;
2081 	case MSR_K7_CLK_CTL:
2082 		/*
2083 		 * Ignore all writes to this no longer documented MSR.
2084 		 * Writes are only relevant for old K7 processors,
2085 		 * all pre-dating SVM, but a recommended workaround from
2086 		 * AMD for these chips. It is possible to specify the
2087 		 * affected processor models on the command line, hence
2088 		 * the need to ignore the workaround.
2089 		 */
2090 		break;
2091 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2092 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2093 	case HV_X64_MSR_CRASH_CTL:
2094 		return kvm_hv_set_msr_common(vcpu, msr, data,
2095 					     msr_info->host_initiated);
2096 	case MSR_IA32_BBL_CR_CTL3:
2097 		/* Drop writes to this legacy MSR -- see rdmsr
2098 		 * counterpart for further detail.
2099 		 */
2100 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2101 		break;
2102 	case MSR_AMD64_OSVW_ID_LENGTH:
2103 		if (!guest_cpuid_has_osvw(vcpu))
2104 			return 1;
2105 		vcpu->arch.osvw.length = data;
2106 		break;
2107 	case MSR_AMD64_OSVW_STATUS:
2108 		if (!guest_cpuid_has_osvw(vcpu))
2109 			return 1;
2110 		vcpu->arch.osvw.status = data;
2111 		break;
2112 	default:
2113 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2114 			return xen_hvm_config(vcpu, data);
2115 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2116 			return kvm_pmu_set_msr(vcpu, msr_info);
2117 		if (!ignore_msrs) {
2118 			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2119 				    msr, data);
2120 			return 1;
2121 		} else {
2122 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2123 				    msr, data);
2124 			break;
2125 		}
2126 	}
2127 	return 0;
2128 }
2129 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2130 
2131 
2132 /*
2133  * Reads an msr value (of 'msr_index') into 'pdata'.
2134  * Returns 0 on success, non-0 otherwise.
2135  * Assumes vcpu_load() was already called.
2136  */
2137 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2138 {
2139 	return kvm_x86_ops->get_msr(vcpu, msr);
2140 }
2141 EXPORT_SYMBOL_GPL(kvm_get_msr);
2142 
2143 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2144 {
2145 	u64 data;
2146 	u64 mcg_cap = vcpu->arch.mcg_cap;
2147 	unsigned bank_num = mcg_cap & 0xff;
2148 
2149 	switch (msr) {
2150 	case MSR_IA32_P5_MC_ADDR:
2151 	case MSR_IA32_P5_MC_TYPE:
2152 		data = 0;
2153 		break;
2154 	case MSR_IA32_MCG_CAP:
2155 		data = vcpu->arch.mcg_cap;
2156 		break;
2157 	case MSR_IA32_MCG_CTL:
2158 		if (!(mcg_cap & MCG_CTL_P))
2159 			return 1;
2160 		data = vcpu->arch.mcg_ctl;
2161 		break;
2162 	case MSR_IA32_MCG_STATUS:
2163 		data = vcpu->arch.mcg_status;
2164 		break;
2165 	default:
2166 		if (msr >= MSR_IA32_MC0_CTL &&
2167 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2168 			u32 offset = msr - MSR_IA32_MC0_CTL;
2169 			data = vcpu->arch.mce_banks[offset];
2170 			break;
2171 		}
2172 		return 1;
2173 	}
2174 	*pdata = data;
2175 	return 0;
2176 }
2177 
2178 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2179 {
2180 	switch (msr_info->index) {
2181 	case MSR_IA32_PLATFORM_ID:
2182 	case MSR_IA32_EBL_CR_POWERON:
2183 	case MSR_IA32_DEBUGCTLMSR:
2184 	case MSR_IA32_LASTBRANCHFROMIP:
2185 	case MSR_IA32_LASTBRANCHTOIP:
2186 	case MSR_IA32_LASTINTFROMIP:
2187 	case MSR_IA32_LASTINTTOIP:
2188 	case MSR_K8_SYSCFG:
2189 	case MSR_K8_TSEG_ADDR:
2190 	case MSR_K8_TSEG_MASK:
2191 	case MSR_K7_HWCR:
2192 	case MSR_VM_HSAVE_PA:
2193 	case MSR_K8_INT_PENDING_MSG:
2194 	case MSR_AMD64_NB_CFG:
2195 	case MSR_FAM10H_MMIO_CONF_BASE:
2196 	case MSR_AMD64_BU_CFG2:
2197 		msr_info->data = 0;
2198 		break;
2199 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2200 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2201 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2202 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2203 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2204 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2205 		msr_info->data = 0;
2206 		break;
2207 	case MSR_IA32_UCODE_REV:
2208 		msr_info->data = 0x100000000ULL;
2209 		break;
2210 	case MSR_MTRRcap:
2211 	case 0x200 ... 0x2ff:
2212 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2213 	case 0xcd: /* fsb frequency */
2214 		msr_info->data = 3;
2215 		break;
2216 		/*
2217 		 * MSR_EBC_FREQUENCY_ID
2218 		 * Conservative value valid for even the basic CPU models.
2219 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2220 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2221 		 * and 266MHz for model 3, or 4. Set Core Clock
2222 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2223 		 * 31:24) even though these are only valid for CPU
2224 		 * models > 2, however guests may end up dividing or
2225 		 * multiplying by zero otherwise.
2226 		 */
2227 	case MSR_EBC_FREQUENCY_ID:
2228 		msr_info->data = 1 << 24;
2229 		break;
2230 	case MSR_IA32_APICBASE:
2231 		msr_info->data = kvm_get_apic_base(vcpu);
2232 		break;
2233 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2234 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2235 		break;
2236 	case MSR_IA32_TSCDEADLINE:
2237 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2238 		break;
2239 	case MSR_IA32_TSC_ADJUST:
2240 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2241 		break;
2242 	case MSR_IA32_MISC_ENABLE:
2243 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2244 		break;
2245 	case MSR_IA32_SMBASE:
2246 		if (!msr_info->host_initiated)
2247 			return 1;
2248 		msr_info->data = vcpu->arch.smbase;
2249 		break;
2250 	case MSR_IA32_PERF_STATUS:
2251 		/* TSC increment by tick */
2252 		msr_info->data = 1000ULL;
2253 		/* CPU multiplier */
2254 		msr_info->data |= (((uint64_t)4ULL) << 40);
2255 		break;
2256 	case MSR_EFER:
2257 		msr_info->data = vcpu->arch.efer;
2258 		break;
2259 	case MSR_KVM_WALL_CLOCK:
2260 	case MSR_KVM_WALL_CLOCK_NEW:
2261 		msr_info->data = vcpu->kvm->arch.wall_clock;
2262 		break;
2263 	case MSR_KVM_SYSTEM_TIME:
2264 	case MSR_KVM_SYSTEM_TIME_NEW:
2265 		msr_info->data = vcpu->arch.time;
2266 		break;
2267 	case MSR_KVM_ASYNC_PF_EN:
2268 		msr_info->data = vcpu->arch.apf.msr_val;
2269 		break;
2270 	case MSR_KVM_STEAL_TIME:
2271 		msr_info->data = vcpu->arch.st.msr_val;
2272 		break;
2273 	case MSR_KVM_PV_EOI_EN:
2274 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2275 		break;
2276 	case MSR_IA32_P5_MC_ADDR:
2277 	case MSR_IA32_P5_MC_TYPE:
2278 	case MSR_IA32_MCG_CAP:
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 get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2283 	case MSR_K7_CLK_CTL:
2284 		/*
2285 		 * Provide expected ramp-up count for K7. All other
2286 		 * are set to zero, indicating minimum divisors for
2287 		 * every field.
2288 		 *
2289 		 * This prevents guest kernels on AMD host with CPU
2290 		 * type 6, model 8 and higher from exploding due to
2291 		 * the rdmsr failing.
2292 		 */
2293 		msr_info->data = 0x20000000;
2294 		break;
2295 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2296 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2297 	case HV_X64_MSR_CRASH_CTL:
2298 		return kvm_hv_get_msr_common(vcpu,
2299 					     msr_info->index, &msr_info->data);
2300 		break;
2301 	case MSR_IA32_BBL_CR_CTL3:
2302 		/* This legacy MSR exists but isn't fully documented in current
2303 		 * silicon.  It is however accessed by winxp in very narrow
2304 		 * scenarios where it sets bit #19, itself documented as
2305 		 * a "reserved" bit.  Best effort attempt to source coherent
2306 		 * read data here should the balance of the register be
2307 		 * interpreted by the guest:
2308 		 *
2309 		 * L2 cache control register 3: 64GB range, 256KB size,
2310 		 * enabled, latency 0x1, configured
2311 		 */
2312 		msr_info->data = 0xbe702111;
2313 		break;
2314 	case MSR_AMD64_OSVW_ID_LENGTH:
2315 		if (!guest_cpuid_has_osvw(vcpu))
2316 			return 1;
2317 		msr_info->data = vcpu->arch.osvw.length;
2318 		break;
2319 	case MSR_AMD64_OSVW_STATUS:
2320 		if (!guest_cpuid_has_osvw(vcpu))
2321 			return 1;
2322 		msr_info->data = vcpu->arch.osvw.status;
2323 		break;
2324 	default:
2325 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2326 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2327 		if (!ignore_msrs) {
2328 			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2329 			return 1;
2330 		} else {
2331 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2332 			msr_info->data = 0;
2333 		}
2334 		break;
2335 	}
2336 	return 0;
2337 }
2338 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2339 
2340 /*
2341  * Read or write a bunch of msrs. All parameters are kernel addresses.
2342  *
2343  * @return number of msrs set successfully.
2344  */
2345 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2346 		    struct kvm_msr_entry *entries,
2347 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2348 				  unsigned index, u64 *data))
2349 {
2350 	int i, idx;
2351 
2352 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2353 	for (i = 0; i < msrs->nmsrs; ++i)
2354 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2355 			break;
2356 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2357 
2358 	return i;
2359 }
2360 
2361 /*
2362  * Read or write a bunch of msrs. Parameters are user addresses.
2363  *
2364  * @return number of msrs set successfully.
2365  */
2366 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2367 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2368 				unsigned index, u64 *data),
2369 		  int writeback)
2370 {
2371 	struct kvm_msrs msrs;
2372 	struct kvm_msr_entry *entries;
2373 	int r, n;
2374 	unsigned size;
2375 
2376 	r = -EFAULT;
2377 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2378 		goto out;
2379 
2380 	r = -E2BIG;
2381 	if (msrs.nmsrs >= MAX_IO_MSRS)
2382 		goto out;
2383 
2384 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2385 	entries = memdup_user(user_msrs->entries, size);
2386 	if (IS_ERR(entries)) {
2387 		r = PTR_ERR(entries);
2388 		goto out;
2389 	}
2390 
2391 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2392 	if (r < 0)
2393 		goto out_free;
2394 
2395 	r = -EFAULT;
2396 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2397 		goto out_free;
2398 
2399 	r = n;
2400 
2401 out_free:
2402 	kfree(entries);
2403 out:
2404 	return r;
2405 }
2406 
2407 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2408 {
2409 	int r;
2410 
2411 	switch (ext) {
2412 	case KVM_CAP_IRQCHIP:
2413 	case KVM_CAP_HLT:
2414 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2415 	case KVM_CAP_SET_TSS_ADDR:
2416 	case KVM_CAP_EXT_CPUID:
2417 	case KVM_CAP_EXT_EMUL_CPUID:
2418 	case KVM_CAP_CLOCKSOURCE:
2419 	case KVM_CAP_PIT:
2420 	case KVM_CAP_NOP_IO_DELAY:
2421 	case KVM_CAP_MP_STATE:
2422 	case KVM_CAP_SYNC_MMU:
2423 	case KVM_CAP_USER_NMI:
2424 	case KVM_CAP_REINJECT_CONTROL:
2425 	case KVM_CAP_IRQ_INJECT_STATUS:
2426 	case KVM_CAP_IOEVENTFD:
2427 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2428 	case KVM_CAP_PIT2:
2429 	case KVM_CAP_PIT_STATE2:
2430 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2431 	case KVM_CAP_XEN_HVM:
2432 	case KVM_CAP_ADJUST_CLOCK:
2433 	case KVM_CAP_VCPU_EVENTS:
2434 	case KVM_CAP_HYPERV:
2435 	case KVM_CAP_HYPERV_VAPIC:
2436 	case KVM_CAP_HYPERV_SPIN:
2437 	case KVM_CAP_PCI_SEGMENT:
2438 	case KVM_CAP_DEBUGREGS:
2439 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2440 	case KVM_CAP_XSAVE:
2441 	case KVM_CAP_ASYNC_PF:
2442 	case KVM_CAP_GET_TSC_KHZ:
2443 	case KVM_CAP_KVMCLOCK_CTRL:
2444 	case KVM_CAP_READONLY_MEM:
2445 	case KVM_CAP_HYPERV_TIME:
2446 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2447 	case KVM_CAP_TSC_DEADLINE_TIMER:
2448 	case KVM_CAP_ENABLE_CAP_VM:
2449 	case KVM_CAP_DISABLE_QUIRKS:
2450 	case KVM_CAP_SET_BOOT_CPU_ID:
2451 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2452 	case KVM_CAP_ASSIGN_DEV_IRQ:
2453 	case KVM_CAP_PCI_2_3:
2454 #endif
2455 		r = 1;
2456 		break;
2457 	case KVM_CAP_X86_SMM:
2458 		/* SMBASE is usually relocated above 1M on modern chipsets,
2459 		 * and SMM handlers might indeed rely on 4G segment limits,
2460 		 * so do not report SMM to be available if real mode is
2461 		 * emulated via vm86 mode.  Still, do not go to great lengths
2462 		 * to avoid userspace's usage of the feature, because it is a
2463 		 * fringe case that is not enabled except via specific settings
2464 		 * of the module parameters.
2465 		 */
2466 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2467 		break;
2468 	case KVM_CAP_COALESCED_MMIO:
2469 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2470 		break;
2471 	case KVM_CAP_VAPIC:
2472 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2473 		break;
2474 	case KVM_CAP_NR_VCPUS:
2475 		r = KVM_SOFT_MAX_VCPUS;
2476 		break;
2477 	case KVM_CAP_MAX_VCPUS:
2478 		r = KVM_MAX_VCPUS;
2479 		break;
2480 	case KVM_CAP_NR_MEMSLOTS:
2481 		r = KVM_USER_MEM_SLOTS;
2482 		break;
2483 	case KVM_CAP_PV_MMU:	/* obsolete */
2484 		r = 0;
2485 		break;
2486 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2487 	case KVM_CAP_IOMMU:
2488 		r = iommu_present(&pci_bus_type);
2489 		break;
2490 #endif
2491 	case KVM_CAP_MCE:
2492 		r = KVM_MAX_MCE_BANKS;
2493 		break;
2494 	case KVM_CAP_XCRS:
2495 		r = cpu_has_xsave;
2496 		break;
2497 	case KVM_CAP_TSC_CONTROL:
2498 		r = kvm_has_tsc_control;
2499 		break;
2500 	default:
2501 		r = 0;
2502 		break;
2503 	}
2504 	return r;
2505 
2506 }
2507 
2508 long kvm_arch_dev_ioctl(struct file *filp,
2509 			unsigned int ioctl, unsigned long arg)
2510 {
2511 	void __user *argp = (void __user *)arg;
2512 	long r;
2513 
2514 	switch (ioctl) {
2515 	case KVM_GET_MSR_INDEX_LIST: {
2516 		struct kvm_msr_list __user *user_msr_list = argp;
2517 		struct kvm_msr_list msr_list;
2518 		unsigned n;
2519 
2520 		r = -EFAULT;
2521 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2522 			goto out;
2523 		n = msr_list.nmsrs;
2524 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2525 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2526 			goto out;
2527 		r = -E2BIG;
2528 		if (n < msr_list.nmsrs)
2529 			goto out;
2530 		r = -EFAULT;
2531 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2532 				 num_msrs_to_save * sizeof(u32)))
2533 			goto out;
2534 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2535 				 &emulated_msrs,
2536 				 num_emulated_msrs * sizeof(u32)))
2537 			goto out;
2538 		r = 0;
2539 		break;
2540 	}
2541 	case KVM_GET_SUPPORTED_CPUID:
2542 	case KVM_GET_EMULATED_CPUID: {
2543 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2544 		struct kvm_cpuid2 cpuid;
2545 
2546 		r = -EFAULT;
2547 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2548 			goto out;
2549 
2550 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2551 					    ioctl);
2552 		if (r)
2553 			goto out;
2554 
2555 		r = -EFAULT;
2556 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2557 			goto out;
2558 		r = 0;
2559 		break;
2560 	}
2561 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2562 		u64 mce_cap;
2563 
2564 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2565 		r = -EFAULT;
2566 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2567 			goto out;
2568 		r = 0;
2569 		break;
2570 	}
2571 	default:
2572 		r = -EINVAL;
2573 	}
2574 out:
2575 	return r;
2576 }
2577 
2578 static void wbinvd_ipi(void *garbage)
2579 {
2580 	wbinvd();
2581 }
2582 
2583 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2584 {
2585 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2586 }
2587 
2588 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2589 {
2590 	/* Address WBINVD may be executed by guest */
2591 	if (need_emulate_wbinvd(vcpu)) {
2592 		if (kvm_x86_ops->has_wbinvd_exit())
2593 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2594 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2595 			smp_call_function_single(vcpu->cpu,
2596 					wbinvd_ipi, NULL, 1);
2597 	}
2598 
2599 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2600 
2601 	/* Apply any externally detected TSC adjustments (due to suspend) */
2602 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2603 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2604 		vcpu->arch.tsc_offset_adjustment = 0;
2605 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2606 	}
2607 
2608 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2609 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2610 				rdtsc() - vcpu->arch.last_host_tsc;
2611 		if (tsc_delta < 0)
2612 			mark_tsc_unstable("KVM discovered backwards TSC");
2613 		if (check_tsc_unstable()) {
2614 			u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2615 						vcpu->arch.last_guest_tsc);
2616 			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2617 			vcpu->arch.tsc_catchup = 1;
2618 		}
2619 		/*
2620 		 * On a host with synchronized TSC, there is no need to update
2621 		 * kvmclock on vcpu->cpu migration
2622 		 */
2623 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2624 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2625 		if (vcpu->cpu != cpu)
2626 			kvm_migrate_timers(vcpu);
2627 		vcpu->cpu = cpu;
2628 	}
2629 
2630 	accumulate_steal_time(vcpu);
2631 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2632 }
2633 
2634 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2635 {
2636 	kvm_x86_ops->vcpu_put(vcpu);
2637 	kvm_put_guest_fpu(vcpu);
2638 	vcpu->arch.last_host_tsc = rdtsc();
2639 }
2640 
2641 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2642 				    struct kvm_lapic_state *s)
2643 {
2644 	kvm_x86_ops->sync_pir_to_irr(vcpu);
2645 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2646 
2647 	return 0;
2648 }
2649 
2650 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2651 				    struct kvm_lapic_state *s)
2652 {
2653 	kvm_apic_post_state_restore(vcpu, s);
2654 	update_cr8_intercept(vcpu);
2655 
2656 	return 0;
2657 }
2658 
2659 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2660 				    struct kvm_interrupt *irq)
2661 {
2662 	if (irq->irq >= KVM_NR_INTERRUPTS)
2663 		return -EINVAL;
2664 	if (irqchip_in_kernel(vcpu->kvm))
2665 		return -ENXIO;
2666 
2667 	kvm_queue_interrupt(vcpu, irq->irq, false);
2668 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2669 
2670 	return 0;
2671 }
2672 
2673 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2674 {
2675 	kvm_inject_nmi(vcpu);
2676 
2677 	return 0;
2678 }
2679 
2680 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2681 {
2682 	kvm_make_request(KVM_REQ_SMI, vcpu);
2683 
2684 	return 0;
2685 }
2686 
2687 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2688 					   struct kvm_tpr_access_ctl *tac)
2689 {
2690 	if (tac->flags)
2691 		return -EINVAL;
2692 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2693 	return 0;
2694 }
2695 
2696 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2697 					u64 mcg_cap)
2698 {
2699 	int r;
2700 	unsigned bank_num = mcg_cap & 0xff, bank;
2701 
2702 	r = -EINVAL;
2703 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2704 		goto out;
2705 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2706 		goto out;
2707 	r = 0;
2708 	vcpu->arch.mcg_cap = mcg_cap;
2709 	/* Init IA32_MCG_CTL to all 1s */
2710 	if (mcg_cap & MCG_CTL_P)
2711 		vcpu->arch.mcg_ctl = ~(u64)0;
2712 	/* Init IA32_MCi_CTL to all 1s */
2713 	for (bank = 0; bank < bank_num; bank++)
2714 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2715 out:
2716 	return r;
2717 }
2718 
2719 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2720 				      struct kvm_x86_mce *mce)
2721 {
2722 	u64 mcg_cap = vcpu->arch.mcg_cap;
2723 	unsigned bank_num = mcg_cap & 0xff;
2724 	u64 *banks = vcpu->arch.mce_banks;
2725 
2726 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2727 		return -EINVAL;
2728 	/*
2729 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2730 	 * reporting is disabled
2731 	 */
2732 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2733 	    vcpu->arch.mcg_ctl != ~(u64)0)
2734 		return 0;
2735 	banks += 4 * mce->bank;
2736 	/*
2737 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2738 	 * reporting is disabled for the bank
2739 	 */
2740 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2741 		return 0;
2742 	if (mce->status & MCI_STATUS_UC) {
2743 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2744 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2745 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2746 			return 0;
2747 		}
2748 		if (banks[1] & MCI_STATUS_VAL)
2749 			mce->status |= MCI_STATUS_OVER;
2750 		banks[2] = mce->addr;
2751 		banks[3] = mce->misc;
2752 		vcpu->arch.mcg_status = mce->mcg_status;
2753 		banks[1] = mce->status;
2754 		kvm_queue_exception(vcpu, MC_VECTOR);
2755 	} else if (!(banks[1] & MCI_STATUS_VAL)
2756 		   || !(banks[1] & MCI_STATUS_UC)) {
2757 		if (banks[1] & MCI_STATUS_VAL)
2758 			mce->status |= MCI_STATUS_OVER;
2759 		banks[2] = mce->addr;
2760 		banks[3] = mce->misc;
2761 		banks[1] = mce->status;
2762 	} else
2763 		banks[1] |= MCI_STATUS_OVER;
2764 	return 0;
2765 }
2766 
2767 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2768 					       struct kvm_vcpu_events *events)
2769 {
2770 	process_nmi(vcpu);
2771 	events->exception.injected =
2772 		vcpu->arch.exception.pending &&
2773 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2774 	events->exception.nr = vcpu->arch.exception.nr;
2775 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2776 	events->exception.pad = 0;
2777 	events->exception.error_code = vcpu->arch.exception.error_code;
2778 
2779 	events->interrupt.injected =
2780 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2781 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2782 	events->interrupt.soft = 0;
2783 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2784 
2785 	events->nmi.injected = vcpu->arch.nmi_injected;
2786 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
2787 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2788 	events->nmi.pad = 0;
2789 
2790 	events->sipi_vector = 0; /* never valid when reporting to user space */
2791 
2792 	events->smi.smm = is_smm(vcpu);
2793 	events->smi.pending = vcpu->arch.smi_pending;
2794 	events->smi.smm_inside_nmi =
2795 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2796 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2797 
2798 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2799 			 | KVM_VCPUEVENT_VALID_SHADOW
2800 			 | KVM_VCPUEVENT_VALID_SMM);
2801 	memset(&events->reserved, 0, sizeof(events->reserved));
2802 }
2803 
2804 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2805 					      struct kvm_vcpu_events *events)
2806 {
2807 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2808 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2809 			      | KVM_VCPUEVENT_VALID_SHADOW
2810 			      | KVM_VCPUEVENT_VALID_SMM))
2811 		return -EINVAL;
2812 
2813 	process_nmi(vcpu);
2814 	vcpu->arch.exception.pending = events->exception.injected;
2815 	vcpu->arch.exception.nr = events->exception.nr;
2816 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2817 	vcpu->arch.exception.error_code = events->exception.error_code;
2818 
2819 	vcpu->arch.interrupt.pending = events->interrupt.injected;
2820 	vcpu->arch.interrupt.nr = events->interrupt.nr;
2821 	vcpu->arch.interrupt.soft = events->interrupt.soft;
2822 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2823 		kvm_x86_ops->set_interrupt_shadow(vcpu,
2824 						  events->interrupt.shadow);
2825 
2826 	vcpu->arch.nmi_injected = events->nmi.injected;
2827 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2828 		vcpu->arch.nmi_pending = events->nmi.pending;
2829 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2830 
2831 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2832 	    kvm_vcpu_has_lapic(vcpu))
2833 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
2834 
2835 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2836 		if (events->smi.smm)
2837 			vcpu->arch.hflags |= HF_SMM_MASK;
2838 		else
2839 			vcpu->arch.hflags &= ~HF_SMM_MASK;
2840 		vcpu->arch.smi_pending = events->smi.pending;
2841 		if (events->smi.smm_inside_nmi)
2842 			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2843 		else
2844 			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2845 		if (kvm_vcpu_has_lapic(vcpu)) {
2846 			if (events->smi.latched_init)
2847 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2848 			else
2849 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2850 		}
2851 	}
2852 
2853 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2854 
2855 	return 0;
2856 }
2857 
2858 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2859 					     struct kvm_debugregs *dbgregs)
2860 {
2861 	unsigned long val;
2862 
2863 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2864 	kvm_get_dr(vcpu, 6, &val);
2865 	dbgregs->dr6 = val;
2866 	dbgregs->dr7 = vcpu->arch.dr7;
2867 	dbgregs->flags = 0;
2868 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2869 }
2870 
2871 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2872 					    struct kvm_debugregs *dbgregs)
2873 {
2874 	if (dbgregs->flags)
2875 		return -EINVAL;
2876 
2877 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2878 	kvm_update_dr0123(vcpu);
2879 	vcpu->arch.dr6 = dbgregs->dr6;
2880 	kvm_update_dr6(vcpu);
2881 	vcpu->arch.dr7 = dbgregs->dr7;
2882 	kvm_update_dr7(vcpu);
2883 
2884 	return 0;
2885 }
2886 
2887 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
2888 
2889 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
2890 {
2891 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2892 	u64 xstate_bv = xsave->header.xfeatures;
2893 	u64 valid;
2894 
2895 	/*
2896 	 * Copy legacy XSAVE area, to avoid complications with CPUID
2897 	 * leaves 0 and 1 in the loop below.
2898 	 */
2899 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
2900 
2901 	/* Set XSTATE_BV */
2902 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
2903 
2904 	/*
2905 	 * Copy each region from the possibly compacted offset to the
2906 	 * non-compacted offset.
2907 	 */
2908 	valid = xstate_bv & ~XSTATE_FPSSE;
2909 	while (valid) {
2910 		u64 feature = valid & -valid;
2911 		int index = fls64(feature) - 1;
2912 		void *src = get_xsave_addr(xsave, feature);
2913 
2914 		if (src) {
2915 			u32 size, offset, ecx, edx;
2916 			cpuid_count(XSTATE_CPUID, index,
2917 				    &size, &offset, &ecx, &edx);
2918 			memcpy(dest + offset, src, size);
2919 		}
2920 
2921 		valid -= feature;
2922 	}
2923 }
2924 
2925 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
2926 {
2927 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2928 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
2929 	u64 valid;
2930 
2931 	/*
2932 	 * Copy legacy XSAVE area, to avoid complications with CPUID
2933 	 * leaves 0 and 1 in the loop below.
2934 	 */
2935 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
2936 
2937 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
2938 	xsave->header.xfeatures = xstate_bv;
2939 	if (cpu_has_xsaves)
2940 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
2941 
2942 	/*
2943 	 * Copy each region from the non-compacted offset to the
2944 	 * possibly compacted offset.
2945 	 */
2946 	valid = xstate_bv & ~XSTATE_FPSSE;
2947 	while (valid) {
2948 		u64 feature = valid & -valid;
2949 		int index = fls64(feature) - 1;
2950 		void *dest = get_xsave_addr(xsave, feature);
2951 
2952 		if (dest) {
2953 			u32 size, offset, ecx, edx;
2954 			cpuid_count(XSTATE_CPUID, index,
2955 				    &size, &offset, &ecx, &edx);
2956 			memcpy(dest, src + offset, size);
2957 		}
2958 
2959 		valid -= feature;
2960 	}
2961 }
2962 
2963 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2964 					 struct kvm_xsave *guest_xsave)
2965 {
2966 	if (cpu_has_xsave) {
2967 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
2968 		fill_xsave((u8 *) guest_xsave->region, vcpu);
2969 	} else {
2970 		memcpy(guest_xsave->region,
2971 			&vcpu->arch.guest_fpu.state.fxsave,
2972 			sizeof(struct fxregs_state));
2973 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2974 			XSTATE_FPSSE;
2975 	}
2976 }
2977 
2978 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2979 					struct kvm_xsave *guest_xsave)
2980 {
2981 	u64 xstate_bv =
2982 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2983 
2984 	if (cpu_has_xsave) {
2985 		/*
2986 		 * Here we allow setting states that are not present in
2987 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
2988 		 * with old userspace.
2989 		 */
2990 		if (xstate_bv & ~kvm_supported_xcr0())
2991 			return -EINVAL;
2992 		load_xsave(vcpu, (u8 *)guest_xsave->region);
2993 	} else {
2994 		if (xstate_bv & ~XSTATE_FPSSE)
2995 			return -EINVAL;
2996 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
2997 			guest_xsave->region, sizeof(struct fxregs_state));
2998 	}
2999 	return 0;
3000 }
3001 
3002 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3003 					struct kvm_xcrs *guest_xcrs)
3004 {
3005 	if (!cpu_has_xsave) {
3006 		guest_xcrs->nr_xcrs = 0;
3007 		return;
3008 	}
3009 
3010 	guest_xcrs->nr_xcrs = 1;
3011 	guest_xcrs->flags = 0;
3012 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3013 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3014 }
3015 
3016 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3017 				       struct kvm_xcrs *guest_xcrs)
3018 {
3019 	int i, r = 0;
3020 
3021 	if (!cpu_has_xsave)
3022 		return -EINVAL;
3023 
3024 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3025 		return -EINVAL;
3026 
3027 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3028 		/* Only support XCR0 currently */
3029 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3030 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3031 				guest_xcrs->xcrs[i].value);
3032 			break;
3033 		}
3034 	if (r)
3035 		r = -EINVAL;
3036 	return r;
3037 }
3038 
3039 /*
3040  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3041  * stopped by the hypervisor.  This function will be called from the host only.
3042  * EINVAL is returned when the host attempts to set the flag for a guest that
3043  * does not support pv clocks.
3044  */
3045 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3046 {
3047 	if (!vcpu->arch.pv_time_enabled)
3048 		return -EINVAL;
3049 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3050 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3051 	return 0;
3052 }
3053 
3054 long kvm_arch_vcpu_ioctl(struct file *filp,
3055 			 unsigned int ioctl, unsigned long arg)
3056 {
3057 	struct kvm_vcpu *vcpu = filp->private_data;
3058 	void __user *argp = (void __user *)arg;
3059 	int r;
3060 	union {
3061 		struct kvm_lapic_state *lapic;
3062 		struct kvm_xsave *xsave;
3063 		struct kvm_xcrs *xcrs;
3064 		void *buffer;
3065 	} u;
3066 
3067 	u.buffer = NULL;
3068 	switch (ioctl) {
3069 	case KVM_GET_LAPIC: {
3070 		r = -EINVAL;
3071 		if (!vcpu->arch.apic)
3072 			goto out;
3073 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3074 
3075 		r = -ENOMEM;
3076 		if (!u.lapic)
3077 			goto out;
3078 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3079 		if (r)
3080 			goto out;
3081 		r = -EFAULT;
3082 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3083 			goto out;
3084 		r = 0;
3085 		break;
3086 	}
3087 	case KVM_SET_LAPIC: {
3088 		r = -EINVAL;
3089 		if (!vcpu->arch.apic)
3090 			goto out;
3091 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3092 		if (IS_ERR(u.lapic))
3093 			return PTR_ERR(u.lapic);
3094 
3095 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3096 		break;
3097 	}
3098 	case KVM_INTERRUPT: {
3099 		struct kvm_interrupt irq;
3100 
3101 		r = -EFAULT;
3102 		if (copy_from_user(&irq, argp, sizeof irq))
3103 			goto out;
3104 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3105 		break;
3106 	}
3107 	case KVM_NMI: {
3108 		r = kvm_vcpu_ioctl_nmi(vcpu);
3109 		break;
3110 	}
3111 	case KVM_SMI: {
3112 		r = kvm_vcpu_ioctl_smi(vcpu);
3113 		break;
3114 	}
3115 	case KVM_SET_CPUID: {
3116 		struct kvm_cpuid __user *cpuid_arg = argp;
3117 		struct kvm_cpuid cpuid;
3118 
3119 		r = -EFAULT;
3120 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3121 			goto out;
3122 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3123 		break;
3124 	}
3125 	case KVM_SET_CPUID2: {
3126 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3127 		struct kvm_cpuid2 cpuid;
3128 
3129 		r = -EFAULT;
3130 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3131 			goto out;
3132 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3133 					      cpuid_arg->entries);
3134 		break;
3135 	}
3136 	case KVM_GET_CPUID2: {
3137 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3138 		struct kvm_cpuid2 cpuid;
3139 
3140 		r = -EFAULT;
3141 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3142 			goto out;
3143 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3144 					      cpuid_arg->entries);
3145 		if (r)
3146 			goto out;
3147 		r = -EFAULT;
3148 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3149 			goto out;
3150 		r = 0;
3151 		break;
3152 	}
3153 	case KVM_GET_MSRS:
3154 		r = msr_io(vcpu, argp, do_get_msr, 1);
3155 		break;
3156 	case KVM_SET_MSRS:
3157 		r = msr_io(vcpu, argp, do_set_msr, 0);
3158 		break;
3159 	case KVM_TPR_ACCESS_REPORTING: {
3160 		struct kvm_tpr_access_ctl tac;
3161 
3162 		r = -EFAULT;
3163 		if (copy_from_user(&tac, argp, sizeof tac))
3164 			goto out;
3165 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3166 		if (r)
3167 			goto out;
3168 		r = -EFAULT;
3169 		if (copy_to_user(argp, &tac, sizeof tac))
3170 			goto out;
3171 		r = 0;
3172 		break;
3173 	};
3174 	case KVM_SET_VAPIC_ADDR: {
3175 		struct kvm_vapic_addr va;
3176 
3177 		r = -EINVAL;
3178 		if (!irqchip_in_kernel(vcpu->kvm))
3179 			goto out;
3180 		r = -EFAULT;
3181 		if (copy_from_user(&va, argp, sizeof va))
3182 			goto out;
3183 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3184 		break;
3185 	}
3186 	case KVM_X86_SETUP_MCE: {
3187 		u64 mcg_cap;
3188 
3189 		r = -EFAULT;
3190 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3191 			goto out;
3192 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3193 		break;
3194 	}
3195 	case KVM_X86_SET_MCE: {
3196 		struct kvm_x86_mce mce;
3197 
3198 		r = -EFAULT;
3199 		if (copy_from_user(&mce, argp, sizeof mce))
3200 			goto out;
3201 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3202 		break;
3203 	}
3204 	case KVM_GET_VCPU_EVENTS: {
3205 		struct kvm_vcpu_events events;
3206 
3207 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3208 
3209 		r = -EFAULT;
3210 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3211 			break;
3212 		r = 0;
3213 		break;
3214 	}
3215 	case KVM_SET_VCPU_EVENTS: {
3216 		struct kvm_vcpu_events events;
3217 
3218 		r = -EFAULT;
3219 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3220 			break;
3221 
3222 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3223 		break;
3224 	}
3225 	case KVM_GET_DEBUGREGS: {
3226 		struct kvm_debugregs dbgregs;
3227 
3228 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3229 
3230 		r = -EFAULT;
3231 		if (copy_to_user(argp, &dbgregs,
3232 				 sizeof(struct kvm_debugregs)))
3233 			break;
3234 		r = 0;
3235 		break;
3236 	}
3237 	case KVM_SET_DEBUGREGS: {
3238 		struct kvm_debugregs dbgregs;
3239 
3240 		r = -EFAULT;
3241 		if (copy_from_user(&dbgregs, argp,
3242 				   sizeof(struct kvm_debugregs)))
3243 			break;
3244 
3245 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3246 		break;
3247 	}
3248 	case KVM_GET_XSAVE: {
3249 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3250 		r = -ENOMEM;
3251 		if (!u.xsave)
3252 			break;
3253 
3254 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3255 
3256 		r = -EFAULT;
3257 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3258 			break;
3259 		r = 0;
3260 		break;
3261 	}
3262 	case KVM_SET_XSAVE: {
3263 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3264 		if (IS_ERR(u.xsave))
3265 			return PTR_ERR(u.xsave);
3266 
3267 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3268 		break;
3269 	}
3270 	case KVM_GET_XCRS: {
3271 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3272 		r = -ENOMEM;
3273 		if (!u.xcrs)
3274 			break;
3275 
3276 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3277 
3278 		r = -EFAULT;
3279 		if (copy_to_user(argp, u.xcrs,
3280 				 sizeof(struct kvm_xcrs)))
3281 			break;
3282 		r = 0;
3283 		break;
3284 	}
3285 	case KVM_SET_XCRS: {
3286 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3287 		if (IS_ERR(u.xcrs))
3288 			return PTR_ERR(u.xcrs);
3289 
3290 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3291 		break;
3292 	}
3293 	case KVM_SET_TSC_KHZ: {
3294 		u32 user_tsc_khz;
3295 
3296 		r = -EINVAL;
3297 		user_tsc_khz = (u32)arg;
3298 
3299 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3300 			goto out;
3301 
3302 		if (user_tsc_khz == 0)
3303 			user_tsc_khz = tsc_khz;
3304 
3305 		kvm_set_tsc_khz(vcpu, user_tsc_khz);
3306 
3307 		r = 0;
3308 		goto out;
3309 	}
3310 	case KVM_GET_TSC_KHZ: {
3311 		r = vcpu->arch.virtual_tsc_khz;
3312 		goto out;
3313 	}
3314 	case KVM_KVMCLOCK_CTRL: {
3315 		r = kvm_set_guest_paused(vcpu);
3316 		goto out;
3317 	}
3318 	default:
3319 		r = -EINVAL;
3320 	}
3321 out:
3322 	kfree(u.buffer);
3323 	return r;
3324 }
3325 
3326 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3327 {
3328 	return VM_FAULT_SIGBUS;
3329 }
3330 
3331 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3332 {
3333 	int ret;
3334 
3335 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3336 		return -EINVAL;
3337 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3338 	return ret;
3339 }
3340 
3341 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3342 					      u64 ident_addr)
3343 {
3344 	kvm->arch.ept_identity_map_addr = ident_addr;
3345 	return 0;
3346 }
3347 
3348 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3349 					  u32 kvm_nr_mmu_pages)
3350 {
3351 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3352 		return -EINVAL;
3353 
3354 	mutex_lock(&kvm->slots_lock);
3355 
3356 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3357 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3358 
3359 	mutex_unlock(&kvm->slots_lock);
3360 	return 0;
3361 }
3362 
3363 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3364 {
3365 	return kvm->arch.n_max_mmu_pages;
3366 }
3367 
3368 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3369 {
3370 	int r;
3371 
3372 	r = 0;
3373 	switch (chip->chip_id) {
3374 	case KVM_IRQCHIP_PIC_MASTER:
3375 		memcpy(&chip->chip.pic,
3376 			&pic_irqchip(kvm)->pics[0],
3377 			sizeof(struct kvm_pic_state));
3378 		break;
3379 	case KVM_IRQCHIP_PIC_SLAVE:
3380 		memcpy(&chip->chip.pic,
3381 			&pic_irqchip(kvm)->pics[1],
3382 			sizeof(struct kvm_pic_state));
3383 		break;
3384 	case KVM_IRQCHIP_IOAPIC:
3385 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3386 		break;
3387 	default:
3388 		r = -EINVAL;
3389 		break;
3390 	}
3391 	return r;
3392 }
3393 
3394 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3395 {
3396 	int r;
3397 
3398 	r = 0;
3399 	switch (chip->chip_id) {
3400 	case KVM_IRQCHIP_PIC_MASTER:
3401 		spin_lock(&pic_irqchip(kvm)->lock);
3402 		memcpy(&pic_irqchip(kvm)->pics[0],
3403 			&chip->chip.pic,
3404 			sizeof(struct kvm_pic_state));
3405 		spin_unlock(&pic_irqchip(kvm)->lock);
3406 		break;
3407 	case KVM_IRQCHIP_PIC_SLAVE:
3408 		spin_lock(&pic_irqchip(kvm)->lock);
3409 		memcpy(&pic_irqchip(kvm)->pics[1],
3410 			&chip->chip.pic,
3411 			sizeof(struct kvm_pic_state));
3412 		spin_unlock(&pic_irqchip(kvm)->lock);
3413 		break;
3414 	case KVM_IRQCHIP_IOAPIC:
3415 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3416 		break;
3417 	default:
3418 		r = -EINVAL;
3419 		break;
3420 	}
3421 	kvm_pic_update_irq(pic_irqchip(kvm));
3422 	return r;
3423 }
3424 
3425 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3426 {
3427 	int r = 0;
3428 
3429 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3430 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3431 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3432 	return r;
3433 }
3434 
3435 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3436 {
3437 	int r = 0;
3438 
3439 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3440 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3441 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3442 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3443 	return r;
3444 }
3445 
3446 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3447 {
3448 	int r = 0;
3449 
3450 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3451 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3452 		sizeof(ps->channels));
3453 	ps->flags = kvm->arch.vpit->pit_state.flags;
3454 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3455 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3456 	return r;
3457 }
3458 
3459 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3460 {
3461 	int r = 0, start = 0;
3462 	u32 prev_legacy, cur_legacy;
3463 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3464 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3465 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3466 	if (!prev_legacy && cur_legacy)
3467 		start = 1;
3468 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3469 	       sizeof(kvm->arch.vpit->pit_state.channels));
3470 	kvm->arch.vpit->pit_state.flags = ps->flags;
3471 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3472 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3473 	return r;
3474 }
3475 
3476 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3477 				 struct kvm_reinject_control *control)
3478 {
3479 	if (!kvm->arch.vpit)
3480 		return -ENXIO;
3481 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3482 	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3483 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3484 	return 0;
3485 }
3486 
3487 /**
3488  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3489  * @kvm: kvm instance
3490  * @log: slot id and address to which we copy the log
3491  *
3492  * Steps 1-4 below provide general overview of dirty page logging. See
3493  * kvm_get_dirty_log_protect() function description for additional details.
3494  *
3495  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3496  * always flush the TLB (step 4) even if previous step failed  and the dirty
3497  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3498  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3499  * writes will be marked dirty for next log read.
3500  *
3501  *   1. Take a snapshot of the bit and clear it if needed.
3502  *   2. Write protect the corresponding page.
3503  *   3. Copy the snapshot to the userspace.
3504  *   4. Flush TLB's if needed.
3505  */
3506 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3507 {
3508 	bool is_dirty = false;
3509 	int r;
3510 
3511 	mutex_lock(&kvm->slots_lock);
3512 
3513 	/*
3514 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3515 	 */
3516 	if (kvm_x86_ops->flush_log_dirty)
3517 		kvm_x86_ops->flush_log_dirty(kvm);
3518 
3519 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3520 
3521 	/*
3522 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3523 	 * kvm_mmu_slot_remove_write_access().
3524 	 */
3525 	lockdep_assert_held(&kvm->slots_lock);
3526 	if (is_dirty)
3527 		kvm_flush_remote_tlbs(kvm);
3528 
3529 	mutex_unlock(&kvm->slots_lock);
3530 	return r;
3531 }
3532 
3533 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3534 			bool line_status)
3535 {
3536 	if (!irqchip_in_kernel(kvm))
3537 		return -ENXIO;
3538 
3539 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3540 					irq_event->irq, irq_event->level,
3541 					line_status);
3542 	return 0;
3543 }
3544 
3545 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3546 				   struct kvm_enable_cap *cap)
3547 {
3548 	int r;
3549 
3550 	if (cap->flags)
3551 		return -EINVAL;
3552 
3553 	switch (cap->cap) {
3554 	case KVM_CAP_DISABLE_QUIRKS:
3555 		kvm->arch.disabled_quirks = cap->args[0];
3556 		r = 0;
3557 		break;
3558 	default:
3559 		r = -EINVAL;
3560 		break;
3561 	}
3562 	return r;
3563 }
3564 
3565 long kvm_arch_vm_ioctl(struct file *filp,
3566 		       unsigned int ioctl, unsigned long arg)
3567 {
3568 	struct kvm *kvm = filp->private_data;
3569 	void __user *argp = (void __user *)arg;
3570 	int r = -ENOTTY;
3571 	/*
3572 	 * This union makes it completely explicit to gcc-3.x
3573 	 * that these two variables' stack usage should be
3574 	 * combined, not added together.
3575 	 */
3576 	union {
3577 		struct kvm_pit_state ps;
3578 		struct kvm_pit_state2 ps2;
3579 		struct kvm_pit_config pit_config;
3580 	} u;
3581 
3582 	switch (ioctl) {
3583 	case KVM_SET_TSS_ADDR:
3584 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3585 		break;
3586 	case KVM_SET_IDENTITY_MAP_ADDR: {
3587 		u64 ident_addr;
3588 
3589 		r = -EFAULT;
3590 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3591 			goto out;
3592 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3593 		break;
3594 	}
3595 	case KVM_SET_NR_MMU_PAGES:
3596 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3597 		break;
3598 	case KVM_GET_NR_MMU_PAGES:
3599 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3600 		break;
3601 	case KVM_CREATE_IRQCHIP: {
3602 		struct kvm_pic *vpic;
3603 
3604 		mutex_lock(&kvm->lock);
3605 		r = -EEXIST;
3606 		if (kvm->arch.vpic)
3607 			goto create_irqchip_unlock;
3608 		r = -EINVAL;
3609 		if (atomic_read(&kvm->online_vcpus))
3610 			goto create_irqchip_unlock;
3611 		r = -ENOMEM;
3612 		vpic = kvm_create_pic(kvm);
3613 		if (vpic) {
3614 			r = kvm_ioapic_init(kvm);
3615 			if (r) {
3616 				mutex_lock(&kvm->slots_lock);
3617 				kvm_destroy_pic(vpic);
3618 				mutex_unlock(&kvm->slots_lock);
3619 				goto create_irqchip_unlock;
3620 			}
3621 		} else
3622 			goto create_irqchip_unlock;
3623 		r = kvm_setup_default_irq_routing(kvm);
3624 		if (r) {
3625 			mutex_lock(&kvm->slots_lock);
3626 			mutex_lock(&kvm->irq_lock);
3627 			kvm_ioapic_destroy(kvm);
3628 			kvm_destroy_pic(vpic);
3629 			mutex_unlock(&kvm->irq_lock);
3630 			mutex_unlock(&kvm->slots_lock);
3631 			goto create_irqchip_unlock;
3632 		}
3633 		/* Write kvm->irq_routing before kvm->arch.vpic.  */
3634 		smp_wmb();
3635 		kvm->arch.vpic = vpic;
3636 	create_irqchip_unlock:
3637 		mutex_unlock(&kvm->lock);
3638 		break;
3639 	}
3640 	case KVM_CREATE_PIT:
3641 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3642 		goto create_pit;
3643 	case KVM_CREATE_PIT2:
3644 		r = -EFAULT;
3645 		if (copy_from_user(&u.pit_config, argp,
3646 				   sizeof(struct kvm_pit_config)))
3647 			goto out;
3648 	create_pit:
3649 		mutex_lock(&kvm->slots_lock);
3650 		r = -EEXIST;
3651 		if (kvm->arch.vpit)
3652 			goto create_pit_unlock;
3653 		r = -ENOMEM;
3654 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3655 		if (kvm->arch.vpit)
3656 			r = 0;
3657 	create_pit_unlock:
3658 		mutex_unlock(&kvm->slots_lock);
3659 		break;
3660 	case KVM_GET_IRQCHIP: {
3661 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3662 		struct kvm_irqchip *chip;
3663 
3664 		chip = memdup_user(argp, sizeof(*chip));
3665 		if (IS_ERR(chip)) {
3666 			r = PTR_ERR(chip);
3667 			goto out;
3668 		}
3669 
3670 		r = -ENXIO;
3671 		if (!irqchip_in_kernel(kvm))
3672 			goto get_irqchip_out;
3673 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3674 		if (r)
3675 			goto get_irqchip_out;
3676 		r = -EFAULT;
3677 		if (copy_to_user(argp, chip, sizeof *chip))
3678 			goto get_irqchip_out;
3679 		r = 0;
3680 	get_irqchip_out:
3681 		kfree(chip);
3682 		break;
3683 	}
3684 	case KVM_SET_IRQCHIP: {
3685 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3686 		struct kvm_irqchip *chip;
3687 
3688 		chip = memdup_user(argp, sizeof(*chip));
3689 		if (IS_ERR(chip)) {
3690 			r = PTR_ERR(chip);
3691 			goto out;
3692 		}
3693 
3694 		r = -ENXIO;
3695 		if (!irqchip_in_kernel(kvm))
3696 			goto set_irqchip_out;
3697 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3698 		if (r)
3699 			goto set_irqchip_out;
3700 		r = 0;
3701 	set_irqchip_out:
3702 		kfree(chip);
3703 		break;
3704 	}
3705 	case KVM_GET_PIT: {
3706 		r = -EFAULT;
3707 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3708 			goto out;
3709 		r = -ENXIO;
3710 		if (!kvm->arch.vpit)
3711 			goto out;
3712 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3713 		if (r)
3714 			goto out;
3715 		r = -EFAULT;
3716 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3717 			goto out;
3718 		r = 0;
3719 		break;
3720 	}
3721 	case KVM_SET_PIT: {
3722 		r = -EFAULT;
3723 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3724 			goto out;
3725 		r = -ENXIO;
3726 		if (!kvm->arch.vpit)
3727 			goto out;
3728 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3729 		break;
3730 	}
3731 	case KVM_GET_PIT2: {
3732 		r = -ENXIO;
3733 		if (!kvm->arch.vpit)
3734 			goto out;
3735 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3736 		if (r)
3737 			goto out;
3738 		r = -EFAULT;
3739 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3740 			goto out;
3741 		r = 0;
3742 		break;
3743 	}
3744 	case KVM_SET_PIT2: {
3745 		r = -EFAULT;
3746 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3747 			goto out;
3748 		r = -ENXIO;
3749 		if (!kvm->arch.vpit)
3750 			goto out;
3751 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3752 		break;
3753 	}
3754 	case KVM_REINJECT_CONTROL: {
3755 		struct kvm_reinject_control control;
3756 		r =  -EFAULT;
3757 		if (copy_from_user(&control, argp, sizeof(control)))
3758 			goto out;
3759 		r = kvm_vm_ioctl_reinject(kvm, &control);
3760 		break;
3761 	}
3762 	case KVM_SET_BOOT_CPU_ID:
3763 		r = 0;
3764 		mutex_lock(&kvm->lock);
3765 		if (atomic_read(&kvm->online_vcpus) != 0)
3766 			r = -EBUSY;
3767 		else
3768 			kvm->arch.bsp_vcpu_id = arg;
3769 		mutex_unlock(&kvm->lock);
3770 		break;
3771 	case KVM_XEN_HVM_CONFIG: {
3772 		r = -EFAULT;
3773 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3774 				   sizeof(struct kvm_xen_hvm_config)))
3775 			goto out;
3776 		r = -EINVAL;
3777 		if (kvm->arch.xen_hvm_config.flags)
3778 			goto out;
3779 		r = 0;
3780 		break;
3781 	}
3782 	case KVM_SET_CLOCK: {
3783 		struct kvm_clock_data user_ns;
3784 		u64 now_ns;
3785 		s64 delta;
3786 
3787 		r = -EFAULT;
3788 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3789 			goto out;
3790 
3791 		r = -EINVAL;
3792 		if (user_ns.flags)
3793 			goto out;
3794 
3795 		r = 0;
3796 		local_irq_disable();
3797 		now_ns = get_kernel_ns();
3798 		delta = user_ns.clock - now_ns;
3799 		local_irq_enable();
3800 		kvm->arch.kvmclock_offset = delta;
3801 		kvm_gen_update_masterclock(kvm);
3802 		break;
3803 	}
3804 	case KVM_GET_CLOCK: {
3805 		struct kvm_clock_data user_ns;
3806 		u64 now_ns;
3807 
3808 		local_irq_disable();
3809 		now_ns = get_kernel_ns();
3810 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3811 		local_irq_enable();
3812 		user_ns.flags = 0;
3813 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3814 
3815 		r = -EFAULT;
3816 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3817 			goto out;
3818 		r = 0;
3819 		break;
3820 	}
3821 	case KVM_ENABLE_CAP: {
3822 		struct kvm_enable_cap cap;
3823 
3824 		r = -EFAULT;
3825 		if (copy_from_user(&cap, argp, sizeof(cap)))
3826 			goto out;
3827 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3828 		break;
3829 	}
3830 	default:
3831 		r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
3832 	}
3833 out:
3834 	return r;
3835 }
3836 
3837 static void kvm_init_msr_list(void)
3838 {
3839 	u32 dummy[2];
3840 	unsigned i, j;
3841 
3842 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
3843 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3844 			continue;
3845 
3846 		/*
3847 		 * Even MSRs that are valid in the host may not be exposed
3848 		 * to the guests in some cases.  We could work around this
3849 		 * in VMX with the generic MSR save/load machinery, but it
3850 		 * is not really worthwhile since it will really only
3851 		 * happen with nested virtualization.
3852 		 */
3853 		switch (msrs_to_save[i]) {
3854 		case MSR_IA32_BNDCFGS:
3855 			if (!kvm_x86_ops->mpx_supported())
3856 				continue;
3857 			break;
3858 		default:
3859 			break;
3860 		}
3861 
3862 		if (j < i)
3863 			msrs_to_save[j] = msrs_to_save[i];
3864 		j++;
3865 	}
3866 	num_msrs_to_save = j;
3867 
3868 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
3869 		switch (emulated_msrs[i]) {
3870 		case MSR_IA32_SMBASE:
3871 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
3872 				continue;
3873 			break;
3874 		default:
3875 			break;
3876 		}
3877 
3878 		if (j < i)
3879 			emulated_msrs[j] = emulated_msrs[i];
3880 		j++;
3881 	}
3882 	num_emulated_msrs = j;
3883 }
3884 
3885 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3886 			   const void *v)
3887 {
3888 	int handled = 0;
3889 	int n;
3890 
3891 	do {
3892 		n = min(len, 8);
3893 		if (!(vcpu->arch.apic &&
3894 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
3895 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
3896 			break;
3897 		handled += n;
3898 		addr += n;
3899 		len -= n;
3900 		v += n;
3901 	} while (len);
3902 
3903 	return handled;
3904 }
3905 
3906 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3907 {
3908 	int handled = 0;
3909 	int n;
3910 
3911 	do {
3912 		n = min(len, 8);
3913 		if (!(vcpu->arch.apic &&
3914 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
3915 					 addr, n, v))
3916 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
3917 			break;
3918 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3919 		handled += n;
3920 		addr += n;
3921 		len -= n;
3922 		v += n;
3923 	} while (len);
3924 
3925 	return handled;
3926 }
3927 
3928 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3929 			struct kvm_segment *var, int seg)
3930 {
3931 	kvm_x86_ops->set_segment(vcpu, var, seg);
3932 }
3933 
3934 void kvm_get_segment(struct kvm_vcpu *vcpu,
3935 		     struct kvm_segment *var, int seg)
3936 {
3937 	kvm_x86_ops->get_segment(vcpu, var, seg);
3938 }
3939 
3940 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
3941 			   struct x86_exception *exception)
3942 {
3943 	gpa_t t_gpa;
3944 
3945 	BUG_ON(!mmu_is_nested(vcpu));
3946 
3947 	/* NPT walks are always user-walks */
3948 	access |= PFERR_USER_MASK;
3949 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
3950 
3951 	return t_gpa;
3952 }
3953 
3954 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3955 			      struct x86_exception *exception)
3956 {
3957 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3958 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3959 }
3960 
3961  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3962 				struct x86_exception *exception)
3963 {
3964 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3965 	access |= PFERR_FETCH_MASK;
3966 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3967 }
3968 
3969 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3970 			       struct x86_exception *exception)
3971 {
3972 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3973 	access |= PFERR_WRITE_MASK;
3974 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3975 }
3976 
3977 /* uses this to access any guest's mapped memory without checking CPL */
3978 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3979 				struct x86_exception *exception)
3980 {
3981 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3982 }
3983 
3984 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3985 				      struct kvm_vcpu *vcpu, u32 access,
3986 				      struct x86_exception *exception)
3987 {
3988 	void *data = val;
3989 	int r = X86EMUL_CONTINUE;
3990 
3991 	while (bytes) {
3992 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3993 							    exception);
3994 		unsigned offset = addr & (PAGE_SIZE-1);
3995 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3996 		int ret;
3997 
3998 		if (gpa == UNMAPPED_GVA)
3999 			return X86EMUL_PROPAGATE_FAULT;
4000 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4001 					       offset, toread);
4002 		if (ret < 0) {
4003 			r = X86EMUL_IO_NEEDED;
4004 			goto out;
4005 		}
4006 
4007 		bytes -= toread;
4008 		data += toread;
4009 		addr += toread;
4010 	}
4011 out:
4012 	return r;
4013 }
4014 
4015 /* used for instruction fetching */
4016 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4017 				gva_t addr, void *val, unsigned int bytes,
4018 				struct x86_exception *exception)
4019 {
4020 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4021 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4022 	unsigned offset;
4023 	int ret;
4024 
4025 	/* Inline kvm_read_guest_virt_helper for speed.  */
4026 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4027 						    exception);
4028 	if (unlikely(gpa == UNMAPPED_GVA))
4029 		return X86EMUL_PROPAGATE_FAULT;
4030 
4031 	offset = addr & (PAGE_SIZE-1);
4032 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4033 		bytes = (unsigned)PAGE_SIZE - offset;
4034 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4035 				       offset, bytes);
4036 	if (unlikely(ret < 0))
4037 		return X86EMUL_IO_NEEDED;
4038 
4039 	return X86EMUL_CONTINUE;
4040 }
4041 
4042 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4043 			       gva_t addr, void *val, unsigned int bytes,
4044 			       struct x86_exception *exception)
4045 {
4046 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4047 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4048 
4049 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4050 					  exception);
4051 }
4052 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4053 
4054 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4055 				      gva_t addr, void *val, unsigned int bytes,
4056 				      struct x86_exception *exception)
4057 {
4058 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4059 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4060 }
4061 
4062 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4063 				       gva_t addr, void *val,
4064 				       unsigned int bytes,
4065 				       struct x86_exception *exception)
4066 {
4067 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4068 	void *data = val;
4069 	int r = X86EMUL_CONTINUE;
4070 
4071 	while (bytes) {
4072 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4073 							     PFERR_WRITE_MASK,
4074 							     exception);
4075 		unsigned offset = addr & (PAGE_SIZE-1);
4076 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4077 		int ret;
4078 
4079 		if (gpa == UNMAPPED_GVA)
4080 			return X86EMUL_PROPAGATE_FAULT;
4081 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4082 		if (ret < 0) {
4083 			r = X86EMUL_IO_NEEDED;
4084 			goto out;
4085 		}
4086 
4087 		bytes -= towrite;
4088 		data += towrite;
4089 		addr += towrite;
4090 	}
4091 out:
4092 	return r;
4093 }
4094 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4095 
4096 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4097 				gpa_t *gpa, struct x86_exception *exception,
4098 				bool write)
4099 {
4100 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4101 		| (write ? PFERR_WRITE_MASK : 0);
4102 
4103 	if (vcpu_match_mmio_gva(vcpu, gva)
4104 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4105 				 vcpu->arch.access, access)) {
4106 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4107 					(gva & (PAGE_SIZE - 1));
4108 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4109 		return 1;
4110 	}
4111 
4112 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4113 
4114 	if (*gpa == UNMAPPED_GVA)
4115 		return -1;
4116 
4117 	/* For APIC access vmexit */
4118 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4119 		return 1;
4120 
4121 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4122 		trace_vcpu_match_mmio(gva, *gpa, write, true);
4123 		return 1;
4124 	}
4125 
4126 	return 0;
4127 }
4128 
4129 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4130 			const void *val, int bytes)
4131 {
4132 	int ret;
4133 
4134 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4135 	if (ret < 0)
4136 		return 0;
4137 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4138 	return 1;
4139 }
4140 
4141 struct read_write_emulator_ops {
4142 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4143 				  int bytes);
4144 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4145 				  void *val, int bytes);
4146 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4147 			       int bytes, void *val);
4148 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4149 				    void *val, int bytes);
4150 	bool write;
4151 };
4152 
4153 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4154 {
4155 	if (vcpu->mmio_read_completed) {
4156 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4157 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4158 		vcpu->mmio_read_completed = 0;
4159 		return 1;
4160 	}
4161 
4162 	return 0;
4163 }
4164 
4165 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4166 			void *val, int bytes)
4167 {
4168 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4169 }
4170 
4171 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4172 			 void *val, int bytes)
4173 {
4174 	return emulator_write_phys(vcpu, gpa, val, bytes);
4175 }
4176 
4177 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4178 {
4179 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4180 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4181 }
4182 
4183 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4184 			  void *val, int bytes)
4185 {
4186 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4187 	return X86EMUL_IO_NEEDED;
4188 }
4189 
4190 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4191 			   void *val, int bytes)
4192 {
4193 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4194 
4195 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4196 	return X86EMUL_CONTINUE;
4197 }
4198 
4199 static const struct read_write_emulator_ops read_emultor = {
4200 	.read_write_prepare = read_prepare,
4201 	.read_write_emulate = read_emulate,
4202 	.read_write_mmio = vcpu_mmio_read,
4203 	.read_write_exit_mmio = read_exit_mmio,
4204 };
4205 
4206 static const struct read_write_emulator_ops write_emultor = {
4207 	.read_write_emulate = write_emulate,
4208 	.read_write_mmio = write_mmio,
4209 	.read_write_exit_mmio = write_exit_mmio,
4210 	.write = true,
4211 };
4212 
4213 static int emulator_read_write_onepage(unsigned long addr, void *val,
4214 				       unsigned int bytes,
4215 				       struct x86_exception *exception,
4216 				       struct kvm_vcpu *vcpu,
4217 				       const struct read_write_emulator_ops *ops)
4218 {
4219 	gpa_t gpa;
4220 	int handled, ret;
4221 	bool write = ops->write;
4222 	struct kvm_mmio_fragment *frag;
4223 
4224 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4225 
4226 	if (ret < 0)
4227 		return X86EMUL_PROPAGATE_FAULT;
4228 
4229 	/* For APIC access vmexit */
4230 	if (ret)
4231 		goto mmio;
4232 
4233 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4234 		return X86EMUL_CONTINUE;
4235 
4236 mmio:
4237 	/*
4238 	 * Is this MMIO handled locally?
4239 	 */
4240 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4241 	if (handled == bytes)
4242 		return X86EMUL_CONTINUE;
4243 
4244 	gpa += handled;
4245 	bytes -= handled;
4246 	val += handled;
4247 
4248 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4249 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4250 	frag->gpa = gpa;
4251 	frag->data = val;
4252 	frag->len = bytes;
4253 	return X86EMUL_CONTINUE;
4254 }
4255 
4256 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4257 			unsigned long addr,
4258 			void *val, unsigned int bytes,
4259 			struct x86_exception *exception,
4260 			const struct read_write_emulator_ops *ops)
4261 {
4262 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4263 	gpa_t gpa;
4264 	int rc;
4265 
4266 	if (ops->read_write_prepare &&
4267 		  ops->read_write_prepare(vcpu, val, bytes))
4268 		return X86EMUL_CONTINUE;
4269 
4270 	vcpu->mmio_nr_fragments = 0;
4271 
4272 	/* Crossing a page boundary? */
4273 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4274 		int now;
4275 
4276 		now = -addr & ~PAGE_MASK;
4277 		rc = emulator_read_write_onepage(addr, val, now, exception,
4278 						 vcpu, ops);
4279 
4280 		if (rc != X86EMUL_CONTINUE)
4281 			return rc;
4282 		addr += now;
4283 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4284 			addr = (u32)addr;
4285 		val += now;
4286 		bytes -= now;
4287 	}
4288 
4289 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4290 					 vcpu, ops);
4291 	if (rc != X86EMUL_CONTINUE)
4292 		return rc;
4293 
4294 	if (!vcpu->mmio_nr_fragments)
4295 		return rc;
4296 
4297 	gpa = vcpu->mmio_fragments[0].gpa;
4298 
4299 	vcpu->mmio_needed = 1;
4300 	vcpu->mmio_cur_fragment = 0;
4301 
4302 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4303 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4304 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4305 	vcpu->run->mmio.phys_addr = gpa;
4306 
4307 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4308 }
4309 
4310 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4311 				  unsigned long addr,
4312 				  void *val,
4313 				  unsigned int bytes,
4314 				  struct x86_exception *exception)
4315 {
4316 	return emulator_read_write(ctxt, addr, val, bytes,
4317 				   exception, &read_emultor);
4318 }
4319 
4320 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4321 			    unsigned long addr,
4322 			    const void *val,
4323 			    unsigned int bytes,
4324 			    struct x86_exception *exception)
4325 {
4326 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4327 				   exception, &write_emultor);
4328 }
4329 
4330 #define CMPXCHG_TYPE(t, ptr, old, new) \
4331 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4332 
4333 #ifdef CONFIG_X86_64
4334 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4335 #else
4336 #  define CMPXCHG64(ptr, old, new) \
4337 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4338 #endif
4339 
4340 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4341 				     unsigned long addr,
4342 				     const void *old,
4343 				     const void *new,
4344 				     unsigned int bytes,
4345 				     struct x86_exception *exception)
4346 {
4347 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4348 	gpa_t gpa;
4349 	struct page *page;
4350 	char *kaddr;
4351 	bool exchanged;
4352 
4353 	/* guests cmpxchg8b have to be emulated atomically */
4354 	if (bytes > 8 || (bytes & (bytes - 1)))
4355 		goto emul_write;
4356 
4357 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4358 
4359 	if (gpa == UNMAPPED_GVA ||
4360 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4361 		goto emul_write;
4362 
4363 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4364 		goto emul_write;
4365 
4366 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4367 	if (is_error_page(page))
4368 		goto emul_write;
4369 
4370 	kaddr = kmap_atomic(page);
4371 	kaddr += offset_in_page(gpa);
4372 	switch (bytes) {
4373 	case 1:
4374 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4375 		break;
4376 	case 2:
4377 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4378 		break;
4379 	case 4:
4380 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4381 		break;
4382 	case 8:
4383 		exchanged = CMPXCHG64(kaddr, old, new);
4384 		break;
4385 	default:
4386 		BUG();
4387 	}
4388 	kunmap_atomic(kaddr);
4389 	kvm_release_page_dirty(page);
4390 
4391 	if (!exchanged)
4392 		return X86EMUL_CMPXCHG_FAILED;
4393 
4394 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4395 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4396 
4397 	return X86EMUL_CONTINUE;
4398 
4399 emul_write:
4400 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4401 
4402 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4403 }
4404 
4405 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4406 {
4407 	/* TODO: String I/O for in kernel device */
4408 	int r;
4409 
4410 	if (vcpu->arch.pio.in)
4411 		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4412 				    vcpu->arch.pio.size, pd);
4413 	else
4414 		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4415 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4416 				     pd);
4417 	return r;
4418 }
4419 
4420 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4421 			       unsigned short port, void *val,
4422 			       unsigned int count, bool in)
4423 {
4424 	vcpu->arch.pio.port = port;
4425 	vcpu->arch.pio.in = in;
4426 	vcpu->arch.pio.count  = count;
4427 	vcpu->arch.pio.size = size;
4428 
4429 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4430 		vcpu->arch.pio.count = 0;
4431 		return 1;
4432 	}
4433 
4434 	vcpu->run->exit_reason = KVM_EXIT_IO;
4435 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4436 	vcpu->run->io.size = size;
4437 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4438 	vcpu->run->io.count = count;
4439 	vcpu->run->io.port = port;
4440 
4441 	return 0;
4442 }
4443 
4444 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4445 				    int size, unsigned short port, void *val,
4446 				    unsigned int count)
4447 {
4448 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4449 	int ret;
4450 
4451 	if (vcpu->arch.pio.count)
4452 		goto data_avail;
4453 
4454 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4455 	if (ret) {
4456 data_avail:
4457 		memcpy(val, vcpu->arch.pio_data, size * count);
4458 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4459 		vcpu->arch.pio.count = 0;
4460 		return 1;
4461 	}
4462 
4463 	return 0;
4464 }
4465 
4466 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4467 				     int size, unsigned short port,
4468 				     const void *val, unsigned int count)
4469 {
4470 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4471 
4472 	memcpy(vcpu->arch.pio_data, val, size * count);
4473 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4474 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4475 }
4476 
4477 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4478 {
4479 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4480 }
4481 
4482 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4483 {
4484 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4485 }
4486 
4487 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4488 {
4489 	if (!need_emulate_wbinvd(vcpu))
4490 		return X86EMUL_CONTINUE;
4491 
4492 	if (kvm_x86_ops->has_wbinvd_exit()) {
4493 		int cpu = get_cpu();
4494 
4495 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4496 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4497 				wbinvd_ipi, NULL, 1);
4498 		put_cpu();
4499 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4500 	} else
4501 		wbinvd();
4502 	return X86EMUL_CONTINUE;
4503 }
4504 
4505 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4506 {
4507 	kvm_x86_ops->skip_emulated_instruction(vcpu);
4508 	return kvm_emulate_wbinvd_noskip(vcpu);
4509 }
4510 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4511 
4512 
4513 
4514 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4515 {
4516 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4517 }
4518 
4519 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4520 			   unsigned long *dest)
4521 {
4522 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4523 }
4524 
4525 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4526 			   unsigned long value)
4527 {
4528 
4529 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4530 }
4531 
4532 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4533 {
4534 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4535 }
4536 
4537 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4538 {
4539 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4540 	unsigned long value;
4541 
4542 	switch (cr) {
4543 	case 0:
4544 		value = kvm_read_cr0(vcpu);
4545 		break;
4546 	case 2:
4547 		value = vcpu->arch.cr2;
4548 		break;
4549 	case 3:
4550 		value = kvm_read_cr3(vcpu);
4551 		break;
4552 	case 4:
4553 		value = kvm_read_cr4(vcpu);
4554 		break;
4555 	case 8:
4556 		value = kvm_get_cr8(vcpu);
4557 		break;
4558 	default:
4559 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4560 		return 0;
4561 	}
4562 
4563 	return value;
4564 }
4565 
4566 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4567 {
4568 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4569 	int res = 0;
4570 
4571 	switch (cr) {
4572 	case 0:
4573 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4574 		break;
4575 	case 2:
4576 		vcpu->arch.cr2 = val;
4577 		break;
4578 	case 3:
4579 		res = kvm_set_cr3(vcpu, val);
4580 		break;
4581 	case 4:
4582 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4583 		break;
4584 	case 8:
4585 		res = kvm_set_cr8(vcpu, val);
4586 		break;
4587 	default:
4588 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4589 		res = -1;
4590 	}
4591 
4592 	return res;
4593 }
4594 
4595 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4596 {
4597 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4598 }
4599 
4600 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4601 {
4602 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4603 }
4604 
4605 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4606 {
4607 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4608 }
4609 
4610 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4611 {
4612 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4613 }
4614 
4615 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4616 {
4617 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4618 }
4619 
4620 static unsigned long emulator_get_cached_segment_base(
4621 	struct x86_emulate_ctxt *ctxt, int seg)
4622 {
4623 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4624 }
4625 
4626 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4627 				 struct desc_struct *desc, u32 *base3,
4628 				 int seg)
4629 {
4630 	struct kvm_segment var;
4631 
4632 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4633 	*selector = var.selector;
4634 
4635 	if (var.unusable) {
4636 		memset(desc, 0, sizeof(*desc));
4637 		return false;
4638 	}
4639 
4640 	if (var.g)
4641 		var.limit >>= 12;
4642 	set_desc_limit(desc, var.limit);
4643 	set_desc_base(desc, (unsigned long)var.base);
4644 #ifdef CONFIG_X86_64
4645 	if (base3)
4646 		*base3 = var.base >> 32;
4647 #endif
4648 	desc->type = var.type;
4649 	desc->s = var.s;
4650 	desc->dpl = var.dpl;
4651 	desc->p = var.present;
4652 	desc->avl = var.avl;
4653 	desc->l = var.l;
4654 	desc->d = var.db;
4655 	desc->g = var.g;
4656 
4657 	return true;
4658 }
4659 
4660 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4661 				 struct desc_struct *desc, u32 base3,
4662 				 int seg)
4663 {
4664 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4665 	struct kvm_segment var;
4666 
4667 	var.selector = selector;
4668 	var.base = get_desc_base(desc);
4669 #ifdef CONFIG_X86_64
4670 	var.base |= ((u64)base3) << 32;
4671 #endif
4672 	var.limit = get_desc_limit(desc);
4673 	if (desc->g)
4674 		var.limit = (var.limit << 12) | 0xfff;
4675 	var.type = desc->type;
4676 	var.dpl = desc->dpl;
4677 	var.db = desc->d;
4678 	var.s = desc->s;
4679 	var.l = desc->l;
4680 	var.g = desc->g;
4681 	var.avl = desc->avl;
4682 	var.present = desc->p;
4683 	var.unusable = !var.present;
4684 	var.padding = 0;
4685 
4686 	kvm_set_segment(vcpu, &var, seg);
4687 	return;
4688 }
4689 
4690 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4691 			    u32 msr_index, u64 *pdata)
4692 {
4693 	struct msr_data msr;
4694 	int r;
4695 
4696 	msr.index = msr_index;
4697 	msr.host_initiated = false;
4698 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4699 	if (r)
4700 		return r;
4701 
4702 	*pdata = msr.data;
4703 	return 0;
4704 }
4705 
4706 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4707 			    u32 msr_index, u64 data)
4708 {
4709 	struct msr_data msr;
4710 
4711 	msr.data = data;
4712 	msr.index = msr_index;
4713 	msr.host_initiated = false;
4714 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4715 }
4716 
4717 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4718 {
4719 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4720 
4721 	return vcpu->arch.smbase;
4722 }
4723 
4724 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4725 {
4726 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4727 
4728 	vcpu->arch.smbase = smbase;
4729 }
4730 
4731 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4732 			      u32 pmc)
4733 {
4734 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4735 }
4736 
4737 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4738 			     u32 pmc, u64 *pdata)
4739 {
4740 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4741 }
4742 
4743 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4744 {
4745 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4746 }
4747 
4748 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4749 {
4750 	preempt_disable();
4751 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4752 	/*
4753 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4754 	 * so it may be clear at this point.
4755 	 */
4756 	clts();
4757 }
4758 
4759 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4760 {
4761 	preempt_enable();
4762 }
4763 
4764 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4765 			      struct x86_instruction_info *info,
4766 			      enum x86_intercept_stage stage)
4767 {
4768 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4769 }
4770 
4771 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4772 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4773 {
4774 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4775 }
4776 
4777 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4778 {
4779 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
4780 }
4781 
4782 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4783 {
4784 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4785 }
4786 
4787 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4788 {
4789 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4790 }
4791 
4792 static const struct x86_emulate_ops emulate_ops = {
4793 	.read_gpr            = emulator_read_gpr,
4794 	.write_gpr           = emulator_write_gpr,
4795 	.read_std            = kvm_read_guest_virt_system,
4796 	.write_std           = kvm_write_guest_virt_system,
4797 	.fetch               = kvm_fetch_guest_virt,
4798 	.read_emulated       = emulator_read_emulated,
4799 	.write_emulated      = emulator_write_emulated,
4800 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
4801 	.invlpg              = emulator_invlpg,
4802 	.pio_in_emulated     = emulator_pio_in_emulated,
4803 	.pio_out_emulated    = emulator_pio_out_emulated,
4804 	.get_segment         = emulator_get_segment,
4805 	.set_segment         = emulator_set_segment,
4806 	.get_cached_segment_base = emulator_get_cached_segment_base,
4807 	.get_gdt             = emulator_get_gdt,
4808 	.get_idt	     = emulator_get_idt,
4809 	.set_gdt             = emulator_set_gdt,
4810 	.set_idt	     = emulator_set_idt,
4811 	.get_cr              = emulator_get_cr,
4812 	.set_cr              = emulator_set_cr,
4813 	.cpl                 = emulator_get_cpl,
4814 	.get_dr              = emulator_get_dr,
4815 	.set_dr              = emulator_set_dr,
4816 	.get_smbase          = emulator_get_smbase,
4817 	.set_smbase          = emulator_set_smbase,
4818 	.set_msr             = emulator_set_msr,
4819 	.get_msr             = emulator_get_msr,
4820 	.check_pmc	     = emulator_check_pmc,
4821 	.read_pmc            = emulator_read_pmc,
4822 	.halt                = emulator_halt,
4823 	.wbinvd              = emulator_wbinvd,
4824 	.fix_hypercall       = emulator_fix_hypercall,
4825 	.get_fpu             = emulator_get_fpu,
4826 	.put_fpu             = emulator_put_fpu,
4827 	.intercept           = emulator_intercept,
4828 	.get_cpuid           = emulator_get_cpuid,
4829 	.set_nmi_mask        = emulator_set_nmi_mask,
4830 };
4831 
4832 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4833 {
4834 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4835 	/*
4836 	 * an sti; sti; sequence only disable interrupts for the first
4837 	 * instruction. So, if the last instruction, be it emulated or
4838 	 * not, left the system with the INT_STI flag enabled, it
4839 	 * means that the last instruction is an sti. We should not
4840 	 * leave the flag on in this case. The same goes for mov ss
4841 	 */
4842 	if (int_shadow & mask)
4843 		mask = 0;
4844 	if (unlikely(int_shadow || mask)) {
4845 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4846 		if (!mask)
4847 			kvm_make_request(KVM_REQ_EVENT, vcpu);
4848 	}
4849 }
4850 
4851 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
4852 {
4853 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4854 	if (ctxt->exception.vector == PF_VECTOR)
4855 		return kvm_propagate_fault(vcpu, &ctxt->exception);
4856 
4857 	if (ctxt->exception.error_code_valid)
4858 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4859 				      ctxt->exception.error_code);
4860 	else
4861 		kvm_queue_exception(vcpu, ctxt->exception.vector);
4862 	return false;
4863 }
4864 
4865 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4866 {
4867 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4868 	int cs_db, cs_l;
4869 
4870 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4871 
4872 	ctxt->eflags = kvm_get_rflags(vcpu);
4873 	ctxt->eip = kvm_rip_read(vcpu);
4874 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
4875 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
4876 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
4877 		     cs_db				? X86EMUL_MODE_PROT32 :
4878 							  X86EMUL_MODE_PROT16;
4879 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
4880 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
4881 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
4882 	ctxt->emul_flags = vcpu->arch.hflags;
4883 
4884 	init_decode_cache(ctxt);
4885 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4886 }
4887 
4888 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4889 {
4890 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4891 	int ret;
4892 
4893 	init_emulate_ctxt(vcpu);
4894 
4895 	ctxt->op_bytes = 2;
4896 	ctxt->ad_bytes = 2;
4897 	ctxt->_eip = ctxt->eip + inc_eip;
4898 	ret = emulate_int_real(ctxt, irq);
4899 
4900 	if (ret != X86EMUL_CONTINUE)
4901 		return EMULATE_FAIL;
4902 
4903 	ctxt->eip = ctxt->_eip;
4904 	kvm_rip_write(vcpu, ctxt->eip);
4905 	kvm_set_rflags(vcpu, ctxt->eflags);
4906 
4907 	if (irq == NMI_VECTOR)
4908 		vcpu->arch.nmi_pending = 0;
4909 	else
4910 		vcpu->arch.interrupt.pending = false;
4911 
4912 	return EMULATE_DONE;
4913 }
4914 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4915 
4916 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4917 {
4918 	int r = EMULATE_DONE;
4919 
4920 	++vcpu->stat.insn_emulation_fail;
4921 	trace_kvm_emulate_insn_failed(vcpu);
4922 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4923 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4924 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4925 		vcpu->run->internal.ndata = 0;
4926 		r = EMULATE_FAIL;
4927 	}
4928 	kvm_queue_exception(vcpu, UD_VECTOR);
4929 
4930 	return r;
4931 }
4932 
4933 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4934 				  bool write_fault_to_shadow_pgtable,
4935 				  int emulation_type)
4936 {
4937 	gpa_t gpa = cr2;
4938 	pfn_t pfn;
4939 
4940 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
4941 		return false;
4942 
4943 	if (!vcpu->arch.mmu.direct_map) {
4944 		/*
4945 		 * Write permission should be allowed since only
4946 		 * write access need to be emulated.
4947 		 */
4948 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4949 
4950 		/*
4951 		 * If the mapping is invalid in guest, let cpu retry
4952 		 * it to generate fault.
4953 		 */
4954 		if (gpa == UNMAPPED_GVA)
4955 			return true;
4956 	}
4957 
4958 	/*
4959 	 * Do not retry the unhandleable instruction if it faults on the
4960 	 * readonly host memory, otherwise it will goto a infinite loop:
4961 	 * retry instruction -> write #PF -> emulation fail -> retry
4962 	 * instruction -> ...
4963 	 */
4964 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4965 
4966 	/*
4967 	 * If the instruction failed on the error pfn, it can not be fixed,
4968 	 * report the error to userspace.
4969 	 */
4970 	if (is_error_noslot_pfn(pfn))
4971 		return false;
4972 
4973 	kvm_release_pfn_clean(pfn);
4974 
4975 	/* The instructions are well-emulated on direct mmu. */
4976 	if (vcpu->arch.mmu.direct_map) {
4977 		unsigned int indirect_shadow_pages;
4978 
4979 		spin_lock(&vcpu->kvm->mmu_lock);
4980 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4981 		spin_unlock(&vcpu->kvm->mmu_lock);
4982 
4983 		if (indirect_shadow_pages)
4984 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4985 
4986 		return true;
4987 	}
4988 
4989 	/*
4990 	 * if emulation was due to access to shadowed page table
4991 	 * and it failed try to unshadow page and re-enter the
4992 	 * guest to let CPU execute the instruction.
4993 	 */
4994 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4995 
4996 	/*
4997 	 * If the access faults on its page table, it can not
4998 	 * be fixed by unprotecting shadow page and it should
4999 	 * be reported to userspace.
5000 	 */
5001 	return !write_fault_to_shadow_pgtable;
5002 }
5003 
5004 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5005 			      unsigned long cr2,  int emulation_type)
5006 {
5007 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5008 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5009 
5010 	last_retry_eip = vcpu->arch.last_retry_eip;
5011 	last_retry_addr = vcpu->arch.last_retry_addr;
5012 
5013 	/*
5014 	 * If the emulation is caused by #PF and it is non-page_table
5015 	 * writing instruction, it means the VM-EXIT is caused by shadow
5016 	 * page protected, we can zap the shadow page and retry this
5017 	 * instruction directly.
5018 	 *
5019 	 * Note: if the guest uses a non-page-table modifying instruction
5020 	 * on the PDE that points to the instruction, then we will unmap
5021 	 * the instruction and go to an infinite loop. So, we cache the
5022 	 * last retried eip and the last fault address, if we meet the eip
5023 	 * and the address again, we can break out of the potential infinite
5024 	 * loop.
5025 	 */
5026 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5027 
5028 	if (!(emulation_type & EMULTYPE_RETRY))
5029 		return false;
5030 
5031 	if (x86_page_table_writing_insn(ctxt))
5032 		return false;
5033 
5034 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5035 		return false;
5036 
5037 	vcpu->arch.last_retry_eip = ctxt->eip;
5038 	vcpu->arch.last_retry_addr = cr2;
5039 
5040 	if (!vcpu->arch.mmu.direct_map)
5041 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5042 
5043 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5044 
5045 	return true;
5046 }
5047 
5048 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5049 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5050 
5051 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5052 {
5053 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5054 		/* This is a good place to trace that we are exiting SMM.  */
5055 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5056 
5057 		if (unlikely(vcpu->arch.smi_pending)) {
5058 			kvm_make_request(KVM_REQ_SMI, vcpu);
5059 			vcpu->arch.smi_pending = 0;
5060 		} else {
5061 			/* Process a latched INIT, if any.  */
5062 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5063 		}
5064 	}
5065 
5066 	kvm_mmu_reset_context(vcpu);
5067 }
5068 
5069 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5070 {
5071 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5072 
5073 	vcpu->arch.hflags = emul_flags;
5074 
5075 	if (changed & HF_SMM_MASK)
5076 		kvm_smm_changed(vcpu);
5077 }
5078 
5079 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5080 				unsigned long *db)
5081 {
5082 	u32 dr6 = 0;
5083 	int i;
5084 	u32 enable, rwlen;
5085 
5086 	enable = dr7;
5087 	rwlen = dr7 >> 16;
5088 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5089 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5090 			dr6 |= (1 << i);
5091 	return dr6;
5092 }
5093 
5094 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5095 {
5096 	struct kvm_run *kvm_run = vcpu->run;
5097 
5098 	/*
5099 	 * rflags is the old, "raw" value of the flags.  The new value has
5100 	 * not been saved yet.
5101 	 *
5102 	 * This is correct even for TF set by the guest, because "the
5103 	 * processor will not generate this exception after the instruction
5104 	 * that sets the TF flag".
5105 	 */
5106 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5107 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5108 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5109 						  DR6_RTM;
5110 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5111 			kvm_run->debug.arch.exception = DB_VECTOR;
5112 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5113 			*r = EMULATE_USER_EXIT;
5114 		} else {
5115 			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5116 			/*
5117 			 * "Certain debug exceptions may clear bit 0-3.  The
5118 			 * remaining contents of the DR6 register are never
5119 			 * cleared by the processor".
5120 			 */
5121 			vcpu->arch.dr6 &= ~15;
5122 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5123 			kvm_queue_exception(vcpu, DB_VECTOR);
5124 		}
5125 	}
5126 }
5127 
5128 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5129 {
5130 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5131 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5132 		struct kvm_run *kvm_run = vcpu->run;
5133 		unsigned long eip = kvm_get_linear_rip(vcpu);
5134 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5135 					   vcpu->arch.guest_debug_dr7,
5136 					   vcpu->arch.eff_db);
5137 
5138 		if (dr6 != 0) {
5139 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5140 			kvm_run->debug.arch.pc = eip;
5141 			kvm_run->debug.arch.exception = DB_VECTOR;
5142 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5143 			*r = EMULATE_USER_EXIT;
5144 			return true;
5145 		}
5146 	}
5147 
5148 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5149 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5150 		unsigned long eip = kvm_get_linear_rip(vcpu);
5151 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5152 					   vcpu->arch.dr7,
5153 					   vcpu->arch.db);
5154 
5155 		if (dr6 != 0) {
5156 			vcpu->arch.dr6 &= ~15;
5157 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5158 			kvm_queue_exception(vcpu, DB_VECTOR);
5159 			*r = EMULATE_DONE;
5160 			return true;
5161 		}
5162 	}
5163 
5164 	return false;
5165 }
5166 
5167 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5168 			    unsigned long cr2,
5169 			    int emulation_type,
5170 			    void *insn,
5171 			    int insn_len)
5172 {
5173 	int r;
5174 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5175 	bool writeback = true;
5176 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5177 
5178 	/*
5179 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5180 	 * never reused.
5181 	 */
5182 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5183 	kvm_clear_exception_queue(vcpu);
5184 
5185 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5186 		init_emulate_ctxt(vcpu);
5187 
5188 		/*
5189 		 * We will reenter on the same instruction since
5190 		 * we do not set complete_userspace_io.  This does not
5191 		 * handle watchpoints yet, those would be handled in
5192 		 * the emulate_ops.
5193 		 */
5194 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5195 			return r;
5196 
5197 		ctxt->interruptibility = 0;
5198 		ctxt->have_exception = false;
5199 		ctxt->exception.vector = -1;
5200 		ctxt->perm_ok = false;
5201 
5202 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5203 
5204 		r = x86_decode_insn(ctxt, insn, insn_len);
5205 
5206 		trace_kvm_emulate_insn_start(vcpu);
5207 		++vcpu->stat.insn_emulation;
5208 		if (r != EMULATION_OK)  {
5209 			if (emulation_type & EMULTYPE_TRAP_UD)
5210 				return EMULATE_FAIL;
5211 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5212 						emulation_type))
5213 				return EMULATE_DONE;
5214 			if (emulation_type & EMULTYPE_SKIP)
5215 				return EMULATE_FAIL;
5216 			return handle_emulation_failure(vcpu);
5217 		}
5218 	}
5219 
5220 	if (emulation_type & EMULTYPE_SKIP) {
5221 		kvm_rip_write(vcpu, ctxt->_eip);
5222 		if (ctxt->eflags & X86_EFLAGS_RF)
5223 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5224 		return EMULATE_DONE;
5225 	}
5226 
5227 	if (retry_instruction(ctxt, cr2, emulation_type))
5228 		return EMULATE_DONE;
5229 
5230 	/* this is needed for vmware backdoor interface to work since it
5231 	   changes registers values  during IO operation */
5232 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5233 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5234 		emulator_invalidate_register_cache(ctxt);
5235 	}
5236 
5237 restart:
5238 	r = x86_emulate_insn(ctxt);
5239 
5240 	if (r == EMULATION_INTERCEPTED)
5241 		return EMULATE_DONE;
5242 
5243 	if (r == EMULATION_FAILED) {
5244 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5245 					emulation_type))
5246 			return EMULATE_DONE;
5247 
5248 		return handle_emulation_failure(vcpu);
5249 	}
5250 
5251 	if (ctxt->have_exception) {
5252 		r = EMULATE_DONE;
5253 		if (inject_emulated_exception(vcpu))
5254 			return r;
5255 	} else if (vcpu->arch.pio.count) {
5256 		if (!vcpu->arch.pio.in) {
5257 			/* FIXME: return into emulator if single-stepping.  */
5258 			vcpu->arch.pio.count = 0;
5259 		} else {
5260 			writeback = false;
5261 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5262 		}
5263 		r = EMULATE_USER_EXIT;
5264 	} else if (vcpu->mmio_needed) {
5265 		if (!vcpu->mmio_is_write)
5266 			writeback = false;
5267 		r = EMULATE_USER_EXIT;
5268 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5269 	} else if (r == EMULATION_RESTART)
5270 		goto restart;
5271 	else
5272 		r = EMULATE_DONE;
5273 
5274 	if (writeback) {
5275 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5276 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5277 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5278 		if (vcpu->arch.hflags != ctxt->emul_flags)
5279 			kvm_set_hflags(vcpu, ctxt->emul_flags);
5280 		kvm_rip_write(vcpu, ctxt->eip);
5281 		if (r == EMULATE_DONE)
5282 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5283 		if (!ctxt->have_exception ||
5284 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5285 			__kvm_set_rflags(vcpu, ctxt->eflags);
5286 
5287 		/*
5288 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5289 		 * do nothing, and it will be requested again as soon as
5290 		 * the shadow expires.  But we still need to check here,
5291 		 * because POPF has no interrupt shadow.
5292 		 */
5293 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5294 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5295 	} else
5296 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5297 
5298 	return r;
5299 }
5300 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5301 
5302 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5303 {
5304 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5305 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5306 					    size, port, &val, 1);
5307 	/* do not return to emulator after return from userspace */
5308 	vcpu->arch.pio.count = 0;
5309 	return ret;
5310 }
5311 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5312 
5313 static void tsc_bad(void *info)
5314 {
5315 	__this_cpu_write(cpu_tsc_khz, 0);
5316 }
5317 
5318 static void tsc_khz_changed(void *data)
5319 {
5320 	struct cpufreq_freqs *freq = data;
5321 	unsigned long khz = 0;
5322 
5323 	if (data)
5324 		khz = freq->new;
5325 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5326 		khz = cpufreq_quick_get(raw_smp_processor_id());
5327 	if (!khz)
5328 		khz = tsc_khz;
5329 	__this_cpu_write(cpu_tsc_khz, khz);
5330 }
5331 
5332 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5333 				     void *data)
5334 {
5335 	struct cpufreq_freqs *freq = data;
5336 	struct kvm *kvm;
5337 	struct kvm_vcpu *vcpu;
5338 	int i, send_ipi = 0;
5339 
5340 	/*
5341 	 * We allow guests to temporarily run on slowing clocks,
5342 	 * provided we notify them after, or to run on accelerating
5343 	 * clocks, provided we notify them before.  Thus time never
5344 	 * goes backwards.
5345 	 *
5346 	 * However, we have a problem.  We can't atomically update
5347 	 * the frequency of a given CPU from this function; it is
5348 	 * merely a notifier, which can be called from any CPU.
5349 	 * Changing the TSC frequency at arbitrary points in time
5350 	 * requires a recomputation of local variables related to
5351 	 * the TSC for each VCPU.  We must flag these local variables
5352 	 * to be updated and be sure the update takes place with the
5353 	 * new frequency before any guests proceed.
5354 	 *
5355 	 * Unfortunately, the combination of hotplug CPU and frequency
5356 	 * change creates an intractable locking scenario; the order
5357 	 * of when these callouts happen is undefined with respect to
5358 	 * CPU hotplug, and they can race with each other.  As such,
5359 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5360 	 * undefined; you can actually have a CPU frequency change take
5361 	 * place in between the computation of X and the setting of the
5362 	 * variable.  To protect against this problem, all updates of
5363 	 * the per_cpu tsc_khz variable are done in an interrupt
5364 	 * protected IPI, and all callers wishing to update the value
5365 	 * must wait for a synchronous IPI to complete (which is trivial
5366 	 * if the caller is on the CPU already).  This establishes the
5367 	 * necessary total order on variable updates.
5368 	 *
5369 	 * Note that because a guest time update may take place
5370 	 * anytime after the setting of the VCPU's request bit, the
5371 	 * correct TSC value must be set before the request.  However,
5372 	 * to ensure the update actually makes it to any guest which
5373 	 * starts running in hardware virtualization between the set
5374 	 * and the acquisition of the spinlock, we must also ping the
5375 	 * CPU after setting the request bit.
5376 	 *
5377 	 */
5378 
5379 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5380 		return 0;
5381 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5382 		return 0;
5383 
5384 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5385 
5386 	spin_lock(&kvm_lock);
5387 	list_for_each_entry(kvm, &vm_list, vm_list) {
5388 		kvm_for_each_vcpu(i, vcpu, kvm) {
5389 			if (vcpu->cpu != freq->cpu)
5390 				continue;
5391 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5392 			if (vcpu->cpu != smp_processor_id())
5393 				send_ipi = 1;
5394 		}
5395 	}
5396 	spin_unlock(&kvm_lock);
5397 
5398 	if (freq->old < freq->new && send_ipi) {
5399 		/*
5400 		 * We upscale the frequency.  Must make the guest
5401 		 * doesn't see old kvmclock values while running with
5402 		 * the new frequency, otherwise we risk the guest sees
5403 		 * time go backwards.
5404 		 *
5405 		 * In case we update the frequency for another cpu
5406 		 * (which might be in guest context) send an interrupt
5407 		 * to kick the cpu out of guest context.  Next time
5408 		 * guest context is entered kvmclock will be updated,
5409 		 * so the guest will not see stale values.
5410 		 */
5411 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5412 	}
5413 	return 0;
5414 }
5415 
5416 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5417 	.notifier_call  = kvmclock_cpufreq_notifier
5418 };
5419 
5420 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5421 					unsigned long action, void *hcpu)
5422 {
5423 	unsigned int cpu = (unsigned long)hcpu;
5424 
5425 	switch (action) {
5426 		case CPU_ONLINE:
5427 		case CPU_DOWN_FAILED:
5428 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5429 			break;
5430 		case CPU_DOWN_PREPARE:
5431 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5432 			break;
5433 	}
5434 	return NOTIFY_OK;
5435 }
5436 
5437 static struct notifier_block kvmclock_cpu_notifier_block = {
5438 	.notifier_call  = kvmclock_cpu_notifier,
5439 	.priority = -INT_MAX
5440 };
5441 
5442 static void kvm_timer_init(void)
5443 {
5444 	int cpu;
5445 
5446 	max_tsc_khz = tsc_khz;
5447 
5448 	cpu_notifier_register_begin();
5449 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5450 #ifdef CONFIG_CPU_FREQ
5451 		struct cpufreq_policy policy;
5452 		memset(&policy, 0, sizeof(policy));
5453 		cpu = get_cpu();
5454 		cpufreq_get_policy(&policy, cpu);
5455 		if (policy.cpuinfo.max_freq)
5456 			max_tsc_khz = policy.cpuinfo.max_freq;
5457 		put_cpu();
5458 #endif
5459 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5460 					  CPUFREQ_TRANSITION_NOTIFIER);
5461 	}
5462 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5463 	for_each_online_cpu(cpu)
5464 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5465 
5466 	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5467 	cpu_notifier_register_done();
5468 
5469 }
5470 
5471 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5472 
5473 int kvm_is_in_guest(void)
5474 {
5475 	return __this_cpu_read(current_vcpu) != NULL;
5476 }
5477 
5478 static int kvm_is_user_mode(void)
5479 {
5480 	int user_mode = 3;
5481 
5482 	if (__this_cpu_read(current_vcpu))
5483 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5484 
5485 	return user_mode != 0;
5486 }
5487 
5488 static unsigned long kvm_get_guest_ip(void)
5489 {
5490 	unsigned long ip = 0;
5491 
5492 	if (__this_cpu_read(current_vcpu))
5493 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5494 
5495 	return ip;
5496 }
5497 
5498 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5499 	.is_in_guest		= kvm_is_in_guest,
5500 	.is_user_mode		= kvm_is_user_mode,
5501 	.get_guest_ip		= kvm_get_guest_ip,
5502 };
5503 
5504 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5505 {
5506 	__this_cpu_write(current_vcpu, vcpu);
5507 }
5508 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5509 
5510 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5511 {
5512 	__this_cpu_write(current_vcpu, NULL);
5513 }
5514 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5515 
5516 static void kvm_set_mmio_spte_mask(void)
5517 {
5518 	u64 mask;
5519 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5520 
5521 	/*
5522 	 * Set the reserved bits and the present bit of an paging-structure
5523 	 * entry to generate page fault with PFER.RSV = 1.
5524 	 */
5525 	 /* Mask the reserved physical address bits. */
5526 	mask = rsvd_bits(maxphyaddr, 51);
5527 
5528 	/* Bit 62 is always reserved for 32bit host. */
5529 	mask |= 0x3ull << 62;
5530 
5531 	/* Set the present bit. */
5532 	mask |= 1ull;
5533 
5534 #ifdef CONFIG_X86_64
5535 	/*
5536 	 * If reserved bit is not supported, clear the present bit to disable
5537 	 * mmio page fault.
5538 	 */
5539 	if (maxphyaddr == 52)
5540 		mask &= ~1ull;
5541 #endif
5542 
5543 	kvm_mmu_set_mmio_spte_mask(mask);
5544 }
5545 
5546 #ifdef CONFIG_X86_64
5547 static void pvclock_gtod_update_fn(struct work_struct *work)
5548 {
5549 	struct kvm *kvm;
5550 
5551 	struct kvm_vcpu *vcpu;
5552 	int i;
5553 
5554 	spin_lock(&kvm_lock);
5555 	list_for_each_entry(kvm, &vm_list, vm_list)
5556 		kvm_for_each_vcpu(i, vcpu, kvm)
5557 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5558 	atomic_set(&kvm_guest_has_master_clock, 0);
5559 	spin_unlock(&kvm_lock);
5560 }
5561 
5562 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5563 
5564 /*
5565  * Notification about pvclock gtod data update.
5566  */
5567 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5568 			       void *priv)
5569 {
5570 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5571 	struct timekeeper *tk = priv;
5572 
5573 	update_pvclock_gtod(tk);
5574 
5575 	/* disable master clock if host does not trust, or does not
5576 	 * use, TSC clocksource
5577 	 */
5578 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5579 	    atomic_read(&kvm_guest_has_master_clock) != 0)
5580 		queue_work(system_long_wq, &pvclock_gtod_work);
5581 
5582 	return 0;
5583 }
5584 
5585 static struct notifier_block pvclock_gtod_notifier = {
5586 	.notifier_call = pvclock_gtod_notify,
5587 };
5588 #endif
5589 
5590 int kvm_arch_init(void *opaque)
5591 {
5592 	int r;
5593 	struct kvm_x86_ops *ops = opaque;
5594 
5595 	if (kvm_x86_ops) {
5596 		printk(KERN_ERR "kvm: already loaded the other module\n");
5597 		r = -EEXIST;
5598 		goto out;
5599 	}
5600 
5601 	if (!ops->cpu_has_kvm_support()) {
5602 		printk(KERN_ERR "kvm: no hardware support\n");
5603 		r = -EOPNOTSUPP;
5604 		goto out;
5605 	}
5606 	if (ops->disabled_by_bios()) {
5607 		printk(KERN_ERR "kvm: disabled by bios\n");
5608 		r = -EOPNOTSUPP;
5609 		goto out;
5610 	}
5611 
5612 	r = -ENOMEM;
5613 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5614 	if (!shared_msrs) {
5615 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5616 		goto out;
5617 	}
5618 
5619 	r = kvm_mmu_module_init();
5620 	if (r)
5621 		goto out_free_percpu;
5622 
5623 	kvm_set_mmio_spte_mask();
5624 
5625 	kvm_x86_ops = ops;
5626 
5627 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5628 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5629 
5630 	kvm_timer_init();
5631 
5632 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5633 
5634 	if (cpu_has_xsave)
5635 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5636 
5637 	kvm_lapic_init();
5638 #ifdef CONFIG_X86_64
5639 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5640 #endif
5641 
5642 	return 0;
5643 
5644 out_free_percpu:
5645 	free_percpu(shared_msrs);
5646 out:
5647 	return r;
5648 }
5649 
5650 void kvm_arch_exit(void)
5651 {
5652 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5653 
5654 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5655 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5656 					    CPUFREQ_TRANSITION_NOTIFIER);
5657 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5658 #ifdef CONFIG_X86_64
5659 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5660 #endif
5661 	kvm_x86_ops = NULL;
5662 	kvm_mmu_module_exit();
5663 	free_percpu(shared_msrs);
5664 }
5665 
5666 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5667 {
5668 	++vcpu->stat.halt_exits;
5669 	if (irqchip_in_kernel(vcpu->kvm)) {
5670 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5671 		return 1;
5672 	} else {
5673 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5674 		return 0;
5675 	}
5676 }
5677 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5678 
5679 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5680 {
5681 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5682 	return kvm_vcpu_halt(vcpu);
5683 }
5684 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5685 
5686 /*
5687  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5688  *
5689  * @apicid - apicid of vcpu to be kicked.
5690  */
5691 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5692 {
5693 	struct kvm_lapic_irq lapic_irq;
5694 
5695 	lapic_irq.shorthand = 0;
5696 	lapic_irq.dest_mode = 0;
5697 	lapic_irq.dest_id = apicid;
5698 	lapic_irq.msi_redir_hint = false;
5699 
5700 	lapic_irq.delivery_mode = APIC_DM_REMRD;
5701 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5702 }
5703 
5704 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5705 {
5706 	unsigned long nr, a0, a1, a2, a3, ret;
5707 	int op_64_bit, r = 1;
5708 
5709 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5710 
5711 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5712 		return kvm_hv_hypercall(vcpu);
5713 
5714 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5715 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5716 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5717 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5718 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5719 
5720 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5721 
5722 	op_64_bit = is_64_bit_mode(vcpu);
5723 	if (!op_64_bit) {
5724 		nr &= 0xFFFFFFFF;
5725 		a0 &= 0xFFFFFFFF;
5726 		a1 &= 0xFFFFFFFF;
5727 		a2 &= 0xFFFFFFFF;
5728 		a3 &= 0xFFFFFFFF;
5729 	}
5730 
5731 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5732 		ret = -KVM_EPERM;
5733 		goto out;
5734 	}
5735 
5736 	switch (nr) {
5737 	case KVM_HC_VAPIC_POLL_IRQ:
5738 		ret = 0;
5739 		break;
5740 	case KVM_HC_KICK_CPU:
5741 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5742 		ret = 0;
5743 		break;
5744 	default:
5745 		ret = -KVM_ENOSYS;
5746 		break;
5747 	}
5748 out:
5749 	if (!op_64_bit)
5750 		ret = (u32)ret;
5751 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5752 	++vcpu->stat.hypercalls;
5753 	return r;
5754 }
5755 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5756 
5757 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5758 {
5759 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5760 	char instruction[3];
5761 	unsigned long rip = kvm_rip_read(vcpu);
5762 
5763 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
5764 
5765 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5766 }
5767 
5768 /*
5769  * Check if userspace requested an interrupt window, and that the
5770  * interrupt window is open.
5771  *
5772  * No need to exit to userspace if we already have an interrupt queued.
5773  */
5774 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5775 {
5776 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5777 		vcpu->run->request_interrupt_window &&
5778 		kvm_arch_interrupt_allowed(vcpu));
5779 }
5780 
5781 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5782 {
5783 	struct kvm_run *kvm_run = vcpu->run;
5784 
5785 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5786 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5787 	kvm_run->cr8 = kvm_get_cr8(vcpu);
5788 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
5789 	if (irqchip_in_kernel(vcpu->kvm))
5790 		kvm_run->ready_for_interrupt_injection = 1;
5791 	else
5792 		kvm_run->ready_for_interrupt_injection =
5793 			kvm_arch_interrupt_allowed(vcpu) &&
5794 			!kvm_cpu_has_interrupt(vcpu) &&
5795 			!kvm_event_needs_reinjection(vcpu);
5796 }
5797 
5798 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5799 {
5800 	int max_irr, tpr;
5801 
5802 	if (!kvm_x86_ops->update_cr8_intercept)
5803 		return;
5804 
5805 	if (!vcpu->arch.apic)
5806 		return;
5807 
5808 	if (!vcpu->arch.apic->vapic_addr)
5809 		max_irr = kvm_lapic_find_highest_irr(vcpu);
5810 	else
5811 		max_irr = -1;
5812 
5813 	if (max_irr != -1)
5814 		max_irr >>= 4;
5815 
5816 	tpr = kvm_lapic_get_cr8(vcpu);
5817 
5818 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5819 }
5820 
5821 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5822 {
5823 	int r;
5824 
5825 	/* try to reinject previous events if any */
5826 	if (vcpu->arch.exception.pending) {
5827 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
5828 					vcpu->arch.exception.has_error_code,
5829 					vcpu->arch.exception.error_code);
5830 
5831 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5832 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5833 					     X86_EFLAGS_RF);
5834 
5835 		if (vcpu->arch.exception.nr == DB_VECTOR &&
5836 		    (vcpu->arch.dr7 & DR7_GD)) {
5837 			vcpu->arch.dr7 &= ~DR7_GD;
5838 			kvm_update_dr7(vcpu);
5839 		}
5840 
5841 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5842 					  vcpu->arch.exception.has_error_code,
5843 					  vcpu->arch.exception.error_code,
5844 					  vcpu->arch.exception.reinject);
5845 		return 0;
5846 	}
5847 
5848 	if (vcpu->arch.nmi_injected) {
5849 		kvm_x86_ops->set_nmi(vcpu);
5850 		return 0;
5851 	}
5852 
5853 	if (vcpu->arch.interrupt.pending) {
5854 		kvm_x86_ops->set_irq(vcpu);
5855 		return 0;
5856 	}
5857 
5858 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5859 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5860 		if (r != 0)
5861 			return r;
5862 	}
5863 
5864 	/* try to inject new event if pending */
5865 	if (vcpu->arch.nmi_pending) {
5866 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
5867 			--vcpu->arch.nmi_pending;
5868 			vcpu->arch.nmi_injected = true;
5869 			kvm_x86_ops->set_nmi(vcpu);
5870 		}
5871 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
5872 		/*
5873 		 * Because interrupts can be injected asynchronously, we are
5874 		 * calling check_nested_events again here to avoid a race condition.
5875 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5876 		 * proposal and current concerns.  Perhaps we should be setting
5877 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
5878 		 */
5879 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5880 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5881 			if (r != 0)
5882 				return r;
5883 		}
5884 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5885 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5886 					    false);
5887 			kvm_x86_ops->set_irq(vcpu);
5888 		}
5889 	}
5890 	return 0;
5891 }
5892 
5893 static void process_nmi(struct kvm_vcpu *vcpu)
5894 {
5895 	unsigned limit = 2;
5896 
5897 	/*
5898 	 * x86 is limited to one NMI running, and one NMI pending after it.
5899 	 * If an NMI is already in progress, limit further NMIs to just one.
5900 	 * Otherwise, allow two (and we'll inject the first one immediately).
5901 	 */
5902 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5903 		limit = 1;
5904 
5905 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5906 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5907 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5908 }
5909 
5910 #define put_smstate(type, buf, offset, val)			  \
5911 	*(type *)((buf) + (offset) - 0x7e00) = val
5912 
5913 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
5914 {
5915 	u32 flags = 0;
5916 	flags |= seg->g       << 23;
5917 	flags |= seg->db      << 22;
5918 	flags |= seg->l       << 21;
5919 	flags |= seg->avl     << 20;
5920 	flags |= seg->present << 15;
5921 	flags |= seg->dpl     << 13;
5922 	flags |= seg->s       << 12;
5923 	flags |= seg->type    << 8;
5924 	return flags;
5925 }
5926 
5927 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
5928 {
5929 	struct kvm_segment seg;
5930 	int offset;
5931 
5932 	kvm_get_segment(vcpu, &seg, n);
5933 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
5934 
5935 	if (n < 3)
5936 		offset = 0x7f84 + n * 12;
5937 	else
5938 		offset = 0x7f2c + (n - 3) * 12;
5939 
5940 	put_smstate(u32, buf, offset + 8, seg.base);
5941 	put_smstate(u32, buf, offset + 4, seg.limit);
5942 	put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
5943 }
5944 
5945 #ifdef CONFIG_X86_64
5946 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
5947 {
5948 	struct kvm_segment seg;
5949 	int offset;
5950 	u16 flags;
5951 
5952 	kvm_get_segment(vcpu, &seg, n);
5953 	offset = 0x7e00 + n * 16;
5954 
5955 	flags = process_smi_get_segment_flags(&seg) >> 8;
5956 	put_smstate(u16, buf, offset, seg.selector);
5957 	put_smstate(u16, buf, offset + 2, flags);
5958 	put_smstate(u32, buf, offset + 4, seg.limit);
5959 	put_smstate(u64, buf, offset + 8, seg.base);
5960 }
5961 #endif
5962 
5963 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
5964 {
5965 	struct desc_ptr dt;
5966 	struct kvm_segment seg;
5967 	unsigned long val;
5968 	int i;
5969 
5970 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
5971 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
5972 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
5973 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
5974 
5975 	for (i = 0; i < 8; i++)
5976 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
5977 
5978 	kvm_get_dr(vcpu, 6, &val);
5979 	put_smstate(u32, buf, 0x7fcc, (u32)val);
5980 	kvm_get_dr(vcpu, 7, &val);
5981 	put_smstate(u32, buf, 0x7fc8, (u32)val);
5982 
5983 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
5984 	put_smstate(u32, buf, 0x7fc4, seg.selector);
5985 	put_smstate(u32, buf, 0x7f64, seg.base);
5986 	put_smstate(u32, buf, 0x7f60, seg.limit);
5987 	put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
5988 
5989 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
5990 	put_smstate(u32, buf, 0x7fc0, seg.selector);
5991 	put_smstate(u32, buf, 0x7f80, seg.base);
5992 	put_smstate(u32, buf, 0x7f7c, seg.limit);
5993 	put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
5994 
5995 	kvm_x86_ops->get_gdt(vcpu, &dt);
5996 	put_smstate(u32, buf, 0x7f74, dt.address);
5997 	put_smstate(u32, buf, 0x7f70, dt.size);
5998 
5999 	kvm_x86_ops->get_idt(vcpu, &dt);
6000 	put_smstate(u32, buf, 0x7f58, dt.address);
6001 	put_smstate(u32, buf, 0x7f54, dt.size);
6002 
6003 	for (i = 0; i < 6; i++)
6004 		process_smi_save_seg_32(vcpu, buf, i);
6005 
6006 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6007 
6008 	/* revision id */
6009 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6010 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6011 }
6012 
6013 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6014 {
6015 #ifdef CONFIG_X86_64
6016 	struct desc_ptr dt;
6017 	struct kvm_segment seg;
6018 	unsigned long val;
6019 	int i;
6020 
6021 	for (i = 0; i < 16; i++)
6022 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6023 
6024 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6025 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6026 
6027 	kvm_get_dr(vcpu, 6, &val);
6028 	put_smstate(u64, buf, 0x7f68, val);
6029 	kvm_get_dr(vcpu, 7, &val);
6030 	put_smstate(u64, buf, 0x7f60, val);
6031 
6032 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6033 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6034 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6035 
6036 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6037 
6038 	/* revision id */
6039 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6040 
6041 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6042 
6043 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6044 	put_smstate(u16, buf, 0x7e90, seg.selector);
6045 	put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6046 	put_smstate(u32, buf, 0x7e94, seg.limit);
6047 	put_smstate(u64, buf, 0x7e98, seg.base);
6048 
6049 	kvm_x86_ops->get_idt(vcpu, &dt);
6050 	put_smstate(u32, buf, 0x7e84, dt.size);
6051 	put_smstate(u64, buf, 0x7e88, dt.address);
6052 
6053 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6054 	put_smstate(u16, buf, 0x7e70, seg.selector);
6055 	put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6056 	put_smstate(u32, buf, 0x7e74, seg.limit);
6057 	put_smstate(u64, buf, 0x7e78, seg.base);
6058 
6059 	kvm_x86_ops->get_gdt(vcpu, &dt);
6060 	put_smstate(u32, buf, 0x7e64, dt.size);
6061 	put_smstate(u64, buf, 0x7e68, dt.address);
6062 
6063 	for (i = 0; i < 6; i++)
6064 		process_smi_save_seg_64(vcpu, buf, i);
6065 #else
6066 	WARN_ON_ONCE(1);
6067 #endif
6068 }
6069 
6070 static void process_smi(struct kvm_vcpu *vcpu)
6071 {
6072 	struct kvm_segment cs, ds;
6073 	struct desc_ptr dt;
6074 	char buf[512];
6075 	u32 cr0;
6076 
6077 	if (is_smm(vcpu)) {
6078 		vcpu->arch.smi_pending = true;
6079 		return;
6080 	}
6081 
6082 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6083 	vcpu->arch.hflags |= HF_SMM_MASK;
6084 	memset(buf, 0, 512);
6085 	if (guest_cpuid_has_longmode(vcpu))
6086 		process_smi_save_state_64(vcpu, buf);
6087 	else
6088 		process_smi_save_state_32(vcpu, buf);
6089 
6090 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6091 
6092 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6093 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6094 	else
6095 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6096 
6097 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6098 	kvm_rip_write(vcpu, 0x8000);
6099 
6100 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6101 	kvm_x86_ops->set_cr0(vcpu, cr0);
6102 	vcpu->arch.cr0 = cr0;
6103 
6104 	kvm_x86_ops->set_cr4(vcpu, 0);
6105 
6106 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6107 	dt.address = dt.size = 0;
6108 	kvm_x86_ops->set_idt(vcpu, &dt);
6109 
6110 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6111 
6112 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6113 	cs.base = vcpu->arch.smbase;
6114 
6115 	ds.selector = 0;
6116 	ds.base = 0;
6117 
6118 	cs.limit    = ds.limit = 0xffffffff;
6119 	cs.type     = ds.type = 0x3;
6120 	cs.dpl      = ds.dpl = 0;
6121 	cs.db       = ds.db = 0;
6122 	cs.s        = ds.s = 1;
6123 	cs.l        = ds.l = 0;
6124 	cs.g        = ds.g = 1;
6125 	cs.avl      = ds.avl = 0;
6126 	cs.present  = ds.present = 1;
6127 	cs.unusable = ds.unusable = 0;
6128 	cs.padding  = ds.padding = 0;
6129 
6130 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6131 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6132 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6133 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6134 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6135 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6136 
6137 	if (guest_cpuid_has_longmode(vcpu))
6138 		kvm_x86_ops->set_efer(vcpu, 0);
6139 
6140 	kvm_update_cpuid(vcpu);
6141 	kvm_mmu_reset_context(vcpu);
6142 }
6143 
6144 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6145 {
6146 	u64 eoi_exit_bitmap[4];
6147 	u32 tmr[8];
6148 
6149 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6150 		return;
6151 
6152 	memset(eoi_exit_bitmap, 0, 32);
6153 	memset(tmr, 0, 32);
6154 
6155 	kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6156 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6157 	kvm_apic_update_tmr(vcpu, tmr);
6158 }
6159 
6160 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6161 {
6162 	++vcpu->stat.tlb_flush;
6163 	kvm_x86_ops->tlb_flush(vcpu);
6164 }
6165 
6166 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6167 {
6168 	struct page *page = NULL;
6169 
6170 	if (!irqchip_in_kernel(vcpu->kvm))
6171 		return;
6172 
6173 	if (!kvm_x86_ops->set_apic_access_page_addr)
6174 		return;
6175 
6176 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6177 	if (is_error_page(page))
6178 		return;
6179 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6180 
6181 	/*
6182 	 * Do not pin apic access page in memory, the MMU notifier
6183 	 * will call us again if it is migrated or swapped out.
6184 	 */
6185 	put_page(page);
6186 }
6187 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6188 
6189 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6190 					   unsigned long address)
6191 {
6192 	/*
6193 	 * The physical address of apic access page is stored in the VMCS.
6194 	 * Update it when it becomes invalid.
6195 	 */
6196 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6197 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6198 }
6199 
6200 /*
6201  * Returns 1 to let vcpu_run() continue the guest execution loop without
6202  * exiting to the userspace.  Otherwise, the value will be returned to the
6203  * userspace.
6204  */
6205 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6206 {
6207 	int r;
6208 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6209 		vcpu->run->request_interrupt_window;
6210 	bool req_immediate_exit = false;
6211 
6212 	if (vcpu->requests) {
6213 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6214 			kvm_mmu_unload(vcpu);
6215 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6216 			__kvm_migrate_timers(vcpu);
6217 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6218 			kvm_gen_update_masterclock(vcpu->kvm);
6219 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6220 			kvm_gen_kvmclock_update(vcpu);
6221 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6222 			r = kvm_guest_time_update(vcpu);
6223 			if (unlikely(r))
6224 				goto out;
6225 		}
6226 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6227 			kvm_mmu_sync_roots(vcpu);
6228 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6229 			kvm_vcpu_flush_tlb(vcpu);
6230 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6231 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6232 			r = 0;
6233 			goto out;
6234 		}
6235 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6236 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6237 			r = 0;
6238 			goto out;
6239 		}
6240 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6241 			vcpu->fpu_active = 0;
6242 			kvm_x86_ops->fpu_deactivate(vcpu);
6243 		}
6244 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6245 			/* Page is swapped out. Do synthetic halt */
6246 			vcpu->arch.apf.halted = true;
6247 			r = 1;
6248 			goto out;
6249 		}
6250 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6251 			record_steal_time(vcpu);
6252 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6253 			process_smi(vcpu);
6254 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6255 			process_nmi(vcpu);
6256 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6257 			kvm_pmu_handle_event(vcpu);
6258 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6259 			kvm_pmu_deliver_pmi(vcpu);
6260 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6261 			vcpu_scan_ioapic(vcpu);
6262 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6263 			kvm_vcpu_reload_apic_access_page(vcpu);
6264 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6265 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6266 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6267 			r = 0;
6268 			goto out;
6269 		}
6270 	}
6271 
6272 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6273 		kvm_apic_accept_events(vcpu);
6274 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6275 			r = 1;
6276 			goto out;
6277 		}
6278 
6279 		if (inject_pending_event(vcpu, req_int_win) != 0)
6280 			req_immediate_exit = true;
6281 		/* enable NMI/IRQ window open exits if needed */
6282 		else if (vcpu->arch.nmi_pending)
6283 			kvm_x86_ops->enable_nmi_window(vcpu);
6284 		else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6285 			kvm_x86_ops->enable_irq_window(vcpu);
6286 
6287 		if (kvm_lapic_enabled(vcpu)) {
6288 			/*
6289 			 * Update architecture specific hints for APIC
6290 			 * virtual interrupt delivery.
6291 			 */
6292 			if (kvm_x86_ops->hwapic_irr_update)
6293 				kvm_x86_ops->hwapic_irr_update(vcpu,
6294 					kvm_lapic_find_highest_irr(vcpu));
6295 			update_cr8_intercept(vcpu);
6296 			kvm_lapic_sync_to_vapic(vcpu);
6297 		}
6298 	}
6299 
6300 	r = kvm_mmu_reload(vcpu);
6301 	if (unlikely(r)) {
6302 		goto cancel_injection;
6303 	}
6304 
6305 	preempt_disable();
6306 
6307 	kvm_x86_ops->prepare_guest_switch(vcpu);
6308 	if (vcpu->fpu_active)
6309 		kvm_load_guest_fpu(vcpu);
6310 	kvm_load_guest_xcr0(vcpu);
6311 
6312 	vcpu->mode = IN_GUEST_MODE;
6313 
6314 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6315 
6316 	/* We should set ->mode before check ->requests,
6317 	 * see the comment in make_all_cpus_request.
6318 	 */
6319 	smp_mb__after_srcu_read_unlock();
6320 
6321 	local_irq_disable();
6322 
6323 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6324 	    || need_resched() || signal_pending(current)) {
6325 		vcpu->mode = OUTSIDE_GUEST_MODE;
6326 		smp_wmb();
6327 		local_irq_enable();
6328 		preempt_enable();
6329 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6330 		r = 1;
6331 		goto cancel_injection;
6332 	}
6333 
6334 	if (req_immediate_exit)
6335 		smp_send_reschedule(vcpu->cpu);
6336 
6337 	__kvm_guest_enter();
6338 
6339 	if (unlikely(vcpu->arch.switch_db_regs)) {
6340 		set_debugreg(0, 7);
6341 		set_debugreg(vcpu->arch.eff_db[0], 0);
6342 		set_debugreg(vcpu->arch.eff_db[1], 1);
6343 		set_debugreg(vcpu->arch.eff_db[2], 2);
6344 		set_debugreg(vcpu->arch.eff_db[3], 3);
6345 		set_debugreg(vcpu->arch.dr6, 6);
6346 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6347 	}
6348 
6349 	trace_kvm_entry(vcpu->vcpu_id);
6350 	wait_lapic_expire(vcpu);
6351 	kvm_x86_ops->run(vcpu);
6352 
6353 	/*
6354 	 * Do this here before restoring debug registers on the host.  And
6355 	 * since we do this before handling the vmexit, a DR access vmexit
6356 	 * can (a) read the correct value of the debug registers, (b) set
6357 	 * KVM_DEBUGREG_WONT_EXIT again.
6358 	 */
6359 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6360 		int i;
6361 
6362 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6363 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6364 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6365 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6366 	}
6367 
6368 	/*
6369 	 * If the guest has used debug registers, at least dr7
6370 	 * will be disabled while returning to the host.
6371 	 * If we don't have active breakpoints in the host, we don't
6372 	 * care about the messed up debug address registers. But if
6373 	 * we have some of them active, restore the old state.
6374 	 */
6375 	if (hw_breakpoint_active())
6376 		hw_breakpoint_restore();
6377 
6378 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6379 							   rdtsc());
6380 
6381 	vcpu->mode = OUTSIDE_GUEST_MODE;
6382 	smp_wmb();
6383 
6384 	/* Interrupt is enabled by handle_external_intr() */
6385 	kvm_x86_ops->handle_external_intr(vcpu);
6386 
6387 	++vcpu->stat.exits;
6388 
6389 	/*
6390 	 * We must have an instruction between local_irq_enable() and
6391 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6392 	 * the interrupt shadow.  The stat.exits increment will do nicely.
6393 	 * But we need to prevent reordering, hence this barrier():
6394 	 */
6395 	barrier();
6396 
6397 	kvm_guest_exit();
6398 
6399 	preempt_enable();
6400 
6401 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6402 
6403 	/*
6404 	 * Profile KVM exit RIPs:
6405 	 */
6406 	if (unlikely(prof_on == KVM_PROFILING)) {
6407 		unsigned long rip = kvm_rip_read(vcpu);
6408 		profile_hit(KVM_PROFILING, (void *)rip);
6409 	}
6410 
6411 	if (unlikely(vcpu->arch.tsc_always_catchup))
6412 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6413 
6414 	if (vcpu->arch.apic_attention)
6415 		kvm_lapic_sync_from_vapic(vcpu);
6416 
6417 	r = kvm_x86_ops->handle_exit(vcpu);
6418 	return r;
6419 
6420 cancel_injection:
6421 	kvm_x86_ops->cancel_injection(vcpu);
6422 	if (unlikely(vcpu->arch.apic_attention))
6423 		kvm_lapic_sync_from_vapic(vcpu);
6424 out:
6425 	return r;
6426 }
6427 
6428 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6429 {
6430 	if (!kvm_arch_vcpu_runnable(vcpu)) {
6431 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6432 		kvm_vcpu_block(vcpu);
6433 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6434 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6435 			return 1;
6436 	}
6437 
6438 	kvm_apic_accept_events(vcpu);
6439 	switch(vcpu->arch.mp_state) {
6440 	case KVM_MP_STATE_HALTED:
6441 		vcpu->arch.pv.pv_unhalted = false;
6442 		vcpu->arch.mp_state =
6443 			KVM_MP_STATE_RUNNABLE;
6444 	case KVM_MP_STATE_RUNNABLE:
6445 		vcpu->arch.apf.halted = false;
6446 		break;
6447 	case KVM_MP_STATE_INIT_RECEIVED:
6448 		break;
6449 	default:
6450 		return -EINTR;
6451 		break;
6452 	}
6453 	return 1;
6454 }
6455 
6456 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6457 {
6458 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6459 		!vcpu->arch.apf.halted);
6460 }
6461 
6462 static int vcpu_run(struct kvm_vcpu *vcpu)
6463 {
6464 	int r;
6465 	struct kvm *kvm = vcpu->kvm;
6466 
6467 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6468 
6469 	for (;;) {
6470 		if (kvm_vcpu_running(vcpu))
6471 			r = vcpu_enter_guest(vcpu);
6472 		else
6473 			r = vcpu_block(kvm, vcpu);
6474 		if (r <= 0)
6475 			break;
6476 
6477 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6478 		if (kvm_cpu_has_pending_timer(vcpu))
6479 			kvm_inject_pending_timer_irqs(vcpu);
6480 
6481 		if (dm_request_for_irq_injection(vcpu)) {
6482 			r = -EINTR;
6483 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6484 			++vcpu->stat.request_irq_exits;
6485 			break;
6486 		}
6487 
6488 		kvm_check_async_pf_completion(vcpu);
6489 
6490 		if (signal_pending(current)) {
6491 			r = -EINTR;
6492 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6493 			++vcpu->stat.signal_exits;
6494 			break;
6495 		}
6496 		if (need_resched()) {
6497 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6498 			cond_resched();
6499 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6500 		}
6501 	}
6502 
6503 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6504 
6505 	return r;
6506 }
6507 
6508 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6509 {
6510 	int r;
6511 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6512 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6513 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6514 	if (r != EMULATE_DONE)
6515 		return 0;
6516 	return 1;
6517 }
6518 
6519 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6520 {
6521 	BUG_ON(!vcpu->arch.pio.count);
6522 
6523 	return complete_emulated_io(vcpu);
6524 }
6525 
6526 /*
6527  * Implements the following, as a state machine:
6528  *
6529  * read:
6530  *   for each fragment
6531  *     for each mmio piece in the fragment
6532  *       write gpa, len
6533  *       exit
6534  *       copy data
6535  *   execute insn
6536  *
6537  * write:
6538  *   for each fragment
6539  *     for each mmio piece in the fragment
6540  *       write gpa, len
6541  *       copy data
6542  *       exit
6543  */
6544 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6545 {
6546 	struct kvm_run *run = vcpu->run;
6547 	struct kvm_mmio_fragment *frag;
6548 	unsigned len;
6549 
6550 	BUG_ON(!vcpu->mmio_needed);
6551 
6552 	/* Complete previous fragment */
6553 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6554 	len = min(8u, frag->len);
6555 	if (!vcpu->mmio_is_write)
6556 		memcpy(frag->data, run->mmio.data, len);
6557 
6558 	if (frag->len <= 8) {
6559 		/* Switch to the next fragment. */
6560 		frag++;
6561 		vcpu->mmio_cur_fragment++;
6562 	} else {
6563 		/* Go forward to the next mmio piece. */
6564 		frag->data += len;
6565 		frag->gpa += len;
6566 		frag->len -= len;
6567 	}
6568 
6569 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6570 		vcpu->mmio_needed = 0;
6571 
6572 		/* FIXME: return into emulator if single-stepping.  */
6573 		if (vcpu->mmio_is_write)
6574 			return 1;
6575 		vcpu->mmio_read_completed = 1;
6576 		return complete_emulated_io(vcpu);
6577 	}
6578 
6579 	run->exit_reason = KVM_EXIT_MMIO;
6580 	run->mmio.phys_addr = frag->gpa;
6581 	if (vcpu->mmio_is_write)
6582 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6583 	run->mmio.len = min(8u, frag->len);
6584 	run->mmio.is_write = vcpu->mmio_is_write;
6585 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6586 	return 0;
6587 }
6588 
6589 
6590 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6591 {
6592 	struct fpu *fpu = &current->thread.fpu;
6593 	int r;
6594 	sigset_t sigsaved;
6595 
6596 	fpu__activate_curr(fpu);
6597 
6598 	if (vcpu->sigset_active)
6599 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6600 
6601 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6602 		kvm_vcpu_block(vcpu);
6603 		kvm_apic_accept_events(vcpu);
6604 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6605 		r = -EAGAIN;
6606 		goto out;
6607 	}
6608 
6609 	/* re-sync apic's tpr */
6610 	if (!irqchip_in_kernel(vcpu->kvm)) {
6611 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6612 			r = -EINVAL;
6613 			goto out;
6614 		}
6615 	}
6616 
6617 	if (unlikely(vcpu->arch.complete_userspace_io)) {
6618 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6619 		vcpu->arch.complete_userspace_io = NULL;
6620 		r = cui(vcpu);
6621 		if (r <= 0)
6622 			goto out;
6623 	} else
6624 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6625 
6626 	r = vcpu_run(vcpu);
6627 
6628 out:
6629 	post_kvm_run_save(vcpu);
6630 	if (vcpu->sigset_active)
6631 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6632 
6633 	return r;
6634 }
6635 
6636 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6637 {
6638 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6639 		/*
6640 		 * We are here if userspace calls get_regs() in the middle of
6641 		 * instruction emulation. Registers state needs to be copied
6642 		 * back from emulation context to vcpu. Userspace shouldn't do
6643 		 * that usually, but some bad designed PV devices (vmware
6644 		 * backdoor interface) need this to work
6645 		 */
6646 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6647 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6648 	}
6649 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6650 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6651 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6652 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6653 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6654 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6655 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6656 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6657 #ifdef CONFIG_X86_64
6658 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6659 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6660 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6661 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6662 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6663 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6664 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6665 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6666 #endif
6667 
6668 	regs->rip = kvm_rip_read(vcpu);
6669 	regs->rflags = kvm_get_rflags(vcpu);
6670 
6671 	return 0;
6672 }
6673 
6674 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6675 {
6676 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6677 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6678 
6679 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6680 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6681 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6682 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6683 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6684 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6685 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6686 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6687 #ifdef CONFIG_X86_64
6688 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6689 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6690 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6691 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6692 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6693 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6694 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6695 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6696 #endif
6697 
6698 	kvm_rip_write(vcpu, regs->rip);
6699 	kvm_set_rflags(vcpu, regs->rflags);
6700 
6701 	vcpu->arch.exception.pending = false;
6702 
6703 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6704 
6705 	return 0;
6706 }
6707 
6708 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6709 {
6710 	struct kvm_segment cs;
6711 
6712 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6713 	*db = cs.db;
6714 	*l = cs.l;
6715 }
6716 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6717 
6718 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6719 				  struct kvm_sregs *sregs)
6720 {
6721 	struct desc_ptr dt;
6722 
6723 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6724 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6725 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6726 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6727 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6728 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6729 
6730 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6731 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6732 
6733 	kvm_x86_ops->get_idt(vcpu, &dt);
6734 	sregs->idt.limit = dt.size;
6735 	sregs->idt.base = dt.address;
6736 	kvm_x86_ops->get_gdt(vcpu, &dt);
6737 	sregs->gdt.limit = dt.size;
6738 	sregs->gdt.base = dt.address;
6739 
6740 	sregs->cr0 = kvm_read_cr0(vcpu);
6741 	sregs->cr2 = vcpu->arch.cr2;
6742 	sregs->cr3 = kvm_read_cr3(vcpu);
6743 	sregs->cr4 = kvm_read_cr4(vcpu);
6744 	sregs->cr8 = kvm_get_cr8(vcpu);
6745 	sregs->efer = vcpu->arch.efer;
6746 	sregs->apic_base = kvm_get_apic_base(vcpu);
6747 
6748 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6749 
6750 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6751 		set_bit(vcpu->arch.interrupt.nr,
6752 			(unsigned long *)sregs->interrupt_bitmap);
6753 
6754 	return 0;
6755 }
6756 
6757 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6758 				    struct kvm_mp_state *mp_state)
6759 {
6760 	kvm_apic_accept_events(vcpu);
6761 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6762 					vcpu->arch.pv.pv_unhalted)
6763 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6764 	else
6765 		mp_state->mp_state = vcpu->arch.mp_state;
6766 
6767 	return 0;
6768 }
6769 
6770 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6771 				    struct kvm_mp_state *mp_state)
6772 {
6773 	if (!kvm_vcpu_has_lapic(vcpu) &&
6774 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6775 		return -EINVAL;
6776 
6777 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6778 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6779 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6780 	} else
6781 		vcpu->arch.mp_state = mp_state->mp_state;
6782 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6783 	return 0;
6784 }
6785 
6786 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6787 		    int reason, bool has_error_code, u32 error_code)
6788 {
6789 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6790 	int ret;
6791 
6792 	init_emulate_ctxt(vcpu);
6793 
6794 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6795 				   has_error_code, error_code);
6796 
6797 	if (ret)
6798 		return EMULATE_FAIL;
6799 
6800 	kvm_rip_write(vcpu, ctxt->eip);
6801 	kvm_set_rflags(vcpu, ctxt->eflags);
6802 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6803 	return EMULATE_DONE;
6804 }
6805 EXPORT_SYMBOL_GPL(kvm_task_switch);
6806 
6807 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6808 				  struct kvm_sregs *sregs)
6809 {
6810 	struct msr_data apic_base_msr;
6811 	int mmu_reset_needed = 0;
6812 	int pending_vec, max_bits, idx;
6813 	struct desc_ptr dt;
6814 
6815 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6816 		return -EINVAL;
6817 
6818 	dt.size = sregs->idt.limit;
6819 	dt.address = sregs->idt.base;
6820 	kvm_x86_ops->set_idt(vcpu, &dt);
6821 	dt.size = sregs->gdt.limit;
6822 	dt.address = sregs->gdt.base;
6823 	kvm_x86_ops->set_gdt(vcpu, &dt);
6824 
6825 	vcpu->arch.cr2 = sregs->cr2;
6826 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6827 	vcpu->arch.cr3 = sregs->cr3;
6828 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6829 
6830 	kvm_set_cr8(vcpu, sregs->cr8);
6831 
6832 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6833 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
6834 	apic_base_msr.data = sregs->apic_base;
6835 	apic_base_msr.host_initiated = true;
6836 	kvm_set_apic_base(vcpu, &apic_base_msr);
6837 
6838 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6839 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6840 	vcpu->arch.cr0 = sregs->cr0;
6841 
6842 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6843 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6844 	if (sregs->cr4 & X86_CR4_OSXSAVE)
6845 		kvm_update_cpuid(vcpu);
6846 
6847 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6848 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6849 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6850 		mmu_reset_needed = 1;
6851 	}
6852 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6853 
6854 	if (mmu_reset_needed)
6855 		kvm_mmu_reset_context(vcpu);
6856 
6857 	max_bits = KVM_NR_INTERRUPTS;
6858 	pending_vec = find_first_bit(
6859 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
6860 	if (pending_vec < max_bits) {
6861 		kvm_queue_interrupt(vcpu, pending_vec, false);
6862 		pr_debug("Set back pending irq %d\n", pending_vec);
6863 	}
6864 
6865 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6866 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6867 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6868 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6869 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6870 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6871 
6872 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6873 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6874 
6875 	update_cr8_intercept(vcpu);
6876 
6877 	/* Older userspace won't unhalt the vcpu on reset. */
6878 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6879 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6880 	    !is_protmode(vcpu))
6881 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6882 
6883 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6884 
6885 	return 0;
6886 }
6887 
6888 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6889 					struct kvm_guest_debug *dbg)
6890 {
6891 	unsigned long rflags;
6892 	int i, r;
6893 
6894 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6895 		r = -EBUSY;
6896 		if (vcpu->arch.exception.pending)
6897 			goto out;
6898 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6899 			kvm_queue_exception(vcpu, DB_VECTOR);
6900 		else
6901 			kvm_queue_exception(vcpu, BP_VECTOR);
6902 	}
6903 
6904 	/*
6905 	 * Read rflags as long as potentially injected trace flags are still
6906 	 * filtered out.
6907 	 */
6908 	rflags = kvm_get_rflags(vcpu);
6909 
6910 	vcpu->guest_debug = dbg->control;
6911 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6912 		vcpu->guest_debug = 0;
6913 
6914 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6915 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
6916 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6917 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6918 	} else {
6919 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6920 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6921 	}
6922 	kvm_update_dr7(vcpu);
6923 
6924 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6925 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6926 			get_segment_base(vcpu, VCPU_SREG_CS);
6927 
6928 	/*
6929 	 * Trigger an rflags update that will inject or remove the trace
6930 	 * flags.
6931 	 */
6932 	kvm_set_rflags(vcpu, rflags);
6933 
6934 	kvm_x86_ops->update_db_bp_intercept(vcpu);
6935 
6936 	r = 0;
6937 
6938 out:
6939 
6940 	return r;
6941 }
6942 
6943 /*
6944  * Translate a guest virtual address to a guest physical address.
6945  */
6946 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6947 				    struct kvm_translation *tr)
6948 {
6949 	unsigned long vaddr = tr->linear_address;
6950 	gpa_t gpa;
6951 	int idx;
6952 
6953 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6954 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6955 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6956 	tr->physical_address = gpa;
6957 	tr->valid = gpa != UNMAPPED_GVA;
6958 	tr->writeable = 1;
6959 	tr->usermode = 0;
6960 
6961 	return 0;
6962 }
6963 
6964 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6965 {
6966 	struct fxregs_state *fxsave =
6967 			&vcpu->arch.guest_fpu.state.fxsave;
6968 
6969 	memcpy(fpu->fpr, fxsave->st_space, 128);
6970 	fpu->fcw = fxsave->cwd;
6971 	fpu->fsw = fxsave->swd;
6972 	fpu->ftwx = fxsave->twd;
6973 	fpu->last_opcode = fxsave->fop;
6974 	fpu->last_ip = fxsave->rip;
6975 	fpu->last_dp = fxsave->rdp;
6976 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6977 
6978 	return 0;
6979 }
6980 
6981 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6982 {
6983 	struct fxregs_state *fxsave =
6984 			&vcpu->arch.guest_fpu.state.fxsave;
6985 
6986 	memcpy(fxsave->st_space, fpu->fpr, 128);
6987 	fxsave->cwd = fpu->fcw;
6988 	fxsave->swd = fpu->fsw;
6989 	fxsave->twd = fpu->ftwx;
6990 	fxsave->fop = fpu->last_opcode;
6991 	fxsave->rip = fpu->last_ip;
6992 	fxsave->rdp = fpu->last_dp;
6993 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6994 
6995 	return 0;
6996 }
6997 
6998 static void fx_init(struct kvm_vcpu *vcpu)
6999 {
7000 	fpstate_init(&vcpu->arch.guest_fpu.state);
7001 	if (cpu_has_xsaves)
7002 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7003 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7004 
7005 	/*
7006 	 * Ensure guest xcr0 is valid for loading
7007 	 */
7008 	vcpu->arch.xcr0 = XSTATE_FP;
7009 
7010 	vcpu->arch.cr0 |= X86_CR0_ET;
7011 }
7012 
7013 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7014 {
7015 	if (vcpu->guest_fpu_loaded)
7016 		return;
7017 
7018 	/*
7019 	 * Restore all possible states in the guest,
7020 	 * and assume host would use all available bits.
7021 	 * Guest xcr0 would be loaded later.
7022 	 */
7023 	kvm_put_guest_xcr0(vcpu);
7024 	vcpu->guest_fpu_loaded = 1;
7025 	__kernel_fpu_begin();
7026 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7027 	trace_kvm_fpu(1);
7028 }
7029 
7030 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7031 {
7032 	kvm_put_guest_xcr0(vcpu);
7033 
7034 	if (!vcpu->guest_fpu_loaded) {
7035 		vcpu->fpu_counter = 0;
7036 		return;
7037 	}
7038 
7039 	vcpu->guest_fpu_loaded = 0;
7040 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7041 	__kernel_fpu_end();
7042 	++vcpu->stat.fpu_reload;
7043 	/*
7044 	 * If using eager FPU mode, or if the guest is a frequent user
7045 	 * of the FPU, just leave the FPU active for next time.
7046 	 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7047 	 * the FPU in bursts will revert to loading it on demand.
7048 	 */
7049 	if (!vcpu->arch.eager_fpu) {
7050 		if (++vcpu->fpu_counter < 5)
7051 			kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7052 	}
7053 	trace_kvm_fpu(0);
7054 }
7055 
7056 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7057 {
7058 	kvmclock_reset(vcpu);
7059 
7060 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7061 	kvm_x86_ops->vcpu_free(vcpu);
7062 }
7063 
7064 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7065 						unsigned int id)
7066 {
7067 	struct kvm_vcpu *vcpu;
7068 
7069 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7070 		printk_once(KERN_WARNING
7071 		"kvm: SMP vm created on host with unstable TSC; "
7072 		"guest TSC will not be reliable\n");
7073 
7074 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7075 
7076 	return vcpu;
7077 }
7078 
7079 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7080 {
7081 	int r;
7082 
7083 	kvm_vcpu_mtrr_init(vcpu);
7084 	r = vcpu_load(vcpu);
7085 	if (r)
7086 		return r;
7087 	kvm_vcpu_reset(vcpu, false);
7088 	kvm_mmu_setup(vcpu);
7089 	vcpu_put(vcpu);
7090 	return r;
7091 }
7092 
7093 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7094 {
7095 	struct msr_data msr;
7096 	struct kvm *kvm = vcpu->kvm;
7097 
7098 	if (vcpu_load(vcpu))
7099 		return;
7100 	msr.data = 0x0;
7101 	msr.index = MSR_IA32_TSC;
7102 	msr.host_initiated = true;
7103 	kvm_write_tsc(vcpu, &msr);
7104 	vcpu_put(vcpu);
7105 
7106 	if (!kvmclock_periodic_sync)
7107 		return;
7108 
7109 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7110 					KVMCLOCK_SYNC_PERIOD);
7111 }
7112 
7113 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7114 {
7115 	int r;
7116 	vcpu->arch.apf.msr_val = 0;
7117 
7118 	r = vcpu_load(vcpu);
7119 	BUG_ON(r);
7120 	kvm_mmu_unload(vcpu);
7121 	vcpu_put(vcpu);
7122 
7123 	kvm_x86_ops->vcpu_free(vcpu);
7124 }
7125 
7126 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7127 {
7128 	vcpu->arch.hflags = 0;
7129 
7130 	atomic_set(&vcpu->arch.nmi_queued, 0);
7131 	vcpu->arch.nmi_pending = 0;
7132 	vcpu->arch.nmi_injected = false;
7133 	kvm_clear_interrupt_queue(vcpu);
7134 	kvm_clear_exception_queue(vcpu);
7135 
7136 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7137 	kvm_update_dr0123(vcpu);
7138 	vcpu->arch.dr6 = DR6_INIT;
7139 	kvm_update_dr6(vcpu);
7140 	vcpu->arch.dr7 = DR7_FIXED_1;
7141 	kvm_update_dr7(vcpu);
7142 
7143 	vcpu->arch.cr2 = 0;
7144 
7145 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7146 	vcpu->arch.apf.msr_val = 0;
7147 	vcpu->arch.st.msr_val = 0;
7148 
7149 	kvmclock_reset(vcpu);
7150 
7151 	kvm_clear_async_pf_completion_queue(vcpu);
7152 	kvm_async_pf_hash_reset(vcpu);
7153 	vcpu->arch.apf.halted = false;
7154 
7155 	if (!init_event) {
7156 		kvm_pmu_reset(vcpu);
7157 		vcpu->arch.smbase = 0x30000;
7158 	}
7159 
7160 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7161 	vcpu->arch.regs_avail = ~0;
7162 	vcpu->arch.regs_dirty = ~0;
7163 
7164 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7165 }
7166 
7167 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7168 {
7169 	struct kvm_segment cs;
7170 
7171 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7172 	cs.selector = vector << 8;
7173 	cs.base = vector << 12;
7174 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7175 	kvm_rip_write(vcpu, 0);
7176 }
7177 
7178 int kvm_arch_hardware_enable(void)
7179 {
7180 	struct kvm *kvm;
7181 	struct kvm_vcpu *vcpu;
7182 	int i;
7183 	int ret;
7184 	u64 local_tsc;
7185 	u64 max_tsc = 0;
7186 	bool stable, backwards_tsc = false;
7187 
7188 	kvm_shared_msr_cpu_online();
7189 	ret = kvm_x86_ops->hardware_enable();
7190 	if (ret != 0)
7191 		return ret;
7192 
7193 	local_tsc = rdtsc();
7194 	stable = !check_tsc_unstable();
7195 	list_for_each_entry(kvm, &vm_list, vm_list) {
7196 		kvm_for_each_vcpu(i, vcpu, kvm) {
7197 			if (!stable && vcpu->cpu == smp_processor_id())
7198 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7199 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7200 				backwards_tsc = true;
7201 				if (vcpu->arch.last_host_tsc > max_tsc)
7202 					max_tsc = vcpu->arch.last_host_tsc;
7203 			}
7204 		}
7205 	}
7206 
7207 	/*
7208 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7209 	 * platforms that reset TSC during suspend or hibernate actions, but
7210 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7211 	 * detect that condition here, which happens early in CPU bringup,
7212 	 * before any KVM threads can be running.  Unfortunately, we can't
7213 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7214 	 * enough into CPU bringup that we know how much real time has actually
7215 	 * elapsed; our helper function, get_kernel_ns() will be using boot
7216 	 * variables that haven't been updated yet.
7217 	 *
7218 	 * So we simply find the maximum observed TSC above, then record the
7219 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7220 	 * the adjustment will be applied.  Note that we accumulate
7221 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7222 	 * gets a chance to run again.  In the event that no KVM threads get a
7223 	 * chance to run, we will miss the entire elapsed period, as we'll have
7224 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7225 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7226 	 * uniform across all VCPUs (not to mention the scenario is extremely
7227 	 * unlikely). It is possible that a second hibernate recovery happens
7228 	 * much faster than a first, causing the observed TSC here to be
7229 	 * smaller; this would require additional padding adjustment, which is
7230 	 * why we set last_host_tsc to the local tsc observed here.
7231 	 *
7232 	 * N.B. - this code below runs only on platforms with reliable TSC,
7233 	 * as that is the only way backwards_tsc is set above.  Also note
7234 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7235 	 * have the same delta_cyc adjustment applied if backwards_tsc
7236 	 * is detected.  Note further, this adjustment is only done once,
7237 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7238 	 * called multiple times (one for each physical CPU bringup).
7239 	 *
7240 	 * Platforms with unreliable TSCs don't have to deal with this, they
7241 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7242 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7243 	 * guarantee that they stay in perfect synchronization.
7244 	 */
7245 	if (backwards_tsc) {
7246 		u64 delta_cyc = max_tsc - local_tsc;
7247 		backwards_tsc_observed = true;
7248 		list_for_each_entry(kvm, &vm_list, vm_list) {
7249 			kvm_for_each_vcpu(i, vcpu, kvm) {
7250 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7251 				vcpu->arch.last_host_tsc = local_tsc;
7252 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7253 			}
7254 
7255 			/*
7256 			 * We have to disable TSC offset matching.. if you were
7257 			 * booting a VM while issuing an S4 host suspend....
7258 			 * you may have some problem.  Solving this issue is
7259 			 * left as an exercise to the reader.
7260 			 */
7261 			kvm->arch.last_tsc_nsec = 0;
7262 			kvm->arch.last_tsc_write = 0;
7263 		}
7264 
7265 	}
7266 	return 0;
7267 }
7268 
7269 void kvm_arch_hardware_disable(void)
7270 {
7271 	kvm_x86_ops->hardware_disable();
7272 	drop_user_return_notifiers();
7273 }
7274 
7275 int kvm_arch_hardware_setup(void)
7276 {
7277 	int r;
7278 
7279 	r = kvm_x86_ops->hardware_setup();
7280 	if (r != 0)
7281 		return r;
7282 
7283 	kvm_init_msr_list();
7284 	return 0;
7285 }
7286 
7287 void kvm_arch_hardware_unsetup(void)
7288 {
7289 	kvm_x86_ops->hardware_unsetup();
7290 }
7291 
7292 void kvm_arch_check_processor_compat(void *rtn)
7293 {
7294 	kvm_x86_ops->check_processor_compatibility(rtn);
7295 }
7296 
7297 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7298 {
7299 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7300 }
7301 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7302 
7303 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7304 {
7305 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7306 }
7307 
7308 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7309 {
7310 	return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7311 }
7312 
7313 struct static_key kvm_no_apic_vcpu __read_mostly;
7314 
7315 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7316 {
7317 	struct page *page;
7318 	struct kvm *kvm;
7319 	int r;
7320 
7321 	BUG_ON(vcpu->kvm == NULL);
7322 	kvm = vcpu->kvm;
7323 
7324 	vcpu->arch.pv.pv_unhalted = false;
7325 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7326 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7327 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7328 	else
7329 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7330 
7331 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7332 	if (!page) {
7333 		r = -ENOMEM;
7334 		goto fail;
7335 	}
7336 	vcpu->arch.pio_data = page_address(page);
7337 
7338 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7339 
7340 	r = kvm_mmu_create(vcpu);
7341 	if (r < 0)
7342 		goto fail_free_pio_data;
7343 
7344 	if (irqchip_in_kernel(kvm)) {
7345 		r = kvm_create_lapic(vcpu);
7346 		if (r < 0)
7347 			goto fail_mmu_destroy;
7348 	} else
7349 		static_key_slow_inc(&kvm_no_apic_vcpu);
7350 
7351 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7352 				       GFP_KERNEL);
7353 	if (!vcpu->arch.mce_banks) {
7354 		r = -ENOMEM;
7355 		goto fail_free_lapic;
7356 	}
7357 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7358 
7359 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7360 		r = -ENOMEM;
7361 		goto fail_free_mce_banks;
7362 	}
7363 
7364 	fx_init(vcpu);
7365 
7366 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7367 	vcpu->arch.pv_time_enabled = false;
7368 
7369 	vcpu->arch.guest_supported_xcr0 = 0;
7370 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7371 
7372 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7373 
7374 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7375 
7376 	kvm_async_pf_hash_reset(vcpu);
7377 	kvm_pmu_init(vcpu);
7378 
7379 	return 0;
7380 
7381 fail_free_mce_banks:
7382 	kfree(vcpu->arch.mce_banks);
7383 fail_free_lapic:
7384 	kvm_free_lapic(vcpu);
7385 fail_mmu_destroy:
7386 	kvm_mmu_destroy(vcpu);
7387 fail_free_pio_data:
7388 	free_page((unsigned long)vcpu->arch.pio_data);
7389 fail:
7390 	return r;
7391 }
7392 
7393 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7394 {
7395 	int idx;
7396 
7397 	kvm_pmu_destroy(vcpu);
7398 	kfree(vcpu->arch.mce_banks);
7399 	kvm_free_lapic(vcpu);
7400 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7401 	kvm_mmu_destroy(vcpu);
7402 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7403 	free_page((unsigned long)vcpu->arch.pio_data);
7404 	if (!irqchip_in_kernel(vcpu->kvm))
7405 		static_key_slow_dec(&kvm_no_apic_vcpu);
7406 }
7407 
7408 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7409 {
7410 	kvm_x86_ops->sched_in(vcpu, cpu);
7411 }
7412 
7413 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7414 {
7415 	if (type)
7416 		return -EINVAL;
7417 
7418 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7419 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7420 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7421 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7422 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7423 
7424 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7425 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7426 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7427 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7428 		&kvm->arch.irq_sources_bitmap);
7429 
7430 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7431 	mutex_init(&kvm->arch.apic_map_lock);
7432 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7433 
7434 	pvclock_update_vm_gtod_copy(kvm);
7435 
7436 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7437 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7438 
7439 	return 0;
7440 }
7441 
7442 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7443 {
7444 	int r;
7445 	r = vcpu_load(vcpu);
7446 	BUG_ON(r);
7447 	kvm_mmu_unload(vcpu);
7448 	vcpu_put(vcpu);
7449 }
7450 
7451 static void kvm_free_vcpus(struct kvm *kvm)
7452 {
7453 	unsigned int i;
7454 	struct kvm_vcpu *vcpu;
7455 
7456 	/*
7457 	 * Unpin any mmu pages first.
7458 	 */
7459 	kvm_for_each_vcpu(i, vcpu, kvm) {
7460 		kvm_clear_async_pf_completion_queue(vcpu);
7461 		kvm_unload_vcpu_mmu(vcpu);
7462 	}
7463 	kvm_for_each_vcpu(i, vcpu, kvm)
7464 		kvm_arch_vcpu_free(vcpu);
7465 
7466 	mutex_lock(&kvm->lock);
7467 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7468 		kvm->vcpus[i] = NULL;
7469 
7470 	atomic_set(&kvm->online_vcpus, 0);
7471 	mutex_unlock(&kvm->lock);
7472 }
7473 
7474 void kvm_arch_sync_events(struct kvm *kvm)
7475 {
7476 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7477 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7478 	kvm_free_all_assigned_devices(kvm);
7479 	kvm_free_pit(kvm);
7480 }
7481 
7482 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7483 {
7484 	int i, r;
7485 	unsigned long hva;
7486 	struct kvm_memslots *slots = kvm_memslots(kvm);
7487 	struct kvm_memory_slot *slot, old;
7488 
7489 	/* Called with kvm->slots_lock held.  */
7490 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7491 		return -EINVAL;
7492 
7493 	slot = id_to_memslot(slots, id);
7494 	if (size) {
7495 		if (WARN_ON(slot->npages))
7496 			return -EEXIST;
7497 
7498 		/*
7499 		 * MAP_SHARED to prevent internal slot pages from being moved
7500 		 * by fork()/COW.
7501 		 */
7502 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7503 			      MAP_SHARED | MAP_ANONYMOUS, 0);
7504 		if (IS_ERR((void *)hva))
7505 			return PTR_ERR((void *)hva);
7506 	} else {
7507 		if (!slot->npages)
7508 			return 0;
7509 
7510 		hva = 0;
7511 	}
7512 
7513 	old = *slot;
7514 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7515 		struct kvm_userspace_memory_region m;
7516 
7517 		m.slot = id | (i << 16);
7518 		m.flags = 0;
7519 		m.guest_phys_addr = gpa;
7520 		m.userspace_addr = hva;
7521 		m.memory_size = size;
7522 		r = __kvm_set_memory_region(kvm, &m);
7523 		if (r < 0)
7524 			return r;
7525 	}
7526 
7527 	if (!size) {
7528 		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7529 		WARN_ON(r < 0);
7530 	}
7531 
7532 	return 0;
7533 }
7534 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7535 
7536 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7537 {
7538 	int r;
7539 
7540 	mutex_lock(&kvm->slots_lock);
7541 	r = __x86_set_memory_region(kvm, id, gpa, size);
7542 	mutex_unlock(&kvm->slots_lock);
7543 
7544 	return r;
7545 }
7546 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7547 
7548 void kvm_arch_destroy_vm(struct kvm *kvm)
7549 {
7550 	if (current->mm == kvm->mm) {
7551 		/*
7552 		 * Free memory regions allocated on behalf of userspace,
7553 		 * unless the the memory map has changed due to process exit
7554 		 * or fd copying.
7555 		 */
7556 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7557 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7558 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7559 	}
7560 	kvm_iommu_unmap_guest(kvm);
7561 	kfree(kvm->arch.vpic);
7562 	kfree(kvm->arch.vioapic);
7563 	kvm_free_vcpus(kvm);
7564 	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7565 }
7566 
7567 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7568 			   struct kvm_memory_slot *dont)
7569 {
7570 	int i;
7571 
7572 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7573 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7574 			kvfree(free->arch.rmap[i]);
7575 			free->arch.rmap[i] = NULL;
7576 		}
7577 		if (i == 0)
7578 			continue;
7579 
7580 		if (!dont || free->arch.lpage_info[i - 1] !=
7581 			     dont->arch.lpage_info[i - 1]) {
7582 			kvfree(free->arch.lpage_info[i - 1]);
7583 			free->arch.lpage_info[i - 1] = NULL;
7584 		}
7585 	}
7586 }
7587 
7588 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7589 			    unsigned long npages)
7590 {
7591 	int i;
7592 
7593 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7594 		unsigned long ugfn;
7595 		int lpages;
7596 		int level = i + 1;
7597 
7598 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7599 				      slot->base_gfn, level) + 1;
7600 
7601 		slot->arch.rmap[i] =
7602 			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7603 		if (!slot->arch.rmap[i])
7604 			goto out_free;
7605 		if (i == 0)
7606 			continue;
7607 
7608 		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7609 					sizeof(*slot->arch.lpage_info[i - 1]));
7610 		if (!slot->arch.lpage_info[i - 1])
7611 			goto out_free;
7612 
7613 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7614 			slot->arch.lpage_info[i - 1][0].write_count = 1;
7615 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7616 			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7617 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7618 		/*
7619 		 * If the gfn and userspace address are not aligned wrt each
7620 		 * other, or if explicitly asked to, disable large page
7621 		 * support for this slot
7622 		 */
7623 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7624 		    !kvm_largepages_enabled()) {
7625 			unsigned long j;
7626 
7627 			for (j = 0; j < lpages; ++j)
7628 				slot->arch.lpage_info[i - 1][j].write_count = 1;
7629 		}
7630 	}
7631 
7632 	return 0;
7633 
7634 out_free:
7635 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7636 		kvfree(slot->arch.rmap[i]);
7637 		slot->arch.rmap[i] = NULL;
7638 		if (i == 0)
7639 			continue;
7640 
7641 		kvfree(slot->arch.lpage_info[i - 1]);
7642 		slot->arch.lpage_info[i - 1] = NULL;
7643 	}
7644 	return -ENOMEM;
7645 }
7646 
7647 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7648 {
7649 	/*
7650 	 * memslots->generation has been incremented.
7651 	 * mmio generation may have reached its maximum value.
7652 	 */
7653 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7654 }
7655 
7656 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7657 				struct kvm_memory_slot *memslot,
7658 				const struct kvm_userspace_memory_region *mem,
7659 				enum kvm_mr_change change)
7660 {
7661 	return 0;
7662 }
7663 
7664 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7665 				     struct kvm_memory_slot *new)
7666 {
7667 	/* Still write protect RO slot */
7668 	if (new->flags & KVM_MEM_READONLY) {
7669 		kvm_mmu_slot_remove_write_access(kvm, new);
7670 		return;
7671 	}
7672 
7673 	/*
7674 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
7675 	 *
7676 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
7677 	 *
7678 	 *  - KVM_MR_CREATE with dirty logging is disabled
7679 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7680 	 *
7681 	 * The reason is, in case of PML, we need to set D-bit for any slots
7682 	 * with dirty logging disabled in order to eliminate unnecessary GPA
7683 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
7684 	 * guarantees leaving PML enabled during guest's lifetime won't have
7685 	 * any additonal overhead from PML when guest is running with dirty
7686 	 * logging disabled for memory slots.
7687 	 *
7688 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7689 	 * to dirty logging mode.
7690 	 *
7691 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7692 	 *
7693 	 * In case of write protect:
7694 	 *
7695 	 * Write protect all pages for dirty logging.
7696 	 *
7697 	 * All the sptes including the large sptes which point to this
7698 	 * slot are set to readonly. We can not create any new large
7699 	 * spte on this slot until the end of the logging.
7700 	 *
7701 	 * See the comments in fast_page_fault().
7702 	 */
7703 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7704 		if (kvm_x86_ops->slot_enable_log_dirty)
7705 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7706 		else
7707 			kvm_mmu_slot_remove_write_access(kvm, new);
7708 	} else {
7709 		if (kvm_x86_ops->slot_disable_log_dirty)
7710 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7711 	}
7712 }
7713 
7714 void kvm_arch_commit_memory_region(struct kvm *kvm,
7715 				const struct kvm_userspace_memory_region *mem,
7716 				const struct kvm_memory_slot *old,
7717 				const struct kvm_memory_slot *new,
7718 				enum kvm_mr_change change)
7719 {
7720 	int nr_mmu_pages = 0;
7721 
7722 	if (!kvm->arch.n_requested_mmu_pages)
7723 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7724 
7725 	if (nr_mmu_pages)
7726 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7727 
7728 	/*
7729 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
7730 	 * sptes have to be split.  If live migration is successful, the guest
7731 	 * in the source machine will be destroyed and large sptes will be
7732 	 * created in the destination. However, if the guest continues to run
7733 	 * in the source machine (for example if live migration fails), small
7734 	 * sptes will remain around and cause bad performance.
7735 	 *
7736 	 * Scan sptes if dirty logging has been stopped, dropping those
7737 	 * which can be collapsed into a single large-page spte.  Later
7738 	 * page faults will create the large-page sptes.
7739 	 */
7740 	if ((change != KVM_MR_DELETE) &&
7741 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7742 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7743 		kvm_mmu_zap_collapsible_sptes(kvm, new);
7744 
7745 	/*
7746 	 * Set up write protection and/or dirty logging for the new slot.
7747 	 *
7748 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7749 	 * been zapped so no dirty logging staff is needed for old slot. For
7750 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7751 	 * new and it's also covered when dealing with the new slot.
7752 	 *
7753 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
7754 	 */
7755 	if (change != KVM_MR_DELETE)
7756 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7757 }
7758 
7759 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7760 {
7761 	kvm_mmu_invalidate_zap_all_pages(kvm);
7762 }
7763 
7764 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7765 				   struct kvm_memory_slot *slot)
7766 {
7767 	kvm_mmu_invalidate_zap_all_pages(kvm);
7768 }
7769 
7770 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
7771 {
7772 	if (!list_empty_careful(&vcpu->async_pf.done))
7773 		return true;
7774 
7775 	if (kvm_apic_has_events(vcpu))
7776 		return true;
7777 
7778 	if (vcpu->arch.pv.pv_unhalted)
7779 		return true;
7780 
7781 	if (atomic_read(&vcpu->arch.nmi_queued))
7782 		return true;
7783 
7784 	if (test_bit(KVM_REQ_SMI, &vcpu->requests))
7785 		return true;
7786 
7787 	if (kvm_arch_interrupt_allowed(vcpu) &&
7788 	    kvm_cpu_has_interrupt(vcpu))
7789 		return true;
7790 
7791 	return false;
7792 }
7793 
7794 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7795 {
7796 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7797 		kvm_x86_ops->check_nested_events(vcpu, false);
7798 
7799 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
7800 }
7801 
7802 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7803 {
7804 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7805 }
7806 
7807 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7808 {
7809 	return kvm_x86_ops->interrupt_allowed(vcpu);
7810 }
7811 
7812 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7813 {
7814 	if (is_64_bit_mode(vcpu))
7815 		return kvm_rip_read(vcpu);
7816 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7817 		     kvm_rip_read(vcpu));
7818 }
7819 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7820 
7821 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7822 {
7823 	return kvm_get_linear_rip(vcpu) == linear_rip;
7824 }
7825 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7826 
7827 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7828 {
7829 	unsigned long rflags;
7830 
7831 	rflags = kvm_x86_ops->get_rflags(vcpu);
7832 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7833 		rflags &= ~X86_EFLAGS_TF;
7834 	return rflags;
7835 }
7836 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7837 
7838 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7839 {
7840 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7841 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7842 		rflags |= X86_EFLAGS_TF;
7843 	kvm_x86_ops->set_rflags(vcpu, rflags);
7844 }
7845 
7846 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7847 {
7848 	__kvm_set_rflags(vcpu, rflags);
7849 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7850 }
7851 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7852 
7853 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7854 {
7855 	int r;
7856 
7857 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7858 	      work->wakeup_all)
7859 		return;
7860 
7861 	r = kvm_mmu_reload(vcpu);
7862 	if (unlikely(r))
7863 		return;
7864 
7865 	if (!vcpu->arch.mmu.direct_map &&
7866 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7867 		return;
7868 
7869 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7870 }
7871 
7872 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7873 {
7874 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7875 }
7876 
7877 static inline u32 kvm_async_pf_next_probe(u32 key)
7878 {
7879 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7880 }
7881 
7882 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7883 {
7884 	u32 key = kvm_async_pf_hash_fn(gfn);
7885 
7886 	while (vcpu->arch.apf.gfns[key] != ~0)
7887 		key = kvm_async_pf_next_probe(key);
7888 
7889 	vcpu->arch.apf.gfns[key] = gfn;
7890 }
7891 
7892 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7893 {
7894 	int i;
7895 	u32 key = kvm_async_pf_hash_fn(gfn);
7896 
7897 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7898 		     (vcpu->arch.apf.gfns[key] != gfn &&
7899 		      vcpu->arch.apf.gfns[key] != ~0); i++)
7900 		key = kvm_async_pf_next_probe(key);
7901 
7902 	return key;
7903 }
7904 
7905 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7906 {
7907 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7908 }
7909 
7910 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7911 {
7912 	u32 i, j, k;
7913 
7914 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7915 	while (true) {
7916 		vcpu->arch.apf.gfns[i] = ~0;
7917 		do {
7918 			j = kvm_async_pf_next_probe(j);
7919 			if (vcpu->arch.apf.gfns[j] == ~0)
7920 				return;
7921 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7922 			/*
7923 			 * k lies cyclically in ]i,j]
7924 			 * |    i.k.j |
7925 			 * |....j i.k.| or  |.k..j i...|
7926 			 */
7927 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7928 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7929 		i = j;
7930 	}
7931 }
7932 
7933 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7934 {
7935 
7936 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7937 				      sizeof(val));
7938 }
7939 
7940 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7941 				     struct kvm_async_pf *work)
7942 {
7943 	struct x86_exception fault;
7944 
7945 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7946 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7947 
7948 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7949 	    (vcpu->arch.apf.send_user_only &&
7950 	     kvm_x86_ops->get_cpl(vcpu) == 0))
7951 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7952 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7953 		fault.vector = PF_VECTOR;
7954 		fault.error_code_valid = true;
7955 		fault.error_code = 0;
7956 		fault.nested_page_fault = false;
7957 		fault.address = work->arch.token;
7958 		kvm_inject_page_fault(vcpu, &fault);
7959 	}
7960 }
7961 
7962 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7963 				 struct kvm_async_pf *work)
7964 {
7965 	struct x86_exception fault;
7966 
7967 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
7968 	if (work->wakeup_all)
7969 		work->arch.token = ~0; /* broadcast wakeup */
7970 	else
7971 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7972 
7973 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7974 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7975 		fault.vector = PF_VECTOR;
7976 		fault.error_code_valid = true;
7977 		fault.error_code = 0;
7978 		fault.nested_page_fault = false;
7979 		fault.address = work->arch.token;
7980 		kvm_inject_page_fault(vcpu, &fault);
7981 	}
7982 	vcpu->arch.apf.halted = false;
7983 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7984 }
7985 
7986 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7987 {
7988 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7989 		return true;
7990 	else
7991 		return !kvm_event_needs_reinjection(vcpu) &&
7992 			kvm_x86_ops->interrupt_allowed(vcpu);
7993 }
7994 
7995 void kvm_arch_start_assignment(struct kvm *kvm)
7996 {
7997 	atomic_inc(&kvm->arch.assigned_device_count);
7998 }
7999 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8000 
8001 void kvm_arch_end_assignment(struct kvm *kvm)
8002 {
8003 	atomic_dec(&kvm->arch.assigned_device_count);
8004 }
8005 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8006 
8007 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8008 {
8009 	return atomic_read(&kvm->arch.assigned_device_count);
8010 }
8011 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8012 
8013 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8014 {
8015 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8016 }
8017 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8018 
8019 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8020 {
8021 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8022 }
8023 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8024 
8025 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8026 {
8027 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8028 }
8029 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8030 
8031 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8032 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8033 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8034 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8035 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8036 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8037 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8038 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8039 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8040 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8041 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8042 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8043 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8044 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8045 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8046