xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 2596e07a)
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 <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <trace/events/kvm.h>
57 
58 #define CREATE_TRACE_POINTS
59 #include "trace.h"
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 
71 #define MAX_IO_MSRS 256
72 #define KVM_MAX_MCE_BANKS 32
73 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
74 
75 #define emul_to_vcpu(ctxt) \
76 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
77 
78 /* EFER defaults:
79  * - enable syscall per default because its emulated by KVM
80  * - enable LME and LMA per default on 64 bit KVM
81  */
82 #ifdef CONFIG_X86_64
83 static
84 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
85 #else
86 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
87 #endif
88 
89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
91 
92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93 static void process_nmi(struct kvm_vcpu *vcpu);
94 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
95 
96 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
98 
99 static bool __read_mostly ignore_msrs = 0;
100 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
101 
102 unsigned int min_timer_period_us = 500;
103 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
104 
105 static bool __read_mostly kvmclock_periodic_sync = true;
106 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
107 
108 bool __read_mostly kvm_has_tsc_control;
109 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
110 u32  __read_mostly kvm_max_guest_tsc_khz;
111 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
112 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
113 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
114 u64  __read_mostly kvm_max_tsc_scaling_ratio;
115 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
116 static u64 __read_mostly kvm_default_tsc_scaling_ratio;
117 
118 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
119 static u32 __read_mostly tsc_tolerance_ppm = 250;
120 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
121 
122 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
123 unsigned int __read_mostly lapic_timer_advance_ns = 0;
124 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
125 
126 static bool __read_mostly backwards_tsc_observed = false;
127 
128 #define KVM_NR_SHARED_MSRS 16
129 
130 struct kvm_shared_msrs_global {
131 	int nr;
132 	u32 msrs[KVM_NR_SHARED_MSRS];
133 };
134 
135 struct kvm_shared_msrs {
136 	struct user_return_notifier urn;
137 	bool registered;
138 	struct kvm_shared_msr_values {
139 		u64 host;
140 		u64 curr;
141 	} values[KVM_NR_SHARED_MSRS];
142 };
143 
144 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
145 static struct kvm_shared_msrs __percpu *shared_msrs;
146 
147 struct kvm_stats_debugfs_item debugfs_entries[] = {
148 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
149 	{ "pf_guest", VCPU_STAT(pf_guest) },
150 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
151 	{ "invlpg", VCPU_STAT(invlpg) },
152 	{ "exits", VCPU_STAT(exits) },
153 	{ "io_exits", VCPU_STAT(io_exits) },
154 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
155 	{ "signal_exits", VCPU_STAT(signal_exits) },
156 	{ "irq_window", VCPU_STAT(irq_window_exits) },
157 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
158 	{ "halt_exits", VCPU_STAT(halt_exits) },
159 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
160 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
161 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
162 	{ "hypercalls", VCPU_STAT(hypercalls) },
163 	{ "request_irq", VCPU_STAT(request_irq_exits) },
164 	{ "irq_exits", VCPU_STAT(irq_exits) },
165 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
166 	{ "efer_reload", VCPU_STAT(efer_reload) },
167 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
168 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
169 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
170 	{ "irq_injections", VCPU_STAT(irq_injections) },
171 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
172 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
173 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
174 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
175 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
176 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
177 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
178 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
179 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
180 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
181 	{ "largepages", VM_STAT(lpages) },
182 	{ NULL }
183 };
184 
185 u64 __read_mostly host_xcr0;
186 
187 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
188 
189 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
190 {
191 	int i;
192 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
193 		vcpu->arch.apf.gfns[i] = ~0;
194 }
195 
196 static void kvm_on_user_return(struct user_return_notifier *urn)
197 {
198 	unsigned slot;
199 	struct kvm_shared_msrs *locals
200 		= container_of(urn, struct kvm_shared_msrs, urn);
201 	struct kvm_shared_msr_values *values;
202 
203 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
204 		values = &locals->values[slot];
205 		if (values->host != values->curr) {
206 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
207 			values->curr = values->host;
208 		}
209 	}
210 	locals->registered = false;
211 	user_return_notifier_unregister(urn);
212 }
213 
214 static void shared_msr_update(unsigned slot, u32 msr)
215 {
216 	u64 value;
217 	unsigned int cpu = smp_processor_id();
218 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
219 
220 	/* only read, and nobody should modify it at this time,
221 	 * so don't need lock */
222 	if (slot >= shared_msrs_global.nr) {
223 		printk(KERN_ERR "kvm: invalid MSR slot!");
224 		return;
225 	}
226 	rdmsrl_safe(msr, &value);
227 	smsr->values[slot].host = value;
228 	smsr->values[slot].curr = value;
229 }
230 
231 void kvm_define_shared_msr(unsigned slot, u32 msr)
232 {
233 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
234 	shared_msrs_global.msrs[slot] = msr;
235 	if (slot >= shared_msrs_global.nr)
236 		shared_msrs_global.nr = slot + 1;
237 }
238 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
239 
240 static void kvm_shared_msr_cpu_online(void)
241 {
242 	unsigned i;
243 
244 	for (i = 0; i < shared_msrs_global.nr; ++i)
245 		shared_msr_update(i, shared_msrs_global.msrs[i]);
246 }
247 
248 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
249 {
250 	unsigned int cpu = smp_processor_id();
251 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
252 	int err;
253 
254 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
255 		return 0;
256 	smsr->values[slot].curr = value;
257 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
258 	if (err)
259 		return 1;
260 
261 	if (!smsr->registered) {
262 		smsr->urn.on_user_return = kvm_on_user_return;
263 		user_return_notifier_register(&smsr->urn);
264 		smsr->registered = true;
265 	}
266 	return 0;
267 }
268 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
269 
270 static void drop_user_return_notifiers(void)
271 {
272 	unsigned int cpu = smp_processor_id();
273 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
274 
275 	if (smsr->registered)
276 		kvm_on_user_return(&smsr->urn);
277 }
278 
279 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
280 {
281 	return vcpu->arch.apic_base;
282 }
283 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
284 
285 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
286 {
287 	u64 old_state = vcpu->arch.apic_base &
288 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
289 	u64 new_state = msr_info->data &
290 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
291 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
292 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
293 
294 	if (!msr_info->host_initiated &&
295 	    ((msr_info->data & reserved_bits) != 0 ||
296 	     new_state == X2APIC_ENABLE ||
297 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
298 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
299 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
300 	      old_state == 0)))
301 		return 1;
302 
303 	kvm_lapic_set_base(vcpu, msr_info->data);
304 	return 0;
305 }
306 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
307 
308 asmlinkage __visible void kvm_spurious_fault(void)
309 {
310 	/* Fault while not rebooting.  We want the trace. */
311 	BUG();
312 }
313 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
314 
315 #define EXCPT_BENIGN		0
316 #define EXCPT_CONTRIBUTORY	1
317 #define EXCPT_PF		2
318 
319 static int exception_class(int vector)
320 {
321 	switch (vector) {
322 	case PF_VECTOR:
323 		return EXCPT_PF;
324 	case DE_VECTOR:
325 	case TS_VECTOR:
326 	case NP_VECTOR:
327 	case SS_VECTOR:
328 	case GP_VECTOR:
329 		return EXCPT_CONTRIBUTORY;
330 	default:
331 		break;
332 	}
333 	return EXCPT_BENIGN;
334 }
335 
336 #define EXCPT_FAULT		0
337 #define EXCPT_TRAP		1
338 #define EXCPT_ABORT		2
339 #define EXCPT_INTERRUPT		3
340 
341 static int exception_type(int vector)
342 {
343 	unsigned int mask;
344 
345 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
346 		return EXCPT_INTERRUPT;
347 
348 	mask = 1 << vector;
349 
350 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
351 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
352 		return EXCPT_TRAP;
353 
354 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
355 		return EXCPT_ABORT;
356 
357 	/* Reserved exceptions will result in fault */
358 	return EXCPT_FAULT;
359 }
360 
361 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
362 		unsigned nr, bool has_error, u32 error_code,
363 		bool reinject)
364 {
365 	u32 prev_nr;
366 	int class1, class2;
367 
368 	kvm_make_request(KVM_REQ_EVENT, vcpu);
369 
370 	if (!vcpu->arch.exception.pending) {
371 	queue:
372 		if (has_error && !is_protmode(vcpu))
373 			has_error = false;
374 		vcpu->arch.exception.pending = true;
375 		vcpu->arch.exception.has_error_code = has_error;
376 		vcpu->arch.exception.nr = nr;
377 		vcpu->arch.exception.error_code = error_code;
378 		vcpu->arch.exception.reinject = reinject;
379 		return;
380 	}
381 
382 	/* to check exception */
383 	prev_nr = vcpu->arch.exception.nr;
384 	if (prev_nr == DF_VECTOR) {
385 		/* triple fault -> shutdown */
386 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
387 		return;
388 	}
389 	class1 = exception_class(prev_nr);
390 	class2 = exception_class(nr);
391 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
392 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
393 		/* generate double fault per SDM Table 5-5 */
394 		vcpu->arch.exception.pending = true;
395 		vcpu->arch.exception.has_error_code = true;
396 		vcpu->arch.exception.nr = DF_VECTOR;
397 		vcpu->arch.exception.error_code = 0;
398 	} else
399 		/* replace previous exception with a new one in a hope
400 		   that instruction re-execution will regenerate lost
401 		   exception */
402 		goto queue;
403 }
404 
405 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
406 {
407 	kvm_multiple_exception(vcpu, nr, false, 0, false);
408 }
409 EXPORT_SYMBOL_GPL(kvm_queue_exception);
410 
411 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
412 {
413 	kvm_multiple_exception(vcpu, nr, false, 0, true);
414 }
415 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
416 
417 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
418 {
419 	if (err)
420 		kvm_inject_gp(vcpu, 0);
421 	else
422 		kvm_x86_ops->skip_emulated_instruction(vcpu);
423 }
424 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
425 
426 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
427 {
428 	++vcpu->stat.pf_guest;
429 	vcpu->arch.cr2 = fault->address;
430 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
431 }
432 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
433 
434 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
435 {
436 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
437 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
438 	else
439 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
440 
441 	return fault->nested_page_fault;
442 }
443 
444 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
445 {
446 	atomic_inc(&vcpu->arch.nmi_queued);
447 	kvm_make_request(KVM_REQ_NMI, vcpu);
448 }
449 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
450 
451 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
452 {
453 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
454 }
455 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
456 
457 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
458 {
459 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
460 }
461 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
462 
463 /*
464  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
465  * a #GP and return false.
466  */
467 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
468 {
469 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
470 		return true;
471 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
472 	return false;
473 }
474 EXPORT_SYMBOL_GPL(kvm_require_cpl);
475 
476 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
477 {
478 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
479 		return true;
480 
481 	kvm_queue_exception(vcpu, UD_VECTOR);
482 	return false;
483 }
484 EXPORT_SYMBOL_GPL(kvm_require_dr);
485 
486 /*
487  * This function will be used to read from the physical memory of the currently
488  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
489  * can read from guest physical or from the guest's guest physical memory.
490  */
491 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
492 			    gfn_t ngfn, void *data, int offset, int len,
493 			    u32 access)
494 {
495 	struct x86_exception exception;
496 	gfn_t real_gfn;
497 	gpa_t ngpa;
498 
499 	ngpa     = gfn_to_gpa(ngfn);
500 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
501 	if (real_gfn == UNMAPPED_GVA)
502 		return -EFAULT;
503 
504 	real_gfn = gpa_to_gfn(real_gfn);
505 
506 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
507 }
508 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
509 
510 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
511 			       void *data, int offset, int len, u32 access)
512 {
513 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
514 				       data, offset, len, access);
515 }
516 
517 /*
518  * Load the pae pdptrs.  Return true is they are all valid.
519  */
520 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
521 {
522 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
523 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
524 	int i;
525 	int ret;
526 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
527 
528 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
529 				      offset * sizeof(u64), sizeof(pdpte),
530 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
531 	if (ret < 0) {
532 		ret = 0;
533 		goto out;
534 	}
535 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
536 		if (is_present_gpte(pdpte[i]) &&
537 		    (pdpte[i] &
538 		     vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
539 			ret = 0;
540 			goto out;
541 		}
542 	}
543 	ret = 1;
544 
545 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
546 	__set_bit(VCPU_EXREG_PDPTR,
547 		  (unsigned long *)&vcpu->arch.regs_avail);
548 	__set_bit(VCPU_EXREG_PDPTR,
549 		  (unsigned long *)&vcpu->arch.regs_dirty);
550 out:
551 
552 	return ret;
553 }
554 EXPORT_SYMBOL_GPL(load_pdptrs);
555 
556 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
557 {
558 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
559 	bool changed = true;
560 	int offset;
561 	gfn_t gfn;
562 	int r;
563 
564 	if (is_long_mode(vcpu) || !is_pae(vcpu))
565 		return false;
566 
567 	if (!test_bit(VCPU_EXREG_PDPTR,
568 		      (unsigned long *)&vcpu->arch.regs_avail))
569 		return true;
570 
571 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
572 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
573 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
574 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
575 	if (r < 0)
576 		goto out;
577 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
578 out:
579 
580 	return changed;
581 }
582 
583 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
584 {
585 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
586 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
587 
588 	cr0 |= X86_CR0_ET;
589 
590 #ifdef CONFIG_X86_64
591 	if (cr0 & 0xffffffff00000000UL)
592 		return 1;
593 #endif
594 
595 	cr0 &= ~CR0_RESERVED_BITS;
596 
597 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
598 		return 1;
599 
600 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
601 		return 1;
602 
603 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
604 #ifdef CONFIG_X86_64
605 		if ((vcpu->arch.efer & EFER_LME)) {
606 			int cs_db, cs_l;
607 
608 			if (!is_pae(vcpu))
609 				return 1;
610 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
611 			if (cs_l)
612 				return 1;
613 		} else
614 #endif
615 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
616 						 kvm_read_cr3(vcpu)))
617 			return 1;
618 	}
619 
620 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
621 		return 1;
622 
623 	kvm_x86_ops->set_cr0(vcpu, cr0);
624 
625 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
626 		kvm_clear_async_pf_completion_queue(vcpu);
627 		kvm_async_pf_hash_reset(vcpu);
628 	}
629 
630 	if ((cr0 ^ old_cr0) & update_bits)
631 		kvm_mmu_reset_context(vcpu);
632 
633 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
634 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
635 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
636 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
637 
638 	return 0;
639 }
640 EXPORT_SYMBOL_GPL(kvm_set_cr0);
641 
642 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
643 {
644 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
645 }
646 EXPORT_SYMBOL_GPL(kvm_lmsw);
647 
648 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
649 {
650 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
651 			!vcpu->guest_xcr0_loaded) {
652 		/* kvm_set_xcr() also depends on this */
653 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
654 		vcpu->guest_xcr0_loaded = 1;
655 	}
656 }
657 
658 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
659 {
660 	if (vcpu->guest_xcr0_loaded) {
661 		if (vcpu->arch.xcr0 != host_xcr0)
662 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
663 		vcpu->guest_xcr0_loaded = 0;
664 	}
665 }
666 
667 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
668 {
669 	u64 xcr0 = xcr;
670 	u64 old_xcr0 = vcpu->arch.xcr0;
671 	u64 valid_bits;
672 
673 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
674 	if (index != XCR_XFEATURE_ENABLED_MASK)
675 		return 1;
676 	if (!(xcr0 & XFEATURE_MASK_FP))
677 		return 1;
678 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
679 		return 1;
680 
681 	/*
682 	 * Do not allow the guest to set bits that we do not support
683 	 * saving.  However, xcr0 bit 0 is always set, even if the
684 	 * emulated CPU does not support XSAVE (see fx_init).
685 	 */
686 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
687 	if (xcr0 & ~valid_bits)
688 		return 1;
689 
690 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
691 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
692 		return 1;
693 
694 	if (xcr0 & XFEATURE_MASK_AVX512) {
695 		if (!(xcr0 & XFEATURE_MASK_YMM))
696 			return 1;
697 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
698 			return 1;
699 	}
700 	kvm_put_guest_xcr0(vcpu);
701 	vcpu->arch.xcr0 = xcr0;
702 
703 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
704 		kvm_update_cpuid(vcpu);
705 	return 0;
706 }
707 
708 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
709 {
710 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
711 	    __kvm_set_xcr(vcpu, index, xcr)) {
712 		kvm_inject_gp(vcpu, 0);
713 		return 1;
714 	}
715 	return 0;
716 }
717 EXPORT_SYMBOL_GPL(kvm_set_xcr);
718 
719 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
720 {
721 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
722 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
723 				   X86_CR4_SMEP | X86_CR4_SMAP;
724 
725 	if (cr4 & CR4_RESERVED_BITS)
726 		return 1;
727 
728 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
729 		return 1;
730 
731 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
732 		return 1;
733 
734 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
735 		return 1;
736 
737 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
738 		return 1;
739 
740 	if (is_long_mode(vcpu)) {
741 		if (!(cr4 & X86_CR4_PAE))
742 			return 1;
743 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
744 		   && ((cr4 ^ old_cr4) & pdptr_bits)
745 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
746 				   kvm_read_cr3(vcpu)))
747 		return 1;
748 
749 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
750 		if (!guest_cpuid_has_pcid(vcpu))
751 			return 1;
752 
753 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
754 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
755 			return 1;
756 	}
757 
758 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
759 		return 1;
760 
761 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
762 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
763 		kvm_mmu_reset_context(vcpu);
764 
765 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
766 		kvm_update_cpuid(vcpu);
767 
768 	return 0;
769 }
770 EXPORT_SYMBOL_GPL(kvm_set_cr4);
771 
772 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
773 {
774 #ifdef CONFIG_X86_64
775 	cr3 &= ~CR3_PCID_INVD;
776 #endif
777 
778 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
779 		kvm_mmu_sync_roots(vcpu);
780 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
781 		return 0;
782 	}
783 
784 	if (is_long_mode(vcpu)) {
785 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
786 			return 1;
787 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
788 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
789 		return 1;
790 
791 	vcpu->arch.cr3 = cr3;
792 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
793 	kvm_mmu_new_cr3(vcpu);
794 	return 0;
795 }
796 EXPORT_SYMBOL_GPL(kvm_set_cr3);
797 
798 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
799 {
800 	if (cr8 & CR8_RESERVED_BITS)
801 		return 1;
802 	if (lapic_in_kernel(vcpu))
803 		kvm_lapic_set_tpr(vcpu, cr8);
804 	else
805 		vcpu->arch.cr8 = cr8;
806 	return 0;
807 }
808 EXPORT_SYMBOL_GPL(kvm_set_cr8);
809 
810 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
811 {
812 	if (lapic_in_kernel(vcpu))
813 		return kvm_lapic_get_cr8(vcpu);
814 	else
815 		return vcpu->arch.cr8;
816 }
817 EXPORT_SYMBOL_GPL(kvm_get_cr8);
818 
819 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
820 {
821 	int i;
822 
823 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
824 		for (i = 0; i < KVM_NR_DB_REGS; i++)
825 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
826 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
827 	}
828 }
829 
830 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
831 {
832 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
833 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
834 }
835 
836 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
837 {
838 	unsigned long dr7;
839 
840 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
841 		dr7 = vcpu->arch.guest_debug_dr7;
842 	else
843 		dr7 = vcpu->arch.dr7;
844 	kvm_x86_ops->set_dr7(vcpu, dr7);
845 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
846 	if (dr7 & DR7_BP_EN_MASK)
847 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
848 }
849 
850 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
851 {
852 	u64 fixed = DR6_FIXED_1;
853 
854 	if (!guest_cpuid_has_rtm(vcpu))
855 		fixed |= DR6_RTM;
856 	return fixed;
857 }
858 
859 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
860 {
861 	switch (dr) {
862 	case 0 ... 3:
863 		vcpu->arch.db[dr] = val;
864 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
865 			vcpu->arch.eff_db[dr] = val;
866 		break;
867 	case 4:
868 		/* fall through */
869 	case 6:
870 		if (val & 0xffffffff00000000ULL)
871 			return -1; /* #GP */
872 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
873 		kvm_update_dr6(vcpu);
874 		break;
875 	case 5:
876 		/* fall through */
877 	default: /* 7 */
878 		if (val & 0xffffffff00000000ULL)
879 			return -1; /* #GP */
880 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
881 		kvm_update_dr7(vcpu);
882 		break;
883 	}
884 
885 	return 0;
886 }
887 
888 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
889 {
890 	if (__kvm_set_dr(vcpu, dr, val)) {
891 		kvm_inject_gp(vcpu, 0);
892 		return 1;
893 	}
894 	return 0;
895 }
896 EXPORT_SYMBOL_GPL(kvm_set_dr);
897 
898 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
899 {
900 	switch (dr) {
901 	case 0 ... 3:
902 		*val = vcpu->arch.db[dr];
903 		break;
904 	case 4:
905 		/* fall through */
906 	case 6:
907 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
908 			*val = vcpu->arch.dr6;
909 		else
910 			*val = kvm_x86_ops->get_dr6(vcpu);
911 		break;
912 	case 5:
913 		/* fall through */
914 	default: /* 7 */
915 		*val = vcpu->arch.dr7;
916 		break;
917 	}
918 	return 0;
919 }
920 EXPORT_SYMBOL_GPL(kvm_get_dr);
921 
922 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
923 {
924 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
925 	u64 data;
926 	int err;
927 
928 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
929 	if (err)
930 		return err;
931 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
932 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
933 	return err;
934 }
935 EXPORT_SYMBOL_GPL(kvm_rdpmc);
936 
937 /*
938  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
939  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
940  *
941  * This list is modified at module load time to reflect the
942  * capabilities of the host cpu. This capabilities test skips MSRs that are
943  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
944  * may depend on host virtualization features rather than host cpu features.
945  */
946 
947 static u32 msrs_to_save[] = {
948 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
949 	MSR_STAR,
950 #ifdef CONFIG_X86_64
951 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
952 #endif
953 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
954 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
955 };
956 
957 static unsigned num_msrs_to_save;
958 
959 static u32 emulated_msrs[] = {
960 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
961 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
962 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
963 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
964 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
965 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
966 	HV_X64_MSR_RESET,
967 	HV_X64_MSR_VP_INDEX,
968 	HV_X64_MSR_VP_RUNTIME,
969 	HV_X64_MSR_SCONTROL,
970 	HV_X64_MSR_STIMER0_CONFIG,
971 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
972 	MSR_KVM_PV_EOI_EN,
973 
974 	MSR_IA32_TSC_ADJUST,
975 	MSR_IA32_TSCDEADLINE,
976 	MSR_IA32_MISC_ENABLE,
977 	MSR_IA32_MCG_STATUS,
978 	MSR_IA32_MCG_CTL,
979 	MSR_IA32_SMBASE,
980 };
981 
982 static unsigned num_emulated_msrs;
983 
984 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
985 {
986 	if (efer & efer_reserved_bits)
987 		return false;
988 
989 	if (efer & EFER_FFXSR) {
990 		struct kvm_cpuid_entry2 *feat;
991 
992 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
993 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
994 			return false;
995 	}
996 
997 	if (efer & EFER_SVME) {
998 		struct kvm_cpuid_entry2 *feat;
999 
1000 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1001 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1002 			return false;
1003 	}
1004 
1005 	return true;
1006 }
1007 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1008 
1009 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1010 {
1011 	u64 old_efer = vcpu->arch.efer;
1012 
1013 	if (!kvm_valid_efer(vcpu, efer))
1014 		return 1;
1015 
1016 	if (is_paging(vcpu)
1017 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1018 		return 1;
1019 
1020 	efer &= ~EFER_LMA;
1021 	efer |= vcpu->arch.efer & EFER_LMA;
1022 
1023 	kvm_x86_ops->set_efer(vcpu, efer);
1024 
1025 	/* Update reserved bits */
1026 	if ((efer ^ old_efer) & EFER_NX)
1027 		kvm_mmu_reset_context(vcpu);
1028 
1029 	return 0;
1030 }
1031 
1032 void kvm_enable_efer_bits(u64 mask)
1033 {
1034        efer_reserved_bits &= ~mask;
1035 }
1036 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1037 
1038 /*
1039  * Writes msr value into into the appropriate "register".
1040  * Returns 0 on success, non-0 otherwise.
1041  * Assumes vcpu_load() was already called.
1042  */
1043 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1044 {
1045 	switch (msr->index) {
1046 	case MSR_FS_BASE:
1047 	case MSR_GS_BASE:
1048 	case MSR_KERNEL_GS_BASE:
1049 	case MSR_CSTAR:
1050 	case MSR_LSTAR:
1051 		if (is_noncanonical_address(msr->data))
1052 			return 1;
1053 		break;
1054 	case MSR_IA32_SYSENTER_EIP:
1055 	case MSR_IA32_SYSENTER_ESP:
1056 		/*
1057 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1058 		 * non-canonical address is written on Intel but not on
1059 		 * AMD (which ignores the top 32-bits, because it does
1060 		 * not implement 64-bit SYSENTER).
1061 		 *
1062 		 * 64-bit code should hence be able to write a non-canonical
1063 		 * value on AMD.  Making the address canonical ensures that
1064 		 * vmentry does not fail on Intel after writing a non-canonical
1065 		 * value, and that something deterministic happens if the guest
1066 		 * invokes 64-bit SYSENTER.
1067 		 */
1068 		msr->data = get_canonical(msr->data);
1069 	}
1070 	return kvm_x86_ops->set_msr(vcpu, msr);
1071 }
1072 EXPORT_SYMBOL_GPL(kvm_set_msr);
1073 
1074 /*
1075  * Adapt set_msr() to msr_io()'s calling convention
1076  */
1077 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1078 {
1079 	struct msr_data msr;
1080 	int r;
1081 
1082 	msr.index = index;
1083 	msr.host_initiated = true;
1084 	r = kvm_get_msr(vcpu, &msr);
1085 	if (r)
1086 		return r;
1087 
1088 	*data = msr.data;
1089 	return 0;
1090 }
1091 
1092 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1093 {
1094 	struct msr_data msr;
1095 
1096 	msr.data = *data;
1097 	msr.index = index;
1098 	msr.host_initiated = true;
1099 	return kvm_set_msr(vcpu, &msr);
1100 }
1101 
1102 #ifdef CONFIG_X86_64
1103 struct pvclock_gtod_data {
1104 	seqcount_t	seq;
1105 
1106 	struct { /* extract of a clocksource struct */
1107 		int vclock_mode;
1108 		cycle_t	cycle_last;
1109 		cycle_t	mask;
1110 		u32	mult;
1111 		u32	shift;
1112 	} clock;
1113 
1114 	u64		boot_ns;
1115 	u64		nsec_base;
1116 };
1117 
1118 static struct pvclock_gtod_data pvclock_gtod_data;
1119 
1120 static void update_pvclock_gtod(struct timekeeper *tk)
1121 {
1122 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1123 	u64 boot_ns;
1124 
1125 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1126 
1127 	write_seqcount_begin(&vdata->seq);
1128 
1129 	/* copy pvclock gtod data */
1130 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1131 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1132 	vdata->clock.mask		= tk->tkr_mono.mask;
1133 	vdata->clock.mult		= tk->tkr_mono.mult;
1134 	vdata->clock.shift		= tk->tkr_mono.shift;
1135 
1136 	vdata->boot_ns			= boot_ns;
1137 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1138 
1139 	write_seqcount_end(&vdata->seq);
1140 }
1141 #endif
1142 
1143 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1144 {
1145 	/*
1146 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1147 	 * vcpu_enter_guest.  This function is only called from
1148 	 * the physical CPU that is running vcpu.
1149 	 */
1150 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1151 }
1152 
1153 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1154 {
1155 	int version;
1156 	int r;
1157 	struct pvclock_wall_clock wc;
1158 	struct timespec boot;
1159 
1160 	if (!wall_clock)
1161 		return;
1162 
1163 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1164 	if (r)
1165 		return;
1166 
1167 	if (version & 1)
1168 		++version;  /* first time write, random junk */
1169 
1170 	++version;
1171 
1172 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1173 		return;
1174 
1175 	/*
1176 	 * The guest calculates current wall clock time by adding
1177 	 * system time (updated by kvm_guest_time_update below) to the
1178 	 * wall clock specified here.  guest system time equals host
1179 	 * system time for us, thus we must fill in host boot time here.
1180 	 */
1181 	getboottime(&boot);
1182 
1183 	if (kvm->arch.kvmclock_offset) {
1184 		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1185 		boot = timespec_sub(boot, ts);
1186 	}
1187 	wc.sec = boot.tv_sec;
1188 	wc.nsec = boot.tv_nsec;
1189 	wc.version = version;
1190 
1191 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1192 
1193 	version++;
1194 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1195 }
1196 
1197 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1198 {
1199 	uint32_t quotient, remainder;
1200 
1201 	/* Don't try to replace with do_div(), this one calculates
1202 	 * "(dividend << 32) / divisor" */
1203 	__asm__ ( "divl %4"
1204 		  : "=a" (quotient), "=d" (remainder)
1205 		  : "0" (0), "1" (dividend), "r" (divisor) );
1206 	return quotient;
1207 }
1208 
1209 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1210 			       s8 *pshift, u32 *pmultiplier)
1211 {
1212 	uint64_t scaled64;
1213 	int32_t  shift = 0;
1214 	uint64_t tps64;
1215 	uint32_t tps32;
1216 
1217 	tps64 = base_khz * 1000LL;
1218 	scaled64 = scaled_khz * 1000LL;
1219 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1220 		tps64 >>= 1;
1221 		shift--;
1222 	}
1223 
1224 	tps32 = (uint32_t)tps64;
1225 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1226 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1227 			scaled64 >>= 1;
1228 		else
1229 			tps32 <<= 1;
1230 		shift++;
1231 	}
1232 
1233 	*pshift = shift;
1234 	*pmultiplier = div_frac(scaled64, tps32);
1235 
1236 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1237 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1238 }
1239 
1240 #ifdef CONFIG_X86_64
1241 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1242 #endif
1243 
1244 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1245 static unsigned long max_tsc_khz;
1246 
1247 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1248 {
1249 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1250 				   vcpu->arch.virtual_tsc_shift);
1251 }
1252 
1253 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1254 {
1255 	u64 v = (u64)khz * (1000000 + ppm);
1256 	do_div(v, 1000000);
1257 	return v;
1258 }
1259 
1260 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1261 {
1262 	u64 ratio;
1263 
1264 	/* Guest TSC same frequency as host TSC? */
1265 	if (!scale) {
1266 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1267 		return 0;
1268 	}
1269 
1270 	/* TSC scaling supported? */
1271 	if (!kvm_has_tsc_control) {
1272 		if (user_tsc_khz > tsc_khz) {
1273 			vcpu->arch.tsc_catchup = 1;
1274 			vcpu->arch.tsc_always_catchup = 1;
1275 			return 0;
1276 		} else {
1277 			WARN(1, "user requested TSC rate below hardware speed\n");
1278 			return -1;
1279 		}
1280 	}
1281 
1282 	/* TSC scaling required  - calculate ratio */
1283 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1284 				user_tsc_khz, tsc_khz);
1285 
1286 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1287 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1288 			  user_tsc_khz);
1289 		return -1;
1290 	}
1291 
1292 	vcpu->arch.tsc_scaling_ratio = ratio;
1293 	return 0;
1294 }
1295 
1296 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1297 {
1298 	u32 thresh_lo, thresh_hi;
1299 	int use_scaling = 0;
1300 
1301 	/* tsc_khz can be zero if TSC calibration fails */
1302 	if (this_tsc_khz == 0) {
1303 		/* set tsc_scaling_ratio to a safe value */
1304 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1305 		return -1;
1306 	}
1307 
1308 	/* Compute a scale to convert nanoseconds in TSC cycles */
1309 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1310 			   &vcpu->arch.virtual_tsc_shift,
1311 			   &vcpu->arch.virtual_tsc_mult);
1312 	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1313 
1314 	/*
1315 	 * Compute the variation in TSC rate which is acceptable
1316 	 * within the range of tolerance and decide if the
1317 	 * rate being applied is within that bounds of the hardware
1318 	 * rate.  If so, no scaling or compensation need be done.
1319 	 */
1320 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1321 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1322 	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1323 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1324 		use_scaling = 1;
1325 	}
1326 	return set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1327 }
1328 
1329 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1330 {
1331 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1332 				      vcpu->arch.virtual_tsc_mult,
1333 				      vcpu->arch.virtual_tsc_shift);
1334 	tsc += vcpu->arch.this_tsc_write;
1335 	return tsc;
1336 }
1337 
1338 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1339 {
1340 #ifdef CONFIG_X86_64
1341 	bool vcpus_matched;
1342 	struct kvm_arch *ka = &vcpu->kvm->arch;
1343 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1344 
1345 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1346 			 atomic_read(&vcpu->kvm->online_vcpus));
1347 
1348 	/*
1349 	 * Once the masterclock is enabled, always perform request in
1350 	 * order to update it.
1351 	 *
1352 	 * In order to enable masterclock, the host clocksource must be TSC
1353 	 * and the vcpus need to have matched TSCs.  When that happens,
1354 	 * perform request to enable masterclock.
1355 	 */
1356 	if (ka->use_master_clock ||
1357 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1358 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1359 
1360 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1361 			    atomic_read(&vcpu->kvm->online_vcpus),
1362 		            ka->use_master_clock, gtod->clock.vclock_mode);
1363 #endif
1364 }
1365 
1366 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1367 {
1368 	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1369 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1370 }
1371 
1372 /*
1373  * Multiply tsc by a fixed point number represented by ratio.
1374  *
1375  * The most significant 64-N bits (mult) of ratio represent the
1376  * integral part of the fixed point number; the remaining N bits
1377  * (frac) represent the fractional part, ie. ratio represents a fixed
1378  * point number (mult + frac * 2^(-N)).
1379  *
1380  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1381  */
1382 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1383 {
1384 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1385 }
1386 
1387 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1388 {
1389 	u64 _tsc = tsc;
1390 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1391 
1392 	if (ratio != kvm_default_tsc_scaling_ratio)
1393 		_tsc = __scale_tsc(ratio, tsc);
1394 
1395 	return _tsc;
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1398 
1399 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1400 {
1401 	u64 tsc;
1402 
1403 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1404 
1405 	return target_tsc - tsc;
1406 }
1407 
1408 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1409 {
1410 	return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1413 
1414 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1415 {
1416 	struct kvm *kvm = vcpu->kvm;
1417 	u64 offset, ns, elapsed;
1418 	unsigned long flags;
1419 	s64 usdiff;
1420 	bool matched;
1421 	bool already_matched;
1422 	u64 data = msr->data;
1423 
1424 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1425 	offset = kvm_compute_tsc_offset(vcpu, data);
1426 	ns = get_kernel_ns();
1427 	elapsed = ns - kvm->arch.last_tsc_nsec;
1428 
1429 	if (vcpu->arch.virtual_tsc_khz) {
1430 		int faulted = 0;
1431 
1432 		/* n.b - signed multiplication and division required */
1433 		usdiff = data - kvm->arch.last_tsc_write;
1434 #ifdef CONFIG_X86_64
1435 		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1436 #else
1437 		/* do_div() only does unsigned */
1438 		asm("1: idivl %[divisor]\n"
1439 		    "2: xor %%edx, %%edx\n"
1440 		    "   movl $0, %[faulted]\n"
1441 		    "3:\n"
1442 		    ".section .fixup,\"ax\"\n"
1443 		    "4: movl $1, %[faulted]\n"
1444 		    "   jmp  3b\n"
1445 		    ".previous\n"
1446 
1447 		_ASM_EXTABLE(1b, 4b)
1448 
1449 		: "=A"(usdiff), [faulted] "=r" (faulted)
1450 		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1451 
1452 #endif
1453 		do_div(elapsed, 1000);
1454 		usdiff -= elapsed;
1455 		if (usdiff < 0)
1456 			usdiff = -usdiff;
1457 
1458 		/* idivl overflow => difference is larger than USEC_PER_SEC */
1459 		if (faulted)
1460 			usdiff = USEC_PER_SEC;
1461 	} else
1462 		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1463 
1464 	/*
1465 	 * Special case: TSC write with a small delta (1 second) of virtual
1466 	 * cycle time against real time is interpreted as an attempt to
1467 	 * synchronize the CPU.
1468          *
1469 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1470 	 * TSC, we add elapsed time in this computation.  We could let the
1471 	 * compensation code attempt to catch up if we fall behind, but
1472 	 * it's better to try to match offsets from the beginning.
1473          */
1474 	if (usdiff < USEC_PER_SEC &&
1475 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1476 		if (!check_tsc_unstable()) {
1477 			offset = kvm->arch.cur_tsc_offset;
1478 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1479 		} else {
1480 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1481 			data += delta;
1482 			offset = kvm_compute_tsc_offset(vcpu, data);
1483 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1484 		}
1485 		matched = true;
1486 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1487 	} else {
1488 		/*
1489 		 * We split periods of matched TSC writes into generations.
1490 		 * For each generation, we track the original measured
1491 		 * nanosecond time, offset, and write, so if TSCs are in
1492 		 * sync, we can match exact offset, and if not, we can match
1493 		 * exact software computation in compute_guest_tsc()
1494 		 *
1495 		 * These values are tracked in kvm->arch.cur_xxx variables.
1496 		 */
1497 		kvm->arch.cur_tsc_generation++;
1498 		kvm->arch.cur_tsc_nsec = ns;
1499 		kvm->arch.cur_tsc_write = data;
1500 		kvm->arch.cur_tsc_offset = offset;
1501 		matched = false;
1502 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1503 			 kvm->arch.cur_tsc_generation, data);
1504 	}
1505 
1506 	/*
1507 	 * We also track th most recent recorded KHZ, write and time to
1508 	 * allow the matching interval to be extended at each write.
1509 	 */
1510 	kvm->arch.last_tsc_nsec = ns;
1511 	kvm->arch.last_tsc_write = data;
1512 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1513 
1514 	vcpu->arch.last_guest_tsc = data;
1515 
1516 	/* Keep track of which generation this VCPU has synchronized to */
1517 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1518 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1519 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1520 
1521 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1522 		update_ia32_tsc_adjust_msr(vcpu, offset);
1523 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1524 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1525 
1526 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1527 	if (!matched) {
1528 		kvm->arch.nr_vcpus_matched_tsc = 0;
1529 	} else if (!already_matched) {
1530 		kvm->arch.nr_vcpus_matched_tsc++;
1531 	}
1532 
1533 	kvm_track_tsc_matching(vcpu);
1534 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1535 }
1536 
1537 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1538 
1539 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1540 					   s64 adjustment)
1541 {
1542 	kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1543 }
1544 
1545 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1546 {
1547 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1548 		WARN_ON(adjustment < 0);
1549 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1550 	kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1551 }
1552 
1553 #ifdef CONFIG_X86_64
1554 
1555 static cycle_t read_tsc(void)
1556 {
1557 	cycle_t ret = (cycle_t)rdtsc_ordered();
1558 	u64 last = pvclock_gtod_data.clock.cycle_last;
1559 
1560 	if (likely(ret >= last))
1561 		return ret;
1562 
1563 	/*
1564 	 * GCC likes to generate cmov here, but this branch is extremely
1565 	 * predictable (it's just a funciton of time and the likely is
1566 	 * very likely) and there's a data dependence, so force GCC
1567 	 * to generate a branch instead.  I don't barrier() because
1568 	 * we don't actually need a barrier, and if this function
1569 	 * ever gets inlined it will generate worse code.
1570 	 */
1571 	asm volatile ("");
1572 	return last;
1573 }
1574 
1575 static inline u64 vgettsc(cycle_t *cycle_now)
1576 {
1577 	long v;
1578 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1579 
1580 	*cycle_now = read_tsc();
1581 
1582 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1583 	return v * gtod->clock.mult;
1584 }
1585 
1586 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1587 {
1588 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1589 	unsigned long seq;
1590 	int mode;
1591 	u64 ns;
1592 
1593 	do {
1594 		seq = read_seqcount_begin(&gtod->seq);
1595 		mode = gtod->clock.vclock_mode;
1596 		ns = gtod->nsec_base;
1597 		ns += vgettsc(cycle_now);
1598 		ns >>= gtod->clock.shift;
1599 		ns += gtod->boot_ns;
1600 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1601 	*t = ns;
1602 
1603 	return mode;
1604 }
1605 
1606 /* returns true if host is using tsc clocksource */
1607 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1608 {
1609 	/* checked again under seqlock below */
1610 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1611 		return false;
1612 
1613 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1614 }
1615 #endif
1616 
1617 /*
1618  *
1619  * Assuming a stable TSC across physical CPUS, and a stable TSC
1620  * across virtual CPUs, the following condition is possible.
1621  * Each numbered line represents an event visible to both
1622  * CPUs at the next numbered event.
1623  *
1624  * "timespecX" represents host monotonic time. "tscX" represents
1625  * RDTSC value.
1626  *
1627  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1628  *
1629  * 1.  read timespec0,tsc0
1630  * 2.					| timespec1 = timespec0 + N
1631  * 					| tsc1 = tsc0 + M
1632  * 3. transition to guest		| transition to guest
1633  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1634  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1635  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1636  *
1637  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1638  *
1639  * 	- ret0 < ret1
1640  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1641  *		...
1642  *	- 0 < N - M => M < N
1643  *
1644  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1645  * always the case (the difference between two distinct xtime instances
1646  * might be smaller then the difference between corresponding TSC reads,
1647  * when updating guest vcpus pvclock areas).
1648  *
1649  * To avoid that problem, do not allow visibility of distinct
1650  * system_timestamp/tsc_timestamp values simultaneously: use a master
1651  * copy of host monotonic time values. Update that master copy
1652  * in lockstep.
1653  *
1654  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1655  *
1656  */
1657 
1658 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1659 {
1660 #ifdef CONFIG_X86_64
1661 	struct kvm_arch *ka = &kvm->arch;
1662 	int vclock_mode;
1663 	bool host_tsc_clocksource, vcpus_matched;
1664 
1665 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1666 			atomic_read(&kvm->online_vcpus));
1667 
1668 	/*
1669 	 * If the host uses TSC clock, then passthrough TSC as stable
1670 	 * to the guest.
1671 	 */
1672 	host_tsc_clocksource = kvm_get_time_and_clockread(
1673 					&ka->master_kernel_ns,
1674 					&ka->master_cycle_now);
1675 
1676 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1677 				&& !backwards_tsc_observed
1678 				&& !ka->boot_vcpu_runs_old_kvmclock;
1679 
1680 	if (ka->use_master_clock)
1681 		atomic_set(&kvm_guest_has_master_clock, 1);
1682 
1683 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1684 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1685 					vcpus_matched);
1686 #endif
1687 }
1688 
1689 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1690 {
1691 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1692 }
1693 
1694 static void kvm_gen_update_masterclock(struct kvm *kvm)
1695 {
1696 #ifdef CONFIG_X86_64
1697 	int i;
1698 	struct kvm_vcpu *vcpu;
1699 	struct kvm_arch *ka = &kvm->arch;
1700 
1701 	spin_lock(&ka->pvclock_gtod_sync_lock);
1702 	kvm_make_mclock_inprogress_request(kvm);
1703 	/* no guest entries from this point */
1704 	pvclock_update_vm_gtod_copy(kvm);
1705 
1706 	kvm_for_each_vcpu(i, vcpu, kvm)
1707 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1708 
1709 	/* guest entries allowed */
1710 	kvm_for_each_vcpu(i, vcpu, kvm)
1711 		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1712 
1713 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1714 #endif
1715 }
1716 
1717 static int kvm_guest_time_update(struct kvm_vcpu *v)
1718 {
1719 	unsigned long flags, this_tsc_khz, tgt_tsc_khz;
1720 	struct kvm_vcpu_arch *vcpu = &v->arch;
1721 	struct kvm_arch *ka = &v->kvm->arch;
1722 	s64 kernel_ns;
1723 	u64 tsc_timestamp, host_tsc;
1724 	struct pvclock_vcpu_time_info guest_hv_clock;
1725 	u8 pvclock_flags;
1726 	bool use_master_clock;
1727 
1728 	kernel_ns = 0;
1729 	host_tsc = 0;
1730 
1731 	/*
1732 	 * If the host uses TSC clock, then passthrough TSC as stable
1733 	 * to the guest.
1734 	 */
1735 	spin_lock(&ka->pvclock_gtod_sync_lock);
1736 	use_master_clock = ka->use_master_clock;
1737 	if (use_master_clock) {
1738 		host_tsc = ka->master_cycle_now;
1739 		kernel_ns = ka->master_kernel_ns;
1740 	}
1741 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1742 
1743 	/* Keep irq disabled to prevent changes to the clock */
1744 	local_irq_save(flags);
1745 	this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1746 	if (unlikely(this_tsc_khz == 0)) {
1747 		local_irq_restore(flags);
1748 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1749 		return 1;
1750 	}
1751 	if (!use_master_clock) {
1752 		host_tsc = rdtsc();
1753 		kernel_ns = get_kernel_ns();
1754 	}
1755 
1756 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1757 
1758 	/*
1759 	 * We may have to catch up the TSC to match elapsed wall clock
1760 	 * time for two reasons, even if kvmclock is used.
1761 	 *   1) CPU could have been running below the maximum TSC rate
1762 	 *   2) Broken TSC compensation resets the base at each VCPU
1763 	 *      entry to avoid unknown leaps of TSC even when running
1764 	 *      again on the same CPU.  This may cause apparent elapsed
1765 	 *      time to disappear, and the guest to stand still or run
1766 	 *	very slowly.
1767 	 */
1768 	if (vcpu->tsc_catchup) {
1769 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1770 		if (tsc > tsc_timestamp) {
1771 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1772 			tsc_timestamp = tsc;
1773 		}
1774 	}
1775 
1776 	local_irq_restore(flags);
1777 
1778 	if (!vcpu->pv_time_enabled)
1779 		return 0;
1780 
1781 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1782 		tgt_tsc_khz = kvm_has_tsc_control ?
1783 			vcpu->virtual_tsc_khz : this_tsc_khz;
1784 		kvm_get_time_scale(NSEC_PER_SEC / 1000, tgt_tsc_khz,
1785 				   &vcpu->hv_clock.tsc_shift,
1786 				   &vcpu->hv_clock.tsc_to_system_mul);
1787 		vcpu->hw_tsc_khz = this_tsc_khz;
1788 	}
1789 
1790 	/* With all the info we got, fill in the values */
1791 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1792 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1793 	vcpu->last_guest_tsc = tsc_timestamp;
1794 
1795 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1796 		&guest_hv_clock, sizeof(guest_hv_clock))))
1797 		return 0;
1798 
1799 	/* This VCPU is paused, but it's legal for a guest to read another
1800 	 * VCPU's kvmclock, so we really have to follow the specification where
1801 	 * it says that version is odd if data is being modified, and even after
1802 	 * it is consistent.
1803 	 *
1804 	 * Version field updates must be kept separate.  This is because
1805 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1806 	 * writes within a string instruction are weakly ordered.  So there
1807 	 * are three writes overall.
1808 	 *
1809 	 * As a small optimization, only write the version field in the first
1810 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1811 	 * version field is the first in the struct.
1812 	 */
1813 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1814 
1815 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1816 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1817 				&vcpu->hv_clock,
1818 				sizeof(vcpu->hv_clock.version));
1819 
1820 	smp_wmb();
1821 
1822 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1823 	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1824 
1825 	if (vcpu->pvclock_set_guest_stopped_request) {
1826 		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1827 		vcpu->pvclock_set_guest_stopped_request = false;
1828 	}
1829 
1830 	/* If the host uses TSC clocksource, then it is stable */
1831 	if (use_master_clock)
1832 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1833 
1834 	vcpu->hv_clock.flags = pvclock_flags;
1835 
1836 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1837 
1838 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1839 				&vcpu->hv_clock,
1840 				sizeof(vcpu->hv_clock));
1841 
1842 	smp_wmb();
1843 
1844 	vcpu->hv_clock.version++;
1845 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1846 				&vcpu->hv_clock,
1847 				sizeof(vcpu->hv_clock.version));
1848 	return 0;
1849 }
1850 
1851 /*
1852  * kvmclock updates which are isolated to a given vcpu, such as
1853  * vcpu->cpu migration, should not allow system_timestamp from
1854  * the rest of the vcpus to remain static. Otherwise ntp frequency
1855  * correction applies to one vcpu's system_timestamp but not
1856  * the others.
1857  *
1858  * So in those cases, request a kvmclock update for all vcpus.
1859  * We need to rate-limit these requests though, as they can
1860  * considerably slow guests that have a large number of vcpus.
1861  * The time for a remote vcpu to update its kvmclock is bound
1862  * by the delay we use to rate-limit the updates.
1863  */
1864 
1865 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1866 
1867 static void kvmclock_update_fn(struct work_struct *work)
1868 {
1869 	int i;
1870 	struct delayed_work *dwork = to_delayed_work(work);
1871 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1872 					   kvmclock_update_work);
1873 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1874 	struct kvm_vcpu *vcpu;
1875 
1876 	kvm_for_each_vcpu(i, vcpu, kvm) {
1877 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1878 		kvm_vcpu_kick(vcpu);
1879 	}
1880 }
1881 
1882 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1883 {
1884 	struct kvm *kvm = v->kvm;
1885 
1886 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1887 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1888 					KVMCLOCK_UPDATE_DELAY);
1889 }
1890 
1891 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1892 
1893 static void kvmclock_sync_fn(struct work_struct *work)
1894 {
1895 	struct delayed_work *dwork = to_delayed_work(work);
1896 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1897 					   kvmclock_sync_work);
1898 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1899 
1900 	if (!kvmclock_periodic_sync)
1901 		return;
1902 
1903 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1904 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1905 					KVMCLOCK_SYNC_PERIOD);
1906 }
1907 
1908 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1909 {
1910 	u64 mcg_cap = vcpu->arch.mcg_cap;
1911 	unsigned bank_num = mcg_cap & 0xff;
1912 
1913 	switch (msr) {
1914 	case MSR_IA32_MCG_STATUS:
1915 		vcpu->arch.mcg_status = data;
1916 		break;
1917 	case MSR_IA32_MCG_CTL:
1918 		if (!(mcg_cap & MCG_CTL_P))
1919 			return 1;
1920 		if (data != 0 && data != ~(u64)0)
1921 			return -1;
1922 		vcpu->arch.mcg_ctl = data;
1923 		break;
1924 	default:
1925 		if (msr >= MSR_IA32_MC0_CTL &&
1926 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
1927 			u32 offset = msr - MSR_IA32_MC0_CTL;
1928 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1929 			 * some Linux kernels though clear bit 10 in bank 4 to
1930 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1931 			 * this to avoid an uncatched #GP in the guest
1932 			 */
1933 			if ((offset & 0x3) == 0 &&
1934 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1935 				return -1;
1936 			vcpu->arch.mce_banks[offset] = data;
1937 			break;
1938 		}
1939 		return 1;
1940 	}
1941 	return 0;
1942 }
1943 
1944 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1945 {
1946 	struct kvm *kvm = vcpu->kvm;
1947 	int lm = is_long_mode(vcpu);
1948 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1949 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1950 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1951 		: kvm->arch.xen_hvm_config.blob_size_32;
1952 	u32 page_num = data & ~PAGE_MASK;
1953 	u64 page_addr = data & PAGE_MASK;
1954 	u8 *page;
1955 	int r;
1956 
1957 	r = -E2BIG;
1958 	if (page_num >= blob_size)
1959 		goto out;
1960 	r = -ENOMEM;
1961 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1962 	if (IS_ERR(page)) {
1963 		r = PTR_ERR(page);
1964 		goto out;
1965 	}
1966 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1967 		goto out_free;
1968 	r = 0;
1969 out_free:
1970 	kfree(page);
1971 out:
1972 	return r;
1973 }
1974 
1975 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1976 {
1977 	gpa_t gpa = data & ~0x3f;
1978 
1979 	/* Bits 2:5 are reserved, Should be zero */
1980 	if (data & 0x3c)
1981 		return 1;
1982 
1983 	vcpu->arch.apf.msr_val = data;
1984 
1985 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1986 		kvm_clear_async_pf_completion_queue(vcpu);
1987 		kvm_async_pf_hash_reset(vcpu);
1988 		return 0;
1989 	}
1990 
1991 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1992 					sizeof(u32)))
1993 		return 1;
1994 
1995 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1996 	kvm_async_pf_wakeup_all(vcpu);
1997 	return 0;
1998 }
1999 
2000 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2001 {
2002 	vcpu->arch.pv_time_enabled = false;
2003 }
2004 
2005 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2006 {
2007 	u64 delta;
2008 
2009 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2010 		return;
2011 
2012 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2013 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2014 	vcpu->arch.st.accum_steal = delta;
2015 }
2016 
2017 static void record_steal_time(struct kvm_vcpu *vcpu)
2018 {
2019 	accumulate_steal_time(vcpu);
2020 
2021 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2022 		return;
2023 
2024 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2025 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2026 		return;
2027 
2028 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2029 	vcpu->arch.st.steal.version += 2;
2030 	vcpu->arch.st.accum_steal = 0;
2031 
2032 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2033 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2034 }
2035 
2036 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2037 {
2038 	bool pr = false;
2039 	u32 msr = msr_info->index;
2040 	u64 data = msr_info->data;
2041 
2042 	switch (msr) {
2043 	case MSR_AMD64_NB_CFG:
2044 	case MSR_IA32_UCODE_REV:
2045 	case MSR_IA32_UCODE_WRITE:
2046 	case MSR_VM_HSAVE_PA:
2047 	case MSR_AMD64_PATCH_LOADER:
2048 	case MSR_AMD64_BU_CFG2:
2049 		break;
2050 
2051 	case MSR_EFER:
2052 		return set_efer(vcpu, data);
2053 	case MSR_K7_HWCR:
2054 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2055 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2056 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2057 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2058 		if (data != 0) {
2059 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2060 				    data);
2061 			return 1;
2062 		}
2063 		break;
2064 	case MSR_FAM10H_MMIO_CONF_BASE:
2065 		if (data != 0) {
2066 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2067 				    "0x%llx\n", data);
2068 			return 1;
2069 		}
2070 		break;
2071 	case MSR_IA32_DEBUGCTLMSR:
2072 		if (!data) {
2073 			/* We support the non-activated case already */
2074 			break;
2075 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2076 			/* Values other than LBR and BTF are vendor-specific,
2077 			   thus reserved and should throw a #GP */
2078 			return 1;
2079 		}
2080 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2081 			    __func__, data);
2082 		break;
2083 	case 0x200 ... 0x2ff:
2084 		return kvm_mtrr_set_msr(vcpu, msr, data);
2085 	case MSR_IA32_APICBASE:
2086 		return kvm_set_apic_base(vcpu, msr_info);
2087 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2088 		return kvm_x2apic_msr_write(vcpu, msr, data);
2089 	case MSR_IA32_TSCDEADLINE:
2090 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2091 		break;
2092 	case MSR_IA32_TSC_ADJUST:
2093 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2094 			if (!msr_info->host_initiated) {
2095 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2096 				adjust_tsc_offset_guest(vcpu, adj);
2097 			}
2098 			vcpu->arch.ia32_tsc_adjust_msr = data;
2099 		}
2100 		break;
2101 	case MSR_IA32_MISC_ENABLE:
2102 		vcpu->arch.ia32_misc_enable_msr = data;
2103 		break;
2104 	case MSR_IA32_SMBASE:
2105 		if (!msr_info->host_initiated)
2106 			return 1;
2107 		vcpu->arch.smbase = data;
2108 		break;
2109 	case MSR_KVM_WALL_CLOCK_NEW:
2110 	case MSR_KVM_WALL_CLOCK:
2111 		vcpu->kvm->arch.wall_clock = data;
2112 		kvm_write_wall_clock(vcpu->kvm, data);
2113 		break;
2114 	case MSR_KVM_SYSTEM_TIME_NEW:
2115 	case MSR_KVM_SYSTEM_TIME: {
2116 		u64 gpa_offset;
2117 		struct kvm_arch *ka = &vcpu->kvm->arch;
2118 
2119 		kvmclock_reset(vcpu);
2120 
2121 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2122 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2123 
2124 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2125 				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2126 					&vcpu->requests);
2127 
2128 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2129 		}
2130 
2131 		vcpu->arch.time = data;
2132 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2133 
2134 		/* we verify if the enable bit is set... */
2135 		if (!(data & 1))
2136 			break;
2137 
2138 		gpa_offset = data & ~(PAGE_MASK | 1);
2139 
2140 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2141 		     &vcpu->arch.pv_time, data & ~1ULL,
2142 		     sizeof(struct pvclock_vcpu_time_info)))
2143 			vcpu->arch.pv_time_enabled = false;
2144 		else
2145 			vcpu->arch.pv_time_enabled = true;
2146 
2147 		break;
2148 	}
2149 	case MSR_KVM_ASYNC_PF_EN:
2150 		if (kvm_pv_enable_async_pf(vcpu, data))
2151 			return 1;
2152 		break;
2153 	case MSR_KVM_STEAL_TIME:
2154 
2155 		if (unlikely(!sched_info_on()))
2156 			return 1;
2157 
2158 		if (data & KVM_STEAL_RESERVED_MASK)
2159 			return 1;
2160 
2161 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2162 						data & KVM_STEAL_VALID_BITS,
2163 						sizeof(struct kvm_steal_time)))
2164 			return 1;
2165 
2166 		vcpu->arch.st.msr_val = data;
2167 
2168 		if (!(data & KVM_MSR_ENABLED))
2169 			break;
2170 
2171 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2172 
2173 		break;
2174 	case MSR_KVM_PV_EOI_EN:
2175 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2176 			return 1;
2177 		break;
2178 
2179 	case MSR_IA32_MCG_CTL:
2180 	case MSR_IA32_MCG_STATUS:
2181 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2182 		return set_msr_mce(vcpu, msr, data);
2183 
2184 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2185 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2186 		pr = true; /* fall through */
2187 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2188 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2189 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2190 			return kvm_pmu_set_msr(vcpu, msr_info);
2191 
2192 		if (pr || data != 0)
2193 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2194 				    "0x%x data 0x%llx\n", msr, data);
2195 		break;
2196 	case MSR_K7_CLK_CTL:
2197 		/*
2198 		 * Ignore all writes to this no longer documented MSR.
2199 		 * Writes are only relevant for old K7 processors,
2200 		 * all pre-dating SVM, but a recommended workaround from
2201 		 * AMD for these chips. It is possible to specify the
2202 		 * affected processor models on the command line, hence
2203 		 * the need to ignore the workaround.
2204 		 */
2205 		break;
2206 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2207 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2208 	case HV_X64_MSR_CRASH_CTL:
2209 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2210 		return kvm_hv_set_msr_common(vcpu, msr, data,
2211 					     msr_info->host_initiated);
2212 	case MSR_IA32_BBL_CR_CTL3:
2213 		/* Drop writes to this legacy MSR -- see rdmsr
2214 		 * counterpart for further detail.
2215 		 */
2216 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2217 		break;
2218 	case MSR_AMD64_OSVW_ID_LENGTH:
2219 		if (!guest_cpuid_has_osvw(vcpu))
2220 			return 1;
2221 		vcpu->arch.osvw.length = data;
2222 		break;
2223 	case MSR_AMD64_OSVW_STATUS:
2224 		if (!guest_cpuid_has_osvw(vcpu))
2225 			return 1;
2226 		vcpu->arch.osvw.status = data;
2227 		break;
2228 	default:
2229 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2230 			return xen_hvm_config(vcpu, data);
2231 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2232 			return kvm_pmu_set_msr(vcpu, msr_info);
2233 		if (!ignore_msrs) {
2234 			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2235 				    msr, data);
2236 			return 1;
2237 		} else {
2238 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2239 				    msr, data);
2240 			break;
2241 		}
2242 	}
2243 	return 0;
2244 }
2245 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2246 
2247 
2248 /*
2249  * Reads an msr value (of 'msr_index') into 'pdata'.
2250  * Returns 0 on success, non-0 otherwise.
2251  * Assumes vcpu_load() was already called.
2252  */
2253 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2254 {
2255 	return kvm_x86_ops->get_msr(vcpu, msr);
2256 }
2257 EXPORT_SYMBOL_GPL(kvm_get_msr);
2258 
2259 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2260 {
2261 	u64 data;
2262 	u64 mcg_cap = vcpu->arch.mcg_cap;
2263 	unsigned bank_num = mcg_cap & 0xff;
2264 
2265 	switch (msr) {
2266 	case MSR_IA32_P5_MC_ADDR:
2267 	case MSR_IA32_P5_MC_TYPE:
2268 		data = 0;
2269 		break;
2270 	case MSR_IA32_MCG_CAP:
2271 		data = vcpu->arch.mcg_cap;
2272 		break;
2273 	case MSR_IA32_MCG_CTL:
2274 		if (!(mcg_cap & MCG_CTL_P))
2275 			return 1;
2276 		data = vcpu->arch.mcg_ctl;
2277 		break;
2278 	case MSR_IA32_MCG_STATUS:
2279 		data = vcpu->arch.mcg_status;
2280 		break;
2281 	default:
2282 		if (msr >= MSR_IA32_MC0_CTL &&
2283 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2284 			u32 offset = msr - MSR_IA32_MC0_CTL;
2285 			data = vcpu->arch.mce_banks[offset];
2286 			break;
2287 		}
2288 		return 1;
2289 	}
2290 	*pdata = data;
2291 	return 0;
2292 }
2293 
2294 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2295 {
2296 	switch (msr_info->index) {
2297 	case MSR_IA32_PLATFORM_ID:
2298 	case MSR_IA32_EBL_CR_POWERON:
2299 	case MSR_IA32_DEBUGCTLMSR:
2300 	case MSR_IA32_LASTBRANCHFROMIP:
2301 	case MSR_IA32_LASTBRANCHTOIP:
2302 	case MSR_IA32_LASTINTFROMIP:
2303 	case MSR_IA32_LASTINTTOIP:
2304 	case MSR_K8_SYSCFG:
2305 	case MSR_K8_TSEG_ADDR:
2306 	case MSR_K8_TSEG_MASK:
2307 	case MSR_K7_HWCR:
2308 	case MSR_VM_HSAVE_PA:
2309 	case MSR_K8_INT_PENDING_MSG:
2310 	case MSR_AMD64_NB_CFG:
2311 	case MSR_FAM10H_MMIO_CONF_BASE:
2312 	case MSR_AMD64_BU_CFG2:
2313 		msr_info->data = 0;
2314 		break;
2315 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2316 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2317 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2318 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2319 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2320 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2321 		msr_info->data = 0;
2322 		break;
2323 	case MSR_IA32_UCODE_REV:
2324 		msr_info->data = 0x100000000ULL;
2325 		break;
2326 	case MSR_MTRRcap:
2327 	case 0x200 ... 0x2ff:
2328 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2329 	case 0xcd: /* fsb frequency */
2330 		msr_info->data = 3;
2331 		break;
2332 		/*
2333 		 * MSR_EBC_FREQUENCY_ID
2334 		 * Conservative value valid for even the basic CPU models.
2335 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2336 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2337 		 * and 266MHz for model 3, or 4. Set Core Clock
2338 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2339 		 * 31:24) even though these are only valid for CPU
2340 		 * models > 2, however guests may end up dividing or
2341 		 * multiplying by zero otherwise.
2342 		 */
2343 	case MSR_EBC_FREQUENCY_ID:
2344 		msr_info->data = 1 << 24;
2345 		break;
2346 	case MSR_IA32_APICBASE:
2347 		msr_info->data = kvm_get_apic_base(vcpu);
2348 		break;
2349 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2350 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2351 		break;
2352 	case MSR_IA32_TSCDEADLINE:
2353 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2354 		break;
2355 	case MSR_IA32_TSC_ADJUST:
2356 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2357 		break;
2358 	case MSR_IA32_MISC_ENABLE:
2359 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2360 		break;
2361 	case MSR_IA32_SMBASE:
2362 		if (!msr_info->host_initiated)
2363 			return 1;
2364 		msr_info->data = vcpu->arch.smbase;
2365 		break;
2366 	case MSR_IA32_PERF_STATUS:
2367 		/* TSC increment by tick */
2368 		msr_info->data = 1000ULL;
2369 		/* CPU multiplier */
2370 		msr_info->data |= (((uint64_t)4ULL) << 40);
2371 		break;
2372 	case MSR_EFER:
2373 		msr_info->data = vcpu->arch.efer;
2374 		break;
2375 	case MSR_KVM_WALL_CLOCK:
2376 	case MSR_KVM_WALL_CLOCK_NEW:
2377 		msr_info->data = vcpu->kvm->arch.wall_clock;
2378 		break;
2379 	case MSR_KVM_SYSTEM_TIME:
2380 	case MSR_KVM_SYSTEM_TIME_NEW:
2381 		msr_info->data = vcpu->arch.time;
2382 		break;
2383 	case MSR_KVM_ASYNC_PF_EN:
2384 		msr_info->data = vcpu->arch.apf.msr_val;
2385 		break;
2386 	case MSR_KVM_STEAL_TIME:
2387 		msr_info->data = vcpu->arch.st.msr_val;
2388 		break;
2389 	case MSR_KVM_PV_EOI_EN:
2390 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2391 		break;
2392 	case MSR_IA32_P5_MC_ADDR:
2393 	case MSR_IA32_P5_MC_TYPE:
2394 	case MSR_IA32_MCG_CAP:
2395 	case MSR_IA32_MCG_CTL:
2396 	case MSR_IA32_MCG_STATUS:
2397 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2398 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2399 	case MSR_K7_CLK_CTL:
2400 		/*
2401 		 * Provide expected ramp-up count for K7. All other
2402 		 * are set to zero, indicating minimum divisors for
2403 		 * every field.
2404 		 *
2405 		 * This prevents guest kernels on AMD host with CPU
2406 		 * type 6, model 8 and higher from exploding due to
2407 		 * the rdmsr failing.
2408 		 */
2409 		msr_info->data = 0x20000000;
2410 		break;
2411 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2412 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2413 	case HV_X64_MSR_CRASH_CTL:
2414 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2415 		return kvm_hv_get_msr_common(vcpu,
2416 					     msr_info->index, &msr_info->data);
2417 		break;
2418 	case MSR_IA32_BBL_CR_CTL3:
2419 		/* This legacy MSR exists but isn't fully documented in current
2420 		 * silicon.  It is however accessed by winxp in very narrow
2421 		 * scenarios where it sets bit #19, itself documented as
2422 		 * a "reserved" bit.  Best effort attempt to source coherent
2423 		 * read data here should the balance of the register be
2424 		 * interpreted by the guest:
2425 		 *
2426 		 * L2 cache control register 3: 64GB range, 256KB size,
2427 		 * enabled, latency 0x1, configured
2428 		 */
2429 		msr_info->data = 0xbe702111;
2430 		break;
2431 	case MSR_AMD64_OSVW_ID_LENGTH:
2432 		if (!guest_cpuid_has_osvw(vcpu))
2433 			return 1;
2434 		msr_info->data = vcpu->arch.osvw.length;
2435 		break;
2436 	case MSR_AMD64_OSVW_STATUS:
2437 		if (!guest_cpuid_has_osvw(vcpu))
2438 			return 1;
2439 		msr_info->data = vcpu->arch.osvw.status;
2440 		break;
2441 	default:
2442 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2443 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2444 		if (!ignore_msrs) {
2445 			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2446 			return 1;
2447 		} else {
2448 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2449 			msr_info->data = 0;
2450 		}
2451 		break;
2452 	}
2453 	return 0;
2454 }
2455 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2456 
2457 /*
2458  * Read or write a bunch of msrs. All parameters are kernel addresses.
2459  *
2460  * @return number of msrs set successfully.
2461  */
2462 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2463 		    struct kvm_msr_entry *entries,
2464 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2465 				  unsigned index, u64 *data))
2466 {
2467 	int i, idx;
2468 
2469 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2470 	for (i = 0; i < msrs->nmsrs; ++i)
2471 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2472 			break;
2473 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2474 
2475 	return i;
2476 }
2477 
2478 /*
2479  * Read or write a bunch of msrs. Parameters are user addresses.
2480  *
2481  * @return number of msrs set successfully.
2482  */
2483 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2484 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2485 				unsigned index, u64 *data),
2486 		  int writeback)
2487 {
2488 	struct kvm_msrs msrs;
2489 	struct kvm_msr_entry *entries;
2490 	int r, n;
2491 	unsigned size;
2492 
2493 	r = -EFAULT;
2494 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2495 		goto out;
2496 
2497 	r = -E2BIG;
2498 	if (msrs.nmsrs >= MAX_IO_MSRS)
2499 		goto out;
2500 
2501 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2502 	entries = memdup_user(user_msrs->entries, size);
2503 	if (IS_ERR(entries)) {
2504 		r = PTR_ERR(entries);
2505 		goto out;
2506 	}
2507 
2508 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2509 	if (r < 0)
2510 		goto out_free;
2511 
2512 	r = -EFAULT;
2513 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2514 		goto out_free;
2515 
2516 	r = n;
2517 
2518 out_free:
2519 	kfree(entries);
2520 out:
2521 	return r;
2522 }
2523 
2524 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2525 {
2526 	int r;
2527 
2528 	switch (ext) {
2529 	case KVM_CAP_IRQCHIP:
2530 	case KVM_CAP_HLT:
2531 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2532 	case KVM_CAP_SET_TSS_ADDR:
2533 	case KVM_CAP_EXT_CPUID:
2534 	case KVM_CAP_EXT_EMUL_CPUID:
2535 	case KVM_CAP_CLOCKSOURCE:
2536 	case KVM_CAP_PIT:
2537 	case KVM_CAP_NOP_IO_DELAY:
2538 	case KVM_CAP_MP_STATE:
2539 	case KVM_CAP_SYNC_MMU:
2540 	case KVM_CAP_USER_NMI:
2541 	case KVM_CAP_REINJECT_CONTROL:
2542 	case KVM_CAP_IRQ_INJECT_STATUS:
2543 	case KVM_CAP_IOEVENTFD:
2544 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2545 	case KVM_CAP_PIT2:
2546 	case KVM_CAP_PIT_STATE2:
2547 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2548 	case KVM_CAP_XEN_HVM:
2549 	case KVM_CAP_ADJUST_CLOCK:
2550 	case KVM_CAP_VCPU_EVENTS:
2551 	case KVM_CAP_HYPERV:
2552 	case KVM_CAP_HYPERV_VAPIC:
2553 	case KVM_CAP_HYPERV_SPIN:
2554 	case KVM_CAP_HYPERV_SYNIC:
2555 	case KVM_CAP_PCI_SEGMENT:
2556 	case KVM_CAP_DEBUGREGS:
2557 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2558 	case KVM_CAP_XSAVE:
2559 	case KVM_CAP_ASYNC_PF:
2560 	case KVM_CAP_GET_TSC_KHZ:
2561 	case KVM_CAP_KVMCLOCK_CTRL:
2562 	case KVM_CAP_READONLY_MEM:
2563 	case KVM_CAP_HYPERV_TIME:
2564 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2565 	case KVM_CAP_TSC_DEADLINE_TIMER:
2566 	case KVM_CAP_ENABLE_CAP_VM:
2567 	case KVM_CAP_DISABLE_QUIRKS:
2568 	case KVM_CAP_SET_BOOT_CPU_ID:
2569  	case KVM_CAP_SPLIT_IRQCHIP:
2570 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2571 	case KVM_CAP_ASSIGN_DEV_IRQ:
2572 	case KVM_CAP_PCI_2_3:
2573 #endif
2574 		r = 1;
2575 		break;
2576 	case KVM_CAP_X86_SMM:
2577 		/* SMBASE is usually relocated above 1M on modern chipsets,
2578 		 * and SMM handlers might indeed rely on 4G segment limits,
2579 		 * so do not report SMM to be available if real mode is
2580 		 * emulated via vm86 mode.  Still, do not go to great lengths
2581 		 * to avoid userspace's usage of the feature, because it is a
2582 		 * fringe case that is not enabled except via specific settings
2583 		 * of the module parameters.
2584 		 */
2585 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2586 		break;
2587 	case KVM_CAP_COALESCED_MMIO:
2588 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2589 		break;
2590 	case KVM_CAP_VAPIC:
2591 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2592 		break;
2593 	case KVM_CAP_NR_VCPUS:
2594 		r = KVM_SOFT_MAX_VCPUS;
2595 		break;
2596 	case KVM_CAP_MAX_VCPUS:
2597 		r = KVM_MAX_VCPUS;
2598 		break;
2599 	case KVM_CAP_NR_MEMSLOTS:
2600 		r = KVM_USER_MEM_SLOTS;
2601 		break;
2602 	case KVM_CAP_PV_MMU:	/* obsolete */
2603 		r = 0;
2604 		break;
2605 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2606 	case KVM_CAP_IOMMU:
2607 		r = iommu_present(&pci_bus_type);
2608 		break;
2609 #endif
2610 	case KVM_CAP_MCE:
2611 		r = KVM_MAX_MCE_BANKS;
2612 		break;
2613 	case KVM_CAP_XCRS:
2614 		r = cpu_has_xsave;
2615 		break;
2616 	case KVM_CAP_TSC_CONTROL:
2617 		r = kvm_has_tsc_control;
2618 		break;
2619 	default:
2620 		r = 0;
2621 		break;
2622 	}
2623 	return r;
2624 
2625 }
2626 
2627 long kvm_arch_dev_ioctl(struct file *filp,
2628 			unsigned int ioctl, unsigned long arg)
2629 {
2630 	void __user *argp = (void __user *)arg;
2631 	long r;
2632 
2633 	switch (ioctl) {
2634 	case KVM_GET_MSR_INDEX_LIST: {
2635 		struct kvm_msr_list __user *user_msr_list = argp;
2636 		struct kvm_msr_list msr_list;
2637 		unsigned n;
2638 
2639 		r = -EFAULT;
2640 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2641 			goto out;
2642 		n = msr_list.nmsrs;
2643 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2644 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2645 			goto out;
2646 		r = -E2BIG;
2647 		if (n < msr_list.nmsrs)
2648 			goto out;
2649 		r = -EFAULT;
2650 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2651 				 num_msrs_to_save * sizeof(u32)))
2652 			goto out;
2653 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2654 				 &emulated_msrs,
2655 				 num_emulated_msrs * sizeof(u32)))
2656 			goto out;
2657 		r = 0;
2658 		break;
2659 	}
2660 	case KVM_GET_SUPPORTED_CPUID:
2661 	case KVM_GET_EMULATED_CPUID: {
2662 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2663 		struct kvm_cpuid2 cpuid;
2664 
2665 		r = -EFAULT;
2666 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2667 			goto out;
2668 
2669 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2670 					    ioctl);
2671 		if (r)
2672 			goto out;
2673 
2674 		r = -EFAULT;
2675 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2676 			goto out;
2677 		r = 0;
2678 		break;
2679 	}
2680 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2681 		u64 mce_cap;
2682 
2683 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2684 		r = -EFAULT;
2685 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2686 			goto out;
2687 		r = 0;
2688 		break;
2689 	}
2690 	default:
2691 		r = -EINVAL;
2692 	}
2693 out:
2694 	return r;
2695 }
2696 
2697 static void wbinvd_ipi(void *garbage)
2698 {
2699 	wbinvd();
2700 }
2701 
2702 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2703 {
2704 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2705 }
2706 
2707 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2708 {
2709 	set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2710 }
2711 
2712 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2713 {
2714 	/* Address WBINVD may be executed by guest */
2715 	if (need_emulate_wbinvd(vcpu)) {
2716 		if (kvm_x86_ops->has_wbinvd_exit())
2717 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2718 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2719 			smp_call_function_single(vcpu->cpu,
2720 					wbinvd_ipi, NULL, 1);
2721 	}
2722 
2723 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2724 
2725 	/* Apply any externally detected TSC adjustments (due to suspend) */
2726 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2727 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2728 		vcpu->arch.tsc_offset_adjustment = 0;
2729 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2730 	}
2731 
2732 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2733 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2734 				rdtsc() - vcpu->arch.last_host_tsc;
2735 		if (tsc_delta < 0)
2736 			mark_tsc_unstable("KVM discovered backwards TSC");
2737 		if (check_tsc_unstable()) {
2738 			u64 offset = kvm_compute_tsc_offset(vcpu,
2739 						vcpu->arch.last_guest_tsc);
2740 			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2741 			vcpu->arch.tsc_catchup = 1;
2742 		}
2743 		/*
2744 		 * On a host with synchronized TSC, there is no need to update
2745 		 * kvmclock on vcpu->cpu migration
2746 		 */
2747 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2748 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2749 		if (vcpu->cpu != cpu)
2750 			kvm_migrate_timers(vcpu);
2751 		vcpu->cpu = cpu;
2752 	}
2753 
2754 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2755 	vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
2756 }
2757 
2758 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2759 {
2760 	kvm_x86_ops->vcpu_put(vcpu);
2761 	kvm_put_guest_fpu(vcpu);
2762 	vcpu->arch.last_host_tsc = rdtsc();
2763 }
2764 
2765 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2766 				    struct kvm_lapic_state *s)
2767 {
2768 	if (vcpu->arch.apicv_active)
2769 		kvm_x86_ops->sync_pir_to_irr(vcpu);
2770 
2771 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2772 
2773 	return 0;
2774 }
2775 
2776 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2777 				    struct kvm_lapic_state *s)
2778 {
2779 	kvm_apic_post_state_restore(vcpu, s);
2780 	update_cr8_intercept(vcpu);
2781 
2782 	return 0;
2783 }
2784 
2785 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2786 {
2787 	return (!lapic_in_kernel(vcpu) ||
2788 		kvm_apic_accept_pic_intr(vcpu));
2789 }
2790 
2791 /*
2792  * if userspace requested an interrupt window, check that the
2793  * interrupt window is open.
2794  *
2795  * No need to exit to userspace if we already have an interrupt queued.
2796  */
2797 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2798 {
2799 	return kvm_arch_interrupt_allowed(vcpu) &&
2800 		!kvm_cpu_has_interrupt(vcpu) &&
2801 		!kvm_event_needs_reinjection(vcpu) &&
2802 		kvm_cpu_accept_dm_intr(vcpu);
2803 }
2804 
2805 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2806 				    struct kvm_interrupt *irq)
2807 {
2808 	if (irq->irq >= KVM_NR_INTERRUPTS)
2809 		return -EINVAL;
2810 
2811 	if (!irqchip_in_kernel(vcpu->kvm)) {
2812 		kvm_queue_interrupt(vcpu, irq->irq, false);
2813 		kvm_make_request(KVM_REQ_EVENT, vcpu);
2814 		return 0;
2815 	}
2816 
2817 	/*
2818 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2819 	 * fail for in-kernel 8259.
2820 	 */
2821 	if (pic_in_kernel(vcpu->kvm))
2822 		return -ENXIO;
2823 
2824 	if (vcpu->arch.pending_external_vector != -1)
2825 		return -EEXIST;
2826 
2827 	vcpu->arch.pending_external_vector = irq->irq;
2828 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2829 	return 0;
2830 }
2831 
2832 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2833 {
2834 	kvm_inject_nmi(vcpu);
2835 
2836 	return 0;
2837 }
2838 
2839 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2840 {
2841 	kvm_make_request(KVM_REQ_SMI, vcpu);
2842 
2843 	return 0;
2844 }
2845 
2846 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2847 					   struct kvm_tpr_access_ctl *tac)
2848 {
2849 	if (tac->flags)
2850 		return -EINVAL;
2851 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2852 	return 0;
2853 }
2854 
2855 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2856 					u64 mcg_cap)
2857 {
2858 	int r;
2859 	unsigned bank_num = mcg_cap & 0xff, bank;
2860 
2861 	r = -EINVAL;
2862 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2863 		goto out;
2864 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2865 		goto out;
2866 	r = 0;
2867 	vcpu->arch.mcg_cap = mcg_cap;
2868 	/* Init IA32_MCG_CTL to all 1s */
2869 	if (mcg_cap & MCG_CTL_P)
2870 		vcpu->arch.mcg_ctl = ~(u64)0;
2871 	/* Init IA32_MCi_CTL to all 1s */
2872 	for (bank = 0; bank < bank_num; bank++)
2873 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2874 out:
2875 	return r;
2876 }
2877 
2878 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2879 				      struct kvm_x86_mce *mce)
2880 {
2881 	u64 mcg_cap = vcpu->arch.mcg_cap;
2882 	unsigned bank_num = mcg_cap & 0xff;
2883 	u64 *banks = vcpu->arch.mce_banks;
2884 
2885 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2886 		return -EINVAL;
2887 	/*
2888 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2889 	 * reporting is disabled
2890 	 */
2891 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2892 	    vcpu->arch.mcg_ctl != ~(u64)0)
2893 		return 0;
2894 	banks += 4 * mce->bank;
2895 	/*
2896 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2897 	 * reporting is disabled for the bank
2898 	 */
2899 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2900 		return 0;
2901 	if (mce->status & MCI_STATUS_UC) {
2902 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2903 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2904 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2905 			return 0;
2906 		}
2907 		if (banks[1] & MCI_STATUS_VAL)
2908 			mce->status |= MCI_STATUS_OVER;
2909 		banks[2] = mce->addr;
2910 		banks[3] = mce->misc;
2911 		vcpu->arch.mcg_status = mce->mcg_status;
2912 		banks[1] = mce->status;
2913 		kvm_queue_exception(vcpu, MC_VECTOR);
2914 	} else if (!(banks[1] & MCI_STATUS_VAL)
2915 		   || !(banks[1] & MCI_STATUS_UC)) {
2916 		if (banks[1] & MCI_STATUS_VAL)
2917 			mce->status |= MCI_STATUS_OVER;
2918 		banks[2] = mce->addr;
2919 		banks[3] = mce->misc;
2920 		banks[1] = mce->status;
2921 	} else
2922 		banks[1] |= MCI_STATUS_OVER;
2923 	return 0;
2924 }
2925 
2926 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2927 					       struct kvm_vcpu_events *events)
2928 {
2929 	process_nmi(vcpu);
2930 	events->exception.injected =
2931 		vcpu->arch.exception.pending &&
2932 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2933 	events->exception.nr = vcpu->arch.exception.nr;
2934 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2935 	events->exception.pad = 0;
2936 	events->exception.error_code = vcpu->arch.exception.error_code;
2937 
2938 	events->interrupt.injected =
2939 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2940 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2941 	events->interrupt.soft = 0;
2942 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2943 
2944 	events->nmi.injected = vcpu->arch.nmi_injected;
2945 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
2946 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2947 	events->nmi.pad = 0;
2948 
2949 	events->sipi_vector = 0; /* never valid when reporting to user space */
2950 
2951 	events->smi.smm = is_smm(vcpu);
2952 	events->smi.pending = vcpu->arch.smi_pending;
2953 	events->smi.smm_inside_nmi =
2954 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2955 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2956 
2957 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2958 			 | KVM_VCPUEVENT_VALID_SHADOW
2959 			 | KVM_VCPUEVENT_VALID_SMM);
2960 	memset(&events->reserved, 0, sizeof(events->reserved));
2961 }
2962 
2963 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2964 					      struct kvm_vcpu_events *events)
2965 {
2966 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2967 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2968 			      | KVM_VCPUEVENT_VALID_SHADOW
2969 			      | KVM_VCPUEVENT_VALID_SMM))
2970 		return -EINVAL;
2971 
2972 	process_nmi(vcpu);
2973 	vcpu->arch.exception.pending = events->exception.injected;
2974 	vcpu->arch.exception.nr = events->exception.nr;
2975 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2976 	vcpu->arch.exception.error_code = events->exception.error_code;
2977 
2978 	vcpu->arch.interrupt.pending = events->interrupt.injected;
2979 	vcpu->arch.interrupt.nr = events->interrupt.nr;
2980 	vcpu->arch.interrupt.soft = events->interrupt.soft;
2981 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2982 		kvm_x86_ops->set_interrupt_shadow(vcpu,
2983 						  events->interrupt.shadow);
2984 
2985 	vcpu->arch.nmi_injected = events->nmi.injected;
2986 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2987 		vcpu->arch.nmi_pending = events->nmi.pending;
2988 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2989 
2990 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2991 	    kvm_vcpu_has_lapic(vcpu))
2992 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
2993 
2994 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2995 		if (events->smi.smm)
2996 			vcpu->arch.hflags |= HF_SMM_MASK;
2997 		else
2998 			vcpu->arch.hflags &= ~HF_SMM_MASK;
2999 		vcpu->arch.smi_pending = events->smi.pending;
3000 		if (events->smi.smm_inside_nmi)
3001 			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3002 		else
3003 			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3004 		if (kvm_vcpu_has_lapic(vcpu)) {
3005 			if (events->smi.latched_init)
3006 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3007 			else
3008 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3009 		}
3010 	}
3011 
3012 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3013 
3014 	return 0;
3015 }
3016 
3017 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3018 					     struct kvm_debugregs *dbgregs)
3019 {
3020 	unsigned long val;
3021 
3022 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3023 	kvm_get_dr(vcpu, 6, &val);
3024 	dbgregs->dr6 = val;
3025 	dbgregs->dr7 = vcpu->arch.dr7;
3026 	dbgregs->flags = 0;
3027 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3028 }
3029 
3030 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3031 					    struct kvm_debugregs *dbgregs)
3032 {
3033 	if (dbgregs->flags)
3034 		return -EINVAL;
3035 
3036 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3037 	kvm_update_dr0123(vcpu);
3038 	vcpu->arch.dr6 = dbgregs->dr6;
3039 	kvm_update_dr6(vcpu);
3040 	vcpu->arch.dr7 = dbgregs->dr7;
3041 	kvm_update_dr7(vcpu);
3042 
3043 	return 0;
3044 }
3045 
3046 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3047 
3048 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3049 {
3050 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3051 	u64 xstate_bv = xsave->header.xfeatures;
3052 	u64 valid;
3053 
3054 	/*
3055 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3056 	 * leaves 0 and 1 in the loop below.
3057 	 */
3058 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3059 
3060 	/* Set XSTATE_BV */
3061 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3062 
3063 	/*
3064 	 * Copy each region from the possibly compacted offset to the
3065 	 * non-compacted offset.
3066 	 */
3067 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3068 	while (valid) {
3069 		u64 feature = valid & -valid;
3070 		int index = fls64(feature) - 1;
3071 		void *src = get_xsave_addr(xsave, feature);
3072 
3073 		if (src) {
3074 			u32 size, offset, ecx, edx;
3075 			cpuid_count(XSTATE_CPUID, index,
3076 				    &size, &offset, &ecx, &edx);
3077 			memcpy(dest + offset, src, size);
3078 		}
3079 
3080 		valid -= feature;
3081 	}
3082 }
3083 
3084 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3085 {
3086 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3087 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3088 	u64 valid;
3089 
3090 	/*
3091 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3092 	 * leaves 0 and 1 in the loop below.
3093 	 */
3094 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3095 
3096 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3097 	xsave->header.xfeatures = xstate_bv;
3098 	if (cpu_has_xsaves)
3099 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3100 
3101 	/*
3102 	 * Copy each region from the non-compacted offset to the
3103 	 * possibly compacted offset.
3104 	 */
3105 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3106 	while (valid) {
3107 		u64 feature = valid & -valid;
3108 		int index = fls64(feature) - 1;
3109 		void *dest = get_xsave_addr(xsave, feature);
3110 
3111 		if (dest) {
3112 			u32 size, offset, ecx, edx;
3113 			cpuid_count(XSTATE_CPUID, index,
3114 				    &size, &offset, &ecx, &edx);
3115 			memcpy(dest, src + offset, size);
3116 		}
3117 
3118 		valid -= feature;
3119 	}
3120 }
3121 
3122 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3123 					 struct kvm_xsave *guest_xsave)
3124 {
3125 	if (cpu_has_xsave) {
3126 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3127 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3128 	} else {
3129 		memcpy(guest_xsave->region,
3130 			&vcpu->arch.guest_fpu.state.fxsave,
3131 			sizeof(struct fxregs_state));
3132 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3133 			XFEATURE_MASK_FPSSE;
3134 	}
3135 }
3136 
3137 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3138 					struct kvm_xsave *guest_xsave)
3139 {
3140 	u64 xstate_bv =
3141 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3142 
3143 	if (cpu_has_xsave) {
3144 		/*
3145 		 * Here we allow setting states that are not present in
3146 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3147 		 * with old userspace.
3148 		 */
3149 		if (xstate_bv & ~kvm_supported_xcr0())
3150 			return -EINVAL;
3151 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3152 	} else {
3153 		if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3154 			return -EINVAL;
3155 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3156 			guest_xsave->region, sizeof(struct fxregs_state));
3157 	}
3158 	return 0;
3159 }
3160 
3161 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3162 					struct kvm_xcrs *guest_xcrs)
3163 {
3164 	if (!cpu_has_xsave) {
3165 		guest_xcrs->nr_xcrs = 0;
3166 		return;
3167 	}
3168 
3169 	guest_xcrs->nr_xcrs = 1;
3170 	guest_xcrs->flags = 0;
3171 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3172 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3173 }
3174 
3175 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3176 				       struct kvm_xcrs *guest_xcrs)
3177 {
3178 	int i, r = 0;
3179 
3180 	if (!cpu_has_xsave)
3181 		return -EINVAL;
3182 
3183 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3184 		return -EINVAL;
3185 
3186 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3187 		/* Only support XCR0 currently */
3188 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3189 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3190 				guest_xcrs->xcrs[i].value);
3191 			break;
3192 		}
3193 	if (r)
3194 		r = -EINVAL;
3195 	return r;
3196 }
3197 
3198 /*
3199  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3200  * stopped by the hypervisor.  This function will be called from the host only.
3201  * EINVAL is returned when the host attempts to set the flag for a guest that
3202  * does not support pv clocks.
3203  */
3204 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3205 {
3206 	if (!vcpu->arch.pv_time_enabled)
3207 		return -EINVAL;
3208 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3209 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3210 	return 0;
3211 }
3212 
3213 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3214 				     struct kvm_enable_cap *cap)
3215 {
3216 	if (cap->flags)
3217 		return -EINVAL;
3218 
3219 	switch (cap->cap) {
3220 	case KVM_CAP_HYPERV_SYNIC:
3221 		return kvm_hv_activate_synic(vcpu);
3222 	default:
3223 		return -EINVAL;
3224 	}
3225 }
3226 
3227 long kvm_arch_vcpu_ioctl(struct file *filp,
3228 			 unsigned int ioctl, unsigned long arg)
3229 {
3230 	struct kvm_vcpu *vcpu = filp->private_data;
3231 	void __user *argp = (void __user *)arg;
3232 	int r;
3233 	union {
3234 		struct kvm_lapic_state *lapic;
3235 		struct kvm_xsave *xsave;
3236 		struct kvm_xcrs *xcrs;
3237 		void *buffer;
3238 	} u;
3239 
3240 	u.buffer = NULL;
3241 	switch (ioctl) {
3242 	case KVM_GET_LAPIC: {
3243 		r = -EINVAL;
3244 		if (!vcpu->arch.apic)
3245 			goto out;
3246 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3247 
3248 		r = -ENOMEM;
3249 		if (!u.lapic)
3250 			goto out;
3251 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3252 		if (r)
3253 			goto out;
3254 		r = -EFAULT;
3255 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3256 			goto out;
3257 		r = 0;
3258 		break;
3259 	}
3260 	case KVM_SET_LAPIC: {
3261 		r = -EINVAL;
3262 		if (!vcpu->arch.apic)
3263 			goto out;
3264 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3265 		if (IS_ERR(u.lapic))
3266 			return PTR_ERR(u.lapic);
3267 
3268 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3269 		break;
3270 	}
3271 	case KVM_INTERRUPT: {
3272 		struct kvm_interrupt irq;
3273 
3274 		r = -EFAULT;
3275 		if (copy_from_user(&irq, argp, sizeof irq))
3276 			goto out;
3277 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3278 		break;
3279 	}
3280 	case KVM_NMI: {
3281 		r = kvm_vcpu_ioctl_nmi(vcpu);
3282 		break;
3283 	}
3284 	case KVM_SMI: {
3285 		r = kvm_vcpu_ioctl_smi(vcpu);
3286 		break;
3287 	}
3288 	case KVM_SET_CPUID: {
3289 		struct kvm_cpuid __user *cpuid_arg = argp;
3290 		struct kvm_cpuid cpuid;
3291 
3292 		r = -EFAULT;
3293 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3294 			goto out;
3295 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3296 		break;
3297 	}
3298 	case KVM_SET_CPUID2: {
3299 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3300 		struct kvm_cpuid2 cpuid;
3301 
3302 		r = -EFAULT;
3303 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3304 			goto out;
3305 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3306 					      cpuid_arg->entries);
3307 		break;
3308 	}
3309 	case KVM_GET_CPUID2: {
3310 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3311 		struct kvm_cpuid2 cpuid;
3312 
3313 		r = -EFAULT;
3314 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3315 			goto out;
3316 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3317 					      cpuid_arg->entries);
3318 		if (r)
3319 			goto out;
3320 		r = -EFAULT;
3321 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3322 			goto out;
3323 		r = 0;
3324 		break;
3325 	}
3326 	case KVM_GET_MSRS:
3327 		r = msr_io(vcpu, argp, do_get_msr, 1);
3328 		break;
3329 	case KVM_SET_MSRS:
3330 		r = msr_io(vcpu, argp, do_set_msr, 0);
3331 		break;
3332 	case KVM_TPR_ACCESS_REPORTING: {
3333 		struct kvm_tpr_access_ctl tac;
3334 
3335 		r = -EFAULT;
3336 		if (copy_from_user(&tac, argp, sizeof tac))
3337 			goto out;
3338 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3339 		if (r)
3340 			goto out;
3341 		r = -EFAULT;
3342 		if (copy_to_user(argp, &tac, sizeof tac))
3343 			goto out;
3344 		r = 0;
3345 		break;
3346 	};
3347 	case KVM_SET_VAPIC_ADDR: {
3348 		struct kvm_vapic_addr va;
3349 
3350 		r = -EINVAL;
3351 		if (!lapic_in_kernel(vcpu))
3352 			goto out;
3353 		r = -EFAULT;
3354 		if (copy_from_user(&va, argp, sizeof va))
3355 			goto out;
3356 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3357 		break;
3358 	}
3359 	case KVM_X86_SETUP_MCE: {
3360 		u64 mcg_cap;
3361 
3362 		r = -EFAULT;
3363 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3364 			goto out;
3365 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3366 		break;
3367 	}
3368 	case KVM_X86_SET_MCE: {
3369 		struct kvm_x86_mce mce;
3370 
3371 		r = -EFAULT;
3372 		if (copy_from_user(&mce, argp, sizeof mce))
3373 			goto out;
3374 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3375 		break;
3376 	}
3377 	case KVM_GET_VCPU_EVENTS: {
3378 		struct kvm_vcpu_events events;
3379 
3380 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3381 
3382 		r = -EFAULT;
3383 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3384 			break;
3385 		r = 0;
3386 		break;
3387 	}
3388 	case KVM_SET_VCPU_EVENTS: {
3389 		struct kvm_vcpu_events events;
3390 
3391 		r = -EFAULT;
3392 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3393 			break;
3394 
3395 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3396 		break;
3397 	}
3398 	case KVM_GET_DEBUGREGS: {
3399 		struct kvm_debugregs dbgregs;
3400 
3401 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3402 
3403 		r = -EFAULT;
3404 		if (copy_to_user(argp, &dbgregs,
3405 				 sizeof(struct kvm_debugregs)))
3406 			break;
3407 		r = 0;
3408 		break;
3409 	}
3410 	case KVM_SET_DEBUGREGS: {
3411 		struct kvm_debugregs dbgregs;
3412 
3413 		r = -EFAULT;
3414 		if (copy_from_user(&dbgregs, argp,
3415 				   sizeof(struct kvm_debugregs)))
3416 			break;
3417 
3418 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3419 		break;
3420 	}
3421 	case KVM_GET_XSAVE: {
3422 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3423 		r = -ENOMEM;
3424 		if (!u.xsave)
3425 			break;
3426 
3427 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3428 
3429 		r = -EFAULT;
3430 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3431 			break;
3432 		r = 0;
3433 		break;
3434 	}
3435 	case KVM_SET_XSAVE: {
3436 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3437 		if (IS_ERR(u.xsave))
3438 			return PTR_ERR(u.xsave);
3439 
3440 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3441 		break;
3442 	}
3443 	case KVM_GET_XCRS: {
3444 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3445 		r = -ENOMEM;
3446 		if (!u.xcrs)
3447 			break;
3448 
3449 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3450 
3451 		r = -EFAULT;
3452 		if (copy_to_user(argp, u.xcrs,
3453 				 sizeof(struct kvm_xcrs)))
3454 			break;
3455 		r = 0;
3456 		break;
3457 	}
3458 	case KVM_SET_XCRS: {
3459 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3460 		if (IS_ERR(u.xcrs))
3461 			return PTR_ERR(u.xcrs);
3462 
3463 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3464 		break;
3465 	}
3466 	case KVM_SET_TSC_KHZ: {
3467 		u32 user_tsc_khz;
3468 
3469 		r = -EINVAL;
3470 		user_tsc_khz = (u32)arg;
3471 
3472 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3473 			goto out;
3474 
3475 		if (user_tsc_khz == 0)
3476 			user_tsc_khz = tsc_khz;
3477 
3478 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3479 			r = 0;
3480 
3481 		goto out;
3482 	}
3483 	case KVM_GET_TSC_KHZ: {
3484 		r = vcpu->arch.virtual_tsc_khz;
3485 		goto out;
3486 	}
3487 	case KVM_KVMCLOCK_CTRL: {
3488 		r = kvm_set_guest_paused(vcpu);
3489 		goto out;
3490 	}
3491 	case KVM_ENABLE_CAP: {
3492 		struct kvm_enable_cap cap;
3493 
3494 		r = -EFAULT;
3495 		if (copy_from_user(&cap, argp, sizeof(cap)))
3496 			goto out;
3497 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3498 		break;
3499 	}
3500 	default:
3501 		r = -EINVAL;
3502 	}
3503 out:
3504 	kfree(u.buffer);
3505 	return r;
3506 }
3507 
3508 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3509 {
3510 	return VM_FAULT_SIGBUS;
3511 }
3512 
3513 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3514 {
3515 	int ret;
3516 
3517 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3518 		return -EINVAL;
3519 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3520 	return ret;
3521 }
3522 
3523 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3524 					      u64 ident_addr)
3525 {
3526 	kvm->arch.ept_identity_map_addr = ident_addr;
3527 	return 0;
3528 }
3529 
3530 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3531 					  u32 kvm_nr_mmu_pages)
3532 {
3533 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3534 		return -EINVAL;
3535 
3536 	mutex_lock(&kvm->slots_lock);
3537 
3538 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3539 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3540 
3541 	mutex_unlock(&kvm->slots_lock);
3542 	return 0;
3543 }
3544 
3545 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3546 {
3547 	return kvm->arch.n_max_mmu_pages;
3548 }
3549 
3550 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3551 {
3552 	int r;
3553 
3554 	r = 0;
3555 	switch (chip->chip_id) {
3556 	case KVM_IRQCHIP_PIC_MASTER:
3557 		memcpy(&chip->chip.pic,
3558 			&pic_irqchip(kvm)->pics[0],
3559 			sizeof(struct kvm_pic_state));
3560 		break;
3561 	case KVM_IRQCHIP_PIC_SLAVE:
3562 		memcpy(&chip->chip.pic,
3563 			&pic_irqchip(kvm)->pics[1],
3564 			sizeof(struct kvm_pic_state));
3565 		break;
3566 	case KVM_IRQCHIP_IOAPIC:
3567 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3568 		break;
3569 	default:
3570 		r = -EINVAL;
3571 		break;
3572 	}
3573 	return r;
3574 }
3575 
3576 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3577 {
3578 	int r;
3579 
3580 	r = 0;
3581 	switch (chip->chip_id) {
3582 	case KVM_IRQCHIP_PIC_MASTER:
3583 		spin_lock(&pic_irqchip(kvm)->lock);
3584 		memcpy(&pic_irqchip(kvm)->pics[0],
3585 			&chip->chip.pic,
3586 			sizeof(struct kvm_pic_state));
3587 		spin_unlock(&pic_irqchip(kvm)->lock);
3588 		break;
3589 	case KVM_IRQCHIP_PIC_SLAVE:
3590 		spin_lock(&pic_irqchip(kvm)->lock);
3591 		memcpy(&pic_irqchip(kvm)->pics[1],
3592 			&chip->chip.pic,
3593 			sizeof(struct kvm_pic_state));
3594 		spin_unlock(&pic_irqchip(kvm)->lock);
3595 		break;
3596 	case KVM_IRQCHIP_IOAPIC:
3597 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3598 		break;
3599 	default:
3600 		r = -EINVAL;
3601 		break;
3602 	}
3603 	kvm_pic_update_irq(pic_irqchip(kvm));
3604 	return r;
3605 }
3606 
3607 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3608 {
3609 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3610 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3611 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3612 	return 0;
3613 }
3614 
3615 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3616 {
3617 	int i;
3618 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3619 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3620 	for (i = 0; i < 3; i++)
3621 		kvm_pit_load_count(kvm, i, ps->channels[i].count, 0);
3622 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3623 	return 0;
3624 }
3625 
3626 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3627 {
3628 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3629 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3630 		sizeof(ps->channels));
3631 	ps->flags = kvm->arch.vpit->pit_state.flags;
3632 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3633 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3634 	return 0;
3635 }
3636 
3637 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3638 {
3639 	int start = 0;
3640 	int i;
3641 	u32 prev_legacy, cur_legacy;
3642 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3643 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3644 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3645 	if (!prev_legacy && cur_legacy)
3646 		start = 1;
3647 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3648 	       sizeof(kvm->arch.vpit->pit_state.channels));
3649 	kvm->arch.vpit->pit_state.flags = ps->flags;
3650 	for (i = 0; i < 3; i++)
3651 		kvm_pit_load_count(kvm, i, kvm->arch.vpit->pit_state.channels[i].count,
3652 				   start && i == 0);
3653 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3654 	return 0;
3655 }
3656 
3657 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3658 				 struct kvm_reinject_control *control)
3659 {
3660 	if (!kvm->arch.vpit)
3661 		return -ENXIO;
3662 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3663 	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3664 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3665 	return 0;
3666 }
3667 
3668 /**
3669  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3670  * @kvm: kvm instance
3671  * @log: slot id and address to which we copy the log
3672  *
3673  * Steps 1-4 below provide general overview of dirty page logging. See
3674  * kvm_get_dirty_log_protect() function description for additional details.
3675  *
3676  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3677  * always flush the TLB (step 4) even if previous step failed  and the dirty
3678  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3679  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3680  * writes will be marked dirty for next log read.
3681  *
3682  *   1. Take a snapshot of the bit and clear it if needed.
3683  *   2. Write protect the corresponding page.
3684  *   3. Copy the snapshot to the userspace.
3685  *   4. Flush TLB's if needed.
3686  */
3687 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3688 {
3689 	bool is_dirty = false;
3690 	int r;
3691 
3692 	mutex_lock(&kvm->slots_lock);
3693 
3694 	/*
3695 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3696 	 */
3697 	if (kvm_x86_ops->flush_log_dirty)
3698 		kvm_x86_ops->flush_log_dirty(kvm);
3699 
3700 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3701 
3702 	/*
3703 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3704 	 * kvm_mmu_slot_remove_write_access().
3705 	 */
3706 	lockdep_assert_held(&kvm->slots_lock);
3707 	if (is_dirty)
3708 		kvm_flush_remote_tlbs(kvm);
3709 
3710 	mutex_unlock(&kvm->slots_lock);
3711 	return r;
3712 }
3713 
3714 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3715 			bool line_status)
3716 {
3717 	if (!irqchip_in_kernel(kvm))
3718 		return -ENXIO;
3719 
3720 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3721 					irq_event->irq, irq_event->level,
3722 					line_status);
3723 	return 0;
3724 }
3725 
3726 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3727 				   struct kvm_enable_cap *cap)
3728 {
3729 	int r;
3730 
3731 	if (cap->flags)
3732 		return -EINVAL;
3733 
3734 	switch (cap->cap) {
3735 	case KVM_CAP_DISABLE_QUIRKS:
3736 		kvm->arch.disabled_quirks = cap->args[0];
3737 		r = 0;
3738 		break;
3739 	case KVM_CAP_SPLIT_IRQCHIP: {
3740 		mutex_lock(&kvm->lock);
3741 		r = -EINVAL;
3742 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3743 			goto split_irqchip_unlock;
3744 		r = -EEXIST;
3745 		if (irqchip_in_kernel(kvm))
3746 			goto split_irqchip_unlock;
3747 		if (atomic_read(&kvm->online_vcpus))
3748 			goto split_irqchip_unlock;
3749 		r = kvm_setup_empty_irq_routing(kvm);
3750 		if (r)
3751 			goto split_irqchip_unlock;
3752 		/* Pairs with irqchip_in_kernel. */
3753 		smp_wmb();
3754 		kvm->arch.irqchip_split = true;
3755 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3756 		r = 0;
3757 split_irqchip_unlock:
3758 		mutex_unlock(&kvm->lock);
3759 		break;
3760 	}
3761 	default:
3762 		r = -EINVAL;
3763 		break;
3764 	}
3765 	return r;
3766 }
3767 
3768 long kvm_arch_vm_ioctl(struct file *filp,
3769 		       unsigned int ioctl, unsigned long arg)
3770 {
3771 	struct kvm *kvm = filp->private_data;
3772 	void __user *argp = (void __user *)arg;
3773 	int r = -ENOTTY;
3774 	/*
3775 	 * This union makes it completely explicit to gcc-3.x
3776 	 * that these two variables' stack usage should be
3777 	 * combined, not added together.
3778 	 */
3779 	union {
3780 		struct kvm_pit_state ps;
3781 		struct kvm_pit_state2 ps2;
3782 		struct kvm_pit_config pit_config;
3783 	} u;
3784 
3785 	switch (ioctl) {
3786 	case KVM_SET_TSS_ADDR:
3787 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3788 		break;
3789 	case KVM_SET_IDENTITY_MAP_ADDR: {
3790 		u64 ident_addr;
3791 
3792 		r = -EFAULT;
3793 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3794 			goto out;
3795 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3796 		break;
3797 	}
3798 	case KVM_SET_NR_MMU_PAGES:
3799 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3800 		break;
3801 	case KVM_GET_NR_MMU_PAGES:
3802 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3803 		break;
3804 	case KVM_CREATE_IRQCHIP: {
3805 		struct kvm_pic *vpic;
3806 
3807 		mutex_lock(&kvm->lock);
3808 		r = -EEXIST;
3809 		if (kvm->arch.vpic)
3810 			goto create_irqchip_unlock;
3811 		r = -EINVAL;
3812 		if (atomic_read(&kvm->online_vcpus))
3813 			goto create_irqchip_unlock;
3814 		r = -ENOMEM;
3815 		vpic = kvm_create_pic(kvm);
3816 		if (vpic) {
3817 			r = kvm_ioapic_init(kvm);
3818 			if (r) {
3819 				mutex_lock(&kvm->slots_lock);
3820 				kvm_destroy_pic(vpic);
3821 				mutex_unlock(&kvm->slots_lock);
3822 				goto create_irqchip_unlock;
3823 			}
3824 		} else
3825 			goto create_irqchip_unlock;
3826 		r = kvm_setup_default_irq_routing(kvm);
3827 		if (r) {
3828 			mutex_lock(&kvm->slots_lock);
3829 			mutex_lock(&kvm->irq_lock);
3830 			kvm_ioapic_destroy(kvm);
3831 			kvm_destroy_pic(vpic);
3832 			mutex_unlock(&kvm->irq_lock);
3833 			mutex_unlock(&kvm->slots_lock);
3834 			goto create_irqchip_unlock;
3835 		}
3836 		/* Write kvm->irq_routing before kvm->arch.vpic.  */
3837 		smp_wmb();
3838 		kvm->arch.vpic = vpic;
3839 	create_irqchip_unlock:
3840 		mutex_unlock(&kvm->lock);
3841 		break;
3842 	}
3843 	case KVM_CREATE_PIT:
3844 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3845 		goto create_pit;
3846 	case KVM_CREATE_PIT2:
3847 		r = -EFAULT;
3848 		if (copy_from_user(&u.pit_config, argp,
3849 				   sizeof(struct kvm_pit_config)))
3850 			goto out;
3851 	create_pit:
3852 		mutex_lock(&kvm->slots_lock);
3853 		r = -EEXIST;
3854 		if (kvm->arch.vpit)
3855 			goto create_pit_unlock;
3856 		r = -ENOMEM;
3857 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3858 		if (kvm->arch.vpit)
3859 			r = 0;
3860 	create_pit_unlock:
3861 		mutex_unlock(&kvm->slots_lock);
3862 		break;
3863 	case KVM_GET_IRQCHIP: {
3864 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3865 		struct kvm_irqchip *chip;
3866 
3867 		chip = memdup_user(argp, sizeof(*chip));
3868 		if (IS_ERR(chip)) {
3869 			r = PTR_ERR(chip);
3870 			goto out;
3871 		}
3872 
3873 		r = -ENXIO;
3874 		if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3875 			goto get_irqchip_out;
3876 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3877 		if (r)
3878 			goto get_irqchip_out;
3879 		r = -EFAULT;
3880 		if (copy_to_user(argp, chip, sizeof *chip))
3881 			goto get_irqchip_out;
3882 		r = 0;
3883 	get_irqchip_out:
3884 		kfree(chip);
3885 		break;
3886 	}
3887 	case KVM_SET_IRQCHIP: {
3888 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3889 		struct kvm_irqchip *chip;
3890 
3891 		chip = memdup_user(argp, sizeof(*chip));
3892 		if (IS_ERR(chip)) {
3893 			r = PTR_ERR(chip);
3894 			goto out;
3895 		}
3896 
3897 		r = -ENXIO;
3898 		if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3899 			goto set_irqchip_out;
3900 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3901 		if (r)
3902 			goto set_irqchip_out;
3903 		r = 0;
3904 	set_irqchip_out:
3905 		kfree(chip);
3906 		break;
3907 	}
3908 	case KVM_GET_PIT: {
3909 		r = -EFAULT;
3910 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3911 			goto out;
3912 		r = -ENXIO;
3913 		if (!kvm->arch.vpit)
3914 			goto out;
3915 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3916 		if (r)
3917 			goto out;
3918 		r = -EFAULT;
3919 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3920 			goto out;
3921 		r = 0;
3922 		break;
3923 	}
3924 	case KVM_SET_PIT: {
3925 		r = -EFAULT;
3926 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3927 			goto out;
3928 		r = -ENXIO;
3929 		if (!kvm->arch.vpit)
3930 			goto out;
3931 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3932 		break;
3933 	}
3934 	case KVM_GET_PIT2: {
3935 		r = -ENXIO;
3936 		if (!kvm->arch.vpit)
3937 			goto out;
3938 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3939 		if (r)
3940 			goto out;
3941 		r = -EFAULT;
3942 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3943 			goto out;
3944 		r = 0;
3945 		break;
3946 	}
3947 	case KVM_SET_PIT2: {
3948 		r = -EFAULT;
3949 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3950 			goto out;
3951 		r = -ENXIO;
3952 		if (!kvm->arch.vpit)
3953 			goto out;
3954 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3955 		break;
3956 	}
3957 	case KVM_REINJECT_CONTROL: {
3958 		struct kvm_reinject_control control;
3959 		r =  -EFAULT;
3960 		if (copy_from_user(&control, argp, sizeof(control)))
3961 			goto out;
3962 		r = kvm_vm_ioctl_reinject(kvm, &control);
3963 		break;
3964 	}
3965 	case KVM_SET_BOOT_CPU_ID:
3966 		r = 0;
3967 		mutex_lock(&kvm->lock);
3968 		if (atomic_read(&kvm->online_vcpus) != 0)
3969 			r = -EBUSY;
3970 		else
3971 			kvm->arch.bsp_vcpu_id = arg;
3972 		mutex_unlock(&kvm->lock);
3973 		break;
3974 	case KVM_XEN_HVM_CONFIG: {
3975 		r = -EFAULT;
3976 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3977 				   sizeof(struct kvm_xen_hvm_config)))
3978 			goto out;
3979 		r = -EINVAL;
3980 		if (kvm->arch.xen_hvm_config.flags)
3981 			goto out;
3982 		r = 0;
3983 		break;
3984 	}
3985 	case KVM_SET_CLOCK: {
3986 		struct kvm_clock_data user_ns;
3987 		u64 now_ns;
3988 		s64 delta;
3989 
3990 		r = -EFAULT;
3991 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3992 			goto out;
3993 
3994 		r = -EINVAL;
3995 		if (user_ns.flags)
3996 			goto out;
3997 
3998 		r = 0;
3999 		local_irq_disable();
4000 		now_ns = get_kernel_ns();
4001 		delta = user_ns.clock - now_ns;
4002 		local_irq_enable();
4003 		kvm->arch.kvmclock_offset = delta;
4004 		kvm_gen_update_masterclock(kvm);
4005 		break;
4006 	}
4007 	case KVM_GET_CLOCK: {
4008 		struct kvm_clock_data user_ns;
4009 		u64 now_ns;
4010 
4011 		local_irq_disable();
4012 		now_ns = get_kernel_ns();
4013 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4014 		local_irq_enable();
4015 		user_ns.flags = 0;
4016 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4017 
4018 		r = -EFAULT;
4019 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4020 			goto out;
4021 		r = 0;
4022 		break;
4023 	}
4024 	case KVM_ENABLE_CAP: {
4025 		struct kvm_enable_cap cap;
4026 
4027 		r = -EFAULT;
4028 		if (copy_from_user(&cap, argp, sizeof(cap)))
4029 			goto out;
4030 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4031 		break;
4032 	}
4033 	default:
4034 		r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4035 	}
4036 out:
4037 	return r;
4038 }
4039 
4040 static void kvm_init_msr_list(void)
4041 {
4042 	u32 dummy[2];
4043 	unsigned i, j;
4044 
4045 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4046 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4047 			continue;
4048 
4049 		/*
4050 		 * Even MSRs that are valid in the host may not be exposed
4051 		 * to the guests in some cases.
4052 		 */
4053 		switch (msrs_to_save[i]) {
4054 		case MSR_IA32_BNDCFGS:
4055 			if (!kvm_x86_ops->mpx_supported())
4056 				continue;
4057 			break;
4058 		case MSR_TSC_AUX:
4059 			if (!kvm_x86_ops->rdtscp_supported())
4060 				continue;
4061 			break;
4062 		default:
4063 			break;
4064 		}
4065 
4066 		if (j < i)
4067 			msrs_to_save[j] = msrs_to_save[i];
4068 		j++;
4069 	}
4070 	num_msrs_to_save = j;
4071 
4072 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4073 		switch (emulated_msrs[i]) {
4074 		case MSR_IA32_SMBASE:
4075 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4076 				continue;
4077 			break;
4078 		default:
4079 			break;
4080 		}
4081 
4082 		if (j < i)
4083 			emulated_msrs[j] = emulated_msrs[i];
4084 		j++;
4085 	}
4086 	num_emulated_msrs = j;
4087 }
4088 
4089 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4090 			   const void *v)
4091 {
4092 	int handled = 0;
4093 	int n;
4094 
4095 	do {
4096 		n = min(len, 8);
4097 		if (!(vcpu->arch.apic &&
4098 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4099 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4100 			break;
4101 		handled += n;
4102 		addr += n;
4103 		len -= n;
4104 		v += n;
4105 	} while (len);
4106 
4107 	return handled;
4108 }
4109 
4110 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4111 {
4112 	int handled = 0;
4113 	int n;
4114 
4115 	do {
4116 		n = min(len, 8);
4117 		if (!(vcpu->arch.apic &&
4118 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4119 					 addr, n, v))
4120 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4121 			break;
4122 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4123 		handled += n;
4124 		addr += n;
4125 		len -= n;
4126 		v += n;
4127 	} while (len);
4128 
4129 	return handled;
4130 }
4131 
4132 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4133 			struct kvm_segment *var, int seg)
4134 {
4135 	kvm_x86_ops->set_segment(vcpu, var, seg);
4136 }
4137 
4138 void kvm_get_segment(struct kvm_vcpu *vcpu,
4139 		     struct kvm_segment *var, int seg)
4140 {
4141 	kvm_x86_ops->get_segment(vcpu, var, seg);
4142 }
4143 
4144 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4145 			   struct x86_exception *exception)
4146 {
4147 	gpa_t t_gpa;
4148 
4149 	BUG_ON(!mmu_is_nested(vcpu));
4150 
4151 	/* NPT walks are always user-walks */
4152 	access |= PFERR_USER_MASK;
4153 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4154 
4155 	return t_gpa;
4156 }
4157 
4158 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4159 			      struct x86_exception *exception)
4160 {
4161 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4162 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4163 }
4164 
4165  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4166 				struct x86_exception *exception)
4167 {
4168 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4169 	access |= PFERR_FETCH_MASK;
4170 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4171 }
4172 
4173 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4174 			       struct x86_exception *exception)
4175 {
4176 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4177 	access |= PFERR_WRITE_MASK;
4178 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4179 }
4180 
4181 /* uses this to access any guest's mapped memory without checking CPL */
4182 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4183 				struct x86_exception *exception)
4184 {
4185 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4186 }
4187 
4188 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4189 				      struct kvm_vcpu *vcpu, u32 access,
4190 				      struct x86_exception *exception)
4191 {
4192 	void *data = val;
4193 	int r = X86EMUL_CONTINUE;
4194 
4195 	while (bytes) {
4196 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4197 							    exception);
4198 		unsigned offset = addr & (PAGE_SIZE-1);
4199 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4200 		int ret;
4201 
4202 		if (gpa == UNMAPPED_GVA)
4203 			return X86EMUL_PROPAGATE_FAULT;
4204 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4205 					       offset, toread);
4206 		if (ret < 0) {
4207 			r = X86EMUL_IO_NEEDED;
4208 			goto out;
4209 		}
4210 
4211 		bytes -= toread;
4212 		data += toread;
4213 		addr += toread;
4214 	}
4215 out:
4216 	return r;
4217 }
4218 
4219 /* used for instruction fetching */
4220 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4221 				gva_t addr, void *val, unsigned int bytes,
4222 				struct x86_exception *exception)
4223 {
4224 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4225 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4226 	unsigned offset;
4227 	int ret;
4228 
4229 	/* Inline kvm_read_guest_virt_helper for speed.  */
4230 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4231 						    exception);
4232 	if (unlikely(gpa == UNMAPPED_GVA))
4233 		return X86EMUL_PROPAGATE_FAULT;
4234 
4235 	offset = addr & (PAGE_SIZE-1);
4236 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4237 		bytes = (unsigned)PAGE_SIZE - offset;
4238 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4239 				       offset, bytes);
4240 	if (unlikely(ret < 0))
4241 		return X86EMUL_IO_NEEDED;
4242 
4243 	return X86EMUL_CONTINUE;
4244 }
4245 
4246 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4247 			       gva_t addr, void *val, unsigned int bytes,
4248 			       struct x86_exception *exception)
4249 {
4250 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4251 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4252 
4253 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4254 					  exception);
4255 }
4256 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4257 
4258 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4259 				      gva_t addr, void *val, unsigned int bytes,
4260 				      struct x86_exception *exception)
4261 {
4262 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4263 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4264 }
4265 
4266 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4267 		unsigned long addr, void *val, unsigned int bytes)
4268 {
4269 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4270 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4271 
4272 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4273 }
4274 
4275 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4276 				       gva_t addr, void *val,
4277 				       unsigned int bytes,
4278 				       struct x86_exception *exception)
4279 {
4280 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4281 	void *data = val;
4282 	int r = X86EMUL_CONTINUE;
4283 
4284 	while (bytes) {
4285 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4286 							     PFERR_WRITE_MASK,
4287 							     exception);
4288 		unsigned offset = addr & (PAGE_SIZE-1);
4289 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4290 		int ret;
4291 
4292 		if (gpa == UNMAPPED_GVA)
4293 			return X86EMUL_PROPAGATE_FAULT;
4294 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4295 		if (ret < 0) {
4296 			r = X86EMUL_IO_NEEDED;
4297 			goto out;
4298 		}
4299 
4300 		bytes -= towrite;
4301 		data += towrite;
4302 		addr += towrite;
4303 	}
4304 out:
4305 	return r;
4306 }
4307 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4308 
4309 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4310 				gpa_t *gpa, struct x86_exception *exception,
4311 				bool write)
4312 {
4313 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4314 		| (write ? PFERR_WRITE_MASK : 0);
4315 
4316 	if (vcpu_match_mmio_gva(vcpu, gva)
4317 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4318 				 vcpu->arch.access, access)) {
4319 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4320 					(gva & (PAGE_SIZE - 1));
4321 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4322 		return 1;
4323 	}
4324 
4325 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4326 
4327 	if (*gpa == UNMAPPED_GVA)
4328 		return -1;
4329 
4330 	/* For APIC access vmexit */
4331 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4332 		return 1;
4333 
4334 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4335 		trace_vcpu_match_mmio(gva, *gpa, write, true);
4336 		return 1;
4337 	}
4338 
4339 	return 0;
4340 }
4341 
4342 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4343 			const void *val, int bytes)
4344 {
4345 	int ret;
4346 
4347 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4348 	if (ret < 0)
4349 		return 0;
4350 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4351 	return 1;
4352 }
4353 
4354 struct read_write_emulator_ops {
4355 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4356 				  int bytes);
4357 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4358 				  void *val, int bytes);
4359 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4360 			       int bytes, void *val);
4361 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4362 				    void *val, int bytes);
4363 	bool write;
4364 };
4365 
4366 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4367 {
4368 	if (vcpu->mmio_read_completed) {
4369 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4370 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4371 		vcpu->mmio_read_completed = 0;
4372 		return 1;
4373 	}
4374 
4375 	return 0;
4376 }
4377 
4378 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4379 			void *val, int bytes)
4380 {
4381 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4382 }
4383 
4384 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4385 			 void *val, int bytes)
4386 {
4387 	return emulator_write_phys(vcpu, gpa, val, bytes);
4388 }
4389 
4390 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4391 {
4392 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4393 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4394 }
4395 
4396 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4397 			  void *val, int bytes)
4398 {
4399 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4400 	return X86EMUL_IO_NEEDED;
4401 }
4402 
4403 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4404 			   void *val, int bytes)
4405 {
4406 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4407 
4408 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4409 	return X86EMUL_CONTINUE;
4410 }
4411 
4412 static const struct read_write_emulator_ops read_emultor = {
4413 	.read_write_prepare = read_prepare,
4414 	.read_write_emulate = read_emulate,
4415 	.read_write_mmio = vcpu_mmio_read,
4416 	.read_write_exit_mmio = read_exit_mmio,
4417 };
4418 
4419 static const struct read_write_emulator_ops write_emultor = {
4420 	.read_write_emulate = write_emulate,
4421 	.read_write_mmio = write_mmio,
4422 	.read_write_exit_mmio = write_exit_mmio,
4423 	.write = true,
4424 };
4425 
4426 static int emulator_read_write_onepage(unsigned long addr, void *val,
4427 				       unsigned int bytes,
4428 				       struct x86_exception *exception,
4429 				       struct kvm_vcpu *vcpu,
4430 				       const struct read_write_emulator_ops *ops)
4431 {
4432 	gpa_t gpa;
4433 	int handled, ret;
4434 	bool write = ops->write;
4435 	struct kvm_mmio_fragment *frag;
4436 
4437 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4438 
4439 	if (ret < 0)
4440 		return X86EMUL_PROPAGATE_FAULT;
4441 
4442 	/* For APIC access vmexit */
4443 	if (ret)
4444 		goto mmio;
4445 
4446 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4447 		return X86EMUL_CONTINUE;
4448 
4449 mmio:
4450 	/*
4451 	 * Is this MMIO handled locally?
4452 	 */
4453 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4454 	if (handled == bytes)
4455 		return X86EMUL_CONTINUE;
4456 
4457 	gpa += handled;
4458 	bytes -= handled;
4459 	val += handled;
4460 
4461 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4462 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4463 	frag->gpa = gpa;
4464 	frag->data = val;
4465 	frag->len = bytes;
4466 	return X86EMUL_CONTINUE;
4467 }
4468 
4469 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4470 			unsigned long addr,
4471 			void *val, unsigned int bytes,
4472 			struct x86_exception *exception,
4473 			const struct read_write_emulator_ops *ops)
4474 {
4475 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4476 	gpa_t gpa;
4477 	int rc;
4478 
4479 	if (ops->read_write_prepare &&
4480 		  ops->read_write_prepare(vcpu, val, bytes))
4481 		return X86EMUL_CONTINUE;
4482 
4483 	vcpu->mmio_nr_fragments = 0;
4484 
4485 	/* Crossing a page boundary? */
4486 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4487 		int now;
4488 
4489 		now = -addr & ~PAGE_MASK;
4490 		rc = emulator_read_write_onepage(addr, val, now, exception,
4491 						 vcpu, ops);
4492 
4493 		if (rc != X86EMUL_CONTINUE)
4494 			return rc;
4495 		addr += now;
4496 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4497 			addr = (u32)addr;
4498 		val += now;
4499 		bytes -= now;
4500 	}
4501 
4502 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4503 					 vcpu, ops);
4504 	if (rc != X86EMUL_CONTINUE)
4505 		return rc;
4506 
4507 	if (!vcpu->mmio_nr_fragments)
4508 		return rc;
4509 
4510 	gpa = vcpu->mmio_fragments[0].gpa;
4511 
4512 	vcpu->mmio_needed = 1;
4513 	vcpu->mmio_cur_fragment = 0;
4514 
4515 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4516 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4517 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4518 	vcpu->run->mmio.phys_addr = gpa;
4519 
4520 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4521 }
4522 
4523 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4524 				  unsigned long addr,
4525 				  void *val,
4526 				  unsigned int bytes,
4527 				  struct x86_exception *exception)
4528 {
4529 	return emulator_read_write(ctxt, addr, val, bytes,
4530 				   exception, &read_emultor);
4531 }
4532 
4533 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4534 			    unsigned long addr,
4535 			    const void *val,
4536 			    unsigned int bytes,
4537 			    struct x86_exception *exception)
4538 {
4539 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4540 				   exception, &write_emultor);
4541 }
4542 
4543 #define CMPXCHG_TYPE(t, ptr, old, new) \
4544 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4545 
4546 #ifdef CONFIG_X86_64
4547 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4548 #else
4549 #  define CMPXCHG64(ptr, old, new) \
4550 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4551 #endif
4552 
4553 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4554 				     unsigned long addr,
4555 				     const void *old,
4556 				     const void *new,
4557 				     unsigned int bytes,
4558 				     struct x86_exception *exception)
4559 {
4560 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4561 	gpa_t gpa;
4562 	struct page *page;
4563 	char *kaddr;
4564 	bool exchanged;
4565 
4566 	/* guests cmpxchg8b have to be emulated atomically */
4567 	if (bytes > 8 || (bytes & (bytes - 1)))
4568 		goto emul_write;
4569 
4570 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4571 
4572 	if (gpa == UNMAPPED_GVA ||
4573 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4574 		goto emul_write;
4575 
4576 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4577 		goto emul_write;
4578 
4579 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4580 	if (is_error_page(page))
4581 		goto emul_write;
4582 
4583 	kaddr = kmap_atomic(page);
4584 	kaddr += offset_in_page(gpa);
4585 	switch (bytes) {
4586 	case 1:
4587 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4588 		break;
4589 	case 2:
4590 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4591 		break;
4592 	case 4:
4593 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4594 		break;
4595 	case 8:
4596 		exchanged = CMPXCHG64(kaddr, old, new);
4597 		break;
4598 	default:
4599 		BUG();
4600 	}
4601 	kunmap_atomic(kaddr);
4602 	kvm_release_page_dirty(page);
4603 
4604 	if (!exchanged)
4605 		return X86EMUL_CMPXCHG_FAILED;
4606 
4607 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4608 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4609 
4610 	return X86EMUL_CONTINUE;
4611 
4612 emul_write:
4613 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4614 
4615 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4616 }
4617 
4618 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4619 {
4620 	/* TODO: String I/O for in kernel device */
4621 	int r;
4622 
4623 	if (vcpu->arch.pio.in)
4624 		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4625 				    vcpu->arch.pio.size, pd);
4626 	else
4627 		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4628 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4629 				     pd);
4630 	return r;
4631 }
4632 
4633 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4634 			       unsigned short port, void *val,
4635 			       unsigned int count, bool in)
4636 {
4637 	vcpu->arch.pio.port = port;
4638 	vcpu->arch.pio.in = in;
4639 	vcpu->arch.pio.count  = count;
4640 	vcpu->arch.pio.size = size;
4641 
4642 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4643 		vcpu->arch.pio.count = 0;
4644 		return 1;
4645 	}
4646 
4647 	vcpu->run->exit_reason = KVM_EXIT_IO;
4648 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4649 	vcpu->run->io.size = size;
4650 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4651 	vcpu->run->io.count = count;
4652 	vcpu->run->io.port = port;
4653 
4654 	return 0;
4655 }
4656 
4657 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4658 				    int size, unsigned short port, void *val,
4659 				    unsigned int count)
4660 {
4661 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4662 	int ret;
4663 
4664 	if (vcpu->arch.pio.count)
4665 		goto data_avail;
4666 
4667 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4668 	if (ret) {
4669 data_avail:
4670 		memcpy(val, vcpu->arch.pio_data, size * count);
4671 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4672 		vcpu->arch.pio.count = 0;
4673 		return 1;
4674 	}
4675 
4676 	return 0;
4677 }
4678 
4679 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4680 				     int size, unsigned short port,
4681 				     const void *val, unsigned int count)
4682 {
4683 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4684 
4685 	memcpy(vcpu->arch.pio_data, val, size * count);
4686 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4687 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4688 }
4689 
4690 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4691 {
4692 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4693 }
4694 
4695 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4696 {
4697 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4698 }
4699 
4700 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4701 {
4702 	if (!need_emulate_wbinvd(vcpu))
4703 		return X86EMUL_CONTINUE;
4704 
4705 	if (kvm_x86_ops->has_wbinvd_exit()) {
4706 		int cpu = get_cpu();
4707 
4708 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4709 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4710 				wbinvd_ipi, NULL, 1);
4711 		put_cpu();
4712 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4713 	} else
4714 		wbinvd();
4715 	return X86EMUL_CONTINUE;
4716 }
4717 
4718 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4719 {
4720 	kvm_x86_ops->skip_emulated_instruction(vcpu);
4721 	return kvm_emulate_wbinvd_noskip(vcpu);
4722 }
4723 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4724 
4725 
4726 
4727 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4728 {
4729 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4730 }
4731 
4732 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4733 			   unsigned long *dest)
4734 {
4735 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4736 }
4737 
4738 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4739 			   unsigned long value)
4740 {
4741 
4742 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4743 }
4744 
4745 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4746 {
4747 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4748 }
4749 
4750 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4751 {
4752 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4753 	unsigned long value;
4754 
4755 	switch (cr) {
4756 	case 0:
4757 		value = kvm_read_cr0(vcpu);
4758 		break;
4759 	case 2:
4760 		value = vcpu->arch.cr2;
4761 		break;
4762 	case 3:
4763 		value = kvm_read_cr3(vcpu);
4764 		break;
4765 	case 4:
4766 		value = kvm_read_cr4(vcpu);
4767 		break;
4768 	case 8:
4769 		value = kvm_get_cr8(vcpu);
4770 		break;
4771 	default:
4772 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4773 		return 0;
4774 	}
4775 
4776 	return value;
4777 }
4778 
4779 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4780 {
4781 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4782 	int res = 0;
4783 
4784 	switch (cr) {
4785 	case 0:
4786 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4787 		break;
4788 	case 2:
4789 		vcpu->arch.cr2 = val;
4790 		break;
4791 	case 3:
4792 		res = kvm_set_cr3(vcpu, val);
4793 		break;
4794 	case 4:
4795 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4796 		break;
4797 	case 8:
4798 		res = kvm_set_cr8(vcpu, val);
4799 		break;
4800 	default:
4801 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4802 		res = -1;
4803 	}
4804 
4805 	return res;
4806 }
4807 
4808 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4809 {
4810 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4811 }
4812 
4813 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4814 {
4815 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4816 }
4817 
4818 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4819 {
4820 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4821 }
4822 
4823 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4824 {
4825 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4826 }
4827 
4828 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4829 {
4830 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4831 }
4832 
4833 static unsigned long emulator_get_cached_segment_base(
4834 	struct x86_emulate_ctxt *ctxt, int seg)
4835 {
4836 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4837 }
4838 
4839 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4840 				 struct desc_struct *desc, u32 *base3,
4841 				 int seg)
4842 {
4843 	struct kvm_segment var;
4844 
4845 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4846 	*selector = var.selector;
4847 
4848 	if (var.unusable) {
4849 		memset(desc, 0, sizeof(*desc));
4850 		return false;
4851 	}
4852 
4853 	if (var.g)
4854 		var.limit >>= 12;
4855 	set_desc_limit(desc, var.limit);
4856 	set_desc_base(desc, (unsigned long)var.base);
4857 #ifdef CONFIG_X86_64
4858 	if (base3)
4859 		*base3 = var.base >> 32;
4860 #endif
4861 	desc->type = var.type;
4862 	desc->s = var.s;
4863 	desc->dpl = var.dpl;
4864 	desc->p = var.present;
4865 	desc->avl = var.avl;
4866 	desc->l = var.l;
4867 	desc->d = var.db;
4868 	desc->g = var.g;
4869 
4870 	return true;
4871 }
4872 
4873 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4874 				 struct desc_struct *desc, u32 base3,
4875 				 int seg)
4876 {
4877 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4878 	struct kvm_segment var;
4879 
4880 	var.selector = selector;
4881 	var.base = get_desc_base(desc);
4882 #ifdef CONFIG_X86_64
4883 	var.base |= ((u64)base3) << 32;
4884 #endif
4885 	var.limit = get_desc_limit(desc);
4886 	if (desc->g)
4887 		var.limit = (var.limit << 12) | 0xfff;
4888 	var.type = desc->type;
4889 	var.dpl = desc->dpl;
4890 	var.db = desc->d;
4891 	var.s = desc->s;
4892 	var.l = desc->l;
4893 	var.g = desc->g;
4894 	var.avl = desc->avl;
4895 	var.present = desc->p;
4896 	var.unusable = !var.present;
4897 	var.padding = 0;
4898 
4899 	kvm_set_segment(vcpu, &var, seg);
4900 	return;
4901 }
4902 
4903 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4904 			    u32 msr_index, u64 *pdata)
4905 {
4906 	struct msr_data msr;
4907 	int r;
4908 
4909 	msr.index = msr_index;
4910 	msr.host_initiated = false;
4911 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4912 	if (r)
4913 		return r;
4914 
4915 	*pdata = msr.data;
4916 	return 0;
4917 }
4918 
4919 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4920 			    u32 msr_index, u64 data)
4921 {
4922 	struct msr_data msr;
4923 
4924 	msr.data = data;
4925 	msr.index = msr_index;
4926 	msr.host_initiated = false;
4927 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4928 }
4929 
4930 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4931 {
4932 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4933 
4934 	return vcpu->arch.smbase;
4935 }
4936 
4937 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4938 {
4939 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4940 
4941 	vcpu->arch.smbase = smbase;
4942 }
4943 
4944 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4945 			      u32 pmc)
4946 {
4947 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4948 }
4949 
4950 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4951 			     u32 pmc, u64 *pdata)
4952 {
4953 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4954 }
4955 
4956 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4957 {
4958 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4959 }
4960 
4961 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4962 {
4963 	preempt_disable();
4964 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4965 	/*
4966 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4967 	 * so it may be clear at this point.
4968 	 */
4969 	clts();
4970 }
4971 
4972 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4973 {
4974 	preempt_enable();
4975 }
4976 
4977 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4978 			      struct x86_instruction_info *info,
4979 			      enum x86_intercept_stage stage)
4980 {
4981 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4982 }
4983 
4984 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4985 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4986 {
4987 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4988 }
4989 
4990 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4991 {
4992 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
4993 }
4994 
4995 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4996 {
4997 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4998 }
4999 
5000 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5001 {
5002 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5003 }
5004 
5005 static const struct x86_emulate_ops emulate_ops = {
5006 	.read_gpr            = emulator_read_gpr,
5007 	.write_gpr           = emulator_write_gpr,
5008 	.read_std            = kvm_read_guest_virt_system,
5009 	.write_std           = kvm_write_guest_virt_system,
5010 	.read_phys           = kvm_read_guest_phys_system,
5011 	.fetch               = kvm_fetch_guest_virt,
5012 	.read_emulated       = emulator_read_emulated,
5013 	.write_emulated      = emulator_write_emulated,
5014 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5015 	.invlpg              = emulator_invlpg,
5016 	.pio_in_emulated     = emulator_pio_in_emulated,
5017 	.pio_out_emulated    = emulator_pio_out_emulated,
5018 	.get_segment         = emulator_get_segment,
5019 	.set_segment         = emulator_set_segment,
5020 	.get_cached_segment_base = emulator_get_cached_segment_base,
5021 	.get_gdt             = emulator_get_gdt,
5022 	.get_idt	     = emulator_get_idt,
5023 	.set_gdt             = emulator_set_gdt,
5024 	.set_idt	     = emulator_set_idt,
5025 	.get_cr              = emulator_get_cr,
5026 	.set_cr              = emulator_set_cr,
5027 	.cpl                 = emulator_get_cpl,
5028 	.get_dr              = emulator_get_dr,
5029 	.set_dr              = emulator_set_dr,
5030 	.get_smbase          = emulator_get_smbase,
5031 	.set_smbase          = emulator_set_smbase,
5032 	.set_msr             = emulator_set_msr,
5033 	.get_msr             = emulator_get_msr,
5034 	.check_pmc	     = emulator_check_pmc,
5035 	.read_pmc            = emulator_read_pmc,
5036 	.halt                = emulator_halt,
5037 	.wbinvd              = emulator_wbinvd,
5038 	.fix_hypercall       = emulator_fix_hypercall,
5039 	.get_fpu             = emulator_get_fpu,
5040 	.put_fpu             = emulator_put_fpu,
5041 	.intercept           = emulator_intercept,
5042 	.get_cpuid           = emulator_get_cpuid,
5043 	.set_nmi_mask        = emulator_set_nmi_mask,
5044 };
5045 
5046 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5047 {
5048 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5049 	/*
5050 	 * an sti; sti; sequence only disable interrupts for the first
5051 	 * instruction. So, if the last instruction, be it emulated or
5052 	 * not, left the system with the INT_STI flag enabled, it
5053 	 * means that the last instruction is an sti. We should not
5054 	 * leave the flag on in this case. The same goes for mov ss
5055 	 */
5056 	if (int_shadow & mask)
5057 		mask = 0;
5058 	if (unlikely(int_shadow || mask)) {
5059 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5060 		if (!mask)
5061 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5062 	}
5063 }
5064 
5065 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5066 {
5067 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5068 	if (ctxt->exception.vector == PF_VECTOR)
5069 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5070 
5071 	if (ctxt->exception.error_code_valid)
5072 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5073 				      ctxt->exception.error_code);
5074 	else
5075 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5076 	return false;
5077 }
5078 
5079 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5080 {
5081 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5082 	int cs_db, cs_l;
5083 
5084 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5085 
5086 	ctxt->eflags = kvm_get_rflags(vcpu);
5087 	ctxt->eip = kvm_rip_read(vcpu);
5088 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5089 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5090 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5091 		     cs_db				? X86EMUL_MODE_PROT32 :
5092 							  X86EMUL_MODE_PROT16;
5093 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5094 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5095 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5096 	ctxt->emul_flags = vcpu->arch.hflags;
5097 
5098 	init_decode_cache(ctxt);
5099 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5100 }
5101 
5102 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5103 {
5104 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5105 	int ret;
5106 
5107 	init_emulate_ctxt(vcpu);
5108 
5109 	ctxt->op_bytes = 2;
5110 	ctxt->ad_bytes = 2;
5111 	ctxt->_eip = ctxt->eip + inc_eip;
5112 	ret = emulate_int_real(ctxt, irq);
5113 
5114 	if (ret != X86EMUL_CONTINUE)
5115 		return EMULATE_FAIL;
5116 
5117 	ctxt->eip = ctxt->_eip;
5118 	kvm_rip_write(vcpu, ctxt->eip);
5119 	kvm_set_rflags(vcpu, ctxt->eflags);
5120 
5121 	if (irq == NMI_VECTOR)
5122 		vcpu->arch.nmi_pending = 0;
5123 	else
5124 		vcpu->arch.interrupt.pending = false;
5125 
5126 	return EMULATE_DONE;
5127 }
5128 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5129 
5130 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5131 {
5132 	int r = EMULATE_DONE;
5133 
5134 	++vcpu->stat.insn_emulation_fail;
5135 	trace_kvm_emulate_insn_failed(vcpu);
5136 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5137 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5138 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5139 		vcpu->run->internal.ndata = 0;
5140 		r = EMULATE_FAIL;
5141 	}
5142 	kvm_queue_exception(vcpu, UD_VECTOR);
5143 
5144 	return r;
5145 }
5146 
5147 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5148 				  bool write_fault_to_shadow_pgtable,
5149 				  int emulation_type)
5150 {
5151 	gpa_t gpa = cr2;
5152 	kvm_pfn_t pfn;
5153 
5154 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5155 		return false;
5156 
5157 	if (!vcpu->arch.mmu.direct_map) {
5158 		/*
5159 		 * Write permission should be allowed since only
5160 		 * write access need to be emulated.
5161 		 */
5162 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5163 
5164 		/*
5165 		 * If the mapping is invalid in guest, let cpu retry
5166 		 * it to generate fault.
5167 		 */
5168 		if (gpa == UNMAPPED_GVA)
5169 			return true;
5170 	}
5171 
5172 	/*
5173 	 * Do not retry the unhandleable instruction if it faults on the
5174 	 * readonly host memory, otherwise it will goto a infinite loop:
5175 	 * retry instruction -> write #PF -> emulation fail -> retry
5176 	 * instruction -> ...
5177 	 */
5178 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5179 
5180 	/*
5181 	 * If the instruction failed on the error pfn, it can not be fixed,
5182 	 * report the error to userspace.
5183 	 */
5184 	if (is_error_noslot_pfn(pfn))
5185 		return false;
5186 
5187 	kvm_release_pfn_clean(pfn);
5188 
5189 	/* The instructions are well-emulated on direct mmu. */
5190 	if (vcpu->arch.mmu.direct_map) {
5191 		unsigned int indirect_shadow_pages;
5192 
5193 		spin_lock(&vcpu->kvm->mmu_lock);
5194 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5195 		spin_unlock(&vcpu->kvm->mmu_lock);
5196 
5197 		if (indirect_shadow_pages)
5198 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5199 
5200 		return true;
5201 	}
5202 
5203 	/*
5204 	 * if emulation was due to access to shadowed page table
5205 	 * and it failed try to unshadow page and re-enter the
5206 	 * guest to let CPU execute the instruction.
5207 	 */
5208 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5209 
5210 	/*
5211 	 * If the access faults on its page table, it can not
5212 	 * be fixed by unprotecting shadow page and it should
5213 	 * be reported to userspace.
5214 	 */
5215 	return !write_fault_to_shadow_pgtable;
5216 }
5217 
5218 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5219 			      unsigned long cr2,  int emulation_type)
5220 {
5221 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5222 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5223 
5224 	last_retry_eip = vcpu->arch.last_retry_eip;
5225 	last_retry_addr = vcpu->arch.last_retry_addr;
5226 
5227 	/*
5228 	 * If the emulation is caused by #PF and it is non-page_table
5229 	 * writing instruction, it means the VM-EXIT is caused by shadow
5230 	 * page protected, we can zap the shadow page and retry this
5231 	 * instruction directly.
5232 	 *
5233 	 * Note: if the guest uses a non-page-table modifying instruction
5234 	 * on the PDE that points to the instruction, then we will unmap
5235 	 * the instruction and go to an infinite loop. So, we cache the
5236 	 * last retried eip and the last fault address, if we meet the eip
5237 	 * and the address again, we can break out of the potential infinite
5238 	 * loop.
5239 	 */
5240 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5241 
5242 	if (!(emulation_type & EMULTYPE_RETRY))
5243 		return false;
5244 
5245 	if (x86_page_table_writing_insn(ctxt))
5246 		return false;
5247 
5248 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5249 		return false;
5250 
5251 	vcpu->arch.last_retry_eip = ctxt->eip;
5252 	vcpu->arch.last_retry_addr = cr2;
5253 
5254 	if (!vcpu->arch.mmu.direct_map)
5255 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5256 
5257 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5258 
5259 	return true;
5260 }
5261 
5262 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5263 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5264 
5265 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5266 {
5267 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5268 		/* This is a good place to trace that we are exiting SMM.  */
5269 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5270 
5271 		if (unlikely(vcpu->arch.smi_pending)) {
5272 			kvm_make_request(KVM_REQ_SMI, vcpu);
5273 			vcpu->arch.smi_pending = 0;
5274 		} else {
5275 			/* Process a latched INIT, if any.  */
5276 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5277 		}
5278 	}
5279 
5280 	kvm_mmu_reset_context(vcpu);
5281 }
5282 
5283 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5284 {
5285 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5286 
5287 	vcpu->arch.hflags = emul_flags;
5288 
5289 	if (changed & HF_SMM_MASK)
5290 		kvm_smm_changed(vcpu);
5291 }
5292 
5293 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5294 				unsigned long *db)
5295 {
5296 	u32 dr6 = 0;
5297 	int i;
5298 	u32 enable, rwlen;
5299 
5300 	enable = dr7;
5301 	rwlen = dr7 >> 16;
5302 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5303 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5304 			dr6 |= (1 << i);
5305 	return dr6;
5306 }
5307 
5308 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5309 {
5310 	struct kvm_run *kvm_run = vcpu->run;
5311 
5312 	/*
5313 	 * rflags is the old, "raw" value of the flags.  The new value has
5314 	 * not been saved yet.
5315 	 *
5316 	 * This is correct even for TF set by the guest, because "the
5317 	 * processor will not generate this exception after the instruction
5318 	 * that sets the TF flag".
5319 	 */
5320 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5321 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5322 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5323 						  DR6_RTM;
5324 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5325 			kvm_run->debug.arch.exception = DB_VECTOR;
5326 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5327 			*r = EMULATE_USER_EXIT;
5328 		} else {
5329 			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5330 			/*
5331 			 * "Certain debug exceptions may clear bit 0-3.  The
5332 			 * remaining contents of the DR6 register are never
5333 			 * cleared by the processor".
5334 			 */
5335 			vcpu->arch.dr6 &= ~15;
5336 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5337 			kvm_queue_exception(vcpu, DB_VECTOR);
5338 		}
5339 	}
5340 }
5341 
5342 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5343 {
5344 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5345 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5346 		struct kvm_run *kvm_run = vcpu->run;
5347 		unsigned long eip = kvm_get_linear_rip(vcpu);
5348 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5349 					   vcpu->arch.guest_debug_dr7,
5350 					   vcpu->arch.eff_db);
5351 
5352 		if (dr6 != 0) {
5353 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5354 			kvm_run->debug.arch.pc = eip;
5355 			kvm_run->debug.arch.exception = DB_VECTOR;
5356 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5357 			*r = EMULATE_USER_EXIT;
5358 			return true;
5359 		}
5360 	}
5361 
5362 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5363 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5364 		unsigned long eip = kvm_get_linear_rip(vcpu);
5365 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5366 					   vcpu->arch.dr7,
5367 					   vcpu->arch.db);
5368 
5369 		if (dr6 != 0) {
5370 			vcpu->arch.dr6 &= ~15;
5371 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5372 			kvm_queue_exception(vcpu, DB_VECTOR);
5373 			*r = EMULATE_DONE;
5374 			return true;
5375 		}
5376 	}
5377 
5378 	return false;
5379 }
5380 
5381 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5382 			    unsigned long cr2,
5383 			    int emulation_type,
5384 			    void *insn,
5385 			    int insn_len)
5386 {
5387 	int r;
5388 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5389 	bool writeback = true;
5390 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5391 
5392 	/*
5393 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5394 	 * never reused.
5395 	 */
5396 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5397 	kvm_clear_exception_queue(vcpu);
5398 
5399 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5400 		init_emulate_ctxt(vcpu);
5401 
5402 		/*
5403 		 * We will reenter on the same instruction since
5404 		 * we do not set complete_userspace_io.  This does not
5405 		 * handle watchpoints yet, those would be handled in
5406 		 * the emulate_ops.
5407 		 */
5408 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5409 			return r;
5410 
5411 		ctxt->interruptibility = 0;
5412 		ctxt->have_exception = false;
5413 		ctxt->exception.vector = -1;
5414 		ctxt->perm_ok = false;
5415 
5416 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5417 
5418 		r = x86_decode_insn(ctxt, insn, insn_len);
5419 
5420 		trace_kvm_emulate_insn_start(vcpu);
5421 		++vcpu->stat.insn_emulation;
5422 		if (r != EMULATION_OK)  {
5423 			if (emulation_type & EMULTYPE_TRAP_UD)
5424 				return EMULATE_FAIL;
5425 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5426 						emulation_type))
5427 				return EMULATE_DONE;
5428 			if (emulation_type & EMULTYPE_SKIP)
5429 				return EMULATE_FAIL;
5430 			return handle_emulation_failure(vcpu);
5431 		}
5432 	}
5433 
5434 	if (emulation_type & EMULTYPE_SKIP) {
5435 		kvm_rip_write(vcpu, ctxt->_eip);
5436 		if (ctxt->eflags & X86_EFLAGS_RF)
5437 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5438 		return EMULATE_DONE;
5439 	}
5440 
5441 	if (retry_instruction(ctxt, cr2, emulation_type))
5442 		return EMULATE_DONE;
5443 
5444 	/* this is needed for vmware backdoor interface to work since it
5445 	   changes registers values  during IO operation */
5446 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5447 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5448 		emulator_invalidate_register_cache(ctxt);
5449 	}
5450 
5451 restart:
5452 	r = x86_emulate_insn(ctxt);
5453 
5454 	if (r == EMULATION_INTERCEPTED)
5455 		return EMULATE_DONE;
5456 
5457 	if (r == EMULATION_FAILED) {
5458 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5459 					emulation_type))
5460 			return EMULATE_DONE;
5461 
5462 		return handle_emulation_failure(vcpu);
5463 	}
5464 
5465 	if (ctxt->have_exception) {
5466 		r = EMULATE_DONE;
5467 		if (inject_emulated_exception(vcpu))
5468 			return r;
5469 	} else if (vcpu->arch.pio.count) {
5470 		if (!vcpu->arch.pio.in) {
5471 			/* FIXME: return into emulator if single-stepping.  */
5472 			vcpu->arch.pio.count = 0;
5473 		} else {
5474 			writeback = false;
5475 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5476 		}
5477 		r = EMULATE_USER_EXIT;
5478 	} else if (vcpu->mmio_needed) {
5479 		if (!vcpu->mmio_is_write)
5480 			writeback = false;
5481 		r = EMULATE_USER_EXIT;
5482 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5483 	} else if (r == EMULATION_RESTART)
5484 		goto restart;
5485 	else
5486 		r = EMULATE_DONE;
5487 
5488 	if (writeback) {
5489 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5490 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5491 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5492 		if (vcpu->arch.hflags != ctxt->emul_flags)
5493 			kvm_set_hflags(vcpu, ctxt->emul_flags);
5494 		kvm_rip_write(vcpu, ctxt->eip);
5495 		if (r == EMULATE_DONE)
5496 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5497 		if (!ctxt->have_exception ||
5498 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5499 			__kvm_set_rflags(vcpu, ctxt->eflags);
5500 
5501 		/*
5502 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5503 		 * do nothing, and it will be requested again as soon as
5504 		 * the shadow expires.  But we still need to check here,
5505 		 * because POPF has no interrupt shadow.
5506 		 */
5507 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5508 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5509 	} else
5510 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5511 
5512 	return r;
5513 }
5514 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5515 
5516 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5517 {
5518 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5519 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5520 					    size, port, &val, 1);
5521 	/* do not return to emulator after return from userspace */
5522 	vcpu->arch.pio.count = 0;
5523 	return ret;
5524 }
5525 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5526 
5527 static void tsc_bad(void *info)
5528 {
5529 	__this_cpu_write(cpu_tsc_khz, 0);
5530 }
5531 
5532 static void tsc_khz_changed(void *data)
5533 {
5534 	struct cpufreq_freqs *freq = data;
5535 	unsigned long khz = 0;
5536 
5537 	if (data)
5538 		khz = freq->new;
5539 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5540 		khz = cpufreq_quick_get(raw_smp_processor_id());
5541 	if (!khz)
5542 		khz = tsc_khz;
5543 	__this_cpu_write(cpu_tsc_khz, khz);
5544 }
5545 
5546 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5547 				     void *data)
5548 {
5549 	struct cpufreq_freqs *freq = data;
5550 	struct kvm *kvm;
5551 	struct kvm_vcpu *vcpu;
5552 	int i, send_ipi = 0;
5553 
5554 	/*
5555 	 * We allow guests to temporarily run on slowing clocks,
5556 	 * provided we notify them after, or to run on accelerating
5557 	 * clocks, provided we notify them before.  Thus time never
5558 	 * goes backwards.
5559 	 *
5560 	 * However, we have a problem.  We can't atomically update
5561 	 * the frequency of a given CPU from this function; it is
5562 	 * merely a notifier, which can be called from any CPU.
5563 	 * Changing the TSC frequency at arbitrary points in time
5564 	 * requires a recomputation of local variables related to
5565 	 * the TSC for each VCPU.  We must flag these local variables
5566 	 * to be updated and be sure the update takes place with the
5567 	 * new frequency before any guests proceed.
5568 	 *
5569 	 * Unfortunately, the combination of hotplug CPU and frequency
5570 	 * change creates an intractable locking scenario; the order
5571 	 * of when these callouts happen is undefined with respect to
5572 	 * CPU hotplug, and they can race with each other.  As such,
5573 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5574 	 * undefined; you can actually have a CPU frequency change take
5575 	 * place in between the computation of X and the setting of the
5576 	 * variable.  To protect against this problem, all updates of
5577 	 * the per_cpu tsc_khz variable are done in an interrupt
5578 	 * protected IPI, and all callers wishing to update the value
5579 	 * must wait for a synchronous IPI to complete (which is trivial
5580 	 * if the caller is on the CPU already).  This establishes the
5581 	 * necessary total order on variable updates.
5582 	 *
5583 	 * Note that because a guest time update may take place
5584 	 * anytime after the setting of the VCPU's request bit, the
5585 	 * correct TSC value must be set before the request.  However,
5586 	 * to ensure the update actually makes it to any guest which
5587 	 * starts running in hardware virtualization between the set
5588 	 * and the acquisition of the spinlock, we must also ping the
5589 	 * CPU after setting the request bit.
5590 	 *
5591 	 */
5592 
5593 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5594 		return 0;
5595 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5596 		return 0;
5597 
5598 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5599 
5600 	spin_lock(&kvm_lock);
5601 	list_for_each_entry(kvm, &vm_list, vm_list) {
5602 		kvm_for_each_vcpu(i, vcpu, kvm) {
5603 			if (vcpu->cpu != freq->cpu)
5604 				continue;
5605 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5606 			if (vcpu->cpu != smp_processor_id())
5607 				send_ipi = 1;
5608 		}
5609 	}
5610 	spin_unlock(&kvm_lock);
5611 
5612 	if (freq->old < freq->new && send_ipi) {
5613 		/*
5614 		 * We upscale the frequency.  Must make the guest
5615 		 * doesn't see old kvmclock values while running with
5616 		 * the new frequency, otherwise we risk the guest sees
5617 		 * time go backwards.
5618 		 *
5619 		 * In case we update the frequency for another cpu
5620 		 * (which might be in guest context) send an interrupt
5621 		 * to kick the cpu out of guest context.  Next time
5622 		 * guest context is entered kvmclock will be updated,
5623 		 * so the guest will not see stale values.
5624 		 */
5625 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5626 	}
5627 	return 0;
5628 }
5629 
5630 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5631 	.notifier_call  = kvmclock_cpufreq_notifier
5632 };
5633 
5634 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5635 					unsigned long action, void *hcpu)
5636 {
5637 	unsigned int cpu = (unsigned long)hcpu;
5638 
5639 	switch (action) {
5640 		case CPU_ONLINE:
5641 		case CPU_DOWN_FAILED:
5642 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5643 			break;
5644 		case CPU_DOWN_PREPARE:
5645 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5646 			break;
5647 	}
5648 	return NOTIFY_OK;
5649 }
5650 
5651 static struct notifier_block kvmclock_cpu_notifier_block = {
5652 	.notifier_call  = kvmclock_cpu_notifier,
5653 	.priority = -INT_MAX
5654 };
5655 
5656 static void kvm_timer_init(void)
5657 {
5658 	int cpu;
5659 
5660 	max_tsc_khz = tsc_khz;
5661 
5662 	cpu_notifier_register_begin();
5663 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5664 #ifdef CONFIG_CPU_FREQ
5665 		struct cpufreq_policy policy;
5666 		memset(&policy, 0, sizeof(policy));
5667 		cpu = get_cpu();
5668 		cpufreq_get_policy(&policy, cpu);
5669 		if (policy.cpuinfo.max_freq)
5670 			max_tsc_khz = policy.cpuinfo.max_freq;
5671 		put_cpu();
5672 #endif
5673 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5674 					  CPUFREQ_TRANSITION_NOTIFIER);
5675 	}
5676 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5677 	for_each_online_cpu(cpu)
5678 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5679 
5680 	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5681 	cpu_notifier_register_done();
5682 
5683 }
5684 
5685 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5686 
5687 int kvm_is_in_guest(void)
5688 {
5689 	return __this_cpu_read(current_vcpu) != NULL;
5690 }
5691 
5692 static int kvm_is_user_mode(void)
5693 {
5694 	int user_mode = 3;
5695 
5696 	if (__this_cpu_read(current_vcpu))
5697 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5698 
5699 	return user_mode != 0;
5700 }
5701 
5702 static unsigned long kvm_get_guest_ip(void)
5703 {
5704 	unsigned long ip = 0;
5705 
5706 	if (__this_cpu_read(current_vcpu))
5707 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5708 
5709 	return ip;
5710 }
5711 
5712 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5713 	.is_in_guest		= kvm_is_in_guest,
5714 	.is_user_mode		= kvm_is_user_mode,
5715 	.get_guest_ip		= kvm_get_guest_ip,
5716 };
5717 
5718 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5719 {
5720 	__this_cpu_write(current_vcpu, vcpu);
5721 }
5722 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5723 
5724 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5725 {
5726 	__this_cpu_write(current_vcpu, NULL);
5727 }
5728 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5729 
5730 static void kvm_set_mmio_spte_mask(void)
5731 {
5732 	u64 mask;
5733 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5734 
5735 	/*
5736 	 * Set the reserved bits and the present bit of an paging-structure
5737 	 * entry to generate page fault with PFER.RSV = 1.
5738 	 */
5739 	 /* Mask the reserved physical address bits. */
5740 	mask = rsvd_bits(maxphyaddr, 51);
5741 
5742 	/* Bit 62 is always reserved for 32bit host. */
5743 	mask |= 0x3ull << 62;
5744 
5745 	/* Set the present bit. */
5746 	mask |= 1ull;
5747 
5748 #ifdef CONFIG_X86_64
5749 	/*
5750 	 * If reserved bit is not supported, clear the present bit to disable
5751 	 * mmio page fault.
5752 	 */
5753 	if (maxphyaddr == 52)
5754 		mask &= ~1ull;
5755 #endif
5756 
5757 	kvm_mmu_set_mmio_spte_mask(mask);
5758 }
5759 
5760 #ifdef CONFIG_X86_64
5761 static void pvclock_gtod_update_fn(struct work_struct *work)
5762 {
5763 	struct kvm *kvm;
5764 
5765 	struct kvm_vcpu *vcpu;
5766 	int i;
5767 
5768 	spin_lock(&kvm_lock);
5769 	list_for_each_entry(kvm, &vm_list, vm_list)
5770 		kvm_for_each_vcpu(i, vcpu, kvm)
5771 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5772 	atomic_set(&kvm_guest_has_master_clock, 0);
5773 	spin_unlock(&kvm_lock);
5774 }
5775 
5776 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5777 
5778 /*
5779  * Notification about pvclock gtod data update.
5780  */
5781 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5782 			       void *priv)
5783 {
5784 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5785 	struct timekeeper *tk = priv;
5786 
5787 	update_pvclock_gtod(tk);
5788 
5789 	/* disable master clock if host does not trust, or does not
5790 	 * use, TSC clocksource
5791 	 */
5792 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5793 	    atomic_read(&kvm_guest_has_master_clock) != 0)
5794 		queue_work(system_long_wq, &pvclock_gtod_work);
5795 
5796 	return 0;
5797 }
5798 
5799 static struct notifier_block pvclock_gtod_notifier = {
5800 	.notifier_call = pvclock_gtod_notify,
5801 };
5802 #endif
5803 
5804 int kvm_arch_init(void *opaque)
5805 {
5806 	int r;
5807 	struct kvm_x86_ops *ops = opaque;
5808 
5809 	if (kvm_x86_ops) {
5810 		printk(KERN_ERR "kvm: already loaded the other module\n");
5811 		r = -EEXIST;
5812 		goto out;
5813 	}
5814 
5815 	if (!ops->cpu_has_kvm_support()) {
5816 		printk(KERN_ERR "kvm: no hardware support\n");
5817 		r = -EOPNOTSUPP;
5818 		goto out;
5819 	}
5820 	if (ops->disabled_by_bios()) {
5821 		printk(KERN_ERR "kvm: disabled by bios\n");
5822 		r = -EOPNOTSUPP;
5823 		goto out;
5824 	}
5825 
5826 	r = -ENOMEM;
5827 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5828 	if (!shared_msrs) {
5829 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5830 		goto out;
5831 	}
5832 
5833 	r = kvm_mmu_module_init();
5834 	if (r)
5835 		goto out_free_percpu;
5836 
5837 	kvm_set_mmio_spte_mask();
5838 
5839 	kvm_x86_ops = ops;
5840 
5841 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5842 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5843 
5844 	kvm_timer_init();
5845 
5846 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5847 
5848 	if (cpu_has_xsave)
5849 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5850 
5851 	kvm_lapic_init();
5852 #ifdef CONFIG_X86_64
5853 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5854 #endif
5855 
5856 	return 0;
5857 
5858 out_free_percpu:
5859 	free_percpu(shared_msrs);
5860 out:
5861 	return r;
5862 }
5863 
5864 void kvm_arch_exit(void)
5865 {
5866 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5867 
5868 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5869 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5870 					    CPUFREQ_TRANSITION_NOTIFIER);
5871 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5872 #ifdef CONFIG_X86_64
5873 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5874 #endif
5875 	kvm_x86_ops = NULL;
5876 	kvm_mmu_module_exit();
5877 	free_percpu(shared_msrs);
5878 }
5879 
5880 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5881 {
5882 	++vcpu->stat.halt_exits;
5883 	if (lapic_in_kernel(vcpu)) {
5884 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5885 		return 1;
5886 	} else {
5887 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5888 		return 0;
5889 	}
5890 }
5891 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5892 
5893 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5894 {
5895 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5896 	return kvm_vcpu_halt(vcpu);
5897 }
5898 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5899 
5900 /*
5901  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5902  *
5903  * @apicid - apicid of vcpu to be kicked.
5904  */
5905 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5906 {
5907 	struct kvm_lapic_irq lapic_irq;
5908 
5909 	lapic_irq.shorthand = 0;
5910 	lapic_irq.dest_mode = 0;
5911 	lapic_irq.dest_id = apicid;
5912 	lapic_irq.msi_redir_hint = false;
5913 
5914 	lapic_irq.delivery_mode = APIC_DM_REMRD;
5915 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5916 }
5917 
5918 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5919 {
5920 	vcpu->arch.apicv_active = false;
5921 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5922 }
5923 
5924 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5925 {
5926 	unsigned long nr, a0, a1, a2, a3, ret;
5927 	int op_64_bit, r = 1;
5928 
5929 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5930 
5931 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5932 		return kvm_hv_hypercall(vcpu);
5933 
5934 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5935 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5936 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5937 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5938 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5939 
5940 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5941 
5942 	op_64_bit = is_64_bit_mode(vcpu);
5943 	if (!op_64_bit) {
5944 		nr &= 0xFFFFFFFF;
5945 		a0 &= 0xFFFFFFFF;
5946 		a1 &= 0xFFFFFFFF;
5947 		a2 &= 0xFFFFFFFF;
5948 		a3 &= 0xFFFFFFFF;
5949 	}
5950 
5951 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5952 		ret = -KVM_EPERM;
5953 		goto out;
5954 	}
5955 
5956 	switch (nr) {
5957 	case KVM_HC_VAPIC_POLL_IRQ:
5958 		ret = 0;
5959 		break;
5960 	case KVM_HC_KICK_CPU:
5961 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5962 		ret = 0;
5963 		break;
5964 	default:
5965 		ret = -KVM_ENOSYS;
5966 		break;
5967 	}
5968 out:
5969 	if (!op_64_bit)
5970 		ret = (u32)ret;
5971 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5972 	++vcpu->stat.hypercalls;
5973 	return r;
5974 }
5975 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5976 
5977 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5978 {
5979 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5980 	char instruction[3];
5981 	unsigned long rip = kvm_rip_read(vcpu);
5982 
5983 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
5984 
5985 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5986 }
5987 
5988 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5989 {
5990 	return vcpu->run->request_interrupt_window &&
5991 		likely(!pic_in_kernel(vcpu->kvm));
5992 }
5993 
5994 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5995 {
5996 	struct kvm_run *kvm_run = vcpu->run;
5997 
5998 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5999 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6000 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6001 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6002 	kvm_run->ready_for_interrupt_injection =
6003 		pic_in_kernel(vcpu->kvm) ||
6004 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
6005 }
6006 
6007 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6008 {
6009 	int max_irr, tpr;
6010 
6011 	if (!kvm_x86_ops->update_cr8_intercept)
6012 		return;
6013 
6014 	if (!vcpu->arch.apic)
6015 		return;
6016 
6017 	if (vcpu->arch.apicv_active)
6018 		return;
6019 
6020 	if (!vcpu->arch.apic->vapic_addr)
6021 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6022 	else
6023 		max_irr = -1;
6024 
6025 	if (max_irr != -1)
6026 		max_irr >>= 4;
6027 
6028 	tpr = kvm_lapic_get_cr8(vcpu);
6029 
6030 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6031 }
6032 
6033 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6034 {
6035 	int r;
6036 
6037 	/* try to reinject previous events if any */
6038 	if (vcpu->arch.exception.pending) {
6039 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6040 					vcpu->arch.exception.has_error_code,
6041 					vcpu->arch.exception.error_code);
6042 
6043 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6044 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6045 					     X86_EFLAGS_RF);
6046 
6047 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6048 		    (vcpu->arch.dr7 & DR7_GD)) {
6049 			vcpu->arch.dr7 &= ~DR7_GD;
6050 			kvm_update_dr7(vcpu);
6051 		}
6052 
6053 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6054 					  vcpu->arch.exception.has_error_code,
6055 					  vcpu->arch.exception.error_code,
6056 					  vcpu->arch.exception.reinject);
6057 		return 0;
6058 	}
6059 
6060 	if (vcpu->arch.nmi_injected) {
6061 		kvm_x86_ops->set_nmi(vcpu);
6062 		return 0;
6063 	}
6064 
6065 	if (vcpu->arch.interrupt.pending) {
6066 		kvm_x86_ops->set_irq(vcpu);
6067 		return 0;
6068 	}
6069 
6070 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6071 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6072 		if (r != 0)
6073 			return r;
6074 	}
6075 
6076 	/* try to inject new event if pending */
6077 	if (vcpu->arch.nmi_pending) {
6078 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
6079 			--vcpu->arch.nmi_pending;
6080 			vcpu->arch.nmi_injected = true;
6081 			kvm_x86_ops->set_nmi(vcpu);
6082 		}
6083 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6084 		/*
6085 		 * Because interrupts can be injected asynchronously, we are
6086 		 * calling check_nested_events again here to avoid a race condition.
6087 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6088 		 * proposal and current concerns.  Perhaps we should be setting
6089 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6090 		 */
6091 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6092 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6093 			if (r != 0)
6094 				return r;
6095 		}
6096 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6097 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6098 					    false);
6099 			kvm_x86_ops->set_irq(vcpu);
6100 		}
6101 	}
6102 	return 0;
6103 }
6104 
6105 static void process_nmi(struct kvm_vcpu *vcpu)
6106 {
6107 	unsigned limit = 2;
6108 
6109 	/*
6110 	 * x86 is limited to one NMI running, and one NMI pending after it.
6111 	 * If an NMI is already in progress, limit further NMIs to just one.
6112 	 * Otherwise, allow two (and we'll inject the first one immediately).
6113 	 */
6114 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6115 		limit = 1;
6116 
6117 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6118 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6119 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6120 }
6121 
6122 #define put_smstate(type, buf, offset, val)			  \
6123 	*(type *)((buf) + (offset) - 0x7e00) = val
6124 
6125 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6126 {
6127 	u32 flags = 0;
6128 	flags |= seg->g       << 23;
6129 	flags |= seg->db      << 22;
6130 	flags |= seg->l       << 21;
6131 	flags |= seg->avl     << 20;
6132 	flags |= seg->present << 15;
6133 	flags |= seg->dpl     << 13;
6134 	flags |= seg->s       << 12;
6135 	flags |= seg->type    << 8;
6136 	return flags;
6137 }
6138 
6139 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6140 {
6141 	struct kvm_segment seg;
6142 	int offset;
6143 
6144 	kvm_get_segment(vcpu, &seg, n);
6145 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6146 
6147 	if (n < 3)
6148 		offset = 0x7f84 + n * 12;
6149 	else
6150 		offset = 0x7f2c + (n - 3) * 12;
6151 
6152 	put_smstate(u32, buf, offset + 8, seg.base);
6153 	put_smstate(u32, buf, offset + 4, seg.limit);
6154 	put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6155 }
6156 
6157 #ifdef CONFIG_X86_64
6158 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6159 {
6160 	struct kvm_segment seg;
6161 	int offset;
6162 	u16 flags;
6163 
6164 	kvm_get_segment(vcpu, &seg, n);
6165 	offset = 0x7e00 + n * 16;
6166 
6167 	flags = process_smi_get_segment_flags(&seg) >> 8;
6168 	put_smstate(u16, buf, offset, seg.selector);
6169 	put_smstate(u16, buf, offset + 2, flags);
6170 	put_smstate(u32, buf, offset + 4, seg.limit);
6171 	put_smstate(u64, buf, offset + 8, seg.base);
6172 }
6173 #endif
6174 
6175 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6176 {
6177 	struct desc_ptr dt;
6178 	struct kvm_segment seg;
6179 	unsigned long val;
6180 	int i;
6181 
6182 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6183 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6184 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6185 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6186 
6187 	for (i = 0; i < 8; i++)
6188 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6189 
6190 	kvm_get_dr(vcpu, 6, &val);
6191 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6192 	kvm_get_dr(vcpu, 7, &val);
6193 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6194 
6195 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6196 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6197 	put_smstate(u32, buf, 0x7f64, seg.base);
6198 	put_smstate(u32, buf, 0x7f60, seg.limit);
6199 	put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6200 
6201 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6202 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6203 	put_smstate(u32, buf, 0x7f80, seg.base);
6204 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6205 	put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6206 
6207 	kvm_x86_ops->get_gdt(vcpu, &dt);
6208 	put_smstate(u32, buf, 0x7f74, dt.address);
6209 	put_smstate(u32, buf, 0x7f70, dt.size);
6210 
6211 	kvm_x86_ops->get_idt(vcpu, &dt);
6212 	put_smstate(u32, buf, 0x7f58, dt.address);
6213 	put_smstate(u32, buf, 0x7f54, dt.size);
6214 
6215 	for (i = 0; i < 6; i++)
6216 		process_smi_save_seg_32(vcpu, buf, i);
6217 
6218 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6219 
6220 	/* revision id */
6221 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6222 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6223 }
6224 
6225 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6226 {
6227 #ifdef CONFIG_X86_64
6228 	struct desc_ptr dt;
6229 	struct kvm_segment seg;
6230 	unsigned long val;
6231 	int i;
6232 
6233 	for (i = 0; i < 16; i++)
6234 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6235 
6236 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6237 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6238 
6239 	kvm_get_dr(vcpu, 6, &val);
6240 	put_smstate(u64, buf, 0x7f68, val);
6241 	kvm_get_dr(vcpu, 7, &val);
6242 	put_smstate(u64, buf, 0x7f60, val);
6243 
6244 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6245 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6246 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6247 
6248 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6249 
6250 	/* revision id */
6251 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6252 
6253 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6254 
6255 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6256 	put_smstate(u16, buf, 0x7e90, seg.selector);
6257 	put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6258 	put_smstate(u32, buf, 0x7e94, seg.limit);
6259 	put_smstate(u64, buf, 0x7e98, seg.base);
6260 
6261 	kvm_x86_ops->get_idt(vcpu, &dt);
6262 	put_smstate(u32, buf, 0x7e84, dt.size);
6263 	put_smstate(u64, buf, 0x7e88, dt.address);
6264 
6265 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6266 	put_smstate(u16, buf, 0x7e70, seg.selector);
6267 	put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6268 	put_smstate(u32, buf, 0x7e74, seg.limit);
6269 	put_smstate(u64, buf, 0x7e78, seg.base);
6270 
6271 	kvm_x86_ops->get_gdt(vcpu, &dt);
6272 	put_smstate(u32, buf, 0x7e64, dt.size);
6273 	put_smstate(u64, buf, 0x7e68, dt.address);
6274 
6275 	for (i = 0; i < 6; i++)
6276 		process_smi_save_seg_64(vcpu, buf, i);
6277 #else
6278 	WARN_ON_ONCE(1);
6279 #endif
6280 }
6281 
6282 static void process_smi(struct kvm_vcpu *vcpu)
6283 {
6284 	struct kvm_segment cs, ds;
6285 	struct desc_ptr dt;
6286 	char buf[512];
6287 	u32 cr0;
6288 
6289 	if (is_smm(vcpu)) {
6290 		vcpu->arch.smi_pending = true;
6291 		return;
6292 	}
6293 
6294 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6295 	vcpu->arch.hflags |= HF_SMM_MASK;
6296 	memset(buf, 0, 512);
6297 	if (guest_cpuid_has_longmode(vcpu))
6298 		process_smi_save_state_64(vcpu, buf);
6299 	else
6300 		process_smi_save_state_32(vcpu, buf);
6301 
6302 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6303 
6304 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6305 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6306 	else
6307 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6308 
6309 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6310 	kvm_rip_write(vcpu, 0x8000);
6311 
6312 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6313 	kvm_x86_ops->set_cr0(vcpu, cr0);
6314 	vcpu->arch.cr0 = cr0;
6315 
6316 	kvm_x86_ops->set_cr4(vcpu, 0);
6317 
6318 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
6319 	dt.address = dt.size = 0;
6320 	kvm_x86_ops->set_idt(vcpu, &dt);
6321 
6322 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6323 
6324 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6325 	cs.base = vcpu->arch.smbase;
6326 
6327 	ds.selector = 0;
6328 	ds.base = 0;
6329 
6330 	cs.limit    = ds.limit = 0xffffffff;
6331 	cs.type     = ds.type = 0x3;
6332 	cs.dpl      = ds.dpl = 0;
6333 	cs.db       = ds.db = 0;
6334 	cs.s        = ds.s = 1;
6335 	cs.l        = ds.l = 0;
6336 	cs.g        = ds.g = 1;
6337 	cs.avl      = ds.avl = 0;
6338 	cs.present  = ds.present = 1;
6339 	cs.unusable = ds.unusable = 0;
6340 	cs.padding  = ds.padding = 0;
6341 
6342 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6343 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6344 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6345 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6346 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6347 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6348 
6349 	if (guest_cpuid_has_longmode(vcpu))
6350 		kvm_x86_ops->set_efer(vcpu, 0);
6351 
6352 	kvm_update_cpuid(vcpu);
6353 	kvm_mmu_reset_context(vcpu);
6354 }
6355 
6356 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6357 {
6358 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6359 }
6360 
6361 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6362 {
6363 	u64 eoi_exit_bitmap[4];
6364 
6365 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6366 		return;
6367 
6368 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6369 
6370 	if (irqchip_split(vcpu->kvm))
6371 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6372 	else {
6373 		if (vcpu->arch.apicv_active)
6374 			kvm_x86_ops->sync_pir_to_irr(vcpu);
6375 		kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6376 	}
6377 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6378 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
6379 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6380 }
6381 
6382 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6383 {
6384 	++vcpu->stat.tlb_flush;
6385 	kvm_x86_ops->tlb_flush(vcpu);
6386 }
6387 
6388 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6389 {
6390 	struct page *page = NULL;
6391 
6392 	if (!lapic_in_kernel(vcpu))
6393 		return;
6394 
6395 	if (!kvm_x86_ops->set_apic_access_page_addr)
6396 		return;
6397 
6398 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6399 	if (is_error_page(page))
6400 		return;
6401 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6402 
6403 	/*
6404 	 * Do not pin apic access page in memory, the MMU notifier
6405 	 * will call us again if it is migrated or swapped out.
6406 	 */
6407 	put_page(page);
6408 }
6409 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6410 
6411 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6412 					   unsigned long address)
6413 {
6414 	/*
6415 	 * The physical address of apic access page is stored in the VMCS.
6416 	 * Update it when it becomes invalid.
6417 	 */
6418 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6419 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6420 }
6421 
6422 /*
6423  * Returns 1 to let vcpu_run() continue the guest execution loop without
6424  * exiting to the userspace.  Otherwise, the value will be returned to the
6425  * userspace.
6426  */
6427 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6428 {
6429 	int r;
6430 	bool req_int_win =
6431 		dm_request_for_irq_injection(vcpu) &&
6432 		kvm_cpu_accept_dm_intr(vcpu);
6433 
6434 	bool req_immediate_exit = false;
6435 
6436 	if (vcpu->requests) {
6437 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6438 			kvm_mmu_unload(vcpu);
6439 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6440 			__kvm_migrate_timers(vcpu);
6441 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6442 			kvm_gen_update_masterclock(vcpu->kvm);
6443 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6444 			kvm_gen_kvmclock_update(vcpu);
6445 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6446 			r = kvm_guest_time_update(vcpu);
6447 			if (unlikely(r))
6448 				goto out;
6449 		}
6450 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6451 			kvm_mmu_sync_roots(vcpu);
6452 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6453 			kvm_vcpu_flush_tlb(vcpu);
6454 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6455 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6456 			r = 0;
6457 			goto out;
6458 		}
6459 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6460 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6461 			r = 0;
6462 			goto out;
6463 		}
6464 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6465 			vcpu->fpu_active = 0;
6466 			kvm_x86_ops->fpu_deactivate(vcpu);
6467 		}
6468 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6469 			/* Page is swapped out. Do synthetic halt */
6470 			vcpu->arch.apf.halted = true;
6471 			r = 1;
6472 			goto out;
6473 		}
6474 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6475 			record_steal_time(vcpu);
6476 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6477 			process_smi(vcpu);
6478 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6479 			process_nmi(vcpu);
6480 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6481 			kvm_pmu_handle_event(vcpu);
6482 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6483 			kvm_pmu_deliver_pmi(vcpu);
6484 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6485 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6486 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
6487 				     vcpu->arch.ioapic_handled_vectors)) {
6488 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6489 				vcpu->run->eoi.vector =
6490 						vcpu->arch.pending_ioapic_eoi;
6491 				r = 0;
6492 				goto out;
6493 			}
6494 		}
6495 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6496 			vcpu_scan_ioapic(vcpu);
6497 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6498 			kvm_vcpu_reload_apic_access_page(vcpu);
6499 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6500 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6501 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6502 			r = 0;
6503 			goto out;
6504 		}
6505 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6506 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6507 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6508 			r = 0;
6509 			goto out;
6510 		}
6511 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6512 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6513 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6514 			r = 0;
6515 			goto out;
6516 		}
6517 
6518 		/*
6519 		 * KVM_REQ_HV_STIMER has to be processed after
6520 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6521 		 * depend on the guest clock being up-to-date
6522 		 */
6523 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6524 			kvm_hv_process_stimers(vcpu);
6525 	}
6526 
6527 	/*
6528 	 * KVM_REQ_EVENT is not set when posted interrupts are set by
6529 	 * VT-d hardware, so we have to update RVI unconditionally.
6530 	 */
6531 	if (kvm_lapic_enabled(vcpu)) {
6532 		/*
6533 		 * Update architecture specific hints for APIC
6534 		 * virtual interrupt delivery.
6535 		 */
6536 		if (vcpu->arch.apicv_active)
6537 			kvm_x86_ops->hwapic_irr_update(vcpu,
6538 				kvm_lapic_find_highest_irr(vcpu));
6539 	}
6540 
6541 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6542 		kvm_apic_accept_events(vcpu);
6543 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6544 			r = 1;
6545 			goto out;
6546 		}
6547 
6548 		if (inject_pending_event(vcpu, req_int_win) != 0)
6549 			req_immediate_exit = true;
6550 		/* enable NMI/IRQ window open exits if needed */
6551 		else if (vcpu->arch.nmi_pending)
6552 			kvm_x86_ops->enable_nmi_window(vcpu);
6553 		else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6554 			kvm_x86_ops->enable_irq_window(vcpu);
6555 
6556 		if (kvm_lapic_enabled(vcpu)) {
6557 			update_cr8_intercept(vcpu);
6558 			kvm_lapic_sync_to_vapic(vcpu);
6559 		}
6560 	}
6561 
6562 	r = kvm_mmu_reload(vcpu);
6563 	if (unlikely(r)) {
6564 		goto cancel_injection;
6565 	}
6566 
6567 	preempt_disable();
6568 
6569 	kvm_x86_ops->prepare_guest_switch(vcpu);
6570 	if (vcpu->fpu_active)
6571 		kvm_load_guest_fpu(vcpu);
6572 	kvm_load_guest_xcr0(vcpu);
6573 
6574 	vcpu->mode = IN_GUEST_MODE;
6575 
6576 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6577 
6578 	/* We should set ->mode before check ->requests,
6579 	 * see the comment in make_all_cpus_request.
6580 	 */
6581 	smp_mb__after_srcu_read_unlock();
6582 
6583 	local_irq_disable();
6584 
6585 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6586 	    || need_resched() || signal_pending(current)) {
6587 		vcpu->mode = OUTSIDE_GUEST_MODE;
6588 		smp_wmb();
6589 		local_irq_enable();
6590 		preempt_enable();
6591 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6592 		r = 1;
6593 		goto cancel_injection;
6594 	}
6595 
6596 	if (req_immediate_exit)
6597 		smp_send_reschedule(vcpu->cpu);
6598 
6599 	trace_kvm_entry(vcpu->vcpu_id);
6600 	wait_lapic_expire(vcpu);
6601 	__kvm_guest_enter();
6602 
6603 	if (unlikely(vcpu->arch.switch_db_regs)) {
6604 		set_debugreg(0, 7);
6605 		set_debugreg(vcpu->arch.eff_db[0], 0);
6606 		set_debugreg(vcpu->arch.eff_db[1], 1);
6607 		set_debugreg(vcpu->arch.eff_db[2], 2);
6608 		set_debugreg(vcpu->arch.eff_db[3], 3);
6609 		set_debugreg(vcpu->arch.dr6, 6);
6610 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6611 	}
6612 
6613 	kvm_x86_ops->run(vcpu);
6614 
6615 	/*
6616 	 * Do this here before restoring debug registers on the host.  And
6617 	 * since we do this before handling the vmexit, a DR access vmexit
6618 	 * can (a) read the correct value of the debug registers, (b) set
6619 	 * KVM_DEBUGREG_WONT_EXIT again.
6620 	 */
6621 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6622 		int i;
6623 
6624 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6625 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6626 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6627 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6628 	}
6629 
6630 	/*
6631 	 * If the guest has used debug registers, at least dr7
6632 	 * will be disabled while returning to the host.
6633 	 * If we don't have active breakpoints in the host, we don't
6634 	 * care about the messed up debug address registers. But if
6635 	 * we have some of them active, restore the old state.
6636 	 */
6637 	if (hw_breakpoint_active())
6638 		hw_breakpoint_restore();
6639 
6640 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6641 
6642 	vcpu->mode = OUTSIDE_GUEST_MODE;
6643 	smp_wmb();
6644 
6645 	/* Interrupt is enabled by handle_external_intr() */
6646 	kvm_x86_ops->handle_external_intr(vcpu);
6647 
6648 	++vcpu->stat.exits;
6649 
6650 	/*
6651 	 * We must have an instruction between local_irq_enable() and
6652 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6653 	 * the interrupt shadow.  The stat.exits increment will do nicely.
6654 	 * But we need to prevent reordering, hence this barrier():
6655 	 */
6656 	barrier();
6657 
6658 	kvm_guest_exit();
6659 
6660 	preempt_enable();
6661 
6662 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6663 
6664 	/*
6665 	 * Profile KVM exit RIPs:
6666 	 */
6667 	if (unlikely(prof_on == KVM_PROFILING)) {
6668 		unsigned long rip = kvm_rip_read(vcpu);
6669 		profile_hit(KVM_PROFILING, (void *)rip);
6670 	}
6671 
6672 	if (unlikely(vcpu->arch.tsc_always_catchup))
6673 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6674 
6675 	if (vcpu->arch.apic_attention)
6676 		kvm_lapic_sync_from_vapic(vcpu);
6677 
6678 	r = kvm_x86_ops->handle_exit(vcpu);
6679 	return r;
6680 
6681 cancel_injection:
6682 	kvm_x86_ops->cancel_injection(vcpu);
6683 	if (unlikely(vcpu->arch.apic_attention))
6684 		kvm_lapic_sync_from_vapic(vcpu);
6685 out:
6686 	return r;
6687 }
6688 
6689 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6690 {
6691 	if (!kvm_arch_vcpu_runnable(vcpu) &&
6692 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6693 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6694 		kvm_vcpu_block(vcpu);
6695 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6696 
6697 		if (kvm_x86_ops->post_block)
6698 			kvm_x86_ops->post_block(vcpu);
6699 
6700 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6701 			return 1;
6702 	}
6703 
6704 	kvm_apic_accept_events(vcpu);
6705 	switch(vcpu->arch.mp_state) {
6706 	case KVM_MP_STATE_HALTED:
6707 		vcpu->arch.pv.pv_unhalted = false;
6708 		vcpu->arch.mp_state =
6709 			KVM_MP_STATE_RUNNABLE;
6710 	case KVM_MP_STATE_RUNNABLE:
6711 		vcpu->arch.apf.halted = false;
6712 		break;
6713 	case KVM_MP_STATE_INIT_RECEIVED:
6714 		break;
6715 	default:
6716 		return -EINTR;
6717 		break;
6718 	}
6719 	return 1;
6720 }
6721 
6722 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6723 {
6724 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6725 		!vcpu->arch.apf.halted);
6726 }
6727 
6728 static int vcpu_run(struct kvm_vcpu *vcpu)
6729 {
6730 	int r;
6731 	struct kvm *kvm = vcpu->kvm;
6732 
6733 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6734 
6735 	for (;;) {
6736 		if (kvm_vcpu_running(vcpu)) {
6737 			r = vcpu_enter_guest(vcpu);
6738 		} else {
6739 			r = vcpu_block(kvm, vcpu);
6740 		}
6741 
6742 		if (r <= 0)
6743 			break;
6744 
6745 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6746 		if (kvm_cpu_has_pending_timer(vcpu))
6747 			kvm_inject_pending_timer_irqs(vcpu);
6748 
6749 		if (dm_request_for_irq_injection(vcpu) &&
6750 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6751 			r = 0;
6752 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6753 			++vcpu->stat.request_irq_exits;
6754 			break;
6755 		}
6756 
6757 		kvm_check_async_pf_completion(vcpu);
6758 
6759 		if (signal_pending(current)) {
6760 			r = -EINTR;
6761 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6762 			++vcpu->stat.signal_exits;
6763 			break;
6764 		}
6765 		if (need_resched()) {
6766 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6767 			cond_resched();
6768 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6769 		}
6770 	}
6771 
6772 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6773 
6774 	return r;
6775 }
6776 
6777 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6778 {
6779 	int r;
6780 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6781 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6782 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6783 	if (r != EMULATE_DONE)
6784 		return 0;
6785 	return 1;
6786 }
6787 
6788 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6789 {
6790 	BUG_ON(!vcpu->arch.pio.count);
6791 
6792 	return complete_emulated_io(vcpu);
6793 }
6794 
6795 /*
6796  * Implements the following, as a state machine:
6797  *
6798  * read:
6799  *   for each fragment
6800  *     for each mmio piece in the fragment
6801  *       write gpa, len
6802  *       exit
6803  *       copy data
6804  *   execute insn
6805  *
6806  * write:
6807  *   for each fragment
6808  *     for each mmio piece in the fragment
6809  *       write gpa, len
6810  *       copy data
6811  *       exit
6812  */
6813 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6814 {
6815 	struct kvm_run *run = vcpu->run;
6816 	struct kvm_mmio_fragment *frag;
6817 	unsigned len;
6818 
6819 	BUG_ON(!vcpu->mmio_needed);
6820 
6821 	/* Complete previous fragment */
6822 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6823 	len = min(8u, frag->len);
6824 	if (!vcpu->mmio_is_write)
6825 		memcpy(frag->data, run->mmio.data, len);
6826 
6827 	if (frag->len <= 8) {
6828 		/* Switch to the next fragment. */
6829 		frag++;
6830 		vcpu->mmio_cur_fragment++;
6831 	} else {
6832 		/* Go forward to the next mmio piece. */
6833 		frag->data += len;
6834 		frag->gpa += len;
6835 		frag->len -= len;
6836 	}
6837 
6838 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6839 		vcpu->mmio_needed = 0;
6840 
6841 		/* FIXME: return into emulator if single-stepping.  */
6842 		if (vcpu->mmio_is_write)
6843 			return 1;
6844 		vcpu->mmio_read_completed = 1;
6845 		return complete_emulated_io(vcpu);
6846 	}
6847 
6848 	run->exit_reason = KVM_EXIT_MMIO;
6849 	run->mmio.phys_addr = frag->gpa;
6850 	if (vcpu->mmio_is_write)
6851 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6852 	run->mmio.len = min(8u, frag->len);
6853 	run->mmio.is_write = vcpu->mmio_is_write;
6854 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6855 	return 0;
6856 }
6857 
6858 
6859 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6860 {
6861 	struct fpu *fpu = &current->thread.fpu;
6862 	int r;
6863 	sigset_t sigsaved;
6864 
6865 	fpu__activate_curr(fpu);
6866 
6867 	if (vcpu->sigset_active)
6868 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6869 
6870 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6871 		kvm_vcpu_block(vcpu);
6872 		kvm_apic_accept_events(vcpu);
6873 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6874 		r = -EAGAIN;
6875 		goto out;
6876 	}
6877 
6878 	/* re-sync apic's tpr */
6879 	if (!lapic_in_kernel(vcpu)) {
6880 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6881 			r = -EINVAL;
6882 			goto out;
6883 		}
6884 	}
6885 
6886 	if (unlikely(vcpu->arch.complete_userspace_io)) {
6887 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6888 		vcpu->arch.complete_userspace_io = NULL;
6889 		r = cui(vcpu);
6890 		if (r <= 0)
6891 			goto out;
6892 	} else
6893 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6894 
6895 	r = vcpu_run(vcpu);
6896 
6897 out:
6898 	post_kvm_run_save(vcpu);
6899 	if (vcpu->sigset_active)
6900 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6901 
6902 	return r;
6903 }
6904 
6905 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6906 {
6907 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6908 		/*
6909 		 * We are here if userspace calls get_regs() in the middle of
6910 		 * instruction emulation. Registers state needs to be copied
6911 		 * back from emulation context to vcpu. Userspace shouldn't do
6912 		 * that usually, but some bad designed PV devices (vmware
6913 		 * backdoor interface) need this to work
6914 		 */
6915 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6916 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6917 	}
6918 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6919 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6920 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6921 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6922 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6923 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6924 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6925 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6926 #ifdef CONFIG_X86_64
6927 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6928 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6929 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6930 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6931 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6932 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6933 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6934 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6935 #endif
6936 
6937 	regs->rip = kvm_rip_read(vcpu);
6938 	regs->rflags = kvm_get_rflags(vcpu);
6939 
6940 	return 0;
6941 }
6942 
6943 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6944 {
6945 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6946 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6947 
6948 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6949 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6950 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6951 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6952 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6953 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6954 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6955 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6956 #ifdef CONFIG_X86_64
6957 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6958 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6959 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6960 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6961 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6962 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6963 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6964 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6965 #endif
6966 
6967 	kvm_rip_write(vcpu, regs->rip);
6968 	kvm_set_rflags(vcpu, regs->rflags);
6969 
6970 	vcpu->arch.exception.pending = false;
6971 
6972 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6973 
6974 	return 0;
6975 }
6976 
6977 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6978 {
6979 	struct kvm_segment cs;
6980 
6981 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6982 	*db = cs.db;
6983 	*l = cs.l;
6984 }
6985 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6986 
6987 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6988 				  struct kvm_sregs *sregs)
6989 {
6990 	struct desc_ptr dt;
6991 
6992 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6993 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6994 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6995 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6996 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6997 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6998 
6999 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7000 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7001 
7002 	kvm_x86_ops->get_idt(vcpu, &dt);
7003 	sregs->idt.limit = dt.size;
7004 	sregs->idt.base = dt.address;
7005 	kvm_x86_ops->get_gdt(vcpu, &dt);
7006 	sregs->gdt.limit = dt.size;
7007 	sregs->gdt.base = dt.address;
7008 
7009 	sregs->cr0 = kvm_read_cr0(vcpu);
7010 	sregs->cr2 = vcpu->arch.cr2;
7011 	sregs->cr3 = kvm_read_cr3(vcpu);
7012 	sregs->cr4 = kvm_read_cr4(vcpu);
7013 	sregs->cr8 = kvm_get_cr8(vcpu);
7014 	sregs->efer = vcpu->arch.efer;
7015 	sregs->apic_base = kvm_get_apic_base(vcpu);
7016 
7017 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7018 
7019 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7020 		set_bit(vcpu->arch.interrupt.nr,
7021 			(unsigned long *)sregs->interrupt_bitmap);
7022 
7023 	return 0;
7024 }
7025 
7026 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7027 				    struct kvm_mp_state *mp_state)
7028 {
7029 	kvm_apic_accept_events(vcpu);
7030 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7031 					vcpu->arch.pv.pv_unhalted)
7032 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7033 	else
7034 		mp_state->mp_state = vcpu->arch.mp_state;
7035 
7036 	return 0;
7037 }
7038 
7039 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7040 				    struct kvm_mp_state *mp_state)
7041 {
7042 	if (!kvm_vcpu_has_lapic(vcpu) &&
7043 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7044 		return -EINVAL;
7045 
7046 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7047 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7048 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7049 	} else
7050 		vcpu->arch.mp_state = mp_state->mp_state;
7051 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7052 	return 0;
7053 }
7054 
7055 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7056 		    int reason, bool has_error_code, u32 error_code)
7057 {
7058 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7059 	int ret;
7060 
7061 	init_emulate_ctxt(vcpu);
7062 
7063 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7064 				   has_error_code, error_code);
7065 
7066 	if (ret)
7067 		return EMULATE_FAIL;
7068 
7069 	kvm_rip_write(vcpu, ctxt->eip);
7070 	kvm_set_rflags(vcpu, ctxt->eflags);
7071 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7072 	return EMULATE_DONE;
7073 }
7074 EXPORT_SYMBOL_GPL(kvm_task_switch);
7075 
7076 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7077 				  struct kvm_sregs *sregs)
7078 {
7079 	struct msr_data apic_base_msr;
7080 	int mmu_reset_needed = 0;
7081 	int pending_vec, max_bits, idx;
7082 	struct desc_ptr dt;
7083 
7084 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7085 		return -EINVAL;
7086 
7087 	dt.size = sregs->idt.limit;
7088 	dt.address = sregs->idt.base;
7089 	kvm_x86_ops->set_idt(vcpu, &dt);
7090 	dt.size = sregs->gdt.limit;
7091 	dt.address = sregs->gdt.base;
7092 	kvm_x86_ops->set_gdt(vcpu, &dt);
7093 
7094 	vcpu->arch.cr2 = sregs->cr2;
7095 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7096 	vcpu->arch.cr3 = sregs->cr3;
7097 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7098 
7099 	kvm_set_cr8(vcpu, sregs->cr8);
7100 
7101 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7102 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7103 	apic_base_msr.data = sregs->apic_base;
7104 	apic_base_msr.host_initiated = true;
7105 	kvm_set_apic_base(vcpu, &apic_base_msr);
7106 
7107 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7108 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7109 	vcpu->arch.cr0 = sregs->cr0;
7110 
7111 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7112 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7113 	if (sregs->cr4 & X86_CR4_OSXSAVE)
7114 		kvm_update_cpuid(vcpu);
7115 
7116 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7117 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7118 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7119 		mmu_reset_needed = 1;
7120 	}
7121 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7122 
7123 	if (mmu_reset_needed)
7124 		kvm_mmu_reset_context(vcpu);
7125 
7126 	max_bits = KVM_NR_INTERRUPTS;
7127 	pending_vec = find_first_bit(
7128 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7129 	if (pending_vec < max_bits) {
7130 		kvm_queue_interrupt(vcpu, pending_vec, false);
7131 		pr_debug("Set back pending irq %d\n", pending_vec);
7132 	}
7133 
7134 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7135 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7136 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7137 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7138 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7139 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7140 
7141 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7142 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7143 
7144 	update_cr8_intercept(vcpu);
7145 
7146 	/* Older userspace won't unhalt the vcpu on reset. */
7147 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7148 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7149 	    !is_protmode(vcpu))
7150 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7151 
7152 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7153 
7154 	return 0;
7155 }
7156 
7157 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7158 					struct kvm_guest_debug *dbg)
7159 {
7160 	unsigned long rflags;
7161 	int i, r;
7162 
7163 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7164 		r = -EBUSY;
7165 		if (vcpu->arch.exception.pending)
7166 			goto out;
7167 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7168 			kvm_queue_exception(vcpu, DB_VECTOR);
7169 		else
7170 			kvm_queue_exception(vcpu, BP_VECTOR);
7171 	}
7172 
7173 	/*
7174 	 * Read rflags as long as potentially injected trace flags are still
7175 	 * filtered out.
7176 	 */
7177 	rflags = kvm_get_rflags(vcpu);
7178 
7179 	vcpu->guest_debug = dbg->control;
7180 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7181 		vcpu->guest_debug = 0;
7182 
7183 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7184 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7185 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7186 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7187 	} else {
7188 		for (i = 0; i < KVM_NR_DB_REGS; i++)
7189 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7190 	}
7191 	kvm_update_dr7(vcpu);
7192 
7193 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7194 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7195 			get_segment_base(vcpu, VCPU_SREG_CS);
7196 
7197 	/*
7198 	 * Trigger an rflags update that will inject or remove the trace
7199 	 * flags.
7200 	 */
7201 	kvm_set_rflags(vcpu, rflags);
7202 
7203 	kvm_x86_ops->update_bp_intercept(vcpu);
7204 
7205 	r = 0;
7206 
7207 out:
7208 
7209 	return r;
7210 }
7211 
7212 /*
7213  * Translate a guest virtual address to a guest physical address.
7214  */
7215 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7216 				    struct kvm_translation *tr)
7217 {
7218 	unsigned long vaddr = tr->linear_address;
7219 	gpa_t gpa;
7220 	int idx;
7221 
7222 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7223 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7224 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7225 	tr->physical_address = gpa;
7226 	tr->valid = gpa != UNMAPPED_GVA;
7227 	tr->writeable = 1;
7228 	tr->usermode = 0;
7229 
7230 	return 0;
7231 }
7232 
7233 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7234 {
7235 	struct fxregs_state *fxsave =
7236 			&vcpu->arch.guest_fpu.state.fxsave;
7237 
7238 	memcpy(fpu->fpr, fxsave->st_space, 128);
7239 	fpu->fcw = fxsave->cwd;
7240 	fpu->fsw = fxsave->swd;
7241 	fpu->ftwx = fxsave->twd;
7242 	fpu->last_opcode = fxsave->fop;
7243 	fpu->last_ip = fxsave->rip;
7244 	fpu->last_dp = fxsave->rdp;
7245 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7246 
7247 	return 0;
7248 }
7249 
7250 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7251 {
7252 	struct fxregs_state *fxsave =
7253 			&vcpu->arch.guest_fpu.state.fxsave;
7254 
7255 	memcpy(fxsave->st_space, fpu->fpr, 128);
7256 	fxsave->cwd = fpu->fcw;
7257 	fxsave->swd = fpu->fsw;
7258 	fxsave->twd = fpu->ftwx;
7259 	fxsave->fop = fpu->last_opcode;
7260 	fxsave->rip = fpu->last_ip;
7261 	fxsave->rdp = fpu->last_dp;
7262 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7263 
7264 	return 0;
7265 }
7266 
7267 static void fx_init(struct kvm_vcpu *vcpu)
7268 {
7269 	fpstate_init(&vcpu->arch.guest_fpu.state);
7270 	if (cpu_has_xsaves)
7271 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7272 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7273 
7274 	/*
7275 	 * Ensure guest xcr0 is valid for loading
7276 	 */
7277 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7278 
7279 	vcpu->arch.cr0 |= X86_CR0_ET;
7280 }
7281 
7282 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7283 {
7284 	if (vcpu->guest_fpu_loaded)
7285 		return;
7286 
7287 	/*
7288 	 * Restore all possible states in the guest,
7289 	 * and assume host would use all available bits.
7290 	 * Guest xcr0 would be loaded later.
7291 	 */
7292 	kvm_put_guest_xcr0(vcpu);
7293 	vcpu->guest_fpu_loaded = 1;
7294 	__kernel_fpu_begin();
7295 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7296 	trace_kvm_fpu(1);
7297 }
7298 
7299 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7300 {
7301 	kvm_put_guest_xcr0(vcpu);
7302 
7303 	if (!vcpu->guest_fpu_loaded) {
7304 		vcpu->fpu_counter = 0;
7305 		return;
7306 	}
7307 
7308 	vcpu->guest_fpu_loaded = 0;
7309 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7310 	__kernel_fpu_end();
7311 	++vcpu->stat.fpu_reload;
7312 	/*
7313 	 * If using eager FPU mode, or if the guest is a frequent user
7314 	 * of the FPU, just leave the FPU active for next time.
7315 	 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7316 	 * the FPU in bursts will revert to loading it on demand.
7317 	 */
7318 	if (!vcpu->arch.eager_fpu) {
7319 		if (++vcpu->fpu_counter < 5)
7320 			kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7321 	}
7322 	trace_kvm_fpu(0);
7323 }
7324 
7325 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7326 {
7327 	kvmclock_reset(vcpu);
7328 
7329 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7330 	kvm_x86_ops->vcpu_free(vcpu);
7331 }
7332 
7333 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7334 						unsigned int id)
7335 {
7336 	struct kvm_vcpu *vcpu;
7337 
7338 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7339 		printk_once(KERN_WARNING
7340 		"kvm: SMP vm created on host with unstable TSC; "
7341 		"guest TSC will not be reliable\n");
7342 
7343 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7344 
7345 	return vcpu;
7346 }
7347 
7348 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7349 {
7350 	int r;
7351 
7352 	kvm_vcpu_mtrr_init(vcpu);
7353 	r = vcpu_load(vcpu);
7354 	if (r)
7355 		return r;
7356 	kvm_vcpu_reset(vcpu, false);
7357 	kvm_mmu_setup(vcpu);
7358 	vcpu_put(vcpu);
7359 	return r;
7360 }
7361 
7362 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7363 {
7364 	struct msr_data msr;
7365 	struct kvm *kvm = vcpu->kvm;
7366 
7367 	if (vcpu_load(vcpu))
7368 		return;
7369 	msr.data = 0x0;
7370 	msr.index = MSR_IA32_TSC;
7371 	msr.host_initiated = true;
7372 	kvm_write_tsc(vcpu, &msr);
7373 	vcpu_put(vcpu);
7374 
7375 	if (!kvmclock_periodic_sync)
7376 		return;
7377 
7378 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7379 					KVMCLOCK_SYNC_PERIOD);
7380 }
7381 
7382 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7383 {
7384 	int r;
7385 	vcpu->arch.apf.msr_val = 0;
7386 
7387 	r = vcpu_load(vcpu);
7388 	BUG_ON(r);
7389 	kvm_mmu_unload(vcpu);
7390 	vcpu_put(vcpu);
7391 
7392 	kvm_x86_ops->vcpu_free(vcpu);
7393 }
7394 
7395 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7396 {
7397 	vcpu->arch.hflags = 0;
7398 
7399 	atomic_set(&vcpu->arch.nmi_queued, 0);
7400 	vcpu->arch.nmi_pending = 0;
7401 	vcpu->arch.nmi_injected = false;
7402 	kvm_clear_interrupt_queue(vcpu);
7403 	kvm_clear_exception_queue(vcpu);
7404 
7405 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7406 	kvm_update_dr0123(vcpu);
7407 	vcpu->arch.dr6 = DR6_INIT;
7408 	kvm_update_dr6(vcpu);
7409 	vcpu->arch.dr7 = DR7_FIXED_1;
7410 	kvm_update_dr7(vcpu);
7411 
7412 	vcpu->arch.cr2 = 0;
7413 
7414 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7415 	vcpu->arch.apf.msr_val = 0;
7416 	vcpu->arch.st.msr_val = 0;
7417 
7418 	kvmclock_reset(vcpu);
7419 
7420 	kvm_clear_async_pf_completion_queue(vcpu);
7421 	kvm_async_pf_hash_reset(vcpu);
7422 	vcpu->arch.apf.halted = false;
7423 
7424 	if (!init_event) {
7425 		kvm_pmu_reset(vcpu);
7426 		vcpu->arch.smbase = 0x30000;
7427 	}
7428 
7429 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7430 	vcpu->arch.regs_avail = ~0;
7431 	vcpu->arch.regs_dirty = ~0;
7432 
7433 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7434 }
7435 
7436 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7437 {
7438 	struct kvm_segment cs;
7439 
7440 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7441 	cs.selector = vector << 8;
7442 	cs.base = vector << 12;
7443 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7444 	kvm_rip_write(vcpu, 0);
7445 }
7446 
7447 int kvm_arch_hardware_enable(void)
7448 {
7449 	struct kvm *kvm;
7450 	struct kvm_vcpu *vcpu;
7451 	int i;
7452 	int ret;
7453 	u64 local_tsc;
7454 	u64 max_tsc = 0;
7455 	bool stable, backwards_tsc = false;
7456 
7457 	kvm_shared_msr_cpu_online();
7458 	ret = kvm_x86_ops->hardware_enable();
7459 	if (ret != 0)
7460 		return ret;
7461 
7462 	local_tsc = rdtsc();
7463 	stable = !check_tsc_unstable();
7464 	list_for_each_entry(kvm, &vm_list, vm_list) {
7465 		kvm_for_each_vcpu(i, vcpu, kvm) {
7466 			if (!stable && vcpu->cpu == smp_processor_id())
7467 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7468 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7469 				backwards_tsc = true;
7470 				if (vcpu->arch.last_host_tsc > max_tsc)
7471 					max_tsc = vcpu->arch.last_host_tsc;
7472 			}
7473 		}
7474 	}
7475 
7476 	/*
7477 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7478 	 * platforms that reset TSC during suspend or hibernate actions, but
7479 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7480 	 * detect that condition here, which happens early in CPU bringup,
7481 	 * before any KVM threads can be running.  Unfortunately, we can't
7482 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7483 	 * enough into CPU bringup that we know how much real time has actually
7484 	 * elapsed; our helper function, get_kernel_ns() will be using boot
7485 	 * variables that haven't been updated yet.
7486 	 *
7487 	 * So we simply find the maximum observed TSC above, then record the
7488 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7489 	 * the adjustment will be applied.  Note that we accumulate
7490 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7491 	 * gets a chance to run again.  In the event that no KVM threads get a
7492 	 * chance to run, we will miss the entire elapsed period, as we'll have
7493 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7494 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7495 	 * uniform across all VCPUs (not to mention the scenario is extremely
7496 	 * unlikely). It is possible that a second hibernate recovery happens
7497 	 * much faster than a first, causing the observed TSC here to be
7498 	 * smaller; this would require additional padding adjustment, which is
7499 	 * why we set last_host_tsc to the local tsc observed here.
7500 	 *
7501 	 * N.B. - this code below runs only on platforms with reliable TSC,
7502 	 * as that is the only way backwards_tsc is set above.  Also note
7503 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7504 	 * have the same delta_cyc adjustment applied if backwards_tsc
7505 	 * is detected.  Note further, this adjustment is only done once,
7506 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7507 	 * called multiple times (one for each physical CPU bringup).
7508 	 *
7509 	 * Platforms with unreliable TSCs don't have to deal with this, they
7510 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7511 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7512 	 * guarantee that they stay in perfect synchronization.
7513 	 */
7514 	if (backwards_tsc) {
7515 		u64 delta_cyc = max_tsc - local_tsc;
7516 		backwards_tsc_observed = true;
7517 		list_for_each_entry(kvm, &vm_list, vm_list) {
7518 			kvm_for_each_vcpu(i, vcpu, kvm) {
7519 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7520 				vcpu->arch.last_host_tsc = local_tsc;
7521 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7522 			}
7523 
7524 			/*
7525 			 * We have to disable TSC offset matching.. if you were
7526 			 * booting a VM while issuing an S4 host suspend....
7527 			 * you may have some problem.  Solving this issue is
7528 			 * left as an exercise to the reader.
7529 			 */
7530 			kvm->arch.last_tsc_nsec = 0;
7531 			kvm->arch.last_tsc_write = 0;
7532 		}
7533 
7534 	}
7535 	return 0;
7536 }
7537 
7538 void kvm_arch_hardware_disable(void)
7539 {
7540 	kvm_x86_ops->hardware_disable();
7541 	drop_user_return_notifiers();
7542 }
7543 
7544 int kvm_arch_hardware_setup(void)
7545 {
7546 	int r;
7547 
7548 	r = kvm_x86_ops->hardware_setup();
7549 	if (r != 0)
7550 		return r;
7551 
7552 	if (kvm_has_tsc_control) {
7553 		/*
7554 		 * Make sure the user can only configure tsc_khz values that
7555 		 * fit into a signed integer.
7556 		 * A min value is not calculated needed because it will always
7557 		 * be 1 on all machines.
7558 		 */
7559 		u64 max = min(0x7fffffffULL,
7560 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7561 		kvm_max_guest_tsc_khz = max;
7562 
7563 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7564 	}
7565 
7566 	kvm_init_msr_list();
7567 	return 0;
7568 }
7569 
7570 void kvm_arch_hardware_unsetup(void)
7571 {
7572 	kvm_x86_ops->hardware_unsetup();
7573 }
7574 
7575 void kvm_arch_check_processor_compat(void *rtn)
7576 {
7577 	kvm_x86_ops->check_processor_compatibility(rtn);
7578 }
7579 
7580 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7581 {
7582 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7583 }
7584 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7585 
7586 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7587 {
7588 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7589 }
7590 
7591 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7592 {
7593 	return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7594 }
7595 
7596 struct static_key kvm_no_apic_vcpu __read_mostly;
7597 
7598 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7599 {
7600 	struct page *page;
7601 	struct kvm *kvm;
7602 	int r;
7603 
7604 	BUG_ON(vcpu->kvm == NULL);
7605 	kvm = vcpu->kvm;
7606 
7607 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7608 	vcpu->arch.pv.pv_unhalted = false;
7609 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7610 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7611 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7612 	else
7613 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7614 
7615 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7616 	if (!page) {
7617 		r = -ENOMEM;
7618 		goto fail;
7619 	}
7620 	vcpu->arch.pio_data = page_address(page);
7621 
7622 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7623 
7624 	r = kvm_mmu_create(vcpu);
7625 	if (r < 0)
7626 		goto fail_free_pio_data;
7627 
7628 	if (irqchip_in_kernel(kvm)) {
7629 		r = kvm_create_lapic(vcpu);
7630 		if (r < 0)
7631 			goto fail_mmu_destroy;
7632 	} else
7633 		static_key_slow_inc(&kvm_no_apic_vcpu);
7634 
7635 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7636 				       GFP_KERNEL);
7637 	if (!vcpu->arch.mce_banks) {
7638 		r = -ENOMEM;
7639 		goto fail_free_lapic;
7640 	}
7641 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7642 
7643 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7644 		r = -ENOMEM;
7645 		goto fail_free_mce_banks;
7646 	}
7647 
7648 	fx_init(vcpu);
7649 
7650 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7651 	vcpu->arch.pv_time_enabled = false;
7652 
7653 	vcpu->arch.guest_supported_xcr0 = 0;
7654 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7655 
7656 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7657 
7658 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7659 
7660 	kvm_async_pf_hash_reset(vcpu);
7661 	kvm_pmu_init(vcpu);
7662 
7663 	vcpu->arch.pending_external_vector = -1;
7664 
7665 	kvm_hv_vcpu_init(vcpu);
7666 
7667 	return 0;
7668 
7669 fail_free_mce_banks:
7670 	kfree(vcpu->arch.mce_banks);
7671 fail_free_lapic:
7672 	kvm_free_lapic(vcpu);
7673 fail_mmu_destroy:
7674 	kvm_mmu_destroy(vcpu);
7675 fail_free_pio_data:
7676 	free_page((unsigned long)vcpu->arch.pio_data);
7677 fail:
7678 	return r;
7679 }
7680 
7681 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7682 {
7683 	int idx;
7684 
7685 	kvm_hv_vcpu_uninit(vcpu);
7686 	kvm_pmu_destroy(vcpu);
7687 	kfree(vcpu->arch.mce_banks);
7688 	kvm_free_lapic(vcpu);
7689 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7690 	kvm_mmu_destroy(vcpu);
7691 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7692 	free_page((unsigned long)vcpu->arch.pio_data);
7693 	if (!lapic_in_kernel(vcpu))
7694 		static_key_slow_dec(&kvm_no_apic_vcpu);
7695 }
7696 
7697 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7698 {
7699 	kvm_x86_ops->sched_in(vcpu, cpu);
7700 }
7701 
7702 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7703 {
7704 	if (type)
7705 		return -EINVAL;
7706 
7707 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7708 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7709 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7710 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7711 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7712 
7713 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7714 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7715 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7716 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7717 		&kvm->arch.irq_sources_bitmap);
7718 
7719 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7720 	mutex_init(&kvm->arch.apic_map_lock);
7721 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7722 
7723 	pvclock_update_vm_gtod_copy(kvm);
7724 
7725 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7726 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7727 
7728 	return 0;
7729 }
7730 
7731 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7732 {
7733 	int r;
7734 	r = vcpu_load(vcpu);
7735 	BUG_ON(r);
7736 	kvm_mmu_unload(vcpu);
7737 	vcpu_put(vcpu);
7738 }
7739 
7740 static void kvm_free_vcpus(struct kvm *kvm)
7741 {
7742 	unsigned int i;
7743 	struct kvm_vcpu *vcpu;
7744 
7745 	/*
7746 	 * Unpin any mmu pages first.
7747 	 */
7748 	kvm_for_each_vcpu(i, vcpu, kvm) {
7749 		kvm_clear_async_pf_completion_queue(vcpu);
7750 		kvm_unload_vcpu_mmu(vcpu);
7751 	}
7752 	kvm_for_each_vcpu(i, vcpu, kvm)
7753 		kvm_arch_vcpu_free(vcpu);
7754 
7755 	mutex_lock(&kvm->lock);
7756 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7757 		kvm->vcpus[i] = NULL;
7758 
7759 	atomic_set(&kvm->online_vcpus, 0);
7760 	mutex_unlock(&kvm->lock);
7761 }
7762 
7763 void kvm_arch_sync_events(struct kvm *kvm)
7764 {
7765 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7766 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7767 	kvm_free_all_assigned_devices(kvm);
7768 	kvm_free_pit(kvm);
7769 }
7770 
7771 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7772 {
7773 	int i, r;
7774 	unsigned long hva;
7775 	struct kvm_memslots *slots = kvm_memslots(kvm);
7776 	struct kvm_memory_slot *slot, old;
7777 
7778 	/* Called with kvm->slots_lock held.  */
7779 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7780 		return -EINVAL;
7781 
7782 	slot = id_to_memslot(slots, id);
7783 	if (size) {
7784 		if (WARN_ON(slot->npages))
7785 			return -EEXIST;
7786 
7787 		/*
7788 		 * MAP_SHARED to prevent internal slot pages from being moved
7789 		 * by fork()/COW.
7790 		 */
7791 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7792 			      MAP_SHARED | MAP_ANONYMOUS, 0);
7793 		if (IS_ERR((void *)hva))
7794 			return PTR_ERR((void *)hva);
7795 	} else {
7796 		if (!slot->npages)
7797 			return 0;
7798 
7799 		hva = 0;
7800 	}
7801 
7802 	old = *slot;
7803 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7804 		struct kvm_userspace_memory_region m;
7805 
7806 		m.slot = id | (i << 16);
7807 		m.flags = 0;
7808 		m.guest_phys_addr = gpa;
7809 		m.userspace_addr = hva;
7810 		m.memory_size = size;
7811 		r = __kvm_set_memory_region(kvm, &m);
7812 		if (r < 0)
7813 			return r;
7814 	}
7815 
7816 	if (!size) {
7817 		r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7818 		WARN_ON(r < 0);
7819 	}
7820 
7821 	return 0;
7822 }
7823 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7824 
7825 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7826 {
7827 	int r;
7828 
7829 	mutex_lock(&kvm->slots_lock);
7830 	r = __x86_set_memory_region(kvm, id, gpa, size);
7831 	mutex_unlock(&kvm->slots_lock);
7832 
7833 	return r;
7834 }
7835 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7836 
7837 void kvm_arch_destroy_vm(struct kvm *kvm)
7838 {
7839 	if (current->mm == kvm->mm) {
7840 		/*
7841 		 * Free memory regions allocated on behalf of userspace,
7842 		 * unless the the memory map has changed due to process exit
7843 		 * or fd copying.
7844 		 */
7845 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7846 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7847 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7848 	}
7849 	kvm_iommu_unmap_guest(kvm);
7850 	kfree(kvm->arch.vpic);
7851 	kfree(kvm->arch.vioapic);
7852 	kvm_free_vcpus(kvm);
7853 	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7854 }
7855 
7856 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7857 			   struct kvm_memory_slot *dont)
7858 {
7859 	int i;
7860 
7861 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7862 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7863 			kvfree(free->arch.rmap[i]);
7864 			free->arch.rmap[i] = NULL;
7865 		}
7866 		if (i == 0)
7867 			continue;
7868 
7869 		if (!dont || free->arch.lpage_info[i - 1] !=
7870 			     dont->arch.lpage_info[i - 1]) {
7871 			kvfree(free->arch.lpage_info[i - 1]);
7872 			free->arch.lpage_info[i - 1] = NULL;
7873 		}
7874 	}
7875 }
7876 
7877 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7878 			    unsigned long npages)
7879 {
7880 	int i;
7881 
7882 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7883 		unsigned long ugfn;
7884 		int lpages;
7885 		int level = i + 1;
7886 
7887 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7888 				      slot->base_gfn, level) + 1;
7889 
7890 		slot->arch.rmap[i] =
7891 			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7892 		if (!slot->arch.rmap[i])
7893 			goto out_free;
7894 		if (i == 0)
7895 			continue;
7896 
7897 		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7898 					sizeof(*slot->arch.lpage_info[i - 1]));
7899 		if (!slot->arch.lpage_info[i - 1])
7900 			goto out_free;
7901 
7902 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7903 			slot->arch.lpage_info[i - 1][0].write_count = 1;
7904 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7905 			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7906 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7907 		/*
7908 		 * If the gfn and userspace address are not aligned wrt each
7909 		 * other, or if explicitly asked to, disable large page
7910 		 * support for this slot
7911 		 */
7912 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7913 		    !kvm_largepages_enabled()) {
7914 			unsigned long j;
7915 
7916 			for (j = 0; j < lpages; ++j)
7917 				slot->arch.lpage_info[i - 1][j].write_count = 1;
7918 		}
7919 	}
7920 
7921 	return 0;
7922 
7923 out_free:
7924 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7925 		kvfree(slot->arch.rmap[i]);
7926 		slot->arch.rmap[i] = NULL;
7927 		if (i == 0)
7928 			continue;
7929 
7930 		kvfree(slot->arch.lpage_info[i - 1]);
7931 		slot->arch.lpage_info[i - 1] = NULL;
7932 	}
7933 	return -ENOMEM;
7934 }
7935 
7936 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7937 {
7938 	/*
7939 	 * memslots->generation has been incremented.
7940 	 * mmio generation may have reached its maximum value.
7941 	 */
7942 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7943 }
7944 
7945 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7946 				struct kvm_memory_slot *memslot,
7947 				const struct kvm_userspace_memory_region *mem,
7948 				enum kvm_mr_change change)
7949 {
7950 	return 0;
7951 }
7952 
7953 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7954 				     struct kvm_memory_slot *new)
7955 {
7956 	/* Still write protect RO slot */
7957 	if (new->flags & KVM_MEM_READONLY) {
7958 		kvm_mmu_slot_remove_write_access(kvm, new);
7959 		return;
7960 	}
7961 
7962 	/*
7963 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
7964 	 *
7965 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
7966 	 *
7967 	 *  - KVM_MR_CREATE with dirty logging is disabled
7968 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7969 	 *
7970 	 * The reason is, in case of PML, we need to set D-bit for any slots
7971 	 * with dirty logging disabled in order to eliminate unnecessary GPA
7972 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
7973 	 * guarantees leaving PML enabled during guest's lifetime won't have
7974 	 * any additonal overhead from PML when guest is running with dirty
7975 	 * logging disabled for memory slots.
7976 	 *
7977 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7978 	 * to dirty logging mode.
7979 	 *
7980 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7981 	 *
7982 	 * In case of write protect:
7983 	 *
7984 	 * Write protect all pages for dirty logging.
7985 	 *
7986 	 * All the sptes including the large sptes which point to this
7987 	 * slot are set to readonly. We can not create any new large
7988 	 * spte on this slot until the end of the logging.
7989 	 *
7990 	 * See the comments in fast_page_fault().
7991 	 */
7992 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7993 		if (kvm_x86_ops->slot_enable_log_dirty)
7994 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7995 		else
7996 			kvm_mmu_slot_remove_write_access(kvm, new);
7997 	} else {
7998 		if (kvm_x86_ops->slot_disable_log_dirty)
7999 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8000 	}
8001 }
8002 
8003 void kvm_arch_commit_memory_region(struct kvm *kvm,
8004 				const struct kvm_userspace_memory_region *mem,
8005 				const struct kvm_memory_slot *old,
8006 				const struct kvm_memory_slot *new,
8007 				enum kvm_mr_change change)
8008 {
8009 	int nr_mmu_pages = 0;
8010 
8011 	if (!kvm->arch.n_requested_mmu_pages)
8012 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8013 
8014 	if (nr_mmu_pages)
8015 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8016 
8017 	/*
8018 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
8019 	 * sptes have to be split.  If live migration is successful, the guest
8020 	 * in the source machine will be destroyed and large sptes will be
8021 	 * created in the destination. However, if the guest continues to run
8022 	 * in the source machine (for example if live migration fails), small
8023 	 * sptes will remain around and cause bad performance.
8024 	 *
8025 	 * Scan sptes if dirty logging has been stopped, dropping those
8026 	 * which can be collapsed into a single large-page spte.  Later
8027 	 * page faults will create the large-page sptes.
8028 	 */
8029 	if ((change != KVM_MR_DELETE) &&
8030 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8031 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8032 		kvm_mmu_zap_collapsible_sptes(kvm, new);
8033 
8034 	/*
8035 	 * Set up write protection and/or dirty logging for the new slot.
8036 	 *
8037 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8038 	 * been zapped so no dirty logging staff is needed for old slot. For
8039 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8040 	 * new and it's also covered when dealing with the new slot.
8041 	 *
8042 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
8043 	 */
8044 	if (change != KVM_MR_DELETE)
8045 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8046 }
8047 
8048 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8049 {
8050 	kvm_mmu_invalidate_zap_all_pages(kvm);
8051 }
8052 
8053 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8054 				   struct kvm_memory_slot *slot)
8055 {
8056 	kvm_mmu_invalidate_zap_all_pages(kvm);
8057 }
8058 
8059 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8060 {
8061 	if (!list_empty_careful(&vcpu->async_pf.done))
8062 		return true;
8063 
8064 	if (kvm_apic_has_events(vcpu))
8065 		return true;
8066 
8067 	if (vcpu->arch.pv.pv_unhalted)
8068 		return true;
8069 
8070 	if (atomic_read(&vcpu->arch.nmi_queued))
8071 		return true;
8072 
8073 	if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8074 		return true;
8075 
8076 	if (kvm_arch_interrupt_allowed(vcpu) &&
8077 	    kvm_cpu_has_interrupt(vcpu))
8078 		return true;
8079 
8080 	if (kvm_hv_has_stimer_pending(vcpu))
8081 		return true;
8082 
8083 	return false;
8084 }
8085 
8086 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8087 {
8088 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8089 		kvm_x86_ops->check_nested_events(vcpu, false);
8090 
8091 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8092 }
8093 
8094 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8095 {
8096 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8097 }
8098 
8099 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8100 {
8101 	return kvm_x86_ops->interrupt_allowed(vcpu);
8102 }
8103 
8104 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8105 {
8106 	if (is_64_bit_mode(vcpu))
8107 		return kvm_rip_read(vcpu);
8108 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8109 		     kvm_rip_read(vcpu));
8110 }
8111 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8112 
8113 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8114 {
8115 	return kvm_get_linear_rip(vcpu) == linear_rip;
8116 }
8117 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8118 
8119 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8120 {
8121 	unsigned long rflags;
8122 
8123 	rflags = kvm_x86_ops->get_rflags(vcpu);
8124 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8125 		rflags &= ~X86_EFLAGS_TF;
8126 	return rflags;
8127 }
8128 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8129 
8130 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8131 {
8132 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8133 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8134 		rflags |= X86_EFLAGS_TF;
8135 	kvm_x86_ops->set_rflags(vcpu, rflags);
8136 }
8137 
8138 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8139 {
8140 	__kvm_set_rflags(vcpu, rflags);
8141 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8142 }
8143 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8144 
8145 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8146 {
8147 	int r;
8148 
8149 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8150 	      work->wakeup_all)
8151 		return;
8152 
8153 	r = kvm_mmu_reload(vcpu);
8154 	if (unlikely(r))
8155 		return;
8156 
8157 	if (!vcpu->arch.mmu.direct_map &&
8158 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8159 		return;
8160 
8161 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8162 }
8163 
8164 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8165 {
8166 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8167 }
8168 
8169 static inline u32 kvm_async_pf_next_probe(u32 key)
8170 {
8171 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8172 }
8173 
8174 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8175 {
8176 	u32 key = kvm_async_pf_hash_fn(gfn);
8177 
8178 	while (vcpu->arch.apf.gfns[key] != ~0)
8179 		key = kvm_async_pf_next_probe(key);
8180 
8181 	vcpu->arch.apf.gfns[key] = gfn;
8182 }
8183 
8184 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8185 {
8186 	int i;
8187 	u32 key = kvm_async_pf_hash_fn(gfn);
8188 
8189 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8190 		     (vcpu->arch.apf.gfns[key] != gfn &&
8191 		      vcpu->arch.apf.gfns[key] != ~0); i++)
8192 		key = kvm_async_pf_next_probe(key);
8193 
8194 	return key;
8195 }
8196 
8197 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8198 {
8199 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8200 }
8201 
8202 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8203 {
8204 	u32 i, j, k;
8205 
8206 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8207 	while (true) {
8208 		vcpu->arch.apf.gfns[i] = ~0;
8209 		do {
8210 			j = kvm_async_pf_next_probe(j);
8211 			if (vcpu->arch.apf.gfns[j] == ~0)
8212 				return;
8213 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8214 			/*
8215 			 * k lies cyclically in ]i,j]
8216 			 * |    i.k.j |
8217 			 * |....j i.k.| or  |.k..j i...|
8218 			 */
8219 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8220 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8221 		i = j;
8222 	}
8223 }
8224 
8225 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8226 {
8227 
8228 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8229 				      sizeof(val));
8230 }
8231 
8232 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8233 				     struct kvm_async_pf *work)
8234 {
8235 	struct x86_exception fault;
8236 
8237 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8238 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8239 
8240 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8241 	    (vcpu->arch.apf.send_user_only &&
8242 	     kvm_x86_ops->get_cpl(vcpu) == 0))
8243 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8244 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8245 		fault.vector = PF_VECTOR;
8246 		fault.error_code_valid = true;
8247 		fault.error_code = 0;
8248 		fault.nested_page_fault = false;
8249 		fault.address = work->arch.token;
8250 		kvm_inject_page_fault(vcpu, &fault);
8251 	}
8252 }
8253 
8254 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8255 				 struct kvm_async_pf *work)
8256 {
8257 	struct x86_exception fault;
8258 
8259 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8260 	if (work->wakeup_all)
8261 		work->arch.token = ~0; /* broadcast wakeup */
8262 	else
8263 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8264 
8265 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8266 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8267 		fault.vector = PF_VECTOR;
8268 		fault.error_code_valid = true;
8269 		fault.error_code = 0;
8270 		fault.nested_page_fault = false;
8271 		fault.address = work->arch.token;
8272 		kvm_inject_page_fault(vcpu, &fault);
8273 	}
8274 	vcpu->arch.apf.halted = false;
8275 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8276 }
8277 
8278 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8279 {
8280 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8281 		return true;
8282 	else
8283 		return !kvm_event_needs_reinjection(vcpu) &&
8284 			kvm_x86_ops->interrupt_allowed(vcpu);
8285 }
8286 
8287 void kvm_arch_start_assignment(struct kvm *kvm)
8288 {
8289 	atomic_inc(&kvm->arch.assigned_device_count);
8290 }
8291 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8292 
8293 void kvm_arch_end_assignment(struct kvm *kvm)
8294 {
8295 	atomic_dec(&kvm->arch.assigned_device_count);
8296 }
8297 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8298 
8299 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8300 {
8301 	return atomic_read(&kvm->arch.assigned_device_count);
8302 }
8303 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8304 
8305 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8306 {
8307 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8308 }
8309 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8310 
8311 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8312 {
8313 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8314 }
8315 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8316 
8317 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8318 {
8319 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8320 }
8321 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8322 
8323 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8324 				      struct irq_bypass_producer *prod)
8325 {
8326 	struct kvm_kernel_irqfd *irqfd =
8327 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8328 
8329 	if (kvm_x86_ops->update_pi_irte) {
8330 		irqfd->producer = prod;
8331 		return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8332 				prod->irq, irqfd->gsi, 1);
8333 	}
8334 
8335 	return -EINVAL;
8336 }
8337 
8338 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8339 				      struct irq_bypass_producer *prod)
8340 {
8341 	int ret;
8342 	struct kvm_kernel_irqfd *irqfd =
8343 		container_of(cons, struct kvm_kernel_irqfd, consumer);
8344 
8345 	if (!kvm_x86_ops->update_pi_irte) {
8346 		WARN_ON(irqfd->producer != NULL);
8347 		return;
8348 	}
8349 
8350 	WARN_ON(irqfd->producer != prod);
8351 	irqfd->producer = NULL;
8352 
8353 	/*
8354 	 * When producer of consumer is unregistered, we change back to
8355 	 * remapped mode, so we can re-use the current implementation
8356 	 * when the irq is masked/disabed or the consumer side (KVM
8357 	 * int this case doesn't want to receive the interrupts.
8358 	*/
8359 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8360 	if (ret)
8361 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8362 		       " fails: %d\n", irqfd->consumer.token, ret);
8363 }
8364 
8365 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8366 				   uint32_t guest_irq, bool set)
8367 {
8368 	if (!kvm_x86_ops->update_pi_irte)
8369 		return -EINVAL;
8370 
8371 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8372 }
8373 
8374 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8375 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8376 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8377 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8378 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8379 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8380 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8381 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8382 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8383 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8384 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8385 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8386 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8387 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8388 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8389 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8390 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8391