xref: /linux/arch/x86/kvm/x86.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
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 
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
33 #include <linux/fs.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <trace/events/kvm.h>
48 
49 #define CREATE_TRACE_POINTS
50 #include "trace.h"
51 
52 #include <asm/debugreg.h>
53 #include <asm/msr.h>
54 #include <asm/desc.h>
55 #include <asm/mtrr.h>
56 #include <asm/mce.h>
57 #include <asm/i387.h>
58 #include <asm/xcr.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
61 
62 #define MAX_IO_MSRS 256
63 #define KVM_MAX_MCE_BANKS 32
64 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
65 
66 #define emul_to_vcpu(ctxt) \
67 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
68 
69 /* EFER defaults:
70  * - enable syscall per default because its emulated by KVM
71  * - enable LME and LMA per default on 64 bit KVM
72  */
73 #ifdef CONFIG_X86_64
74 static
75 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
76 #else
77 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
78 #endif
79 
80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
82 
83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
85 				    struct kvm_cpuid_entry2 __user *entries);
86 
87 struct kvm_x86_ops *kvm_x86_ops;
88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
89 
90 int ignore_msrs = 0;
91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
92 
93 bool kvm_has_tsc_control;
94 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
95 u32  kvm_max_guest_tsc_khz;
96 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
97 
98 #define KVM_NR_SHARED_MSRS 16
99 
100 struct kvm_shared_msrs_global {
101 	int nr;
102 	u32 msrs[KVM_NR_SHARED_MSRS];
103 };
104 
105 struct kvm_shared_msrs {
106 	struct user_return_notifier urn;
107 	bool registered;
108 	struct kvm_shared_msr_values {
109 		u64 host;
110 		u64 curr;
111 	} values[KVM_NR_SHARED_MSRS];
112 };
113 
114 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
115 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
116 
117 struct kvm_stats_debugfs_item debugfs_entries[] = {
118 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
119 	{ "pf_guest", VCPU_STAT(pf_guest) },
120 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
121 	{ "invlpg", VCPU_STAT(invlpg) },
122 	{ "exits", VCPU_STAT(exits) },
123 	{ "io_exits", VCPU_STAT(io_exits) },
124 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
125 	{ "signal_exits", VCPU_STAT(signal_exits) },
126 	{ "irq_window", VCPU_STAT(irq_window_exits) },
127 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
128 	{ "halt_exits", VCPU_STAT(halt_exits) },
129 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
130 	{ "hypercalls", VCPU_STAT(hypercalls) },
131 	{ "request_irq", VCPU_STAT(request_irq_exits) },
132 	{ "irq_exits", VCPU_STAT(irq_exits) },
133 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
134 	{ "efer_reload", VCPU_STAT(efer_reload) },
135 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
136 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
137 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
138 	{ "irq_injections", VCPU_STAT(irq_injections) },
139 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
140 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
141 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
142 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
143 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
144 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
145 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
146 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
147 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
148 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
149 	{ "largepages", VM_STAT(lpages) },
150 	{ NULL }
151 };
152 
153 u64 __read_mostly host_xcr0;
154 
155 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
156 
157 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
158 {
159 	int i;
160 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
161 		vcpu->arch.apf.gfns[i] = ~0;
162 }
163 
164 static void kvm_on_user_return(struct user_return_notifier *urn)
165 {
166 	unsigned slot;
167 	struct kvm_shared_msrs *locals
168 		= container_of(urn, struct kvm_shared_msrs, urn);
169 	struct kvm_shared_msr_values *values;
170 
171 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
172 		values = &locals->values[slot];
173 		if (values->host != values->curr) {
174 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
175 			values->curr = values->host;
176 		}
177 	}
178 	locals->registered = false;
179 	user_return_notifier_unregister(urn);
180 }
181 
182 static void shared_msr_update(unsigned slot, u32 msr)
183 {
184 	struct kvm_shared_msrs *smsr;
185 	u64 value;
186 
187 	smsr = &__get_cpu_var(shared_msrs);
188 	/* only read, and nobody should modify it at this time,
189 	 * so don't need lock */
190 	if (slot >= shared_msrs_global.nr) {
191 		printk(KERN_ERR "kvm: invalid MSR slot!");
192 		return;
193 	}
194 	rdmsrl_safe(msr, &value);
195 	smsr->values[slot].host = value;
196 	smsr->values[slot].curr = value;
197 }
198 
199 void kvm_define_shared_msr(unsigned slot, u32 msr)
200 {
201 	if (slot >= shared_msrs_global.nr)
202 		shared_msrs_global.nr = slot + 1;
203 	shared_msrs_global.msrs[slot] = msr;
204 	/* we need ensured the shared_msr_global have been updated */
205 	smp_wmb();
206 }
207 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
208 
209 static void kvm_shared_msr_cpu_online(void)
210 {
211 	unsigned i;
212 
213 	for (i = 0; i < shared_msrs_global.nr; ++i)
214 		shared_msr_update(i, shared_msrs_global.msrs[i]);
215 }
216 
217 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
218 {
219 	struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
220 
221 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
222 		return;
223 	smsr->values[slot].curr = value;
224 	wrmsrl(shared_msrs_global.msrs[slot], value);
225 	if (!smsr->registered) {
226 		smsr->urn.on_user_return = kvm_on_user_return;
227 		user_return_notifier_register(&smsr->urn);
228 		smsr->registered = true;
229 	}
230 }
231 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
232 
233 static void drop_user_return_notifiers(void *ignore)
234 {
235 	struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
236 
237 	if (smsr->registered)
238 		kvm_on_user_return(&smsr->urn);
239 }
240 
241 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
242 {
243 	if (irqchip_in_kernel(vcpu->kvm))
244 		return vcpu->arch.apic_base;
245 	else
246 		return vcpu->arch.apic_base;
247 }
248 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
249 
250 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
251 {
252 	/* TODO: reserve bits check */
253 	if (irqchip_in_kernel(vcpu->kvm))
254 		kvm_lapic_set_base(vcpu, data);
255 	else
256 		vcpu->arch.apic_base = data;
257 }
258 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
259 
260 #define EXCPT_BENIGN		0
261 #define EXCPT_CONTRIBUTORY	1
262 #define EXCPT_PF		2
263 
264 static int exception_class(int vector)
265 {
266 	switch (vector) {
267 	case PF_VECTOR:
268 		return EXCPT_PF;
269 	case DE_VECTOR:
270 	case TS_VECTOR:
271 	case NP_VECTOR:
272 	case SS_VECTOR:
273 	case GP_VECTOR:
274 		return EXCPT_CONTRIBUTORY;
275 	default:
276 		break;
277 	}
278 	return EXCPT_BENIGN;
279 }
280 
281 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
282 		unsigned nr, bool has_error, u32 error_code,
283 		bool reinject)
284 {
285 	u32 prev_nr;
286 	int class1, class2;
287 
288 	kvm_make_request(KVM_REQ_EVENT, vcpu);
289 
290 	if (!vcpu->arch.exception.pending) {
291 	queue:
292 		vcpu->arch.exception.pending = true;
293 		vcpu->arch.exception.has_error_code = has_error;
294 		vcpu->arch.exception.nr = nr;
295 		vcpu->arch.exception.error_code = error_code;
296 		vcpu->arch.exception.reinject = reinject;
297 		return;
298 	}
299 
300 	/* to check exception */
301 	prev_nr = vcpu->arch.exception.nr;
302 	if (prev_nr == DF_VECTOR) {
303 		/* triple fault -> shutdown */
304 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
305 		return;
306 	}
307 	class1 = exception_class(prev_nr);
308 	class2 = exception_class(nr);
309 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
310 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
311 		/* generate double fault per SDM Table 5-5 */
312 		vcpu->arch.exception.pending = true;
313 		vcpu->arch.exception.has_error_code = true;
314 		vcpu->arch.exception.nr = DF_VECTOR;
315 		vcpu->arch.exception.error_code = 0;
316 	} else
317 		/* replace previous exception with a new one in a hope
318 		   that instruction re-execution will regenerate lost
319 		   exception */
320 		goto queue;
321 }
322 
323 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
324 {
325 	kvm_multiple_exception(vcpu, nr, false, 0, false);
326 }
327 EXPORT_SYMBOL_GPL(kvm_queue_exception);
328 
329 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330 {
331 	kvm_multiple_exception(vcpu, nr, false, 0, true);
332 }
333 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
334 
335 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
336 {
337 	if (err)
338 		kvm_inject_gp(vcpu, 0);
339 	else
340 		kvm_x86_ops->skip_emulated_instruction(vcpu);
341 }
342 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
343 
344 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
345 {
346 	++vcpu->stat.pf_guest;
347 	vcpu->arch.cr2 = fault->address;
348 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
349 }
350 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
351 
352 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
353 {
354 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
355 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
356 	else
357 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
358 }
359 
360 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
361 {
362 	kvm_make_request(KVM_REQ_EVENT, vcpu);
363 	vcpu->arch.nmi_pending = 1;
364 }
365 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
366 
367 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
368 {
369 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
370 }
371 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
372 
373 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
374 {
375 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
376 }
377 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
378 
379 /*
380  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
381  * a #GP and return false.
382  */
383 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
384 {
385 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
386 		return true;
387 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
388 	return false;
389 }
390 EXPORT_SYMBOL_GPL(kvm_require_cpl);
391 
392 /*
393  * This function will be used to read from the physical memory of the currently
394  * running guest. The difference to kvm_read_guest_page is that this function
395  * can read from guest physical or from the guest's guest physical memory.
396  */
397 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
398 			    gfn_t ngfn, void *data, int offset, int len,
399 			    u32 access)
400 {
401 	gfn_t real_gfn;
402 	gpa_t ngpa;
403 
404 	ngpa     = gfn_to_gpa(ngfn);
405 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
406 	if (real_gfn == UNMAPPED_GVA)
407 		return -EFAULT;
408 
409 	real_gfn = gpa_to_gfn(real_gfn);
410 
411 	return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
412 }
413 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
414 
415 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
416 			       void *data, int offset, int len, u32 access)
417 {
418 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
419 				       data, offset, len, access);
420 }
421 
422 /*
423  * Load the pae pdptrs.  Return true is they are all valid.
424  */
425 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
426 {
427 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
428 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
429 	int i;
430 	int ret;
431 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
432 
433 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
434 				      offset * sizeof(u64), sizeof(pdpte),
435 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
436 	if (ret < 0) {
437 		ret = 0;
438 		goto out;
439 	}
440 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
441 		if (is_present_gpte(pdpte[i]) &&
442 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
443 			ret = 0;
444 			goto out;
445 		}
446 	}
447 	ret = 1;
448 
449 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
450 	__set_bit(VCPU_EXREG_PDPTR,
451 		  (unsigned long *)&vcpu->arch.regs_avail);
452 	__set_bit(VCPU_EXREG_PDPTR,
453 		  (unsigned long *)&vcpu->arch.regs_dirty);
454 out:
455 
456 	return ret;
457 }
458 EXPORT_SYMBOL_GPL(load_pdptrs);
459 
460 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
461 {
462 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
463 	bool changed = true;
464 	int offset;
465 	gfn_t gfn;
466 	int r;
467 
468 	if (is_long_mode(vcpu) || !is_pae(vcpu))
469 		return false;
470 
471 	if (!test_bit(VCPU_EXREG_PDPTR,
472 		      (unsigned long *)&vcpu->arch.regs_avail))
473 		return true;
474 
475 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
476 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
477 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
478 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
479 	if (r < 0)
480 		goto out;
481 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
482 out:
483 
484 	return changed;
485 }
486 
487 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
488 {
489 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
490 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
491 				    X86_CR0_CD | X86_CR0_NW;
492 
493 	cr0 |= X86_CR0_ET;
494 
495 #ifdef CONFIG_X86_64
496 	if (cr0 & 0xffffffff00000000UL)
497 		return 1;
498 #endif
499 
500 	cr0 &= ~CR0_RESERVED_BITS;
501 
502 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
503 		return 1;
504 
505 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
506 		return 1;
507 
508 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
509 #ifdef CONFIG_X86_64
510 		if ((vcpu->arch.efer & EFER_LME)) {
511 			int cs_db, cs_l;
512 
513 			if (!is_pae(vcpu))
514 				return 1;
515 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
516 			if (cs_l)
517 				return 1;
518 		} else
519 #endif
520 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
521 						 kvm_read_cr3(vcpu)))
522 			return 1;
523 	}
524 
525 	kvm_x86_ops->set_cr0(vcpu, cr0);
526 
527 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
528 		kvm_clear_async_pf_completion_queue(vcpu);
529 		kvm_async_pf_hash_reset(vcpu);
530 	}
531 
532 	if ((cr0 ^ old_cr0) & update_bits)
533 		kvm_mmu_reset_context(vcpu);
534 	return 0;
535 }
536 EXPORT_SYMBOL_GPL(kvm_set_cr0);
537 
538 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
539 {
540 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
541 }
542 EXPORT_SYMBOL_GPL(kvm_lmsw);
543 
544 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
545 {
546 	u64 xcr0;
547 
548 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
549 	if (index != XCR_XFEATURE_ENABLED_MASK)
550 		return 1;
551 	xcr0 = xcr;
552 	if (kvm_x86_ops->get_cpl(vcpu) != 0)
553 		return 1;
554 	if (!(xcr0 & XSTATE_FP))
555 		return 1;
556 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
557 		return 1;
558 	if (xcr0 & ~host_xcr0)
559 		return 1;
560 	vcpu->arch.xcr0 = xcr0;
561 	vcpu->guest_xcr0_loaded = 0;
562 	return 0;
563 }
564 
565 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
566 {
567 	if (__kvm_set_xcr(vcpu, index, xcr)) {
568 		kvm_inject_gp(vcpu, 0);
569 		return 1;
570 	}
571 	return 0;
572 }
573 EXPORT_SYMBOL_GPL(kvm_set_xcr);
574 
575 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
576 {
577 	struct kvm_cpuid_entry2 *best;
578 
579 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
580 	return best && (best->ecx & bit(X86_FEATURE_XSAVE));
581 }
582 
583 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
584 {
585 	struct kvm_cpuid_entry2 *best;
586 
587 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
588 	return best && (best->ebx & bit(X86_FEATURE_SMEP));
589 }
590 
591 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
592 {
593 	struct kvm_cpuid_entry2 *best;
594 
595 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
596 	return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
597 }
598 
599 static void update_cpuid(struct kvm_vcpu *vcpu)
600 {
601 	struct kvm_cpuid_entry2 *best;
602 
603 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
604 	if (!best)
605 		return;
606 
607 	/* Update OSXSAVE bit */
608 	if (cpu_has_xsave && best->function == 0x1) {
609 		best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
610 		if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
611 			best->ecx |= bit(X86_FEATURE_OSXSAVE);
612 	}
613 }
614 
615 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
616 {
617 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
618 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
619 				   X86_CR4_PAE | X86_CR4_SMEP;
620 	if (cr4 & CR4_RESERVED_BITS)
621 		return 1;
622 
623 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
624 		return 1;
625 
626 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
627 		return 1;
628 
629 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
630 		return 1;
631 
632 	if (is_long_mode(vcpu)) {
633 		if (!(cr4 & X86_CR4_PAE))
634 			return 1;
635 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
636 		   && ((cr4 ^ old_cr4) & pdptr_bits)
637 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
638 				   kvm_read_cr3(vcpu)))
639 		return 1;
640 
641 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
642 		return 1;
643 
644 	if ((cr4 ^ old_cr4) & pdptr_bits)
645 		kvm_mmu_reset_context(vcpu);
646 
647 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
648 		update_cpuid(vcpu);
649 
650 	return 0;
651 }
652 EXPORT_SYMBOL_GPL(kvm_set_cr4);
653 
654 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
655 {
656 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
657 		kvm_mmu_sync_roots(vcpu);
658 		kvm_mmu_flush_tlb(vcpu);
659 		return 0;
660 	}
661 
662 	if (is_long_mode(vcpu)) {
663 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
664 			return 1;
665 	} else {
666 		if (is_pae(vcpu)) {
667 			if (cr3 & CR3_PAE_RESERVED_BITS)
668 				return 1;
669 			if (is_paging(vcpu) &&
670 			    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
671 				return 1;
672 		}
673 		/*
674 		 * We don't check reserved bits in nonpae mode, because
675 		 * this isn't enforced, and VMware depends on this.
676 		 */
677 	}
678 
679 	/*
680 	 * Does the new cr3 value map to physical memory? (Note, we
681 	 * catch an invalid cr3 even in real-mode, because it would
682 	 * cause trouble later on when we turn on paging anyway.)
683 	 *
684 	 * A real CPU would silently accept an invalid cr3 and would
685 	 * attempt to use it - with largely undefined (and often hard
686 	 * to debug) behavior on the guest side.
687 	 */
688 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
689 		return 1;
690 	vcpu->arch.cr3 = cr3;
691 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
692 	vcpu->arch.mmu.new_cr3(vcpu);
693 	return 0;
694 }
695 EXPORT_SYMBOL_GPL(kvm_set_cr3);
696 
697 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
698 {
699 	if (cr8 & CR8_RESERVED_BITS)
700 		return 1;
701 	if (irqchip_in_kernel(vcpu->kvm))
702 		kvm_lapic_set_tpr(vcpu, cr8);
703 	else
704 		vcpu->arch.cr8 = cr8;
705 	return 0;
706 }
707 EXPORT_SYMBOL_GPL(kvm_set_cr8);
708 
709 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
710 {
711 	if (irqchip_in_kernel(vcpu->kvm))
712 		return kvm_lapic_get_cr8(vcpu);
713 	else
714 		return vcpu->arch.cr8;
715 }
716 EXPORT_SYMBOL_GPL(kvm_get_cr8);
717 
718 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
719 {
720 	switch (dr) {
721 	case 0 ... 3:
722 		vcpu->arch.db[dr] = val;
723 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
724 			vcpu->arch.eff_db[dr] = val;
725 		break;
726 	case 4:
727 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
728 			return 1; /* #UD */
729 		/* fall through */
730 	case 6:
731 		if (val & 0xffffffff00000000ULL)
732 			return -1; /* #GP */
733 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
734 		break;
735 	case 5:
736 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
737 			return 1; /* #UD */
738 		/* fall through */
739 	default: /* 7 */
740 		if (val & 0xffffffff00000000ULL)
741 			return -1; /* #GP */
742 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
743 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
744 			kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
745 			vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
746 		}
747 		break;
748 	}
749 
750 	return 0;
751 }
752 
753 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
754 {
755 	int res;
756 
757 	res = __kvm_set_dr(vcpu, dr, val);
758 	if (res > 0)
759 		kvm_queue_exception(vcpu, UD_VECTOR);
760 	else if (res < 0)
761 		kvm_inject_gp(vcpu, 0);
762 
763 	return res;
764 }
765 EXPORT_SYMBOL_GPL(kvm_set_dr);
766 
767 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
768 {
769 	switch (dr) {
770 	case 0 ... 3:
771 		*val = vcpu->arch.db[dr];
772 		break;
773 	case 4:
774 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
775 			return 1;
776 		/* fall through */
777 	case 6:
778 		*val = vcpu->arch.dr6;
779 		break;
780 	case 5:
781 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
782 			return 1;
783 		/* fall through */
784 	default: /* 7 */
785 		*val = vcpu->arch.dr7;
786 		break;
787 	}
788 
789 	return 0;
790 }
791 
792 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
793 {
794 	if (_kvm_get_dr(vcpu, dr, val)) {
795 		kvm_queue_exception(vcpu, UD_VECTOR);
796 		return 1;
797 	}
798 	return 0;
799 }
800 EXPORT_SYMBOL_GPL(kvm_get_dr);
801 
802 /*
803  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
804  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
805  *
806  * This list is modified at module load time to reflect the
807  * capabilities of the host cpu. This capabilities test skips MSRs that are
808  * kvm-specific. Those are put in the beginning of the list.
809  */
810 
811 #define KVM_SAVE_MSRS_BEGIN	9
812 static u32 msrs_to_save[] = {
813 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
814 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
815 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
816 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
817 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
818 	MSR_STAR,
819 #ifdef CONFIG_X86_64
820 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
821 #endif
822 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
823 };
824 
825 static unsigned num_msrs_to_save;
826 
827 static u32 emulated_msrs[] = {
828 	MSR_IA32_MISC_ENABLE,
829 	MSR_IA32_MCG_STATUS,
830 	MSR_IA32_MCG_CTL,
831 };
832 
833 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
834 {
835 	u64 old_efer = vcpu->arch.efer;
836 
837 	if (efer & efer_reserved_bits)
838 		return 1;
839 
840 	if (is_paging(vcpu)
841 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
842 		return 1;
843 
844 	if (efer & EFER_FFXSR) {
845 		struct kvm_cpuid_entry2 *feat;
846 
847 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
848 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
849 			return 1;
850 	}
851 
852 	if (efer & EFER_SVME) {
853 		struct kvm_cpuid_entry2 *feat;
854 
855 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
856 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
857 			return 1;
858 	}
859 
860 	efer &= ~EFER_LMA;
861 	efer |= vcpu->arch.efer & EFER_LMA;
862 
863 	kvm_x86_ops->set_efer(vcpu, efer);
864 
865 	vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
866 
867 	/* Update reserved bits */
868 	if ((efer ^ old_efer) & EFER_NX)
869 		kvm_mmu_reset_context(vcpu);
870 
871 	return 0;
872 }
873 
874 void kvm_enable_efer_bits(u64 mask)
875 {
876        efer_reserved_bits &= ~mask;
877 }
878 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
879 
880 
881 /*
882  * Writes msr value into into the appropriate "register".
883  * Returns 0 on success, non-0 otherwise.
884  * Assumes vcpu_load() was already called.
885  */
886 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
887 {
888 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
889 }
890 
891 /*
892  * Adapt set_msr() to msr_io()'s calling convention
893  */
894 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
895 {
896 	return kvm_set_msr(vcpu, index, *data);
897 }
898 
899 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
900 {
901 	int version;
902 	int r;
903 	struct pvclock_wall_clock wc;
904 	struct timespec boot;
905 
906 	if (!wall_clock)
907 		return;
908 
909 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
910 	if (r)
911 		return;
912 
913 	if (version & 1)
914 		++version;  /* first time write, random junk */
915 
916 	++version;
917 
918 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
919 
920 	/*
921 	 * The guest calculates current wall clock time by adding
922 	 * system time (updated by kvm_guest_time_update below) to the
923 	 * wall clock specified here.  guest system time equals host
924 	 * system time for us, thus we must fill in host boot time here.
925 	 */
926 	getboottime(&boot);
927 
928 	wc.sec = boot.tv_sec;
929 	wc.nsec = boot.tv_nsec;
930 	wc.version = version;
931 
932 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
933 
934 	version++;
935 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
936 }
937 
938 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
939 {
940 	uint32_t quotient, remainder;
941 
942 	/* Don't try to replace with do_div(), this one calculates
943 	 * "(dividend << 32) / divisor" */
944 	__asm__ ( "divl %4"
945 		  : "=a" (quotient), "=d" (remainder)
946 		  : "0" (0), "1" (dividend), "r" (divisor) );
947 	return quotient;
948 }
949 
950 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
951 			       s8 *pshift, u32 *pmultiplier)
952 {
953 	uint64_t scaled64;
954 	int32_t  shift = 0;
955 	uint64_t tps64;
956 	uint32_t tps32;
957 
958 	tps64 = base_khz * 1000LL;
959 	scaled64 = scaled_khz * 1000LL;
960 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
961 		tps64 >>= 1;
962 		shift--;
963 	}
964 
965 	tps32 = (uint32_t)tps64;
966 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
967 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
968 			scaled64 >>= 1;
969 		else
970 			tps32 <<= 1;
971 		shift++;
972 	}
973 
974 	*pshift = shift;
975 	*pmultiplier = div_frac(scaled64, tps32);
976 
977 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
978 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
979 }
980 
981 static inline u64 get_kernel_ns(void)
982 {
983 	struct timespec ts;
984 
985 	WARN_ON(preemptible());
986 	ktime_get_ts(&ts);
987 	monotonic_to_bootbased(&ts);
988 	return timespec_to_ns(&ts);
989 }
990 
991 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
992 unsigned long max_tsc_khz;
993 
994 static inline int kvm_tsc_changes_freq(void)
995 {
996 	int cpu = get_cpu();
997 	int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
998 		  cpufreq_quick_get(cpu) != 0;
999 	put_cpu();
1000 	return ret;
1001 }
1002 
1003 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1004 {
1005 	if (vcpu->arch.virtual_tsc_khz)
1006 		return vcpu->arch.virtual_tsc_khz;
1007 	else
1008 		return __this_cpu_read(cpu_tsc_khz);
1009 }
1010 
1011 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1012 {
1013 	u64 ret;
1014 
1015 	WARN_ON(preemptible());
1016 	if (kvm_tsc_changes_freq())
1017 		printk_once(KERN_WARNING
1018 		 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1019 	ret = nsec * vcpu_tsc_khz(vcpu);
1020 	do_div(ret, USEC_PER_SEC);
1021 	return ret;
1022 }
1023 
1024 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1025 {
1026 	/* Compute a scale to convert nanoseconds in TSC cycles */
1027 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1028 			   &vcpu->arch.tsc_catchup_shift,
1029 			   &vcpu->arch.tsc_catchup_mult);
1030 }
1031 
1032 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1033 {
1034 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1035 				      vcpu->arch.tsc_catchup_mult,
1036 				      vcpu->arch.tsc_catchup_shift);
1037 	tsc += vcpu->arch.last_tsc_write;
1038 	return tsc;
1039 }
1040 
1041 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1042 {
1043 	struct kvm *kvm = vcpu->kvm;
1044 	u64 offset, ns, elapsed;
1045 	unsigned long flags;
1046 	s64 sdiff;
1047 
1048 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1049 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1050 	ns = get_kernel_ns();
1051 	elapsed = ns - kvm->arch.last_tsc_nsec;
1052 	sdiff = data - kvm->arch.last_tsc_write;
1053 	if (sdiff < 0)
1054 		sdiff = -sdiff;
1055 
1056 	/*
1057 	 * Special case: close write to TSC within 5 seconds of
1058 	 * another CPU is interpreted as an attempt to synchronize
1059 	 * The 5 seconds is to accommodate host load / swapping as
1060 	 * well as any reset of TSC during the boot process.
1061 	 *
1062 	 * In that case, for a reliable TSC, we can match TSC offsets,
1063 	 * or make a best guest using elapsed value.
1064 	 */
1065 	if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1066 	    elapsed < 5ULL * NSEC_PER_SEC) {
1067 		if (!check_tsc_unstable()) {
1068 			offset = kvm->arch.last_tsc_offset;
1069 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1070 		} else {
1071 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1072 			offset += delta;
1073 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1074 		}
1075 		ns = kvm->arch.last_tsc_nsec;
1076 	}
1077 	kvm->arch.last_tsc_nsec = ns;
1078 	kvm->arch.last_tsc_write = data;
1079 	kvm->arch.last_tsc_offset = offset;
1080 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1081 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1082 
1083 	/* Reset of TSC must disable overshoot protection below */
1084 	vcpu->arch.hv_clock.tsc_timestamp = 0;
1085 	vcpu->arch.last_tsc_write = data;
1086 	vcpu->arch.last_tsc_nsec = ns;
1087 }
1088 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1089 
1090 static int kvm_guest_time_update(struct kvm_vcpu *v)
1091 {
1092 	unsigned long flags;
1093 	struct kvm_vcpu_arch *vcpu = &v->arch;
1094 	void *shared_kaddr;
1095 	unsigned long this_tsc_khz;
1096 	s64 kernel_ns, max_kernel_ns;
1097 	u64 tsc_timestamp;
1098 
1099 	/* Keep irq disabled to prevent changes to the clock */
1100 	local_irq_save(flags);
1101 	kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1102 	kernel_ns = get_kernel_ns();
1103 	this_tsc_khz = vcpu_tsc_khz(v);
1104 	if (unlikely(this_tsc_khz == 0)) {
1105 		local_irq_restore(flags);
1106 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1107 		return 1;
1108 	}
1109 
1110 	/*
1111 	 * We may have to catch up the TSC to match elapsed wall clock
1112 	 * time for two reasons, even if kvmclock is used.
1113 	 *   1) CPU could have been running below the maximum TSC rate
1114 	 *   2) Broken TSC compensation resets the base at each VCPU
1115 	 *      entry to avoid unknown leaps of TSC even when running
1116 	 *      again on the same CPU.  This may cause apparent elapsed
1117 	 *      time to disappear, and the guest to stand still or run
1118 	 *	very slowly.
1119 	 */
1120 	if (vcpu->tsc_catchup) {
1121 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1122 		if (tsc > tsc_timestamp) {
1123 			kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1124 			tsc_timestamp = tsc;
1125 		}
1126 	}
1127 
1128 	local_irq_restore(flags);
1129 
1130 	if (!vcpu->time_page)
1131 		return 0;
1132 
1133 	/*
1134 	 * Time as measured by the TSC may go backwards when resetting the base
1135 	 * tsc_timestamp.  The reason for this is that the TSC resolution is
1136 	 * higher than the resolution of the other clock scales.  Thus, many
1137 	 * possible measurments of the TSC correspond to one measurement of any
1138 	 * other clock, and so a spread of values is possible.  This is not a
1139 	 * problem for the computation of the nanosecond clock; with TSC rates
1140 	 * around 1GHZ, there can only be a few cycles which correspond to one
1141 	 * nanosecond value, and any path through this code will inevitably
1142 	 * take longer than that.  However, with the kernel_ns value itself,
1143 	 * the precision may be much lower, down to HZ granularity.  If the
1144 	 * first sampling of TSC against kernel_ns ends in the low part of the
1145 	 * range, and the second in the high end of the range, we can get:
1146 	 *
1147 	 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1148 	 *
1149 	 * As the sampling errors potentially range in the thousands of cycles,
1150 	 * it is possible such a time value has already been observed by the
1151 	 * guest.  To protect against this, we must compute the system time as
1152 	 * observed by the guest and ensure the new system time is greater.
1153 	 */
1154 	max_kernel_ns = 0;
1155 	if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1156 		max_kernel_ns = vcpu->last_guest_tsc -
1157 				vcpu->hv_clock.tsc_timestamp;
1158 		max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1159 				    vcpu->hv_clock.tsc_to_system_mul,
1160 				    vcpu->hv_clock.tsc_shift);
1161 		max_kernel_ns += vcpu->last_kernel_ns;
1162 	}
1163 
1164 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1165 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1166 				   &vcpu->hv_clock.tsc_shift,
1167 				   &vcpu->hv_clock.tsc_to_system_mul);
1168 		vcpu->hw_tsc_khz = this_tsc_khz;
1169 	}
1170 
1171 	if (max_kernel_ns > kernel_ns)
1172 		kernel_ns = max_kernel_ns;
1173 
1174 	/* With all the info we got, fill in the values */
1175 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1176 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1177 	vcpu->last_kernel_ns = kernel_ns;
1178 	vcpu->last_guest_tsc = tsc_timestamp;
1179 	vcpu->hv_clock.flags = 0;
1180 
1181 	/*
1182 	 * The interface expects us to write an even number signaling that the
1183 	 * update is finished. Since the guest won't see the intermediate
1184 	 * state, we just increase by 2 at the end.
1185 	 */
1186 	vcpu->hv_clock.version += 2;
1187 
1188 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1189 
1190 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1191 	       sizeof(vcpu->hv_clock));
1192 
1193 	kunmap_atomic(shared_kaddr, KM_USER0);
1194 
1195 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1196 	return 0;
1197 }
1198 
1199 static bool msr_mtrr_valid(unsigned msr)
1200 {
1201 	switch (msr) {
1202 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1203 	case MSR_MTRRfix64K_00000:
1204 	case MSR_MTRRfix16K_80000:
1205 	case MSR_MTRRfix16K_A0000:
1206 	case MSR_MTRRfix4K_C0000:
1207 	case MSR_MTRRfix4K_C8000:
1208 	case MSR_MTRRfix4K_D0000:
1209 	case MSR_MTRRfix4K_D8000:
1210 	case MSR_MTRRfix4K_E0000:
1211 	case MSR_MTRRfix4K_E8000:
1212 	case MSR_MTRRfix4K_F0000:
1213 	case MSR_MTRRfix4K_F8000:
1214 	case MSR_MTRRdefType:
1215 	case MSR_IA32_CR_PAT:
1216 		return true;
1217 	case 0x2f8:
1218 		return true;
1219 	}
1220 	return false;
1221 }
1222 
1223 static bool valid_pat_type(unsigned t)
1224 {
1225 	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1226 }
1227 
1228 static bool valid_mtrr_type(unsigned t)
1229 {
1230 	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1231 }
1232 
1233 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1234 {
1235 	int i;
1236 
1237 	if (!msr_mtrr_valid(msr))
1238 		return false;
1239 
1240 	if (msr == MSR_IA32_CR_PAT) {
1241 		for (i = 0; i < 8; i++)
1242 			if (!valid_pat_type((data >> (i * 8)) & 0xff))
1243 				return false;
1244 		return true;
1245 	} else if (msr == MSR_MTRRdefType) {
1246 		if (data & ~0xcff)
1247 			return false;
1248 		return valid_mtrr_type(data & 0xff);
1249 	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1250 		for (i = 0; i < 8 ; i++)
1251 			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1252 				return false;
1253 		return true;
1254 	}
1255 
1256 	/* variable MTRRs */
1257 	return valid_mtrr_type(data & 0xff);
1258 }
1259 
1260 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1261 {
1262 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1263 
1264 	if (!mtrr_valid(vcpu, msr, data))
1265 		return 1;
1266 
1267 	if (msr == MSR_MTRRdefType) {
1268 		vcpu->arch.mtrr_state.def_type = data;
1269 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1270 	} else if (msr == MSR_MTRRfix64K_00000)
1271 		p[0] = data;
1272 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1273 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
1274 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1275 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1276 	else if (msr == MSR_IA32_CR_PAT)
1277 		vcpu->arch.pat = data;
1278 	else {	/* Variable MTRRs */
1279 		int idx, is_mtrr_mask;
1280 		u64 *pt;
1281 
1282 		idx = (msr - 0x200) / 2;
1283 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1284 		if (!is_mtrr_mask)
1285 			pt =
1286 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1287 		else
1288 			pt =
1289 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1290 		*pt = data;
1291 	}
1292 
1293 	kvm_mmu_reset_context(vcpu);
1294 	return 0;
1295 }
1296 
1297 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1298 {
1299 	u64 mcg_cap = vcpu->arch.mcg_cap;
1300 	unsigned bank_num = mcg_cap & 0xff;
1301 
1302 	switch (msr) {
1303 	case MSR_IA32_MCG_STATUS:
1304 		vcpu->arch.mcg_status = data;
1305 		break;
1306 	case MSR_IA32_MCG_CTL:
1307 		if (!(mcg_cap & MCG_CTL_P))
1308 			return 1;
1309 		if (data != 0 && data != ~(u64)0)
1310 			return -1;
1311 		vcpu->arch.mcg_ctl = data;
1312 		break;
1313 	default:
1314 		if (msr >= MSR_IA32_MC0_CTL &&
1315 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1316 			u32 offset = msr - MSR_IA32_MC0_CTL;
1317 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1318 			 * some Linux kernels though clear bit 10 in bank 4 to
1319 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1320 			 * this to avoid an uncatched #GP in the guest
1321 			 */
1322 			if ((offset & 0x3) == 0 &&
1323 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1324 				return -1;
1325 			vcpu->arch.mce_banks[offset] = data;
1326 			break;
1327 		}
1328 		return 1;
1329 	}
1330 	return 0;
1331 }
1332 
1333 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1334 {
1335 	struct kvm *kvm = vcpu->kvm;
1336 	int lm = is_long_mode(vcpu);
1337 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1338 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1339 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1340 		: kvm->arch.xen_hvm_config.blob_size_32;
1341 	u32 page_num = data & ~PAGE_MASK;
1342 	u64 page_addr = data & PAGE_MASK;
1343 	u8 *page;
1344 	int r;
1345 
1346 	r = -E2BIG;
1347 	if (page_num >= blob_size)
1348 		goto out;
1349 	r = -ENOMEM;
1350 	page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1351 	if (!page)
1352 		goto out;
1353 	r = -EFAULT;
1354 	if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1355 		goto out_free;
1356 	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1357 		goto out_free;
1358 	r = 0;
1359 out_free:
1360 	kfree(page);
1361 out:
1362 	return r;
1363 }
1364 
1365 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1366 {
1367 	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1368 }
1369 
1370 static bool kvm_hv_msr_partition_wide(u32 msr)
1371 {
1372 	bool r = false;
1373 	switch (msr) {
1374 	case HV_X64_MSR_GUEST_OS_ID:
1375 	case HV_X64_MSR_HYPERCALL:
1376 		r = true;
1377 		break;
1378 	}
1379 
1380 	return r;
1381 }
1382 
1383 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1384 {
1385 	struct kvm *kvm = vcpu->kvm;
1386 
1387 	switch (msr) {
1388 	case HV_X64_MSR_GUEST_OS_ID:
1389 		kvm->arch.hv_guest_os_id = data;
1390 		/* setting guest os id to zero disables hypercall page */
1391 		if (!kvm->arch.hv_guest_os_id)
1392 			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1393 		break;
1394 	case HV_X64_MSR_HYPERCALL: {
1395 		u64 gfn;
1396 		unsigned long addr;
1397 		u8 instructions[4];
1398 
1399 		/* if guest os id is not set hypercall should remain disabled */
1400 		if (!kvm->arch.hv_guest_os_id)
1401 			break;
1402 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1403 			kvm->arch.hv_hypercall = data;
1404 			break;
1405 		}
1406 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1407 		addr = gfn_to_hva(kvm, gfn);
1408 		if (kvm_is_error_hva(addr))
1409 			return 1;
1410 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1411 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1412 		if (__copy_to_user((void __user *)addr, instructions, 4))
1413 			return 1;
1414 		kvm->arch.hv_hypercall = data;
1415 		break;
1416 	}
1417 	default:
1418 		pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1419 			  "data 0x%llx\n", msr, data);
1420 		return 1;
1421 	}
1422 	return 0;
1423 }
1424 
1425 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1426 {
1427 	switch (msr) {
1428 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
1429 		unsigned long addr;
1430 
1431 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1432 			vcpu->arch.hv_vapic = data;
1433 			break;
1434 		}
1435 		addr = gfn_to_hva(vcpu->kvm, data >>
1436 				  HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1437 		if (kvm_is_error_hva(addr))
1438 			return 1;
1439 		if (__clear_user((void __user *)addr, PAGE_SIZE))
1440 			return 1;
1441 		vcpu->arch.hv_vapic = data;
1442 		break;
1443 	}
1444 	case HV_X64_MSR_EOI:
1445 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1446 	case HV_X64_MSR_ICR:
1447 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1448 	case HV_X64_MSR_TPR:
1449 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1450 	default:
1451 		pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1452 			  "data 0x%llx\n", msr, data);
1453 		return 1;
1454 	}
1455 
1456 	return 0;
1457 }
1458 
1459 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1460 {
1461 	gpa_t gpa = data & ~0x3f;
1462 
1463 	/* Bits 2:5 are resrved, Should be zero */
1464 	if (data & 0x3c)
1465 		return 1;
1466 
1467 	vcpu->arch.apf.msr_val = data;
1468 
1469 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1470 		kvm_clear_async_pf_completion_queue(vcpu);
1471 		kvm_async_pf_hash_reset(vcpu);
1472 		return 0;
1473 	}
1474 
1475 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1476 		return 1;
1477 
1478 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1479 	kvm_async_pf_wakeup_all(vcpu);
1480 	return 0;
1481 }
1482 
1483 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1484 {
1485 	if (vcpu->arch.time_page) {
1486 		kvm_release_page_dirty(vcpu->arch.time_page);
1487 		vcpu->arch.time_page = NULL;
1488 	}
1489 }
1490 
1491 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1492 {
1493 	u64 delta;
1494 
1495 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1496 		return;
1497 
1498 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1499 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
1500 	vcpu->arch.st.accum_steal = delta;
1501 }
1502 
1503 static void record_steal_time(struct kvm_vcpu *vcpu)
1504 {
1505 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1506 		return;
1507 
1508 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1509 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1510 		return;
1511 
1512 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1513 	vcpu->arch.st.steal.version += 2;
1514 	vcpu->arch.st.accum_steal = 0;
1515 
1516 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1517 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1518 }
1519 
1520 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1521 {
1522 	switch (msr) {
1523 	case MSR_EFER:
1524 		return set_efer(vcpu, data);
1525 	case MSR_K7_HWCR:
1526 		data &= ~(u64)0x40;	/* ignore flush filter disable */
1527 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
1528 		if (data != 0) {
1529 			pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1530 				data);
1531 			return 1;
1532 		}
1533 		break;
1534 	case MSR_FAM10H_MMIO_CONF_BASE:
1535 		if (data != 0) {
1536 			pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1537 				"0x%llx\n", data);
1538 			return 1;
1539 		}
1540 		break;
1541 	case MSR_AMD64_NB_CFG:
1542 		break;
1543 	case MSR_IA32_DEBUGCTLMSR:
1544 		if (!data) {
1545 			/* We support the non-activated case already */
1546 			break;
1547 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1548 			/* Values other than LBR and BTF are vendor-specific,
1549 			   thus reserved and should throw a #GP */
1550 			return 1;
1551 		}
1552 		pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1553 			__func__, data);
1554 		break;
1555 	case MSR_IA32_UCODE_REV:
1556 	case MSR_IA32_UCODE_WRITE:
1557 	case MSR_VM_HSAVE_PA:
1558 	case MSR_AMD64_PATCH_LOADER:
1559 		break;
1560 	case 0x200 ... 0x2ff:
1561 		return set_msr_mtrr(vcpu, msr, data);
1562 	case MSR_IA32_APICBASE:
1563 		kvm_set_apic_base(vcpu, data);
1564 		break;
1565 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1566 		return kvm_x2apic_msr_write(vcpu, msr, data);
1567 	case MSR_IA32_MISC_ENABLE:
1568 		vcpu->arch.ia32_misc_enable_msr = data;
1569 		break;
1570 	case MSR_KVM_WALL_CLOCK_NEW:
1571 	case MSR_KVM_WALL_CLOCK:
1572 		vcpu->kvm->arch.wall_clock = data;
1573 		kvm_write_wall_clock(vcpu->kvm, data);
1574 		break;
1575 	case MSR_KVM_SYSTEM_TIME_NEW:
1576 	case MSR_KVM_SYSTEM_TIME: {
1577 		kvmclock_reset(vcpu);
1578 
1579 		vcpu->arch.time = data;
1580 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1581 
1582 		/* we verify if the enable bit is set... */
1583 		if (!(data & 1))
1584 			break;
1585 
1586 		/* ...but clean it before doing the actual write */
1587 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1588 
1589 		vcpu->arch.time_page =
1590 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1591 
1592 		if (is_error_page(vcpu->arch.time_page)) {
1593 			kvm_release_page_clean(vcpu->arch.time_page);
1594 			vcpu->arch.time_page = NULL;
1595 		}
1596 		break;
1597 	}
1598 	case MSR_KVM_ASYNC_PF_EN:
1599 		if (kvm_pv_enable_async_pf(vcpu, data))
1600 			return 1;
1601 		break;
1602 	case MSR_KVM_STEAL_TIME:
1603 
1604 		if (unlikely(!sched_info_on()))
1605 			return 1;
1606 
1607 		if (data & KVM_STEAL_RESERVED_MASK)
1608 			return 1;
1609 
1610 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1611 							data & KVM_STEAL_VALID_BITS))
1612 			return 1;
1613 
1614 		vcpu->arch.st.msr_val = data;
1615 
1616 		if (!(data & KVM_MSR_ENABLED))
1617 			break;
1618 
1619 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
1620 
1621 		preempt_disable();
1622 		accumulate_steal_time(vcpu);
1623 		preempt_enable();
1624 
1625 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1626 
1627 		break;
1628 
1629 	case MSR_IA32_MCG_CTL:
1630 	case MSR_IA32_MCG_STATUS:
1631 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1632 		return set_msr_mce(vcpu, msr, data);
1633 
1634 	/* Performance counters are not protected by a CPUID bit,
1635 	 * so we should check all of them in the generic path for the sake of
1636 	 * cross vendor migration.
1637 	 * Writing a zero into the event select MSRs disables them,
1638 	 * which we perfectly emulate ;-). Any other value should be at least
1639 	 * reported, some guests depend on them.
1640 	 */
1641 	case MSR_P6_EVNTSEL0:
1642 	case MSR_P6_EVNTSEL1:
1643 	case MSR_K7_EVNTSEL0:
1644 	case MSR_K7_EVNTSEL1:
1645 	case MSR_K7_EVNTSEL2:
1646 	case MSR_K7_EVNTSEL3:
1647 		if (data != 0)
1648 			pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1649 				"0x%x data 0x%llx\n", msr, data);
1650 		break;
1651 	/* at least RHEL 4 unconditionally writes to the perfctr registers,
1652 	 * so we ignore writes to make it happy.
1653 	 */
1654 	case MSR_P6_PERFCTR0:
1655 	case MSR_P6_PERFCTR1:
1656 	case MSR_K7_PERFCTR0:
1657 	case MSR_K7_PERFCTR1:
1658 	case MSR_K7_PERFCTR2:
1659 	case MSR_K7_PERFCTR3:
1660 		pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1661 			"0x%x data 0x%llx\n", msr, data);
1662 		break;
1663 	case MSR_K7_CLK_CTL:
1664 		/*
1665 		 * Ignore all writes to this no longer documented MSR.
1666 		 * Writes are only relevant for old K7 processors,
1667 		 * all pre-dating SVM, but a recommended workaround from
1668 		 * AMD for these chips. It is possible to speicify the
1669 		 * affected processor models on the command line, hence
1670 		 * the need to ignore the workaround.
1671 		 */
1672 		break;
1673 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1674 		if (kvm_hv_msr_partition_wide(msr)) {
1675 			int r;
1676 			mutex_lock(&vcpu->kvm->lock);
1677 			r = set_msr_hyperv_pw(vcpu, msr, data);
1678 			mutex_unlock(&vcpu->kvm->lock);
1679 			return r;
1680 		} else
1681 			return set_msr_hyperv(vcpu, msr, data);
1682 		break;
1683 	case MSR_IA32_BBL_CR_CTL3:
1684 		/* Drop writes to this legacy MSR -- see rdmsr
1685 		 * counterpart for further detail.
1686 		 */
1687 		pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1688 		break;
1689 	default:
1690 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1691 			return xen_hvm_config(vcpu, data);
1692 		if (!ignore_msrs) {
1693 			pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1694 				msr, data);
1695 			return 1;
1696 		} else {
1697 			pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1698 				msr, data);
1699 			break;
1700 		}
1701 	}
1702 	return 0;
1703 }
1704 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1705 
1706 
1707 /*
1708  * Reads an msr value (of 'msr_index') into 'pdata'.
1709  * Returns 0 on success, non-0 otherwise.
1710  * Assumes vcpu_load() was already called.
1711  */
1712 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1713 {
1714 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1715 }
1716 
1717 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1718 {
1719 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1720 
1721 	if (!msr_mtrr_valid(msr))
1722 		return 1;
1723 
1724 	if (msr == MSR_MTRRdefType)
1725 		*pdata = vcpu->arch.mtrr_state.def_type +
1726 			 (vcpu->arch.mtrr_state.enabled << 10);
1727 	else if (msr == MSR_MTRRfix64K_00000)
1728 		*pdata = p[0];
1729 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1730 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1731 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1732 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1733 	else if (msr == MSR_IA32_CR_PAT)
1734 		*pdata = vcpu->arch.pat;
1735 	else {	/* Variable MTRRs */
1736 		int idx, is_mtrr_mask;
1737 		u64 *pt;
1738 
1739 		idx = (msr - 0x200) / 2;
1740 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1741 		if (!is_mtrr_mask)
1742 			pt =
1743 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1744 		else
1745 			pt =
1746 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1747 		*pdata = *pt;
1748 	}
1749 
1750 	return 0;
1751 }
1752 
1753 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1754 {
1755 	u64 data;
1756 	u64 mcg_cap = vcpu->arch.mcg_cap;
1757 	unsigned bank_num = mcg_cap & 0xff;
1758 
1759 	switch (msr) {
1760 	case MSR_IA32_P5_MC_ADDR:
1761 	case MSR_IA32_P5_MC_TYPE:
1762 		data = 0;
1763 		break;
1764 	case MSR_IA32_MCG_CAP:
1765 		data = vcpu->arch.mcg_cap;
1766 		break;
1767 	case MSR_IA32_MCG_CTL:
1768 		if (!(mcg_cap & MCG_CTL_P))
1769 			return 1;
1770 		data = vcpu->arch.mcg_ctl;
1771 		break;
1772 	case MSR_IA32_MCG_STATUS:
1773 		data = vcpu->arch.mcg_status;
1774 		break;
1775 	default:
1776 		if (msr >= MSR_IA32_MC0_CTL &&
1777 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1778 			u32 offset = msr - MSR_IA32_MC0_CTL;
1779 			data = vcpu->arch.mce_banks[offset];
1780 			break;
1781 		}
1782 		return 1;
1783 	}
1784 	*pdata = data;
1785 	return 0;
1786 }
1787 
1788 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1789 {
1790 	u64 data = 0;
1791 	struct kvm *kvm = vcpu->kvm;
1792 
1793 	switch (msr) {
1794 	case HV_X64_MSR_GUEST_OS_ID:
1795 		data = kvm->arch.hv_guest_os_id;
1796 		break;
1797 	case HV_X64_MSR_HYPERCALL:
1798 		data = kvm->arch.hv_hypercall;
1799 		break;
1800 	default:
1801 		pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1802 		return 1;
1803 	}
1804 
1805 	*pdata = data;
1806 	return 0;
1807 }
1808 
1809 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1810 {
1811 	u64 data = 0;
1812 
1813 	switch (msr) {
1814 	case HV_X64_MSR_VP_INDEX: {
1815 		int r;
1816 		struct kvm_vcpu *v;
1817 		kvm_for_each_vcpu(r, v, vcpu->kvm)
1818 			if (v == vcpu)
1819 				data = r;
1820 		break;
1821 	}
1822 	case HV_X64_MSR_EOI:
1823 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1824 	case HV_X64_MSR_ICR:
1825 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1826 	case HV_X64_MSR_TPR:
1827 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1828 	default:
1829 		pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1830 		return 1;
1831 	}
1832 	*pdata = data;
1833 	return 0;
1834 }
1835 
1836 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1837 {
1838 	u64 data;
1839 
1840 	switch (msr) {
1841 	case MSR_IA32_PLATFORM_ID:
1842 	case MSR_IA32_UCODE_REV:
1843 	case MSR_IA32_EBL_CR_POWERON:
1844 	case MSR_IA32_DEBUGCTLMSR:
1845 	case MSR_IA32_LASTBRANCHFROMIP:
1846 	case MSR_IA32_LASTBRANCHTOIP:
1847 	case MSR_IA32_LASTINTFROMIP:
1848 	case MSR_IA32_LASTINTTOIP:
1849 	case MSR_K8_SYSCFG:
1850 	case MSR_K7_HWCR:
1851 	case MSR_VM_HSAVE_PA:
1852 	case MSR_P6_PERFCTR0:
1853 	case MSR_P6_PERFCTR1:
1854 	case MSR_P6_EVNTSEL0:
1855 	case MSR_P6_EVNTSEL1:
1856 	case MSR_K7_EVNTSEL0:
1857 	case MSR_K7_PERFCTR0:
1858 	case MSR_K8_INT_PENDING_MSG:
1859 	case MSR_AMD64_NB_CFG:
1860 	case MSR_FAM10H_MMIO_CONF_BASE:
1861 		data = 0;
1862 		break;
1863 	case MSR_MTRRcap:
1864 		data = 0x500 | KVM_NR_VAR_MTRR;
1865 		break;
1866 	case 0x200 ... 0x2ff:
1867 		return get_msr_mtrr(vcpu, msr, pdata);
1868 	case 0xcd: /* fsb frequency */
1869 		data = 3;
1870 		break;
1871 		/*
1872 		 * MSR_EBC_FREQUENCY_ID
1873 		 * Conservative value valid for even the basic CPU models.
1874 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1875 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1876 		 * and 266MHz for model 3, or 4. Set Core Clock
1877 		 * Frequency to System Bus Frequency Ratio to 1 (bits
1878 		 * 31:24) even though these are only valid for CPU
1879 		 * models > 2, however guests may end up dividing or
1880 		 * multiplying by zero otherwise.
1881 		 */
1882 	case MSR_EBC_FREQUENCY_ID:
1883 		data = 1 << 24;
1884 		break;
1885 	case MSR_IA32_APICBASE:
1886 		data = kvm_get_apic_base(vcpu);
1887 		break;
1888 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1889 		return kvm_x2apic_msr_read(vcpu, msr, pdata);
1890 		break;
1891 	case MSR_IA32_MISC_ENABLE:
1892 		data = vcpu->arch.ia32_misc_enable_msr;
1893 		break;
1894 	case MSR_IA32_PERF_STATUS:
1895 		/* TSC increment by tick */
1896 		data = 1000ULL;
1897 		/* CPU multiplier */
1898 		data |= (((uint64_t)4ULL) << 40);
1899 		break;
1900 	case MSR_EFER:
1901 		data = vcpu->arch.efer;
1902 		break;
1903 	case MSR_KVM_WALL_CLOCK:
1904 	case MSR_KVM_WALL_CLOCK_NEW:
1905 		data = vcpu->kvm->arch.wall_clock;
1906 		break;
1907 	case MSR_KVM_SYSTEM_TIME:
1908 	case MSR_KVM_SYSTEM_TIME_NEW:
1909 		data = vcpu->arch.time;
1910 		break;
1911 	case MSR_KVM_ASYNC_PF_EN:
1912 		data = vcpu->arch.apf.msr_val;
1913 		break;
1914 	case MSR_KVM_STEAL_TIME:
1915 		data = vcpu->arch.st.msr_val;
1916 		break;
1917 	case MSR_IA32_P5_MC_ADDR:
1918 	case MSR_IA32_P5_MC_TYPE:
1919 	case MSR_IA32_MCG_CAP:
1920 	case MSR_IA32_MCG_CTL:
1921 	case MSR_IA32_MCG_STATUS:
1922 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1923 		return get_msr_mce(vcpu, msr, pdata);
1924 	case MSR_K7_CLK_CTL:
1925 		/*
1926 		 * Provide expected ramp-up count for K7. All other
1927 		 * are set to zero, indicating minimum divisors for
1928 		 * every field.
1929 		 *
1930 		 * This prevents guest kernels on AMD host with CPU
1931 		 * type 6, model 8 and higher from exploding due to
1932 		 * the rdmsr failing.
1933 		 */
1934 		data = 0x20000000;
1935 		break;
1936 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1937 		if (kvm_hv_msr_partition_wide(msr)) {
1938 			int r;
1939 			mutex_lock(&vcpu->kvm->lock);
1940 			r = get_msr_hyperv_pw(vcpu, msr, pdata);
1941 			mutex_unlock(&vcpu->kvm->lock);
1942 			return r;
1943 		} else
1944 			return get_msr_hyperv(vcpu, msr, pdata);
1945 		break;
1946 	case MSR_IA32_BBL_CR_CTL3:
1947 		/* This legacy MSR exists but isn't fully documented in current
1948 		 * silicon.  It is however accessed by winxp in very narrow
1949 		 * scenarios where it sets bit #19, itself documented as
1950 		 * a "reserved" bit.  Best effort attempt to source coherent
1951 		 * read data here should the balance of the register be
1952 		 * interpreted by the guest:
1953 		 *
1954 		 * L2 cache control register 3: 64GB range, 256KB size,
1955 		 * enabled, latency 0x1, configured
1956 		 */
1957 		data = 0xbe702111;
1958 		break;
1959 	default:
1960 		if (!ignore_msrs) {
1961 			pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1962 			return 1;
1963 		} else {
1964 			pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1965 			data = 0;
1966 		}
1967 		break;
1968 	}
1969 	*pdata = data;
1970 	return 0;
1971 }
1972 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1973 
1974 /*
1975  * Read or write a bunch of msrs. All parameters are kernel addresses.
1976  *
1977  * @return number of msrs set successfully.
1978  */
1979 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1980 		    struct kvm_msr_entry *entries,
1981 		    int (*do_msr)(struct kvm_vcpu *vcpu,
1982 				  unsigned index, u64 *data))
1983 {
1984 	int i, idx;
1985 
1986 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1987 	for (i = 0; i < msrs->nmsrs; ++i)
1988 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
1989 			break;
1990 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1991 
1992 	return i;
1993 }
1994 
1995 /*
1996  * Read or write a bunch of msrs. Parameters are user addresses.
1997  *
1998  * @return number of msrs set successfully.
1999  */
2000 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2001 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2002 				unsigned index, u64 *data),
2003 		  int writeback)
2004 {
2005 	struct kvm_msrs msrs;
2006 	struct kvm_msr_entry *entries;
2007 	int r, n;
2008 	unsigned size;
2009 
2010 	r = -EFAULT;
2011 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2012 		goto out;
2013 
2014 	r = -E2BIG;
2015 	if (msrs.nmsrs >= MAX_IO_MSRS)
2016 		goto out;
2017 
2018 	r = -ENOMEM;
2019 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2020 	entries = kmalloc(size, GFP_KERNEL);
2021 	if (!entries)
2022 		goto out;
2023 
2024 	r = -EFAULT;
2025 	if (copy_from_user(entries, user_msrs->entries, size))
2026 		goto out_free;
2027 
2028 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2029 	if (r < 0)
2030 		goto out_free;
2031 
2032 	r = -EFAULT;
2033 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2034 		goto out_free;
2035 
2036 	r = n;
2037 
2038 out_free:
2039 	kfree(entries);
2040 out:
2041 	return r;
2042 }
2043 
2044 int kvm_dev_ioctl_check_extension(long ext)
2045 {
2046 	int r;
2047 
2048 	switch (ext) {
2049 	case KVM_CAP_IRQCHIP:
2050 	case KVM_CAP_HLT:
2051 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2052 	case KVM_CAP_SET_TSS_ADDR:
2053 	case KVM_CAP_EXT_CPUID:
2054 	case KVM_CAP_CLOCKSOURCE:
2055 	case KVM_CAP_PIT:
2056 	case KVM_CAP_NOP_IO_DELAY:
2057 	case KVM_CAP_MP_STATE:
2058 	case KVM_CAP_SYNC_MMU:
2059 	case KVM_CAP_USER_NMI:
2060 	case KVM_CAP_REINJECT_CONTROL:
2061 	case KVM_CAP_IRQ_INJECT_STATUS:
2062 	case KVM_CAP_ASSIGN_DEV_IRQ:
2063 	case KVM_CAP_IRQFD:
2064 	case KVM_CAP_IOEVENTFD:
2065 	case KVM_CAP_PIT2:
2066 	case KVM_CAP_PIT_STATE2:
2067 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2068 	case KVM_CAP_XEN_HVM:
2069 	case KVM_CAP_ADJUST_CLOCK:
2070 	case KVM_CAP_VCPU_EVENTS:
2071 	case KVM_CAP_HYPERV:
2072 	case KVM_CAP_HYPERV_VAPIC:
2073 	case KVM_CAP_HYPERV_SPIN:
2074 	case KVM_CAP_PCI_SEGMENT:
2075 	case KVM_CAP_DEBUGREGS:
2076 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2077 	case KVM_CAP_XSAVE:
2078 	case KVM_CAP_ASYNC_PF:
2079 	case KVM_CAP_GET_TSC_KHZ:
2080 		r = 1;
2081 		break;
2082 	case KVM_CAP_COALESCED_MMIO:
2083 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2084 		break;
2085 	case KVM_CAP_VAPIC:
2086 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2087 		break;
2088 	case KVM_CAP_NR_VCPUS:
2089 		r = KVM_MAX_VCPUS;
2090 		break;
2091 	case KVM_CAP_NR_MEMSLOTS:
2092 		r = KVM_MEMORY_SLOTS;
2093 		break;
2094 	case KVM_CAP_PV_MMU:	/* obsolete */
2095 		r = 0;
2096 		break;
2097 	case KVM_CAP_IOMMU:
2098 		r = iommu_found();
2099 		break;
2100 	case KVM_CAP_MCE:
2101 		r = KVM_MAX_MCE_BANKS;
2102 		break;
2103 	case KVM_CAP_XCRS:
2104 		r = cpu_has_xsave;
2105 		break;
2106 	case KVM_CAP_TSC_CONTROL:
2107 		r = kvm_has_tsc_control;
2108 		break;
2109 	default:
2110 		r = 0;
2111 		break;
2112 	}
2113 	return r;
2114 
2115 }
2116 
2117 long kvm_arch_dev_ioctl(struct file *filp,
2118 			unsigned int ioctl, unsigned long arg)
2119 {
2120 	void __user *argp = (void __user *)arg;
2121 	long r;
2122 
2123 	switch (ioctl) {
2124 	case KVM_GET_MSR_INDEX_LIST: {
2125 		struct kvm_msr_list __user *user_msr_list = argp;
2126 		struct kvm_msr_list msr_list;
2127 		unsigned n;
2128 
2129 		r = -EFAULT;
2130 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2131 			goto out;
2132 		n = msr_list.nmsrs;
2133 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2134 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2135 			goto out;
2136 		r = -E2BIG;
2137 		if (n < msr_list.nmsrs)
2138 			goto out;
2139 		r = -EFAULT;
2140 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2141 				 num_msrs_to_save * sizeof(u32)))
2142 			goto out;
2143 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2144 				 &emulated_msrs,
2145 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2146 			goto out;
2147 		r = 0;
2148 		break;
2149 	}
2150 	case KVM_GET_SUPPORTED_CPUID: {
2151 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2152 		struct kvm_cpuid2 cpuid;
2153 
2154 		r = -EFAULT;
2155 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2156 			goto out;
2157 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2158 						      cpuid_arg->entries);
2159 		if (r)
2160 			goto out;
2161 
2162 		r = -EFAULT;
2163 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2164 			goto out;
2165 		r = 0;
2166 		break;
2167 	}
2168 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2169 		u64 mce_cap;
2170 
2171 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2172 		r = -EFAULT;
2173 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2174 			goto out;
2175 		r = 0;
2176 		break;
2177 	}
2178 	default:
2179 		r = -EINVAL;
2180 	}
2181 out:
2182 	return r;
2183 }
2184 
2185 static void wbinvd_ipi(void *garbage)
2186 {
2187 	wbinvd();
2188 }
2189 
2190 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2191 {
2192 	return vcpu->kvm->arch.iommu_domain &&
2193 		!(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2194 }
2195 
2196 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2197 {
2198 	/* Address WBINVD may be executed by guest */
2199 	if (need_emulate_wbinvd(vcpu)) {
2200 		if (kvm_x86_ops->has_wbinvd_exit())
2201 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2202 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2203 			smp_call_function_single(vcpu->cpu,
2204 					wbinvd_ipi, NULL, 1);
2205 	}
2206 
2207 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2208 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2209 		/* Make sure TSC doesn't go backwards */
2210 		s64 tsc_delta;
2211 		u64 tsc;
2212 
2213 		kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2214 		tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2215 			     tsc - vcpu->arch.last_guest_tsc;
2216 
2217 		if (tsc_delta < 0)
2218 			mark_tsc_unstable("KVM discovered backwards TSC");
2219 		if (check_tsc_unstable()) {
2220 			kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2221 			vcpu->arch.tsc_catchup = 1;
2222 		}
2223 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2224 		if (vcpu->cpu != cpu)
2225 			kvm_migrate_timers(vcpu);
2226 		vcpu->cpu = cpu;
2227 	}
2228 
2229 	accumulate_steal_time(vcpu);
2230 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2231 }
2232 
2233 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2234 {
2235 	kvm_x86_ops->vcpu_put(vcpu);
2236 	kvm_put_guest_fpu(vcpu);
2237 	kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2238 }
2239 
2240 static int is_efer_nx(void)
2241 {
2242 	unsigned long long efer = 0;
2243 
2244 	rdmsrl_safe(MSR_EFER, &efer);
2245 	return efer & EFER_NX;
2246 }
2247 
2248 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2249 {
2250 	int i;
2251 	struct kvm_cpuid_entry2 *e, *entry;
2252 
2253 	entry = NULL;
2254 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2255 		e = &vcpu->arch.cpuid_entries[i];
2256 		if (e->function == 0x80000001) {
2257 			entry = e;
2258 			break;
2259 		}
2260 	}
2261 	if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2262 		entry->edx &= ~(1 << 20);
2263 		printk(KERN_INFO "kvm: guest NX capability removed\n");
2264 	}
2265 }
2266 
2267 /* when an old userspace process fills a new kernel module */
2268 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2269 				    struct kvm_cpuid *cpuid,
2270 				    struct kvm_cpuid_entry __user *entries)
2271 {
2272 	int r, i;
2273 	struct kvm_cpuid_entry *cpuid_entries;
2274 
2275 	r = -E2BIG;
2276 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2277 		goto out;
2278 	r = -ENOMEM;
2279 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2280 	if (!cpuid_entries)
2281 		goto out;
2282 	r = -EFAULT;
2283 	if (copy_from_user(cpuid_entries, entries,
2284 			   cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2285 		goto out_free;
2286 	for (i = 0; i < cpuid->nent; i++) {
2287 		vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2288 		vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2289 		vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2290 		vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2291 		vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2292 		vcpu->arch.cpuid_entries[i].index = 0;
2293 		vcpu->arch.cpuid_entries[i].flags = 0;
2294 		vcpu->arch.cpuid_entries[i].padding[0] = 0;
2295 		vcpu->arch.cpuid_entries[i].padding[1] = 0;
2296 		vcpu->arch.cpuid_entries[i].padding[2] = 0;
2297 	}
2298 	vcpu->arch.cpuid_nent = cpuid->nent;
2299 	cpuid_fix_nx_cap(vcpu);
2300 	r = 0;
2301 	kvm_apic_set_version(vcpu);
2302 	kvm_x86_ops->cpuid_update(vcpu);
2303 	update_cpuid(vcpu);
2304 
2305 out_free:
2306 	vfree(cpuid_entries);
2307 out:
2308 	return r;
2309 }
2310 
2311 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2312 				     struct kvm_cpuid2 *cpuid,
2313 				     struct kvm_cpuid_entry2 __user *entries)
2314 {
2315 	int r;
2316 
2317 	r = -E2BIG;
2318 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2319 		goto out;
2320 	r = -EFAULT;
2321 	if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2322 			   cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2323 		goto out;
2324 	vcpu->arch.cpuid_nent = cpuid->nent;
2325 	kvm_apic_set_version(vcpu);
2326 	kvm_x86_ops->cpuid_update(vcpu);
2327 	update_cpuid(vcpu);
2328 	return 0;
2329 
2330 out:
2331 	return r;
2332 }
2333 
2334 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2335 				     struct kvm_cpuid2 *cpuid,
2336 				     struct kvm_cpuid_entry2 __user *entries)
2337 {
2338 	int r;
2339 
2340 	r = -E2BIG;
2341 	if (cpuid->nent < vcpu->arch.cpuid_nent)
2342 		goto out;
2343 	r = -EFAULT;
2344 	if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2345 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2346 		goto out;
2347 	return 0;
2348 
2349 out:
2350 	cpuid->nent = vcpu->arch.cpuid_nent;
2351 	return r;
2352 }
2353 
2354 static void cpuid_mask(u32 *word, int wordnum)
2355 {
2356 	*word &= boot_cpu_data.x86_capability[wordnum];
2357 }
2358 
2359 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2360 			   u32 index)
2361 {
2362 	entry->function = function;
2363 	entry->index = index;
2364 	cpuid_count(entry->function, entry->index,
2365 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2366 	entry->flags = 0;
2367 }
2368 
2369 static bool supported_xcr0_bit(unsigned bit)
2370 {
2371 	u64 mask = ((u64)1 << bit);
2372 
2373 	return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2374 }
2375 
2376 #define F(x) bit(X86_FEATURE_##x)
2377 
2378 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2379 			 u32 index, int *nent, int maxnent)
2380 {
2381 	unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2382 #ifdef CONFIG_X86_64
2383 	unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2384 				? F(GBPAGES) : 0;
2385 	unsigned f_lm = F(LM);
2386 #else
2387 	unsigned f_gbpages = 0;
2388 	unsigned f_lm = 0;
2389 #endif
2390 	unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2391 
2392 	/* cpuid 1.edx */
2393 	const u32 kvm_supported_word0_x86_features =
2394 		F(FPU) | F(VME) | F(DE) | F(PSE) |
2395 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2396 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2397 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2398 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2399 		0 /* Reserved, DS, ACPI */ | F(MMX) |
2400 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2401 		0 /* HTT, TM, Reserved, PBE */;
2402 	/* cpuid 0x80000001.edx */
2403 	const u32 kvm_supported_word1_x86_features =
2404 		F(FPU) | F(VME) | F(DE) | F(PSE) |
2405 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2406 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2407 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2408 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
2409 		f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2410 		F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2411 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2412 	/* cpuid 1.ecx */
2413 	const u32 kvm_supported_word4_x86_features =
2414 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2415 		0 /* DS-CPL, VMX, SMX, EST */ |
2416 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2417 		0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2418 		0 /* Reserved, DCA */ | F(XMM4_1) |
2419 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2420 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2421 		F(F16C) | F(RDRAND);
2422 	/* cpuid 0x80000001.ecx */
2423 	const u32 kvm_supported_word6_x86_features =
2424 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2425 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2426 		F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2427 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2428 
2429 	/* cpuid 0xC0000001.edx */
2430 	const u32 kvm_supported_word5_x86_features =
2431 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2432 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2433 		F(PMM) | F(PMM_EN);
2434 
2435 	/* cpuid 7.0.ebx */
2436 	const u32 kvm_supported_word9_x86_features =
2437 		F(SMEP) | F(FSGSBASE) | F(ERMS);
2438 
2439 	/* all calls to cpuid_count() should be made on the same cpu */
2440 	get_cpu();
2441 	do_cpuid_1_ent(entry, function, index);
2442 	++*nent;
2443 
2444 	switch (function) {
2445 	case 0:
2446 		entry->eax = min(entry->eax, (u32)0xd);
2447 		break;
2448 	case 1:
2449 		entry->edx &= kvm_supported_word0_x86_features;
2450 		cpuid_mask(&entry->edx, 0);
2451 		entry->ecx &= kvm_supported_word4_x86_features;
2452 		cpuid_mask(&entry->ecx, 4);
2453 		/* we support x2apic emulation even if host does not support
2454 		 * it since we emulate x2apic in software */
2455 		entry->ecx |= F(X2APIC);
2456 		break;
2457 	/* function 2 entries are STATEFUL. That is, repeated cpuid commands
2458 	 * may return different values. This forces us to get_cpu() before
2459 	 * issuing the first command, and also to emulate this annoying behavior
2460 	 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2461 	case 2: {
2462 		int t, times = entry->eax & 0xff;
2463 
2464 		entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2465 		entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2466 		for (t = 1; t < times && *nent < maxnent; ++t) {
2467 			do_cpuid_1_ent(&entry[t], function, 0);
2468 			entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2469 			++*nent;
2470 		}
2471 		break;
2472 	}
2473 	/* function 4 has additional index. */
2474 	case 4: {
2475 		int i, cache_type;
2476 
2477 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2478 		/* read more entries until cache_type is zero */
2479 		for (i = 1; *nent < maxnent; ++i) {
2480 			cache_type = entry[i - 1].eax & 0x1f;
2481 			if (!cache_type)
2482 				break;
2483 			do_cpuid_1_ent(&entry[i], function, i);
2484 			entry[i].flags |=
2485 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2486 			++*nent;
2487 		}
2488 		break;
2489 	}
2490 	case 7: {
2491 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2492 		/* Mask ebx against host capbability word 9 */
2493 		if (index == 0) {
2494 			entry->ebx &= kvm_supported_word9_x86_features;
2495 			cpuid_mask(&entry->ebx, 9);
2496 		} else
2497 			entry->ebx = 0;
2498 		entry->eax = 0;
2499 		entry->ecx = 0;
2500 		entry->edx = 0;
2501 		break;
2502 	}
2503 	case 9:
2504 		break;
2505 	/* function 0xb has additional index. */
2506 	case 0xb: {
2507 		int i, level_type;
2508 
2509 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2510 		/* read more entries until level_type is zero */
2511 		for (i = 1; *nent < maxnent; ++i) {
2512 			level_type = entry[i - 1].ecx & 0xff00;
2513 			if (!level_type)
2514 				break;
2515 			do_cpuid_1_ent(&entry[i], function, i);
2516 			entry[i].flags |=
2517 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2518 			++*nent;
2519 		}
2520 		break;
2521 	}
2522 	case 0xd: {
2523 		int idx, i;
2524 
2525 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2526 		for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2527 			do_cpuid_1_ent(&entry[i], function, idx);
2528 			if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2529 				continue;
2530 			entry[i].flags |=
2531 			       KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2532 			++*nent;
2533 			++i;
2534 		}
2535 		break;
2536 	}
2537 	case KVM_CPUID_SIGNATURE: {
2538 		char signature[12] = "KVMKVMKVM\0\0";
2539 		u32 *sigptr = (u32 *)signature;
2540 		entry->eax = 0;
2541 		entry->ebx = sigptr[0];
2542 		entry->ecx = sigptr[1];
2543 		entry->edx = sigptr[2];
2544 		break;
2545 	}
2546 	case KVM_CPUID_FEATURES:
2547 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2548 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
2549 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
2550 			     (1 << KVM_FEATURE_ASYNC_PF) |
2551 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2552 
2553 		if (sched_info_on())
2554 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2555 
2556 		entry->ebx = 0;
2557 		entry->ecx = 0;
2558 		entry->edx = 0;
2559 		break;
2560 	case 0x80000000:
2561 		entry->eax = min(entry->eax, 0x8000001a);
2562 		break;
2563 	case 0x80000001:
2564 		entry->edx &= kvm_supported_word1_x86_features;
2565 		cpuid_mask(&entry->edx, 1);
2566 		entry->ecx &= kvm_supported_word6_x86_features;
2567 		cpuid_mask(&entry->ecx, 6);
2568 		break;
2569 	case 0x80000008: {
2570 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2571 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2572 		unsigned phys_as = entry->eax & 0xff;
2573 
2574 		if (!g_phys_as)
2575 			g_phys_as = phys_as;
2576 		entry->eax = g_phys_as | (virt_as << 8);
2577 		entry->ebx = entry->edx = 0;
2578 		break;
2579 	}
2580 	case 0x80000019:
2581 		entry->ecx = entry->edx = 0;
2582 		break;
2583 	case 0x8000001a:
2584 		break;
2585 	case 0x8000001d:
2586 		break;
2587 	/*Add support for Centaur's CPUID instruction*/
2588 	case 0xC0000000:
2589 		/*Just support up to 0xC0000004 now*/
2590 		entry->eax = min(entry->eax, 0xC0000004);
2591 		break;
2592 	case 0xC0000001:
2593 		entry->edx &= kvm_supported_word5_x86_features;
2594 		cpuid_mask(&entry->edx, 5);
2595 		break;
2596 	case 3: /* Processor serial number */
2597 	case 5: /* MONITOR/MWAIT */
2598 	case 6: /* Thermal management */
2599 	case 0xA: /* Architectural Performance Monitoring */
2600 	case 0x80000007: /* Advanced power management */
2601 	case 0xC0000002:
2602 	case 0xC0000003:
2603 	case 0xC0000004:
2604 	default:
2605 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2606 		break;
2607 	}
2608 
2609 	kvm_x86_ops->set_supported_cpuid(function, entry);
2610 
2611 	put_cpu();
2612 }
2613 
2614 #undef F
2615 
2616 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2617 				     struct kvm_cpuid_entry2 __user *entries)
2618 {
2619 	struct kvm_cpuid_entry2 *cpuid_entries;
2620 	int limit, nent = 0, r = -E2BIG;
2621 	u32 func;
2622 
2623 	if (cpuid->nent < 1)
2624 		goto out;
2625 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2626 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2627 	r = -ENOMEM;
2628 	cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2629 	if (!cpuid_entries)
2630 		goto out;
2631 
2632 	do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2633 	limit = cpuid_entries[0].eax;
2634 	for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2635 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
2636 			     &nent, cpuid->nent);
2637 	r = -E2BIG;
2638 	if (nent >= cpuid->nent)
2639 		goto out_free;
2640 
2641 	do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2642 	limit = cpuid_entries[nent - 1].eax;
2643 	for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2644 		do_cpuid_ent(&cpuid_entries[nent], func, 0,
2645 			     &nent, cpuid->nent);
2646 
2647 
2648 
2649 	r = -E2BIG;
2650 	if (nent >= cpuid->nent)
2651 		goto out_free;
2652 
2653 	/* Add support for Centaur's CPUID instruction. */
2654 	if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2655 		do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2656 				&nent, cpuid->nent);
2657 
2658 		r = -E2BIG;
2659 		if (nent >= cpuid->nent)
2660 			goto out_free;
2661 
2662 		limit = cpuid_entries[nent - 1].eax;
2663 		for (func = 0xC0000001;
2664 			func <= limit && nent < cpuid->nent; ++func)
2665 			do_cpuid_ent(&cpuid_entries[nent], func, 0,
2666 					&nent, cpuid->nent);
2667 
2668 		r = -E2BIG;
2669 		if (nent >= cpuid->nent)
2670 			goto out_free;
2671 	}
2672 
2673 	do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2674 		     cpuid->nent);
2675 
2676 	r = -E2BIG;
2677 	if (nent >= cpuid->nent)
2678 		goto out_free;
2679 
2680 	do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2681 		     cpuid->nent);
2682 
2683 	r = -E2BIG;
2684 	if (nent >= cpuid->nent)
2685 		goto out_free;
2686 
2687 	r = -EFAULT;
2688 	if (copy_to_user(entries, cpuid_entries,
2689 			 nent * sizeof(struct kvm_cpuid_entry2)))
2690 		goto out_free;
2691 	cpuid->nent = nent;
2692 	r = 0;
2693 
2694 out_free:
2695 	vfree(cpuid_entries);
2696 out:
2697 	return r;
2698 }
2699 
2700 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2701 				    struct kvm_lapic_state *s)
2702 {
2703 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2704 
2705 	return 0;
2706 }
2707 
2708 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2709 				    struct kvm_lapic_state *s)
2710 {
2711 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2712 	kvm_apic_post_state_restore(vcpu);
2713 	update_cr8_intercept(vcpu);
2714 
2715 	return 0;
2716 }
2717 
2718 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2719 				    struct kvm_interrupt *irq)
2720 {
2721 	if (irq->irq < 0 || irq->irq >= 256)
2722 		return -EINVAL;
2723 	if (irqchip_in_kernel(vcpu->kvm))
2724 		return -ENXIO;
2725 
2726 	kvm_queue_interrupt(vcpu, irq->irq, false);
2727 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2728 
2729 	return 0;
2730 }
2731 
2732 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2733 {
2734 	kvm_inject_nmi(vcpu);
2735 
2736 	return 0;
2737 }
2738 
2739 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2740 					   struct kvm_tpr_access_ctl *tac)
2741 {
2742 	if (tac->flags)
2743 		return -EINVAL;
2744 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2745 	return 0;
2746 }
2747 
2748 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2749 					u64 mcg_cap)
2750 {
2751 	int r;
2752 	unsigned bank_num = mcg_cap & 0xff, bank;
2753 
2754 	r = -EINVAL;
2755 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2756 		goto out;
2757 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2758 		goto out;
2759 	r = 0;
2760 	vcpu->arch.mcg_cap = mcg_cap;
2761 	/* Init IA32_MCG_CTL to all 1s */
2762 	if (mcg_cap & MCG_CTL_P)
2763 		vcpu->arch.mcg_ctl = ~(u64)0;
2764 	/* Init IA32_MCi_CTL to all 1s */
2765 	for (bank = 0; bank < bank_num; bank++)
2766 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2767 out:
2768 	return r;
2769 }
2770 
2771 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2772 				      struct kvm_x86_mce *mce)
2773 {
2774 	u64 mcg_cap = vcpu->arch.mcg_cap;
2775 	unsigned bank_num = mcg_cap & 0xff;
2776 	u64 *banks = vcpu->arch.mce_banks;
2777 
2778 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2779 		return -EINVAL;
2780 	/*
2781 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2782 	 * reporting is disabled
2783 	 */
2784 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2785 	    vcpu->arch.mcg_ctl != ~(u64)0)
2786 		return 0;
2787 	banks += 4 * mce->bank;
2788 	/*
2789 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2790 	 * reporting is disabled for the bank
2791 	 */
2792 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2793 		return 0;
2794 	if (mce->status & MCI_STATUS_UC) {
2795 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2796 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2797 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2798 			return 0;
2799 		}
2800 		if (banks[1] & MCI_STATUS_VAL)
2801 			mce->status |= MCI_STATUS_OVER;
2802 		banks[2] = mce->addr;
2803 		banks[3] = mce->misc;
2804 		vcpu->arch.mcg_status = mce->mcg_status;
2805 		banks[1] = mce->status;
2806 		kvm_queue_exception(vcpu, MC_VECTOR);
2807 	} else if (!(banks[1] & MCI_STATUS_VAL)
2808 		   || !(banks[1] & MCI_STATUS_UC)) {
2809 		if (banks[1] & MCI_STATUS_VAL)
2810 			mce->status |= MCI_STATUS_OVER;
2811 		banks[2] = mce->addr;
2812 		banks[3] = mce->misc;
2813 		banks[1] = mce->status;
2814 	} else
2815 		banks[1] |= MCI_STATUS_OVER;
2816 	return 0;
2817 }
2818 
2819 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2820 					       struct kvm_vcpu_events *events)
2821 {
2822 	events->exception.injected =
2823 		vcpu->arch.exception.pending &&
2824 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2825 	events->exception.nr = vcpu->arch.exception.nr;
2826 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2827 	events->exception.pad = 0;
2828 	events->exception.error_code = vcpu->arch.exception.error_code;
2829 
2830 	events->interrupt.injected =
2831 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2832 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2833 	events->interrupt.soft = 0;
2834 	events->interrupt.shadow =
2835 		kvm_x86_ops->get_interrupt_shadow(vcpu,
2836 			KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2837 
2838 	events->nmi.injected = vcpu->arch.nmi_injected;
2839 	events->nmi.pending = vcpu->arch.nmi_pending;
2840 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2841 	events->nmi.pad = 0;
2842 
2843 	events->sipi_vector = vcpu->arch.sipi_vector;
2844 
2845 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2846 			 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2847 			 | KVM_VCPUEVENT_VALID_SHADOW);
2848 	memset(&events->reserved, 0, sizeof(events->reserved));
2849 }
2850 
2851 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2852 					      struct kvm_vcpu_events *events)
2853 {
2854 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2855 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2856 			      | KVM_VCPUEVENT_VALID_SHADOW))
2857 		return -EINVAL;
2858 
2859 	vcpu->arch.exception.pending = events->exception.injected;
2860 	vcpu->arch.exception.nr = events->exception.nr;
2861 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2862 	vcpu->arch.exception.error_code = events->exception.error_code;
2863 
2864 	vcpu->arch.interrupt.pending = events->interrupt.injected;
2865 	vcpu->arch.interrupt.nr = events->interrupt.nr;
2866 	vcpu->arch.interrupt.soft = events->interrupt.soft;
2867 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2868 		kvm_x86_ops->set_interrupt_shadow(vcpu,
2869 						  events->interrupt.shadow);
2870 
2871 	vcpu->arch.nmi_injected = events->nmi.injected;
2872 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2873 		vcpu->arch.nmi_pending = events->nmi.pending;
2874 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2875 
2876 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2877 		vcpu->arch.sipi_vector = events->sipi_vector;
2878 
2879 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2880 
2881 	return 0;
2882 }
2883 
2884 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2885 					     struct kvm_debugregs *dbgregs)
2886 {
2887 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2888 	dbgregs->dr6 = vcpu->arch.dr6;
2889 	dbgregs->dr7 = vcpu->arch.dr7;
2890 	dbgregs->flags = 0;
2891 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2892 }
2893 
2894 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2895 					    struct kvm_debugregs *dbgregs)
2896 {
2897 	if (dbgregs->flags)
2898 		return -EINVAL;
2899 
2900 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2901 	vcpu->arch.dr6 = dbgregs->dr6;
2902 	vcpu->arch.dr7 = dbgregs->dr7;
2903 
2904 	return 0;
2905 }
2906 
2907 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2908 					 struct kvm_xsave *guest_xsave)
2909 {
2910 	if (cpu_has_xsave)
2911 		memcpy(guest_xsave->region,
2912 			&vcpu->arch.guest_fpu.state->xsave,
2913 			xstate_size);
2914 	else {
2915 		memcpy(guest_xsave->region,
2916 			&vcpu->arch.guest_fpu.state->fxsave,
2917 			sizeof(struct i387_fxsave_struct));
2918 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2919 			XSTATE_FPSSE;
2920 	}
2921 }
2922 
2923 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2924 					struct kvm_xsave *guest_xsave)
2925 {
2926 	u64 xstate_bv =
2927 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2928 
2929 	if (cpu_has_xsave)
2930 		memcpy(&vcpu->arch.guest_fpu.state->xsave,
2931 			guest_xsave->region, xstate_size);
2932 	else {
2933 		if (xstate_bv & ~XSTATE_FPSSE)
2934 			return -EINVAL;
2935 		memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2936 			guest_xsave->region, sizeof(struct i387_fxsave_struct));
2937 	}
2938 	return 0;
2939 }
2940 
2941 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2942 					struct kvm_xcrs *guest_xcrs)
2943 {
2944 	if (!cpu_has_xsave) {
2945 		guest_xcrs->nr_xcrs = 0;
2946 		return;
2947 	}
2948 
2949 	guest_xcrs->nr_xcrs = 1;
2950 	guest_xcrs->flags = 0;
2951 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2952 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2953 }
2954 
2955 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2956 				       struct kvm_xcrs *guest_xcrs)
2957 {
2958 	int i, r = 0;
2959 
2960 	if (!cpu_has_xsave)
2961 		return -EINVAL;
2962 
2963 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2964 		return -EINVAL;
2965 
2966 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2967 		/* Only support XCR0 currently */
2968 		if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2969 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2970 				guest_xcrs->xcrs[0].value);
2971 			break;
2972 		}
2973 	if (r)
2974 		r = -EINVAL;
2975 	return r;
2976 }
2977 
2978 long kvm_arch_vcpu_ioctl(struct file *filp,
2979 			 unsigned int ioctl, unsigned long arg)
2980 {
2981 	struct kvm_vcpu *vcpu = filp->private_data;
2982 	void __user *argp = (void __user *)arg;
2983 	int r;
2984 	union {
2985 		struct kvm_lapic_state *lapic;
2986 		struct kvm_xsave *xsave;
2987 		struct kvm_xcrs *xcrs;
2988 		void *buffer;
2989 	} u;
2990 
2991 	u.buffer = NULL;
2992 	switch (ioctl) {
2993 	case KVM_GET_LAPIC: {
2994 		r = -EINVAL;
2995 		if (!vcpu->arch.apic)
2996 			goto out;
2997 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2998 
2999 		r = -ENOMEM;
3000 		if (!u.lapic)
3001 			goto out;
3002 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3003 		if (r)
3004 			goto out;
3005 		r = -EFAULT;
3006 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3007 			goto out;
3008 		r = 0;
3009 		break;
3010 	}
3011 	case KVM_SET_LAPIC: {
3012 		r = -EINVAL;
3013 		if (!vcpu->arch.apic)
3014 			goto out;
3015 		u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3016 		r = -ENOMEM;
3017 		if (!u.lapic)
3018 			goto out;
3019 		r = -EFAULT;
3020 		if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3021 			goto out;
3022 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3023 		if (r)
3024 			goto out;
3025 		r = 0;
3026 		break;
3027 	}
3028 	case KVM_INTERRUPT: {
3029 		struct kvm_interrupt irq;
3030 
3031 		r = -EFAULT;
3032 		if (copy_from_user(&irq, argp, sizeof irq))
3033 			goto out;
3034 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3035 		if (r)
3036 			goto out;
3037 		r = 0;
3038 		break;
3039 	}
3040 	case KVM_NMI: {
3041 		r = kvm_vcpu_ioctl_nmi(vcpu);
3042 		if (r)
3043 			goto out;
3044 		r = 0;
3045 		break;
3046 	}
3047 	case KVM_SET_CPUID: {
3048 		struct kvm_cpuid __user *cpuid_arg = argp;
3049 		struct kvm_cpuid cpuid;
3050 
3051 		r = -EFAULT;
3052 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3053 			goto out;
3054 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3055 		if (r)
3056 			goto out;
3057 		break;
3058 	}
3059 	case KVM_SET_CPUID2: {
3060 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3061 		struct kvm_cpuid2 cpuid;
3062 
3063 		r = -EFAULT;
3064 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3065 			goto out;
3066 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3067 					      cpuid_arg->entries);
3068 		if (r)
3069 			goto out;
3070 		break;
3071 	}
3072 	case KVM_GET_CPUID2: {
3073 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3074 		struct kvm_cpuid2 cpuid;
3075 
3076 		r = -EFAULT;
3077 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3078 			goto out;
3079 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3080 					      cpuid_arg->entries);
3081 		if (r)
3082 			goto out;
3083 		r = -EFAULT;
3084 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3085 			goto out;
3086 		r = 0;
3087 		break;
3088 	}
3089 	case KVM_GET_MSRS:
3090 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
3091 		break;
3092 	case KVM_SET_MSRS:
3093 		r = msr_io(vcpu, argp, do_set_msr, 0);
3094 		break;
3095 	case KVM_TPR_ACCESS_REPORTING: {
3096 		struct kvm_tpr_access_ctl tac;
3097 
3098 		r = -EFAULT;
3099 		if (copy_from_user(&tac, argp, sizeof tac))
3100 			goto out;
3101 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3102 		if (r)
3103 			goto out;
3104 		r = -EFAULT;
3105 		if (copy_to_user(argp, &tac, sizeof tac))
3106 			goto out;
3107 		r = 0;
3108 		break;
3109 	};
3110 	case KVM_SET_VAPIC_ADDR: {
3111 		struct kvm_vapic_addr va;
3112 
3113 		r = -EINVAL;
3114 		if (!irqchip_in_kernel(vcpu->kvm))
3115 			goto out;
3116 		r = -EFAULT;
3117 		if (copy_from_user(&va, argp, sizeof va))
3118 			goto out;
3119 		r = 0;
3120 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3121 		break;
3122 	}
3123 	case KVM_X86_SETUP_MCE: {
3124 		u64 mcg_cap;
3125 
3126 		r = -EFAULT;
3127 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3128 			goto out;
3129 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3130 		break;
3131 	}
3132 	case KVM_X86_SET_MCE: {
3133 		struct kvm_x86_mce mce;
3134 
3135 		r = -EFAULT;
3136 		if (copy_from_user(&mce, argp, sizeof mce))
3137 			goto out;
3138 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3139 		break;
3140 	}
3141 	case KVM_GET_VCPU_EVENTS: {
3142 		struct kvm_vcpu_events events;
3143 
3144 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3145 
3146 		r = -EFAULT;
3147 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3148 			break;
3149 		r = 0;
3150 		break;
3151 	}
3152 	case KVM_SET_VCPU_EVENTS: {
3153 		struct kvm_vcpu_events events;
3154 
3155 		r = -EFAULT;
3156 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3157 			break;
3158 
3159 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3160 		break;
3161 	}
3162 	case KVM_GET_DEBUGREGS: {
3163 		struct kvm_debugregs dbgregs;
3164 
3165 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3166 
3167 		r = -EFAULT;
3168 		if (copy_to_user(argp, &dbgregs,
3169 				 sizeof(struct kvm_debugregs)))
3170 			break;
3171 		r = 0;
3172 		break;
3173 	}
3174 	case KVM_SET_DEBUGREGS: {
3175 		struct kvm_debugregs dbgregs;
3176 
3177 		r = -EFAULT;
3178 		if (copy_from_user(&dbgregs, argp,
3179 				   sizeof(struct kvm_debugregs)))
3180 			break;
3181 
3182 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3183 		break;
3184 	}
3185 	case KVM_GET_XSAVE: {
3186 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3187 		r = -ENOMEM;
3188 		if (!u.xsave)
3189 			break;
3190 
3191 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3192 
3193 		r = -EFAULT;
3194 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3195 			break;
3196 		r = 0;
3197 		break;
3198 	}
3199 	case KVM_SET_XSAVE: {
3200 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3201 		r = -ENOMEM;
3202 		if (!u.xsave)
3203 			break;
3204 
3205 		r = -EFAULT;
3206 		if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3207 			break;
3208 
3209 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3210 		break;
3211 	}
3212 	case KVM_GET_XCRS: {
3213 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3214 		r = -ENOMEM;
3215 		if (!u.xcrs)
3216 			break;
3217 
3218 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3219 
3220 		r = -EFAULT;
3221 		if (copy_to_user(argp, u.xcrs,
3222 				 sizeof(struct kvm_xcrs)))
3223 			break;
3224 		r = 0;
3225 		break;
3226 	}
3227 	case KVM_SET_XCRS: {
3228 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3229 		r = -ENOMEM;
3230 		if (!u.xcrs)
3231 			break;
3232 
3233 		r = -EFAULT;
3234 		if (copy_from_user(u.xcrs, argp,
3235 				   sizeof(struct kvm_xcrs)))
3236 			break;
3237 
3238 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3239 		break;
3240 	}
3241 	case KVM_SET_TSC_KHZ: {
3242 		u32 user_tsc_khz;
3243 
3244 		r = -EINVAL;
3245 		if (!kvm_has_tsc_control)
3246 			break;
3247 
3248 		user_tsc_khz = (u32)arg;
3249 
3250 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3251 			goto out;
3252 
3253 		kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3254 
3255 		r = 0;
3256 		goto out;
3257 	}
3258 	case KVM_GET_TSC_KHZ: {
3259 		r = -EIO;
3260 		if (check_tsc_unstable())
3261 			goto out;
3262 
3263 		r = vcpu_tsc_khz(vcpu);
3264 
3265 		goto out;
3266 	}
3267 	default:
3268 		r = -EINVAL;
3269 	}
3270 out:
3271 	kfree(u.buffer);
3272 	return r;
3273 }
3274 
3275 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3276 {
3277 	int ret;
3278 
3279 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3280 		return -1;
3281 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3282 	return ret;
3283 }
3284 
3285 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3286 					      u64 ident_addr)
3287 {
3288 	kvm->arch.ept_identity_map_addr = ident_addr;
3289 	return 0;
3290 }
3291 
3292 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3293 					  u32 kvm_nr_mmu_pages)
3294 {
3295 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3296 		return -EINVAL;
3297 
3298 	mutex_lock(&kvm->slots_lock);
3299 	spin_lock(&kvm->mmu_lock);
3300 
3301 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3302 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3303 
3304 	spin_unlock(&kvm->mmu_lock);
3305 	mutex_unlock(&kvm->slots_lock);
3306 	return 0;
3307 }
3308 
3309 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3310 {
3311 	return kvm->arch.n_max_mmu_pages;
3312 }
3313 
3314 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3315 {
3316 	int r;
3317 
3318 	r = 0;
3319 	switch (chip->chip_id) {
3320 	case KVM_IRQCHIP_PIC_MASTER:
3321 		memcpy(&chip->chip.pic,
3322 			&pic_irqchip(kvm)->pics[0],
3323 			sizeof(struct kvm_pic_state));
3324 		break;
3325 	case KVM_IRQCHIP_PIC_SLAVE:
3326 		memcpy(&chip->chip.pic,
3327 			&pic_irqchip(kvm)->pics[1],
3328 			sizeof(struct kvm_pic_state));
3329 		break;
3330 	case KVM_IRQCHIP_IOAPIC:
3331 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3332 		break;
3333 	default:
3334 		r = -EINVAL;
3335 		break;
3336 	}
3337 	return r;
3338 }
3339 
3340 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3341 {
3342 	int r;
3343 
3344 	r = 0;
3345 	switch (chip->chip_id) {
3346 	case KVM_IRQCHIP_PIC_MASTER:
3347 		spin_lock(&pic_irqchip(kvm)->lock);
3348 		memcpy(&pic_irqchip(kvm)->pics[0],
3349 			&chip->chip.pic,
3350 			sizeof(struct kvm_pic_state));
3351 		spin_unlock(&pic_irqchip(kvm)->lock);
3352 		break;
3353 	case KVM_IRQCHIP_PIC_SLAVE:
3354 		spin_lock(&pic_irqchip(kvm)->lock);
3355 		memcpy(&pic_irqchip(kvm)->pics[1],
3356 			&chip->chip.pic,
3357 			sizeof(struct kvm_pic_state));
3358 		spin_unlock(&pic_irqchip(kvm)->lock);
3359 		break;
3360 	case KVM_IRQCHIP_IOAPIC:
3361 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3362 		break;
3363 	default:
3364 		r = -EINVAL;
3365 		break;
3366 	}
3367 	kvm_pic_update_irq(pic_irqchip(kvm));
3368 	return r;
3369 }
3370 
3371 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3372 {
3373 	int r = 0;
3374 
3375 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3376 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3377 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3378 	return r;
3379 }
3380 
3381 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3382 {
3383 	int r = 0;
3384 
3385 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3386 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3387 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3388 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3389 	return r;
3390 }
3391 
3392 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3393 {
3394 	int r = 0;
3395 
3396 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3397 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3398 		sizeof(ps->channels));
3399 	ps->flags = kvm->arch.vpit->pit_state.flags;
3400 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3401 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3402 	return r;
3403 }
3404 
3405 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3406 {
3407 	int r = 0, start = 0;
3408 	u32 prev_legacy, cur_legacy;
3409 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3410 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3411 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3412 	if (!prev_legacy && cur_legacy)
3413 		start = 1;
3414 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3415 	       sizeof(kvm->arch.vpit->pit_state.channels));
3416 	kvm->arch.vpit->pit_state.flags = ps->flags;
3417 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3418 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3419 	return r;
3420 }
3421 
3422 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3423 				 struct kvm_reinject_control *control)
3424 {
3425 	if (!kvm->arch.vpit)
3426 		return -ENXIO;
3427 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3428 	kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3429 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3430 	return 0;
3431 }
3432 
3433 /*
3434  * Get (and clear) the dirty memory log for a memory slot.
3435  */
3436 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3437 				      struct kvm_dirty_log *log)
3438 {
3439 	int r, i;
3440 	struct kvm_memory_slot *memslot;
3441 	unsigned long n;
3442 	unsigned long is_dirty = 0;
3443 
3444 	mutex_lock(&kvm->slots_lock);
3445 
3446 	r = -EINVAL;
3447 	if (log->slot >= KVM_MEMORY_SLOTS)
3448 		goto out;
3449 
3450 	memslot = &kvm->memslots->memslots[log->slot];
3451 	r = -ENOENT;
3452 	if (!memslot->dirty_bitmap)
3453 		goto out;
3454 
3455 	n = kvm_dirty_bitmap_bytes(memslot);
3456 
3457 	for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3458 		is_dirty = memslot->dirty_bitmap[i];
3459 
3460 	/* If nothing is dirty, don't bother messing with page tables. */
3461 	if (is_dirty) {
3462 		struct kvm_memslots *slots, *old_slots;
3463 		unsigned long *dirty_bitmap;
3464 
3465 		dirty_bitmap = memslot->dirty_bitmap_head;
3466 		if (memslot->dirty_bitmap == dirty_bitmap)
3467 			dirty_bitmap += n / sizeof(long);
3468 		memset(dirty_bitmap, 0, n);
3469 
3470 		r = -ENOMEM;
3471 		slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3472 		if (!slots)
3473 			goto out;
3474 		memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3475 		slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3476 		slots->generation++;
3477 
3478 		old_slots = kvm->memslots;
3479 		rcu_assign_pointer(kvm->memslots, slots);
3480 		synchronize_srcu_expedited(&kvm->srcu);
3481 		dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3482 		kfree(old_slots);
3483 
3484 		spin_lock(&kvm->mmu_lock);
3485 		kvm_mmu_slot_remove_write_access(kvm, log->slot);
3486 		spin_unlock(&kvm->mmu_lock);
3487 
3488 		r = -EFAULT;
3489 		if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3490 			goto out;
3491 	} else {
3492 		r = -EFAULT;
3493 		if (clear_user(log->dirty_bitmap, n))
3494 			goto out;
3495 	}
3496 
3497 	r = 0;
3498 out:
3499 	mutex_unlock(&kvm->slots_lock);
3500 	return r;
3501 }
3502 
3503 long kvm_arch_vm_ioctl(struct file *filp,
3504 		       unsigned int ioctl, unsigned long arg)
3505 {
3506 	struct kvm *kvm = filp->private_data;
3507 	void __user *argp = (void __user *)arg;
3508 	int r = -ENOTTY;
3509 	/*
3510 	 * This union makes it completely explicit to gcc-3.x
3511 	 * that these two variables' stack usage should be
3512 	 * combined, not added together.
3513 	 */
3514 	union {
3515 		struct kvm_pit_state ps;
3516 		struct kvm_pit_state2 ps2;
3517 		struct kvm_pit_config pit_config;
3518 	} u;
3519 
3520 	switch (ioctl) {
3521 	case KVM_SET_TSS_ADDR:
3522 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3523 		if (r < 0)
3524 			goto out;
3525 		break;
3526 	case KVM_SET_IDENTITY_MAP_ADDR: {
3527 		u64 ident_addr;
3528 
3529 		r = -EFAULT;
3530 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3531 			goto out;
3532 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3533 		if (r < 0)
3534 			goto out;
3535 		break;
3536 	}
3537 	case KVM_SET_NR_MMU_PAGES:
3538 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3539 		if (r)
3540 			goto out;
3541 		break;
3542 	case KVM_GET_NR_MMU_PAGES:
3543 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3544 		break;
3545 	case KVM_CREATE_IRQCHIP: {
3546 		struct kvm_pic *vpic;
3547 
3548 		mutex_lock(&kvm->lock);
3549 		r = -EEXIST;
3550 		if (kvm->arch.vpic)
3551 			goto create_irqchip_unlock;
3552 		r = -ENOMEM;
3553 		vpic = kvm_create_pic(kvm);
3554 		if (vpic) {
3555 			r = kvm_ioapic_init(kvm);
3556 			if (r) {
3557 				mutex_lock(&kvm->slots_lock);
3558 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3559 							  &vpic->dev);
3560 				mutex_unlock(&kvm->slots_lock);
3561 				kfree(vpic);
3562 				goto create_irqchip_unlock;
3563 			}
3564 		} else
3565 			goto create_irqchip_unlock;
3566 		smp_wmb();
3567 		kvm->arch.vpic = vpic;
3568 		smp_wmb();
3569 		r = kvm_setup_default_irq_routing(kvm);
3570 		if (r) {
3571 			mutex_lock(&kvm->slots_lock);
3572 			mutex_lock(&kvm->irq_lock);
3573 			kvm_ioapic_destroy(kvm);
3574 			kvm_destroy_pic(kvm);
3575 			mutex_unlock(&kvm->irq_lock);
3576 			mutex_unlock(&kvm->slots_lock);
3577 		}
3578 	create_irqchip_unlock:
3579 		mutex_unlock(&kvm->lock);
3580 		break;
3581 	}
3582 	case KVM_CREATE_PIT:
3583 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3584 		goto create_pit;
3585 	case KVM_CREATE_PIT2:
3586 		r = -EFAULT;
3587 		if (copy_from_user(&u.pit_config, argp,
3588 				   sizeof(struct kvm_pit_config)))
3589 			goto out;
3590 	create_pit:
3591 		mutex_lock(&kvm->slots_lock);
3592 		r = -EEXIST;
3593 		if (kvm->arch.vpit)
3594 			goto create_pit_unlock;
3595 		r = -ENOMEM;
3596 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3597 		if (kvm->arch.vpit)
3598 			r = 0;
3599 	create_pit_unlock:
3600 		mutex_unlock(&kvm->slots_lock);
3601 		break;
3602 	case KVM_IRQ_LINE_STATUS:
3603 	case KVM_IRQ_LINE: {
3604 		struct kvm_irq_level irq_event;
3605 
3606 		r = -EFAULT;
3607 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
3608 			goto out;
3609 		r = -ENXIO;
3610 		if (irqchip_in_kernel(kvm)) {
3611 			__s32 status;
3612 			status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3613 					irq_event.irq, irq_event.level);
3614 			if (ioctl == KVM_IRQ_LINE_STATUS) {
3615 				r = -EFAULT;
3616 				irq_event.status = status;
3617 				if (copy_to_user(argp, &irq_event,
3618 							sizeof irq_event))
3619 					goto out;
3620 			}
3621 			r = 0;
3622 		}
3623 		break;
3624 	}
3625 	case KVM_GET_IRQCHIP: {
3626 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3627 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3628 
3629 		r = -ENOMEM;
3630 		if (!chip)
3631 			goto out;
3632 		r = -EFAULT;
3633 		if (copy_from_user(chip, argp, sizeof *chip))
3634 			goto get_irqchip_out;
3635 		r = -ENXIO;
3636 		if (!irqchip_in_kernel(kvm))
3637 			goto get_irqchip_out;
3638 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3639 		if (r)
3640 			goto get_irqchip_out;
3641 		r = -EFAULT;
3642 		if (copy_to_user(argp, chip, sizeof *chip))
3643 			goto get_irqchip_out;
3644 		r = 0;
3645 	get_irqchip_out:
3646 		kfree(chip);
3647 		if (r)
3648 			goto out;
3649 		break;
3650 	}
3651 	case KVM_SET_IRQCHIP: {
3652 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3653 		struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3654 
3655 		r = -ENOMEM;
3656 		if (!chip)
3657 			goto out;
3658 		r = -EFAULT;
3659 		if (copy_from_user(chip, argp, sizeof *chip))
3660 			goto set_irqchip_out;
3661 		r = -ENXIO;
3662 		if (!irqchip_in_kernel(kvm))
3663 			goto set_irqchip_out;
3664 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3665 		if (r)
3666 			goto set_irqchip_out;
3667 		r = 0;
3668 	set_irqchip_out:
3669 		kfree(chip);
3670 		if (r)
3671 			goto out;
3672 		break;
3673 	}
3674 	case KVM_GET_PIT: {
3675 		r = -EFAULT;
3676 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3677 			goto out;
3678 		r = -ENXIO;
3679 		if (!kvm->arch.vpit)
3680 			goto out;
3681 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3682 		if (r)
3683 			goto out;
3684 		r = -EFAULT;
3685 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3686 			goto out;
3687 		r = 0;
3688 		break;
3689 	}
3690 	case KVM_SET_PIT: {
3691 		r = -EFAULT;
3692 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3693 			goto out;
3694 		r = -ENXIO;
3695 		if (!kvm->arch.vpit)
3696 			goto out;
3697 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3698 		if (r)
3699 			goto out;
3700 		r = 0;
3701 		break;
3702 	}
3703 	case KVM_GET_PIT2: {
3704 		r = -ENXIO;
3705 		if (!kvm->arch.vpit)
3706 			goto out;
3707 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3708 		if (r)
3709 			goto out;
3710 		r = -EFAULT;
3711 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3712 			goto out;
3713 		r = 0;
3714 		break;
3715 	}
3716 	case KVM_SET_PIT2: {
3717 		r = -EFAULT;
3718 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3719 			goto out;
3720 		r = -ENXIO;
3721 		if (!kvm->arch.vpit)
3722 			goto out;
3723 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3724 		if (r)
3725 			goto out;
3726 		r = 0;
3727 		break;
3728 	}
3729 	case KVM_REINJECT_CONTROL: {
3730 		struct kvm_reinject_control control;
3731 		r =  -EFAULT;
3732 		if (copy_from_user(&control, argp, sizeof(control)))
3733 			goto out;
3734 		r = kvm_vm_ioctl_reinject(kvm, &control);
3735 		if (r)
3736 			goto out;
3737 		r = 0;
3738 		break;
3739 	}
3740 	case KVM_XEN_HVM_CONFIG: {
3741 		r = -EFAULT;
3742 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3743 				   sizeof(struct kvm_xen_hvm_config)))
3744 			goto out;
3745 		r = -EINVAL;
3746 		if (kvm->arch.xen_hvm_config.flags)
3747 			goto out;
3748 		r = 0;
3749 		break;
3750 	}
3751 	case KVM_SET_CLOCK: {
3752 		struct kvm_clock_data user_ns;
3753 		u64 now_ns;
3754 		s64 delta;
3755 
3756 		r = -EFAULT;
3757 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3758 			goto out;
3759 
3760 		r = -EINVAL;
3761 		if (user_ns.flags)
3762 			goto out;
3763 
3764 		r = 0;
3765 		local_irq_disable();
3766 		now_ns = get_kernel_ns();
3767 		delta = user_ns.clock - now_ns;
3768 		local_irq_enable();
3769 		kvm->arch.kvmclock_offset = delta;
3770 		break;
3771 	}
3772 	case KVM_GET_CLOCK: {
3773 		struct kvm_clock_data user_ns;
3774 		u64 now_ns;
3775 
3776 		local_irq_disable();
3777 		now_ns = get_kernel_ns();
3778 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3779 		local_irq_enable();
3780 		user_ns.flags = 0;
3781 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3782 
3783 		r = -EFAULT;
3784 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3785 			goto out;
3786 		r = 0;
3787 		break;
3788 	}
3789 
3790 	default:
3791 		;
3792 	}
3793 out:
3794 	return r;
3795 }
3796 
3797 static void kvm_init_msr_list(void)
3798 {
3799 	u32 dummy[2];
3800 	unsigned i, j;
3801 
3802 	/* skip the first msrs in the list. KVM-specific */
3803 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3804 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3805 			continue;
3806 		if (j < i)
3807 			msrs_to_save[j] = msrs_to_save[i];
3808 		j++;
3809 	}
3810 	num_msrs_to_save = j;
3811 }
3812 
3813 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3814 			   const void *v)
3815 {
3816 	int handled = 0;
3817 	int n;
3818 
3819 	do {
3820 		n = min(len, 8);
3821 		if (!(vcpu->arch.apic &&
3822 		      !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3823 		    && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3824 			break;
3825 		handled += n;
3826 		addr += n;
3827 		len -= n;
3828 		v += n;
3829 	} while (len);
3830 
3831 	return handled;
3832 }
3833 
3834 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3835 {
3836 	int handled = 0;
3837 	int n;
3838 
3839 	do {
3840 		n = min(len, 8);
3841 		if (!(vcpu->arch.apic &&
3842 		      !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3843 		    && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3844 			break;
3845 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3846 		handled += n;
3847 		addr += n;
3848 		len -= n;
3849 		v += n;
3850 	} while (len);
3851 
3852 	return handled;
3853 }
3854 
3855 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3856 			struct kvm_segment *var, int seg)
3857 {
3858 	kvm_x86_ops->set_segment(vcpu, var, seg);
3859 }
3860 
3861 void kvm_get_segment(struct kvm_vcpu *vcpu,
3862 		     struct kvm_segment *var, int seg)
3863 {
3864 	kvm_x86_ops->get_segment(vcpu, var, seg);
3865 }
3866 
3867 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3868 {
3869 	return gpa;
3870 }
3871 
3872 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3873 {
3874 	gpa_t t_gpa;
3875 	struct x86_exception exception;
3876 
3877 	BUG_ON(!mmu_is_nested(vcpu));
3878 
3879 	/* NPT walks are always user-walks */
3880 	access |= PFERR_USER_MASK;
3881 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3882 
3883 	return t_gpa;
3884 }
3885 
3886 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3887 			      struct x86_exception *exception)
3888 {
3889 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3890 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3891 }
3892 
3893  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3894 				struct x86_exception *exception)
3895 {
3896 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3897 	access |= PFERR_FETCH_MASK;
3898 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3899 }
3900 
3901 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3902 			       struct x86_exception *exception)
3903 {
3904 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3905 	access |= PFERR_WRITE_MASK;
3906 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3907 }
3908 
3909 /* uses this to access any guest's mapped memory without checking CPL */
3910 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3911 				struct x86_exception *exception)
3912 {
3913 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3914 }
3915 
3916 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3917 				      struct kvm_vcpu *vcpu, u32 access,
3918 				      struct x86_exception *exception)
3919 {
3920 	void *data = val;
3921 	int r = X86EMUL_CONTINUE;
3922 
3923 	while (bytes) {
3924 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3925 							    exception);
3926 		unsigned offset = addr & (PAGE_SIZE-1);
3927 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3928 		int ret;
3929 
3930 		if (gpa == UNMAPPED_GVA)
3931 			return X86EMUL_PROPAGATE_FAULT;
3932 		ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3933 		if (ret < 0) {
3934 			r = X86EMUL_IO_NEEDED;
3935 			goto out;
3936 		}
3937 
3938 		bytes -= toread;
3939 		data += toread;
3940 		addr += toread;
3941 	}
3942 out:
3943 	return r;
3944 }
3945 
3946 /* used for instruction fetching */
3947 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3948 				gva_t addr, void *val, unsigned int bytes,
3949 				struct x86_exception *exception)
3950 {
3951 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3952 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3953 
3954 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3955 					  access | PFERR_FETCH_MASK,
3956 					  exception);
3957 }
3958 
3959 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3960 			       gva_t addr, void *val, unsigned int bytes,
3961 			       struct x86_exception *exception)
3962 {
3963 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3964 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3965 
3966 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3967 					  exception);
3968 }
3969 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3970 
3971 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3972 				      gva_t addr, void *val, unsigned int bytes,
3973 				      struct x86_exception *exception)
3974 {
3975 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3976 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3977 }
3978 
3979 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3980 				       gva_t addr, void *val,
3981 				       unsigned int bytes,
3982 				       struct x86_exception *exception)
3983 {
3984 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3985 	void *data = val;
3986 	int r = X86EMUL_CONTINUE;
3987 
3988 	while (bytes) {
3989 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3990 							     PFERR_WRITE_MASK,
3991 							     exception);
3992 		unsigned offset = addr & (PAGE_SIZE-1);
3993 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3994 		int ret;
3995 
3996 		if (gpa == UNMAPPED_GVA)
3997 			return X86EMUL_PROPAGATE_FAULT;
3998 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3999 		if (ret < 0) {
4000 			r = X86EMUL_IO_NEEDED;
4001 			goto out;
4002 		}
4003 
4004 		bytes -= towrite;
4005 		data += towrite;
4006 		addr += towrite;
4007 	}
4008 out:
4009 	return r;
4010 }
4011 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4012 
4013 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4014 				gpa_t *gpa, struct x86_exception *exception,
4015 				bool write)
4016 {
4017 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4018 
4019 	if (vcpu_match_mmio_gva(vcpu, gva) &&
4020 		  check_write_user_access(vcpu, write, access,
4021 		  vcpu->arch.access)) {
4022 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4023 					(gva & (PAGE_SIZE - 1));
4024 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4025 		return 1;
4026 	}
4027 
4028 	if (write)
4029 		access |= PFERR_WRITE_MASK;
4030 
4031 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4032 
4033 	if (*gpa == UNMAPPED_GVA)
4034 		return -1;
4035 
4036 	/* For APIC access vmexit */
4037 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4038 		return 1;
4039 
4040 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4041 		trace_vcpu_match_mmio(gva, *gpa, write, true);
4042 		return 1;
4043 	}
4044 
4045 	return 0;
4046 }
4047 
4048 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4049 				  unsigned long addr,
4050 				  void *val,
4051 				  unsigned int bytes,
4052 				  struct x86_exception *exception)
4053 {
4054 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4055 	gpa_t gpa;
4056 	int handled, ret;
4057 
4058 	if (vcpu->mmio_read_completed) {
4059 		memcpy(val, vcpu->mmio_data, bytes);
4060 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4061 			       vcpu->mmio_phys_addr, *(u64 *)val);
4062 		vcpu->mmio_read_completed = 0;
4063 		return X86EMUL_CONTINUE;
4064 	}
4065 
4066 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, false);
4067 
4068 	if (ret < 0)
4069 		return X86EMUL_PROPAGATE_FAULT;
4070 
4071 	if (ret)
4072 		goto mmio;
4073 
4074 	if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
4075 	    == X86EMUL_CONTINUE)
4076 		return X86EMUL_CONTINUE;
4077 
4078 mmio:
4079 	/*
4080 	 * Is this MMIO handled locally?
4081 	 */
4082 	handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
4083 
4084 	if (handled == bytes)
4085 		return X86EMUL_CONTINUE;
4086 
4087 	gpa += handled;
4088 	bytes -= handled;
4089 	val += handled;
4090 
4091 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4092 
4093 	vcpu->mmio_needed = 1;
4094 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4095 	vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4096 	vcpu->mmio_size = bytes;
4097 	vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4098 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
4099 	vcpu->mmio_index = 0;
4100 
4101 	return X86EMUL_IO_NEEDED;
4102 }
4103 
4104 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4105 			const void *val, int bytes)
4106 {
4107 	int ret;
4108 
4109 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4110 	if (ret < 0)
4111 		return 0;
4112 	kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4113 	return 1;
4114 }
4115 
4116 static int emulator_write_emulated_onepage(unsigned long addr,
4117 					   const void *val,
4118 					   unsigned int bytes,
4119 					   struct x86_exception *exception,
4120 					   struct kvm_vcpu *vcpu)
4121 {
4122 	gpa_t gpa;
4123 	int handled, ret;
4124 
4125 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, true);
4126 
4127 	if (ret < 0)
4128 		return X86EMUL_PROPAGATE_FAULT;
4129 
4130 	/* For APIC access vmexit */
4131 	if (ret)
4132 		goto mmio;
4133 
4134 	if (emulator_write_phys(vcpu, gpa, val, bytes))
4135 		return X86EMUL_CONTINUE;
4136 
4137 mmio:
4138 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4139 	/*
4140 	 * Is this MMIO handled locally?
4141 	 */
4142 	handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
4143 	if (handled == bytes)
4144 		return X86EMUL_CONTINUE;
4145 
4146 	gpa += handled;
4147 	bytes -= handled;
4148 	val += handled;
4149 
4150 	vcpu->mmio_needed = 1;
4151 	memcpy(vcpu->mmio_data, val, bytes);
4152 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4153 	vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4154 	vcpu->mmio_size = bytes;
4155 	vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4156 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
4157 	memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4158 	vcpu->mmio_index = 0;
4159 
4160 	return X86EMUL_CONTINUE;
4161 }
4162 
4163 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4164 			    unsigned long addr,
4165 			    const void *val,
4166 			    unsigned int bytes,
4167 			    struct x86_exception *exception)
4168 {
4169 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4170 
4171 	/* Crossing a page boundary? */
4172 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4173 		int rc, now;
4174 
4175 		now = -addr & ~PAGE_MASK;
4176 		rc = emulator_write_emulated_onepage(addr, val, now, exception,
4177 						     vcpu);
4178 		if (rc != X86EMUL_CONTINUE)
4179 			return rc;
4180 		addr += now;
4181 		val += now;
4182 		bytes -= now;
4183 	}
4184 	return emulator_write_emulated_onepage(addr, val, bytes, exception,
4185 					       vcpu);
4186 }
4187 
4188 #define CMPXCHG_TYPE(t, ptr, old, new) \
4189 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4190 
4191 #ifdef CONFIG_X86_64
4192 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4193 #else
4194 #  define CMPXCHG64(ptr, old, new) \
4195 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4196 #endif
4197 
4198 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4199 				     unsigned long addr,
4200 				     const void *old,
4201 				     const void *new,
4202 				     unsigned int bytes,
4203 				     struct x86_exception *exception)
4204 {
4205 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4206 	gpa_t gpa;
4207 	struct page *page;
4208 	char *kaddr;
4209 	bool exchanged;
4210 
4211 	/* guests cmpxchg8b have to be emulated atomically */
4212 	if (bytes > 8 || (bytes & (bytes - 1)))
4213 		goto emul_write;
4214 
4215 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4216 
4217 	if (gpa == UNMAPPED_GVA ||
4218 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4219 		goto emul_write;
4220 
4221 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4222 		goto emul_write;
4223 
4224 	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4225 	if (is_error_page(page)) {
4226 		kvm_release_page_clean(page);
4227 		goto emul_write;
4228 	}
4229 
4230 	kaddr = kmap_atomic(page, KM_USER0);
4231 	kaddr += offset_in_page(gpa);
4232 	switch (bytes) {
4233 	case 1:
4234 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4235 		break;
4236 	case 2:
4237 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4238 		break;
4239 	case 4:
4240 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4241 		break;
4242 	case 8:
4243 		exchanged = CMPXCHG64(kaddr, old, new);
4244 		break;
4245 	default:
4246 		BUG();
4247 	}
4248 	kunmap_atomic(kaddr, KM_USER0);
4249 	kvm_release_page_dirty(page);
4250 
4251 	if (!exchanged)
4252 		return X86EMUL_CMPXCHG_FAILED;
4253 
4254 	kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4255 
4256 	return X86EMUL_CONTINUE;
4257 
4258 emul_write:
4259 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4260 
4261 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4262 }
4263 
4264 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4265 {
4266 	/* TODO: String I/O for in kernel device */
4267 	int r;
4268 
4269 	if (vcpu->arch.pio.in)
4270 		r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4271 				    vcpu->arch.pio.size, pd);
4272 	else
4273 		r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4274 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4275 				     pd);
4276 	return r;
4277 }
4278 
4279 
4280 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4281 				    int size, unsigned short port, void *val,
4282 				    unsigned int count)
4283 {
4284 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4285 
4286 	if (vcpu->arch.pio.count)
4287 		goto data_avail;
4288 
4289 	trace_kvm_pio(0, port, size, count);
4290 
4291 	vcpu->arch.pio.port = port;
4292 	vcpu->arch.pio.in = 1;
4293 	vcpu->arch.pio.count  = count;
4294 	vcpu->arch.pio.size = size;
4295 
4296 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4297 	data_avail:
4298 		memcpy(val, vcpu->arch.pio_data, size * count);
4299 		vcpu->arch.pio.count = 0;
4300 		return 1;
4301 	}
4302 
4303 	vcpu->run->exit_reason = KVM_EXIT_IO;
4304 	vcpu->run->io.direction = KVM_EXIT_IO_IN;
4305 	vcpu->run->io.size = size;
4306 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4307 	vcpu->run->io.count = count;
4308 	vcpu->run->io.port = port;
4309 
4310 	return 0;
4311 }
4312 
4313 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4314 				     int size, unsigned short port,
4315 				     const void *val, unsigned int count)
4316 {
4317 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4318 
4319 	trace_kvm_pio(1, port, size, count);
4320 
4321 	vcpu->arch.pio.port = port;
4322 	vcpu->arch.pio.in = 0;
4323 	vcpu->arch.pio.count = count;
4324 	vcpu->arch.pio.size = size;
4325 
4326 	memcpy(vcpu->arch.pio_data, val, size * count);
4327 
4328 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4329 		vcpu->arch.pio.count = 0;
4330 		return 1;
4331 	}
4332 
4333 	vcpu->run->exit_reason = KVM_EXIT_IO;
4334 	vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4335 	vcpu->run->io.size = size;
4336 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4337 	vcpu->run->io.count = count;
4338 	vcpu->run->io.port = port;
4339 
4340 	return 0;
4341 }
4342 
4343 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4344 {
4345 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4346 }
4347 
4348 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4349 {
4350 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4351 }
4352 
4353 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4354 {
4355 	if (!need_emulate_wbinvd(vcpu))
4356 		return X86EMUL_CONTINUE;
4357 
4358 	if (kvm_x86_ops->has_wbinvd_exit()) {
4359 		int cpu = get_cpu();
4360 
4361 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4362 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4363 				wbinvd_ipi, NULL, 1);
4364 		put_cpu();
4365 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4366 	} else
4367 		wbinvd();
4368 	return X86EMUL_CONTINUE;
4369 }
4370 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4371 
4372 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4373 {
4374 	kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4375 }
4376 
4377 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4378 {
4379 	return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4380 }
4381 
4382 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4383 {
4384 
4385 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4386 }
4387 
4388 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4389 {
4390 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4391 }
4392 
4393 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4394 {
4395 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4396 	unsigned long value;
4397 
4398 	switch (cr) {
4399 	case 0:
4400 		value = kvm_read_cr0(vcpu);
4401 		break;
4402 	case 2:
4403 		value = vcpu->arch.cr2;
4404 		break;
4405 	case 3:
4406 		value = kvm_read_cr3(vcpu);
4407 		break;
4408 	case 4:
4409 		value = kvm_read_cr4(vcpu);
4410 		break;
4411 	case 8:
4412 		value = kvm_get_cr8(vcpu);
4413 		break;
4414 	default:
4415 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4416 		return 0;
4417 	}
4418 
4419 	return value;
4420 }
4421 
4422 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4423 {
4424 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4425 	int res = 0;
4426 
4427 	switch (cr) {
4428 	case 0:
4429 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4430 		break;
4431 	case 2:
4432 		vcpu->arch.cr2 = val;
4433 		break;
4434 	case 3:
4435 		res = kvm_set_cr3(vcpu, val);
4436 		break;
4437 	case 4:
4438 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4439 		break;
4440 	case 8:
4441 		res = kvm_set_cr8(vcpu, val);
4442 		break;
4443 	default:
4444 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4445 		res = -1;
4446 	}
4447 
4448 	return res;
4449 }
4450 
4451 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4452 {
4453 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4454 }
4455 
4456 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4457 {
4458 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4459 }
4460 
4461 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4462 {
4463 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4464 }
4465 
4466 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4467 {
4468 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4469 }
4470 
4471 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4472 {
4473 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4474 }
4475 
4476 static unsigned long emulator_get_cached_segment_base(
4477 	struct x86_emulate_ctxt *ctxt, int seg)
4478 {
4479 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4480 }
4481 
4482 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4483 				 struct desc_struct *desc, u32 *base3,
4484 				 int seg)
4485 {
4486 	struct kvm_segment var;
4487 
4488 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4489 	*selector = var.selector;
4490 
4491 	if (var.unusable)
4492 		return false;
4493 
4494 	if (var.g)
4495 		var.limit >>= 12;
4496 	set_desc_limit(desc, var.limit);
4497 	set_desc_base(desc, (unsigned long)var.base);
4498 #ifdef CONFIG_X86_64
4499 	if (base3)
4500 		*base3 = var.base >> 32;
4501 #endif
4502 	desc->type = var.type;
4503 	desc->s = var.s;
4504 	desc->dpl = var.dpl;
4505 	desc->p = var.present;
4506 	desc->avl = var.avl;
4507 	desc->l = var.l;
4508 	desc->d = var.db;
4509 	desc->g = var.g;
4510 
4511 	return true;
4512 }
4513 
4514 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4515 				 struct desc_struct *desc, u32 base3,
4516 				 int seg)
4517 {
4518 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4519 	struct kvm_segment var;
4520 
4521 	var.selector = selector;
4522 	var.base = get_desc_base(desc);
4523 #ifdef CONFIG_X86_64
4524 	var.base |= ((u64)base3) << 32;
4525 #endif
4526 	var.limit = get_desc_limit(desc);
4527 	if (desc->g)
4528 		var.limit = (var.limit << 12) | 0xfff;
4529 	var.type = desc->type;
4530 	var.present = desc->p;
4531 	var.dpl = desc->dpl;
4532 	var.db = desc->d;
4533 	var.s = desc->s;
4534 	var.l = desc->l;
4535 	var.g = desc->g;
4536 	var.avl = desc->avl;
4537 	var.present = desc->p;
4538 	var.unusable = !var.present;
4539 	var.padding = 0;
4540 
4541 	kvm_set_segment(vcpu, &var, seg);
4542 	return;
4543 }
4544 
4545 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4546 			    u32 msr_index, u64 *pdata)
4547 {
4548 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4549 }
4550 
4551 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4552 			    u32 msr_index, u64 data)
4553 {
4554 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4555 }
4556 
4557 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4558 {
4559 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4560 }
4561 
4562 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4563 {
4564 	preempt_disable();
4565 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4566 	/*
4567 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4568 	 * so it may be clear at this point.
4569 	 */
4570 	clts();
4571 }
4572 
4573 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4574 {
4575 	preempt_enable();
4576 }
4577 
4578 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4579 			      struct x86_instruction_info *info,
4580 			      enum x86_intercept_stage stage)
4581 {
4582 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4583 }
4584 
4585 static struct x86_emulate_ops emulate_ops = {
4586 	.read_std            = kvm_read_guest_virt_system,
4587 	.write_std           = kvm_write_guest_virt_system,
4588 	.fetch               = kvm_fetch_guest_virt,
4589 	.read_emulated       = emulator_read_emulated,
4590 	.write_emulated      = emulator_write_emulated,
4591 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
4592 	.invlpg              = emulator_invlpg,
4593 	.pio_in_emulated     = emulator_pio_in_emulated,
4594 	.pio_out_emulated    = emulator_pio_out_emulated,
4595 	.get_segment         = emulator_get_segment,
4596 	.set_segment         = emulator_set_segment,
4597 	.get_cached_segment_base = emulator_get_cached_segment_base,
4598 	.get_gdt             = emulator_get_gdt,
4599 	.get_idt	     = emulator_get_idt,
4600 	.set_gdt             = emulator_set_gdt,
4601 	.set_idt	     = emulator_set_idt,
4602 	.get_cr              = emulator_get_cr,
4603 	.set_cr              = emulator_set_cr,
4604 	.cpl                 = emulator_get_cpl,
4605 	.get_dr              = emulator_get_dr,
4606 	.set_dr              = emulator_set_dr,
4607 	.set_msr             = emulator_set_msr,
4608 	.get_msr             = emulator_get_msr,
4609 	.halt                = emulator_halt,
4610 	.wbinvd              = emulator_wbinvd,
4611 	.fix_hypercall       = emulator_fix_hypercall,
4612 	.get_fpu             = emulator_get_fpu,
4613 	.put_fpu             = emulator_put_fpu,
4614 	.intercept           = emulator_intercept,
4615 };
4616 
4617 static void cache_all_regs(struct kvm_vcpu *vcpu)
4618 {
4619 	kvm_register_read(vcpu, VCPU_REGS_RAX);
4620 	kvm_register_read(vcpu, VCPU_REGS_RSP);
4621 	kvm_register_read(vcpu, VCPU_REGS_RIP);
4622 	vcpu->arch.regs_dirty = ~0;
4623 }
4624 
4625 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4626 {
4627 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4628 	/*
4629 	 * an sti; sti; sequence only disable interrupts for the first
4630 	 * instruction. So, if the last instruction, be it emulated or
4631 	 * not, left the system with the INT_STI flag enabled, it
4632 	 * means that the last instruction is an sti. We should not
4633 	 * leave the flag on in this case. The same goes for mov ss
4634 	 */
4635 	if (!(int_shadow & mask))
4636 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4637 }
4638 
4639 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4640 {
4641 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4642 	if (ctxt->exception.vector == PF_VECTOR)
4643 		kvm_propagate_fault(vcpu, &ctxt->exception);
4644 	else if (ctxt->exception.error_code_valid)
4645 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4646 				      ctxt->exception.error_code);
4647 	else
4648 		kvm_queue_exception(vcpu, ctxt->exception.vector);
4649 }
4650 
4651 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4652 			      const unsigned long *regs)
4653 {
4654 	memset(&ctxt->twobyte, 0,
4655 	       (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4656 	memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4657 
4658 	ctxt->fetch.start = 0;
4659 	ctxt->fetch.end = 0;
4660 	ctxt->io_read.pos = 0;
4661 	ctxt->io_read.end = 0;
4662 	ctxt->mem_read.pos = 0;
4663 	ctxt->mem_read.end = 0;
4664 }
4665 
4666 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4667 {
4668 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4669 	int cs_db, cs_l;
4670 
4671 	/*
4672 	 * TODO: fix emulate.c to use guest_read/write_register
4673 	 * instead of direct ->regs accesses, can save hundred cycles
4674 	 * on Intel for instructions that don't read/change RSP, for
4675 	 * for example.
4676 	 */
4677 	cache_all_regs(vcpu);
4678 
4679 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4680 
4681 	ctxt->eflags = kvm_get_rflags(vcpu);
4682 	ctxt->eip = kvm_rip_read(vcpu);
4683 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
4684 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
4685 		     cs_l				? X86EMUL_MODE_PROT64 :
4686 		     cs_db				? X86EMUL_MODE_PROT32 :
4687 							  X86EMUL_MODE_PROT16;
4688 	ctxt->guest_mode = is_guest_mode(vcpu);
4689 
4690 	init_decode_cache(ctxt, vcpu->arch.regs);
4691 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4692 }
4693 
4694 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4695 {
4696 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4697 	int ret;
4698 
4699 	init_emulate_ctxt(vcpu);
4700 
4701 	ctxt->op_bytes = 2;
4702 	ctxt->ad_bytes = 2;
4703 	ctxt->_eip = ctxt->eip + inc_eip;
4704 	ret = emulate_int_real(ctxt, irq);
4705 
4706 	if (ret != X86EMUL_CONTINUE)
4707 		return EMULATE_FAIL;
4708 
4709 	ctxt->eip = ctxt->_eip;
4710 	memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4711 	kvm_rip_write(vcpu, ctxt->eip);
4712 	kvm_set_rflags(vcpu, ctxt->eflags);
4713 
4714 	if (irq == NMI_VECTOR)
4715 		vcpu->arch.nmi_pending = false;
4716 	else
4717 		vcpu->arch.interrupt.pending = false;
4718 
4719 	return EMULATE_DONE;
4720 }
4721 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4722 
4723 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4724 {
4725 	int r = EMULATE_DONE;
4726 
4727 	++vcpu->stat.insn_emulation_fail;
4728 	trace_kvm_emulate_insn_failed(vcpu);
4729 	if (!is_guest_mode(vcpu)) {
4730 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4731 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4732 		vcpu->run->internal.ndata = 0;
4733 		r = EMULATE_FAIL;
4734 	}
4735 	kvm_queue_exception(vcpu, UD_VECTOR);
4736 
4737 	return r;
4738 }
4739 
4740 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4741 {
4742 	gpa_t gpa;
4743 
4744 	if (tdp_enabled)
4745 		return false;
4746 
4747 	/*
4748 	 * if emulation was due to access to shadowed page table
4749 	 * and it failed try to unshadow page and re-entetr the
4750 	 * guest to let CPU execute the instruction.
4751 	 */
4752 	if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4753 		return true;
4754 
4755 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4756 
4757 	if (gpa == UNMAPPED_GVA)
4758 		return true; /* let cpu generate fault */
4759 
4760 	if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4761 		return true;
4762 
4763 	return false;
4764 }
4765 
4766 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4767 			    unsigned long cr2,
4768 			    int emulation_type,
4769 			    void *insn,
4770 			    int insn_len)
4771 {
4772 	int r;
4773 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4774 	bool writeback = true;
4775 
4776 	kvm_clear_exception_queue(vcpu);
4777 
4778 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4779 		init_emulate_ctxt(vcpu);
4780 		ctxt->interruptibility = 0;
4781 		ctxt->have_exception = false;
4782 		ctxt->perm_ok = false;
4783 
4784 		ctxt->only_vendor_specific_insn
4785 			= emulation_type & EMULTYPE_TRAP_UD;
4786 
4787 		r = x86_decode_insn(ctxt, insn, insn_len);
4788 
4789 		trace_kvm_emulate_insn_start(vcpu);
4790 		++vcpu->stat.insn_emulation;
4791 		if (r)  {
4792 			if (emulation_type & EMULTYPE_TRAP_UD)
4793 				return EMULATE_FAIL;
4794 			if (reexecute_instruction(vcpu, cr2))
4795 				return EMULATE_DONE;
4796 			if (emulation_type & EMULTYPE_SKIP)
4797 				return EMULATE_FAIL;
4798 			return handle_emulation_failure(vcpu);
4799 		}
4800 	}
4801 
4802 	if (emulation_type & EMULTYPE_SKIP) {
4803 		kvm_rip_write(vcpu, ctxt->_eip);
4804 		return EMULATE_DONE;
4805 	}
4806 
4807 	/* this is needed for vmware backdoor interface to work since it
4808 	   changes registers values  during IO operation */
4809 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4810 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4811 		memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4812 	}
4813 
4814 restart:
4815 	r = x86_emulate_insn(ctxt);
4816 
4817 	if (r == EMULATION_INTERCEPTED)
4818 		return EMULATE_DONE;
4819 
4820 	if (r == EMULATION_FAILED) {
4821 		if (reexecute_instruction(vcpu, cr2))
4822 			return EMULATE_DONE;
4823 
4824 		return handle_emulation_failure(vcpu);
4825 	}
4826 
4827 	if (ctxt->have_exception) {
4828 		inject_emulated_exception(vcpu);
4829 		r = EMULATE_DONE;
4830 	} else if (vcpu->arch.pio.count) {
4831 		if (!vcpu->arch.pio.in)
4832 			vcpu->arch.pio.count = 0;
4833 		else
4834 			writeback = false;
4835 		r = EMULATE_DO_MMIO;
4836 	} else if (vcpu->mmio_needed) {
4837 		if (!vcpu->mmio_is_write)
4838 			writeback = false;
4839 		r = EMULATE_DO_MMIO;
4840 	} else if (r == EMULATION_RESTART)
4841 		goto restart;
4842 	else
4843 		r = EMULATE_DONE;
4844 
4845 	if (writeback) {
4846 		toggle_interruptibility(vcpu, ctxt->interruptibility);
4847 		kvm_set_rflags(vcpu, ctxt->eflags);
4848 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4849 		memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4850 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4851 		kvm_rip_write(vcpu, ctxt->eip);
4852 	} else
4853 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4854 
4855 	return r;
4856 }
4857 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4858 
4859 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4860 {
4861 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4862 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4863 					    size, port, &val, 1);
4864 	/* do not return to emulator after return from userspace */
4865 	vcpu->arch.pio.count = 0;
4866 	return ret;
4867 }
4868 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4869 
4870 static void tsc_bad(void *info)
4871 {
4872 	__this_cpu_write(cpu_tsc_khz, 0);
4873 }
4874 
4875 static void tsc_khz_changed(void *data)
4876 {
4877 	struct cpufreq_freqs *freq = data;
4878 	unsigned long khz = 0;
4879 
4880 	if (data)
4881 		khz = freq->new;
4882 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4883 		khz = cpufreq_quick_get(raw_smp_processor_id());
4884 	if (!khz)
4885 		khz = tsc_khz;
4886 	__this_cpu_write(cpu_tsc_khz, khz);
4887 }
4888 
4889 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4890 				     void *data)
4891 {
4892 	struct cpufreq_freqs *freq = data;
4893 	struct kvm *kvm;
4894 	struct kvm_vcpu *vcpu;
4895 	int i, send_ipi = 0;
4896 
4897 	/*
4898 	 * We allow guests to temporarily run on slowing clocks,
4899 	 * provided we notify them after, or to run on accelerating
4900 	 * clocks, provided we notify them before.  Thus time never
4901 	 * goes backwards.
4902 	 *
4903 	 * However, we have a problem.  We can't atomically update
4904 	 * the frequency of a given CPU from this function; it is
4905 	 * merely a notifier, which can be called from any CPU.
4906 	 * Changing the TSC frequency at arbitrary points in time
4907 	 * requires a recomputation of local variables related to
4908 	 * the TSC for each VCPU.  We must flag these local variables
4909 	 * to be updated and be sure the update takes place with the
4910 	 * new frequency before any guests proceed.
4911 	 *
4912 	 * Unfortunately, the combination of hotplug CPU and frequency
4913 	 * change creates an intractable locking scenario; the order
4914 	 * of when these callouts happen is undefined with respect to
4915 	 * CPU hotplug, and they can race with each other.  As such,
4916 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4917 	 * undefined; you can actually have a CPU frequency change take
4918 	 * place in between the computation of X and the setting of the
4919 	 * variable.  To protect against this problem, all updates of
4920 	 * the per_cpu tsc_khz variable are done in an interrupt
4921 	 * protected IPI, and all callers wishing to update the value
4922 	 * must wait for a synchronous IPI to complete (which is trivial
4923 	 * if the caller is on the CPU already).  This establishes the
4924 	 * necessary total order on variable updates.
4925 	 *
4926 	 * Note that because a guest time update may take place
4927 	 * anytime after the setting of the VCPU's request bit, the
4928 	 * correct TSC value must be set before the request.  However,
4929 	 * to ensure the update actually makes it to any guest which
4930 	 * starts running in hardware virtualization between the set
4931 	 * and the acquisition of the spinlock, we must also ping the
4932 	 * CPU after setting the request bit.
4933 	 *
4934 	 */
4935 
4936 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4937 		return 0;
4938 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4939 		return 0;
4940 
4941 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4942 
4943 	raw_spin_lock(&kvm_lock);
4944 	list_for_each_entry(kvm, &vm_list, vm_list) {
4945 		kvm_for_each_vcpu(i, vcpu, kvm) {
4946 			if (vcpu->cpu != freq->cpu)
4947 				continue;
4948 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4949 			if (vcpu->cpu != smp_processor_id())
4950 				send_ipi = 1;
4951 		}
4952 	}
4953 	raw_spin_unlock(&kvm_lock);
4954 
4955 	if (freq->old < freq->new && send_ipi) {
4956 		/*
4957 		 * We upscale the frequency.  Must make the guest
4958 		 * doesn't see old kvmclock values while running with
4959 		 * the new frequency, otherwise we risk the guest sees
4960 		 * time go backwards.
4961 		 *
4962 		 * In case we update the frequency for another cpu
4963 		 * (which might be in guest context) send an interrupt
4964 		 * to kick the cpu out of guest context.  Next time
4965 		 * guest context is entered kvmclock will be updated,
4966 		 * so the guest will not see stale values.
4967 		 */
4968 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4969 	}
4970 	return 0;
4971 }
4972 
4973 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4974 	.notifier_call  = kvmclock_cpufreq_notifier
4975 };
4976 
4977 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4978 					unsigned long action, void *hcpu)
4979 {
4980 	unsigned int cpu = (unsigned long)hcpu;
4981 
4982 	switch (action) {
4983 		case CPU_ONLINE:
4984 		case CPU_DOWN_FAILED:
4985 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4986 			break;
4987 		case CPU_DOWN_PREPARE:
4988 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
4989 			break;
4990 	}
4991 	return NOTIFY_OK;
4992 }
4993 
4994 static struct notifier_block kvmclock_cpu_notifier_block = {
4995 	.notifier_call  = kvmclock_cpu_notifier,
4996 	.priority = -INT_MAX
4997 };
4998 
4999 static void kvm_timer_init(void)
5000 {
5001 	int cpu;
5002 
5003 	max_tsc_khz = tsc_khz;
5004 	register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5005 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5006 #ifdef CONFIG_CPU_FREQ
5007 		struct cpufreq_policy policy;
5008 		memset(&policy, 0, sizeof(policy));
5009 		cpu = get_cpu();
5010 		cpufreq_get_policy(&policy, cpu);
5011 		if (policy.cpuinfo.max_freq)
5012 			max_tsc_khz = policy.cpuinfo.max_freq;
5013 		put_cpu();
5014 #endif
5015 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5016 					  CPUFREQ_TRANSITION_NOTIFIER);
5017 	}
5018 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5019 	for_each_online_cpu(cpu)
5020 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5021 }
5022 
5023 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5024 
5025 static int kvm_is_in_guest(void)
5026 {
5027 	return percpu_read(current_vcpu) != NULL;
5028 }
5029 
5030 static int kvm_is_user_mode(void)
5031 {
5032 	int user_mode = 3;
5033 
5034 	if (percpu_read(current_vcpu))
5035 		user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5036 
5037 	return user_mode != 0;
5038 }
5039 
5040 static unsigned long kvm_get_guest_ip(void)
5041 {
5042 	unsigned long ip = 0;
5043 
5044 	if (percpu_read(current_vcpu))
5045 		ip = kvm_rip_read(percpu_read(current_vcpu));
5046 
5047 	return ip;
5048 }
5049 
5050 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5051 	.is_in_guest		= kvm_is_in_guest,
5052 	.is_user_mode		= kvm_is_user_mode,
5053 	.get_guest_ip		= kvm_get_guest_ip,
5054 };
5055 
5056 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5057 {
5058 	percpu_write(current_vcpu, vcpu);
5059 }
5060 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5061 
5062 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5063 {
5064 	percpu_write(current_vcpu, NULL);
5065 }
5066 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5067 
5068 static void kvm_set_mmio_spte_mask(void)
5069 {
5070 	u64 mask;
5071 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5072 
5073 	/*
5074 	 * Set the reserved bits and the present bit of an paging-structure
5075 	 * entry to generate page fault with PFER.RSV = 1.
5076 	 */
5077 	mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5078 	mask |= 1ull;
5079 
5080 #ifdef CONFIG_X86_64
5081 	/*
5082 	 * If reserved bit is not supported, clear the present bit to disable
5083 	 * mmio page fault.
5084 	 */
5085 	if (maxphyaddr == 52)
5086 		mask &= ~1ull;
5087 #endif
5088 
5089 	kvm_mmu_set_mmio_spte_mask(mask);
5090 }
5091 
5092 int kvm_arch_init(void *opaque)
5093 {
5094 	int r;
5095 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5096 
5097 	if (kvm_x86_ops) {
5098 		printk(KERN_ERR "kvm: already loaded the other module\n");
5099 		r = -EEXIST;
5100 		goto out;
5101 	}
5102 
5103 	if (!ops->cpu_has_kvm_support()) {
5104 		printk(KERN_ERR "kvm: no hardware support\n");
5105 		r = -EOPNOTSUPP;
5106 		goto out;
5107 	}
5108 	if (ops->disabled_by_bios()) {
5109 		printk(KERN_ERR "kvm: disabled by bios\n");
5110 		r = -EOPNOTSUPP;
5111 		goto out;
5112 	}
5113 
5114 	r = kvm_mmu_module_init();
5115 	if (r)
5116 		goto out;
5117 
5118 	kvm_set_mmio_spte_mask();
5119 	kvm_init_msr_list();
5120 
5121 	kvm_x86_ops = ops;
5122 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5123 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5124 
5125 	kvm_timer_init();
5126 
5127 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5128 
5129 	if (cpu_has_xsave)
5130 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5131 
5132 	return 0;
5133 
5134 out:
5135 	return r;
5136 }
5137 
5138 void kvm_arch_exit(void)
5139 {
5140 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5141 
5142 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5143 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5144 					    CPUFREQ_TRANSITION_NOTIFIER);
5145 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5146 	kvm_x86_ops = NULL;
5147 	kvm_mmu_module_exit();
5148 }
5149 
5150 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5151 {
5152 	++vcpu->stat.halt_exits;
5153 	if (irqchip_in_kernel(vcpu->kvm)) {
5154 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5155 		return 1;
5156 	} else {
5157 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5158 		return 0;
5159 	}
5160 }
5161 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5162 
5163 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5164 			   unsigned long a1)
5165 {
5166 	if (is_long_mode(vcpu))
5167 		return a0;
5168 	else
5169 		return a0 | ((gpa_t)a1 << 32);
5170 }
5171 
5172 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5173 {
5174 	u64 param, ingpa, outgpa, ret;
5175 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5176 	bool fast, longmode;
5177 	int cs_db, cs_l;
5178 
5179 	/*
5180 	 * hypercall generates UD from non zero cpl and real mode
5181 	 * per HYPER-V spec
5182 	 */
5183 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5184 		kvm_queue_exception(vcpu, UD_VECTOR);
5185 		return 0;
5186 	}
5187 
5188 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5189 	longmode = is_long_mode(vcpu) && cs_l == 1;
5190 
5191 	if (!longmode) {
5192 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5193 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5194 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5195 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5196 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5197 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5198 	}
5199 #ifdef CONFIG_X86_64
5200 	else {
5201 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5202 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5203 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5204 	}
5205 #endif
5206 
5207 	code = param & 0xffff;
5208 	fast = (param >> 16) & 0x1;
5209 	rep_cnt = (param >> 32) & 0xfff;
5210 	rep_idx = (param >> 48) & 0xfff;
5211 
5212 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5213 
5214 	switch (code) {
5215 	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5216 		kvm_vcpu_on_spin(vcpu);
5217 		break;
5218 	default:
5219 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
5220 		break;
5221 	}
5222 
5223 	ret = res | (((u64)rep_done & 0xfff) << 32);
5224 	if (longmode) {
5225 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5226 	} else {
5227 		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5228 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5229 	}
5230 
5231 	return 1;
5232 }
5233 
5234 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5235 {
5236 	unsigned long nr, a0, a1, a2, a3, ret;
5237 	int r = 1;
5238 
5239 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5240 		return kvm_hv_hypercall(vcpu);
5241 
5242 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5243 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5244 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5245 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5246 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5247 
5248 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5249 
5250 	if (!is_long_mode(vcpu)) {
5251 		nr &= 0xFFFFFFFF;
5252 		a0 &= 0xFFFFFFFF;
5253 		a1 &= 0xFFFFFFFF;
5254 		a2 &= 0xFFFFFFFF;
5255 		a3 &= 0xFFFFFFFF;
5256 	}
5257 
5258 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5259 		ret = -KVM_EPERM;
5260 		goto out;
5261 	}
5262 
5263 	switch (nr) {
5264 	case KVM_HC_VAPIC_POLL_IRQ:
5265 		ret = 0;
5266 		break;
5267 	case KVM_HC_MMU_OP:
5268 		r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5269 		break;
5270 	default:
5271 		ret = -KVM_ENOSYS;
5272 		break;
5273 	}
5274 out:
5275 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5276 	++vcpu->stat.hypercalls;
5277 	return r;
5278 }
5279 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5280 
5281 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5282 {
5283 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5284 	char instruction[3];
5285 	unsigned long rip = kvm_rip_read(vcpu);
5286 
5287 	/*
5288 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
5289 	 * to ensure that the updated hypercall appears atomically across all
5290 	 * VCPUs.
5291 	 */
5292 	kvm_mmu_zap_all(vcpu->kvm);
5293 
5294 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
5295 
5296 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5297 }
5298 
5299 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5300 {
5301 	struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5302 	int j, nent = vcpu->arch.cpuid_nent;
5303 
5304 	e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5305 	/* when no next entry is found, the current entry[i] is reselected */
5306 	for (j = i + 1; ; j = (j + 1) % nent) {
5307 		struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5308 		if (ej->function == e->function) {
5309 			ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5310 			return j;
5311 		}
5312 	}
5313 	return 0; /* silence gcc, even though control never reaches here */
5314 }
5315 
5316 /* find an entry with matching function, matching index (if needed), and that
5317  * should be read next (if it's stateful) */
5318 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5319 	u32 function, u32 index)
5320 {
5321 	if (e->function != function)
5322 		return 0;
5323 	if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5324 		return 0;
5325 	if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5326 	    !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5327 		return 0;
5328 	return 1;
5329 }
5330 
5331 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5332 					      u32 function, u32 index)
5333 {
5334 	int i;
5335 	struct kvm_cpuid_entry2 *best = NULL;
5336 
5337 	for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5338 		struct kvm_cpuid_entry2 *e;
5339 
5340 		e = &vcpu->arch.cpuid_entries[i];
5341 		if (is_matching_cpuid_entry(e, function, index)) {
5342 			if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5343 				move_to_next_stateful_cpuid_entry(vcpu, i);
5344 			best = e;
5345 			break;
5346 		}
5347 	}
5348 	return best;
5349 }
5350 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5351 
5352 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5353 {
5354 	struct kvm_cpuid_entry2 *best;
5355 
5356 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5357 	if (!best || best->eax < 0x80000008)
5358 		goto not_found;
5359 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5360 	if (best)
5361 		return best->eax & 0xff;
5362 not_found:
5363 	return 36;
5364 }
5365 
5366 /*
5367  * If no match is found, check whether we exceed the vCPU's limit
5368  * and return the content of the highest valid _standard_ leaf instead.
5369  * This is to satisfy the CPUID specification.
5370  */
5371 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5372                                                   u32 function, u32 index)
5373 {
5374 	struct kvm_cpuid_entry2 *maxlevel;
5375 
5376 	maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5377 	if (!maxlevel || maxlevel->eax >= function)
5378 		return NULL;
5379 	if (function & 0x80000000) {
5380 		maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5381 		if (!maxlevel)
5382 			return NULL;
5383 	}
5384 	return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5385 }
5386 
5387 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5388 {
5389 	u32 function, index;
5390 	struct kvm_cpuid_entry2 *best;
5391 
5392 	function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5393 	index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5394 	kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5395 	kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5396 	kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5397 	kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5398 	best = kvm_find_cpuid_entry(vcpu, function, index);
5399 
5400 	if (!best)
5401 		best = check_cpuid_limit(vcpu, function, index);
5402 
5403 	if (best) {
5404 		kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5405 		kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5406 		kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5407 		kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5408 	}
5409 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5410 	trace_kvm_cpuid(function,
5411 			kvm_register_read(vcpu, VCPU_REGS_RAX),
5412 			kvm_register_read(vcpu, VCPU_REGS_RBX),
5413 			kvm_register_read(vcpu, VCPU_REGS_RCX),
5414 			kvm_register_read(vcpu, VCPU_REGS_RDX));
5415 }
5416 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5417 
5418 /*
5419  * Check if userspace requested an interrupt window, and that the
5420  * interrupt window is open.
5421  *
5422  * No need to exit to userspace if we already have an interrupt queued.
5423  */
5424 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5425 {
5426 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5427 		vcpu->run->request_interrupt_window &&
5428 		kvm_arch_interrupt_allowed(vcpu));
5429 }
5430 
5431 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5432 {
5433 	struct kvm_run *kvm_run = vcpu->run;
5434 
5435 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5436 	kvm_run->cr8 = kvm_get_cr8(vcpu);
5437 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
5438 	if (irqchip_in_kernel(vcpu->kvm))
5439 		kvm_run->ready_for_interrupt_injection = 1;
5440 	else
5441 		kvm_run->ready_for_interrupt_injection =
5442 			kvm_arch_interrupt_allowed(vcpu) &&
5443 			!kvm_cpu_has_interrupt(vcpu) &&
5444 			!kvm_event_needs_reinjection(vcpu);
5445 }
5446 
5447 static void vapic_enter(struct kvm_vcpu *vcpu)
5448 {
5449 	struct kvm_lapic *apic = vcpu->arch.apic;
5450 	struct page *page;
5451 
5452 	if (!apic || !apic->vapic_addr)
5453 		return;
5454 
5455 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5456 
5457 	vcpu->arch.apic->vapic_page = page;
5458 }
5459 
5460 static void vapic_exit(struct kvm_vcpu *vcpu)
5461 {
5462 	struct kvm_lapic *apic = vcpu->arch.apic;
5463 	int idx;
5464 
5465 	if (!apic || !apic->vapic_addr)
5466 		return;
5467 
5468 	idx = srcu_read_lock(&vcpu->kvm->srcu);
5469 	kvm_release_page_dirty(apic->vapic_page);
5470 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5471 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
5472 }
5473 
5474 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5475 {
5476 	int max_irr, tpr;
5477 
5478 	if (!kvm_x86_ops->update_cr8_intercept)
5479 		return;
5480 
5481 	if (!vcpu->arch.apic)
5482 		return;
5483 
5484 	if (!vcpu->arch.apic->vapic_addr)
5485 		max_irr = kvm_lapic_find_highest_irr(vcpu);
5486 	else
5487 		max_irr = -1;
5488 
5489 	if (max_irr != -1)
5490 		max_irr >>= 4;
5491 
5492 	tpr = kvm_lapic_get_cr8(vcpu);
5493 
5494 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5495 }
5496 
5497 static void inject_pending_event(struct kvm_vcpu *vcpu)
5498 {
5499 	/* try to reinject previous events if any */
5500 	if (vcpu->arch.exception.pending) {
5501 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
5502 					vcpu->arch.exception.has_error_code,
5503 					vcpu->arch.exception.error_code);
5504 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5505 					  vcpu->arch.exception.has_error_code,
5506 					  vcpu->arch.exception.error_code,
5507 					  vcpu->arch.exception.reinject);
5508 		return;
5509 	}
5510 
5511 	if (vcpu->arch.nmi_injected) {
5512 		kvm_x86_ops->set_nmi(vcpu);
5513 		return;
5514 	}
5515 
5516 	if (vcpu->arch.interrupt.pending) {
5517 		kvm_x86_ops->set_irq(vcpu);
5518 		return;
5519 	}
5520 
5521 	/* try to inject new event if pending */
5522 	if (vcpu->arch.nmi_pending) {
5523 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
5524 			vcpu->arch.nmi_pending = false;
5525 			vcpu->arch.nmi_injected = true;
5526 			kvm_x86_ops->set_nmi(vcpu);
5527 		}
5528 	} else if (kvm_cpu_has_interrupt(vcpu)) {
5529 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5530 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5531 					    false);
5532 			kvm_x86_ops->set_irq(vcpu);
5533 		}
5534 	}
5535 }
5536 
5537 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5538 {
5539 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5540 			!vcpu->guest_xcr0_loaded) {
5541 		/* kvm_set_xcr() also depends on this */
5542 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5543 		vcpu->guest_xcr0_loaded = 1;
5544 	}
5545 }
5546 
5547 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5548 {
5549 	if (vcpu->guest_xcr0_loaded) {
5550 		if (vcpu->arch.xcr0 != host_xcr0)
5551 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5552 		vcpu->guest_xcr0_loaded = 0;
5553 	}
5554 }
5555 
5556 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5557 {
5558 	int r;
5559 	bool nmi_pending;
5560 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5561 		vcpu->run->request_interrupt_window;
5562 
5563 	if (vcpu->requests) {
5564 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5565 			kvm_mmu_unload(vcpu);
5566 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5567 			__kvm_migrate_timers(vcpu);
5568 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5569 			r = kvm_guest_time_update(vcpu);
5570 			if (unlikely(r))
5571 				goto out;
5572 		}
5573 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5574 			kvm_mmu_sync_roots(vcpu);
5575 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5576 			kvm_x86_ops->tlb_flush(vcpu);
5577 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5578 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5579 			r = 0;
5580 			goto out;
5581 		}
5582 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5583 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5584 			r = 0;
5585 			goto out;
5586 		}
5587 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5588 			vcpu->fpu_active = 0;
5589 			kvm_x86_ops->fpu_deactivate(vcpu);
5590 		}
5591 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5592 			/* Page is swapped out. Do synthetic halt */
5593 			vcpu->arch.apf.halted = true;
5594 			r = 1;
5595 			goto out;
5596 		}
5597 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5598 			record_steal_time(vcpu);
5599 
5600 	}
5601 
5602 	r = kvm_mmu_reload(vcpu);
5603 	if (unlikely(r))
5604 		goto out;
5605 
5606 	/*
5607 	 * An NMI can be injected between local nmi_pending read and
5608 	 * vcpu->arch.nmi_pending read inside inject_pending_event().
5609 	 * But in that case, KVM_REQ_EVENT will be set, which makes
5610 	 * the race described above benign.
5611 	 */
5612 	nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5613 
5614 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5615 		inject_pending_event(vcpu);
5616 
5617 		/* enable NMI/IRQ window open exits if needed */
5618 		if (nmi_pending)
5619 			kvm_x86_ops->enable_nmi_window(vcpu);
5620 		else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5621 			kvm_x86_ops->enable_irq_window(vcpu);
5622 
5623 		if (kvm_lapic_enabled(vcpu)) {
5624 			update_cr8_intercept(vcpu);
5625 			kvm_lapic_sync_to_vapic(vcpu);
5626 		}
5627 	}
5628 
5629 	preempt_disable();
5630 
5631 	kvm_x86_ops->prepare_guest_switch(vcpu);
5632 	if (vcpu->fpu_active)
5633 		kvm_load_guest_fpu(vcpu);
5634 	kvm_load_guest_xcr0(vcpu);
5635 
5636 	vcpu->mode = IN_GUEST_MODE;
5637 
5638 	/* We should set ->mode before check ->requests,
5639 	 * see the comment in make_all_cpus_request.
5640 	 */
5641 	smp_mb();
5642 
5643 	local_irq_disable();
5644 
5645 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5646 	    || need_resched() || signal_pending(current)) {
5647 		vcpu->mode = OUTSIDE_GUEST_MODE;
5648 		smp_wmb();
5649 		local_irq_enable();
5650 		preempt_enable();
5651 		kvm_x86_ops->cancel_injection(vcpu);
5652 		r = 1;
5653 		goto out;
5654 	}
5655 
5656 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5657 
5658 	kvm_guest_enter();
5659 
5660 	if (unlikely(vcpu->arch.switch_db_regs)) {
5661 		set_debugreg(0, 7);
5662 		set_debugreg(vcpu->arch.eff_db[0], 0);
5663 		set_debugreg(vcpu->arch.eff_db[1], 1);
5664 		set_debugreg(vcpu->arch.eff_db[2], 2);
5665 		set_debugreg(vcpu->arch.eff_db[3], 3);
5666 	}
5667 
5668 	trace_kvm_entry(vcpu->vcpu_id);
5669 	kvm_x86_ops->run(vcpu);
5670 
5671 	/*
5672 	 * If the guest has used debug registers, at least dr7
5673 	 * will be disabled while returning to the host.
5674 	 * If we don't have active breakpoints in the host, we don't
5675 	 * care about the messed up debug address registers. But if
5676 	 * we have some of them active, restore the old state.
5677 	 */
5678 	if (hw_breakpoint_active())
5679 		hw_breakpoint_restore();
5680 
5681 	kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5682 
5683 	vcpu->mode = OUTSIDE_GUEST_MODE;
5684 	smp_wmb();
5685 	local_irq_enable();
5686 
5687 	++vcpu->stat.exits;
5688 
5689 	/*
5690 	 * We must have an instruction between local_irq_enable() and
5691 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5692 	 * the interrupt shadow.  The stat.exits increment will do nicely.
5693 	 * But we need to prevent reordering, hence this barrier():
5694 	 */
5695 	barrier();
5696 
5697 	kvm_guest_exit();
5698 
5699 	preempt_enable();
5700 
5701 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5702 
5703 	/*
5704 	 * Profile KVM exit RIPs:
5705 	 */
5706 	if (unlikely(prof_on == KVM_PROFILING)) {
5707 		unsigned long rip = kvm_rip_read(vcpu);
5708 		profile_hit(KVM_PROFILING, (void *)rip);
5709 	}
5710 
5711 
5712 	kvm_lapic_sync_from_vapic(vcpu);
5713 
5714 	r = kvm_x86_ops->handle_exit(vcpu);
5715 out:
5716 	return r;
5717 }
5718 
5719 
5720 static int __vcpu_run(struct kvm_vcpu *vcpu)
5721 {
5722 	int r;
5723 	struct kvm *kvm = vcpu->kvm;
5724 
5725 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5726 		pr_debug("vcpu %d received sipi with vector # %x\n",
5727 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5728 		kvm_lapic_reset(vcpu);
5729 		r = kvm_arch_vcpu_reset(vcpu);
5730 		if (r)
5731 			return r;
5732 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5733 	}
5734 
5735 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5736 	vapic_enter(vcpu);
5737 
5738 	r = 1;
5739 	while (r > 0) {
5740 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5741 		    !vcpu->arch.apf.halted)
5742 			r = vcpu_enter_guest(vcpu);
5743 		else {
5744 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5745 			kvm_vcpu_block(vcpu);
5746 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5747 			if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5748 			{
5749 				switch(vcpu->arch.mp_state) {
5750 				case KVM_MP_STATE_HALTED:
5751 					vcpu->arch.mp_state =
5752 						KVM_MP_STATE_RUNNABLE;
5753 				case KVM_MP_STATE_RUNNABLE:
5754 					vcpu->arch.apf.halted = false;
5755 					break;
5756 				case KVM_MP_STATE_SIPI_RECEIVED:
5757 				default:
5758 					r = -EINTR;
5759 					break;
5760 				}
5761 			}
5762 		}
5763 
5764 		if (r <= 0)
5765 			break;
5766 
5767 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5768 		if (kvm_cpu_has_pending_timer(vcpu))
5769 			kvm_inject_pending_timer_irqs(vcpu);
5770 
5771 		if (dm_request_for_irq_injection(vcpu)) {
5772 			r = -EINTR;
5773 			vcpu->run->exit_reason = KVM_EXIT_INTR;
5774 			++vcpu->stat.request_irq_exits;
5775 		}
5776 
5777 		kvm_check_async_pf_completion(vcpu);
5778 
5779 		if (signal_pending(current)) {
5780 			r = -EINTR;
5781 			vcpu->run->exit_reason = KVM_EXIT_INTR;
5782 			++vcpu->stat.signal_exits;
5783 		}
5784 		if (need_resched()) {
5785 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5786 			kvm_resched(vcpu);
5787 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5788 		}
5789 	}
5790 
5791 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5792 
5793 	vapic_exit(vcpu);
5794 
5795 	return r;
5796 }
5797 
5798 static int complete_mmio(struct kvm_vcpu *vcpu)
5799 {
5800 	struct kvm_run *run = vcpu->run;
5801 	int r;
5802 
5803 	if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5804 		return 1;
5805 
5806 	if (vcpu->mmio_needed) {
5807 		vcpu->mmio_needed = 0;
5808 		if (!vcpu->mmio_is_write)
5809 			memcpy(vcpu->mmio_data + vcpu->mmio_index,
5810 			       run->mmio.data, 8);
5811 		vcpu->mmio_index += 8;
5812 		if (vcpu->mmio_index < vcpu->mmio_size) {
5813 			run->exit_reason = KVM_EXIT_MMIO;
5814 			run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5815 			memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5816 			run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5817 			run->mmio.is_write = vcpu->mmio_is_write;
5818 			vcpu->mmio_needed = 1;
5819 			return 0;
5820 		}
5821 		if (vcpu->mmio_is_write)
5822 			return 1;
5823 		vcpu->mmio_read_completed = 1;
5824 	}
5825 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5826 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5827 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5828 	if (r != EMULATE_DONE)
5829 		return 0;
5830 	return 1;
5831 }
5832 
5833 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5834 {
5835 	int r;
5836 	sigset_t sigsaved;
5837 
5838 	if (!tsk_used_math(current) && init_fpu(current))
5839 		return -ENOMEM;
5840 
5841 	if (vcpu->sigset_active)
5842 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5843 
5844 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5845 		kvm_vcpu_block(vcpu);
5846 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5847 		r = -EAGAIN;
5848 		goto out;
5849 	}
5850 
5851 	/* re-sync apic's tpr */
5852 	if (!irqchip_in_kernel(vcpu->kvm)) {
5853 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5854 			r = -EINVAL;
5855 			goto out;
5856 		}
5857 	}
5858 
5859 	r = complete_mmio(vcpu);
5860 	if (r <= 0)
5861 		goto out;
5862 
5863 	if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5864 		kvm_register_write(vcpu, VCPU_REGS_RAX,
5865 				     kvm_run->hypercall.ret);
5866 
5867 	r = __vcpu_run(vcpu);
5868 
5869 out:
5870 	post_kvm_run_save(vcpu);
5871 	if (vcpu->sigset_active)
5872 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5873 
5874 	return r;
5875 }
5876 
5877 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5878 {
5879 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5880 		/*
5881 		 * We are here if userspace calls get_regs() in the middle of
5882 		 * instruction emulation. Registers state needs to be copied
5883 		 * back from emulation context to vcpu. Usrapace shouldn't do
5884 		 * that usually, but some bad designed PV devices (vmware
5885 		 * backdoor interface) need this to work
5886 		 */
5887 		struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5888 		memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5889 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5890 	}
5891 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5892 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5893 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5894 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5895 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5896 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5897 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5898 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5899 #ifdef CONFIG_X86_64
5900 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5901 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5902 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5903 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5904 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5905 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5906 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5907 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5908 #endif
5909 
5910 	regs->rip = kvm_rip_read(vcpu);
5911 	regs->rflags = kvm_get_rflags(vcpu);
5912 
5913 	return 0;
5914 }
5915 
5916 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5917 {
5918 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5919 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5920 
5921 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5922 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5923 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5924 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5925 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5926 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5927 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5928 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5929 #ifdef CONFIG_X86_64
5930 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5931 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5932 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5933 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5934 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5935 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5936 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5937 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5938 #endif
5939 
5940 	kvm_rip_write(vcpu, regs->rip);
5941 	kvm_set_rflags(vcpu, regs->rflags);
5942 
5943 	vcpu->arch.exception.pending = false;
5944 
5945 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5946 
5947 	return 0;
5948 }
5949 
5950 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5951 {
5952 	struct kvm_segment cs;
5953 
5954 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5955 	*db = cs.db;
5956 	*l = cs.l;
5957 }
5958 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5959 
5960 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5961 				  struct kvm_sregs *sregs)
5962 {
5963 	struct desc_ptr dt;
5964 
5965 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5966 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5967 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5968 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5969 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5970 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5971 
5972 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5973 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5974 
5975 	kvm_x86_ops->get_idt(vcpu, &dt);
5976 	sregs->idt.limit = dt.size;
5977 	sregs->idt.base = dt.address;
5978 	kvm_x86_ops->get_gdt(vcpu, &dt);
5979 	sregs->gdt.limit = dt.size;
5980 	sregs->gdt.base = dt.address;
5981 
5982 	sregs->cr0 = kvm_read_cr0(vcpu);
5983 	sregs->cr2 = vcpu->arch.cr2;
5984 	sregs->cr3 = kvm_read_cr3(vcpu);
5985 	sregs->cr4 = kvm_read_cr4(vcpu);
5986 	sregs->cr8 = kvm_get_cr8(vcpu);
5987 	sregs->efer = vcpu->arch.efer;
5988 	sregs->apic_base = kvm_get_apic_base(vcpu);
5989 
5990 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5991 
5992 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5993 		set_bit(vcpu->arch.interrupt.nr,
5994 			(unsigned long *)sregs->interrupt_bitmap);
5995 
5996 	return 0;
5997 }
5998 
5999 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6000 				    struct kvm_mp_state *mp_state)
6001 {
6002 	mp_state->mp_state = vcpu->arch.mp_state;
6003 	return 0;
6004 }
6005 
6006 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6007 				    struct kvm_mp_state *mp_state)
6008 {
6009 	vcpu->arch.mp_state = mp_state->mp_state;
6010 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6011 	return 0;
6012 }
6013 
6014 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6015 		    bool has_error_code, u32 error_code)
6016 {
6017 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6018 	int ret;
6019 
6020 	init_emulate_ctxt(vcpu);
6021 
6022 	ret = emulator_task_switch(ctxt, tss_selector, reason,
6023 				   has_error_code, error_code);
6024 
6025 	if (ret)
6026 		return EMULATE_FAIL;
6027 
6028 	memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6029 	kvm_rip_write(vcpu, ctxt->eip);
6030 	kvm_set_rflags(vcpu, ctxt->eflags);
6031 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6032 	return EMULATE_DONE;
6033 }
6034 EXPORT_SYMBOL_GPL(kvm_task_switch);
6035 
6036 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6037 				  struct kvm_sregs *sregs)
6038 {
6039 	int mmu_reset_needed = 0;
6040 	int pending_vec, max_bits, idx;
6041 	struct desc_ptr dt;
6042 
6043 	dt.size = sregs->idt.limit;
6044 	dt.address = sregs->idt.base;
6045 	kvm_x86_ops->set_idt(vcpu, &dt);
6046 	dt.size = sregs->gdt.limit;
6047 	dt.address = sregs->gdt.base;
6048 	kvm_x86_ops->set_gdt(vcpu, &dt);
6049 
6050 	vcpu->arch.cr2 = sregs->cr2;
6051 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6052 	vcpu->arch.cr3 = sregs->cr3;
6053 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6054 
6055 	kvm_set_cr8(vcpu, sregs->cr8);
6056 
6057 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6058 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
6059 	kvm_set_apic_base(vcpu, sregs->apic_base);
6060 
6061 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6062 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6063 	vcpu->arch.cr0 = sregs->cr0;
6064 
6065 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6066 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6067 	if (sregs->cr4 & X86_CR4_OSXSAVE)
6068 		update_cpuid(vcpu);
6069 
6070 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6071 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6072 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6073 		mmu_reset_needed = 1;
6074 	}
6075 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6076 
6077 	if (mmu_reset_needed)
6078 		kvm_mmu_reset_context(vcpu);
6079 
6080 	max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6081 	pending_vec = find_first_bit(
6082 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
6083 	if (pending_vec < max_bits) {
6084 		kvm_queue_interrupt(vcpu, pending_vec, false);
6085 		pr_debug("Set back pending irq %d\n", pending_vec);
6086 	}
6087 
6088 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6089 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6090 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6091 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6092 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6093 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6094 
6095 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6096 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6097 
6098 	update_cr8_intercept(vcpu);
6099 
6100 	/* Older userspace won't unhalt the vcpu on reset. */
6101 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6102 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6103 	    !is_protmode(vcpu))
6104 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6105 
6106 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6107 
6108 	return 0;
6109 }
6110 
6111 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6112 					struct kvm_guest_debug *dbg)
6113 {
6114 	unsigned long rflags;
6115 	int i, r;
6116 
6117 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6118 		r = -EBUSY;
6119 		if (vcpu->arch.exception.pending)
6120 			goto out;
6121 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6122 			kvm_queue_exception(vcpu, DB_VECTOR);
6123 		else
6124 			kvm_queue_exception(vcpu, BP_VECTOR);
6125 	}
6126 
6127 	/*
6128 	 * Read rflags as long as potentially injected trace flags are still
6129 	 * filtered out.
6130 	 */
6131 	rflags = kvm_get_rflags(vcpu);
6132 
6133 	vcpu->guest_debug = dbg->control;
6134 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6135 		vcpu->guest_debug = 0;
6136 
6137 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6138 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
6139 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6140 		vcpu->arch.switch_db_regs =
6141 			(dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6142 	} else {
6143 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6144 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6145 		vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6146 	}
6147 
6148 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6149 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6150 			get_segment_base(vcpu, VCPU_SREG_CS);
6151 
6152 	/*
6153 	 * Trigger an rflags update that will inject or remove the trace
6154 	 * flags.
6155 	 */
6156 	kvm_set_rflags(vcpu, rflags);
6157 
6158 	kvm_x86_ops->set_guest_debug(vcpu, dbg);
6159 
6160 	r = 0;
6161 
6162 out:
6163 
6164 	return r;
6165 }
6166 
6167 /*
6168  * Translate a guest virtual address to a guest physical address.
6169  */
6170 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6171 				    struct kvm_translation *tr)
6172 {
6173 	unsigned long vaddr = tr->linear_address;
6174 	gpa_t gpa;
6175 	int idx;
6176 
6177 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6178 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6179 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6180 	tr->physical_address = gpa;
6181 	tr->valid = gpa != UNMAPPED_GVA;
6182 	tr->writeable = 1;
6183 	tr->usermode = 0;
6184 
6185 	return 0;
6186 }
6187 
6188 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6189 {
6190 	struct i387_fxsave_struct *fxsave =
6191 			&vcpu->arch.guest_fpu.state->fxsave;
6192 
6193 	memcpy(fpu->fpr, fxsave->st_space, 128);
6194 	fpu->fcw = fxsave->cwd;
6195 	fpu->fsw = fxsave->swd;
6196 	fpu->ftwx = fxsave->twd;
6197 	fpu->last_opcode = fxsave->fop;
6198 	fpu->last_ip = fxsave->rip;
6199 	fpu->last_dp = fxsave->rdp;
6200 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6201 
6202 	return 0;
6203 }
6204 
6205 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6206 {
6207 	struct i387_fxsave_struct *fxsave =
6208 			&vcpu->arch.guest_fpu.state->fxsave;
6209 
6210 	memcpy(fxsave->st_space, fpu->fpr, 128);
6211 	fxsave->cwd = fpu->fcw;
6212 	fxsave->swd = fpu->fsw;
6213 	fxsave->twd = fpu->ftwx;
6214 	fxsave->fop = fpu->last_opcode;
6215 	fxsave->rip = fpu->last_ip;
6216 	fxsave->rdp = fpu->last_dp;
6217 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6218 
6219 	return 0;
6220 }
6221 
6222 int fx_init(struct kvm_vcpu *vcpu)
6223 {
6224 	int err;
6225 
6226 	err = fpu_alloc(&vcpu->arch.guest_fpu);
6227 	if (err)
6228 		return err;
6229 
6230 	fpu_finit(&vcpu->arch.guest_fpu);
6231 
6232 	/*
6233 	 * Ensure guest xcr0 is valid for loading
6234 	 */
6235 	vcpu->arch.xcr0 = XSTATE_FP;
6236 
6237 	vcpu->arch.cr0 |= X86_CR0_ET;
6238 
6239 	return 0;
6240 }
6241 EXPORT_SYMBOL_GPL(fx_init);
6242 
6243 static void fx_free(struct kvm_vcpu *vcpu)
6244 {
6245 	fpu_free(&vcpu->arch.guest_fpu);
6246 }
6247 
6248 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6249 {
6250 	if (vcpu->guest_fpu_loaded)
6251 		return;
6252 
6253 	/*
6254 	 * Restore all possible states in the guest,
6255 	 * and assume host would use all available bits.
6256 	 * Guest xcr0 would be loaded later.
6257 	 */
6258 	kvm_put_guest_xcr0(vcpu);
6259 	vcpu->guest_fpu_loaded = 1;
6260 	unlazy_fpu(current);
6261 	fpu_restore_checking(&vcpu->arch.guest_fpu);
6262 	trace_kvm_fpu(1);
6263 }
6264 
6265 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6266 {
6267 	kvm_put_guest_xcr0(vcpu);
6268 
6269 	if (!vcpu->guest_fpu_loaded)
6270 		return;
6271 
6272 	vcpu->guest_fpu_loaded = 0;
6273 	fpu_save_init(&vcpu->arch.guest_fpu);
6274 	++vcpu->stat.fpu_reload;
6275 	kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6276 	trace_kvm_fpu(0);
6277 }
6278 
6279 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6280 {
6281 	kvmclock_reset(vcpu);
6282 
6283 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6284 	fx_free(vcpu);
6285 	kvm_x86_ops->vcpu_free(vcpu);
6286 }
6287 
6288 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6289 						unsigned int id)
6290 {
6291 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6292 		printk_once(KERN_WARNING
6293 		"kvm: SMP vm created on host with unstable TSC; "
6294 		"guest TSC will not be reliable\n");
6295 	return kvm_x86_ops->vcpu_create(kvm, id);
6296 }
6297 
6298 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6299 {
6300 	int r;
6301 
6302 	vcpu->arch.mtrr_state.have_fixed = 1;
6303 	vcpu_load(vcpu);
6304 	r = kvm_arch_vcpu_reset(vcpu);
6305 	if (r == 0)
6306 		r = kvm_mmu_setup(vcpu);
6307 	vcpu_put(vcpu);
6308 
6309 	return r;
6310 }
6311 
6312 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6313 {
6314 	vcpu->arch.apf.msr_val = 0;
6315 
6316 	vcpu_load(vcpu);
6317 	kvm_mmu_unload(vcpu);
6318 	vcpu_put(vcpu);
6319 
6320 	fx_free(vcpu);
6321 	kvm_x86_ops->vcpu_free(vcpu);
6322 }
6323 
6324 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6325 {
6326 	vcpu->arch.nmi_pending = false;
6327 	vcpu->arch.nmi_injected = false;
6328 
6329 	vcpu->arch.switch_db_regs = 0;
6330 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6331 	vcpu->arch.dr6 = DR6_FIXED_1;
6332 	vcpu->arch.dr7 = DR7_FIXED_1;
6333 
6334 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6335 	vcpu->arch.apf.msr_val = 0;
6336 	vcpu->arch.st.msr_val = 0;
6337 
6338 	kvmclock_reset(vcpu);
6339 
6340 	kvm_clear_async_pf_completion_queue(vcpu);
6341 	kvm_async_pf_hash_reset(vcpu);
6342 	vcpu->arch.apf.halted = false;
6343 
6344 	return kvm_x86_ops->vcpu_reset(vcpu);
6345 }
6346 
6347 int kvm_arch_hardware_enable(void *garbage)
6348 {
6349 	struct kvm *kvm;
6350 	struct kvm_vcpu *vcpu;
6351 	int i;
6352 
6353 	kvm_shared_msr_cpu_online();
6354 	list_for_each_entry(kvm, &vm_list, vm_list)
6355 		kvm_for_each_vcpu(i, vcpu, kvm)
6356 			if (vcpu->cpu == smp_processor_id())
6357 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6358 	return kvm_x86_ops->hardware_enable(garbage);
6359 }
6360 
6361 void kvm_arch_hardware_disable(void *garbage)
6362 {
6363 	kvm_x86_ops->hardware_disable(garbage);
6364 	drop_user_return_notifiers(garbage);
6365 }
6366 
6367 int kvm_arch_hardware_setup(void)
6368 {
6369 	return kvm_x86_ops->hardware_setup();
6370 }
6371 
6372 void kvm_arch_hardware_unsetup(void)
6373 {
6374 	kvm_x86_ops->hardware_unsetup();
6375 }
6376 
6377 void kvm_arch_check_processor_compat(void *rtn)
6378 {
6379 	kvm_x86_ops->check_processor_compatibility(rtn);
6380 }
6381 
6382 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6383 {
6384 	struct page *page;
6385 	struct kvm *kvm;
6386 	int r;
6387 
6388 	BUG_ON(vcpu->kvm == NULL);
6389 	kvm = vcpu->kvm;
6390 
6391 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6392 	vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6393 	vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6394 	vcpu->arch.mmu.translate_gpa = translate_gpa;
6395 	vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6396 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6397 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6398 	else
6399 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6400 
6401 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6402 	if (!page) {
6403 		r = -ENOMEM;
6404 		goto fail;
6405 	}
6406 	vcpu->arch.pio_data = page_address(page);
6407 
6408 	kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6409 
6410 	r = kvm_mmu_create(vcpu);
6411 	if (r < 0)
6412 		goto fail_free_pio_data;
6413 
6414 	if (irqchip_in_kernel(kvm)) {
6415 		r = kvm_create_lapic(vcpu);
6416 		if (r < 0)
6417 			goto fail_mmu_destroy;
6418 	}
6419 
6420 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6421 				       GFP_KERNEL);
6422 	if (!vcpu->arch.mce_banks) {
6423 		r = -ENOMEM;
6424 		goto fail_free_lapic;
6425 	}
6426 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6427 
6428 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6429 		goto fail_free_mce_banks;
6430 
6431 	kvm_async_pf_hash_reset(vcpu);
6432 
6433 	return 0;
6434 fail_free_mce_banks:
6435 	kfree(vcpu->arch.mce_banks);
6436 fail_free_lapic:
6437 	kvm_free_lapic(vcpu);
6438 fail_mmu_destroy:
6439 	kvm_mmu_destroy(vcpu);
6440 fail_free_pio_data:
6441 	free_page((unsigned long)vcpu->arch.pio_data);
6442 fail:
6443 	return r;
6444 }
6445 
6446 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6447 {
6448 	int idx;
6449 
6450 	kfree(vcpu->arch.mce_banks);
6451 	kvm_free_lapic(vcpu);
6452 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6453 	kvm_mmu_destroy(vcpu);
6454 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6455 	free_page((unsigned long)vcpu->arch.pio_data);
6456 }
6457 
6458 int kvm_arch_init_vm(struct kvm *kvm)
6459 {
6460 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6461 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6462 
6463 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6464 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6465 
6466 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6467 
6468 	return 0;
6469 }
6470 
6471 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6472 {
6473 	vcpu_load(vcpu);
6474 	kvm_mmu_unload(vcpu);
6475 	vcpu_put(vcpu);
6476 }
6477 
6478 static void kvm_free_vcpus(struct kvm *kvm)
6479 {
6480 	unsigned int i;
6481 	struct kvm_vcpu *vcpu;
6482 
6483 	/*
6484 	 * Unpin any mmu pages first.
6485 	 */
6486 	kvm_for_each_vcpu(i, vcpu, kvm) {
6487 		kvm_clear_async_pf_completion_queue(vcpu);
6488 		kvm_unload_vcpu_mmu(vcpu);
6489 	}
6490 	kvm_for_each_vcpu(i, vcpu, kvm)
6491 		kvm_arch_vcpu_free(vcpu);
6492 
6493 	mutex_lock(&kvm->lock);
6494 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6495 		kvm->vcpus[i] = NULL;
6496 
6497 	atomic_set(&kvm->online_vcpus, 0);
6498 	mutex_unlock(&kvm->lock);
6499 }
6500 
6501 void kvm_arch_sync_events(struct kvm *kvm)
6502 {
6503 	kvm_free_all_assigned_devices(kvm);
6504 	kvm_free_pit(kvm);
6505 }
6506 
6507 void kvm_arch_destroy_vm(struct kvm *kvm)
6508 {
6509 	kvm_iommu_unmap_guest(kvm);
6510 	kfree(kvm->arch.vpic);
6511 	kfree(kvm->arch.vioapic);
6512 	kvm_free_vcpus(kvm);
6513 	if (kvm->arch.apic_access_page)
6514 		put_page(kvm->arch.apic_access_page);
6515 	if (kvm->arch.ept_identity_pagetable)
6516 		put_page(kvm->arch.ept_identity_pagetable);
6517 }
6518 
6519 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6520 				struct kvm_memory_slot *memslot,
6521 				struct kvm_memory_slot old,
6522 				struct kvm_userspace_memory_region *mem,
6523 				int user_alloc)
6524 {
6525 	int npages = memslot->npages;
6526 	int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6527 
6528 	/* Prevent internal slot pages from being moved by fork()/COW. */
6529 	if (memslot->id >= KVM_MEMORY_SLOTS)
6530 		map_flags = MAP_SHARED | MAP_ANONYMOUS;
6531 
6532 	/*To keep backward compatibility with older userspace,
6533 	 *x86 needs to hanlde !user_alloc case.
6534 	 */
6535 	if (!user_alloc) {
6536 		if (npages && !old.rmap) {
6537 			unsigned long userspace_addr;
6538 
6539 			down_write(&current->mm->mmap_sem);
6540 			userspace_addr = do_mmap(NULL, 0,
6541 						 npages * PAGE_SIZE,
6542 						 PROT_READ | PROT_WRITE,
6543 						 map_flags,
6544 						 0);
6545 			up_write(&current->mm->mmap_sem);
6546 
6547 			if (IS_ERR((void *)userspace_addr))
6548 				return PTR_ERR((void *)userspace_addr);
6549 
6550 			memslot->userspace_addr = userspace_addr;
6551 		}
6552 	}
6553 
6554 
6555 	return 0;
6556 }
6557 
6558 void kvm_arch_commit_memory_region(struct kvm *kvm,
6559 				struct kvm_userspace_memory_region *mem,
6560 				struct kvm_memory_slot old,
6561 				int user_alloc)
6562 {
6563 
6564 	int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6565 
6566 	if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6567 		int ret;
6568 
6569 		down_write(&current->mm->mmap_sem);
6570 		ret = do_munmap(current->mm, old.userspace_addr,
6571 				old.npages * PAGE_SIZE);
6572 		up_write(&current->mm->mmap_sem);
6573 		if (ret < 0)
6574 			printk(KERN_WARNING
6575 			       "kvm_vm_ioctl_set_memory_region: "
6576 			       "failed to munmap memory\n");
6577 	}
6578 
6579 	if (!kvm->arch.n_requested_mmu_pages)
6580 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6581 
6582 	spin_lock(&kvm->mmu_lock);
6583 	if (nr_mmu_pages)
6584 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6585 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6586 	spin_unlock(&kvm->mmu_lock);
6587 }
6588 
6589 void kvm_arch_flush_shadow(struct kvm *kvm)
6590 {
6591 	kvm_mmu_zap_all(kvm);
6592 	kvm_reload_remote_mmus(kvm);
6593 }
6594 
6595 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6596 {
6597 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6598 		!vcpu->arch.apf.halted)
6599 		|| !list_empty_careful(&vcpu->async_pf.done)
6600 		|| vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6601 		|| vcpu->arch.nmi_pending ||
6602 		(kvm_arch_interrupt_allowed(vcpu) &&
6603 		 kvm_cpu_has_interrupt(vcpu));
6604 }
6605 
6606 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6607 {
6608 	int me;
6609 	int cpu = vcpu->cpu;
6610 
6611 	if (waitqueue_active(&vcpu->wq)) {
6612 		wake_up_interruptible(&vcpu->wq);
6613 		++vcpu->stat.halt_wakeup;
6614 	}
6615 
6616 	me = get_cpu();
6617 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6618 		if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6619 			smp_send_reschedule(cpu);
6620 	put_cpu();
6621 }
6622 
6623 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6624 {
6625 	return kvm_x86_ops->interrupt_allowed(vcpu);
6626 }
6627 
6628 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6629 {
6630 	unsigned long current_rip = kvm_rip_read(vcpu) +
6631 		get_segment_base(vcpu, VCPU_SREG_CS);
6632 
6633 	return current_rip == linear_rip;
6634 }
6635 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6636 
6637 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6638 {
6639 	unsigned long rflags;
6640 
6641 	rflags = kvm_x86_ops->get_rflags(vcpu);
6642 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6643 		rflags &= ~X86_EFLAGS_TF;
6644 	return rflags;
6645 }
6646 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6647 
6648 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6649 {
6650 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6651 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6652 		rflags |= X86_EFLAGS_TF;
6653 	kvm_x86_ops->set_rflags(vcpu, rflags);
6654 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6655 }
6656 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6657 
6658 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6659 {
6660 	int r;
6661 
6662 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6663 	      is_error_page(work->page))
6664 		return;
6665 
6666 	r = kvm_mmu_reload(vcpu);
6667 	if (unlikely(r))
6668 		return;
6669 
6670 	if (!vcpu->arch.mmu.direct_map &&
6671 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6672 		return;
6673 
6674 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6675 }
6676 
6677 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6678 {
6679 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6680 }
6681 
6682 static inline u32 kvm_async_pf_next_probe(u32 key)
6683 {
6684 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6685 }
6686 
6687 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6688 {
6689 	u32 key = kvm_async_pf_hash_fn(gfn);
6690 
6691 	while (vcpu->arch.apf.gfns[key] != ~0)
6692 		key = kvm_async_pf_next_probe(key);
6693 
6694 	vcpu->arch.apf.gfns[key] = gfn;
6695 }
6696 
6697 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6698 {
6699 	int i;
6700 	u32 key = kvm_async_pf_hash_fn(gfn);
6701 
6702 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6703 		     (vcpu->arch.apf.gfns[key] != gfn &&
6704 		      vcpu->arch.apf.gfns[key] != ~0); i++)
6705 		key = kvm_async_pf_next_probe(key);
6706 
6707 	return key;
6708 }
6709 
6710 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6711 {
6712 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6713 }
6714 
6715 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6716 {
6717 	u32 i, j, k;
6718 
6719 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6720 	while (true) {
6721 		vcpu->arch.apf.gfns[i] = ~0;
6722 		do {
6723 			j = kvm_async_pf_next_probe(j);
6724 			if (vcpu->arch.apf.gfns[j] == ~0)
6725 				return;
6726 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6727 			/*
6728 			 * k lies cyclically in ]i,j]
6729 			 * |    i.k.j |
6730 			 * |....j i.k.| or  |.k..j i...|
6731 			 */
6732 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6733 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6734 		i = j;
6735 	}
6736 }
6737 
6738 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6739 {
6740 
6741 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6742 				      sizeof(val));
6743 }
6744 
6745 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6746 				     struct kvm_async_pf *work)
6747 {
6748 	struct x86_exception fault;
6749 
6750 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6751 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6752 
6753 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6754 	    (vcpu->arch.apf.send_user_only &&
6755 	     kvm_x86_ops->get_cpl(vcpu) == 0))
6756 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6757 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6758 		fault.vector = PF_VECTOR;
6759 		fault.error_code_valid = true;
6760 		fault.error_code = 0;
6761 		fault.nested_page_fault = false;
6762 		fault.address = work->arch.token;
6763 		kvm_inject_page_fault(vcpu, &fault);
6764 	}
6765 }
6766 
6767 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6768 				 struct kvm_async_pf *work)
6769 {
6770 	struct x86_exception fault;
6771 
6772 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
6773 	if (is_error_page(work->page))
6774 		work->arch.token = ~0; /* broadcast wakeup */
6775 	else
6776 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6777 
6778 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6779 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6780 		fault.vector = PF_VECTOR;
6781 		fault.error_code_valid = true;
6782 		fault.error_code = 0;
6783 		fault.nested_page_fault = false;
6784 		fault.address = work->arch.token;
6785 		kvm_inject_page_fault(vcpu, &fault);
6786 	}
6787 	vcpu->arch.apf.halted = false;
6788 }
6789 
6790 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6791 {
6792 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6793 		return true;
6794 	else
6795 		return !kvm_event_needs_reinjection(vcpu) &&
6796 			kvm_x86_ops->interrupt_allowed(vcpu);
6797 }
6798 
6799 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6800 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6801 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6802 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6803 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6804 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6805 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6806 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6807 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6808 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6809 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6810 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
6811